paper_id
stringlengths
9
16
version
stringclasses
26 values
yymm
stringclasses
311 values
created
timestamp[s]
title
stringlengths
6
335
secondary_subfield
sequencelengths
1
8
abstract
stringlengths
25
3.93k
primary_subfield
stringclasses
124 values
field
stringclasses
20 values
fulltext
stringlengths
0
2.84M
0910.3607
2
0910
2011-03-14T11:21:47
Multigraded Factorial Rings and Fano varieties with torus action
[ "math.AG", "math.AC" ]
In a first result, we describe all finitely generated factorial algebras over an algebraically closed field of characteristic zero that come with an effective multigrading of complexity one by means of generators and relations. This enables us to construct systematically varieties with free divisor class group and a complexity one torus action via their Cox rings. For the Fano varieties of this type that have a free divisor class group of rank one, we provide explicit bounds for the number of possible deformation types depending on the dimension and the index of the Picard group in the divisor class group. As a consequence, one can produce classification lists for fixed dimension and Picard index. We carry this out expemplarily in the following cases. There are 15 non-toric surfaces with Picard index at most six. Moreover, there are 116 non-toric threefolds with Picard index at most two; nine of them are locally factorial, i.e. of Picard index one, and among these one is smooth, six have canonical singularities and two have non-canonical singularities. Finally, there are 67 non-toric locally factorial fourfolds and two one-dimensional families of non-toric locally factorial fourfolds. In all cases, we list the Cox rings explicitly.
math.AG
math
MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES WITH TORUS ACTION J URGEN HAUSEN, ELAINE HERPPICH, AND HENDRIK S USS Abstract. In a first result, we describe all finitely generated factorial alge- bras over an algebraically closed field of characteristic zero that come with an effective multigrading of complexity one by means of generators and relations. This enables us to construct systematically varieties with free divisor class group and a complexity one torus action via their Cox rings. For the Fano varieties of this type that have a free divisor class group of rank one, we pro- vide explicit bounds for the number of possible deformation types depending on the dimension and the index of the Picard group in the divisor class group. As a consequence, one can produce classification lists for fixed dimension and Picard index. We carry this out expemplarily in the following cases. There are 15 non-toric surfaces with Picard index at most six. Moreover, there are 116 non-toric threefolds with Picard index at most two; nine of them are locally factorial, i.e. of Picard index one, and among these one is smooth, six have canonical singularities and two have non-canonical singularities. Finally, there are 67 non-toric locally factorial fourfolds and two one-dimensional families of non-toric locally factorial fourfolds. In all cases, we list the Cox rings explicitly. Introduction Let K be an algebraically closed field of characteristic zero. A first aim of this paper is to determine all finitely generated factorial K-algebras R with an effec- tive complexity one multigrading R = ⊕u∈M Ru satisfying R0 = K; here effective complexity one multigrading means that with d := dim R we have M ∼= Zd−1 and the u ∈ M with Ru 6= 0 generate M as a Z-module. Our result extends work by Mori [23] and Ishida [17], who settled the cases d = 2 and d = 3. An obvious class of multigraded factorial algebras as above is given by polynomial rings. A much larger class is obtained as follows. Take a sequence A = (a0, . . . , ar) of vectors ai ∈ K2 such that (ai, ak) is linearly independent whenever k 6= i, a sequence n = (n0, . . . , nr) of positive integers and a family L = (lij) of positive integers, where 0 ≤ i ≤ r and 1 ≤ j ≤ ni. For every 0 ≤ i ≤ r, we define a monomial fi := T li1 i1 · · · T lini ini ∈ K[Tij; 0 ≤ i ≤ r, 1 ≤ j ≤ ni], for any two indices 0 ≤ i, j ≤ r, we set αij := det(ai, aj), and for any three indices 0 ≤ i < j < k ≤ r, we define a trinomial gi,j,k := αjkfi + αkifj + αij fk ∈ K[Tij; 0 ≤ i ≤ r, 1 ≤ j ≤ ni]. Note that the coefficients of gi,j,k are all nonzero. The triple (A, n, L) then defines a K-algebra R(A, n, L) := K[Tij; 0 ≤ i ≤ r, 1 ≤ j ≤ ni] / hgi,i+1,i+2; 0 ≤ i ≤ r − 2i. It turns out that R(A, n, L) is a normal complete intersection, see Proposition 1.2. In particular, it is of dimension dim R(A, n, L) = n0 + . . . + nr − r + 1. 2000 Mathematics Subject Classification. 13A02, 13F15, 14J45. 1 2 J. HAUSEN, E. HERPPICH, AND H. S USS If the triple (A, n, L) is admissible, i.e., the numbers gcd(li1, . . . , lini), where 0 ≤ i ≤ r, are pairwise coprime, then R(A, n, L) admits a canonical effective complexity one grading by a lattice K, see Construction 1.7. Our first result is the following. Theorem 1.9. Up to isomorphy, the finitely generated factorial K-algebras with an effective complexity one grading R = ⊕M Ru and R0 = K are (i) the polynomial algebras K[T1, . . . , Td] with a grading deg(Ti) = ui ∈ Zd−1 such that u1, . . . , ud generate Zd−1 as a lattice and the convex cone on Qd−1 generated by u1, . . . , ud is pointed, (ii) the (K × Zm)-graded algebras R(A, n, L)[S1, . . . , Sm], where R(A, n, L) is the K-graded algebra defined by an admissible triple (A, n, L) and deg Sj ∈ Zm is the j-th canonical base vector. The further paper is devoted to normal (possibly singular) d-dimensional Fano varieties X with an effective action of an algebraic torus T . In the case dim T = d, we have the meanwhile extensively studied class of toric Fano varieties, see [3], [27] and [4] for the initiating work. Our aim is to show that the above Theorem provides an approach to classification results for the case dim T = d − 1, that means Fano varieties with a complexity one torus action. Here, we treat the case of divisor class group Cl(X) ∼= Z; note that in the toric setting this gives precisely the weighted projective spaces. The idea is to consider the Cox ring R(X) = MD∈Cl(X) Γ(X, OX (D)). The ring R(X) is factorial, finitely generated as a K-algebra and the T -action on X gives rise to an effective complexity one multigrading of R(X) refining the Cl(X)-grading, see [5] and [15]. Consequently, R(X) is one of the rings listed in the first Theorem. Moreover, X can be easily reconstructed from R(X); it is the homogeneous spectrum with respect to the Cl(X)-grading of R(X). Thus, in order to construct Fano varieties, we firstly have to figure out the Cox rings among the rings occuring in the first Theorem and then find those, which belong to a Fano variety; this is done in Propositions 1.11 and 2.5. In order to produce classification results via this approach, we need explicit bounds on the number of deformation types of Fano varieties with prescribed dis- crete invariants. Besides the dimension, in our setting, a suitable invariant is the Picard index [Cl(X) : Pic(X)]. Denoting by ξ(µ) the number of primes less or equal to µ, we obtain the following bound, see Corollary 2.2: for any pair (d, µ) ∈ Z2 >0, the number δ(d, µ) of different deformation types of d-dimensional Fano varieties with a complexity one torus action such that Cl(X) ∼= Z and µ = [Cl(X) : Pic(X)] hold is bounded by δ(d, µ) ≤ (6dµ)2ξ(3dµ)+d−2µξ(µ)2+2ξ((d+2)µ)+2d+2. In particular, we conclude that for fixed µ ∈ Z>0, the number δ(d) of different de- formation types of d-dimensional Fano varieties with a complexity one torus action Cl(X) ∼= Z and Picard index µ is asymptotically bounded by dAd with a constant A depending only on µ, see Corollary 2.4. In fact, in Theorem 2.1 we even obtain explicit bounds for the discrete input data of the rings R(A, n, L)[S1, . . . , Sm]. This allows us to construct all Fano varieties X with prescribed dimension and Picard index that come with an effective complexity one torus action and have divisor class group Z. Note that, by the approach, we get the Cox rings of the resulting Fano varieties X for free. In Section 3, we give some explicit classifications. We list all non-toric surfaces X with Picard index at most six and the non-toric threefolds X with Picard index up at most two. They all have a Cox ring defined by a single relation; in fact, for surfaces the first Cox ring MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 3 with more than one relation occurs for Picard index 29, and for the threefolds this happens with Picard index 3, see Proposition 3.5 as well as Examples 3.4 and 3.7. Moreover, we determine all locally factorial fourfolds X, i.e. those of Picard index one: 67 of them occur sporadic and there are two one-dimensional families. Here comes the result on the locally factorial threefolds; in the table, we denote by wi the Cl(X)-degree of the variable Ti. Theorem 3.2. The following table lists the Cox rings R(X) of the three-dimensional locally factorial non-toric Fano varieties X with an effective two torus action and Cl(X) = Z. No. 1 2 3 4 5 6 7 8 9 2 + T 3 3 + T 3 3 + T 3 R(X) K[T1, . . . , T5] / hT1T 5 3 + T 2 4 i 4 + T 2 K[T1, . . . , T5] / hT1T2T 4 5 i K[T1, . . . , T5] / hT1T 2 2 T 3 4 + T 2 5 i K[T1, . . . , T5] / hT1T2 + T3T4 + T 2 5 i 4 + T 3 K[T1, . . . , T5] / hT1T 2 5 i K[T1, . . . , T5] / hT1T 3 4 + T 4 5 i 4 + T 2 K[T1, . . . , T5] / hT1T 3 5 i 4 + T 2 K[T1, . . . , T5] / hT1T 5 5 i K[T1, . . . , T5] / hT1T 5 4 + T 2 5 i 2 + T3T 2 2 + T3T 3 2 + T3T 3 2 + T3T 5 2 + T 3 3 T 3 (w1, . . . , w5) (1, 1, 2, 3, 1) (1, 1, 1, 2, 3) (1, 1, 1, 2, 3) (1, 1, 1, 1, 1) (1, 1, 1, 1, 1) (1, 1, 1, 1, 1) (1, 1, 1, 1, 2) (1, 1, 1, 1, 3) (1, 1, 1, 1, 3) (−KX)3 8 8 8 54 24 4 16 2 2 Note that each of these varieties X is a hypersurface in the respective weighted projective space P(w1, . . . , w5). Except number 4, none of them is quasismooth in the sense that Spec R(X) is singular at most in the origin; quasismooth hypersur- faces of weighted projective spaces were studied in [21] and [7]. In Section 4, we take a closer look at the singularities of the threefolds listed above. It turns out that number 1,3,5,7 and 9 are singular with only canonical singularities and all of them admit a crepant resolution. Number 6 and 8 are singular with non-canonical singularities but admit a smooth relative minimal model. Number two is singular with only canonical singularities, one of them of type cA1, and it admits only a singular relative minimal model. Moreover, in all cases, we determine the Cox rings of the resolutions. The authors would like to thank Ivan Arzhantsev for helpful comments and discussions and also the referee for valuable remarks and many references. 1. UFDs with complexity one multigrading As mentioned before, we work over an algebraically closed field K of characteristic zero. In Theorem 1.9, we describe all factorial finitely generated K-algebras R with an effective complexity one grading and R0 = K. Moreover, we characterize the possible Cox rings among these algebras, see Proposition 1.11. First we recall the construction sketched in the introduction. Construction 1.1. Consider a sequence A = (a0, . . . , ar) of vectors ai = (bi, ci) in K2 such that any pair (ai, ak) with k 6= i is linearly independent, a sequence n = (n0, . . . , nr) of positive integers and a family L = (lij ) of positive integers, where 0 ≤ i ≤ r and 1 ≤ j ≤ ni. For every 0 ≤ i ≤ r, define a monomial fi := T li1 i1 · · · T lini ini ∈ K[Tij; 0 ≤ i ≤ r, 1 ≤ j ≤ ni], 4 J. HAUSEN, E. HERPPICH, AND H. S USS for any two indices 0 ≤ i, j ≤ r, set αij := det(ai, aj) = bicj − bjci and for any three indices 0 ≤ i < j < k ≤ r define a trinomial gi,j,k := αjkfi + αkifj + αij fk ∈ K[Tij; 0 ≤ i ≤ r, 1 ≤ j ≤ ni]. Note that the coefficients of this trinomial are all nonzero. The triple (A, n, L) then defines a ring R(A, n, L) := K[Tij; 0 ≤ i ≤ r, 1 ≤ j ≤ ni] / hgi,i+1,i+2; 0 ≤ i ≤ r − 2i. Proposition 1.2. For every triple (A, n, L) as in 1.1, the ring R(A, n, L) is a normal complete intersection of dimension dim R(A, n, L) = n − r + 1, n := n0 + . . . + nr. Lemma 1.3. In the setting of 1.1, one has for any 0 ≤ i < j < k < l ≤ r the identities gi,k,l = αkl · gi,j,k + αik · gj,k,l, gi,j,l = αjl · gi,j,k + αij · gj,k,l. In particular, every trinomial gi,j,k, where 0 ≤ i < j < k ≤ r is contained in the ideal hgi,i+1,i+2; 0 ≤ i ≤ r − 2i. Proof. The identities are easily obtained by direct computation; note that for this one may assume aj = (1, 0) and ak = (0, 1). The supplement then follows by repeated application of the identities. (cid:3) Lemma 1.4. In the notation of 1.1 and 1.2, set X := V (Kn, g0, . . . , gr−2), and let z ∈ X. If we have fi(z) = fj(z) = 0 for two 0 ≤ i < j ≤ r, then fk(z) = 0 holds for all 0 ≤ k ≤ r. Proof. If i < k < j holds, then, according to Lemma 1.3, we have gi,k,j(z) = 0, which implies fk(z) = 0. The cases k < i and j < k are obtained similarly. (cid:3) Proof of Proposition 1.2. Set X := V (Kn; g0, . . . , gr−2), where gi := gi,i+1,i+2. Then we have to show that X is a connected complete intersection with at most normal singularities. In order to see that X is connected, set ℓ := Q niQ lij and l−1 ij . Then X ⊆ Kn is invariant under the K∗-action given by ζij := ℓn−1 i t · z := (tζij zij) and the point 0 ∈ Kn lies in the closure of any orbit K∗·x ⊆ X, which implies con- nectedness. To proceed, consider the Jacobian Jg of g := (g0, . . . , gr−2). According to Serre's criterion, we have to show that the set of points of z ∈ X with Jg(z) not of full rank is of codimension at least two in X. Note that the Jacobian Jg is of the shape Jg =   δ0 0 0 δ0 1 δ1 1 δ0 2 δ1 2 0 δ1 3 0 ... 0 0 0 δr−3 r−3 0 δr−3 r−2 δr−2 r−2 δr−3 r−1 δr−2 r−1 0 δr−2 r   where δti is a nonzero multiple of the gradient δi := grad fi. Consider z ∈ X with Jg(z) not of full rank. Then δi(z) = 0 = δk(z) holds with some 0 ≤ i < k ≤ r. This implies zij = 0 = zkl for some 1 ≤ j ≤ ni and 1 ≤ l ≤ nk. Thus, we have fi(z) = 0 = fk(z). Lemma 1.4 gives fs(z) = 0, for all 0 ≤ s ≤ r. Thus, some coordinate zst must vanish for every 0 ≤ s ≤ r. This shows that z belongs to a closed subset of X having codimension at least two in X. (cid:3) MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 5 Lemma 1.5. Notation as in 1.1. Then the variable Tij defines a prime ideal in R(A, n, L) if and only if the numbers gcd(lk1, . . . , lknk ), where k 6= i, are pairwise coprime. Proof. We treat exemplarily T01. Using Lemma 1.3, we see that the ideal of relations of R(A, n, L) can be presented as follows hgs,s+1,s+2; 0 ≤ s ≤ r − 2i = hg0,s,s+1; 1 ≤ s ≤ r − 1i. Thus, the ideal hT01i ⊆ R(A, n, L) is prime if and only if the following binomial ideal is prime a := hαs+1 0fs + α0sfs+1; 1 ≤ s ≤ r − 1i ⊆ K[Tij; (i, j) 6= (0, 1)]. Set li := (li1, . . . , lini ). Then the ideal a is prime if and only if the following family can be complemented to a lattice basis (l1, −l2, 0, . . . , 0), . . . , (0, . . . , 0, lr−1, −lr). This in turn is equivalent to the statement that the numbers gcd(lk1, . . . , lknk ), where 1 ≤ k ≤ r, are pairwise coprime. (cid:3) Definition 1.6. We say that a triple (A, n, L) as in 1.1 is admissible if the numbers gcd(li1, . . . , lini ), where 0 ≤ i ≤ r, are pairwise coprime. Construction 1.7. Let (A, n, L) be an admissible triple and consider the following free abelian groups E := rMi=0 niMj=1 Z·eij, K := n0Mj=1 Z·u0j ⊕ rMi=1 ni−1Mj=1 Z·uij and define vectors uini := u01 + . . . + u0r − ui1 − . . . − uini−1 ∈ K. Then there is an epimorphism λ : E → K fitting into a commutative diagram with exact rows eij 7→eij 0 0 / E eij 7→uij η / K eij 7→lij eij α β E λ / K Li,j Z/lijZ /Li,j Z/lijZ ∼= 0 / 0 Define a K-grading of K[Tij; 0 ≤ i ≤ r, 1 ≤ j ≤ ni] by setting deg Tij := λ(eij ). Then every fi = T li1 is K-homogeneous of degree i1 · · · T lini ini deg fi = li1λ(ei1) + . . . + lini λ(eini ) = l01λ(e01) + . . . + l0n0λ(e0n0 ) ∈ K. Thus, the polynomials gi,j,k of 1.1 are all K-homogeneous of the same degree and we obtain an effective K-grading of complexity one of R(A, n, L). Proof. Only for the existence of the commutative diagram there is something to show. Write for short li := (li1, . . . , lini). By the admissibility condition, the vectors vi := (0, . . . , 0, li, −li+1, 0, . . . , 0), where 0 ≤ i ≤ r − 1, can be completed to a lattice basis for E. Consequently, we find an epimorphism λ : E → K having precisely lin(v0, . . . , vr−1) as its kernel. By construction, ker(λ) equals α(ker(η)). Using this, we obtain the induced morphism β : K → K and the desired properties. (cid:3) Lemma 1.8. Notation as in 1.7. Then R(A, n, L)0 = K and R(A, n, L)∗ = K∗ hold. Moreover, the Tij define pairwise nonassociated prime elements in R(A, n, L). / / /   / /   / / O O   / / / / 6 J. HAUSEN, E. HERPPICH, AND H. S USS Proof. The fact that all elements of degree zero are constant is due to the fact that all degrees deg Tij = uij ∈ K are non-zero and generate a pointed convex cone in KQ. As a consequence, we obtain that all units in R(A, n, L) are constant. The Tij are prime by the admissibility condition and Lemma 1.5, and they are pairwise nonassociated because they have pairwise different degrees and all units are constant. (cid:3) Theorem 1.9. Up to isomorphy, the finitely generated factorial K-algebras with an effective complexity one grading R = ⊕M Ru and R0 = K are (i) the polynomial algebras K[T1, . . . , Td] with a grading deg(Ti) = ui ∈ Zd−1 such that u1, . . . , ud generate Zd−1 as a lattice and the convex cone on Qd−1 generated by u1, . . . , ud is pointed, (ii) the (K × Zm)-graded algebras R(A, n, L)[S1, . . . , Sm], where R(A, n, L) is the K-graded algebra defined by an admissible triple (A, n, L) as in 1.1 and 1.7 and deg Sj ∈ Zm is the j-th canonical base vector. Proof. We first show that for any admissible triple (A, n, L) the ring R(A, n, L) is a unique factorization domain. If lij = 1 holds for any two i, j, then, by [15, Prop. 2.4], the ring R(A, n, L) is the Cox ring of a space P1(A, n) and hence is a unique factorization domain. Now, let (A, n, L) be arbitrary admissible data and let λ : E → K be an epimor- phism as in 1.7. Set n := n0 + . . . + nr and consider the diagonalizable groups H0 := Spec K[⊕i,jZ/lijZ]. Tn := Spec K[E], H := Spec K[K], Then Tn = (K∗)n is the standard n-torus and H0 is the direct product of the cyclic subgroups Hij := Spec K[Z/lijZ]. Moreover, the diagram in 1.7 gives rise to a commutative diagram with exact rows 0 0 (t lij ij )←[(tij ) Tn  H Tn ı H H0 ∼= H0 0 0 where tij = χeij are the coordinates of Tn corresponding to the characters eij ∈ E and the maps ı,  are the closed embeddings corresponding to the epimorphisms η, λ respectively. Setting deg Tij := eij defines an action of Tn on Kn = Spec K[Tij]; in terms of the coordinates zij corresponding to Tij this action is given by t · z = (tij zij). The torus H acts effectively on Kn via the embedding  : H → Tn. The generic isotropy group of H along V (Kn, Tij) is the subgroup Hij ⊆ H corresponding to K → K/λ(Eij ), where Eij ⊆ E denotes the sublattice generated by all ekl with (k, l) 6= (i, j); recall that we have K/λ(Eij ) ∼= Z/lijZ. Now, set l′ ij := 1 for any two i, j and consider the spectra X := Spec R(A, n, L) and X ′ := Spec R(A, n, L′). Then the canonical surjections K[Tij] → R(A, n, L) and K[Tij] → R(A, n, L′) define embeddings X → Kn and X ′ → Kn. These embeddings fit into the following commutative diagram (z lij ij )←[(zij ) π Kn o X ′ o Kn X The action of H leaves X invariant and the induced H-action on X is the one given by the K-grading of R(A, n, L). Moreover, π : Kn → Kn is the quotient map for the o o o o o o o o o o O O o o O O o o   O O o o o o O O O O MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 7 induced action of H0 ⊆ H on Kn, we have X = π−1(X ′), and hence the restriction π : X → X ′ is a quotient map for the induced action of H0 on X. Removing all subsets V (X; Tij, Tkl), where (i, j) 6= (k, l) from X, we obtain an open subset U ⊆ X. By Lemma 1.8, the complement X \ U is of codimension at least two and each V (U, Tij) is irreducible. By construction, the only isotropy groups of the H-action on U are the groups Hij of the points of V (U, Tij). The image U ′ := π(U ) is open in X ′, the complement X ′ \ U ′ is as well of codimension at least two and H/H0 acts freely on U ′. According to [22, Cor. 5.3], we have two exact sequences fitting into the following diagram 1 Pic(U ′) π∗ 1 / X(H0) α / / PicH0 (U ) β Pic(U ) δ Qi,j X(Hij) Since X ′ is factorial, the Picard group Pic(U ′) is trivial and we obtain that δ is injective. Since H0 is the direct product of the isotropy groups Hij of the Luna strata V (U, Tij), we see that δ ◦ α is an isomorphism. It follows that δ is surjective and hence an isomorphism. This in turn shows that α is an isomorphism. Now, every bundle on U is H-linearizable. Since H0 acts as a subgroup of H, we obtain that every bundle is H0-linearizable. It follows that β is surjective and hence Pic(U ) is trivial. We conclude Cl(X) = Pic(U ) = 0, which means that R(A, n, L) admits unique factorization. The second thing we have to show is that any finitely generated factorial K- algebra R with an effective complexity one multigrading satisfying R0 = K is as claimed. Consider the action of the torus G on X = Spec R defined by the multi- grading, and let X0 ⊆ X be the set of points having finite isotropy Gx. Then [15, Prop 3.3] provides a graded splitting R ∼= R′[S1, . . . , Sm], where the variables Sj are identified with the homogeneous functions defining the prime divisors Ej inside the boundary X \ X0 and R′ is the ring of functions of X0, which are invariant under the subtorus G0 ⊆ G generated by the generic isotropy groups Gj of Ej. Since R′ 0 = R0 = K holds, the orbit space X0/G has only constant functions and thus is a space P1(A, n) as constructed in [15, Section 2]. This allows us to proceed exactly as in the proof of Theorem [15, Thm 1.3] and gives R′ = R(A, n, L). The admissibility condition follows from Lemma 1.5 and the fact that each Tij defines a prime element in R′. (cid:3) Remark 1.10. Let (A, n, L) be an admissible triple with n = (1, . . . , 1). Then K = Z holds, the admissibility condition just means that the numbers lij are pairwise coprime and we have dim R(A, n, L) = n0 + . . . + nr − r + 1 = 2. Consequently, for two-dimensional rings, Theorem 1.9 specializes to Mori's descrip- tion of almost geometrically graded two-dimensional unique factorization domains provided in [23].     / / /   8 J. HAUSEN, E. HERPPICH, AND H. S USS Proposition 1.11. Let (A, n, L) be an admissible triple, consider the associated (K ×Zm)-graded ring R(A, n, L)[S1, . . . , Sm] as in Theorem 1.9 and let µ : K ×Zm → K ′ be a surjection onto an abelian group K ′. Then the following statements are equivalent. (i) The K ′-graded ring R(A, n, L)[S1, . . . , Sm] is the Cox ring of a projective variety X ′ with Cl(X ′) ∼= K ′. (ii) For every pair i, j with 0 ≤ i ≤ r and 1 ≤ j ≤ ni, the group K ′ is generated by the elements µ(λ(ekl)) and µ(es), where (i, j) 6= (k, l) and 1 ≤ s ≤ m, for every 1 ≤ t ≤ m, the group K ′ is generated by the elements µ(λ(eij )) and µ(es), where 0 ≤ i ≤ r, 1 ≤ j ≤ ni and s 6= t, and, finally the following cone is of full dimension in K ′ Q: \(k,l) cone(µ(λ(eij )), µ(es); (i, j) 6= (k, l)) ∩ \t cone(µ(λ(eij )), µ(es); s 6= t). Proof. Suppose that (i) holds, let p : bX ′ → X ′ denote the universal torsor and let X ′′ ⊆ X ′ be the set of smooth points. According to [14, Prop. 2.2], the group H ′ = Spec K[K ′] acts freely on p−1(X ′′), which is a big open subset of the total coordinate space Spec R(A, n, L)[S1, . . . , Sm]. This implies the first condition of (ii). Moreover, by [14, Prop. 4.1], the displayed cone is the moving cone of X ′ and hence of full dimension. Conversely, if (ii) holds, then the K ′-graded ring R(A, n, L)[S1, . . . , Sm] can be made into a bunched ring and hence is the Cox ring of a projective variety, use [14, Thm. 3.6]. (cid:3) 2. Bounds for Fano varieties We consider d-dimensional Fano varieties X that come with a complexity one torus action and have divisor class group Cl(X) ∼= Z. Then the Cox ring R(X) of X is factorial [5, Prop. 8.4] and has an effective complexity one grading, which refines the Cl(X)-grading, see [15, Prop. 2.6]. Thus, according to Theorem 1.9, it is of the form R(X) ∼= K[Tij; 0 ≤ i ≤ r, 1 ≤ j ≤ ni][S1, . . . , Sm] / hgi,i+1,i+2; 0 ≤ i ≤ r − 2i, gi,j,k := αjkT li1 i1 · · · T lini ini + αkiT lj1 j1 · · · T ljnj jnj + αij T lk1 k1 · · · T lknk knk . Here, we may (and will) assume n0 ≥ . . . ≥ nr ≥ 1. With n := n0 + . . . + nr, we have n + m = d + r. For the degrees of the variables in Cl(X) ∼= Z, we write wij := deg Tij for 0 ≤ i ≤ r, 1 ≤ j ≤ ni and uk = deg Sk for 1 ≤ k ≤ m. Moreover, for µ ∈ Z>0, we denote by ξ(µ) the number of primes in {2, . . . , µ}. The following result provides bounds for the discrete data of the Cox ring. Theorem 2.1. In the above situation, fix the dimension d = dim(X) and the Picard index µ = [Cl(X) : Pic(X)]. Then we have uk ≤ µ for 1 ≤ k ≤ m. Moreover, for the degree γ of the relations, the weights wij and the exponents lij , where 0 ≤ i ≤ r and 1 ≤ j ≤ ni one obtains the following. (i) Suppose that r = 0, 1 holds. Then n + m ≤ d + 1 holds and one has the bounds wij ≤ µ for 0 ≤ i ≤ r and 1 ≤ j ≤ ni, and the Picard index is given by µ = lcm(wij , uk; 0 ≤ i ≤ r, 1 ≤ j ≤ ni, 1 ≤ k ≤ m). MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 9 (ii) Suppose that r ≥ 2 and n0 = 1 hold. Then r ≤ ξ(µ) − 1 and n = r + 1 and m = d − 1 hold and one has wi1 ≤ µr for 0 ≤ i ≤ r, l01 · · · lr1 µ, l01 · · · lr1 γ ≤ µr+1, and the Picard index is given by µ = lcm(gcd(wj1; j 6= i), uk; 0 ≤ i ≤ r, 1 ≤ k ≤ m). (iii) Suppose that r ≥ 2 and n0 > n1 = 1 hold. Then we may assume l11 > . . . > lr1 ≥ 2, we have r ≤ ξ(3dµ) − 1 and n0 + m = d and the bounds w01, . . . , w0n0 ≤ µ, l01, . . . , l0n0 < 6dµ, w11, l21 < 2dµ, w21, l11 < 3dµ, wi1 < 6dµ, li1 < 2dµ for 2 ≤ i ≤ r, l11 · · · lr1 γ < 6dµ, and the Picard index is given by µ = lcm(w0j , gcd(w11, . . . , wr1), uk; 1 ≤ j ≤ n0, 1 ≤ k ≤ m). (iv) Suppose that n1 > n2 = 1 holds. Then we may assume l21 > . . . > lr1 ≥ 2, we have r ≤ ξ(2(d + 1)µ) − 1 and n0 + n1 + m = d + 1 and the bounds wij ≤ µ for i = 0, 1 and 1 ≤ j ≤ ni, w21 < (d + 1)µ, wij , lij < 2(d + 1)µ for 0 ≤ i ≤ r and 1 ≤ j ≤ ni, l21 · · · lr1 γ < 2(d + 1)µ, and the Picard index is given by µ = lcm(wij , uk; 0 ≤ i ≤ 1, 1 ≤ j ≤ ni, 1 ≤ k ≤ m). (v) Suppose that n2 > 1 holds and let s be the maximal number with ns > 1. Then one may assume ls+1,1 > . . . > lr1 ≥ 2, we have r ≤ ξ((d + 2)µ) − 1 and n0 + . . . + ns + m = d + s and the bounds wij ≤ µ, for 0 ≤ i ≤ s, wij , lij < (d + 2)µ for 0 ≤ i ≤ r and 1 ≤ j ≤ ni, ls+1,1 · · · lr1 γ < (d + 2)µ, and the Picard index is given by µ = lcm(wij , uk; 0 ≤ i ≤ s, 1 ≤ j ≤ ni, 1 ≤ k ≤ m). Putting all the bounds of the theorem together, we obtain the following (raw) bound for the number of deformation types. Corollary 2.2. For any pair (d, µ) ∈ Z2 >0, the number δ(d, µ) of different deforma- tion types of d-dimensional Fano varieties with a complexity one torus action such that Cl(X) ∼= Z and [Cl(X) : Pic(X)] = µ hold is bounded by δ(d, µ) ≤ (6dµ)2ξ(3dµ)+d−2µξ(µ)2+2ξ((d+2)µ)+2d+2. Proof. By Theorem 2.1 the discrete data r, n, L and m occuring in R(X) are bounded as in the assertion. The continuous data in R(X) are the coefficients αij ; they stem from the family A = (a0, . . . , ar) of points ai ∈ K2. Varying the ai provides flat families of Cox rings and hence, by passing to the homogeneous spectra, flat families of the resulting Fano varieties X. (cid:3) Corollary 2.3. Fix d ∈ Z>0. Then the number δ(µ) of different deformation types of d-dimensional Fano varieties with a complexity one torus action, Cl(X) ∼= Z and Picard index µ := [Cl(X) : Pic(X)] is asymptotically bounded by µAµ2/ log2 µ with a constant A depending only on d. 10 J. HAUSEN, E. HERPPICH, AND H. S USS Corollary 2.4. Fix µ ∈ Z>0. Then the number δ(d) of different deformation types of d-dimensional Fano varieties with a complexity one torus action, Cl(X) ∼= Z and Picard index µ := [Cl(X) : Pic(X)] is asymptotically bounded by dAd with a constant A depending only on µ. We first recall the necessary facts on Cox rings, for details, we refer to [14]. Let X be a complete d-dimensional variety with divisor class group Cl(X) ∼= Z. Then the Cox ring R(X) is finitely generated and the total coordinate space X := Spec R(X) is a factorial affine variety coming with an action of K∗ defined by the Cl(X)-grading of R(X). Choose a system f1, . . . , fν of homogeneous pairwise nonassociated prime generators for R(X). This provides an K∗-equivariant embedding X → Kν, x 7→ (f1(x), . . . , fν(x)). where K∗ acts diagonally with the weights wi = deg(fi) ∈ Cl(X) ∼= Z on Kν. point x ∈ X, we mean the group of Weil divisors WDiv(X) modulo those that are principal near x. Moreover, X is the geometric K∗-quotient of bX := X \ {0}, and the quotient map p : bX → X is a universal torsor. By the local divisor class group Cl(X, x) of a Proposition 2.5. For any x = (x1, . . . , xν ) ∈ bX the local divisor class group Cl(X, x) of x := p(x) is finite of order gcd(wi; xi 6= 0). The index of the Picard group Pic(X) in Cl(X) is given by [Cl(X) : Pic(X)] = lcmx∈X ( Cl(X, x)). Suppose that the ideal of X ⊆ Kν is generated by Cl(X)-homogeneous polynomials g1, . . . , gν−d−1 of degree γj := deg(gj). Then one obtains (−KX )d =   νXi=1 wi − ν−d−1Xj=1 d γ1 · · · γν−d−1 w1 · · · wν γj  −KX = νXi=1 wi − ν−d−1Xj=1 γj, for the anticanonical class −KX ∈ Cl(X) ∼= Z. In particular, X is a Fano variety if and only if the following inequality holds ν−d−1Xj=1 γj < νXi=1 wi. Proof. Using [14, Prop. 2.2, Thm. 4.19], we observe that X arises from the bunched ring (R, F, Φ), where R = R(X), F = (f1, . . . , fν) and Φ = {Q≥0}. The descriptions of local class groups, the Picard index and the anticanonical class are then special cases of [14, Prop. 4.7, Cor. 4.9 and Cor. 4.16]. The anticanonical self-intersection number is easily computed in the ambient weighted projective space P(w1, . . . , wν ), use [14, Constr. 3.13, Cor. 4.13]. (cid:3) Remark 2.6. If the ideal of X ⊆ Kν is generated by Cl(X)-homogeneous polynomials g1, . . . , gν−d−1, then [14, Constr. 3.13, Cor. 4.13] show that X is a well formed complete intersection in the weighted projective space P(w1, . . . , wν ) in the sense of [16, Def. 6.9]. We turn back to the case that X comes with a complexity one torus action as at the beginning of this section. We consider the case n0 = . . . = nr = 1, that means that each relation gi,j,k of the Cox ring R(X) depends only on three variables. Then we may write Ti instead of Ti1 and wi instead of wi1, etc.. In this setting, we obtain the following bounds for the numbers of possible varieties X (Fano or not). MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 11 Proposition 2.7. For any pair (d, µ) ∈ Z2 >0 there is, up to deformation, only a finite number of complete d-dimensional varieties with divisor class group Z, Picard index [Cl(X) : Pic(X)] = µ and Cox ring K[T0, . . . , Tr, S1, . . . , Sm] / hαi+1,i+2T li i+2 ; 0 ≤ i ≤ r − 2i. In this situation we have r ≤ ξ(µ) − 1. Moreover, for the weights wi := deg Ti, where 0 ≤ i ≤ r and uk := deg Sk, where 1 ≤ k ≤ m, the exponents li and the degree γ := l0w0 of the relation one has i+1 + αi,i+1T li+2 i + αi+2,iT li+1 l0 · · · lr γ, l0 · · · lr µ, wi ≤ µξ(µ)−1, uk ≤ µ. Proof. Consider the total coordinate space X ⊆ Kr+1+n and the universal tor- sor p : bX → X as discussed before. For each 0 ≤ i ≤ r fix a point x(i) = (x0, . . . , xr, 0, . . . , 0) in bX such that xi = 0 and xj 6= 0 for j 6= i hold. Then, denoting x(i) := p(x(i)), we obtain gcd(wj ; j 6= i) = Cl(X, x(i)) µ. Consider i, j with j 6= i. Since all relations are homogeneous of the same degree, we have liwi = ljwj. Moreover, by the admissibility condition, li and lj are coprime. We conclude liwj for all j 6= i and hence li gcd(wj ; j 6= i). This implies l0 · · · lr l0w0 = γ, l0 · · · lr µ. We turn to the bounds for the wi, and first verify w0 ≤ µr. Using the relation liwi = l0w0, we obtain for every li a presentation li = l0 · w0 · · · wi−1 w1 · · · wi = ηi · gcd(w0, . . . , wi−1) gcd(w0, . . . , wi) with suitable integers 1 ≤ ηi ≤ µ. In particular, the very last fraction is bounded by µ. This gives the desired estimate: w0 = w0 gcd(w0, w1) · gcd(w0, w1) gcd(w0, w1, w2) · · · gcd(w0, . . . , wr−2) gcd(w0, . . . , wr−1) · gcd(w0, . . . , wr−1) ≤ µr. Similarly, we obtain wi ≤ µr for 1 ≤ i ≤ r. Then we only have to show that r + 1 is bounded by ξ(µ), but this follows immediately from the fact that l0, . . . , lr are pairwise coprime. Finally, to estimate the uk, consider the points x(k) ∈ bX having the (r + k)-th coordinate one and all others zero. Set x(k) := p(x(k)). Then Cl(X, x(k)) is of order uk, which implies uk ≤ µ. (cid:3) Lemma 2.8. Consider the ring K[Tij; 0 ≤ i ≤ 2, 1 ≤ j ≤ ni][S1, . . . , Sk]/hgi where n0 ≥ n1 ≥ n2 ≥ 1 holds. Suppose that g is homogeneous with respect to a Z-grading of K[Tij, Sk] given by deg Tij = wij ∈ Z>0 and deg Sk = uk ∈ Z>0, and assume deg g < 2Xi=0 niXj=1 wij + mXi=1 ui. Let µ ∈ Z>1, assume wij ≤ µ whenever ni > 1, 1 ≤ j ≤ ni and uk ≤ µ for 1 ≤ k ≤ m and set d := n0 + n1 + n2 + m − 2. Depending on the shape of g, one obtains the following bounds. (i) Suppose that g = η0T l01 21 with n0 > 1 and coeffi- cients ηi ∈ K∗ holds, we have l11 ≥ l21 ≥ 2 and l11, l21 are coprime. Then, one has 01 · · · T l0n0 0n0 + η1T l11 11 + η2T l21 w11, l21 < 2dµ, w21, l11 < 3dµ, deg g < 6dµ. 12 J. HAUSEN, E. HERPPICH, AND H. S USS (ii) Suppose that g = η0T l01 01 · · · T and coefficients ηi ∈ K∗ holds and we have l21 ≥ 2. Then one has l0n0 0n0 + η1T l11 11 · · · T l1n1 1n1 + η2T l21 21 with n1 > 1 w21 < (d + 1)µ, deg g < 2(d + 1)µ. Proof. We prove (i). Set for short c := (n0 + m)µ = dµ. Then, using homogeneity of g and the assumed inequality, we obtain l11w11 = l21w21 = deg g < 2Xi=0 niXj=1 wij + mXi=1 ui ≤ c + w11 + w21. Since l11 and l21 are coprime, we have l11 > l21 ≥ 2. Plugging this into the above inequalities, we arrive at 2w11 < c + w21 and w21 < c + w11. We conclude w11 < 2c and w21 < 3c. Moreover, l11w11 = l21w21 and gcd(l11, l21) = 1 imply l11w21 and l21w11. This shows l11 < 3c and l21 < 2c. Finally, we obtain deg g < c + w11 + w21 < 6c. We prove (ii). Here we set c := (n0 + n1 + m)µ = (d + 1)µ. Then the assumed inequality gives l21w21 = deg g < 1Xi=0 niXj=1 wij + mXi=1 ui + w21 ≤ c + w21. Since we assumed l21 ≥ 2, we can conclude w21 < c. This in turn gives us deg g < 2c for the degree of the relation. (cid:3) Proof of Theorem 2.1. As before, we denote by X ⊆ Kn+m the total coordinate We first consider the case that X is a toric variety. Then the Cox ring is a polynomial ring, R(X) = K[S1, . . . , Sm]. For each 1 ≤ k ≤ m, consider the point space and by p : bX → X the universal torsor. x(k) ∈ bX having the k-th coordinate one and all others zero and set x(k) := p(x(k)). Then, by Proposition 2.5, the local class group Cl(X, x(k)) is of order uk where uk := deg Sk. This implies uk ≤ µ for 1 ≤ k ≤ m and settles Assertion (i). Now we treat the non-toric case, which means r ≥ 2. Note that we have n ≥ 3. The case n0 = 1 is done in Proposition 2.7. So, we are left with n0 > 1. For ij-coordinate Tij equal to one and all others equal to zero, and thus we have the point x(i, j) := p(x(i, j)) ∈ X. Moreover, for every 1 ≤ k ≤ m, we have the every i with ni > 1 and every 1 ≤ j ≤ ni, there is the point x(i, j) ∈ bX with point x(k) ∈ bX having the k-coordinate Sk equal to one and all others zero; we set x(k) := p(x(k)). Proposition 2.5 provides the bounds wij = deg Tij = Cl(X, x(i, j)) ≤ µ for ni > 1, 1 ≤ j ≤ ni, uk = deg Sk = Cl(X, x(k)) ≤ µ for 1 ≤ k ≤ m. Let 0 ≤ s ≤ r be the maximal number with ns > 1. Then gs−2,s−1,s is the last polynomial such that each of its three monomials depends on more than one variable. For any t ≥ s, we have the "cut ring" Rt := K[Tij; 0 ≤ i ≤ t, 1 ≤ j ≤ ni][S1, . . . , Sm] / hgi,i+1,i+2; 0 ≤ i ≤ t − 2i MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 13 where the relations gi,i+1,i+2 depend on only three variables as soon as i > s holds. For the degree γ of the relations we have (r − 1)γ = (t − 1)γ + (r − t)γ = (t − 1)γ + lt+1,1wt+1,1 + . . . + lr1wr1 < = rXi=0 tXi=0 niXj=1 niXj=1 wij + mXi=1 ui wij + wt+1,1 + . . . + wr1 + mXi=1 ui. Since li1wi1 > wi1 holds in particular for t + 1 ≤ i ≤ r, we derive from this the inequality γ < 1 t − 1  tXi=0  niXj=1 wij + mXi=1 ui  . To obtain the bounds in Assertions (iii) and (iv), we consider the cut ring Rt with t = 2 and apply Lemma 2.8; note that we have d = n0 + n1 + n2 + m − 2 for the dimension d = dim(X) and that l22 ≥ 0 is due to the fact that X is non-toric. The bounds wij, l0j < 6dµ in Assertion (iii) follow from lij wij = γ < 6dµ and li1 < 2dµ follows from li1 w21 for 3 ≤ i ≤ r. Moreover, li1 w11 for 2 ≤ i ≤ r implies l11 · · · lr1 γ = l11w11. Similarly wij , lij < 2(d + 1)µ in Assertion (iv) follow from lijwij = γ < 2(d + 1)dµ and l21 · · · lr1 γ = l21w21 follows from li1 w21 for 3 ≤ i ≤ r. The bounds on r in (iii) in (iv) are as well consequences of the admissibility condition. To obtain the bounds in Assertion (v), we consider the cut ring Rt with t = s. Using ni = 1 for i ≥ t + 1, we can estimate the degree of the relation as follows: γ ≤ (n0 + . . . + nt + m)µ t − 1 = (d + t)µ t − 1 ≤ (d + 2)µ. Since we have wij lij ≤ deg g0 for any 0 ≤ i ≤ r and any 1 ≤ j ≤ ni, we see that all wij and lij are bounded by (d + 2)µ. As before, ls+1,1 · · · lr1 γ is a consequence of li1 γ for i = s + 2, . . . , r and also the bound on r follows from the admissibility condition. Finally, we have to express the Picard index µ in terms of the weights wij and uk as claimed in the Assertions. This is a direct application of the formula of Proposition 2.5. Observe that it suffices to work with the p-images of the following Tij equal to one and all others equal to zero, for every 0 ≤ i ≤ r with ni = 1 points: For every 0 ≤ i ≤ r with ni > 1 take a point x(i, j) ∈ bX with ij-coordinate whenever ni = 1 take x(i, j) ∈ bX with ij-coordinate Tij equal to zero, all other Tst point x(k) ∈ bX having the k-coordinate Sk equal to one and all others zero. equal to one and coordinates Sk equal to zero, and, for every 1 ≤ k ≤ m, take a (cid:3) We conclude the section with discussing some aspects of the not necessarily Fano varieties of Proposition 2.7. Recall that we considered admissible triples (A, n, L) with n0 = . . . = nr = 1 and thus rings R of the form i + αi+2,iT li+1 K[T0, . . . , Tr, S1, . . . , Sm] / hαi+1,i+2T li i+1 + αi,i+1T li+2 i+2 ; 0 ≤ i ≤ r − 2i. Proposition 2.9. Suppose that the ring R as above is the Cox ring of a non-toric variety X with Cl(X) = Z. Then we have m ≥ 1 and µ := [Cl(X) : Pic(X)] ≥ 30. Moreover, if X is a surface, then we have m = 1 and wi = l−1 l0 · · · lr. i 14 J. HAUSEN, E. HERPPICH, AND H. S USS Proof. The homogeneity condition liwi = ljwj together with the admissibility con- dition gcd(li, lj) = 1 for 0 ≤ i 6= j ≤ r gives us li gcd(wj ; j 6= i). Moreover, by Proposition 1.11, every set of m + r weights wi has to generate the class group Z, so they must have greatest common divisor one. Since X is non-toric, li ≥ 2 holds and we obtain m ≥ 1. To proceed, we infer l0 · · · lr µ and l0 · · · lr deg gijk from Proposition 2.5. As a consequence, the minimal value for µ and deg gijk is obviously 2 · 3 · 5 = 30. what really can be received as the following example shows. Note that if X is a surface we have m = 1 and gcd(wi; 0 ≤ i ≤ r) = 1. Thus, liwi = ljwj gives us deg gijk = l0 · · · lr and wi = l−1 (cid:3) l0 · · · lr. i The bound [Cl(X) : Pic(X)] ≥ 30 given in the above proposition is even sharp; the surface discussed below realizes it. Example 2.10. Consider X with R(X) = K[T0, T1, T2, T3]/hgi with g = T 2 T 5 2 and the grading 0 + T 3 1 + deg T0 = 15, deg T1 = 10, deg T2 = 6, deg T3 = 1. Then we have gcd(15, 10) = 5, gcd(15, 6) = 3 and gcd(10, 6) = 2 and therefore [Cl(X) : Pic(X)] = 30. Further X is Fano because of deg g = 30 < 32 = deg T0 + . . . + deg T3. Let us have a look at the geometric meaning of the condition n0 = . . . = nr = 1. For a variety X with an action of a torus T , we denote by X0 ⊆ X the union of all orbits with at most finite isotropy. Then there is a possibly non-separated orbit space X0/T ; we call it the maximal orbit space. From [15], we infer that n0 = . . . = nr = 1 holds if and only if X0/T is separated. Combining this with Propositions 2.7 and 2.9 gives the following. Corollary 2.11. For any pair (d, µ) ∈ Z2 >0 there is, up to deformation, only a finite number of d-dimensional complete varieties X with a complexity one torus action having divisor class group Z, Picard index [Cl(X) : Pic(X)] = µ and maximal orbit space P1 and for each of these varieties the complement X \ X0 contains divisors. Finally, we present a couple of examples showing that there are also non-Fano va- rieties with a complexity one torus action having divisor class group Z and maximal orbit space P1. Example 2.12. Consider X with R(X) = K[T0, T1, T2, T3]/hgi with g = T 2 T 7 2 and the grading 0 + T 3 1 + deg T0 = 21, deg T1 = 14, deg T2 = 6, deg T3 = 1. Then we have gcd(21, 14) = 7, gcd(21, 6) = 3 and gcd(14, 6) = 2 and therefore [Cl(X) : Pic(X)] = 42. Moreover, X is not Fano, because its canonical class KX is trivial KX = deg g − deg T0 − . . . − deg T3 = 0. Example 2.13. Consider X with R(X) = K[T0, T1, T2, T3]/hgi with g = T 2 T 11 2 and the grading 0 + T 3 1 + deg T0 = 33, deg T1 = 22, deg T2 = 6, deg T3 = 1. Then we have gcd(22, 33) = 11, gcd(33, 6) = 3 and gcd(22, 6) = 2 and therefore [Cl(X) : Pic(X)] = 66. The canonical class KX of X is even ample: KX = deg g − deg T0 − . . . − deg T3 = 4. The following example shows that the Fano assumption is essential for the finite- ness results in Theorem 2.1. MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 15 Remark 2.14. For any pair p, q of coprime positive integers, we obtain a locally factorial K∗-surface X(p, q) with Cl(X) = Z and Cox ring R(X(p, q)) = K[T01, T02, T11, T21] / hgi, g = T01T pq−1 02 + T q 11 + T p 21; the Cl(X)-grading is given by deg T01 = deg T02 = 1, deg T11 = p and deg T21 = q. Note that deg g = pq holds and for p, q ≥ 3, the canonical class KX satisfies KX = deg g − deg T01 − deg T02 − deg T11 − deg T21 = pq − 2 − p − q ≥ 0. 3. Classification results In this section, we give classification results for Fano varieties X with Cl(X) ∼= Z that come with a complexity one torus action; note that they are necessarily ratio- nal. The procedure to obtain classification lists for prescribed dimension d = dim X and Picard index µ = [Cl(X) : Pic(X)] is always the following. By Theorem 1.9, we know that their Cox rings are of the form R(X) ∼= R(A, n, L)[S1, . . . , Sm] with admissible triples (A, n, L). Note that for the family A = (a0, . . . , ar) of points ai ∈ K2, we may assume a0 = (1, 0), a1 = (1, 1), a2 = (0, 1). The bounds on the input data of (A, n, L) provided by Theorem 2.1 as well as the criteria of Propositions 1.11 and 2.5 allow us to generate all the possible Cox rings R(X) of the Fano varieties X in question for fixed dimension d and Picard index µ. Note that X can be reconstructed from R(X) = R(A, n, L)[S1, . . . , Sn] as the homogeneous spectrum with respect to the Cl(X)-grading. Thus X is classified by its Cox ring R(X). In the following tables, we present the Cox rings as K[T1, . . . , Ts] modulo relations and fix the Z-gradings by giving the weight vector (w1, . . . , ws), where wi := deg Ti. The first classification result concerns surfaces. Theorem 3.1. Let X be a non-toric Fano surface with an effective K∗-action such that Cl(X) = Z and [Cl(X) : Pic(X)] ≤ 6 hold. Then its Cox ring is precisely one of the following. [Cl(X) : Pic(X)] = 1 No. 1 No. 2 No. 3 4 5 R(X) K[T1, . . . , T4]/hT1T 5 2 + T 3 3 + T 2 4 i [Cl(X) : Pic(X)] = 2 R(X) K[T1, . . . , T4]/hT 4 1 T2 + T 3 3 + T 2 4 i [Cl(X) : Pic(X)] = 3 R(X) K[T1, . . . , T4]/hT 3 K[T1, . . . , T4]/hT1T 3 K[T1, . . . , T4]/hT 7 1 T2 + T 3 2 + T 5 1 T2 + T 5 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i [Cl(X) : Pic(X)] = 4 (w1, . . . , w4) (−KX)2 (1, 1, 2, 3) 1 (w1, . . . , w4) (−KX)2 (1, 2, 2, 3) 2 (w1, . . . , w4) (−KX)2 (1, 3, 2, 3) (1, 3, 2, 5) (1, 3, 2, 5) 3 1/3 1/3 16 J. HAUSEN, E. HERPPICH, AND H. S USS No. 6 7 No. 8 9 10 11 No. 12 13 14 15 R(X) K[T1, . . . , T4]/hT 2 K[T1, . . . , T4]/hT 6 1 T2 + T 3 1 T2 + T 5 3 + T 2 4 i 3 + T 2 4 i (w1, . . . , w4) (−KX)2 (1, 4, 2, 3) (1, 4, 2, 5) 4 1 [Cl(X) : Pic(X)] = 5 R(X) K[T1, . . . , T4]/hT1T2 + T 3 1 T2 + T 5 K[T1, . . . , T4]/hT 5 1 T2 + T 7 K[T1, . . . , T4]/hT 9 K[T1, . . . , T4]/hT 7 1 T2 + T 4 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 3 4 i (w1, . . . , w4) (−KX)2 (1, 5, 2, 3) (1, 5, 2, 5) (1, 5, 2, 7) (1, 5, 3, 4) 5 9/5 1/5 1/5 [Cl(X) : Pic(X)] = 6 R(X) K[T1, . . . , T4]/hT 4 K[T1, . . . , T4]/hT 8 K[T1, . . . , T4]/hT 6 K[T1, . . . , T4]/hT 9 1 T2 + T 5 1 T2 + T 7 1 T2 + T 4 1 T2 + T 3 3 + T 2 4 i 3 + T 2 4 i 3 + T 3 4 i 3 + T 2 4 i (w1, . . . , w4) (−KX)2 (1, 6, 2, 5) (1, 6, 2, 7) (1, 6, 3, 4) (1, 3, 4, 6) 8/3 2/3 2/3 2/3 Proof. As mentioned, Theorems 1.9, 2.1 and Propositions 1.11, 2.5 produce a list of all Cox rings of surfaces with the prescribed data. Doing this computation, we obtain the list of the assertion. Note that none of the Cox rings listed is a polynomial ring and hence none of the resulting surfaces X is a toric variety. To show that different members of the list are not isomorphic to each other, we use the following two facts. Firstly, observe that any two minimal systems of homogeneous generators of the Cox ring have (up to reordering) the same list of degrees, and thus the list of generator degrees is invariant under isomorphism (up to reordering). Secondly, by Construction 1.7, the exponents lij > 1 are precisely the orders of the non-trivial isotropy groups of one-codimensional orbits of the action of the torus T on X. Using both principles and going through the list, we see that different members X cannot be T -equivariantly isomorphic to each other. Since all listed X are non-toric, the effective complexity one torus action on each X corresponds to a maximal torus in the linear algebraic group Aut(X). Any two maximal tori in the automorphism group are conjugate, and thus we can conclude that two members are isomorphic if and only if they are T -equivariantly isomorphic. (cid:3) We remark that in [28, Section 4], log del Pezzo surfaces with an effective K∗- action and Picard number 1 and Gorenstein index less than 4 were classified. The above list contains six such surfaces, namely no. 1-4, 6 and 8; these are exactly the ones where the maximal exponents of the monomials form a platonic triple, i.e., are of the form (1, k, l), (2, 2, k), (2, 3, 3), (2, 3, 4) or (2, 3, 5). The remaining ones, i.e., no. 5, 7, and 9-15 have non-log-terminal and thus non-rational singularities; to check this one may compute the resolutions via resolution of the ambient weighted projective space as in [14, Ex. 7.5]. MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 17 With the same scheme of proof as in the surface case, one establishes the following classification results on Fano threefolds. Theorem 3.2. Let X be a three-dimensional locally factorial non-toric Fano variety with an effective two torus action such that Cl(X) = Z holds. Then its Cox ring is precisely one of the following. No. 1 2 3 4 5 6 7 8 9 2 + T 3 3 + T 3 3 + T 3 R(X) K[T1, . . . , T5] / hT1T 5 3 + T 2 4 i K[T1, . . . , T5] / hT1T2T 4 4 + T 2 5 i K[T1, . . . , T5] / hT1T 2 2 T 3 4 + T 2 5 i K[T1, . . . , T5] / hT1T2 + T3T4 + T 2 5 i 4 + T 3 K[T1, . . . , T5] / hT1T 2 5 i K[T1, . . . , T5] / hT1T 3 4 + T 4 5 i 4 + T 2 K[T1, . . . , T5] / hT1T 3 5 i 4 + T 2 K[T1, . . . , T5] / hT1T 5 5 i K[T1, . . . , T5] / hT1T 5 4 + T 2 5 i 2 + T3T 2 2 + T3T 3 2 + T3T 3 2 + T3T 5 2 + T 3 3 T 3 (w1, . . . , w5) (1, 1, 2, 3, 1) (1, 1, 1, 2, 3) (1, 1, 1, 2, 3) (1, 1, 1, 1, 1) (1, 1, 1, 1, 1) (1, 1, 1, 1, 1) (1, 1, 1, 1, 2) (1, 1, 1, 1, 3) (1, 1, 1, 1, 3) (−KX)3 8 8 8 54 24 4 16 2 2 The singular threefolds listed in this theorem are rational degenerations of smooth Fano threefolds from [18]. The (smooth) general Fano threefolds of the correspond- ing families are non-rational see [12] for no. 1-3, [8] for no. 5, [20] for no. 6, [30, 29] for no. 7 and [19] for no. 8-9. Even if one allows certain mild singularities, one still has non-rationality in some cases, see [13], [9, 25], [10], [6]. Theorem 3.3. Let X be a three-dimensional non-toric Fano variety with an effec- tive two torus action such that Cl(X) = Z and [Cl(X) : Pic(X)] = 2 hold. Then its Cox ring is precisely one of the following. No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 R(X) K[T1, . . . , T5]/hT 4 K[T1, . . . , T5]/hT 4 K[T1, . . . , T5]/hT 8 K[T1, . . . , T5]/hT 4 K[T1, . . . , T5]/hT 4 K[T1, . . . , T5]/hT 8 K[T1, . . . , T5]/hT1T 5 K[T1, . . . , T5]/hT1T 9 K[T1, . . . , T5]/hT 3 1 T 7 K[T1, . . . , T5]/hT1T 11 1 T 7 K[T1, . . . , T5]/hT 5 K[T1, . . . , T5]/hT1T 11 1 T 7 K[T1, . . . , T5]/hT 5 K[T1, . . . , T5]/hT 2 1 T 5 1 T2 + T 3 1 T 3 2 + T 5 1 T2 + T 5 1 T2 + T 3 1 T 3 2 + T 5 1 T2 + T 5 2 + T 3 2 + T 5 2 + T 5 2 + T 3 2 + T 3 2 + T 3 2 + T 3 2 + T 3 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i 3 + T 2 4 i (w1, . . . , w5) (−KX)3 (1, 2, 2, 3, 1) 27/2 (1, 2, 2, 5, 1) (1, 2, 2, 5, 1) (1, 2, 2, 3, 2) (1, 2, 2, 5, 2) (1, 2, 2, 5, 2) 1/2 1/2 16 2 2 (1, 1, 2, 3, 2) 27/2 (1, 1, 2, 5, 2) (1, 1, 2, 5, 2) (1, 1, 4, 6, 1) (1, 1, 4, 6, 1) (1, 1, 4, 6, 2) (1, 1, 4, 6, 2) (1, 2, 4, 6, 1) 1/2 1/2 1/2 1/2 2 2 2 18 J. HAUSEN, E. HERPPICH, AND H. S USS 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 3 + T 2 1 T2 + T 3 K[T1, . . . , T5]/hT 10 4 i 3 + T 2 2 + T 3 K[T1, . . . , T5]/hT1T 2 4 i 3 + T 2 2 + T 5 K[T1, . . . , T5]/hT1T 4 4 i 2 + T 5 3 + T 2 K[T1, . . . , T5]/hT 2 1 T 3 4 i 2 + T3T4 + T 3 K[T1, . . . , T5]/hT1T 2 5 i 4 + T 5 2 + T3T 2 K[T1, . . . , T5]/hT1T 4 5 i 4 + T 5 2 + T3T 2 K[T1, . . . , T5]/hT 2 1 T 3 5 i 3 T4 + T 4 2 + T 2 K[T1, . . . , T5]/hT1T 3 5 i 3 T4 + T 5 2 + T 3 K[T1, . . . , T5]/hT1T 4 5 i 3 T4 + T 5 2 + T 3 K[T1, . . . , T5]/hT 2 1 T 3 5 i 2 + T 2 3 T4 + T 2 K[T1, . . . , T5]/hT1T 3 5 i 4 + T 3 3 T 2 2 + T 2 K[T1, . . . , T5]/hT1T 5 5 i 3 T4 + T 3 2 + T 4 K[T1, . . . , T5]/hT1T 5 5 i 2 + T 4 3 T4 + T 3 K[T1, . . . , T5]/hT 2 1 T 4 5 i 3 T4 + T 2 2 + T 4 K[T1, . . . , T5]/hT1T 5 5 i 3 T4 + T 2 2 + T 4 1 T 3 K[T1, . . . , T5]/hT 3 5 i 2 + T 2 3 T 3 4 + T 2 K[T1, . . . , T5]/hT1T 7 5 i 4 + T 2 3 T 3 2 + T 2 1 T 5 K[T1, . . . , T5]/hT 3 5 i 3 T4 + T 2 2 + T 6 K[T1, . . . , T5]/hT1T 7 5 i 2 + T 6 3 T4 + T 2 K[T1, . . . , T5]/hT 3 1 T 5 5 i 2 + T3T4 + T 4 K[T1, . . . , T5]/hT1T 3 5 i 4 + T 6 2 + T3T 2 K[T1, . . . , T5]/hT1T 5 5 i 2 + T3T4 + T 2 K[T1, . . . , T5]/hT1T 3 5 i 4 + T 3 2 + T3T 2 K[T1, . . . , T5]/hT1T 5 5 i 4 + T 3 2 + T3T 2 1 T 4 K[T1, . . . , T5]/hT 2 5 i 2 + T3T 2 4 + T 2 K[T1, . . . , T5]/hT 3 1 T 3 5 i 4 + T 2 2 + T3T 3 K[T1, . . . , T5]/hT 3 1 T 5 5 i 4 + T 2 2 + T3T 2 K[T1, . . . , T5]/hT1T 5 5 i 2 + T3T 3 4 + T 2 K[T1, . . . , T5]/hT1T 7 5 i 4 + T 2 2 + T3T 4 K[T1, . . . , T5]/hT1T 9 5 i 4 + T 2 2 + T 2 3 T 3 K[T1, . . . , T5]/hT1T 9 5 i 2 + T3T 4 4 + T 2 K[T1, . . . , T5]/hT 3 1 T 7 5 i 4 + T 2 2 + T 2 3 T 3 1 T 7 K[T1, . . . , T5]/hT 3 5 i 4 + T 2 2 + T3T 4 1 T 5 K[T1, . . . , T5]/hT 5 5 i 2 + T 2 3 T 3 4 + T 2 K[T1, . . . , T5]/hT 5 1 T 5 5 i K[T1, . . . , T5]/hT1T2 + T3T4 + T 3 5 i 3 T4 + T 4 K[T1, . . . , T5]/hT 2 5 i K[T1, . . . , T5]/hT1T 2 4 + T 5 5 i 2 + T3T 2 1 T2 + T 2 (1, 2, 4, 6, 1) (2, 2, 2, 3, 1) (2, 2, 2, 5, 1) (2, 2, 2, 5, 1) 2 16 2 2 (1, 1, 1, 2, 1) 81/2 (1, 1, 1, 2, 1) (1, 1, 1, 2, 1) (1, 1, 1, 2, 1) (1, 1, 1, 2, 1) (1, 1, 1, 2, 1) (1, 1, 1, 2, 2) (1, 1, 1, 2, 2) (1, 1, 1, 2, 2) (1, 1, 1, 2, 2) (1, 1, 1, 2, 3) (1, 1, 1, 2, 3) (1, 1, 1, 2, 4) (1, 1, 1, 2, 4) (1, 1, 1, 2, 4) (1, 1, 1, 2, 4) (1, 1, 2, 2, 1) (1, 1, 2, 2, 1) (1, 1, 2, 2, 2) (1, 1, 2, 2, 2) (1, 1, 2, 2, 2) 5/2 5/2 16 5/2 5/2 27 3/2 3/2 3/2 8 8 1 1 1 1 27 3/2 16 6 6 (1, 1, 2, 2, 2) 27/2 (1, 1, 2, 2, 2) (1, 1, 2, 2, 3) (1, 1, 2, 2, 4) (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) (1, 2, 1, 2, 1) (1, 2, 1, 2, 1) (1, 2, 1, 2, 1) 32 4 32 1/2 1/2 1/2 1/2 1/2 1/2 48 27 10 MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 19 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 3 T4 + T 5 2 + T 3 K[T1, . . . , T5]/hT1T 2 5 i 3 T4 + T 5 1 T2 + T 3 K[T1, . . . , T5]/hT 3 5 i 3 T4 + T 6 1 T2 + T 4 K[T1, . . . , T5]/hT 4 5 i 3 T4 + T 2 1 T2 + T 2 K[T1, . . . , T5]/hT 2 5 i 3 T4 + T 3 1 T 2 2 + T 4 K[T1, . . . , T5]/hT 2 5 i 3 T4 + T 3 1 T2 + T 4 K[T1, . . . , T5]/hT 4 5 i 3 T4 + T 2 1 T2 + T 4 K[T1, . . . , T5]/hT 4 5 i 3 T 3 4 + T 2 1 T 3 2 + T 2 K[T1, . . . , T5]/hT 2 5 i 3 T4 + T 2 1 T 3 2 + T 6 K[T1, . . . , T5]/hT 2 5 i 3 T4 + T 2 1 T2 + T 6 K[T1, . . . , T5]/hT 6 5 i 3 T 3 4 + T 2 1 T 3 2 + T 4 K[T1, . . . , T5]/hT 4 5 i 4 + T 2 3 T 3 1 T2 + T 4 K[T1, . . . , T5]/hT 8 5 i 3 T4 + T 2 1 T2 + T 8 K[T1, . . . , T5]/hT 8 5 i 1 T2 + T3T4 + T 4 K[T1, . . . , T5]/hT 2 5 i 4 + T 6 1 T2 + T3T 2 K[T1, . . . , T5]/hT 4 5 i 4 + T 2 1 T2 + T3T 2 K[T1, . . . , T5]/hT 4 5 i 1 T 3 2 + T3T 4 4 + T 2 K[T1, . . . , T5]/hT 4 5 i 4 + T 2 3 T 3 2 + T 2 1 T 3 K[T1, . . . , T5]/hT 4 5 i 4 + T 2 1 T2 + T3T 4 K[T1, . . . , T5]/hT 8 5 i 1 T2 + T 2 3 T 3 4 + T 2 K[T1, . . . , T5]/hT 8 5 i 4 + T 2 3 + T 3 K[T1, . . . , T5]/hT1T2T 10 5 i 4 + T 2 3 + T 3 2 T 9 K[T1, . . . , T5]/hT1T 2 5 i 4 + T 2 3 + T 3 2 T 8 K[T1, . . . , T5]/hT1T 3 5 i 4 + T 2 3 + T 3 2 T 7 K[T1, . . . , T5]/hT1T 4 5 i 4 + T 2 3 + T 3 2 T 6 K[T1, . . . , T5]/hT1T 5 5 i 3 + T 3 4 + T 2 K[T1, . . . , T5]/hT 2 1 T 3 2 T 7 5 i 4 + T 2 3 + T 3 2 T 5 1 T 5 K[T1, . . . , T5]/hT 2 5 i 4 + T 2 3 + T 3 K[T1, . . . , T5]/hT 3 1 T 4 2 T 5 5 i 3 + T 3 4 + T 2 K[T1, . . . , T5]/hT1T2T 2 5 i 4 + T 2 2 T3 + T 3 K[T1, . . . , T5]/hT1T 3 5 i 4 + T 2 2 T3 + T 3 K[T1, . . . , T5]/hT 2 1 T 2 5 i 3 + T 5 4 + T 2 K[T1, . . . , T5]/hT1T2T 4 5 i 4 + T 2 3 + T 5 2 T 3 K[T1, . . . , T5]/hT1T 3 5 i 4 + T 2 2 T 2 K[T1, . . . , T5]/hT1T 5 3 + T 5 5 i 2 T3 + T 5 K[T1, . . . , T5]/hT1T 7 4 + T 2 5 i 4 + T 2 3 + T 5 2 T 3 1 T 2 K[T1, . . . , T5]/hT 2 5 i 4 + T 2 2 T3 + T 5 1 T 6 K[T1, . . . , T5]/hT 2 5 i K[T1, . . . , T5]/hT 3 1 T 3 2 T 2 3 + T 5 4 + T 2 5 i (1, 2, 1, 2, 1) (1, 2, 1, 2, 1) (1, 2, 1, 2, 1) (1, 2, 1, 2, 2) (1, 2, 1, 2, 2) (1, 2, 1, 2, 2) 10 10 3/2 32 6 6 (1, 2, 1, 2, 3) 27/2 (1, 2, 1, 2, 4) (1, 2, 1, 2, 4) (1, 2, 1, 2, 4) (1, 2, 1, 2, 5) (1, 2, 1, 2, 5) (1, 2, 1, 2, 5) (1, 2, 2, 2, 1) (1, 2, 2, 2, 1) (1, 2, 2, 2, 3) (1, 2, 2, 2, 5) (1, 2, 2, 2, 5) (1, 2, 2, 2, 5) (1, 2, 2, 2, 5) (1, 1, 1, 4, 6) (1, 1, 1, 4, 6) (1, 1, 1, 4, 6) (1, 1, 1, 4, 6) (1, 1, 1, 4, 6) (1, 1, 1, 4, 6) (1, 1, 1, 4, 6) (1, 1, 1, 4, 6) (1, 1, 2, 2, 3) (1, 1, 2, 2, 3) (1, 1, 2, 2, 3) (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) 4 4 4 1/2 1/2 1/2 32 6 16 2 2 2 2 1/2 1/2 1/2 1/2 1/2 1/2 1/2 1/2 27/2 27/2 27/2 1/2 1/2 1/2 1/2 1/2 1/2 1/2 20 J. HAUSEN, E. HERPPICH, AND H. S USS 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 1 T 5 1 T 4 2 T3 + T 5 K[T1, . . . , T5]/hT 3 2 T3 + T 5 K[T1, . . . , T5]/hT 4 3 + T 3 K[T1, . . . , T5]/hT1T2T 5 3 + T 3 2 T 4 K[T1, . . . , T5]/hT1T 3 3 + T 3 K[T1, . . . , T5]/hT1T 5 2 T 3 2 T 2 3 + T 3 K[T1, . . . , T5]/hT1T 7 2 T3 + T 3 K[T1, . . . , T5]/hT1T 9 K[T1, . . . , T5]/hT 2 1 T 4 2 T 3 3 + T 3 2 T3 + T 3 1 T 8 K[T1, . . . , T5]/hT 2 2 T 2 1 T 5 K[T1, . . . , T5]/hT 3 3 + T 3 K[T1, . . . , T5]/hT 3 1 T 7 2 T3 + T 3 2 T3 + T 3 1 T 6 K[T1, . . . , T5]/hT 4 2 T3 + T 3 1 T 5 K[T1, . . . , T5]/hT 5 K[T1, . . . , T5]/hT 2 1 T2T3 + T 3 3 + T 5 1 T2T 3 K[T1, . . . , T5]/hT 2 3 + T 5 1 T2T 2 K[T1, . . . , T5]/hT 4 K[T1, . . . , T5]/hT 6 1 T2T3 + T 5 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i (1, 1, 2, 2, 5) (1, 1, 2, 2, 5) (1, 1, 2, 4, 6) (1, 1, 2, 4, 6) (1, 1, 2, 4, 6) (1, 1, 2, 4, 6) (1, 1, 2, 4, 6) (1, 1, 2, 4, 6) (1, 1, 2, 4, 6) (1, 1, 2, 4, 6) (1, 1, 2, 4, 6) (1, 1, 2, 4, 6) (1, 1, 2, 4, 6) (1, 2, 2, 2, 3) (1, 2, 2, 2, 5) (1, 2, 2, 2, 5) (1, 2, 2, 2, 5) 1/2 1/2 2 2 2 2 2 2 2 2 2 2 2 16 2 2 2 The varieties no. 2,3 and 25, 26 are rational degenerations of quasismooth va- rieties from the list in [16]. In [11] the non-rationality of a general (quasismooth) element of the corresponding family was proved. The varieties listed so far might suggest that we always obtain only one relation in the Cox ring. We discuss now some examples, showing that for a Picard index big enough, we need in general more than one relation, where this refers always to a presentation as in Theorem 1.9 (ii). Example 3.4. A Fano K∗-surface X with Cl(X) = Z such that the Cox ring R(X) needs two relations. Consider the Z-graded ring R = K[T01, T02, T11, T21, T31]/hg0, g1i, where the degrees of T01, T02, T11, T21, T31 are 29, 1, 6, 10, 15, respectively, and the relations g0, g1 are given by g0 := T01T02 + T 5 11 + T 3 21, g1 := α23T 5 11 + α31T 3 21 + α12T 2 31 Then R is the Cox ring of a Fano K∗-surface. Note that the Picard index is given by [Cl(X) : Pic(X)] = lcm(29, 1) = 29. Proposition 3.5. Let X be a non-toric Fano surface with an effective K∗-action such that Cl(X) ∼= Z and [Cl(X) : Pic(X)] < 29 hold. Then the Cox ring of X is of the form R(X) ∼= K[T1, . . . , T4]/hT l1 4 i. 1 T l2 2 + T l3 3 + T l4 Proof. The Cox ring R(X) is as in Theorem 1.9, and, in the notation used there, we have n0 + . . . + nr + m = 2 + r. This leaves us with the possibilities n0 = m = 1 and n0 = 2, m = 0. In the first case, Proposition 2.9 tells us that the Picard index of X is at least 30. MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 21 So, consider the case n0 = 2 and m = 0. Then, according to Theorem 1.9, the Cox ring R(X) is K[T01, T02, T1 . . . , Tr] divided by relations 01 T l02 02 + T l1 1 + T l2 2 , g0,1,2 = T l01 i+1 + αi,i+1T li+2 i+2 , where 1 ≤ i ≤ r − 2. We have to show that r = 2 holds. Set µ := [Cl(X) : Pic(X)] and let γ ∈ Z denote the degree of the relations. Then we have γ = wili for 1 ≤ i ≤ r, where wi := deg Ti. With w0i := deg T0i, Proposition 2.5 gives us gi,i+1,i+2 = αi+1,i+2T li i + αi+2,iT li+1 (r − 1)γ < w01 + w02 + w1 + . . . + wr. We claim that w01 and w02 are coprime. Otherwise they had a common prime divisor p. This p divides γ = liwi. Since l1, . . . , lr are pairwise coprime, p divides at least r − 1 of the weights w1, . . . , wr. This contradicts the Cox ring condition that any r + 1 of the r + 2 weights generate the class group Z. Thus, w01 and w02 are coprime and we obtain µ ≥ lcm(w01, w02) = w01 · w02 ≥ w01 + w02 − 1. Now assume that r ≥ 3 holds. Then we can conclude 2γ < w01 + w02 + w1 + w2 + w3 ≤ µ + 1 + γ(cid:18) 1 l1 + 1 l2 + 1 l3(cid:19) Since the numbers li are pairwise coprime, we obtain l1 ≥ 5, l2 ≥ 3 and l3 ≥ 2. Moreover, liwi = ljwj implies li wj and hence l1l2l3 γ. Thus, we have γ ≥ 30. Plugging this in the above inequality gives µ ≥ γ(cid:18)2 − 1 l1 − 1 l2 − 1 l3(cid:19) − 1 = 29. (cid:3) The Fano assumption is essential in this result; if we omit it, then we may even construct locally factorial surfaces with a Cox ring that needs more then one relation. Example 3.6. A locally factorial K∗-surface X with Cl(X) = Z such that the Cox ring R(X) needs two relations. Consider the Z-graded ring R = K[T01, T02, T11, T21, T31]/hg0, g1i, where the degrees of T01, T02, T11, T21, T31 are 1, 1, 6, 10, 15, respectively, and the relations g0, g1 are given by 02 + T 5 g1 := α23T 5 21 + α12T 2 31 11 + α31T 3 g0 := T 7 11 + T 3 21, 01T 23 Then R is the Cox ring of a non Fano K∗-surface X of Picard index one, i.e, X is locally factorial. For non-toric Fano threefolds X with an effective 2-torus action Cl(X) ∼= Z, the classifications 3.2 and 3.3 show that for Picard indices one and two we only obtain hypersurfaces as Cox rings. The following example shows that this stops at Picard index three. Example 3.7. A Fano threefold X with Cl(X) = Z and a 2-torus action such that the Cox ring R(X) needs two relations. Consider R = K[T01, T02, T11, T12, T21, T31]/hg0, g1i where the degrees of T01, T02, T11, T12, T21, T31 are 1, 1, 3, 3, 2, 3, respectively, and the relations are given by g0 = T 5 01T02 + T11T12 + T 3 21, g1 = α23T11T12 + α31T 3 21 + α12T 2 31. 22 J. HAUSEN, E. HERPPICH, AND H. S USS Then R is the Cox ring of a Fano threefold with a 2-torus action. Note that the Picard index is given by [Cl(X) : Pic(X)] = lcm(1, 1, 3, 3) = 3. Finally, we turn to locally factorial Fano fourfolds. Here we observe more than one relation in the Cox ring even in the locally factorial case. Theorem 3.8. Let X be a four-dimensional locally factorial non-toric Fano variety with an effective three torus action such that Cl(X) = Z holds. Then its Cox ring is precisely one of the following. No. R(X) 2 3 4 5 8 1 6 7 2 + T 3 2 + T 2 2 + T 2 3 + T 3 3 + T 3 3 + T 5 3 + T 5 3 + T 5 3 + T 5 3 + T 5 3 + T 5 K[T1, . . . , T6]/hT1T 5 K[T1, . . . , T6]/hT1T 9 K[T1, . . . , T6]/hT 3 1 T 7 K[T1, . . . , T6]/hT1T2T 4 2 T 3 K[T1, . . . , T6]/hT1T 2 K[T1, . . . , T6]/hT1T2T 8 2 T 7 K[T1, . . . , T6]/hT1T 2 2 T 6 K[T1, . . . , T6]/hT1T 3 K[T1, . . . , T6]/hT1T 4 2 T 5 2 T 5 1 T 3 1 T 3 2 T 4 3 + T 2 4 i 3 + T 5 4 i 3 + T 5 4 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 4 + T 2 5 i 9 4 + T 2 10 K[T1, . . . , T6]/hT 2 5 i 11 K[T1, . . . , T6]/hT 3 4 + T 2 5 i 12 K[T1, . . . , T6]/hT1T2 + T3T4 + T 2 5 i 4 + T 3 2 + T3T 2 13 K[T1, . . . , T6]/hT1T 2 5 i 4 + T 4 2 + T3T 3 14 K[T1, . . . , T6]/hT1T 3 5 i 2 + T3T 4 15 K[T1, . . . , T6]/hT1T 4 4 + T 5 5 i 3 T 3 2 + T 2 4 + T 5 16 K[T1, . . . , T6]/hT1T 4 5 i 4 + T 5 3 T 3 2 + T 2 1 T 3 17 K[T1, . . . , T6]/hT 2 5 i 2 + T3T 3 18 K[T1, . . . , T6]/hT1T 3 4 + T 2 5 i 2 + T3T 5 4 + T 3 19 K[T1, . . . , T6]/hT1T 5 5 i 2 + T 2 3 T 4 4 + T 3 20 K[T1, . . . , T6]/hT1T 5 5 i 2 + T3T 5 21 K[T1, . . . , T6]/hT1T 5 4 + T 2 5 i 2 + T 3 3 T 3 4 + T 2 22 K[T1, . . . , T6]/hT1T 5 5 i 2 + T3T 7 4 + T 2 23 K[T1, . . . , T6]/hT1T 7 5 i 2 + T 3 3 T 5 24 K[T1, . . . , T6]/hT1T 7 4 + T 2 5 i 3 T 5 25 K[T1, . . . , T6]/hT 3 4 + T 2 2 + T 3 1 T 5 5 i 26 K[T1, . . . , T6]/hT1T2T3T 3 5 + T 2 4 + T 3 6 i 27 K[T1, . . . , T6]/hT1T2T 2 3 T 2 4 + T 3 5 + T 2 6 i 5 + T 2 4 + T 5 28 K[T1, . . . , T6]/hT1T2T3T 7 6 i 29 K[T1, . . . , T6]/hT1T2T 2 3 T 6 4 + T 5 5 + T 2 6 i 5 + T 2 4 + T 5 3 T 5 30 K[T1, . . . , T6]/hT1T2T 3 6 i (w1, . . . , w6) (−KX )4 (1, 1, 2, 3, 1, 1) 81 (1, 1, 2, 5, 1, 1) (1, 1, 2, 5, 1, 1) (1, 1, 1, 2, 3, 1) (1, 1, 1, 2, 3, 1) (1, 1, 1, 2, 5, 1) (1, 1, 1, 2, 5, 1) (1, 1, 1, 2, 5, 1) (1, 1, 1, 2, 5, 1) (1, 1, 1, 2, 5, 1) (1, 1, 1, 2, 5, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) 1 1 81 81 1 1 1 1 1 1 512 243 64 5 5 5 (1, 1, 1, 1, 2, 1) 162 (1, 1, 1, 1, 2, 1) (1, 1, 1, 1, 2, 1) (1, 1, 1, 1, 3, 1) (1, 1, 1, 1, 3, 1) (1, 1, 1, 1, 4, 1) (1, 1, 1, 1, 4, 1) (1, 1, 1, 1, 4, 1) (1, 1, 1, 1, 2, 3) (1, 1, 1, 1, 2, 3) (1, 1, 1, 1, 2, 5) (1, 1, 1, 1, 2, 5) (1, 1, 1, 1, 2, 5) 3 3 32 32 2 2 2 81 81 1 1 1 MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 23 31 K[T1, . . . , T6]/hT1T2T 4 4 + T 5 32 K[T1, . . . , T6]/hT1T 2 2 T 2 4 + T 5 2 T 3 33 K[T1, . . . , T6]/hT1T 2 4 + T 5 2 T 3 34 K[T1, . . . , T6]/hT1T 3 4 + T 5 35 K[T1, . . . , T6]/hT 2 1 T 2 2 T 3 4 + T 5 36 K[T1, . . . , T6]/hT1T2T3 + T4T 2 3 + T4T 3 37 K[T1, . . . , T6]/hT1T2T 2 38 K[T1, . . . , T6]/hT1T2T 3 3 + T4T 4 3 + T 2 39 K[T1, . . . , T6]/hT1T2T 3 4 T 3 3 + T4T 4 40 K[T1, . . . , T6]/hT1T 2 2 T 2 41 K[T1, . . . , T6]/hT1T 2 2 T 2 3 + T 2 4 T 3 3 + T4T 3 42 K[T1, . . . , T6]/hT1T2T 2 3 + T4T 5 43 K[T1, . . . , T6]/hT1T2T 4 44 K[T1, . . . , T6]/hT1T2T 4 3 + T 2 4 T 4 3 + T4T 5 2 T 3 45 K[T1, . . . , T6]/hT1T 2 3 + T 2 2 T 3 46 K[T1, . . . , T6]/hT1T 2 4 T 4 47 K[T1, . . . , T6]/hT 2 1 T 2 2 T 2 3 + T4T 5 4 T 3 3 + T 3 2 T 3 48 K[T1, . . . , T6]/hT1T 2 3 + T4T 5 49 K[T1, . . . , T6]/hT1T 2 2 T 3 50 K[T1, . . . , T6]/hT1T2T 4 3 + T 3 4 T 3 3 + T4T 5 51 K[T1, . . . , T6]/hT1T2T 4 3 + T4T 7 52 K[T1, . . . , T6]/hT1T2T 6 4 T 5 3 + T 3 53 K[T1, . . . , T6]/hT1T2T 6 3 + T4T 7 2 T 5 54 K[T1, . . . , T6]/hT1T 2 4 T 5 3 + T 3 2 T 5 55 K[T1, . . . , T6]/hT1T 2 3 + T4T 7 2 T 4 56 K[T1, . . . , T6]/hT1T 3 3 + T 3 2 T 4 57 K[T1, . . . , T6]/hT1T 3 4 T 5 58 K[T1, . . . , T6]/hT 2 1 T 3 2 T 3 3 + T4T 7 4 T 5 3 + T 3 2 T 3 1 T 3 59 K[T1, . . . , T6]/hT 2 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 3 6 i 5 + T 4 6 i 5 + T 5 6 i 5 + T 5 6 i 5 + T 5 6 i 5 + T 5 6 i 5 + T 2 6 i 5 + T 3 6 i 5 + T 3 6 i 5 + T 3 6 i 5 + T 3 6 i 5 + T 3 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 5 + T 2 6 i 60 K[T1, . . . , T6]/hT1T2 + T3T4 + T5T6i 4 + T5T 2 61 K[T1, . . . , T6]/hT1T 2 6 i 4 + T5T 3 62 K[T1, . . . , T6]/hT1T 3 6 i 5 T 2 4 + T 2 63 K[T1, . . . , T6]/hT1T 3 6 i 64 K[T1, . . . , T6]/hT1T 4 4 + T5T 4 6 i 65 K[T1, . . . , T6]/hT1T 4 4 + T 2 5 T 3 6 i 5 T 3 4 + T 2 66 K[T1, . . . , T6]/hT1T 4 6 i 67 K[T1, . . . , T6]/hT 2 1 T 3 4 + T 2 5 T 3 6 i 3 T 4 3 T 5 3 T 4 3 T 3 3 T 3 2 + T3T 2 2 + T3T 3 2 + T3T 3 2 + T3T 4 2 + T3T 4 3 T 3 2 + T 2 2 + T 2 3 T 3 (1, 1, 1, 1, 2, 5) (1, 1, 1, 1, 2, 5) (1, 1, 1, 1, 2, 5) (1, 1, 1, 1, 2, 5) (1, 1, 1, 1, 2, 5) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) 1 1 1 1 1 243 64 5 5 5 5 (1, 1, 1, 1, 1, 2) 162 (1, 1, 1, 1, 1, 2) (1, 1, 1, 1, 1, 2) (1, 1, 1, 1, 1, 2) (1, 1, 1, 1, 1, 2) (1, 1, 1, 1, 1, 2) (1, 1, 1, 1, 1, 3) (1, 1, 1, 1, 1, 3) (1, 1, 1, 1, 1, 3) (1, 1, 1, 1, 1, 3) (1, 1, 1, 1, 1, 4) (1, 1, 1, 1, 1, 4) (1, 1, 1, 1, 1, 4) (1, 1, 1, 1, 1, 4) (1, 1, 1, 1, 1, 4) (1, 1, 1, 1, 1, 4) (1, 1, 1, 1, 1, 4) (1, 1, 1, 1, 1, 4) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) (1, 1, 1, 1, 1, 1) 3 3 3 3 3 32 32 32 32 2 2 2 2 2 2 2 2 512 243 64 64 5 5 5 5 24 J. HAUSEN, E. HERPPICH, AND H. S USS 68 K[T1, . . . , T7]/D T1T2+T3T4+T5T6, 7 E 69 K[T1, . . . , T7]/D T1T 2 7 E 4 +T5T 2 6 , 6 +T 3 αT3T4+T5T6+T 2 4 +T5T 2 2 +T3T 2 αT3T 2 (1, 1, 1, 1, 1, 1, 1) 324 (1, 1, 1, 1, 1, 1, 1) 9 where in the last two rows of the table the parameter α can be any element from K∗ \ {1}. By the result of [26], the singular quintics of this list are rational degenerations of smooth non-rational Fano fourfolds. 4. Geometry of the locally factorial threefolds In this section, we take a closer look at the (factorial) singularities of the Fano varieties X listed in Theorem 3.2. Recall that the discrepancies of a resolution ϕ : eX → X of a singularity are the coefficients of K eX − ϕ∗KX, where KX and K eX are canonical divisors such that K eX − ϕ∗KX is supported on the exceptional locus of ϕ. A resolution is called crepant, if its discrepancies vanish and a singularity is called canonical (terminal), if it admits a resolution with nonnegative (positive) discrepancies. By a relative minimal model we mean a projective morphism eX → X such that eX has at most terminal singularities and its relative canonical divisor is relatively nef. Theorem 4.1. For the nine 3-dimensional Fano varieties listed in Theorem 3.2, we have the following statements. (i) No. 4 is a smooth quadric in P4. (ii) Nos. 1,3,5,7 and 9 are singular with only canonical singularities and all admit a crepant resolution. (iii) Nos. 6 and 8 are singular with non-canonical singularities but admit a smooth relative minimal model. (iv) No. 2 is singular with only canonical singularities, one of them of type cA1, and admits only a singular relative minimal model. The Cox ring of the relative minimal model eX as well as the the Fano degree of X itself are given in the following table. R(eX) 8 T9 + T 2 10T11i 7 T 2 7 + T 2 8 i (−KX)3 8 8 8 54 24 4 16 2 2 No. 1 2 3 4 5 6 7 8 9 5 + T6T 2 7 i 5 T 5 6 + T 3 6 T 3 4 T 4 4 + T5T 2 3 T 3 3 T 4 3 + T4T 3 K[T1, . . . , T14]/(T1T2T 2 K[T1, . . . , T9]/hT1T2T 2 K[T1, . . . , T8]/hT1T 2 2 T 3 K[T1, . . . , T5]/hT1T2 + T3T4 + T 2 5 i 4 + T 3 K[T1, . . . , T6]/hT1T 2 5 T6i K[T1, . . . , T6]/hT1T 3 4 + T 4 5 T6i 4 + T 2 K[T1, . . . , T7]/hT1T 3 5 T6i 4 + T 2 K[T1, . . . , T7]/hT1T 5 5 T6i 4 T 2 5 T 3 6 T 3 24T 3 19···T 2 2 + T3T 2 2 + T3T 3 2 + T3T 3 2 + T3T 5 K[T1, . . . , T46]/D + T11···T18T 2 T1T2T3T 2 7 T 4 25T 3 10 + 9 T 5 8 T 4 26 + T27···T32T 2 33E MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 25 For the proof, it is convenient to work in the language of polyhedral divisors introduced in [1] and [2]. As we are interested in rational varieties with a complexity one torus action, we only have to consider polyhedral divisors on the projective line Y = P1. This considerably simplifies the general definitions and allows us to give a short summary. In the sequel, N ∼= Zn denotes a lattice and M = Hom(N, Z) its dual. For the associated rational vector spaces we write NQ and MQ. A polyhedral divisor on the projective line Y := P1 is a formal sum D = Xy∈Y Dy · y, where the coefficients Dy ⊆ NQ are (possibly empty) convex polyhedra all sharing the same tail (i.e. recession) cone DY = σ ⊆ NQ, and only finitely many Dy differ from σ. The locus of D is the open subset Y (D) ⊆ Y obtained by removing all points y ⊆ Y with Dy = ∅. For every u ∈ σ∨ ∩ M we have the evaluation D(u) := Xy∈Y min v∈Dy hu, vi·y, which is a usual rational divisor on Y (D). We call the polyhedral divisor D on Y proper if deg D ( σ holds, where the polyhedral degree is defined by deg D := Xy∈Y Dy. Every proper polyhedral divisor D on Y defines a normal affine variety X(D) of dimension rk (N ) + 1 coming with an effective action of the torus T = Spec K[M ]: set X(D) := Spec A(D), where A(D) := Mu∈σ∨∩M Γ(Y (D), O(D(u))) ⊆ Mu∈M K(Y ) · χu. A divisorial fan, is a finite set Ξ of polyhedral divisors D on Y , all having their polyhedral coefficients Dy in the same NQ and fulfilling certain compatibility conditions, see [2]. In particular, for every point y ∈ Y , the slice Ξy := {Dy; D ∈ Ξ} must be a polyhedral subdivision. The tail fan is the set ΞY of the tail cones DY of the D ∈ Ξ; it is a fan in the usual sense. Given a divisorial fan Ξ, the affine varieties X(D), where D ∈ Ξ, glue equivariantly together to a normal variety X(Ξ), and we obtain every rational normal variety with a complexity one torus action this way. Smoothness of X = X(Ξ) is checked locally. For a proper polyhedral divisor D on Y , we infer the following from [28, Theorem 3.3]. If Y (D) is affine, then X(D) is smooth if and only if cone({1} × Dy) ⊆ Q × NQ, the convex, polyhedral cone generated by {1} × Dy, is regular for every y ∈ Y (D). If Y (D) = Y holds, then X(D) is smooth if and only if there are y, z ∈ Y such that D = Dyy + Dzz holds and cone({1} × Dy) + cone({−1} × Dz) is a regular cone in Q × NQ. Similarly to toric geometry, singularities of X(D) are resolved by means of subdividing D. This means to consider divisorial fans Ξ such that for any y ∈ Y , the slice Ξy is a subdivision of Dy. Such a Ξ defines a dominant morphism X(Ξ) → X(D) and a slight generalization of [2, Thm. 7.5.] yields that this morphism is proper. Proposition 4.2. The 3-dimensional Fano varieties No. 1-8 listed in Theorem 3.2 and their relative minimal models arise from divisorial fans having the following slices and tail cones. 26 1 (−1,1) J. HAUSEN, E. HERPPICH, AND H. S USS (-1,6) (0,1) − 4 5 − 3 4− 2 3− 1 2 2 (−1,1) (− 1 4 , 1 2 ) (0, 1 2 ) (− 1 2 ,0) (− 2 3 ,0) 1 3 1 2 1 1 2 1 (-1,1) (-4,3) (1,− 1 2 ) (1,0) (1,0) (2,-1) (1,-1) (2,-2.6) (1,1) (0,-1) (0,1) (0,1) 2 , 1 ( 1 2 ) ( 2 3 , 1 3 ) (− 1 2 ,0) (0,− 1 3 ) (0,−1) (0,− 1 2 ) (0,− 1 3 ) 3 4 (−1,0) 5 (− 1 2 ,0) 6 (− 1 3 ,0) 7 3 , 1 ( 2 3 ) (− 1 2 ,0) (0,− 1 3 ) (-1,0) (1,0) (-3,-2) (0,-1) (-1,1) (1,1) ( 1 2 , 1 2 ) (-1,-1) (1,-1) (-1,2) (1,1) 3 , 1 ( 1 3 ) (-1,0) (-1,-1) (0,-1) (-1,3) (2,-1) (1,1) ( 1 4 , 1 4 ) (-1,0) (3,-1) (2,1) (1,0) (-1,-1) (0,-1) (1,2) (-1,0) (-3,-2) (0,-1) b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 27 8 5 , 1 ( 3 5 ) (− 1 2 ,0) (0,− 1 5 ) (-1,0) (-5,-2) (1,2) (3,1) (1,0) (0,-1) The above table should be interpreted as follows. The first three pictures in each row are the slices at 0, 1 and ∞ and the last one is the tail fan. The divisorial fan of the fano variety itself is given by the solid polyhedra in the pictures. Here, all polyhedra of the same gray scale belong to the same polyhedral divisor. The subdivisions for the relative minimal models are sketched with dashed lines. In general, polyhedra with the same tail cone belong all to a unique polyhedral divisor with complete locus. For the white cones inside the tail fan we have another rule: for every polyhedron ∆ ∈ Ξy with the given white cone as its tail there is a polyhedral divisor ∆ · y + ∅ · z ∈ Ξ, with z ∈ {0, 1, ∞} \ {y}. Here, different choices of z lead to isomorphic varieties, only the affine covering given by the X(D) changes. In order to prove Theorem 4.1, we also have to understand invariant divisors on X = X(Ξ) in terms of Ξ, see [15, Prop. 4.11 and 4.12] for details. A first type of invariant prime divisors, is in bijection Dy,v ↔ (y, v) with the vertices (y, v), where y ∈ Y and v ∈ Ξy is of dimension zero. The order of the generic isotropy group along Dy,v equals the minimal positive integer µ(v) with µ(v)v ∈ N . A second type of invariant prime divisors, is in D ↔ with the extremal rays ∈ ΞY , where a ray ∈ ΞY is called extremal if there is a D ∈ Ξ such that ⊆ DY and deg D ∩ = ∅ holds. The set of extremal rays is denoted by Ξ× Y . The divisor of a semi-invariant function f · χu ∈ K(X) is then given by div(f · χu) = −Xy∈Y Xv∈Ξ(0) y µ(v) · (hv, ui + ordyf ) · Dy,v − X∈Ξ× Y hn, ui · D. Next we describe the canonical divisor. Choose a point y0 ∈ Y such that Ξy0 = ΞY holds. Then a canonical divisor on X = X(Ξ) is given by KX = (s − 2) · y0 − XΞy 6=ΞY Xv∈Ξ(0) i Dy,v − X∈Ξ× Y E. Proposition 4.3. Let D be a proper polyhedral divisor with Y (D) = P1, let Ξ be a refinement of D and denote by y1, . . . , ys ∈ Y the points with Ξyi 6= ΞY . Then the associated morphism ϕ : X(Ξ) → X(D) satisfies the following. (i) The prime divisors in the exceptional locus of ϕ are the divisors Dyi,v and D corresponding to v ∈ Ξ(0) yi \ D(0) yi and ∈ Ξ× Y \ D× respectively. (ii) Then the discrepancies along the prime divisors Dyi,v and D of (i) are computed as dyi,v = −µ(v) · (hv, u′i + αy) − 1, d = −hv, u′i − 1, b b b b b 28 J. HAUSEN, E. HERPPICH, AND H. S USS where the numbers αi are determined by   −1 µ(v1 1) ... µ(vr1 1 ) −1 0 ... 0 0 ... 0 0 ... 0 0 ... 0 0 ... 0 . . . . . . . . . . . . . . . −1 0 ... 0 0 µ(v1 1)v1 1 ... µ(vr1 1 )vr1 1 µ(v1 s )v1 µ(v1 s ) s ... ... s )vrs s ) µ(vrs . . . µ(vrs s n1 0 . . . ... ... nr 0 . . .   ·   αy1 ... αys u   =   2 − s 1 ... 1 1 ... 1   Proof. The first claim is obvious by the characterization of invariant prime divisors. For the second claim note that by [24, Theorem 3.1] every Cartier divisor on X(D) is principal. Hence, we may assume ℓ · KX = div(f · χu), div(f ) = Xy αy · y. Then our formulae for div(f · χu) and KX provide a row for every vertex vj i ∈ Ξyi , i = 0, . . . , s, and for every extremal ray i ∈ Ξ×, and ℓ−1(α, u) is the (unique) solution of the above system. (cid:3) Note, that in the above Proposition, the variety X(D) is Q-Gorenstein if and only if the linear system of equations has a solution. Proof of Theorem 4.1 and Proposition 4.2. We exemplarily discuss variety number eight. Recall that its Cox ring is given as R(X) = K[T1, . . . , T5]/(T1T 5 2 + T3T 5 4 + T 2 5 ) with the degrees 1, 1, 1, 1, 3. In particular, X is a hypersurface of degree 6 in P(1, 1, 1, 1, 3), and the self-intersection of the anti-canonical divisor can be calcu- lated as (−K 3 X ) = 6 · (1 + 1 + 1 + 1 + 3 − 6)3 1 · 1 · 1 · 1 · 3 = 2. The embedding X ⊆ P(1, 1, 1, 1, 3) is equivariant, and thus we can use the tech- nique described in [1, Sec. 11] to calculate a divisorial fan Ξ for X. The result is the following divisorial fan; we draw its slices and indicate the polyhedral divisors with affine locus by colouring their tail cones DY ∈ ΞY white: ( 3 5 , 1 5 ) Ξ0 Ξ1 (− 1 2 ,0) (0,− 1 5 ) Ξ∞ (-1,0) (-5,-2) ΞY (1,2) (1,0) One may also use [15, Cor. 4.9.] it computes the Cox ring in terms of Ξ, and, indeed, we obtain again R(X). Now we subdivide and obtain a divisorial fan having the refined slices as indicated in the following picture. to verify that Ξ is the right divisorial fan: b b b b b MULTIGRADED FACTORIAL RINGS AND FANO VARIETIES 29 5 , 1 ( 3 5 ) (− 1 2 ,0) (0,− 1 5 ) (-1,0) (-5,-2) (1,2) (3,1) (1,0) Here, the white ray Q≥0 · (1, 0) indicates that the polyhedral divisors with that tail have affine loci. According to [15, Cor. 4.9.], the corresponding Cox ring is given by (0,-1) R(eX) = K[T1, . . . , T7]/hT1T 5 2 + T3T 5 4 + T 2 5 T6i. We have to check that eX is smooth. Let us do this explicitly for the affine chart defined by the polyhedral divisor D with tail cone DY = cone((1, 2), (3, 1)). Then D is given by D = (cid:18)(cid:18) 3 5 , 1 5(cid:19) + σ(cid:19) · {0} + (cid:18)(cid:20)− , 0(cid:21) × 0 + σ(cid:19) · {∞}. 1 2 Thus, cone({1} × D0) + cone({−1} × D∞) is generated by (5, 3, 1), (−2, −1, 0) and (−1, 0, 0); in particular, it is a regular cone. This implies smoothness of the affine chart X(D). Furthermore, we look at the affine charts defined by the polyhedral divisors D with tail cone DY = cone(1, 0). Since they have affine locus, we have to check cone({1} × Dy), where y ∈ Y . For y 6= 0, 1, we have Dy = DY . In this case, cone({1} × Dy) is generated by (1, 1, 0), (0, 1, 0) and thus is regular. For y = 0, we obtain that cone({1} × Dy) is generated by (5, 3, 1), (1, 0, 0), (0, 1, 0) and this is regular. For y = 1 we get the same result. Hence, the polyhedral divisors with tail cone Dy = cone(1, 0) give rise to smooth affine charts. Now we compute the discrepancies according to Proposition 4.3. The resolution has two exceptional divisors D∞,0 and E(1,0). We work in the chart defined by the divisor D ∈ Ξ with tail cone DY = cone((1, 2), (1, 0)). The resulting system of linear equations and its unique solution are given by The formula for the discrepancies yields d∞,0 = −1 and d(1,0) = −2. In particular, X has non-canonical singularities. By a criterion from [24, Sec. 3.4.], we know that D∞,0 + 2 · E(1,0) is a nef divisor. It follows that eX is a minimal model over X. (cid:3) References [1] K. Altmann, J. Hausen: Polyhedral divisors and algebraic torus actions. Math. Ann. 334 (2006), no. 3, 557 -- 607. [2] K. Altmann, J. Hausen, H. Suss: Gluing Affine Torus Actions Via Divisorial Fans. Transfor- mation Groups 13 (2008), no. 2, 215 -- 242. [3] V.V. Batyrev: Toric Fano threefolds. Izv. Akad. Nauk SSSR Ser. Mat. 45 (1981), no. 4, 704 -- 717, 927. [4] V.V. Batyrev: On the classification of toric Fano 4-folds. J. Math. Sci. New York 94, 1021 -- 1050 (1999). [5] F. Berchtold, J. Hausen: Homogeneous coordinates for algebraic varieties. J. Algebra 266 (2003), no. 2, 636 -- 670. [6] Cheltsov, Ivan; Park, Jihun Sextic double solids. In. Bogomolov, Fedor (ed.) et al.: Coho- mological and geometric approaches to rationality problems. New Perspectives. Boston, MA: Birkhauser. Progress in Mathematics 282, 75 -- 132 (2010). [7] J.J. Chen, J.A. Chen, M. Chen: On quasismooth weighted complete intersections. Preprint, arXiv:0908.1439.   −1 −1 −1 5 0 0 0 0 0 2 −1 0 0 −1 0 1 1 3 1 0 0 0 −1 1 0 1 0 1 5 0   ,   α0 α1 α∞ u   =   0 1 0 −1 4   . b b b b b 30 J. HAUSEN, E. HERPPICH, AND H. S USS [8] C.H. Clemens, P.A. Griffiths: The intermediate Jacobian of the cubic threefold. Ann. Math. (2) 95 (1972), 281 -- 356. [9] A. Corti: Singularities of linear systems and 3-fold birational geometry. Explicit birational geometry of 3-folds, 259 -- 312, London Math. Soc. Lecture Note Ser., 281, Cambridge Univ. Press, Cambridge, 2000. [10] A. Corti, M. Mella: Birational geometry of terminal quartic 3-folds I. Am. J. Math. 126 (2004), No. 4, 739 -- 761. [11] A. Corti, A. Pukhlikov, M. Reid: Fano 3-fold hypersurfaces. Explicit birational geometry of 3-folds, 175 -- 258, London Math. Soc. Lecture Note Ser., 281, Cambridge Univ. Press, Cambridge, 2000. [12] M.M. Grinenko: Mori structures on a Fano threefold of index 2 and degree 1. Tr. Mat. Inst. Steklova 246 (2004), Algebr. Geom. Metody, Svyazi i Prilozh., 116 -- 141; translation in Proc. Steklov Inst. Math. 2004, no. 3 (246), 103 -- 128. [13] M.M. Grinenko: Birational automorphisms of a three-dimensional double cone. Mat. Sb. 189 (1998), no. 7, 37 -- 52; translation in Sb. Math. 189 (1998), no. 7-8, 991 -- 1007. [14] J. Hausen: Cox rings and combinatorics II. Mosc. Math. J., 8 (2008), 711 -- 757. [15] J. Hausen, H. Suss: The Cox ring of an algebraic variety with torus action. Adv. Math. 225 (2010), no. 2, 977 -- 1012. [16] A.R. Iano-Fletcher: Working with weighted complete intersections. Explicit birational geom- etry of 3-folds, 101 -- 173, London Math. Soc. Lecture Note Ser., 281, Cambridge Univ. Press, Cambridge, 2000. [17] M.-N. Ishida: Graded factorial rings of dimension 3 of a restricted type. J. Math. Kyoto Univ. 17 (1977), no. 3, 441 -- 456. [18] V.A. Iskovskih: Fano threefolds. II. Izv. Akad. Nauk SSSR Ser. Mat. 42 (1978), no. 3, 506- -- 549. [19] V.A. Iskovskih: Birational automorphisms of three-dimensional algebraic varieties. Current problems in mathematics, Vol. 12 (Russian), pp. 159 -- 236, 239 (loose errata), VINITI, Moscow, 1979. [20] V.A. Iskovskih, Yu.I. Manin: Three-dimensional quartics and counterexamples to the Luroth problem. Mat. Sb. (N.S.) 86 (128) (1971), 140 -- 166. [21] J.M. Johnson, J. Koll´ar: Fano hypersurfaces in weighted projective 4-spaces. Experiment. Math. 10 (2001), no. 1, 151 -- 158. [22] F. Knop, H Kraft, T. Vust: The Picard group of a G-variety. In: Algebraic Transformation Groups and Invariant Theory, DMV Seminar, Band 3, Birkhauser. [23] S. Mori: Graded factorial domains. Japan J. Math. 3 (1977), no. 2, 223 -- 238. [24] L. Petersen, H. Suss: Torus invariant divisors. Preprint, arXiv:0811.0517 (2008), to appear in Israel J. Math. [25] A.V. Pukhlikov: Birational automorphisms of a three-dimensional quartic with a simple sin- gularity. Mat. Sb. (N.S.) 135(177) (1988), no. 4, 472 -- 496, 559; translation in Math. USSR-Sb. 63 (1989), no. 2, 457 -- 482. [26] A.V. Pukhlikov: Birational isomorphisms of four-dimensional quintics. Invent. Math. 87, no. 2 (1987), 303 -- 329. [27] K. Watanabe, M. Watanabe: The classification of Fano 3-folds with torus embeddings. Tokyo J. Math. 5, 37 -- 48 (1982). [28] H. Suss: Canonical divisors on T -varieties. Preprint, arXiv:0811.0626v1. [29] A.S. Tikhomirov: The intermediate Jacobian of double P3 that is branched in a quartic. Izv. Akad. Nauk SSSR Ser. Mat. 44 (1980), no. 6, 1329 -- 1377, 1439. [30] Voisin, Claire: Sur la jacobienne interm´ediaire du double solide d'indice deux. Duke Math. J. 57 (1988), no. 2, 629 -- 646. Mathematisches Institut, Universitat Tubingen, Auf der Morgenstelle 10, 72076 Tubingen, Germany E-mail address: [email protected] Mathematisches Institut, Universitat Tubingen, Auf der Morgenstelle 10, 72076 Tubingen, Germany E-mail address: [email protected] Institut fur Mathematik, LS Algebra und Geometrie, Brandenburgische Technische Universitat Cottbus, PF 10 13 44, 03013 Cottbus, Germany E-mail address: [email protected]
1810.06882
1
1810
2018-10-16T08:53:48
Free rational curves on low degree hypersurfaces and the circle method
[ "math.AG", "math.NT" ]
We use a function field version of the Hardy-Littlewood circle method to study the locus of free rational curves on an arbitrary smooth projective hypersurface of sufficiently low degree. On the one hand this allows us to bound the dimension of the singular locus of the moduli space of rational curves on such hypersurfaces and, on the other hand, it sheds light on Peyre's reformulation of the Batyrev-Manin conjecture in terms of slopes with respect to the tangent bundle.
math.AG
math
FREE RATIONAL CURVES ON LOW DEGREE HYPERSURFACES AND THE CIRCLE METHOD TIM BROWNING AND WILL SAWIN Abstract. We use a function field version of the Hardy -- Littlewood circle method to study the locus of free rational curves on an arbitrary smooth projective hypersurface of sufficiently low degree. On the one hand this allows us to bound the dimension of the singular locus of the moduli space of rational curves on such hypersurfaces and, on the other hand, it sheds light on Peyre's reformulation of the Batyrev -- Manin conjecture in terms of slopes with respect to the tangent bundle. Contents Introduction 1. 2. Examples 3. Vector bundles on P1 4. The circle method: identification of major arcs 5. The circle method: minor arcs 6. Peyre's freedom counting function References 1 6 9 12 18 31 34 1. Introduction Let X ⊂ Pn−1 be a smooth hypersurface of degree d > 3, over a field K whose characteristic exceeds d if it is positive. This paper has two aspects. On the one hand, motivated by questions in algebraic geometry, we shall be interested in the locus of points corresponding to free rational curves inside the moduli space M0,0(X, e) of degree e rational curves on X. On the other hand, by working over a finite field, we shall establish a function field analogue of a recent conjecture due to Peyre [13] about the distribution of "sufficiently free" rational points of bounded height on Fano varieties. Date: October 17, 2018. 2010 Mathematics Subject Classification. 14H10 (11D45, 11P55, 14G05, 14J70). 2 TIM BROWNING AND WILL SAWIN 1.1. Geometry. The expected dimension of M0,0(X, e) is (n − d)e + n − 5, a fact that is known to hold for generic X if n > d + 3, thanks to Riedl and Yang [14]. It follows from work of Browning and Vishe [2] that M0,0(X, e) is irreducible and has the expected dimension for any smooth X, provided that n > (5d − 4)2d−1. Our first result strengthens this. Theorem 1.1. Let d > 3, let e > 1 and let n > (2d −1)2d−1. Then M0,0(X, e) is an irreducible locally complete intersection of the expected dimension. We can also bound the dimension of the singular locus of M0,0(X, e), as follows. Theorem 1.2. Let d > 3, let e > 1 and let n > 3(d − 1)2d−1. Then the space M0,0(X, e) is smooth outside a set of codimension at least 2d−2 − 6(d − 1)(cid:17)(cid:18)1 +(cid:22) e + 1 (cid:16) n d − 1(cid:23)(cid:19) . In particular, whenever these inequalities are satisfied, it is generically smooth and reduced. 1 For n > 2d + 1 and generic X of degree d > 3, Harris, Roth and Starr [6] have also shown that M0,0(X, e) is generically smooth. Note that, provided n > 3(d − 1)2d−1, the codimension goes to ∞ in Theorem 1.2 when either e or n does, with d fixed. Moreover, when both e and n are large with respect to 2d−2(d−1) of the total dimension. d, the codimension is at least approximately Our work addresses some questions of Eisenbud and Harris [4, §6.8.1] con- cerning the Fano variety of lines F1(X) = M0,0(X, 1) associated to a smooth hypersurface X ⊂ Pn−1 of degree d. Specifically, their question (a) asks whether F1(X) is reduced and irreducible if n > d + 1 and (b) asks whether the dimension of the singular locus of F1(X) can be bounded in terms of d alone. Theorems 1.1 and 1.2 answer the first question affirmatively for n > 3(d − 1)2d−1 and give some weak evidence in support of the second ques- tion, by showing that it grows with n more slowly than the dimension of the whole space. Furthermore, we handle the analogous conjectures with higher degree curves, with no loss in the dependence on n, meaning that for large enough e we do better than their predicted bound d 6 n/e. By comparison, Starr [15] has proved that if n > d + e and X is generic, then M0,0(X, e) has canonical singularities, which implies in particular that it is smooth outside a set of codimension at least 2. It does not seem possible that our method will prove that M0,0(X, e) has canonical singularities. By Mustata [11] and Lang-Weil [9] this is equivalent to the conjunction of an infinite sequence of Diophantine estimates (in the spirit of Definition 3.7), but for fixed n, d and e it seems unlikely that the circle method is able to handle more than finitely many of them. In unpublished work, Starr and FREE RATIONAL CURVES AND THE CIRCLE METHOD 3 Tian use a bend-and-break approach to produce a less restrictive lower bound for the codimension of the singular locus for a general hypersurface X ⊂ Pn−1 of degree d. However, their method never proves a lower bound for the codimension greater than n, whereas our work achieves this if e is sufficiently large. Comparing the various results, we see that Theorem 1.2 holds for a much more restricted range of n (unless e is very large relative to d) but it is valid for an arbitrary smooth hypersurface, rather than just a general one. It should be possible to adapt our strategy to prove results about moduli spaces of genus g curves on X. However, the codimension we obtain for the whole moduli space will not be any better than the codimension we can prove for the space of maps from a fixed genus g curve to X. In particular the codimension will shrink as g grows, so the bound obtained would only be suitable for e sufficiently large with respect to g. Our remaining result deals specifically with free curves and so we recall the definition here. Let TX be the tangent bundle associated to the smooth hypersurface X ⊂ Pn−1 (as defined in [7, p. 180], for example). The following is extracted from Debarre [3, Def. 4.5]. Definition 1.3. Let c : P1 → X be a rational curve and let ∈ Z. We say that c is -free if c∗TX ⊗ OP1(−) is globally generated. Following [3] we say that c is free if it is 0-free, and very free if it is 1-free. One easily checks that Definition 1.3 agrees with the standard definition that c is free if c∗TX is globally generated and very free if c∗TX is ample. Remark 1.4. If c is a -free rational curve on X then it follows from Defi- nition 1.3 that deg(c∗TX ) > rank(c∗TX ). In general, the pull-back of the tangent bundle has rank n − 2 and degree e(n − d). In this way we see that no degree e rational curve on X is ever (⌊ e(n−d) n−2 ⌋ + 1)-free. If d > 2 then this implies that 6 e, for any -free rational curve P1 → X. We let U ⊂ M0,0(X, e) be the Zariski open set that parameterises degree e maps from P1 to X that are -free. We write Z = M0,0(X, e) \ U for the complement. This is the closed set parameterising degree e maps P1 → X that are not -free. We shall prove the following bound for its dimension. Theorem 1.5. Let d > 3 and n > 3(d − 1)2d−1. Assume that > −1 and e > ( + 1)(cid:18)2 + 1 d − 2(cid:19) . (1.1) 4 Then TIM BROWNING AND WILL SAWIN dim Z 6 (n − d)e + n − 5 + 2(d − 1)(cid:22) + 1 2 (cid:23) d − 1(cid:23) −(cid:22) + 1 2d−2 − 6(d − 1)(cid:17)(cid:18)1 +(cid:22) e − −(cid:16) n 2 (cid:23)(cid:19) . (1.2) The notion of free rational curves was originally introduced as a tool to study uniruled and rational connectedness properties of varieties. Taking = 1 it follows from Theorems 1.1 and 1.5 that U1 6= ∅ if K is algebraically closed and e is sufficiently large. Hence, by appealing to [3, Cor. 4.17], we deduce that any smooth hypersurface X ⊂ Pn−1 of degree d is rationally connected if d > 3 and n > 3(d − 1)2d−1. This recovers a weak form of the well-known result due to Koll´ar, Miyaoka and Mori [8] that Fano varieties are rationally connected. In fact both proofs use reduction to characteristic p, but they use different properties of characteristic p varieties, with [8] relying on Frobenius pull-back and our work using the Lang -- Weil estimates. Theorem 1.2 is derived from Theorem 1.5, which is proved using analytic number theory and builds on an approach employed by Browning and Vishe [2]. (Theorem 1.1 uses essentially the same approach as [2], with one improve- ment to a key lemma.) One begins by working over a finite field K = Fq of characteristic > d. We bound the dimension of Z by counting the number In §3, we will of points defined over a finite extension of Fq that lie in it. give an explicit description of this locus in terms of a system of two Diophan- tine equations defined over the function field Fq(T ). Let f ∈ Fq[x1, . . . , xn] be a non-singular form of degree d that defines the hypersurface X ⊂ Pn−1. Given ∈ Z, we shall see that the primary counting function of interest to us, denoted N(q, e, f ), is the one that counts vectors (g, h) ∈ Fq[T ]2n, where g1, . . . , gn have degree at most e and no common zero, with at least one of degree exactly e, and where h1, . . . , hn have degree at most e − 1 − , such that f (g1, . . . , gn) = 0 and hi ∂f ∂xi (g1, . . . , gn) = 0. (1.3) n Xi=1 Since each partial derivative of f is a degree d − 1 polynomial, we obtain a linear equation for h ∈ Fq[T ]n where the coefficients have degree at most (d − 1)e in T . Standard heuristics lead us to expect that, for typical g, the number of available h is q(e−)(n−1)−(d−1)e = qe(n−d)−(n−1). (In fact, we shall see in Lemmas 3.4 and 3.5 that this is true only if the map P1 → X repre- sented by g is -free.) Thus we expect that N(q, e, f ) is approximated by qe(n−d)−(n−1)N(q, e, f ), where N(q, e, f ) is the number of vectors g ∈ Fq[T ]n such that f (g) = 0, where g1, . . . , gn have degree at most e and no common zero, with at least one of degree exactly e. FREE RATIONAL CURVES AND THE CIRCLE METHOD 5 In §4, we apply the function field version of the Hardy -- Littlewood circle method to study the system of degree d equations (1.3), expressing the number of solutions as an integral of an exponential sum. We shall show that the major arc contribution to this integral cancels almost exactly with the expected approximation qe(n−d)−(n−1)N(q, e, f ). In §5, we prove an upper bound on all other arcs, taking special care to make all of our implied constants depend explicitly on the size of the finite field q. The standard way of proceeding involves d−1 applications of Weyl differencing, a process that would ultimately require n > 3(d − 1)2d variables overall. We shall gain a 50% reduction in the number of variables by exploiting the special shape of the Diophantine system (1.3). Finally, we bring everything together and apply the Lang -- Weil estimates [9] to turn the bound for #Z(Fq) into a bound for the dimension of Z. An application of spreading-out shows that the dimension bound holds over an arbitrary base field K such that char(K) > d if it is positive. 1.2. Arithmetic. In our geometric investigation of Z we take the point of view that e and are fixed and q → ∞. In this subsection we assume that the finite field is fixed, but we allow the parameters e and to tend to infinity appropriately. Suppose that V is a smooth projective geometrically integral Fano variety defined over a number field K. For suitable Zariski open subsets U ⊂ V the Batyrev -- Manin conjecture [5] makes a precise prediction about the asymptotic behaviour of the counting function NU (B) = #{x ∈ U(K) : Hω−1 V (x) 6 B}, V : V (K) → R is an anticanonical height function. as B → ∞, where Hω−1 These conjectures are flawed, however, since it has been discovered that the presence of Zariski dense thin sets in V (K) may skew the expected asymp- totics. Recently, Peyre [13] has embarked on an ambitious programme to repair the conjecture by associating a measure of "freeness" ℓ(x) ∈ [0, 1] to any x ∈ V (K) and only counting those rational points for which ℓ(x) > εB, where εB is a function of B decreasing to zero sufficiently slowly. (See [13, Def. 6.11] for a precise statement.) Peyre's function ℓ(x) is defined using Arakelov geometry and the theory of slopes associated to the tangent bundle TV . We can lend support to Peyre's freedom prediction [13, §6] by studying smooth hypersurfaces of low degree in the setting of global fields of positive characteristic. Let X ⊂ Pn−1 be a smooth hypersurface of degree d defined over a finite field Fq whose characteristic exceeds d. We put N ε-free X (B) = #(cid:8)x ∈ X(K) : ℓ(x) > ε, H−ωX (x) 6 qB(cid:9) , (1.4) 6 TIM BROWNING AND WILL SAWIN where K = Fq(T ) is the rational function field and ℓ(x) will be defined in §6. The expectation is that for a suitable range of ε, N ε-free (B) should have the same asymptotic behaviour as the usual counting function NX(B), as B → ∞. The following result confirms this and will be proved in §6. X Theorem 1.6. Let d > 3, let n > 3(d − 1)2d−1 and let 0 6 ε < n − 1 (n − d)(d − 1)22d−1 . Then there exists δ > 0 such that as B → ∞, where cX is the function field analogue of the constant predicted by Peyre [12]. Furthermore, the implied constant only depends on q and f . N ε-free X (B) = cXqB + O(cid:0)q(1−δ)B(cid:1) , Note that this result does not require εB to decrease to zero, but only to stay below some fixed constant. This may be because the hypersurface X has Picard rank one, since Peyre has shown in [13, §7.2] that for the product P1 × P1 one requires εB → 0 for the asymptotic formula to be true. Finally, one can see from the arguments in Theorem 1.6 that we can take the upper bound for ε to be significantly greater than (n−d)(d−1)22d−1 when n is large. (In fact, the cutoff is allowed to approach 1 n−1 d+1 as n → ∞.) With appropriate adjustments it seems likely that our proof of Theorem 1.6 can be extended to handle the corresponding result for smooth hypersurfaces of low degree defined over Q, with Poisson summation taking the place of the Riemann -- Roch arguments that feature in §3. There will be additional complexities arising from smooth weights and the need to work in certain congruence classes modulo the primes of bad reduction of X in the analogue of Proposition 3.8, but these can be controlled easily in the circle method. Acknowledgements. The authors are grateful to Paul Nelson, Per Salberger and Jason Starr for useful comments. While working on this paper the first author was supported by EPRSC grant EP/P026710/1. The research was partially conducted during the period the second author served as a Clay Research Fellow, and partially conducted during the period he was supported by Dr. Max Rossler, the Walter Haefner Foundation and the ETH Zurich Foundation. 2. Examples As usual, X ⊂ Pn−1 is assumed to be a smooth hypersurface of degree d > 3, over a field K whose characteristic is either 0 or > d. While the latter condition arises very naturally in our argument (as explained in Remark 5.5), the following result shows that the statement of Theorem 1.5 is actually false when it is dropped. FREE RATIONAL CURVES AND THE CIRCLE METHOD 7 Lemma 2.1. Let K = Fp for a prime p and let X ⊂ Pn−1 be the Fermat hypersurface xd 1 + · · · + xd n = 0. Assume that p ∤ d and d 6= apr − 1 for any r ∈ N and a ∈ {0, . . . , p − 1}. Then X is smooth, none of the curves in dim M0,0(X, 1) are (−1)-free, and dim M0,0(X, 1) > 2n − d − 5. 1 +· · ·+xd Proof. The moduli space of n-tuples of polynomials of degree 6 1 satisfying the equation xd n = 0 is a GL2-bundle over the moduli stack M0,0(X, 1) parameterising lines in X, because for each line we can choose any basis of the corresponding two-dimensional vector space. Thus its dimension is equal to 4 + dim M0,0(X, 1). This space is is cut out by d + 1 equations in 2n variables, of i is greater than the corresponding base p digit of d. In this way we see where (cid:0)d theorem it follows that p (cid:0)d that p (cid:0)d i(cid:1) divides all coefficients of the ith equation, for 0 6 i 6 d. By Lucas' i(cid:1) if and only if at least one of the base p digits i(cid:1) for some 0 6 i 6 d if and only if d does not take the form apr − 1 for some a ∈ {0, . . . , p − 1}. But then the space is cut out by fewer than d + 1 equations in 2n variables. This implies that it has dimension greater than 2n − d − 1, whence dim M0,0(X, 1) > 2n − d − 5. Furthermore, since the dimension near each curve is greater than the expected dimension, it follows from Lemma 3.1 that they are not (−1)-free. Finally, the Fermat hypersurface is smooth over K if and only if p ∤ d. (cid:3) This example generalises a discussion of Debarre [3, §2.15]. It shows that for typical p < d the statements of Theorems 1.1 and 1.5 are false for fields of characteristic p. Returning to the general setting, the following result provides examples of curves that are not -free. Lemma 2.2. Let d, m, n ∈ N with d > 3 and m 6 n/2. Let K be an infinite field. There exists a non-singular form f (x1, . . . , xn) over K of degree d, such that f (x1, . . . , xm, 0, . . . , 0) = (x1, . . . , xm, 0 . . . , 0) = 0 ∂f ∂xj for all x1, . . . , xm and all j 6 n − m. For such a polynomial, every map c : P1 → X of degree e that factors through Pm−1 ⊆ X ⊆ Pn−1 fails to be (⌊ e(m−d) m−1 ⌋ + 1)-free. The moduli space of such rational curves has dimension m(e + 1) − 4. Let X ⊂ Pn−1 be a smooth hypersurface with underlying polynomial f , as in the lemma. Taking m = d and = 0, we see that when n > 2d the space Z1 of non-very-free rational curves P1 → X of degree e has dimension at least d(e + 1) − 4. 8 TIM BROWNING AND WILL SAWIN Proof of Lemma 2.2. Without the non-singularity condition, the space of such polynomials is linear. The singular polynomials form a closed subset. To prove the existence, it is sufficient to show that this subset has codimension 1. The set of singular polynomials is the projection from the product of this linear space with Pn−1 of the set of pairs of a point and a polynomial singular at that point. For elements in Pm−1 ⊆ Pn−1, the space of polynomials singular at that point has codimension m, as it is defined by the m independent conditions ∂f (x1, . . . , xm, 0 . . . , 0) = 0 for n − m + 1 6 j 6 n. For all other elements, ∂xj we claim that the n conditions ∂f (x1, . . . , xn) = 0 for 1 6 j 6 n define a ∂xj codimension n subspace. To see this we may take a linear form l in the last n−m coordinates that is nonzero at that point. Then the n-dimensional space of polynomials generated by xjld−1 for 1 6 j 6 n lie in the linear subspace, since d − 1 > 2. But only the zero element in that subspace satisfies all n conditions. It follows that the singular locus is the union of the projection of a codimension m bundle on Pm−1 and a codimension n bundle on its complement in Pn−1. Thus the singular locus has codimension at least one, as desired. For the freeness, we use the Euler exact sequence 0 → OPn−1 → OPn−1(1)n → TPn−1 → 0. (2.1) Consider the map OPn−1(1)n → OPn−1(1)m given by projection onto the last m factors. Because m 6 n/2 the composition of this projection with the map OPn−1 → OPn−1(1)n vanishes on Pm−1. So over Pm−1, we obtain a map TPn−1 → OPn−1(1)m. Next consider the exact sequence 0 → TPn−1 → TX → OX(d) → 0 on X. The second map of this sequence is the dot product with the derivative of f . By assumption on f , restricted to Pm−1, this map factors through the projection onto the last m vectors. Hence we obtain an exact sequence 0 → V → OPm−1(1)m → OPm−1(d) → 0 whose kernel V is a vector bundle on Pm−1 of degree m − d, which arises as a quotient of TX . For c : P1 → X a map of degree e whose image lies in Pm−1, c∗V is a vector bundle of degree e(m − d) on P1 which arises as a quotient of c∗TX. Because c∗V splits as a direct sum of m − 1 line bundles, it must contain some line bundle summand of degree at most e(m−d) m−1 , and we can round down to the nearest integer. Hence c∗TX has some line bundle summand of degree at most ⌊ e(m−d) m−1 ⌋ and hence c is not (⌊ e(m−d) The dimension estimate is the standard calculation for the moduli space of (cid:3) rational curves in projective space. m−1 ⌋ + 1)-free. Even for a general hypersurface there are some non-very-free curves. Indeed, for such a variety, the moduli space of lines has dimension 2n − d − 5, and FREE RATIONAL CURVES AND THE CIRCLE METHOD 9 each line admits a (2e + 1)-dimensional moduli space of degree e maps from P1 to that line. Because the pull-back of the tangent bundle to a line has rank n − 2 and degree n − d, it contains some summand of degree at most 0 as soon as d > 2, and so every pull-back of it has a summand of the same degree, and so these degree e coverings of lines fail to be 1-free. Hence, for a general hypersurface X ⊂ Pn−1 of degree d, we have dim Z1 > 2(n + e) − d − 7. These examples show that the dimension of the moduli space of non-very- free curves can grow linearly in n and it can grow linearly in e. We do not know if it can grow linearly in ne, as the dimension of M0,0(X, e) does. 3. Vector bundles on P1 Let f be a homogeneous polynomial of degree d in n variables over a field K and let X ⊂ Pn−1 be its projective zero locus. Assume that X is smooth and let TX be its tangent bundle. In this section we investigate the geometry of -free rational curves c : P1 → X, in the sense of Definition 1.3. It turns out that there is a natural characterization of the (−1)-free curves, which we recall here. Lemma 3.1. A rational curve c : P1 → X of degree e is (−1)-free if and only if, in a neighborhood of c, the moduli space of rational curves on X is smooth of dimension (n − d)e + n − 5. Under the assumptions of Theorem 1.1 or [14], M0,0(X, e) has dimension (n − d)e + n − 5, so this is simply equivalent to M0,0(X, e) being smooth at c. Proof of Lemma 3.1. The tangent space of the moduli space of rational curves at c is H 0(P1, c∗TX )/H 0(P1, TP1). Note that H 0(P1, TP1) has dimension 3. By Riemann -- Roch, dim H 0(P1, c∗TX ) − dim H 1(P1, c∗TX) = dim(c∗TX ) + deg(c∗TX ) = n − 2 + e(n − d). Hence if c is a smooth point on a component of dimension n − 5 + e(n − d) then H 0(P1, c∗TX ) has dimension n − 2 + e(n − d) and so H 1(P1, c∗TX) vanishes. Thus [3, Remark 4.6] implies that c is (−1)-free. Conversely if c is (−1)-free then H 1(P1, c∗TX ) vanishes by [3, Remark 4.6], so deformations are unobstructed. Thus the moduli space is smooth at c, and the dimension of the tangent space to the moduli space is n − 5 + e(n − d). (cid:3) Let TX be the inverse image of TX ⊆ TPn−1 under the map OPn−1(1)n → TPn−1 in the Euler sequence (2.1). This yields 0 → OX → TX → TX → 0, 10 TIM BROWNING AND WILL SAWIN so that in particular TX is a vector bundle of rank n − 1 on X. With this in mind, we refine Definition 1.3 as follows. Definition 3.2. We say that c : P1 → X is strongly -free if c∗ TX ⊗ OP1(−) is globally generated. We thank Paul Nelson for asking a question that suggested the above def- inition, and which turns out to simplify our argument compared to studying the tangent bundle directly. Lemma 3.3. If c is strongly -free, then it is -free. Proof. This follows from the fact that TX is a quotient of TX and if a vector bundle is globally generated then every quotient is globally generated. (cid:3) Lemma 3.4. We have dim H 0(P1, c∗ TX ⊗ OP1(−1 − )) > e(n − d) − (n − 1) with equality if and only if c is strongly -free. Proof. Because TX is the kernel of the map df : TPn−1 → OPn−1(d), TX is the kernel of a map OPn−1(1)n → OPn−1(d) and hence has degree n − d. Thus c∗ TX has degree e(n − d). Because it has rank n − 1, its tensor product with with OP1(−1 − ) has degree e(n − d) − (n − 1) − (n − 1). Hence by Riemann -- Roch, the dimension of its space of global sections is dim H 0(P1, c∗ TX ⊗ OP1(−1 − )) = e(n − d) − (n − 1) + dim H 1(P1, c∗ TX ⊗ OP1(−1 − )). It now suffices to show that H 1(P1, c∗ TX ⊗ OP1(−1 − ) vanishes if and only if c∗ TX ⊗ OP1(−) is globally generated. We can assume that n−1 c∗ TX = OP1(ki). Mi=1 Then H 1(P1, c∗ TX ⊗ OP1(−1 − )) = 0 if and only if ki − 1 − > −1 for all i, which happens if and only if ki − > 0 for all i, which occurs if and only if c∗ TX ⊗ OP1(−) is globally generated. (cid:3) Vector notation such as g or h will denote n-tuples of polynomials in T . Let g be an n-tuple of polynomials in T of degree at most e, at least one of degree e, with no common zero, and such that f (g) = 0. These conditions ensure that (g1 : · · · : gn) defines a degree e map c : P1 → X. Lemma 3.5. H 0(P1, c∗ TX ⊗ OP1(−1 − )) is isomorphic to the space of n- tuples h of polynomials in T of degree 6 e − 1 − , such that ∇f (g) · h = 0. FREE RATIONAL CURVES AND THE CIRCLE METHOD 11 Proof. In this proof it will be convenient to set B = c∗ TX ⊗ OP1(−1 − )). We have an exact sequence 0 → TX → OPn−1(1)n → OPn−1(d) → 0, with the last map given by multiplication by the gradient of f . Thus we obtain an exact sequence 0 → B → c∗OPn−1(1)n ⊗ OP1(−1 − ) → c∗OPn−1(d) ⊗ OP1(−1 − ) → 0 which simplifies to 0 → B → OP1(e − 1 − )n → OP1(de − 1 − ) → 0, because c has degree e. Applying the cohomology long exact sequence, we see that H 0(P1, B) is the kernel of the natural map H 0(P1, OP1(e − 1 − )n) → H 0(P1, OP1(de − 1 − )), given by multiplication by the gradient of f . Since H 0(P1, OP1(e − 1 − )n) = H 0(P1, OP1(e − 1 − ))n is the space of n-tuples of polynomials of degree at most e − 1 − , this is exactly the stated space. (cid:3) We now assume K = Fq is a finite field. Thus f ∈ Fq[x1, . . . , xn] is a non-singular form of degree d > 3. We assume throughout that char(Fq) > d. Definition 3.6. Let N(q, e, f ) be the number of tuples of n polynomials g1, . . . , gn over Fq, of degree at most e, at least one of degree exactly e, with no common zero, such that f (g1, . . . , gn) = 0. Definition 3.7. For each integer , let N(q, e, f ) be the number of pairs of a tuple of polynomials g1, . . . , gn over Fq, of degree at most e, at least one of degree exactly e, with no common zero and a tuple of polynomials h1, . . . , hn over Fq, of degree at most e − 1 − , such that (1.3) holds. Proposition 3.8. (1) The number of Fq-points on M0,0(X, e) is N(q, e, f ) (q − 1)(q3 − q) . (2) The number of Fq-points on Z is at most N(q, e, f )q(n−1)−e(n−d) − N(q, e, f ) (q − 1)2(q3 − q) . Proof. Each point of M0,0(X, e) corresponds to PGL2(Fq) = q3 − q distinct maps P1 → X. Thus in (1) we will count the number of maps P1 → X, and in (2) we will count the number of maps P1 → X that are not -free, and in each case then divide by q3 − q. For (1), it is sufficient to note that for any such tuple g, (g1 : · · · : gn) are the projective coordinates of a degree e map P1 → X. All such maps arise this 12 TIM BROWNING AND WILL SAWIN way, and two tuples define the same map if and only if one is the multiple of the other by a non-zero scalar. For (2), it follows from Lemma 3.3 that it suffices to consider the space of degree e maps c : P1 → X that are not strongly -free. Note that N(q, e, f ) is the sum over tuples of polynomials (g1, . . . , gn), defining maps c, of q raised to the dimension of the vector space of possible h1, . . . , hn. By Lemma 3.5 this exponent is dim H 0(P1, c∗ TX ⊗ OP1(−1 − )). By Lemma 3.4, q to the power of this dimension is equal to qe(n−d)−(n−1) if c is strongly -free and is at least qe(n−d)−(n−1)+1 otherwise. Hence N(q, e, f )q(n−1)−e(n−d) > Xg∈Fq[T ]n g=e q + Xg∈Fq[T ]n g=e 1 c not strongly -free c strongly -free = N(q, e, f ) + (q − 1) Xg∈Fq[T ]n g=e 1. c not strongly -free The proposition follows on noting that there are (q − 1) tuples g for each map c : P1 → X. (cid:3) 4. The circle method: identification of major arcs For e > 1 we have N(q, e, f ) = #{g ∈ Fq[T ]n : g = qe, f (g) = 0, gcd(g1, . . . , gn) = 1}, where g = (g1, . . . , gn) and g = max16i6n gi. In particular only non-zero vectors g occur. Similarly, we may write N(q, e, f ) = Xg∈Fq[T ]n g=qe f (g)=0 1, Xh∈Fq[T ]n h<qe− h.∇f (g)=0 gcd(g1,...,gn)=1 where once again we note that only non-zero vectors g occur. We may use the function field analogue of the Mobius function µ : Fq[T ] → {0, ±1} to detect FREE RATIONAL CURVES AND THE CIRCLE METHOD 13 the coprimality condition gcd(g1, . . . , gn) = 1. This gives N(q, e, f ) = Xk∈Fq[T ] k monic =Xj>0 Xk∈Fq[T ] k=qj k monic 0<g=qe/k µ(k) Xg∈Fq[T ]n µ(k) Xg∈Fq[T ]n f (g)=0 1 h<qe− h.∇f (g)=0 Xh∈Fq[T ]n Xh∈Fq[T ]n 0<g=qe−j f (g)=0 h<qe− h.∇f (g)=0 In view of the elementary identity Xk∈Fq[T ] k=qj k monic 1 −q 0 µ(k) =  if j = 0, if j = 1, if j > 1, it readily follows that where and cjN(e − j + 1, e − ), N(q, e, f ) =Xj>0 cj =  N(u, v) = Xg∈Fq[T ]n 1 −(q + 1) q 0 0<g<qu f (g)=0 if j = 0, if j = 1, if j = 2, if j > 2 1, Xh∈Fq[T ]n h<qv h.∇f (g)=0 1. (4.1) (4.2) for any integers u, v > 1. We have 1 =ZT Xh∈Fq[T ]n h<qe− h.∇f (g)=0 S(β)dβ, where S(β) = Xh∈Fq[T ]n h<qe− ψ(βh.∇f (g)). Here the integral is over the space T of formal Laurent series in T −1 of degree less than 0, against the Haar measure with total mass 1, and ψ is the additive character of Fq((T −1)) that sends a formal Laurent series in T −1 to a fixed 14 TIM BROWNING AND WILL SAWIN non-trivial additive character of Fq applied to the coefficient of T −1. With this notation we now have N(q, e, f ) =Xj>0 cj Xg∈Fq[T ]n 0<g<qe−j+1 ZT f (g)=0 S(β)dβ. (4.3) Our plan will be to define a set of major arcs whose total contribution to q(n−1)−e(n−d)N(q, e, f ) is matched by N(q, e, f ). We note that the sum over g is empty unless e > j, so we will be able to assume this whenever dealing with this sum. In what follows we shall frequently make use of the basic orthogonality property ψ(γb) =(qB 0 if kγk < q−B, otherwise, Xb∈Fq[T ] b<qB (4.4) which is valid for any integer B > 0 and any γ ∈ Fq((T −1)). Here we recall that kγk = Pi6−1 biti for any γ =Pi6N biti ∈ Fq((T −1)). Let g ∈ Fq[T ]n be a non-zero vector such that f (g) = 0. The next result is the first step towards defining the relevant set of major arcs for our problem. Lemma 4.1. Suppose that β = a/r + θ for coprime polynomials a, r ∈ Fq[T ] such that a < r 6 qe−. Assume that rθ < q−(d−1)(e−j). Then S(β) =(qn(e−) 0 if r gcd(g1, . . . , gn)d−1 and θ < q−e/gd−1, otherwise. Proof. We break the sum into residue classes modulo r, by writing h = u + rv for u < r and v < qe−/r. Then S(β) = Xu∈Fq[T ]n u<r ψ(βu.∇f (g)) Xv∈Fq[T ]n v<qe−/r ψ(rθv.∇f (g)) Since rθ < q−(d−1)(e−j) we have rθ∇f (g) 6 rθq(d−1)(e−j) < 1. Thus krθ∇f (g)k = rθ∇f (g) and it follows from (4.4) that ψ(rθv.∇f (g)) =(r−nqn(e−) 0 if θ∇f (g) < q−e, otherwise. Xv∈Fq[T ]n v<qe−/r We claim that ∇f (g) = gd−1. To see this suppose that g = qm for a non-negative integer m and let g∗ ∈ Fn q be the (non-zero) leading coefficient of FREE RATIONAL CURVES AND THE CIRCLE METHOD 15 g. In particular f (g∗) = 0 since f (g) = 0. Since f has degree d it follows that the coefficient of T m(d−1) in ∇f (g) is ∇f (g∗) 6= 0, since f is non-singular. Our argument so far shows that S(β) =(r−nqn(e−)T (β) 0 if θ < q−e/gd−1, otherwise, where T (β) = Xu∈Fq[T ]n u<r When θ < q−e/gd−1 it follows that ψ(βu.∇f (g)). θu.∇f (g) 6 q−1θr∇f (g) 6 q−2+−er 6 q−2, since r 6 qe−. Hence, since a and r are coprime, we deduce that T (β) = Xu∈Fq[T ]n u<r ψ(cid:18)au.∇f (g) r (cid:19) =(rn 0 if r ∇f (g), otherwise, Since f is a non-singular form, the statement of the lemma follows on noting that r ∇f (g) if and only if r gcd(g1, . . . , gn)d−1. (cid:3) Lemma 4.2. Suppose that e > and a1 r1 + θ1 = a2 r2 + θ2, with r1, r2 gcd(g1, . . . , gn)d−1 and θ1, θ2 < q−e/gd−1. Then in fact a1 r1 (and so θ1 = θ2). = a2 r2 Proof. By clearing denominators, we may assume r1 = r2 = gcd(g1, . . . , gn)d−1. Then a1 − a2 = gcd(g1, . . . , gn)d−1(θ2 − θ1), so that a1 − a2 < q−e gcd(g1, . . . , gn)d−1 6 q−e 6 1. gd−1 This implies that a1 = a2, as required. We take as major arcs the union Nj = [r∈Fq[T ] monic r6qe− [a<r gcd(a,r)=1(cid:8)β ∈ Fq((T −1)) : rβ − a < q−(d−1)(e−j)(cid:9) , (cid:3) (4.5) for j > 0. It follows from Lemma 4.1 that S(β) is non-zero for β ∈ Nj if and only if there is some pair (a/r, θ) such that β = a/r + θ and all the conditions r 6 qe−, θ < r−1q−(d−1)(e−j), a < r, gcd(a, r) = 1, 16 and TIM BROWNING AND WILL SAWIN r gcd(g1, . . . , gn)d−1, θ < q−e/gd−1 are satisfied. By Lemma 4.2, pairs satisfying these conditions (or even the last three conditions) are unique. Hence we can rewrite the integral over the major arcs as ZNj S(β)dβ = qn(e−) Xr6qe− r monic gcd(a,r)=1 Zθ<min{q−e/gd−1, r−1q−(d−1)(e−j)} Xa<r ϕ(r)Zθ<min{q−e/gd−1, r−1q−(d−1)(e−j)} dθ, dθ rgcd(g1,...,gn)d−1 = qn(e−) Xr6qe− r monic rgcd(g1,...,gn)d−1 for any non-zero vector g ∈ Fq[T ]n such that f (g) = 0, where ϕ(r) is the function field analogue of the Euler totient function. We want to replace the integral over θ by Zθ<q−e/gd−1 dθ = q−e gd−1 . The error in doing this is at most this volume multiplied by the indicator function for the inequality r−1q−(d−1)(e−j) < q−e/gd−1. Since r gcd(g1, . . . , gn)d−1 this inequality implies that qj+D+1g 6 gcd(g1, . . . , gn)qe, where At this point we observe that d − 1(cid:23) . D =(cid:22) e − (4.6) (4.7) ϕ(r) = gcd(g1, . . . , gn)d−1, Xr ∈ Fq[T ] monic rgcd(g1,...,gn)d−1 since g 6= 0. Note that when r gcd(g1, . . . , gn)d−1 and r > qe− we must have gcd(g1, . . . , gn) > qD+1, (4.8) with D as above. Putting everything together it follows that S(β)dβ = ZNj q(n−1)(e−) gcd(g1, . . . , gn)d−1 gd−1 (1 + ǫj1j(g)) (4.9) FREE RATIONAL CURVES AND THE CIRCLE METHOD 17 for ǫj ∈ [−1, 1], where 1j(g) =(1 0 if (4.6) or (4.8) hold, otherwise. Let N major (q, e, f ) denote the contribution to the right hand side of (4.3) from (4.9) for each j. We now see that N major (q, e, f ) = q(n−1)(e−)Xj>0 cj Xg∈Fq[T ]n 0<g<qe−j+1 f (g)=0 gcd(g1, . . . , gn)d−1 gd−1 (1 + ǫj1j(g)) . On noting that (n − 1)(e − ) − e(d − 1) = e(n − d) − (n − 1), the main term is seen to be qe(n−d)−(n−1)(cid:16) N(e) − qd N (e − 1)(cid:17) , where for u > 0 we set N (u) = Xg∈Fq[T ]n g=qu f (g)=0 gcd(g1, . . . , gn)d−1 = Xk∈Fq[T ] k6qu k monic kd−1N(q, u − deg(k), f ) qdℓN(q, u − ℓ, f ), = u Xℓ=0 in the notation of Definition 3.6. Hence e N (e) − qd N (e − 1) = qdℓN(q, e − ℓ, f ) − Xℓ=0 = N(q, e, f ). qd(ℓ+1)N(q, e − 1 − ℓ, f ) e−1 Xℓ=0 Remark 4.3. The cancellation here is not miraculous. The terms corresponding to g with g < qe or gcd(g1, . . . , gn) > 1 disappear precisely because cj were the coefficients defined in (4.2) to sieve out these terms in the first place. Turning to the error term we can combine (4.6) and (4.8) to deduce that gcd(g1, . . . , gn) > qD+1 min(1, qj−eg) = qD+1+j−eg whenever 1j(g) = 1. Hence N major (q, e, f ) − qe(n−d)−(n−1)N(q, e, f ) 6 q(n−1)(e−)Xj>0 cjEj, 18 where TIM BROWNING AND WILL SAWIN Ej = X06u6e−j Xk∈Fq[T ] monic = X06u6e−j Xℓ>D+1+j−e+u k>qD+1+j−e+u kd−1 qu(d−1) #(cid:26)g ∈ Fq[T ]n : g = qu, f (g) = 0 k = gcd(g1 . . . , gn) (cid:27) qℓ q(u−ℓ)(d−1) N(q, u − ℓ, f ). Invoking [1, Lemma 2.8], we deduce that N(q, e, f ) = Of (q(e+1)(n−1)) for any n > 3, where the implied constant depends at most on f . Hence, since we may clearly assume that n > d + 1, it follows that q−ℓ(n−d−1) qu(n−d)+n−1 Xℓ>D+1+j−e+u Ej ≪f X06u6e−j ≪f q−(D+1)(n−d−1) X06u6e−j ≪f q(e−j)(n−d)+n−1−(D+1)(n−d−1). qu(n−d)+n−1 q(u−e+j)(n−d−1) (4.10) The implied constant in this estimate depends only on f and not on q. Thus q(n−1)(e−)Xj>0 cjEj ≪f q2e(n−d)−(n−1)+de−e+n−1−(D+1)(n−d−1). Putting everything together, we may conclude as follows. Lemma 4.4. Let ∈ Z and assume that e > . Then N major (q, e, f ) = qe(n−d)−(n−1)(cid:0)N(q, e, f ) + Of (q(e+1)(n−1)−(D+1)(n−d−1))(cid:1) , where D is given by (4.7). 5. The circle method: minor arcs It remains to study the quantity N minor (q, e, f ) =Xj>0 cj Xg∈Fq[T ]n 0<g<qe−j+1 Znj f (g)=0 S(β)dβ, (5.1) where nj is the complement in T = {β ∈ Fq((T −1)) : β < 1} of the major arcs Nj that we defined in (4.5). Indeed, in view of Proposition 3.8(2), the following result is now a direct consequence of (4.3) and Lemma 4.4. Lemma 5.1. Assume that d > 3 and e > . Then #Z(Fq) 6 q(n−1)−e(n−d) (q − 1)2(q3 − q) N minor (q, e, f ) + Of (q(e+1)(n−1)−5−(D+1)(n−d−1)), FREE RATIONAL CURVES AND THE CIRCLE METHOD 19 where D is given by (4.7). We have Znj S(β)dβ =ZTZnj Xg∈Fq[T ]n 0<g<qe−j+1 f (g)=0 where (S(α, β) − qn(e−))dαdβ, (5.2) S(α, β) = Xg∈Fq[T ]n g<qe−j+1 Xh∈Fq[T ]n h<qe− ψ(αf (g) + βh.∇f (g)). (5.3) Viewed as polynomials in the 2n variables (g, h) the pair of polynomials f (g) and h.∇f (g) are homogeneous of degree d. The obvious thing to do at this point is to apply Weyl differencing d − 1 times in the spirit of Birch. This requires one to work with a simultaneous Diophantine approximation of α and β, which is somewhat wasteful. It bears fruit provided that 2n − dim V ∗ > 3(d − 1)2d, where V ∗ is the (affine) "Birch singular locus". In this setting V ∗ is the locus of (g, h) ∈ A2n such that the pair of vectors (∇f (g), 0) and (h.∇2f (g), ∇f (g)) are proportional. Since f is non-singular, it follows that V ∗ is the set of (g, h) ∈ A2n such that g = 0, so that dim V ∗ = n. In this way we see that the standard approach would require n > 3(d − 1)2d variables overall, although there are additional difficulties associated to having lopsided boxes. In our work we shall exploit the special shape of our polynomials in such a way that our estimates are only sensitive to the Diophantine approximation properties of α or β independently. This allows us to handle half the number of variables when dealing with the sum S(α, β). In what follows it will be convenient to define the monomials P0(T ) = T e−j, P (T ) = T e−j+1 and Q(T ) = T e−. Let M(J) = [r∈Fq[T ] monic r6qJ [a<r gcd(a,r)=1(cid:8)α ∈ Fq((T −1)) : rα − a < qJ P0−d(cid:9) , (5.4) for any integer J. Note that M(−1) = ∅. Let M =(cid:24)d(e − j) 2 According to the function field version of Dirichlet's approximation theorem any element of T has a representation a/r + θ with a < r 6 qM and (cid:25) . (5.5) 20 TIM BROWNING AND WILL SAWIN rθ < q−M . Hence we can cover T by a union of arcs M(J + 1) \ M(J) for integers J such that −1 6 J 6 M − 1. Next, let N(K) = [r∈Fq[T ] monic r6qK [a<r gcd(a,r)=1 (cid:26)β ∈ Fq((T −1)) : rβ − a < qK P0d−1Q(cid:27) , (5.6) for any integer K. We note that N(e − ) = Nj, in the notation of (4.5). Let N =(cid:24)(e − j)(d − 1) + e − 2 (cid:25) . (5.7) It now follows from Dirichlet's approximation theorem that the minor arcs nj can be covered by the union of arcs N(K + 1) \ N(K) for integers K such that e − 6 K 6 N − 1. Observe in particular that if any minor arcs exist then e − < N so (d − 1)(e − j) > e − . (5.8) We may thus assume (5.8) when dealing with the minor arcs. Keeping the assumptions d > 3 and e > , we see in particular that P , Q > 1. Our plan is to produce two estimates for S(α, β): one for when α belongs to M(J + 1) \ M(J) and one for when β belongs to N(K + 1) \ N(K). Before proceeding further we note that and meas (M(J)) 6 q2J P0−d meas (N(K)) 6 q2KP0−d+1Q−1, (5.9) (5.10) for any integers J, K > 0. Suppose that n f (x) = ci1,...,idxi1 . . . xid, Xi1,...,id=1 with symmetric coefficients ci1,...,id ∈ Fq. Associated to f are the multilinear forms Ψi(x(1), . . . , x(d−1)) = d! n Xi1,...,id−1=1 ci1,...,id−1,ix(1) i1 . . . x(d−1) id−1 , (5.11) for 1 6 i 6 n. Our first estimate for S(α, β) involves summing trivially over h and then applying Weyl differencing d − 1 times to the sum over g. This eliminates the effect of the lower degree term βh.∇f (g) and leads one to a FREE RATIONAL CURVES AND THE CIRCLE METHOD 21 family of linear exponential sums with phase vectors (αΨ1(g), . . . , αΨn(g)), for g = (g1, . . . , gd−1) ∈ Fq[T ](d−1)n. This approach closely parallels [2]. An alternative estimate for S(α, β) is obtained by applying Weyl differencing d − 2 times to the sum over g. After a further application of Cauchy -- Schwarz one then brings the h-sum inside, giving a family of linear exponential sums with phase vectors (βΨ1(g), . . . , βΨn(g)), for g ∈ Fq[T ](d−1)n. This brings the Diophantine properties of β into play but extra difficulties arise from the fact that P and Q need not have the same degree. 5.1. Geometry-of-numbers redux. We shall need to begin by revisiting a function field lattice point counting result that played a key role in [2]. A lattice in Fq((T −1))N is a set of points of the form x = Λu where Λ is an N × N invertible matrix over Fq((T −1)) and u runs over elements of Fq[T ]n. Given a lattice Λ, the adjoint lattice is defined as the lattice associated to the inverse transpose matrix Λ−T . Remark 5.2. We can view lattices as vector bundles on P1 by viewing the matrix Λ as giving gluing data for gluing the trivial vector bundle on A1 and the trivial vector bundle on a formal neighborhood of ∞, using the Beauville -- Laszlo theorem. The adjoint lattice corresponds to the dual vector bundle, and the geometry-of-numbers computations in this section could instead be stated in this language. Bearing our notation in mind we have the following refinement of the shrink- ing lemma [2, Lemma 4.2]. Lemma 5.3. Let γ be a symmetric n×n matrix with entries in Fq((T −1)). Let a, c, s ∈ Z such that c > 0 and s > 0. Let Nγ,a,c be the number of x ∈ Fq[T ]n such that x < qa and kγxk < q−c. Then Nγ,a,c Nγ,a−s,c+s 6 qns+n max(⌊ a−c 2 ⌋,0). Proof. The bound is trivial when a 6 0 since then the left hand side is 1. Hence we may assume that a > 0 in what follows. It will be convenient to adopt the notation R = qR for any R ∈ R. Let 0 tc−aΛ−T a,c =(cid:18)tcIn −tcγ 0 In 0(cid:19) Λ(cid:18) 0 −In In 0(cid:19)−1 . so that We note that Λ−T tcγ Λa,c =(cid:18)t−aIn tcIn(cid:19) , a,c =(cid:18)taIn −taγ t−cIn(cid:19) . t−aIn(cid:19) =(cid:18) 0 −In 0 22 TIM BROWNING AND WILL SAWIN Let R1 6 . . . 6 R2n denote the successive minima of the lattice corresponding to Λa,c. Then qc−a/ R2n 6 . . . 6 qc−a/ R1 are the successive minima of the lattice corresponding to tc−aΛ−T a,c . Since the lattices are equal up to left and right multiplication by a matrix in GL2n(Fq), we must have Ri = qc−a/ R2n+1−i for all 1 6 i 6 2n. Taking i = n + 1 we deduce that q⌈ c−a 2 ⌉ 6 Rn+1. Now Nγ,a,c is simply the number of vectors in the lattice Λa,c of norm < 1, while Nγ,a−s,c−s is the number of norm < q−s. Hence, as established in Lee [10, Lemma 3.3.5], we have 2n 2n max(1, R−1 i ) and Nγ,a−s,c+s = max(1, q−s R−1 i ). Nγ,a,c = Yi=1 Yi=1 Dividing term by term, we see that each i contributes at most qs and each i > n + 1 contributes at most qmax(⌊ a−c 2 ⌋,0). Thus the total contribution is at most qns+n max(⌊ a−c (cid:3) 2 ⌋,0), as desired. For any α ∈ Fq((T −1)) and any r > 0, we set N(α; r) = #(cid:26)g ∈ Fq[T ](d−1)n : g1, . . . , gd−1 < P kαΨi(g)k < q−r (∀i 6 n) (cid:27) . (5.12) Furthermore, for an integer s > 0, we put Ns(α; r) = #(cid:26)g ∈ Fq[T ](d−1)n : g1, . . . , gd−1 < P /qs kαΨi(g)k < q−r−(d−1)s (∀i 6 n) (cid:27) . We can use the shrinking lemma to bound the ratio of these two quantities as follows. Lemma 5.4. For r > 0 and s > max(0, e − j + 1 − r), we have N(α, r) Ns(α, r) 6 q(d−1)ns+n max(0,⌊ e−j+1−r 2 ⌋). Proof. For each v ∈ {0, . . . , d − 1}, let N (v)(α, r) be the number of vectors g ∈ Fq[T ](d−1)n such that g1, . . . , gv < P /qs, gv+1, . . . , gd−1 < P (5.13) and kαΨi(g)k < q−r−v, for 1 6 i 6 n. Thus we have N (0)(α, r) = N(α, r) and N (d−1)(α, r) = Ns(α, r). FREE RATIONAL CURVES AND THE CIRCLE METHOD 23 Fix a choice of v ∈ {1, . . . , d−1} and let g1, . . . , gv−1, gv+1, . . . , gd−1 ∈ Fq[T ]n such that (5.13) holds. We consider the linear forms Li(g) = αΨi(g1, . . . , gv−1, g, gv+1, . . . , gd−1), for 1 6 i 6 n. These form an n × n matrix. Because Ψi is the dual- ization in one variable of a symmetric d-linear form, this n × n matrix is symmetric. The contribution to N (v−1)(α, r) from tuples with the chosen g1, . . . , gv−1, gv+1, . . . , gd−1 ∈ Fq[T ]n is Nγ,e−j+1,r+(v−1)s while the contribu- tion to N (v)(α, r) from tuples of the same form is Nγ,e−j+1−s,r+vs. Note that r + (v − 1)s > r > 0 for v > 1 and so Lemma 5.3 is applicable. We deduce that N (v−1)(α, r) N (v)(α, r) 6 qns+n max(⌊ e−j+1−r−(v−1)s 2 ⌋,0) for 1 6 v 6 d − 1. We take the product of this inequality over all v from 1 to d − 1. The first term in the exponent contributes (d − 1)ns. The second contributes n max(⌊ e−j+1−r ⌋, 0) for v = 1 and 0 for all other values of v, on assuming that s > e − j + 1 − r. Thus we get the stated bound. (cid:3) 2 5.2. Weyl differencing. Our fundamental tool for estimating S(α, β) is Weyl differencing. We recall first that P , Q > 1 in this exponential sum. Appeal- ing to [2, Eq. (5.2)] first, Weyl differencing d − 1 times gives S(α, β) 6 P nQn(cid:0)P −(d−1)nN(α, e − j + 1)(cid:1)1/2d−1 in the notation of (5.12). Note that as N(α, e − j + 1) > 1 and 2d−1 > (d − 1), the right side is > Qn. Thus we have , S(α, β) − qn(e−) 6 2P nQn(cid:0)P −(d−1)nN(α, e − j + 1)(cid:1)1/2d−1 We can also obtain an upper bound for S(α, β) that only uses information , (5.14) about β. Let us put so that T (h) = Xg<P ψ(αf (g) + βh.∇f (g)), with P, Q are as before. It follows from Cauchy -- Schwarz that S(α, β) = Xh<Q T (h), S(α, β)2d−2 6 Q(2d−2−1)n Xh<Q T (h)2d−2 . (5.15) 24 TIM BROWNING AND WILL SAWIN After d − 3 applications of Weyl differencing we obtain T (h)2d−3 6 P (2d−3−d+2)n Xg1,...,gd−3(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Xg ψ (D(g))(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) , where D(g) = Dg1,...,gd−3(αf (g) + βh.∇f (g)) and Dg1,...,gd−3 is the usual differ- encing operator. Here g1, . . . , gd−3, g each run over vectors in Fq[T ]n formed from polynomials of degree less than e − j + 1. A further application of Cauchy -- Schwarz now yields Differencing once more therefore leads to the expression T (h)2d−2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Xg where 2 ψ (D(g))(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Xg 6 P (2d−2−d+1)n Xg1,...,gd−3(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = Xgd−2,gd−1 2 . ψ (D(g))(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ψ(cid:0)Dg1,...,gd−2(αf (gd−1) + βh.∇f (gd−1)(cid:1) , Dg1,...,gd−2(h.∇f (gd−1) = hiΨi(g1, . . . , gd−1), n Xi=1 in the notation of (5.11). Returning to (5.15) we ignore the Diophantine approximation properties of α and instead execute the linear exponential sum over h. This leads to the expression in the notation of (5.12). Again, N(β, e − ) > 1 and 2d−2 > (d − 1) so the right side is > Qn, whence S(α, β) 6 P nQn(cid:0)P −(d−1)nN(β, e − )(cid:1)1/2d−2 S(α, β) − qn(e−) 6 2P nQn(cid:0)P −(d−1)nN(β, e − )(cid:1)1/2d−2 Remark 5.5. When char(Fq) 6 d the polynomials Ψi are identically zero for 1 6 i 6 n, so that (5.14) and (5.16) give nothing beyond the trivial bound for the exponential sum S(α, β). . (5.16) , Recall the definitions (5.4) and (5.6) of M(J) and N(K), respectively. We want to bound the size of S(α, β) when α 6∈ M(J) and β 6∈ N(K). To do this it will be convenient to introduce two parameters s1 and s2. Associated to these are the quantities l1 = e − j + 1 − s1 and l2 = e − j + 1 − s2. We can use our geometry-of-numbers shrinking result to establishing the fol- lowing pair of estimates. FREE RATIONAL CURVES AND THE CIRCLE METHOD 25 Lemma 5.6. Let α 6∈ M(J) and let l1 ∈ Z be such that l1 6 1 + J d − 1 and l1 6 e − j + 1. Then there exists a constant cd,n > 0 such that. N(α, e − j + 1) 6 cd,nq−nl1P (d−1)n. Proof. It follows from Lemma 5.4 that N(α, e − j + 1) is at most q(d−1)ns1#(cid:26)g ∈ Fq[T ](d−1)n : g1, . . . , gd−1 < P /qs1 kαΨi(g)k < P −1q−(d−1)s1 (∀i 6 n) (cid:27) , for any s1 > 0. Note that P /qs1 = ql1 and qs1 = P0/ql1−1. Suppose that g1, . . . , gd−1 < P /qs1 and kαΨi(g)k < P −1q−(d−1)s1 but Ψi(g) 6= 0. Let r = Ψi(g) and let a be the integer part of αΨi(g), each divided through by any common factors that they might share. Then r 6 q(d−1)(l1−1) and rα − a < P −1q−(d−1)s1 = q(d−1)(l1−1)−1P −d 0 . This contradicts the assumption that α 6∈ M(J), if J > (d − 1)(l1 − 1). Hence, if J > (d − 1)(l1 − 1) and α 6∈ M(J), we have N(α, e − j + 1) 6 q(d−1)ns1#(cid:26)g ∈ Fq[T ](d−1)n : g1, . . . , gd−1 < ql1 Ψi(g) = 0 (∀i 6 n) (cid:27) . The statement of the lemma follows on noting that the remaining cardinality is O(q(d−2)nl1) for dimensionality reasons, where the implied constant depends only on d and n. (cid:3) Lemma 5.7. Let β 6∈ N(K) and let l2 ∈ Z be such that l2 6 1 + K d − 1 and l2 6 e − j + 1 − max(0, − j + 1). Then there exists a constant cd,n > 0 such that. N(β, e − ) 6 cd,nq−nl2+n max(0,⌊ −j+1 2 ⌋)P (d−1)n. Proof. This time we take r = e − in Lemma 5.4 and deduce that N(β, e − ) 6 q(d−1)ns2+n max(0,⌊ −j+1 ⌋) 2 × #(cid:26)g ∈ Fq[T ](d−1)n : g1, . . . , gd−1 < ql2 kβΨi(g)k < Q−1q−(d−1)s2 (∀i 6 n) (cid:27) , for any s2 > max(0, − j + 1). Arguing as in the previous result it is simple to check that we must in fact have Ψi(g) = 0 for all 1 6 i 6 n whenever β 6∈ N(K) 26 TIM BROWNING AND WILL SAWIN and K > (d − 1)(l2 − 1). But then there are O(q(d−2)nl2) possible vectors g ∈ Fq[T ](d−1)n that contribute. The statement of the lemma follows. (cid:3) In our work we shall take l1 = 1 +(cid:22) J d − 1(cid:23) , l2 = 1 +(cid:22) K d − 1(cid:23) . (5.17) We need to check that the remaining conditions on l1 and l2 are satisfied in Lemmas 5.6 and 5.7. To begin with we note that J 6(cid:24)d(e − j) 2 (cid:25) − 1 6 d(e − j) 2 − 1 2 Hence for Lemma 5.6 to be applicable it suffices to have d(e − j) − 1 6 2(d − 1)(e − j). But this is equivalent to 0 6 1 + (d − 2)(e − j) which follows from (5.8). Next, we note that K 6(cid:24)(e − j)(d − 1) + e − 2 so that Lemma 5.7 is applicable if (cid:25) − 1 6 (e − j)(d − 1) + e − 2 − 1 2 , e − − 1 6 (d − 1)(e − j − 2 max(0, − j + 1)). Thus it suffices to have and e − − 1 6 (d − 1)(e − j) (5.18) (5.19) However, (5.18) follows from (5.8), so it suffices to assume that (5.19) holds. Inserting Lemmas 5.6 and 5.7 into our Weyl differencing bounds (5.14) and e − − 1 6 (d − 1)(e + j − 2 − 2). (5.16), we deduce that there exists a constant cd,n > 0 such that S(α, β) − qn(e−) 6 cd,nP nQn min(cid:16)q−nl1/2d−1 , q−n(l2−n max(0,⌊ −j+1 2 ⌋))/2d−2(cid:17) = cd,nP nQn/ max(cid:16)ql1, q2l2−2 max(0,⌊ −j+1 2 ⌋)(cid:17)n/2d−1 , whenever (α, β) ∈ M(J + 1) \ M(J) × N(K + 1) \ N(K) and (5.19) holds. We shall proceed under the assumption that the parameter l2 satisfies l2 − max(cid:18)0,(cid:22) − j + 1 2 (cid:23)(cid:19) > 0. (5.20) This is precisely the circumstance under which our β-treatment is non-trivial. Assume that n > (d − 1)2d, so that 2d(d − 1)/n < 1. If (5.20) holds we can FREE RATIONAL CURVES AND THE CIRCLE METHOD 27 invoke the inequality max(A, B) > A2d(d−1)/nB1−2d(d−1)/n, which is valid for any A, B > 1. Thus it follows that S(α, β) − qn(e−) 6 cd,n q2(d−1)(2l2−l1)−4(d−1) max(0,⌊ −j+1 ⌋))n/2d−2 q(l2−max(0,⌊ −j+1 2 2 ⌋)P nQn . Returning to (5.2) we see that Znj Xg∈Fq[T ]n 0<g<qe−j+1 f (g)=0 S(β)dβ 6 M −1 XJ=−1 N −1 XK=e− E(J, K) where we recall from (5.5) and (5.7) that M =(cid:24)d(e − j) 2 (cid:25) , N =(cid:24)(e − j)(d − 1) + e − 2 (cid:25) , and E(J, K) =ZM(J+1)\M(J)ZN(K+1)\N(K) S(α, β) − qn(e−)dαdβ. The measure of all (α, β) in the integral is at most q4+2J+2KP0−2d+1Q−1, by (5.9) and (5.10). Let us consider the total contribution (d−1)l1−1 (d−1)l2−1 El1,l2 = XJ=max((d−1)(l1−1),−1) XK=max((d−1)(l2−1),e−) EJ,K, from J, K associated to integers l1 > 0 and l2 > 1 via (5.17). Then El1,l2 ≪ q6(d−1)l2 P nQn−1P0−2d+1q−4(d−1) max(0,⌊ −j+1 2 ⌋) q(l2−max(0,⌊ −j+1 2 ⌋))n/2d−2 = q∆j−l2(n/2d−2−6(d−1))+max(0,⌊ −j+1 2 ⌋)(n/2d−2−4(d−1)), where we have put ∆j = (e − j)(n − 2d + 1) + (e − )(n − 1) + n. Because K > e − , we have In particular our condition (5.20) is satisfied when d − 1(cid:23) . l2 > 1 +(cid:22) e − 1 +(cid:22) e − d − 1(cid:23) > max(cid:18)0,(cid:22) − j + 1 2 (cid:23)(cid:19) . (5.21) 28 TIM BROWNING AND WILL SAWIN Furthermore, assuming n > 3(d − 1)2d−1, the bound is decreasing in l2, so the dominant contribution occurs when Since there are O(e) choices for l1, our work has therefore shown that l2 = 1 +(cid:22) e − d − 1(cid:23) . Znj S(β)dβ ≪ eq∆j −Γj , Xg∈Fq[T ]n 0<g<qe−j+1 where f (g)=0 d − 1(cid:23)(cid:19) 2d−2 − 6(d − 1)(cid:17)(cid:18)1 +(cid:22) e − Γj = (cid:16) n 2d−2 − 4(d − 1)(cid:17) max(cid:18)0,(cid:22) − j + 1 −(cid:16) n 2d−2 − 6(d − 1)(cid:17)(cid:18)1 +(cid:22) e − = (cid:16) n − 2(d − 1) max(cid:18)0,(cid:22) − j + 1 (cid:23)(cid:19) . 2 2 (cid:23)(cid:19) d − 1(cid:23) − max(cid:18)0,(cid:22) − j + 1 2 (cid:23)(cid:19)(cid:19) Thus we certainly require (5.21) to hold in order to expect any saving in our minor arc estimate. We summarise our argument in the following result. Lemma 5.8. Let d > 3 and n > 3(d − 1)2d−1. Assume that > −1 and e > max(cid:18) + (d − 1)(cid:22) + 1 2 (cid:23) , ( + 1)(cid:18)2 + 1 d − 2(cid:19)(cid:19) . (5.22) N minor (q, e, f ) ≪ eq∆0−Γ0 ∆0 = 2e(n − d) − (n − 1) + n Then where and Γ0 =(cid:16) n 2d−2 − 6(d − 1)(cid:17)(cid:18)1 +(cid:22) e − d − 1(cid:23) −(cid:22) + 1 2 (cid:23)(cid:19) − 2(d − 1)(cid:22) + 1 2 (cid:23) . Proof. Recall (5.1) and note that ∆j = ∆0 − j(n − 2d + 1). Hence for the range of n in which we are interested we deduce from (4.2) that cjq∆j−Γj ≪ q∆0−Γ0, for all j > 0. Moreover, Γ0 takes the value recorded in the statement of the lemma when > −1 and the condition (5.22) on e is enough to ensure that FREE RATIONAL CURVES AND THE CIRCLE METHOD 29 (5.19) and (5.21) both hold for every j ∈ {0, 1, 2}. (For (5.19) we note that it suffices to have e(d − 2) > ( + 1)(2d − 3).) This completes the proof. (cid:3) 5.3. Deduction of Theorem 1.1. We assume that n > (2d − 1)2d−1. We revisit the argument deployed in [1] to establish the irreducibility and dimen- sion of M0,0(X, e). This is based on a counting argument over a finite field Fq whose characteristic is greater than the degree d of the non-singular form f ∈ Fq[x1, . . . , xn] that defines X. According to [1, Eq. (3.3)], in order to deduce that M0,0(X, e) is irreducible and of the expected dimension it suffices to show that q−(n−d)e−n+1 N (q, e, f ) 6 1, lim q→∞ (5.23) where N (q, e, f ) is the number of g ∈ Fq[T ]n such that g < qe+1 and f (g) = 0. We have where N (q, e, f ) =ZT SBV(α)dα, SBV(α) = Xg∈Fq[T ]n g<qe+1 ψ(αf (g)) = q−n(e−)S(α, 0), in the notation of (5.3), with j = 0. Take j = 0 in the major arcs M(J) that were defined in (5.4). A straightforward calculation shows that the contribu- tion from the major arc around 0 is ZM(0) SBV(α)dα = Xg∈Fq[T ]n g<qe+1 Zθ<q−de ψ(θf (g))dθ = qne−de(cid:0)qn−1 + O(qn/2)(cid:1) . In order to complete the proof of (5.23) it therefore suffices to show that q−(n−d)e−n+1 lim q→∞ SBV(α)dα < 1 M −1 XJ=0 ZM(J+1)\M(J) where M = ⌈ de Thus it follows from (5.14) and Lemma 5.6 that 2 ⌉ is given by (5.5). To do this we may apply our previous work. SBV(α) ≪ P nq−nl1/2d−1 , if α 6∈ M(J) and l1 is any integer such that l1 6 1 + J/(d − 1) and l1 6 e + 1. The choice l1 = 1+⌊J/(d−1)⌋ is acceptable since J 6 ⌈ de , whence 2 ⌉−1 6 de−1 2 l1 6 1 + de − 1 2(d − 1) = 1 + e, 30 TIM BROWNING AND WILL SAWIN for d > 3. Since J > 0 we are clearly only interested in integers l1 > 1. Appealing to (5.9) to estimate the volume of M(J + 1), we deduce that for given l1 > 1 the total associated contribution is (d−1)l1−1 XJ=(d−1)(l1−1)ZM(J+1)\M(J) (d−1)l1−1 XJ=(d−1)(l1−1) SBV(α)dα ≪ q2J+2−de.P nq−nl1/2d−1 ≪ q−de+n(e+1)+(2(d−1)−n/2d−1)l1. This is decreasing with l1 if n > (d − 1)2d and we may therefore sum over l1 > 1 to finally deduce that q−(n−d)e−n+1 SBV(α)dα ≪ q1+2(d−1)−n/2d−1 . M −1 XJ=0 ZM(J+1)\M(J) The exponent of q is negative if n > (2d − 1)2d−1, which thereby concludes the proof of (5.23), whence M0,0(X, e) is indeed irreducible and of the expected dimension. It follows from the same method used in [6, p. 2] that M0,0(X, e) is locally a complete intersection. Indeed, since M0,0(X, e) is locally the intersec- tion of de+1 equations in M0,0(Pn−1, e), a smooth stack of dimension ne−4, it is a locally complete intersection if and only if its dimension is (n − d)e + n − 5. 5.4. Deduction of Theorem 1.5. Assume that d > 3, n > 3(d − 1)2d−1, which is needed for Lemma 5.1. In view of Theorem 1.1, the stated bound is trivial unless 1 + ⌊ e− > −1, and e > ( + 1)(cid:0)2 + 1 d−1 ⌋ −(cid:4) +1 and thus e > + (d − 1)(cid:4) +1 d−2(cid:1). In particular, this implies that e > , 2 (cid:5) > 0 d−1⌋ −(cid:4) +1 2 (cid:5) > 0, so we may assume that ⌊ e− 2 (cid:5). Hence we may assume that (5.22) holds. Combining Lemmas 5.1 and 5.8 we deduce that #Z(Fq) ≪ eqe(n−d)+n−5−min(µ1(n),µ2(n)), (5.24) with and µ1(n) =(cid:16) n 2d−2 − 6(d − 1)(cid:17)(cid:18)1 + D −(cid:22) + 1 2 (cid:23)(cid:19) − 2(d − 1)(cid:22) + 1 2 (cid:23) µ2(n) = (1 + D)(n − d − 1) − de + e + 1. Here we recall that D is given by (4.7) as ⌊ e− d−1 ⌋. We claim that µ1(n) 6 µ2(n). They are both increasing affine functions of n, with µ1(n) of lesser slope than µ2(n). Hence to check that µ1(n) is the minimum, it suffices to check that µ2(n) > 0 and µ1(n) 6 0 when n = 3(d − 1)2d−1. In other words, we must show that 3(d − 1)2d−1 > d + 1 + e(d − 1) − 1 1 + D . FREE RATIONAL CURVES AND THE CIRCLE METHOD 31 To do this, observe that because e > + (d − 1)(cid:4) +1 2 (cid:5) we have e > d+1 that 2 , so > e + 1 − 2 d − 1 d+1e > e . d + 1 1 + D > Thus e + 1 − d − 1 d + 1 + e(d − 1) − 1 1 + D 6 d + 1 + e(d − 1) e/(d + 1) = d(d + 1), so it suffices to check 3(d − 1)2d−1 > d(d + 1). But it is clear that this holds for all d > 3, whence µ2(n) > µ1(n). By Lang -- Weil [9], it now follows from (5.24) that dim Z 6 e(n − d) + n − 5 − µ1(n) for any smooth hypersurface defined over a finite field. For a general hypersur- face, we can spread it out to a family defined over a ring finitely-generated over Z. The dimension of Z in this family is manifestly constant on some open subset of the spectrum of this ring, which must contain a finite-field valued point, so dim Z is at most e(n − d) + n − 5 − µ1(n) for the generic point and thus for the original hypersurface. This completes the proof of Theorem 1.5. 5.5. Deduction of Theorem 1.2. We consider the effect of taking = −1 in Theorem 1.5. Clearly (1.1) is equivalent to e > 0 and can be ignored. Note that Z−1 contains the singular locus of M0,0(X, e) by [3, Thm. 2.6]. Thus the codi- mension of the singular locus is at least dim M0,0(X, e)−dim Z−1. Theorem 1.2 therefore follows from applying Theorem 1.1 to calculate dim M0,0(X, e) and Theorem 1.5 to bound dim Z−1. Because the lower bound for the codimension of the singular locus is strictly positive, the moduli space is generically smooth. Any generically smooth lo- cally complete intersection scheme is reduced, which thereby completes the proof of Theorem 1.1. 6. Peyre's freedom counting function In this section we prove the asymptotic formula in Theorem 1.6 for the counting function (1.4), by piecing together our work above and the main results in Lee's thesis [10]. We have N ε-free X (B) = NX(B) − Eε(B), where Eε(B) counts the number of x ∈ X(Fq(T )) with Hω−1 that ℓ(x) < ε. V (6.1) (x) 6 qB such 32 TIM BROWNING AND WILL SAWIN Let us begin by studying NX (B). As usual we suppose that X is defined by a non-singular form f ∈ Fq[x1, . . . , xn] of degree d > 3. It follows from the proof of part (1) of Proposition 3.8 that NX(B) = 1 q − 1 #(cid:26)g ∈ Fq[T ]n : gcd(g1, . . . , gn) = 1 gn−d < qB+1, f (g) = 0 (cid:27) . Using the Mobius function to detect the coprimality condition we obtain 1 NX(B) = = k monic q − 1 Xk∈Fq[T ] q − 1Xj>0 Xk∈Fq[T ] 1 µ(k)#(cid:8)g ∈ Fq[T ]n : 0 < kgn−d < qB+1, f (g) = 0(cid:9) (cid:27) . µ(k)#(cid:26)g ∈ Fq[T ]n : 0 < gn−d < qB+1−j(n−d) f (g) = 0 k=qj k monic Put m = n − (d − 1)2d and assume that m > 0. Then, on appealing to Lee's thesis [10, Thm. 4.1.1], it follows that #(cid:8)g ∈ Fq[T ]n : 0 < gn−d < qR+1, f (g) = 0(cid:9) = qR(cid:16)cf + O(q−mR/(2d+1(d−1)(n−d)))(cid:17) , (6.2) for any R > 0, where cf is the usual product of singular series and singular integral. Using (4.1) to handle the sum over j and k, it now follows from (6.2) that there exists δ > 0 such that NX (B) = cf (q − 1)ζFq(T )(n − d) qB + O(cid:0)q(1−δ)B(cid:1) , where ζFq(T )(s) = (1 − q1−s)−1 is the rational zeta function. Arguing along standard lines (as in Peyre [12, §5.4], for example), one readily confirms that this agrees with the Batyrev -- Manin -- Peyre prediction for the hypersurface X. It remains to produce an upper bound for the quantity Eε(B) in (6.1). Let x ∈ X(Fq(T )) and suppose that it defines a map c : P1 → X of degree e. Then it follows from [13, Notation 5.7] that ℓ(x) = (n − 1) e(n − d) if and only if c is -free but not ( + 1)-free. (In particular, Remark 1.4 implies that ℓ(x) ∈ [0, 1].) We deduce that Eε(B) is at most the number of rational maps from P1 → X with degree at most B/(n − d) which are not -free, with =(cid:22) εB n − 1(cid:23) + 2. (6.3) FREE RATIONAL CURVES AND THE CIRCLE METHOD 33 We may therefore appeal to the proof of Proposition 3.8(2) to estimate this quantity, finding that Eε(B) 6 N(q, B/(n − d), f )q(n−1)−B − N(q, B/(n − d), f ) (q − 1)2 , with given by (6.3). In what follows it will be convenient to set e = B/(n−d) and to assume that e ∈ N. All of the implied constants that follow are allowed to depend on q and f , but not on e or . We seek conditions on n and under which we can deduce that there exists δ > 0 such that Eε(B) = O(q(1−δ)e(n−d)). First we improve our treatment of Lemma 4.4 slightly by revisiting the argument (4.10). Since we no longer care about a dependence on the finite field, rather than invoking a trivial bound we may apply (6.2) to deduce that N(q, u − ℓ, f ) ≪ q(u−ℓ)(n−d) if n > (d − 1)2d. But then (4.10) can be replaced by the bound Ej ≪f X06u6e−j qu(n−2d+1) Xℓ>D+1+j−e+u ≪f q(e−j)(n−2d+1)−(D+1)(n−2d), q−ℓ(n−2d) where D is given by (4.7), whence q(n−1)(e−)Xj>0 cjEj ≪f q2e(n−d)−(n−1)−(D+1)(n−2d) ≪f q2e(n−d)−(n−1)−(e−)(n−2d)/(d−1). It now follows from (4.3) and our modified version of Lemma 4.4 that Eε(B) ≪ qe(n−d)−(e−)(n−2d)/(d−1) + q−e(n−d)+(n−1)N minor (q, e, f ), provided that e > . Note that Γ0 = γ0 + Od,n(1), with γ0 =(cid:16) n 2d−2 − 6(d − 1)(cid:17)(cid:18) e − d − 1 Appealing now to Lemma 5.8 we therefore deduce that − 2(cid:19) − (d − 1). Eε(B) ≪ qe(n−d)−(e−)(n−2d)/(d−1) + eqe(n−d)−γ0 if (5.22) holds. Recall that n > 3(d − 1)2d−1. Then n/2d−2 − 6(d − 1) > 2−d+2 and we can ensure that γ0 > δe for a small parameter δ > 0 (that depends only on d) provided that (6.4) This is also enough to ensure that (e−)(n−2d)/(d−1) > δe. This inequality is clearly much stronger than (5.22). The statement of Theorem 1.6 now follows on taking e = B/(n − d) and noting that the hypothesis on ε in the theorem is e > (d − 1)22d−1. 34 TIM BROWNING AND WILL SAWIN enough to ensure that (6.4) holds when is given by (6.3) and B is sufficiently large. References [1] T.D. Browning and P. Vishe, Cubic hypersurfaces over Fq(t). Geom. Funct. Anal. 25 (2015), 671 -- 732. [2] T.D. Browning and P. Vishe, Rational curves on smooth hypersurfaces of low degree. Algebra & Number Theory 11 (2017), 1657 -- 1675. [3] O. Debarre, Higher-dimensional algebraic geometry. Springer-Verlag, 2001. [4] D. Eisenbud and J. Harris, 3264 and all that. Cambridge University Press, 2016. [5] J. Franke, Y.I. Manin and Y. Tschinkel, Rational points of bounded height on Fano varieties. Invent. Math. 95 (1989), 421 -- 435. [6] J. Harris, M. Roth and J. Starr, Rational curves on hypersurfaces of low degree. J. reine angew. Math. 571 (2004), 73 -- 106. [7] R. Hartshorne, Algebraic geometry. Springer-Verlag, 1977. [8] J. Koll´ar, Y. Miyaoka and S. Mori, Rationally connected varieties. J. Algebraic Geom. 1 (1992), 429 -- 448. [9] S. Lang and A. Weil, Number of points of varieties in finite fields. Amer. J. Math. 76 (1954), 819 -- 827. [10] S.A. Lee, On the applications of the circle method to function fields, and related topics. Ph.D. thesis, University of Bristol, 2013. [11] M. Mustata, Jet schemes of locally complete intersection canonical singularities. In- vent. Math. 145 (2001), 397 -- 424. [12] E. Peyre, Hauteurs et nombres de Tamagawa sur les vari´et´es de Fano. Duke Math. J. 79 (1995), 101 -- 218. [13] E. Peyre, Libert´e et accumulation. Documenta Math. 22 (2017) 1615 -- 1659. [14] E. Riedl and D. Yang, Kontsevich spaces of rational curves on Fano hypersurfaces. J. reine angew. Math., to appear. [15] J. Starr, The Kodaira dimension of spaces of rational curves on low degree hypersur- faces. (arXiv:math/0305432) IST Austria, Am Campus 1, 3400 Klosterneuburg, Austria E-mail address: [email protected] Columbia University, Department of Mathematics, 2990 Broadway, New York, NY 10027, USA E-mail address: [email protected]
1311.0502
3
1311
2017-09-08T07:58:48
Skeleta in non-Archimedean and tropical geometry
[ "math.AG" ]
I describe an algebro-geometric theory of skeleta, which provides a unified setting for the study of tropical varieties, skeleta of non-Archimedean analytic spaces, and affine manifolds with singularities. Skeleta are spaces equipped with a structure sheaf of topological semirings, and are locally modelled on the spectra of the same. The primary result of this paper is that the topological space underlying a non-Archimedean analytic space may locally be recovered from the sheaf of `pointwise valuations' of its analytic functions.
math.AG
math
Skeleta in non-Archimedean and tropical geometry Andrew W. Macpherson October 31, 2018 Abstract I describe an algebro-geometric theory of skeleta, which provides a unified setting for the study of tropical varieties, skeleta of non-Archimedean analytic spaces, and affine manifolds with singularities. Skeleta are spaces equipped with a structure sheaf of topological semirings, and are locally modelled on the spectra of the same. The primary result of this paper is that the topological space X underlying a non-Archimedean an- alytic space may locally be recovered from the sheaf OX of pointwise valuations of its analytic functions; in other words, (X, OX ) is a skeleton. Contents 1 Introduction 1.1 How to read this paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Preliminaries and conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 On sites and topoi 2.2 Non-Archimedean geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Subobjects and B-modules 3.1 B-modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Orders and lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Finiteness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Noetherian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Adjunction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Topological lattices 4.1 Topological modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Semirings 5.1 The tensor sum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Action by contraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Projective tensor sum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Localisation 6.1 Bounded localisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Cellular localisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Prime spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4 Blow-up formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 7 8 8 9 10 11 13 14 17 17 19 21 22 23 27 30 33 34 36 38 39 7 Skeleta 7.1 Spectrum of a semiring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Integral skeleta and cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 Universal skeleton of a formal scheme . . . . . . . . . . . . . . . . . . . . . . . . 7.4 Universal skeleton of an adic space . . . . . . . . . . . . . . . . . . . . . . . . . . 7.5 Shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Examples & applications 8.1 Polytopes and fans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Dual intersection skeleta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3 Tropicalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4 Circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 42 44 47 48 50 51 51 54 56 58 1 Introduction There are several areas in modern geometry in which one is led to consider spaces with affine or piecewise affine structure. The three with which I am in particular concerned are, in order of increasing subtlety: • skeleta of non-Archimedean analytic spaces ([Ber99]); • tropical geometry ([Mik06],[BPR11]); • affine manifolds with singularities ([Gro11],[KS06]). These cases share the following features: • they are piecewise manifolds; • it makes sense to ask which continuous, real-valued functions are piecewise affine; • they admit a stratification on which it makes sense to ask which of these are convex. Moreover, in each case the structure is determined entirely by an underlying space B, together with a sheaf OB ⊆ C0 (B; R ⊔ {−∞}) of piecewise-affine, convex (where this is defined) functions. The sheaf OB is naturally a sheaf of idempotent semirings under the operations of pointwise maximum and addition. It has long been understood, at least in the tropical ge- ometry community (cf. e.g. [Mik06]) that such semirings are the correct algebraic structures to associate to piecewise-affine geometries like B. A natural question to ask is whether this sheaf-theoretic language can be pushed further in this setting and, as in algebraic geometry, the underlying space B recovered from the semirings of local sections of OB. In this paper, I provide an affirmative answer to this question, though, as for the passage from classical algebraic geometry to scheme theory, it will require us to alter our expectations of what type of space underlies a piecewise-affine geometry. The resulting theory is what I call the theory of skeleta. The relationship between the theories of schemes and of skeleta goes beyond mere anal- ogy: they can in fact be couched within the same theoretical framework (appendix 2.1), à 2 la Grothendieck (cf. also [TV09], [Dur07]). Within this framework, one need only specify which semiring homomorphisms Γ(U; OU ) → Γ(V , OV ) are dual to open immersions V ,→ U of skeleta. This is enough to associate to every semiring α a quasi-compact topological space, its spectrum Spec α. Skeleta can then be defined to be those semiringed spaces locally modelled by the spectra of semirings. My main contention in this paper is that the primary source of skeleta is the non- Archimedean geometry, and this is why I have adopted the terminology of this field. The initial concept that links non-Archimedean and piecewise-affine geometry is that of a val- uation. Indeed, semirings are the natural recipients of valuations, while topological rings are the sources. The topology of skeleta is selected so as to ensure that there is a unique functor sk : Ad → Sk from the category Ad of adic spaces to the category Sk of skeleta, a natural homeomorphism Xf→skX for X ∈ Ad, and a universal valuation OX → OX . This universal skeleton skX of an analytic space X can be thought of as the skeleton whose functions are the pointwise logarithmic norms of analytic functions on X . In particular, X is locally the spectrum of the semiring of these functions. The existence of this functor is the primary result of this paper. I also recover within the category of skeleta certain further examples that already existed in the literature: the dual intersection or Clemens complex of a degeneration (§8.2), and the tropicalisation of a subvariety of a toric variety (§8.3). Gist The categories of skeleta (section 7) and of non-Archimedean analytic spaces may be con- structed in the same way: as a category of locally representable sheaves on some site whose underlying category is opposite to a category of algebras (à la [TV09]). As such, to build a bridge between the two categories, it is enough to build a bridge between the categories 1 2 Ringt of topological semirings (defs. 5.2, 5.23) and nA of non-Archimedean rings (def. 2.2), and to check that it satisfies certain compatibility conditions. One can associate to any non-Archimedean ring (A, A+) a free semiring Bc(A; A+), which, as a partially ordered set, is the set of finitely generated A+-submodules of A. The addition on Bc(A; A+) comes from the multiplication on A. It comes with a valuation A → B c(A; A+), f 7→ (f ) universal among continuous semivaluations of A into a semiring whose values on A+ are negative (or zero). In other words, Bc(A; A+) corepresents the functor Val(A, A+; −) : 1 2 Ringt → Set which takes a topological semiring α to the set of continuous semivaluations val : A → α satisfying val A+ ≤ 0. 3 In particular, if A = A+, then Bc A+ := Bc(A+; A+) is the set of finitely generated ideals of A+, or of finitely presented subschemes of Spf A+. Everything in the above paragraph may also be phrased in the internal logic of topoi so that, for example, it makes sense to replace A and α with sheaves of non-Archimedean rings and semirings on a space. Thus if X is a non-Archimedean analytic space, then OX : U 7→ B c(OU; O + U) is a sheaf of topological semirings on X , universal among those receiving a continuous semivaluation from OX . Theorem (7.21). Let X be quasi-compact and quasi-separated. There is a natural homeo- morphism c(OX ; O + X ) Xf→Spec B which matches the structure sheaf on the right with OX on the left. In particular, if X is a qcqs formal scheme, then the spectrum of the semiring Bc OX of ideal sheaves on X is naturally homeomorphic to X itself. This skeleton Spec Bc(OX ; O + X ) is called the universal skeleton skX of X . It follows from the universal property of its structure sheaf that the real points skX (R∨) can be identified canonically with the Berkovich analytic space associated to X [Ber93, §1.6], provided such a thing exists; see theorem 7.25. A natural geometric counterpart to the universality of Bc might be to say that the uni- versal skeleton of an analytic space is universal among skeleta B equipped with a contin- uous map ι : B → X and valuation OX → ι∗OB. However, my point of view is that the very construction of the universal semiring diminishes the importance of valuation theory in getting a handle on the geometry of X . It tends to be easier, and perhaps more natural, to construct skeleta B with a morphism X → B in the opposite direction. For example, let X + → Spf OK be a simple normal crossings degeneration over a DVR OK , with general fibre j : X → X + (so X is an analytic space, smooth over K). The ir- reducible components E i of the central fibre X + 0 of the degeneration generate a subring Osk(X,X +)◦ ⊆ Bc OX + whose elements are the ideals monomial with respect to normal co- 0 . Base-changing over K yields i ). Their supports are the strata of X + i=1 xni ordinates (t =Qk the dual intersection semiring Cl(X , X +) ,→ B c(OX + ⊗ K; OX +) and, dually, dual intersection skeleton SpecCl(X , X +) =: ∆(X , X +) µ ← X (definition 8.3). That X is defined over K means that the universal skeleton, dual intersection skeleton, and morphism µ are defined over its value group: the semifield of integers Z∨ := Z ⊔ {−∞}. The real points sk(X , X +)(R∨) of the dual intersection skeleton are Z∨-semialgebra homo- morphisms Osk(X,X +) → R∨ to the real semifield R∨ =:= R ⊔ {−∞}. They can be identified with the points of the naïve dual intersection complex as defined in, for example, [KS06, §A.3]. Indeed, the simplices of this complex are defined by the logarithms of local equations for the intersections of X + 0 : K{x1,..., xn} i=1 xni i ) (t =Qk Z∨{X1,..., X n} i=1 ni X i) (−1 =Pk 4 , where the curly braces on the right-hand side signify that X i ≤ 0. The latter equation i=1 ni X i = 0 cuts the dual intersection simplex 1 +Pk conv{(0,...,0, −1/ni,0,...,0)}k i=1 ⊂ R n ≤0 out of the negative orthant in Rn. Under this identification, the elements of the dual inter- section semiring correspond to integral, piecewise-affine functions whose restriction to each cell is convex. The construction of such skeleta, perhaps partial skeleta of X , is the crux of the theory. At this point I know of only a few examples (§8). An elliptic curve ks, which I denote {Di}n Let us consider now the case that X + = E+/OK is an elliptic curve degenerating semistably to a cycle of n > 3 P1 i=1 ∈ Z∨{X ; X +}. Its general fibre E/K is a Tate elliptic curve. The dual intersection skeleton ∆(E, E+) is, at the level of real points, a cycle of n unit intervals joined at their endpoints. The vertices {vi}n i=1 correspond to the lines Di. Functions are allowed to be concave at these vertices. In particular, the function Di takes the value -1 at vi and zero at the other vertices. Now let us collapse one of the Dis pi : E+ → E+ i ks meeting transver- to an A1 singularity. The special fibre of E+ sally except at the discriminant locus of the blow-up, which now has the local equation (xy − t2). With these co-ordinates, pi is the blow-up of the ideal (x, y, t). i is now a cycle of (n − 1) P1 In the semialgebraic notation, the ideal is (x, t, y) = D′ i−1 ∨ −1 ∨ D′ i+1 ∈ Cl(E, E+), j denotes the divisor whose strict transform under pi is D j, so p∗ where D′ i D′ The blow-up is monomial, and hence induces a pullback homomorphism p∗ Cl(E, E+), and dually, a morphism i±1 = Di±1 + Di. i : Cl(E, E+) → of the dual intersection skeleta. pi : ∆(E, E+) → ∆(E, E+ i ) The segment of the dual intersection complex corresponding to the singular intersection Di−1 ∩ Di+1 is an interval I of affine length two. Considered as a function on I, the blow-up ideal D′ i+1 has real values as the absolute value i−1 ∨ −1 ∨ D′ − : I ≃ [−1,1] → R. It has a kink in the middle. Because, in ∆(E, E+ function is not allowed; while of course the pullback Di is invertible on ∆(E, E+). i ), there is no vertex here, the inverse of this In fact, Cl(E, E+) is a localisation of Cl(E, E+ i ) at D′ i−1 ∨ −1 ∨ D′ i+1, and ∆(E, E+) is an open subset of ∆(E, E+ i ). 5 Varying i, we obtain therefore an atlas ∆(E, E+) ⇒ nai6= j=1 ∆(E, E+ i ) nai=1 for a skeleton B which compactifies ∆(E, E+). Functions on B are required to be convex everywhere, and B(R∨) is, as an affine manifold, the flat circle R/nZ. This skeleton is a kind of Calabi-Yau skeleton of E, and it depends only on the intrinsic geometry of E and not on any choices of model. See also section 8.4. Mirror symmetry context Conjectures arising from homological mirror symmetry [KS00] suggest that a Calabi-Yau n- fold X approaching a so-called large complex structure limit acquires the structure of a com- pletely integrable system µ : X → B with singularities in (real) codimension one, shrinking to two in the limit. The base B therefore acquires the structure of a Riemannian n-manifold with integral affine co-ordinates yi, away from the singular fibres, given by the Hamiltoni- ans of the system. The metric is locally the Hessian, with respect to these co-ordinates, of a convex function K, and satisfies the Monge-Ampère equation d detµ ∂2K ∂yi∂yj¶ = 0 which can be thought of as the 'tropicalisation' of the complex Monge-Ampère equation satisfied by the Yau metric. The central idea of [KS06] is that the skeleton B can be constructed, with the Legen- dre dual affine structure yi, from the non-Archimedean geometry of X an or, what is the same thing, the birational geometry of its formal models. Indeed, Kontsevich noted that the Gromov-Hausdorff limit of X should resemble the dual intersection complex of a cer- tain 'crepant' model thereof. To be precise, the real points of B should be embeddable into X an(R∨) as the dual intersection complex of any dlt minimal model of X [NX13]. Its struc- ture as a dual intersection complex also endows it with the correct affine structure, away from a subset of codimension one which contains the singularities. More subtle is to construct the correct non-Archimedean torus fibration µ : X → B. This would also determine the affine structure of B in the sense that OB ∼= Im(µ∗OX → µ∗OX ). Such a µ is determined by a choice of minimal model. Unfortunately, in dimensions greater than one, the morphisms µ coming from various models differ. The affine structures they induce are related by so-called worm deformations, which move the singularities of the affine structure along their monodromy-invariant lines. These deformations correspond to flops in birational geometry. This forms the basis of a dictionary, motivated by mirror symmetry, between concepts in birational geometry and the tropical geometry of affine manifolds. This dictionary has been 6 partially developed along combinatorial lines in the Gross-Siebert programme.1 However, geometrically interesting examples present enormous combinatorial complexity, already for the case of K3 surfaces. I propose that a more geometric approach, such as outlined in this paper, will be more robust in such applications. There is some hope that, armed with a suitably flexible language, the birational geom- etry of X together with a polarisation can be used to construct solutions to a real Monge- Ampére equation on B. 1.1 How to read this paper The structure of the paper is as follows. In the first three sections, we establish the theory of semirings and their (semi)modules as a theory of commutative algebras in a certain closed monoidal category, the category of B-modules (ModB, ⊕). The objects of ModB are also known in the literature as 'join-semilattices'. Since we wish to compare with non-Archimedean geometry, we actually need to work with topological B-modules (§4). At this paper's level of sophistication, this causes few complications. Apart from establishing the formal properties of the categories of B-modules and semir- ings, the secondary thrust of this part is to introduce various versions of the subobject and free functors B, B c :ModA −→ ModB 1 2 Ring Ring −→ etc. which will pave the major highway linking algebraic and 'semialgebraic' geometry. I have spelled out in some detail the functoriality of these constructions, though they are mostly self-evident. Section 6 sets about defining the localisation theory of semirings, which is designed to parallel the one used for topological rings in non-Archimedean geometry. These bounded localisations factorise into two types: cellular, and free. The latter resemble ordinary lo- calisations of algebras, and the algebraically-minded reader will be unsurprised by their presence. The cellular localisations, on the other hand, may be less familiar: they involve the non-flat operation of setting a variable equal to zero. Thinking of a skeleton as a poly- hedral or cell complex, these localisations will be dual to the inclusions of cells (of possibly lower dimension). Perhaps confusingly, these are the semiring homomorphisms that cor- respond, under Bc, to open immersions of formal schemes. The precise statements are the Zariski-open (6.18) and cellular cover (7.9) formulas. With some understanding of the 'cellular topology' we are able, as an aside, to describe the spectrum of contracting semirings in terms of a naïve construction: the prime spectrum 1In general the Gross-Siebert programme [GS07] bypasses the non-Archimedean geometry to give a direct construction of the affine structure of B, up to worm deformations, in terms of toric geometry. Using this approach, they were able to obtain many results with a combinatorial flavour, and even a reconstruction of X (as an algebraic variety) from B together with some cocycle data. To mimic at least their basic construction in the context of skeleta is not difficult, but beyond the scope of this paper. 7 §6.3. This makes clear the relationship between the topological space underlying a formal scheme X + and the spectrum of the ideal (sheaf) semiring Bc O + X . It is also easy to describe the free localisations in terms of the polyhedral complex pic- ture. Inverting a strictly convex function has the effect of destroying the affine structure along its non-differentiability locus (or 'tropical set'); we therefore think of it as further subdividing our complex into the cells on which the function is affine. We can also give an algebro-geometric interpretation of these subdivisions: it is given by the blow-up formula (6.22). In the setting of a formal scheme X + over a DVR OK , it says that blowing up an ideal sheaf J supported on the reduction has the effect of inverting J in Bc(OX + ⊗ K; OX +). Intuitively, the blow-up of J is the universal way to make it an invertible sheaf. In §7 we meet the category Sk of skeleta, and introduce some universal constructions of certain skeleta from adic spaces and their models. The construction of this category follows the general programme of glueing objects inside a topos, as outlined in [TV09]. The main result 7.21 - which concerns the main skeletal invariant of an analytic space X , the universal skeleton skX - boils down to proving that for reasonable values of X , the topological space underlying X can be identified with that of Spec Γ(X ; OX ). The technical part of the proof is based on the Zariski-open, blow-up, and cellular cover formulas, which together allow us to explicitly match the open subsets of X with those of its skeleton. As an artefact of the proof, we may notice that a surprisingly many skeleta - those associated to any quasi-compact, quasi-separated analytic space - are affine. As an aside in §7 I was able to obtain a kind of quantification (thm. 7.11) of this observation. We also glance at the relationship (thm. 7.25) between skeleta and the theory of Berkovich. In the examples section 8, we reconstruct some well-known 'tropical spaces' as skeleta: the dual intersection complexes of locally toric degenerations (§8.2), and the tropicalisations of subvarieties of a toric variety [Pay09] (§8.3). I have also attempted to couch the construc- tion of an affine manifold from a Tate elliptic curve, summarised above, in more general terms (§8.4). This forms the first test case of an ongoing project. Acknowledgements I would like to thank my PhD supervisors, Alessio Corti and Richard Thomas, for their support. I thank also Mark Gross, Sam Payne, and Jeff Giansiracusa, for interesting con- versations. I also thank the Cecil King foundation for funding my visit to Mark Gross in UCSD, where some of the aforementioned conversations, as well as part of the work writing this paper, occurred. 2 Preliminaries and conventions 2.1 On sites and topoi Our general notational conventions on sites and topoi follow the canonical [AGV70]. The central glueing construction of the paper revolves around the notion of a locally repre- sentable sheaf, defined in [TV09, def. 2.15]. I only wish to replace the input, the authors' notion of Zariski-open immersion, with something a bit more flexible. 8 2.1 Definition. Let U be a class of monomorphisms in a category C stable for composition and base change. One defines the structure of a Grothendieck site on C whose generating coverings are finite families of morphisms in U that form a covering for the canonical site. morphisms in U . An open immersion in the associated topos Ce is a morphism locally representable by An object of Ce is locally representable if it is a union of representable open subobjects. Much of the theory of [TV09] is valid with this more flexible input, notably proposition 2.18. I warn the reader only that without a locality requirement for our definition of affine open immersion, part 2 of [TV09, prop. 2.17] is false. This is the case, for example, for the category of adic spaces (§2.2). The resulting class of objects can also be characterised in terms of point-set topology via a modern analogue of Stone's construction: it has enough points. ii) Since the morphisms in U were assumed to be monic, the small topos of an object i) By construction, Ce is a coherent topos and so by Deligne's theorem [MM92, §IX.11.3] X ∈ Ce is localic. iii) Being localic and having enough points, the small topos of an object is therefore equiv- alent to a uniquely determined sober topological space [MM92, §IX.3.1-4]. This deter- mines a functor that takes morphisms in U to open immersions. Ce → Top iv) The topological space associated to a representable (or more generally, compact) object is quasi-compact and quasi-separated. v) Being locally representable corresponds to having a basis of open sets coming from open immersions with representable source. A covering - or, more precisely, two-term hypercovering - of a space X will be denoted U• ։ X , with the bullet ranging over a partially ordered set of indices. 2.2 Non-Archimedean geometry The perspective on non-Archimedean geometry taken in this paper was influenced by the foundational works [Hub96] and [FK13]. Broadly speaking, I have adopted the categorical localisation constructions of the latter (after the approach of Raynaud), but the language and notation of the former - in particular, the nomenclature adic spaces. I introduce the following innovations in terminology: 2.2 Definitions. A marked formal scheme is a pair (X +, Z) consisting of a formal scheme X + and a collection Z of Cartier divisors. A morphism of marked formal schemes is a 2 such that (f −1Z2)red ⊆ Z1. An admissible blow-up is a finite type morphism f : X + blow-up whose centre has underlying reduced scheme contained in Z. 1 → X + 9 A non-Archimedean ring is a pair (A, A+) consisting of an adic ring A+ and a localisation A of A+. We only consider locally convex (A, A+)-modules, that is, complete topological A- modules whose topology is generated by A+-submodules; the category of such is denoted LC(A,A+), or just LCA for short. • The category Ad of adic spaces is defined by the same means as the category Rf of [FK13, §II.2], with the modification that the input to the localisation construction is instead the category of coherent marked formal schemes at admissible blow-ups, as in def. 2.2. This ensures that the notion of adic space is a generalisation of that of formal scheme. • The glueing construction of [FK13, §II.2.2(c)], although expressed in less standard language, is identical to the locally representable sheaves story of §2.1. • Following Huber, the sheaf of functions extending over a model is denoted O + (rather than O int as in [FK13, §II.3.2(a)]). The structure sheaf of the adic topos Ade is a pair (O, O +). It is a sheaf of non-Archimedean rings in the sense of def. 2.2. • Accordingly, we also adopt the notation X + for models of a adic space X . The cate- gory of models is denoted MdlX +; if X is qcqs, it is cofiltered. The map j : X → X + exhibiting the model is a morphism of adic spaces. • An affine adic space X is one admitting an affine formal model whose marking divisors are principal. By definition, this space is the spectrum of the non-Archimedean ring A := ΓOX ; following Huber, this spectrum is denoted Spa A. The key aspect of this construction that we will use is that for quasi-compact, quasi- separated X , as topologically ringed sites, (X , O + X ) ≃ lim X +∈Mdl(X) X + (1) where the limit is over all formal models of X . 3 Subobjects and B-modules The theory of B-modules plays the same rôle in tropical geometry that the theory of Abelian groups plays in algebraic geometry: while rings are commutative monoids in the category of Abelian groups, semirings are commutative monoids instead in the category of B-modules. This is the fundamental point of departure of the two theories. There is therefore a tempta- tion to try to treat B-modules as "broken" Abelian groups, and to literally translate as many concepts and constructions from the category Ab as will survive the transition. In this paper, I adopt a different perspective. A B-module is a particular type of partially ordered set which axiomatises some properties of subobject posets in Abelian and similar categories. In particular, there is a functor B : Ab → ModB which associates to an Abelian group its B-module of subgroups. As such, I propose to treat B-modules as though they are lower categorical shadows of structures in the category of Abelian groups, rather than simply as elements of a single Abelian group. The theory of B-modules is a naïve form of category theory, rather than a weak form of group theory. 10 There is also a dual, or more precisely, adjoint, perspective, which is that a B-module is the natural recipient of a non-Archimedean seminorm from an Abelian group. This fits well with traditional perspectives on non-Archimedean geometry. In keeping with the ahistori- cal nature of this paper, I barely touch upon this idea here (but see example 5.5). 3.1 B-modules 3.1 Definition. A B-module is an idempotent commutative monoid. In other words, it is a commutative monoid (α, ∨, −∞) in which the identity X ∨ X = X holds for all X ∈ α, and where −∞ is the identity for ∨. The category of B-modules and their homomorphisms will be denoted ModB. A B-module is automatically a partially ordered set with the relation X ≤ Y ⇔ X ∨ Y = Y . It has all finite joins (suprema). Conversely, any poset with finite joins is a B-module un- der the binary join operation. They are more commonly called join semilattices or simply semilattices.2 We may therefore introduce immediately a path to category theory in the form of an essentially equivalent definition. 3.2 Definition. A B-module is a preorder with finite colimits. A B-module homomorphism is a right exact functor. 3.3 Examples. The null or trivial B-module is the B-module with one element {−∞}. The Boolean semifield is the partial order B = {−∞,0} ≃ {false,true}. The integer, rational, and real semifields Z∨, Q∨, R∨ are obtained by disjointly affixing −∞ to Z, Q, R, respectively. More generally, we can obtain a semifield H∨ by adjoining −∞ to any totally ordered Abelian group H. These semifields are totally ordered B-modules (in fact, semirings; cf. e.g. 5.3). If X is a topological space, the set C0(X , R∨) of continuous functions X → R∨, where R∨ is equipped with the order topology, is a B-module. So too are the subsets of bounded above functions, or of functions bounded above by some fixed constant C ∈ R. Suppose that X is a manifold (with boundary). The subset C1(X , R∨) of differentiable functions is not a B-module, since the pointwise maximum f ∨ g of two differentiable func- tions f , g needn't be differentiable. One must allow piecewise differentiable PC1 (or piece- wise smooth PC∞) functions to obtain submodules of C0(X , R∨). Since convexity is pre- served under ∨, the subsets of convex functions CPCr(X ; R∨) are also submodules. We can also endow X with some kind of affine structure [KS06, §2.1], which gives rise to B-modules CPA∗(X , R∨), ∗ = R, Q, Z of piecewise-affine, convex functions (with real, rational, or integer slopes, respectively). If X = ∆ ⊂ Rn is a polytope, then it has a notion of integer points, and so one can define a B-module CPAZ(X , Z∨) of piecewise-affine, convex functions with integer slopes and which take integer values on lattice points Zn ∩ ∆. Note that any function in this B-module that attains the value −∞ must in fact be constant. 2I abandon this terminology for a number of reasons, but one could be the inconsistency of the rôles of the modifier semi in the words 'semiring' and 'semilattice'. 11 3.4 Example. Let S be a set. The subset B-module BS is the power set of S; its join operation is union. The free B-module BcS ⊆ BS is the set of finite subsets of S. Its elements may be written uniquely (up to permutation of terms) as idempotent linear expressions "with coefficients in B," i.e. as X1 ∨ · · · ∨ X k for some X1,..., X k ∈ S. Both constructions are functorial in S, so we have functors B, Bc : Set → ModB; the latter is left adjoint to the forgetful functor. The theory of B-modules is a finitary algebraic theory, and so limits, filtered colimits, and quotients by groupoid relations are computed in Set; this remains true with Set replaced by any topos. The following (proposition 3.5) also remains true in that generality. For any B-modules α, β, we can construct the direct join α ∨ β as the B-module whose underlying set is the Cartesian product α × β and whose join is defined by the law (X1,Y1) ∨ (X2,Y2) := (X1 ∨ X2,Y1 ∨ Y2). I simply write X1 ∨ X2 for (X1, X2) where this is not likely to cause confusion. There are natural B-module homomorphisms α → α ∨ β → α defined by X 7→ X ∨ (−∞), X ∨ Y 7→ X , and similarly for β, which make the direct join into a coproduct and product in ModB. In particular, there are natural homomorphisms ∆ −→ α ∨ α ∨ −→ α, α the diagonal and the map defining the B-module structure, respectively. I use also the direct join notation for a pushout α ∨β γ := α ⊔β γ. The null B-module is the empty direct join, or zero object, of ModB. The kernel and cokernel of a morphism f : α → β of B-modules are defined: ker f := f −1(−∞),coker f = β ∨α {−∞}. If f , g ∈ Hom(α, β), then their 'sum' is given by the composition id×id −→ α ∨ α −→ β ∨ β id⊔id −→ β α which takes X ∈ α to f (X ) ∨ g(X ). This description establishes that the monoid Hom(α, β) is in fact a B-module in which f ≤ g if and only if f (X ) ≤ g(X ) ∈ β ∀X ∈ α; moreover Hom(−, −) is a bifunctor from ModB to itself. 3.5 Proposition. The category ModB is semiadditive.3 It is complete and cocomplete. It is harder to obtain an explicit description of general coequalisers; see §3.5.2. 3.1.1 Subobjects Beyond the geometric examples 3.3, the primary source of B-modules are the subobject posets in certain finitely cocomplete categories. One could formulate a general theory of subobjects in certain kinds of categories; however, for the purposes of this paper we only need to know the version for modules over a commutative ring (possibly in a Grothendieck topos). 3A category is semiadditive if it admits finite products and coproducts and the natural map × → ⊔ is an isomorphism of bifunctors. 12 3.6 Definition. Let A be a ring, M an A-module. I write B(M; A) for the submodule lattice of M, the partially ordered set of all A-submodules of M; its join operation is submodule sum. I abbreviate B(A; A) to BA, the ideal semiring of A. The submodule lattice is functorial in A-module homomorphisms f : M1 → M2 Bf : B(M1; A) → B(M2; A), N 7→ Im(f N) and ring maps g : A → B Bg : B(M; A) → B(M ⊗A B; B), N 7→ Im(N ⊗A B → M ⊗A B). In particular, BA → BB. Typically, A =: OX will be a sheaf of rings on some space X and M an OX -module, in which case B(M; OX ) is the lattice of OX -subsheaves of M. The submodule lattice is then functorial for maps defined in the sheaf category Xe, but also for morphisms g : (Y , OY ) → (X , OX ) of ringed spaces. In the latter case, I will write g∗ = Bg : B(M; OX ) → B(g∗M; OY ), N 7→ Im(g∗N → g∗M) for the induced map of lattices, though this should not be confused with the functor of pullback of OY -modules, which it equals only when g is flat. 3.7 Example (Discs). Let K be a complete, valued field, V a K-vector space. I would like to be able to say that the subobjects of V are the discs [Bou62, §2.2]. If K is non-Archimedean with ring of integers OK , then a disc is the same thing as an OK -submodule, and so the set of discs is B(V ; OK) (which in loc. cit. is called D(V )). If K is Archimedean, then we need an alternative theory of 'abstract discs' or 'convex sets'. Following [Dur07], one can describe it as a theory of modules for a certain algebraic monad. For instance, if K = R = Q∞, the corresponding monad is that Z∞ (also written OR) of convex, balanced sets [Dur07, §2.14]. An object of ModZ∞ is a set M equipped with a way of evaluating convex linear combinations λi xi, xi ∈ M, λi ∈ R, kXi=1 λi ≤ 1 kXi=1 of its elements. A subset of V is a disc if and only if it is stable for the action of Z∞. In other words, D(V ) = B(V ; Z∞), in a mild generalisation of definition 3.6. 3.2 Orders and lattices The alternative definition 3.2 puts B-module theory in the broader context of order theory. In particular, there are sometimes defined infinitary operations (X i)i∈I 7→ sup i∈I X i. I reserve the notationWk The following definitions are standard in order theory: i=1 X i for the (always defined) operation of finite supremum or join. 13 3.8 Definitions. A map of posets is monotone if it preserves the order. A monotone map of posets is the same as a functor of preorders. The category of posets and monotone maps is denoted POSet. A B-module is a complete lattice if it has all suprema. A complete lattice is the same thing as a cocomplete poset. In particular, meets exist. A lattice homomorphism is a map of complete lattices preserving all suprema, that is, a colimit-preserving functor. The category of complete lattices and homomorphisms is denoted Lat ⊂ Span. Let α be a B-module, S, T ⊆ α. The lower slice set S≤T := {Y ∈ S∃X ∈ T s.t. X ∨ Y = X } = {Y ∈ S∃X ∈ T s.t. Y ≤ X } (2) is the B-module of all elements contained in S that are bounded above by an element of T. The upper slice set S≥T := {Y ∈ S∃X ∈ T s.t. X ∨ Y = Y } is defined dually. A subset S is said to be lower (resp. upper) if S = α≤S (resp. α≥S). A lower submodule of α is called an ideal of α. The subset S is called coinitial (resp. cofinal) if all lower (resp. upper) slice sets are non-empty, that is, ∀X ∈ α, ∃Y ∈ S such that Y ≤ X (resp. X ≤ Y ). 3.9 Example. A quotient of a B-module α by an ideal ι, that is, the cokernel of the inclu- sion ι ,→ α, is easy to make explicit: it is simply the set-theoretic quotient α/ι of α by the equivalence relation ι ∼ −∞. If ι = α≤T is a slice set, we may also write α/T. The cokernel of a B-module homomorphism f : α → β is the quotient of β by β≤f (α), the smallest ideal containing f (α). In particular, α is an ideal if and only if it is the kernel of its cokernel. The set of all ideals of α can be thought of as a subobject poset in the category of B- modules. It is a complete lattice. 3.10 Definition. The lattice of ideals Bα of a B-module α is called the lattice completion of α. The lattice completion defines a left adjoint B : ModB → Lat to the inclusion of Lat into ModB. The unit id → B of the adjunction is an injective homomorphism As a preorder, the lattice completion of α is its category of ind-objects [AGV70, §I.8.2]. α → Bα, X 7→ α≤X . 3.3 Finiteness In ordinary category theory, the notion of finite presentation of objects is captured by com- pact objects, that is, objects whose associated co-representable functor preserves filtered colimits. One then seeks to try to understand all objects of the category in terms of its com- pact objects. In particular, we like to work with compactly generated categories: those for which every object is a colimit of compact objects. A compactly generated category C is equivalent C ∼= Ind(Cc) 14 to its category of ind-compact objects. In particular, filtered colimits are exact. The order-theoretic version of compactness is finiteness. Its basic behaviour can be de- rived by applying the above results directly to the special case of objects in pre-orders. 3.11 Definitions. An element X of a complete lattice α is finite if, for any formula X ≤ supi∈I X i in α, with the X i a filtered family, there exists an index i such that X ≤ X i. A lattice is algebraic if every element is a supremum of finite elements. A homomorphism f : α → β preserves finiteness if f (X ) is finite whenever X is. Be warned that it is not, in general, equivalent to replace the inequalities in the above definition with equalities. An element X ∈ α can be finite as an element of α≤X without being finite in α. 3.12 Lemma. A finite join of finite elements is finite. Let α be a complete lattice. I denote by αc its subset of finite elements; by the lemma, αc is a B-module. It is functorial for B-module homomorphisms that preserve finiteness. 3.13 Proposition. Let α ∈ Lat. The following are equivalent: i) α is algebraic; ii) sup : B(αc) → α is an isomorphism; iii) Every element of α is a supremum of elements X that are finite in their slice set α≤X , and finite meets distribute over filtered suprema in α. Let α be any B-module. A B-module ideal ι ,→ α is finite as an element of Bα if and only if it is principal, that is, equal to some slice set α≤X . Therefore, α → Bα identifies α with the B-module of finite elements of Bα. This sets up an equivalence of categories B : ModB ⇆ Latal : (−)c between ModB and the category Latal of algebraic lattices. 3.14 Examples. Let S be a set. A subset of S is finite as an element of BS if and only if it has finitely many elements; (BS)c ∼= BcS in the notation of example 3.3. The power set BS ∼= BBcS is an algebraic lattice. A submodule of a module M over a ring A is finite if and only if it is finitely generated; B(M; A) is an algebraic lattice. 3.15 Definition. The finite submodule or free B-module on M is the B-module Bc(M; A) ∼= (B(M; A))c of finitely generated A-submodules of M; since a sum of finite submodules is finite, this is closed in B(M; A) under joins. By algebraicity, BBc(M; A) ∼= B(M; A). We abbre- viate Bc(A; A) to Bc A. 3.16 Example (Seminorms). Let A be an Abelian group. A (logarithmic) non-Archimedean seminorm on A with values in a B-module α is a map of sets val : A → α satisfying the 15 ultrametric inequality val(f + g) ≤ val f ∨ val g. One can take the supremum of any (non- Archimedean) seminorm on A over any finitely generated subgroup X ⊆ A; indeed, if X = (x1,..., xn), then This supremum defines a B-module homomorphism Bc(A; Z) → α. val f = sup f ∈X val xn. n_i=1 This correspondence exhibits the natural seminorm A → B c(A; Z), a 7→ (a) as universal among seminorms of A into any B-module. In other words, Bc(A; Z) corepre- sents the functor 1 2 Nm(A, −) : ModB → Set of seminorms on A. 3.17 Example. Let K be a non-Archimedean field with ring of integers OK and value group K ⊆ R. The given valuation induces a B-module isomorphism Bc(K; OK)f→K ∨. In fact, the same holds if K is Archimedean, cf. e.g. 3.7. 3.18 Example (Not enough finites). Let K be a complete, discrete valuation field with uni- formiser t. Let K be an algebraic closure with ring of integers O ∨ = Q≤0 ⊔ {−∞} (cf. def. 5.10) is the set of principal ideals generated by positive rational powers tq of the uniformiser. The 'traditional' way to complete Q∨◦ would be to embed it in its set ∨ of Dedekind cuts. The latter is a complete lattice with no finite elements. R◦ K . Then Bc O ∼= Q◦ K Of course, it is more sensible in this case to consider Q◦ ∨ as the set of finite elements in the well-behaved lattice BQ◦ ∨ ∈ Latal. One can show that if the above statements are interpreted in the usual semantics within the topos of sheaves on a space X , one obtains the following set-theoretic characterisation of the finite submodule B-module (sheaf). Let OX be a sheaf of rings on X , M an OX -module. 3.19 Definition. A submodule N ,→ M is locally finitely generated if there exists a covering {f i : Ui → X }i∈I and epimorphisms O i N for some numbers ni ∈ N. ։ f ∗ ni Ui The finite submodule or free B-module on M is the sheaf c(M; OX ) : U 7→ B c(M(U); OX (U)) B of locally finitely generated OX -submodules of M. One may simply take this as a set-theoretic definition of Bc, verifying directly that Bc(M; OX ) is a sheaf. 3.20 Example (Local seminorms). Let X be a space, A a sheaf of Abelian groups on X . A seminorm on A with values in a sheaf α of B-modules is a map A → α of sheaves which induces over each U ⊆ X a non-Archimedean seminorm on A(U) (e.g. 3.16). One can define a universal seminorm A → Bc(A;ModOX ), which, for a given U ⊆ X , takes f ∈ A(U) to the subsheaf of AU that it locally generates. Any seminorm val : A → α factors uniquely through this universal one, with the factoring arrow taking any finite subsheaf F ⊆ AV to sup f •∈F(U•)¯¯val f •¯¯ =¯¯¯¯¯ n•_i=1 16 i¯¯¯¯¯ val f • ∈ α(U•) ∼= α(V ) In this formula, U• ։ V is a covering on which F is defined, and (f • finite system of generators for F(U•). (Note that the numbers n• need not be bounded.) n•) denotes a locally 1 ,..., f • 3.4 Noetherian 3.21 Definition. A B-module is called Noetherian if the slice sets satisfy the ascending chain condition, that is, if every bounded, totally ordered subset has a maximum. 3.22 Proposition. Let {X i}i∈I ⊆ α be a bounded family of elements of a B-module α. If α is Proof. We proceed by contraposition. Suppose that for all finite J ⊆ I, there is some i(J) ∈ Noetherian, then supi∈I X i =Wi∈J X i for some finite J ⊆ I. I \ J such that X i(J) 6≤W j∈J X j, that is, such that W j∈J X j < X i(J) ∨W j∈J X j. Then I is infinite, and starting from any index 0 ∈ I we can inductively construct an infinite, strictly increasing sequence where n := i({0,..., n − 1}) ∈ I. Therefore α is not Noetherian. X0 < (X1 ∨ X0) < (X2 ∨ X1 ∨ X0) < · · · 3.23 Corollary. The following are equivalent for a bounded B-module α: i) α is Noetherian; ii) α is a complete lattice, and αc = α; iii) αf→Bα. A B-module is Noetherian if and only if its every bounded ideal is Noetherian. 3.24 Example. Let A be a ring. The following are equivalent: i) A is Noetherian; ii) BA is Noetherian; iii) Bc A is Noetherian; iv) Bc M is Noetherian for all A-modules M; In this case, Bc M = BM if and only if M is finitely generated. 3.5 Adjunction As in category theory, the notion of adjoint map is central to the theory of B-modules. 3.25 Definition. Let f : α → β be a monotone map of B-modules. We say that a monotone map g : β → α is right adjoint to f , and write f † := g, if idα ≤ g f and f g ≤ idβ. In this situation, we also say † g := f is left adjoint to g. 17 If α is a complete lattice, then by the adjoint functor theorem a right adjoint exists for f if and only if it preserves arbitrary suprema. We have the formula X 7→ f †X = sup α≤f −1(X). Alternatively, Bf always preserves suprema, and therefore we can always find an adjoint (Bf )† : Bβ → Bα, ι 7→ f −1 ι at the level of the lattice completions. The restriction of (Bf )† to β is an ind-adjoint in the sense of [AGV70, §I.8.11]. If an ordinary right adjoint to f exists, then the ind-adjoint is the composite of this with the inclusion α → Bα; I therefore denote the ind-adjoint also by f † in general, since no confusion can arise. In particular, any B-module homomorphism gives rise to a diagram Bα `❅ ❅ ❅ f † f α in POSet, and idα ≤ f † f . 3.5.1 Pullback and pushforward ❅ ❅ ❅ ❅ ❅ β Suppose that A is a ring, f : M1 → M2 an A-module homomorphism. If N ,→ M2 is a sub- module, then so is N ×M2 M1 → M1. The fibre product is a monotone map f † = f −1 : B(M2; A) → B(M1; A), N 7→ N ×M2 M1, right adjoint to the image functor Bf . It happens to be a lattice homomorphism. Secondly, let g : X → Y be a morphism of ringed spaces, A = OX . Then the pushforward functor f∗ is right adjoint to f ∗ on the category ModO of modules. Correspondingly, g∗ : B(M; OX ) → B(g∗M; OY ), N 7→ g∗N is right adjoint to the lattice homomorphism g∗ = Bg. Since pushforward is left exact, this lattice homomorphism does agree with the functor on modules. 3.26 Example. If X ,→ Y is an open immersion of schemes, then the right adjoint to f ∗ : BOY → BOX sends a closed subscheme of X to its scheme-theoretic closure in Y . 3.5.2 B-module quotients In the theory of categorical localisation, certain types of adjunction can provide a substi- tute for a linear calculus of quotients of categories. One can apply a similar technique to semilinear algebra in order to provide explicit descriptions of B-module coequalisers and quotients. Let s, t : α ⇒ β be a pair of B-module homomorphisms. 3.27 Definition. An ideal ι ,→ β is invariant for the pair s, t if, for all X ∈ α, sX ∈ ι ⇔ tX ∈ ι. 18 / / O O ` Since s and t are B-module homomorphisms, the subset Bβ/(s ∼ t) ⊆ Bβ of invariant ideals is closed under arbitrary suprema. The right adjoint to the inclusion is a self- homomorphism p := sup n∈N³(ts†)n ∨ (st†)n´ : Bβ → Bβ taking an ideal to the smallest invariant ideal containing it. It coequalises s, t. In fact, for any B-module homomorphism f : β → γ coequalising s, t, Bf is independent of the action of s, t, that is, factors uniquely through p. Setting p : β → β/(s = t) := p(β) ⊆ Bβ/(s ∼ t) ⊆ Bβ where p(β) is the set-theoretic image, we therefore obtain: 3.28 Lemma. β/(s = t) is a coequaliser for s, t. In the special case α = B, where s, t are determined by some elements S, T ∈ β, we write also as usual β/(S = T) for the B-module quotient (by the congruence relation generated by the relation S = T). Specialisations of the above construction will come into play in later sections; see, for example, §5.2. 4 Topological lattices A topological space with linear structure is linearly topologised if its topology is generated by linear subspaces. In other words, a linear topology on a space is one that can be defined in terms of a certain decoration - a principal topology - on its subobject lattice. Let α be a complete lattice, αu ⊆ α a non-empty, upper subset, closed under finite meets. Such an αu is called a fundamental system of opens, or just fundamental system, for short. 4.1 Lemma. The collection of slice sets α≤X for X in a fundamental system, together with ;, are a topology on α for which ∨ is continuous. Proof. It is clear that these sets define a topology; for continuity of ∨ : α×α → α, note simply that ∨−1(α≤X ) = α≤X × α≤X . The topology in the lemma is that defined by the fundamental system. Any intersection of fundamental systems is a fundamental system. Therefore, for any family f i : α → βi of maps of complete lattices and fundamental systems βu i on βi, there is a smallest fundamental system on α such that the f i are continuous for the induced topology. Explicitly, it is given by the closure of the upper set under finite meets. [i,X ∈βu i α≥f † i (X) Dually, any union of fundamental systems generates a new fundamental system under finite meets. This coincides with the usual notion of generation of new topologies. Hence, for any family gi : αi → β of homomorphisms and fundamental systems αu i , there is a largest fundamental system β≥fi(αu i ) β u :=\i 19 on β making the gi continuous, and the topology it defines is simply the strong topology on the underlying set. 4.2 Definitions. A complete lattice equipped with a principal topology, that is, a topol- ogy defined by a fundamental system, is called simply a topological lattice. A topological B-module is a B-module α equipped with an ideal topology, that is, is the subspace topol- ogy with respect to some principal topology on Bα ⊇ α. A fundamental system for α is a fundamental system for Bα, and we write αu := (Bα)u. A homomorphism of topological B-modules (resp. lattices), is a continuous B-module homomorphism (resp. lattice homorphism). The category of topological lattices is denoted Latt, the category of topological B-modules ModB,t. A topological B-module is a B-module whose inhabited open sets are ideals, and in which every neighbourhood (of −∞) is open. A topological lattice is the same, except that inhabited open sets are principal ideals. There is also an obvious notion of principal topology on a possibly incomplete B-module. 4.3 Example. The semifields H∨ (e.g. 3.3) will always come equipped with the (principal) topology Hu ∨ = H. (In particular, B is discrete.) A net {X i}i∈I converges to −∞ if and only if it does so with respect to the order; in other words, if ∀λ ∈ H ∃i ∈ I such that X j ≤ λ for all j > i. The category of topological B-modules (resp. lattices) comes with a forgetful functor ? : ModB,t → ModB which I do not suppress from the notation. Its left adjoint is given by equipping a lattice α with the discrete topology αu = α, its right adjoint by the trivial topology αu triv = {sup α}. Both adjoints are fully faithful. We will treat the category of B-modules as the (full) subcategory of discrete objects inside ModB,t. In particular, limits (resp. colimits) in ModB,t are computed by equipping the limits (resp. colimits) of the underlying discrete B-modules with weak (resp. strong) topologies. 4.4 Example. A non-trivial topological B-module is never Hausdorff in the sense of point-set topology, since every open set contains −∞. Let us instead say that a B-module α is Haus- dorff if infαu = −∞. The category ModB,t of Hausdorff B-modules is a reflective subcategory of ModB,t.4 4.5 Definitions. Let f i : αi → β be a family of continuous B-module homomorphisms. We say that β carries the strong topology with respect to the f i, or that the family f i is strong, if its topology is the strongest ideal topology such that the f i are continuous. In particular, if f is just a single map, f : α → β is strong if and only if it sends αu into βu ⊆ Bβ. In this case, we may also say that f is open - although beware that it may fail to be open in the sense of general topology. If g j : α → β j are a family of continuous B-module homomorphisms, then α carries the weak topology with respect to the g j, or that the family gi is weak, if its topology is the weakest ideal topology such that the gi are continuous. From the definition of ideal topology, it follows: 4The notation t follows Bourbaki [Bou62]. 20 4.6 Lemma. A family f i is weak (resp. strong) if and only if the induced family Bf i on the lattice completions is weak (resp. strong). In particular, weak and strong topologies, and hence limits and colimits, always exist. 4.7 Proposition. The category ModB,t is complete, cocomplete, and semi-additive. Filtered colimits are exact. Proof. We need to check that the product and coproduct topology on the direct join agree. The explicit formulas show that (α × β)u = {(X ,sup β) ∧ (sup α,Y )X ∈ α u,Y ∈ β u} = α u × β u = (α ⊔ β)u which proves that ModB,t is semi-additive. Now let αi, βi → γi be a filtered system, with α, β → γ its colimit. We will confuse αi, βi with their image in α ×γ β. To show that the natural map colimi(αi ×γi βi) → α ×γ β is a homeomorphism, it will suffice to show that it is open. Let U ∈µcolim i (αi ×γi βi)¶u =\i α ×γ β≥αu i ∧βu i so there exist X i ∈ αu suprema distribute over meets (cf. 3.3), i ,Yi ∈ βu i such that supi(X i ∧ Yi) ≤ U. Since, in B(α ×γ β), filtered U ≥ (sup i X i) ∧ (sup i Yi) ∈Ã\i α ×γ β≥αu i! ∧Ã\i α ×γ β≥βu i! = (α ×γ β)u and is therefore open. 4.8 Example. There are two obvious ways to topologise the function B-module C0(X , R∨) on a topological space X (and similarly PCr(X ; −), CPCr(X ; −), etc., cf. e.g. 3.3): a topology of pointwise convergence, which is the weak topology with respect to evaluation maps evx : C0(X , R∨) → R∨, and one of uniform convergence, which is the strong topology with respect to the inclusion R∨ ,→ C0(X , R∨) of constants. In the important case CPA∗(X , R∨) of convex, piecewise-affine functions, when X is compact with affine structure, these two topologies agree. 4.1 Topological modules Let A be a non-Archimedean ring (def. 2.2), M a (complete locally convex) A-module. We equip B(M; A+) with a principal topology (BM)u := {U ,→ MU open}, which, by definition of local convexity, is enough to recover the topology on M. We also consider Bc(M; A+) as a topological B-module with respect to the subspace topology. This topology is natural for continuous module homomorphisms, and hence lifts B(c) to a functor (c)(−; M) : LCA → ModB,t. B 21 If g : A → B is a ring homomorphism, then the base extension must be replaced with a completed base extension −b⊗AB : LCA → LCB (that is, ordinary base extension followed by topological completion with respect to the projective tensor product topology). Correspond- ingly, there is a lattice (resp. B-module) homomorphism Bg : B (c)(M; A+) → B (c)(Mb⊗A+B+; B+), N 7→ Im(N ⊗A+ B+ → Mb⊗AB); in the case M = A this agrees with the homomorphism B(c)(A; A+) → B(c)(B; B+) defined previously without taking into account the topology. The same functoriality extends to morphisms of nA-ringed spaces. Beware that the elements of Bc(M; A+) correspond to not necessarily closed submodules of M, and hence might not be represented by a subobject in LCA. 4.9 Example (Continuous seminorms). Let A be a linearly topologised Abelian group. It follows immediately from the definition of the topology on B(A; Z) that the universal semi- norm A → Bc(A; Z) (e.g. 3.16) on A is continuous. In fact, Bc(A; Z) carries the strong topology with respect to this map. In other words, if A → α is any continuous seminorm into some α ∈ ModB,t, then the factorisation Bc(A; Z) → α is also continuous. The topological free B-module Bc(A; Z) corepresents the functor of continuous semi- norms 1 2 Nm(A, −) : ModB,t → Set. As we know, we may also use a seminorm ν : A → α to induce a coarser topology on A, the weak topology with respect to Bc(A Z)ri ghtarrowα. This is called the induced topology with respect to ν. It is Hausdorff if and only if the image of Bc(A; Z) in α is. 4.10 Example. Let K be a complete, rank one valuation field. The isomorphism Bc(K; OK) ∼= K ∨ of example 3.17 is a homeomorphism. One can also formulate a theory of pro-discrete completion for B-modules and lattices to correspond to the completion operation for non-Archimedean rings and their modules. Followed to the conclusion of this paper, this would yield a different category of skeleta. However, in situations typically encountered in geometry, one only has to deal with rings A that have an ideal of definition I, and are therefore in particular first countable. In this situation, one can use the axiom of dependent choice to show that B(−; A+) is automatically pro-discrete. Moreover, Nakayama's lemma ensures that in these situations, even the free B-module Bc(−; A+) is pro-discrete. Indeed, if M ։ M/I is a quotient of discrete A+-modules, any finite system of generators for M/I lifts to generators for M. A pro-finite, I-adic A+- module is therefore finitely generated. Conversely, any finite topological A-module is I- adically complete. It follows that Bc(M; A+) is pro-discrete for any complete A-module M. The main results 7.21, 7.25 of this paper remain true, under such first-countability hypotheses, if we work instead with pro-discrete B-modules. 5 Semirings Any symmetric monoidal category C gives rise to a theory of commutative algebras Alg(C) and their modules. In this section, I describe a closed, symmetric monoidal structure on the category of B-modules; the corresponding theories are those of semirings and their semi- modules. This semialgebra will provide the algebraic underpinning of the theory of skeleta. 22 Let C a category equipped with a monoidal structure ⊗ with unit 1 = 1C. One has a category Alg(C) of monoids or algebras in C, which are objects A of C equipped with structural morphisms A ⊗ A µ → A e ← 1 satisfying various usual constraints, and morphisms respecting these. If A ∈ Alg(C), there is also a category ModA(C) of A-modules in C, which comes equipped with a free-forgetful adjunction − ⊗ A : C ⇄ ModA(C) :?A. The (right adjoint) forgetful functor ?A is conservative. If ⊗ is symmetric, then there is also a category CAlg(C) of commutative algebras. The module category ModA(C) over A ∈ CAlg(C) acquires its own symmetric monoidal structure, the relative tensor product − ⊗A − = coeq(− ⊗ − ⊗ A ⇒ − ⊗ −) (as long as C has coequalisers). If ⊗ is closed, that is, − ⊗ A has a right adjoint HomC(−, A), then ?A also commutes with colimits and therefore − ⊗ A preserves compactness. Limits and colimits of modules are computed in the underlying category. 5.1 The tensor sum The category of B-modules carries a closed symmetric monoidal structure given by the ten- sor sum operation ⊕ which, by definition, is characterised by a natural isomorphism HomB(α ⊕ β, γ) ∼= HomB(α,HomB(β, γ)) where we use the internal Hom functor defined in section 3. Alternatively, it is charac- terised as universal with respect to order-preserving maps α × β → γ that are right exact in each variable, that is, such that for each X ∈ β the composite α → α × {X } → γ is a B-module homomorphism, and similarly the transpose of this property. There is a canonical monotone map α × β → α ⊕ β such that for any such map, there is a unique extension α ⊕ β ❈ ❈ ❈ ❈ α × β ❈ ❈ ❈ !❈ / γ to a commuting diagram of sets. It identifies α × {−∞} ∪ {−∞} × β with {−∞}. Explicitly, α ⊕ β is generated by symbols X ⊕ Y with X ∈ α,Y ∈ β subject to the relations X ⊕ (Y1 ∨ Y2) = (X ⊕ Y1) ∨ (X ⊕ Y2); X ⊕ (−∞) = −∞ which ensure that the map [f : α ⊕ β → γ] 7→ [X 7→ [Y 7→ f (X ⊕ Y )]] is well-defined and determines the promised adjunction. 23 ! O O / 5.1 Proposition. The tensor sum defines a closed, symmetric monoidal structure on ModB. Proof. The argument is routine; I reproduce here the unit and counit of the adjunction − ⊕ α ⊣ Hom(α, −). First, we have maps β → Hom(α, α ⊕ β), X 7→ [Y 7→ Y ⊕ X ] which is a B-module homomorphism by the relations above. Second, one checks that the map ev : Hom(α, β) × α → β preserves joins in each variable, and so descends to a homomorphism Hom(α, β)⊕α → β. The definitions of semirings and semimodules are those of algebras and their modules in the category (ModB, ⊕). I spell out some of these definitions here, in order to fix notation. 5.2 Definitions. An idempotent semiring α, or, more briefly, semiring, is a commutative monoid object (α, +,0) in the monoidal category (ModB, ⊕). Explicitly, it is a B-module equipped with an additional commutative monoidal operation +, called addition, with iden- tity 0, that satisfies X + (Y1 ∨ Y2) = (X + Y1) ∨ (X + Y2) X + (−∞) = −∞ ∀X . In notation, addition takes priority over joins: X + Y ∨ Z = (X + Y ) ∨ Z. A semiring homo- morphism is a monoid homomorphism. The category of semirings is denoted 1 2 Ring. We will also have occasion to use a category 1 2 Alg := Alg(ModB, ⊕) of possibly non- commutative semialgebras. A right semimodule over a semiring α, or (right) α-module, is a B-module µ equipped with an action µ ⊕ α → µ of α, written X ⊕ Z 7→ X + Z. A homomorphism of semimodules is a module homomorphism. The category of α-modules is denoted Modα. The relative tensor sum ⊕α on Modα is the quotient µ ⊕α ν ∼= coeq[µ ⊕ ν ⊕ α ⇒ µ ⊕ ν] in ModB. A commutative monoid in Modα is an α-algebra; it consists of the same data as a semiring β equipped with a semiring homomorphism α → β. The category of α-algebras is denoted 1 2 Ringα. The tensor sum of two α-algebras over α has a semiring structure. 5.3 Examples. The Boolean semifield B = {−∞,0} is a unit for the tensor sum operation. It therefore carries a unique semiring structure, of which the notation is indicative, rendering it initial in the category of semirings. That is, B plays the rôle in the category of semirings that Z plays in the category of rings. Any B-module is in a canonical and unique way a module over B, with 0 acting as the identity and −∞ as the constant map −∞; whence the terminology of B-modules. More generally, the semifield H∨ = H ⊔ {−∞} associated to a totally ordered Abelian group H (e.g. 3.3) carries an addition induced by the group operation on H. If H can be embedded into the additive group R, H∨ is a rank one semifield; these semi- fields play the rôle in tropical geometry that ordinary fields play in algebraic geometry. Of 24 particular interest are Z∨, Q∨, R∨, the value semifields of DVFs, their algebraic closures, and of Novikov fields, respectively. Other semifields that arise from geometry, for example lex, that is, Zk with the lexi- in Huber's work [Hub96], include those with H of the form Zk cographic ordering and −∞ adjoined. These semifields are non-Noetherian. They fit into a tower (Z k lex)∨ → (Z k−1 lex )∨ → · · · → Z∨ of semiring homomorphisms which successively kill each irreducible convex subgroup. See also [FK13, §0.6.1.(a)]. From general principles about algebra in monoidal categories, it follows: 5.4 Proposition. The category of semirings is complete and cocomplete. Limits and filtered colimits are computed in ModB, and the latter are exact. Pushouts are computed by the relative tensor sum. 5.1.1 Free semimodules Let A be a ring, M1, M2 ∈ ModA. There are natural homomorphisms m : B (c)(M1; A) ⊕ B (c)(M2; A) → B (c)(M1 ⊗A M2; A), [N1] ⊕ [N2] 7→ Im(N1 ⊗ N2 → M1 ⊗A M2) which in the case of the subobject B-module B is a lattice homomorphism. These homomor- phisms upgrade B(c) to lax monoidal functors (c) : (ModA, ⊗A) → (ModB, ⊕). B It is therefore compatible with algebra on both sides, in the following ways: i) If B is an A-algebra, then the multiplication µ on B induces a semiring structure on B(c)(B; A) [N1] + [N2] = µ(N1 ⊗ N2) ⊆ B and therefore the subobject (resp. free) B-module functors are upgraded to functors (c) : CAlgA → B 1 2 Ring. Beware that the sum [N1] + [N2] of elements of this submodule semiring corresponds to a product in B, and should not be confused with the set of sums of elements of N1 and N2, which corresponds instead to ∨. ii) If M is a B-module, then the B-action on M induces a B(c)(B; A)-module structure on B(c)(M; A). (c) : ModB → ModB(c)(B;A) B With respect to the relative tensor sum ⊕B(c)(B;A), these functors are lax monoidal. In particular, B(c)(B; A) is a B(c) A-algebra. Be warned that B(c) is not strongly monoidal: usually (c)(M1; A) ⊕B(c) A B B (c)(M2; A) 6∼= B (c)(M1 ⊗A M2; A). Similarly, it does not commute with most base changes - but see prop. 5.15. 25 5.5 Example (Seminormed vector spaces). Let V be a vector space over a complete, valued field K, considered as an OK -module as in example 3.7. Let us discuss seminorms on V with values in K ∨ = K ⊔ {−∞}, the value semifield of K. Note that K ∨ acts on the set of discs B(c)(V ; OK ) (cf. §5.1.1). If K is non-Archimedean, then in the same vein as the previous example 3.16, the ul- trametric inequality for a seminorm can be rephrased as sup z∈〈x,y〉 νz = νx ∨ νy, where 〈x, y〉 denotes the OK -module span of x and y. In other words, a seminorm is the same thing as a B-module homomorphism Bc(V ; OK) → K ∨, compatible with the actions of K ∨ on both sides. On the other hand, if K is Archimedean, and therefore either R or C, then the subobjects are the convex, balanced discs. The join of two discs is their convex hull, and a disc is finite if it is the convex hull of finitely many 'vertices'. Note that this implies that, for example, the unit disc of a K-Banach space V is infinite as soon as dimV > 1. The same triangle inequality as for the non-Archimedean case works if we replace the OK -module span 〈x, y〉 by the convex hull conv(x, y). An Archimedean seminorm is therefore once again a K ∨-module homomorphism Bc(V ; OK) → K ∨. 3.17). In either case, the valuation on K induces a semiring isomorphism Bc(K; OK)f→K ∨ (e.g. The space of seminorms is the hom-space Hom(BcV , K ∨). The unit disc associated to a seminorm ν is ν†0. Conversely, if D ∈ B(V ; OK ) is a disc, then the K ∨-action thereon determines a homomorphism K ∨ → B(V ; OK ), r 7→ rD, where we interpret r as the disc of radius r in K. SinceTr>r0 rD = r0D, this homomorphism preserves infima. If K = Z or R, then K ∨ has all infima, and hence this homomorphism has a left adjoint ν. Its behaviour on elements of V is νx = inf{r ∈ K ∨x ∈ r}. It therefore maps BcV into K ∨ if and only if the disc D absorbs in the sense that K D = V ; in this case, ν is a seminorm. This correspondence recovers the well-known dictionary between seminorms and absorbing discs in the theory of vector spaces over valued fields [Bou62, §2.1.2]. 5.1.2 Free semirings Let α be a semiring. The forgetful functor 1 fore has a left adjoint α[−]. It is the set of 'tropical polynomials' 2 Ringα → Set commutes with limits and there- α[S] ∼=( _n∈NS XX ∈S nX X + Cn¯¯¯¯¯ with the evident join and plus operations. Cn ∈ α, Cn = −∞ for n ≫ 0) 26 5.6 Definition. Let α be a semiring, S a set; α[S] is called the free semiring on S. The free semiring construction commutes with colimits; in particular we have the base change and composition α[S] ∼= α ⊕ B[S] α[S ⊔ T] = α[S] ⊕α α[T] for any α ∈ 1 2 Ring. There is similarly a free functor T 7→ B[T] for a B-module T; intuitively, it is the free semiring generated by the set T, subject to the order relations that exist in T. 5.2 Action by contraction The concept of contracting operator is natural in analysis, and is intimately related to the operator norm. In the context of this paper, we use this concept to control the bounds of tropical functions, and hence the radii of convergence of analytic functions. 5.7 Definition. An endomorphism f of a B-module α is contracting if, for each ideal ι ,→ α, f (ι) ⊆ ι. That is, f is contracting if and only if f (X ) ≤ X for all X ∈ α. 5.8 Example. Let A be an algebra and M an A-module. An A-linear endomorphism of M induces a contracting endomorphism of B(M; A) if and only if it preserves all A-submodules; that is, if it is an element of A. Let now α be a semiring, µ a semimodule. Let ι ,→ α be an ideal. 5.9 Definition. We say that ι contracts µ if it acts by contracting endomorphisms, or equiv- alently, every ideal of µ is ι-invariant. If ι = α, we say that µ is a contracting α-module. If also µ = α, we say simply that α is contracting (as a semiring). In particular, α is contracted by an ideal ι if and only if ι ≤ 0, and α itself is contracting if and only if 0 is a maximal element. Let Modα{ι} denote the full subcategory of Modα on whose objects ι contracts. This sub- category is closed under limits and the tensor sum, and so its inclusion has a lax monoidal left adjoint Modα → Modα{ι}, µ 7→ µ{ι}, the contraction functor. In particular, α{ι} is an α-algebra, and an α-module µ is contracting if and only if its action factors through the structure homomorphism α → α{ι}. In other words, Modα{ι} really is the category of modules over the contraction α{ι} of α. The inclusion into 1 2 Ring of the full subcategory 1 2 Ring≤0 of contracting semirings com- mutes with limits and colimits, and hence has left and right adjoints Left : α 7→ ◦α := α{α} Right : α 7→ α◦ := α≤0 and unit and counit α◦ ,→ α → ◦α. We will also write ◦(−) := (−){α} ∼= − ⊕α ◦α for the corresponding functor Modα → Mod◦α; but beware that this notation hides the de- pendence on α. 27 5.10 Definition. The subring α◦ is the semiring of integers of α. The (universal) contracting quotient is ◦α. 5.11 Proposition. The semiring of integers functor commutes with limits and filtered col- imits. The contraction functor Modα → Modα{ι} defined above can be described explicitly in terms of the ind-adjoint to µ → µ{ι} (compare §3.5.2). To be precise, the semiring homomor- phism α◦ → α◦[ι] induces a homomorphism (−)[ι] : Bµ = B(µ; α◦) → B(µ; α◦[ι]), where we write B(µ; α) for the set of ideals of µ that are also α-submodules. Its right adjoint identifies the term on the right with the set of ι-invariant ideals of µ. Any α-module homo- morphism µ → ν to a semimodule ν contracted by ι factors uniquely through the image of µ in B(µ, α◦[ι]). Thus, µ{ι} ⊆ Bµ is the subset of ι-invariant ideals that are generated as such by a single element. 5.12 Lemma. The image of µ in B(µ, α◦[ι]) is uniquely isomorphic to µ{ι}. 5.13 Example. The ideal semiring B(c) A of a ring A is a contracting semiring. If B is any A-algebra, then B(B; A)◦ is the image of BA → B(B; A). Indeed, the additive identity of BB is precisely the image of the unit A → B of the algebra. 5.14 Example (Semivaluations). Let A be a ring. A semivaluation on A is a map val : A → α into a semiring α which is a seminorm of the underlying Abelian group, and for which It is said to be contracting or integral if α is a contracting semiring. val(f g) = val f + val g. Let A now be a non-Archimedean ring. A (non-Archimedean) semivaluation of A is a continuous valuation on A whose restriction to A+ is integral. Any such valuation factors uniquely through the adic semiring Bc(A; A+) (def. 5.23). That is, this semiring corepresents the functor Hom(B c(A; A+), −) ∼= of continuous semivaluations on A. 1 2 Val(A, A+, −) : Ringt → Set 1 2 5.15 Proposition. Let f : A → B be a ring homomorphism. The extension of scalars trans- formation B(−; A) → B(−; B) induces an isomorphism B(−; A) ⊕B(B;A) BB ∼= ◦B(−; A) ∼= B(−; B) of functors ModB → ModB(B;A), and similarly ◦B c(−; A) ∼= B c(−; B) as functors ModB → ModBc(B;A). 28 Proof. Let M ∈ ModB. We will see that the morphism B(M; A) → B(M; B) satisifies the uni- versal property of ◦B(M; A). Let p : B(M; A) → α be a B(B; A)-module map. Precomposing with the forgetful map Bf † : B(−; B) → B(−; A) gives a map pBf † : B(M; B) → α. Now Bf †Bf is not the identity on B(−; A), but the endomorphism id + A ≥ id. However, since α is contracting, the diagram B(M; B) Bf ● ● ● pBf † ● ● ● ● ● B(M; A) p #● / α nonetheless commutes. In other words, pBf † exhibits B(−; B) as ◦B(−; A). As for the finite version, since BBc(−; A) ∼= B(−; A), applying B across the board embeds the picture into the one above. 5.2.1 Freely contracting semirings Let α be a contracting semiring. The forgetful functor¡ 1 2 Ringα → Set com- mutes with limits and therefore has a left adjoint α{−}. It is the composite of left adjoints α 7→ α[−] 7→ ◦α[S]. 2 Ring≤0¢α → 1 5.16 Definition. Let α be a contracting semiring, S a set (or α-module); α{S} is called the freely contracting semiring on S. If α is any semiring, we may also write α{S} := α ⊕α◦ α◦{S}. Note α{S} ∼= ◦(α◦[S]) ⊕α◦ α ∼= α[S]/(S ≤ 0) = α[S]/(S ∨ 0 = 0) (semiring quotient). The freely contracting functor commutes with colimits; in particular we have the base change and composition α{S} ∼= α ⊕ B{S} α{S ⊔ T} = α{S} ⊕α α{T} 2 Ring. for any α ∈ 1 5.17 Example. If A is a complete DVR with maximal ideal m, then its ideal semiring Bc A is freely contracting on the element m. This can be understood as an explicit construction of a freely contracting semiring on one element. More generally, B{S} for arbitrary S can be described as the semiring of monomial ideals in a polynomial ring k[S] on the same set of variables. 5.18 Example. Let ∆ = [−∞,0] denote the infinite half-line, and consider the semiring CPAZ(∆, R∨) of its convex, piecewise-affine functions with integer slopes (e.g. 3.3). It is generated over R∨ by a single, contracting element X . However, this generation is not free: it satisfies additional relations, such as n(Y1 ∨ Y2) = nY1 ∨ nY2 29 # O O / for all n ∈ N and Yi ∈ CPAZ(∆, R∨). We can see that these relations are not satisfied in R∨{X } by thinking of it as the set of monomial OK {x}-submodules of K{x}, where K is any non-Archimedean field with value group K = R. The key difference between free semirings and function semirings is that the latter are cancellative, while the former are not. In the present example, cancellativity can be en- forced by imposing the above list of relations in R∨{X }. The resulting universal cancellative quotient R∨{X } → CPAZ(∆, R∨) is infinitely presented. In particular, CPAZ(∆, R∨) is not a finitely presented R∨-algebra. 5.3 Projective tensor sum The join of two continuous B-module homomorphisms is continuous. The category of topo- logical B-modules is therefore enriched over ModB. We extend this to an internal Hom func- tor by equipping the continuous homomorphism B-module HomModB,t(α, β) with the weak topology with respect to the evaluation maps evX : f 7→ f (X ) for X ∈ α. In other words, it carries the topology of pointwise convergence. A fundamental system for this topology is given HomModB,t(α, β)u := {UX,Y := {f f (X ) ⊆ Y }X ∈ α,Y ∈ βu}, a formula that should evoke the compact-open topology of mapping spaces in general topol- ogy. 5.19 Example. This is not the only reasonable way of topologising the continuous Hom B- module, though it is of course the weakest. For instance, one could also define a topology of uniform convergence as the weak topology with respect to the natural embedding Hom(α, β) → Hom(Bα, Bβ), where the right-hand term is equipped with the usual topology. These topologies are in general inequivalent; in fact, this embedding is not always continuous in the topology of pointwise convergence. For example, a net {fn}n∈N in Hom(Z∨, Z∨) tends to −∞ as n → ∞ if and only if fn(x) → −∞ for all x ∈ Z. For the same net to die away in Hom(BZ∨, BZ∨), in addition {supx∈Z fn(x)}n∈N must tend to −∞ (and in particular, be finite for cofinal n ∈ N). We can also extend the monoidal structure to ModB,t. The projective topological tensor sum of topological B-modules α, β is tensor sum ?α⊕?β equipped with the strong topology with respect to the maps eY : α → α ⊕ β, X 7→ X ⊕ Y for Y ∈ β and e X for X ∈ α. If α, β are lattices, a fundamental system is generated by elements X ⊕ β ∨ α ⊕ Y , X ∈ α u,Y ∈ β u. It is more difficult to give a fundamental system for general α and β. 30 5.20 Example. The ideal B-module functor B is not lax monoidal for the projective topology. For instance, the B-module Z∨ ⊕ Z∨ is topologised so that a net X n ⊕ Yn dies away if and only if either X n dies and Yn is bounded, or vice versa. However, from the description of the fundamental system it follows that for the same net to die away in BZ∨ ⊕ BZ∨ it is enough that either X n or Yn does. The natural lattice homomorphism BZ∨ ⊕ BZ∨ → B(Z∨ ⊕ Z∨) is discontinuous. It is, however, lax monoidal on bounded B-modules, and in particular, lattices. 5.21 Proposition. The topological tensor sum and continuous internal Hom define a closed, symmetric monoidal structure on ModB,t extending that of ModB. Proof. We only need to check that the unit and counit maps of proposition 5.1 are continu- ous. For the unit α → Hom(β, α⊕β), which by the definition of the projective topology factors through the continuous Hom module, it is enough that the compositions e X : α → α ⊕ β with the evaluations at X ∈ β are continuous. Continuity of the counit is similarly tautologi- cal. 5.22 Proposition. Let α → β be strong. Then for any topological B-module γ, α ⊕ γ → β ⊕ γ is strong. Proof. This follows from the fact that if f g and g are strong (families of) maps, then f is strong. 5.23 Definitions. A topological semiring is a commutative algebra in (ModB,t, ⊕). A topo- logical semiring α is adic if αu is stable in Bα under addition, that is, if addition by an open element is an open map (def. ??). The category of adic semirings and continuous homomor- phisms is denoted 1 2 Ringt. By proposition 5.22, it is stable in the category of all topological semirings under tensor sum. In the sequel, all semirings will be assumed adically topologised, and so we will typi- cally omit the adjectives 'topological' and 'adic'. A non-Archimedean ring A (def. 2.2), resp. homomorphism f : A → B, is adic if and only if Bc(A; A+) is adic, resp. Bf is strong. 5.24 Example. The semifields H∨ associated to totally ordered Abelian groups (e.g. 5.3) are adic with respect to the topology of e.g. 4.3. All our examples of adic semirings will be adic over some H∨. The convergence condition for such semirings will therefore be that a net X n ∈ α converges to −∞ if and only if for each 'constant' r ∈ H∨, cofinally many X n ≤ r in α. For instance, the semirings R∨ → CPA∗(X , R∨) (e.g. 4.8) are of this form. Any continuous semiring homomorphism H∨ → B (where B is as always discrete) is an isomorphism. On the other hand, if H ⊆ R has rank one, then there is always a unique homomorphism H◦ ∨ → B, the reduction map. One can still define this map for general totally ordered semifields, but it is no longer unique. 5.25 Example. An element of definition of an adic semiring α is a principal open I ∈ αu ∩ α such that α is Z◦ ∨-adic with respect to the induced homomorphism Z◦ ∨ → α, −1 7→ I. 31 The join of two elements of definition is an element of definition. If α is Noetherian and has an element of definition, there is therefore also a largest element of definition, and hence a ∨-algebra structure on α. It thereby attains also a canonical reduction canonical largest Z◦ ∨-algebra structure need not be unique or functorial, even for B. Note that this Z◦ α = α◦ ⊕Z◦ adic semiring homomorphisms. ∨ If X is any Noetherian formal scheme, Bc OX attains a canonical Z◦ ∨-algebra structure, X . Again, this is not to say that Bc defines a functor with and the reduction BcOX values in AlgZ◦ 5.26 Example. The free and freely contracting semirings α[X ], α{X } over an adic semiring α are topologised adically over α. ∼= Bc O . ∨ Let A be a non-Archimedean ring. The convergent power series ring A{x} may be con- structed as a certain completion of A[x]; in terms of semirings, it is the completion with respect to the topology induced by A[x] → B c A[X ] → B c A{X }, where the left-hand map is the unique valuation sending x to X . 5.27 Example (Discrete valuations). Let X be an irreducible variety over a field k. A classic result of birational geometry states that 'algebraic' discrete valuations val : K → Z∨ on the function field K of X , integral on OX , are in one-to-one correspondence with prime Cartier divisors on blow-ups of X . this way as the formal spectrum of the completed ring of integers. We can couch this correspondence in terms of semiring theory as follows. Let U := v on K, then provided that the associated residue field is of the correct dimension over k More specifically, let eX → X be a blow-up, D ⊂ eX a prime Cartier divisor, and consider the formal completion i : bD → eX . Then the order of vanishing against D is a continuous discrete valuation on the sheaf i∗K of ObD-modules. Conversely, given any discrete valuation (the algebraicity condition), one can construct the generic point of a bD giving rise to v in eX \ D, and consider (bOU;bOeX ) as a sheaf of non-Archimedean O-algebras on the completion bD. The reduction D corresponds to an invertible element I ∈ Bc(bOU;bOeX ), and induces an of semirings over bD; here Z∨ denote the locally constant sheaf. By Krull's intersection theorem,Tn∈N I n = 0, that is, ν† preserves infima. It therefore c(bOU;bOeX ) adic homomorphism has a left adjoint ν† : Z∨ ,→ B ν : B c(bOU;bOeX ) → BZ∨ = Z∨ ⊔ {∞}, J 7→ inf{n ∈ NJ ≤ nI}. In fact, this adjoint is finite (i.e. does not achieve the value ∞), since every section of OU becomes a section of OeX after multiplication by a power of I; moreover ν−1(−∞) = {−∞}. We have therefore defined a complete, discrete norm over bD. For this norm to define a valuation, the left adjoint ν must commute with addition. In general this property is much more delicate than the existence and finiteness of ν. In ν :bOU → Z∨ 32 our setting, a study of the local algebra shows directly that this happens exactly when D is prime. In this case, if Spec A = V ⊆ X is an affine subset meeting D, then localisation induces an extension K → Z∨ of the induced discrete valuation on A. This extension is not left adjoint to the obvious map Z∨ → Bc(K; OX ), which is typically infinite (not to mention discontinuous). For the converse statement, note only that discrete valuations on K, integral on some model X = Spec A, are the same thing as homomorphisms v : B c(K; A) → Z∨. This homomorphism has a (discontinuous) right ind-adjoint v†; the algebraicity condition is equivalent to this ind-adjoint being the extension of an ordinary adjoint, in which case v†(−1) is a finitely generated ideal on Spec A which may be blown up to obtain D. 5.3.1 Projective tensor product Let A be a non-Archimedean ring. The projective tensor product M ⊗A N of locally convex A-modules M and N is strongly topologised with respect to the map c(M; A+) ⊕ B B c(N; A+) → B(M ⊗A N; A+). We can describe this topology in terms of linear algebra alone: it is the strong topology with respect to the maps e y : M → M ⊗A N, x 7→ x ⊗ y for y ∈ N, and similarly e x for x ∈ M. A sequence converges to zero in M ⊗A N if and only if it is a sum of sequences of the form xn ⊗ y and x ⊗ yn, where xn and yn converge to zero in M and N, respectively. With this definition, the monoidal functoriality of the free B-module Bc spelled out in §5.1.1 lifts to the topological setting; for example, Bc(M; A+) is a topological Bc(A; A+)- module. The corresponding statements for B are false unless A = A+. Similarly, we topologise Hom(M1, M2) weakly with respect to HomA(M1, M2) → HomModB,t(B c M1; B cM2), f 7→ B c f . A sequence of maps {fn}n∈N converges to zero if and only if for every finitely generated submodule N ⊆ M1, every sequence xn ∈ fn(N) converges to zero. This 'finite-open' topology is the weak topology with respect to the evaluation maps evx : HomA(M1, M2) → M2, f 7→ f (x) for x ∈ M1. 6 Localisation Let C be a category with filtered colimits, M an object. In this setting, we can define the (free) localisation of M at an endomorphism s ∈ EndC(M) as the sequential colimit M[s−1] := colimhM s 33 → M s → · · ·i It is universal among objects under M for which s extends to an automorphism. More generally, by composing colimits the localisation with respect to any set S of commuting endomorphisms is defined. If C is a category of modules over some algebra A, then in particular we can localise modules with respect to an element s ∈ A. If A is commutative, and M carries its own A-algebra structure, then the localisation M[s−1] is also an (M-)algebra. The general theory specialises to the case of topological semirings; we write α[−S] for the localisation of α at an element S. 6.1 Example. Let A be a domain, Bc A the finite ideal semiring. If s ∈ A, then Bc A[−(s)] ∼= Bc(A[s−1]; A). In order to obtain the ideal semiring of A[s−1], we need to enforce a contrac- tion (s) ≤ 0. 6.2 Example. Suppose that S ∈ α is open. Then α[−S] is an adic α-algebra (def. 5.23). This corresponds to the fact that if A is an adic, linearly topologised ring, and f ∈ A generates an open ideal, then A[f −1] is an adic A-algebra. 6.3 Definition. A topological semiring is Tate if α◦ is adic, and α is a free localisation of α◦ at an additive family of open elements. The full subcategory of 1 2 Ringt whose objects are Tate is denoted 1 2 RingT. In particular, any contracting semiring is Tate. A non-Archimedean ring A is Tate (def. 2.2) if and only if Bc(A; A+) is. 6.1 Bounded localisation In non-Archimedean geometry, localisations must be supplemented by certain completions, which control the radii of convergence of the inverted functions. For the geometry of skeleta to reflect analytic geometry, there must therefore be a corresponding concept for semirings. 6.4 Definition. Let α be an adic semiring. An element T ∈ α◦ that is invertible in α is called an admissible bound, or simply a bound. Invertible elements S = (+(−S))−1(0) in adic semirings - in particular, bounds - are al- ways open. A localisation µ → µ[−S] is adic if and only if S is open. If T ∈ αu is an open ideal, then S is open as an endomorphism of the semiring quotient µ/(T ≤ S) = µ/(T ∨ S = S), since T ≤ S forces S to be open. The bounded localisation µ → µ/(T ≤ S)[−S] is therefore adic. 6.5 Definition. Let S ∈ α◦ and T ∈ α a bound. Let µ be an α-module. A bounded localisation of µ at S with bound T is an α-module homomorphism µ → µ{T − S} = µ[−S]{T − S}, universal among those under which S becomes invertible with inverse bounded (above) by −T. It is called a cellular localisation if T = 0. 34 If T ≤ S in α◦, then the bounded localisation is isomorphic to an ordinary, or free locali- sation. In this case, we will often call it a subdivision. Note that only free localisations at elements that are bounded below by an admissible bound are allowed. More generally, the above definition makes sense if we replace S with an arbitrary additive subset of α◦ and T with an additive set of bounds in bijection with S. 6.6 Lemma. Any bounded localisation can be factored as a cellular localisation followed by a subdivision. Proof. Factor α → α{T − S} as α → α{T − (S ∨ T)}{−(S − (S ∨ T))}. In fact, this factorisation is natural in α, S, and T. 6.7 Example (Intervals). Consider the semiring CPAZ(∆, R∨) (e.g. 3.3), and for simplicity, specialise to the case that ∆ = [a, b] is an interval with a, b ∈ Z (but see also §8.1). The admissible bounds of CPAZ([a, b], R∨) are the affine functions mX + c, m ∈ Z, c ∈ R. Since every convex function on [a, b] is bounded below by an affine function, any element of CPAZ(∆, R∨) may be freely inverted by a bounded localisation. Let us invert the function X ∨ r for some r ∈ [a, b]. The resulting semiring, which we denote CPAZ([a, r, b], R∨), now consists of integer-sloped, piecewise-affine functions on [a, b] which are convex except possibly at r. I would like to think of this as a ring of functions on the polyhedral complex obtained by joining the intervals [a, r] and [r, b] at their end- points, or alternatively, by subdividing [a, b] into two subintervals meeting at r. The affine structure does not extend over the join point. This is the motivation for the terminology 'subdivision'. More generally, the free bounded localisations of CPAZ([a, b], R∨) are in one-to-one cor- respondence with finite sequences of rationals r1,..., r k ∈ (a, b): CPAZ([a, b], R∨) → CPAZ([a, r1,..., r k, b], R∨), that is with subdivisions of [a, b] in the sense of rational polyhedral complexes. Now let's compose this with the cellular localisation at S = −(0 ∨ (X − r)). This has the effect of imposing the relation X ≤ r. In other words, the localisation is naturally CPAZ([a, r], R∨), the semiring of functions on the lower cell [a, r]. In particular, when r = a, the subdivision has no effect (since in that case X ∨ −a = X is already invertible), and the cellular localisation is just the evaluation CPAZ([a, b], R∨) → R∨ at a. The composite of both localisations can be expressed more succinctly as CPAZ(∆, R∨){X − r}, from which we can read that r is the upper bound for the interval they cut out. More generally, every cellular localisation of CPAZ([a, r1,..., r k, b], R∨) is determined by the union of cells on which a defining function vanishes. 6.8 Example. In the limiting case of the above example ∆ = R, the only functions bounded by zero are the constants R◦ ∨ = CPA∗(R, R∨)◦. The semiring CPA∗(R, R∨) therefore has no completed localisations; it is a poor semialgebraic model for the real line. 6.9 Example. We have seen (e.g. 5.18) that the semirings CPA are not finitely presented over R∨. It may therefore be easier to work instead with finitely presented models of them; for example, R∨{X } instead of CPAZ(R◦ ∨, R∨). 35 However, the free localisation theory of these semirings is much more complicated than their cancellative counterparts - it depends on more than just the 'kink set' of the func- tion being inverted. For example, inverting X ∨ (−1) and nX ∨ (−n) define non-isomorphic localisations for n > 1 (though the former factors through the latter). This could be regarded as a problem with the theory as I have set it up. I will not make any serious attempt to address it in this paper, as it does not directly affect the main results - but see e.g. 7.5. 6.10 Example. Let K be a non-Archimedean field with uniformiser t, K{x} the Tate algebra in one variable. It is complete with respect to the valuation K{x} → K ∨{X } of example 5.26. A completed localisation of the Tate algebra at x has the form K{x, t−kx−1} for some k ∈ K . This k is a bound in the sense of definition 6.4. The completed localisation is a completion of K{x}[x−1] with respect to the topology induced by its natural valuation into K ∨{X , k − X }. The number ek (or pk when the residue characteristic is p > 0) is conventionally called the inner radius of the annulus Spa K{x, tkx−1}. In other words, bounds in semiring theory arise intuitively as the 'logarithms' of radii of convergence in analytic geometry. 6.11 Example (Admissible blow-ups). Let X be a quasi-compact adic space, T ∈ Bc(OX ; O + X ) an admissible bound. Let j : X → X + be a formal model on which T is defined. Then T ≤ 0 corresponds to a subscheme of X + whose pullback to X is empty. In other words, the admissible bounds of Bc(OX ; O + X ) that are defined on X + are exactly the centres for admissible blow-ups of X + (cf. 2.2). The following elementary properties of bounded localisation are a consequence of the universal properties. 6.12 Lemma. Let α be a topological semiring, µ an α-module. i) α{T − S} is a semiring, and µ{T − S} ∼= µ ⊕α α{T − S} as an α{T − S}-module. ii) Localisation commutes with contraction. That is, µ{ι}[−S] ∼= µ[−S]{ι}. iii) Let S1, S2 ∈ α◦, T1, T2 two bounds. Then µ{T1 − S1, T2 − S2} ∼= µ{T1 − S1}{T2 − S2}. It follows also from the discussion above: 6.13 Lemma. Let α be adic. Then µ → µ{T − S} is adic. 6.2 Cellular localisation Let α be a contracting semiring. Then the only invertible element, and hence only ad- missible bound, is 0. All localisations of a contracting semiring are therefore cellular: α → α/(S = 0). 6.14 Example. Let X be a coherent topological space [FK13, def. 0.2.2.1], so that the B- module OX of quasi-compact open subsets of X has finite meets that distribute over joins. Its lattice completion BOX is the lattice of all open subsets of X (or the opposite to the lattice of all closed subsets of X ). If X is quasi-compact, then it is an identity for the meet operation on OX ; in other words, intersection of open subsets is a contracting semiring operation on OX , and X = 0. Note that this addition is idempotent. Let us describe the localisations of OX . 36 Let S ∈ OX . The inclusion ι : S ,→ X induces adjoint semiring homomorphisms ι! : OS ⇆ OX : ι∗ by composition with and pullback along ι, respectively. They satisfy the identities ι∗ι! = id and ι!ι∗ = (−) + S. The right adjoint ι∗ identifies S with 0. Moreover, any semiring homo- morphism f : OX → α with this effect admits a factorisation f = f + S = f ι!ι∗ through OS , necessarily unique since ι∗ is surjective. In other words, OS is a cellular localisation of OX at S. Alternatively, and more in the spirit of what follows, one can argue this using the right ind-adjoint f † : OX {−S} → BOX to the localisation f . This map is easier to describe in terms of closed subsets: if Z ∈ OX {−S}, then f †Z is the smallest closed subset of X whose image in OX {−S} is Z. It identifies the localised semiring with the image of the composite f † f , which is the set of subsets K ⊆ X equal to the closure of their intersections with S, K = K ∩ S. Closure puts OS in one-to-one correspondence with this set. The latter method of this example can be abstracted, in line with the methods of §3.5.2 and §5.2. Let α ∈ 1 2 Ringt, µ an α-module, S ∈ α◦. 6.15 Definitions. An ideal ι ,→ µ is −S-invariant if X + S ∈ ι ⇒ X ∈ ι. The −S-span of an ideal ι is that is, the smallest −S-invariant ideal containing ι. (+S)−n(ι), [n∈N If S is invertible, then being −S-invariant is the same as being invariant under the ac- tion of −S. In particular, the set of −S-invariant ideals of µ[−S] is the lattice B(µ[−S]; α◦[−S]) of α◦[−S]-submodule ideals of µ[−S]. Moreover, 6.16 Lemma. The right adjoint to the localisation map f → B(µ[−S]; α◦[−S]) Bµ identifies the latter with the set of −S-invariant ideals of µ. Proof. Let ι ,→ µ be −S-invariant. Every element of ι[−S] ,→ µ[−S] is of the form X − nS with X ∈ ι. If X − nS = f (Y ) for some Y ∈ µ, then f (Y + nS) = X ∈ ι and hence Y ∈ ι. This proves that f † f ι = ι. Since in the cellular localisation, −S ≤ 0, every ideal is automatically −S-invariant. By lemma 5.12, the contraction (−){−S} induces isomorphisms B(µ[−S]; α◦[−S])f→B(µ{−S}; α◦{−S}) ∼= Bµ{−S}. This identifies the cellular localisation µ{−S} with the image of µ in B(µ[−S]; α◦[−S]). We have obtained a characterisation of cellular localisations in terms of ideals: 37 6.17 Lemma. A homomorphism f : µ → ν of α-modules is a cellular localisation of µ at S ∈ α◦ if and only if f † identifies Bν with the −S-invariants of Bµ. Note only that the 'if' part of the statement follows from the fidelity of B. 6.18 Example (Zariski-open formula). Let X be a quasi-compact formal scheme, i : U ,→ X a quasi-compact open subset. Let I be a finite ideal sheaf cosupported inside X \ U. The restriction ρ : Bc OX → i∗Bc OU evidently factors through Bc OX {−I}. Now suppose that U = X \ Z(I) is exactly the complement of the zeroes of I. Then ρ† identifies i∗Bc OU with the sheaf of subschemes Z ,→ X equal to the scheme-theoretic closure of their intersection with U (cf. e.g. 3.26). These subschemes are the −I-invariants of Bc OX . Indeed, suppose that f is some local function on X such that f I vanishes on Z. Then over U, f I = (f ), that is, f vanishes on Z ∩U and therefore on Z. By lemma 6.17, the natural semiring homomorphism is an isomorphism. 6.3 Prime spectrum c OU B c OX {−I}f→i∗B The purpose of this section is to discuss a special case of the general theory of the following section 7, in which constructions can be made particularly explicit. It therefore perhaps would logically have its place after that section. For this reason, the discussion here is relatively informal. In algebraic geometry, the underlying space of a formal scheme can be described in terms of open primes. A strong analogy holds in the setting of contracting semirings. 6.19 Definition. Let α be a semiring. A semiring ideal ι ,→ α is an ideal and an α- submodule. It is further a prime ideal if α \ ι is closed under addition. Let α be a contracting semiring, p : α → B a (continuous) semiring homomorphism. The kernel p−1(−∞) is an open prime ideal. Conversely, given an open prime ideal p E α, one can define a semiring homomorphism α → B, X 7→½ −∞, X ∈ p X ∉ p 0, This sets up an order-reversing, bijective correspondence between the poset Specp α := Hom(α, B) and that of open prime ideals p ⊳ α. In other words, every point in the prime spectrum of a contracting semiring is represented by a B-point. Let us write D1 B for the Sierpinski space, whose underlying set is the Boolean semi- field, but equipped with the topology is generated instead by the open set {0} instead of the semiring topology. The Sierpinski space underlies the unit disc over B. We now topologise the prime spectrum of a contracting semiring α weakly with respect B, defined by identifying the underlying set of B with to the evaluation maps Specp α → D1 that of D1 B. In other words, a sub-base for the topology is given by the open sets UX := {f : α → Bf (X ) = 0}, 38 and UX ∨Y = UX ∪UY . This upgrades the prime spectrum to a contravariant functor Specp : 1 2 Ring≤0 → Top. The continuous map of prime spectra induced by a homomorphism f : α → β can be de- scribed in terms of prime ideals as Specp f : β ⊲ p 7→ f −1(p) ⊳ α, just as in the case of formal schemes. By construction the localisation morphism Specp α{−S} → Specp α induces an identifica- tion Specp α{−S} ∼= US ⊆ Specp α as topological spaces. This allows us to define a presheaf O of semirings on the site U/Specp α of affine subsets of the prime spectrum. By proposition 7.9, below, it is actually a sheaf. In summary, the prime spectrum construction allows us to contravariantly associate to each contracting topological semiring α a topological space Specp α equipped with a sheaf of semirings whose global sections are naturally α. 6.20 Examples. First, it is of course easy to describe the spectrum of a freely contracting semiring: by the adjoint property, Hom(B{X1,..., X k}, B) = Dk B is the polydisc of i=1 X i = 0 is a kind of combinatorial simplex, in the sense that its poset of irreducible closed subsets is isomorphic to that of the faces of a k-simplex. See also §8.1. dimension k over B. The open subset defined by Wk B :=Qk i=1 D1 Similar statements hold for free contracting H◦ ∨-algebras, where H∨ is a rank one semi- field. Indeed, the unique continuous homomorphism H◦ ∨ → B induces a homeomorphism Specp α ⊕H◦ ∨ B → Specp α for any α over H◦ is finite. ∨. If α is of finite type, then in particular the set underlying the spectrum 6.21 Example. The prime spectrum of a Noetherian semiring is a Noetherian topological space. As such, it has well-behaved notions of dimension and decomposition into irreducible components, cf. [Gro60, §0.2]. In particular, it is quasi-compact.5 6.4 Blow-up formula Let X be a formal scheme, I a finite ideal sheaf. The blow-up p : eX → X of X along I is constructed as ProjX RI, where RI is the Rees algebra RI :=Mn∈N I ntn ⊆ OX [t]. One associates in the usual fashion [Gro60, §II.2.5] a quasi-coherent sheaf on eX to any quasi-coherent, graded RI-module on X ; in particular, if M is quasi-coherent over OX , then 5In fact, one can conclude from Zorn's lemma that any prime spectrum is quasi-compact. I omit an argument, since anyway the definitions of this section will ultimately be superseded. 39 p∗M is associated to M ⊗OX RI. If we write B(c)(M; RI) for the set of (finitely generated) ho- mogeneous RI-submodules of M, the associated module functor is induces a natural trans- formation of functors ModRI → ModB,t over X . (c)(−; RI) → B B (c)(−; OeX ) By following the algebra through, we can obtain an explicit formula relating the subob- ideal R+ The dependence of the associated sheaf to a graded module is only 'up to' the irrelevant jects of quasi-coherent sheaves on X to those of their pullbacks to eX . I =Ln>0 I ntn. For example, let M be quasi-coherent and homogeneous over RI, and let N1, N2 ,→ M be finite, homogeneous submodules. Then N1 = N2 as sections of Bc(M; OeX ) if and only if Ni + kR+ I ≤ N j for all i, j and k ≫ 0 in Bc(M; RI). It is equivalent that the high degree graded pieces (Ni)k, k ≫ 0 agree. In other words, Bc(M; RI) → Bc(M; OeX ) descends to an isomorphism c(M; OeX ). c(M; RI)/(R+ I = 0)f→B homogeneous OX -submodules of M ⊗ RI is itself graded B Now suppose that M is quasi-coherent on X . The B-module Bc(M ⊗ RI; OX ) of finite, B c(M ⊗ RI; OX ) ∼= _n∈N c(M ⊗ I n; OX ) + nT, B where T = (t) is a formal variable to keep track of the grading. It is a module over the graded semiring B c(RI; OX ) ∼= _n∈N B in which the irrelevant ideal is written R+ By proposition 5.15, c OX )≤nI + nT (B c(I n; OX ) + nT ∼= _n∈N I =Wn∈Z>0 n(I + T). c(M ⊗ RI; OX )f→B ◦B c(M ⊗ RI; OX ){R+ B I } ∼= c(M ⊗ RI; RI) in the category of Bc(RI; OX )-modules (cf. def. 5.10 for notation). Composing these identifications, we therefore have for any M a factorisation B c(M; OX ) →µ_n∈N B c(M ⊗ I n) + nT¶(± _n∈Z>0 n(I + T))f→B c¡p∗M; OeX¢ of Bc OX -module homomorphisms. The isomorphism on the right is the general blow-up formula. In the context of adic spaces and their models, a more elegant form is available. 6.22 Proposition (Blow-up formula). Let X be an adic space, j : X → X + a quasi-compact formal model. Let I ∈ Bc j∗O + X be an ideal sheaf cosupported away from X , i.e. such that j∗I = OX . Let j : X → eX + → X + be the blow-up of X + along I. Then the pullback homomorphism c( j∗OX ; OX +) → B B c( j∗OX ; OeX +) is a free localisation at I. 40 Proof. First, the preimage of I on eX + is an invertible sheaf, and therefore invertible in Bc( j∗OX ; OeX +); hence this semiring homomorphism at least factors though the localisation Moreover, ϕ is injective; if two sections Ji − ni I, i = 1,2 become identical on eX +, then by the general blow-up formula, Ji + kI are already equal on X for k ≫ 0. c( j∗OX ; OeX +). Surjectivity, on the other hand, follows from this c( j∗OX ; OX +)[−I] → B ϕ : B 6.23 Lemma. If I is finite, then for any N ∈ Bc p∗M, N + kI is in the image of p∗ for k ≫ 0. Suppose that N is generated in degrees less than k. Then N ′ = k_i=0 Ni + (k − i)I is finite, and satisfies the inequalities p∗N ′ ≤ N + kI ≤ p∗N ′ + kR+. It is therefore a lift for N + kI. In fact, the proof of this lemma shows more: it gives a recipe for exactly which modules on X pull back to which modules on BlI X . Following this recipe yields a generalisation. First, observe thatej∗OX is the OeX +-algebra associated to the graded RI algebra j∗OX tn K I := j∗OX [t] ≃ Mn≫0 on X +. We therefore obtain surjective homomorphisms c( j∗OX ; OX +)[T] ։ B c(K I; RI) ։ B B c(ej∗OX ; OeX +) K I that the Ji ti generate. i=0 iT + Ji the RI-submodule of 6.24 Definition. Let α ⊆ Bc( j∗OX ; OX +) be a subring containing I. The strict transform in which the left-hand arrow associates to a polynomialWk semiring eα of α is subring of Bc(ej∗OX ; OeX +) whose objects can be written as graded RI- with Jn ∈ α. It is the image of α[T] → Bc(ej∗OX ; OeX +). The strict transform semiring contains the inverse of I: it is defined by the formula submodules of K I in the form Mn∈N Jntn ⊆ K I −(p∗ I) ≃ Mn≫0 I n−1tn. The argument of lemma 6.23 therefore establishes: 6.25 Corollary. The strict transform semiring eα is a free localisation of α at I. 41 7 Skeleta 7.1 Spectrum of a semiring Let 1 2 Ring denote the category of Tate semirings (def. 6.3; the subscript T is held to be implicit from hereon in), Skaff its opposite. We say that a morphism f : X → Y in Skaff is an open immersion if it is dual to a bounded localisation f ♯ : OY → OY {Ti − Si}k i=1 of the semiring OY dual to Y at finitely many variables Si, Ti ∈ OY . Paraphrasing lemma 6.12 above: 7.1 Lemma. The class of open immersions is closed under composition and base change. Following the general principles outlined in the preliminaries §2.1, and in more detail in [TV09], we obtain the structure of a Grothendieck site on Skaff generated by those finite canonical covers of the form {Ui → X }k i=1 where Ui → X is an open immersion for each i ∈ [k]. The tautological presheaf O of Tate semirings on Skaff is a sheaf, by the definition of canonical coverings. 7.2 Definitions. The category Skaff, considered equipped with this topology, is called the skeletal site. Its sheaf category is denoted Ske. An affine skeleton, resp. skeleton, is a representable, resp. locally representable sheaf on the skeletal site (cf. def. 2.1, [TV09, def. 2.15]). If α is a semiring, the dual affine skeleton is called its spectrum and denoted Spec α The category of skeleta is denoted Sk. Of course, the Yoneda embedding identifies Skaff with the category of affine skeleta. More general arguments (cf. §2.1) equip each skeleton X with a small topos Xe, equiva- lent to the category of sheaves on a uniquely determined sober topological space with lattice of open sets U/X . I will abuse notation and denote this topological space also by X . This fact allows us to alternatively interpret the Grothendieck site structure on Skaff in terms of a contravariant functor Skaff → Top into the category of sober, quasi-compact, and quasi-separated topological spaces equipped with a sheaf of Tate semirings. A skeleton is then a topological space X equipped with a sheaf OX of Tate semirings, locally isomorphic to an affine skeleton. The sections of OX may be called convex functions on X . 7.3 Proposition ([TV09]). An affine skeleton is qcqs and sober, and affine open subsets form a basis for the topology. The category of skeleta has all fibre products. 7.4 Example. The spectrum ∆[a,b] = SpecCPAZ([a, b], R∨) of the semiring of convex, piecewise- affine functions on an interval [a, b] ⊆ R with rational endpoints (cf. e.g. 6.7) is homeomor- phic to a certain Grothendieck site structure on the poset of closed subintervals of ∆ ⊂ R. Indeed, we already observed in example 6.7 that every subdivision of ∆[a,b] is deter- mined by a subdivision of [a, b] as a rational polyhedron; meanwhile, by the cellular cover 42 formula of the next section (proposition 7.9), the cellular topology of ∆[a,r1,...,rk,b] is generated by the inclusions [r i, r i+1] → [r0, r k]. It remains to say when a collection of affine subsets U j = {[a ji, b ji]}k j i=1 covers X . Conjecture. The U j cover X if and only if [a, b] =Si, j[a ji, b ji] and (a, b) =Si, j(a ji, b ji). With the cellular cover fomula, the only part in question is the condition for a family of subdivisions to cover ∆. The proposed criterion says that a collection of subdivisions covers if and only if there are no common 'kink' points, that is, if \j k j[i=1 {a ji, b ji} = ;. Indeed, in that case, the intersection over j (in, say, the set of continuous functions) of the semirings of piecewise-affine functions convex on U j is exactly the set of such functions convex on ∆. In other words, CPAZ(∆, R∨) →Yi OUi ⇒Yi, j OUi ∩U j is an equaliser of semirings. I do not know how to show that this equaliser is universal. As was pointed out in e.g. 6.8, the spectrum of CPA∗(R, R∨) consists of a single point. One can obtain a better model for the affine real line R as the increasing union in Sk. Like the analytic torus over a non-Archimedean field, it is not quasi-compact. skR := [a→∞ [−a, a] 7.5 Example (Dichotomy). The skeleton constructed in the above example 7.4, although relatively easy to describe, is not finitely presented over Spec R∨ (cf. 5.18). In the vein of example 6.9, we can replace CPAZ([a, b], R∨) with its finitely presented cousin R∨{[a, b]} := R∨{X − b, a − X }. They are related by an (infinitely presented) morphism SpecCPAZ(∆, R∨) → Spec R∨{[a, b]}. We also saw in 6.9 that this morphism is not a homeomorphism, and that in fact the topology of the target is difficult to describe. It seems to be possible to modify the definition of skeleton, by introducing another con- dition into our definition of semiring, so as to make this morphism a homeomorphism. This condition is the semiring version of the algebraic notion of relative normality (integral clo- sure of A+ in A), which is used in non-Archimedean geometry to define a good Spec functor. However, I wish to defer a serious pursuit of this approach to a later paper, since this issue does not directly affect any of the results here. The examples in §8 all more closely resemble finitely presented skeleta like Spec R∨{∆}, but I will often only describe open subsets in a way that depends only on their pullbacks to a 'geometric' counterpart SpecCPAZ(∆, R∨). 43 7.2 Integral skeleta and cells 7.6 Definition. A skeleton that admits a covering by spectra of contracting semirings is said to be integral. The full subcategory of Sk whose objects are integral is denoted Skint. The tropical site Skaff carries a tautological sheaf O ◦ of contracting semirings, whose sections over Spec α are α◦. Taking the spectrum defines a functor Spec O ◦ : Skaff → Skint. Moreover, any covering of an object Spec O ◦(X ) lifts, by base extension O ◦ → O , to a covering of X , that is, Spec O ◦ is cocontinuous. It therefore extends to the pushforward functor of a morphism of the corresponding topoi. (−)◦ : Ske → (Skint)e an affine open cover X =Si Ui such that (Ui ∩ U j)◦ ,→ U ◦ This functor takes a skeleton X to an integral skeleton X ◦ if and only if there exists i is an open immersion, in which In algebraic terms, we need that X admit an affine case U ◦ atlas each of whose structure maps is dual to a localisation α → α{T − S} that restricts to a localisation α◦ → α{T − S}◦ ∼= α◦{T − S} of the semiring of integers. This occurs if and only if the localisation is cellular, that is, if (up to isomorphism) T is invertible in α◦ and therefore zero. • provides an atlas for X ◦. 7.7 Definitions. An open immersion of skeleta is cellular if it is locally dual to a cellular localisation of semirings. A skeleton that admits a cover by affine, cellular-open subsets is said to be a cell complex. In particular, any affine skeleton is a cell complex. If every open subset is cellular, it is a spine. By the discussion above, any integral skeleton is a spine. The categories of spines, resp. cell complexes are denoted Sksp ,→ Skcel. There is a functor (−)◦ : Skcel → Skint, left adjoint to the inclusion, which associates to a cell complex X its integral model X ◦. The unit of the adjunction is a morphism j : X → X ◦. The cellular open subsets of X are those pulled back along j. We have access to a reasonably concrete description of the 'cellular topology'. 7.8 Lemma. Let α be a semiring, {Si}k i=1 ⊆ α◦ a finite list of contracting elements. Write is a universal equaliser of semirings. Proof. The lemma 6.17 yields an embedding of forks α Bα Qi α{−Si} /Qi Bα 44 Qi, j α{−Si, −S j} /Qi, j Bα i=1 Si. Then S =Wk α{−S} →Yi α{−Si} ⇒Yi, j α{−Si, −S j} / /   / / / /     / / / / in which the ith arrow in the lower row takes an ideal to its −Si-span. Let eq ⊆Qi Bα denote the equaliser of the second row, f : Bα → eq the natural B-module homomorphism. An element of eq is a finite list {ιi}k i=1 of −Si-invariant ideals, such that for each i and j the −S j-span of ιi is equal to the −Si-span of ι j. The right adjoint f † to f sends such a list to their intersection in α. Since localisation commutes with base change, the fork in the statement is a universal equaliser as soon as it is an equaliser. By 6.17, it is equivalent to show that f † identifies eq with the set of −S-invariant ideals of α. On the one hand, the elements f †(eq) are certainly −S-invariant. Indeed, suppose X + nS ∈ ι = ∩iιi. Then X + nSi ≤ X + nS ∈ ιi, and so X ∈ ιi for each i. Furthermore, since the −Si-span of ι j contains ιi, if X ∈ ιi, then X + nSi ∈ ι j for some n. Therefore, for n ≫ 0, X + nSi ∈ ι, and ιi is the −Si-span of ι. Conversely, suppose that ι is −S-invariant. Let X ∈ f † f ι ⊇ ι. Then for n ≫ 0, X + nSi ∈ ι for all i, and therefore X + nS ∈ ι, so X ∈ ι. Therefore, f and f † are inverse. In geometric terms: 7.9 Proposition (Cellular cover formula). Let α be a semiring, {Si}k i=1 ⊆ α◦ a finite list of contracting elements. Write S =Wk i=1 Si. Then Spec α{−S} = as subsets of Spec α. Spec α{−Si} k[i=1 7.10 Corollary. Let U be a quasi-compact cell complex. If U can be embedded as an open subset of an affine skeleton, then U is affine. In fact, this result can be greatly improved. 7.11 Theorem. Let X be a quasi-separated cell complex, j : X → X ◦ its integral model. Let us confuse X ◦ with its site U qc /X ◦ of quasi-compact open subsets. Then: i) B( j∗OX ) is flabby; ii) X is affine if and only if it is quasi-compact and j∗OX is flabby. 7.12 Corollary. Any quasi-compact, integral skeleton with Noetherian structure sheaf is affine. Proof. In this case X = X ◦ and OX = BOX . 7.13 Corollary. Let H∨ be a rank one semifield. Any integral skeleton finitely presented over Spec H◦ ∨ is affine. Proof. Such admits a model over some finitely generated subring of H∨. If we make the assumption that all H∨-algebras α satisfy α ∼= α◦ ⊕H◦ H∨, then this last corollary applies also to any cell complex finitely presented over Spec H∨ (which is, in this case, simply the base change of its integral model). ∨ 45 Proof of 7.11. Let f : U• ։ V be a finite, affine, cellular hypercover of some quasi-compact V ⊆ X . By corollary 7.10, we may in fact assume that U• is the nerve of an ordinary cover i=1 Ui ։ X . Let α = Γ(V , OX ). We have an equaliser ⊔i f i :`k α → αi ⇒ αi{−Si j} kYi=1 kYi, j=1 commuting with isomorphisms αi{−Si j} ∼= α j{−S ji}. There are unique elements S j ∈ α whose images in αi are Si j. Since αi → αi{−Si j} is surjective, its right ind-adjoint is injective. The compositions αi ρ → αi{−Si j}f→α j{−S ji} ρ† → Bα j therefore together yield a section Bαi → Bα• of the projection. Since this holds for any quasi-compact V , B( j∗OX ) is flabby. Now set V = X . For the second part, it will be enough to show that each f i is a cellular localisation of α at Si, since in this case the equaliser will be a covering, and hence induce an isomorphism Spec αf→X . We will show this using the characterisation 6.17. : Bαi → Bα has image in the set of −Si-invariant ideals. Since OX is i is also injective. We need only show that it is surjective. The argument is based Certainly, f † i flabby, f † on two lemmata. 7.14 Lemma. Let f : α → β, S ∈ α◦. If f is surjective, Bf preserves −S-invariance. 7.15 Lemma. Let f : µ → µ{−S} be a cellular localisation of α-modules, g : µ → ν a surjective homomorphism. The diagram f † f † µ{−S} g ν{−S} g µ ν commutes. Proof. The right adjoints embed µ{−S}, ν{−S} into Bµ, Bν as the set of −S-invariant ideals, which are preserved under g by 7.14. Let ι ,→ α be a −Si-invariant ideal, ι• its image in α•. By 7.14, ι• is −S•i-invariant, and hence ι• = f † i f iι• = f † i f•ιi = f• f † i ιi where the last equality follows from 7.15. Therefore ι = ι• = f † i ιi. Let α be a contracting semiring. The arguments of §6.3 show that we have a natural continuous functor U/Spec α → U/Specp α and hence a morphism of semiringed spaces Specp α → Spec α. If Specp α is quasi-compact for all α ∈ 1 2 Ring≤0, then this is an isomorphism by the unicity of canonical topologies. This is true for Noetherian semirings by 6.21. The general case is implied by Zorn's lemma. 46     o o o o semiringed spaces. 7.16 Proposition. Let α be a (Noetherian) contracting semiring. Then Specp αf→Spec α, as Topologising as before the set X (B) weakly with the respect to evaluations X (B) → D1 B, we obtain: 7.17 Corollary. Let X be an integral skeleton. Then X (B) → X is a homeomorphism. 7.3 Universal skeleton of a formal scheme 7.18 Example. Let us return to the quasi-compact, coherent space X of example 6.14 and its semiring OX of quasi-compact open subsets. We saw there that the inclusion S ,→ X of a quasi-compact subspace induces a localisation OX → OS at S. The cellular cover formula i=1 Si is a finite union of open subsets, then 7.9 implies that if S =Sk It follows that the functor Spec OS = Spec OSi . k[i=1 Spec O : U qc /X → U qc /skX ,→ Skaff /skX preserves coverings, and hence induces a homeomorphism of X with skX := Spec OX . As we have seen (cor. 7.17), every point of skX is represented uniquely by a B-point. In fact, the stalk of the structure sheaf OX at any point p ∈ skX is canonically isomorphic to B, with 0 (resp. −∞) represented by an open subset containing (resp. not containing) p. Under the homeomorphism skXf→X , OX can be identified with the semiring C0(X , D1 of continuous maps from X to the Sierpinski space D1 functions of open subsets. B) B, that is, with the set of indicator It follows from the functoriality of the sheaves Bc OX associated on formal schemes X , as outlined in sections 3, 4, 5, that they assemble to a sheaf O of contracting semirings on the large formal site (in fact, with the fpqc topology). Its sections over a quasi-compact, quasi-separated formal scheme X are the semiring of finite type ideal sheaves on X . This can be thought of as a geometric version of the sheaf O of the above example, which is simply an avatar of the correspondence between (certain) frames and locales. Let X be any formal scheme, U qc /X its corresponding small site, OX the restriction of O . The Zariski-open formula 6.18 implies that if V ,→ X is a quasi-compact open subset, then OX puts the (necessarily cellular) bounded localisations of OX (V ) into one-to-one correspondence with quasi-compact Zariski-open subsets of V . The cellular cover formula implies that if I i ∈ OX (V ) is a finite family of finite-type ideal sheaves on X , Ui = V \ Z(I i) the complementary quasi-compact opens, and then as subsets of Spec OX (V ). In other words, U 7→ Spec OX (U) defines a cover-preserving equivalence of categories between U/Spec OX (U) and U/X . This proves: Ui, k[i=1 U = V \ Zà k_i=1 I i! = k[i=1 Spec OX (U) = Spec OX (Ui) 47 7.19 Lemma. Let X be a quasi-compact formal scheme. Then Spec O(X ) → X is a homeo- morphism. 7.20 Theorem. Let X be any formal scheme. Then skX := (X , OX ) is a skeleton. Of course, skX is actually an integral skeleton. 7.4 Universal skeleton of an adic space Let Adqcqs denote the quasi-compact, quasi-separated adic site. The sheaves Bc(OX ; O + X ) on each adic space X assemble to a sheaf O of Tate semirings on Adqcqs, extending the one with the same name introduced in the previous section. Note that, unlike the case of formal schemes, this sheaf does not restrict to the presheaf pre O = B c : nA → 1 2 Ring defined in terms of the section spaces of O, since the O +-submodules it parametrises are, on the whole, not quasi-coherent. Naturally, O is the sheafification of O pre. In this section, we will derive the following generalisation of theorem 7.20: 7.21 Theorem. Let X be any adic space. Then skX := (X , OX ) is a skeleton. 7.22 Definition. The skeleton skX is called the universal skeleton of X . The proof rests on a limit formula, following from the fundamental limit 1 of §2.2. 7.23 Lemma. Let X be a qcqs adic space. Then B c(OX ; O + X ) ∼= colim j∈Mdl(X) B c( j∗OX ; j∗O + X ) in 1 2 Ringt. Proof. Indeed, the limit formula states explicitly that j : Xf→lim X + as locales, and that X as sheaves on X . Any finitely generated ideal of O + X = colim j∗ j∗O + O + back from some level j∗O + X . X is therefore pulled Since, by +normality, the morphisms j∗ j∗O + X → OX are injective, then any two such ideals have the same image in OX if and only if they agree on any cover, that is, on any model on which they are both defined. Proof of 7.21. Let X ∈ Adqcqs. We need to show that the localisations of O (X ) are in one- to-one correspondence with the quasi-compact subsets of X . Let S ,→ X be a quasi-compact subset. There exists a formal model j : X → X + and open subset S+ ,→ X + such that S ∼= X ×X + S+, and B c( j∗OX ; j∗O + X ) → B c( j∗OS; j∗O + S ) is a cellular localisation at some (any) finite ideal I cosupported on X + \ S+. This remains true when we modify X +. Since S is quasi-compact, every formal model jS : S → S+ can be extended to a model j of X , and so the colimit formula 7.23 implies that O (X ) ∼= colim j∈Mdl(X)/X + B c( j∗OX ; j∗O + X ) → colim j∈Mdl(X)/X + B c( jS∗OS; jS∗O + S ) ∼= O (S) 48 is a localisation at I. Conversely, any I ∈ O (X ) is representable by some finite ideal sheaf on a qcqs formal model j : X → X + of X , whence O (X ){−I} ∼= O (U), where U ∼= X ×X + (X + \ Z(I)). The cellular cover formula 7.9 shows that this correspondence preserves coverings, and hence induces a homeomorphism of X with Spec O (X ). The argument also shows: 7.24 Corollary. The universal skeleton of an adic space is a spine (def. 7.7). 7.4.1 Real points of the universal skeleton Let X be any skeleton. We can topologise the set X (R∨) of real points of X with respect to the evaluation maps f : X (R∨) → R∨ associated to functions f ∈ OX , where on the right- hand side R∨ is equipped with the usual order topology (rather than the semiring topology). If X is defined over some rank one semifield H∨ ⊆ R∨, then we may rigidify by considering R∨-points over H∨; the subset X H∨(R∨) ⊆ X (R∨) similarly acquires a topology. The natural map X (R∨) → X is often discontinuous with respect to this topology. If X is now an adic space, we can consider (following e.g. 5.14) the space skX (R∨) of real points of the universal skeleton as a space of real valuations of OX . For this to be geomet- rically interesting, we usually want to consider this equipped with some H∨-structure. For instance, if X is Noetherian, then skX carries a canonical 'maximal' morphism to Spec Z∨ (e.g. 5.25). The corresponding valuations send irreducible topological nilpotents to −1. Al- ternatively, if X is defined over a rank one non-Archimedean field K → H∨, then skX is defined over H∨, and the real points are valuations extending the valuation of the ground field. Where there is no possibility of confusion, I will abbreviate skX H∨ (R∨) to X (R∨). To the reader familiar with analytic geometry in the sense of Berkovich [Ber93] the following theorem will come as no surprise: 7.25 Theorem. Let X Berk be a Hausdorff Berkovich analytic space over a non-Archimedean field K, X the corresponding quasi-separated adic space [Ber93, thm. 1.6.1]. The composi- tion X (R∨) → X → X Berk is a homeomorphism. Proof. It is enough to show that the restriction of this map to every affinoid subdomain is a homeomorphism. This follows from the definitions and the identity for affine X = Spa A, which holds because Bc(OX ; O + X ) is a localisation of Bc(A; A+). X ), R∨)f→Hom(B Hom(B c(OX ; O + c(A; A+), R∨) 7.26 Proposition. Let X be integral and adic over an adic space with a Noetherian formal model. Every function on skX is determined by its rational values. 49 In classical terms this means the following: let j : X → X + be a formal model of X , Z1, Z2 ,→ X + two finitely presented subschemes; then if for all continuous rational valua- tions val : OX → Q∨, val(I1) = val(I2), then Z1 = Z2 after some further blow-up of X +. Proof. The statement is clear when X + is Noetherian and the Zi are both supported away from X ; in this case, we may blow-up each Zi to obtain Cartier divisors, which by the Noetherian hypothesis factorise into prime divisors. Knowing that the Zi are Cartier divi- sors, they are therefore determined by the multiplicities of each prime divisor therein, that is, the values of local functions for the Zi under the corresponding discrete valuations. Moreover, any formal subschemes Zi are, by definition, formal inductive limits of sub- schemes supported away from X , and so determined by a (possibly infinite) set of valuations. Finally, for the general case we may assume that Zi are pulled back from some Noethe- rian formal scheme X + → Y + over which X + is integral. Since rational valuations admit unique extensions along integral ring maps, the discrete valuations on OY determining the Zi extend to rational valuations on OX . 7.27 Corollary. The universal skeleton of an adic space is cancellative. 7.28 Corollary. Let X be as in 7.26. Then X (R∨) satisfies the conclusion of Urysohn's lemma. Proof. The proposition implies that OX injects into the the set C0(X (R∨), R∨) of continu- ous, real-valued functions. By definition, two points of X (R∨) agree only every element of OX takes the same value at both points. In other words, distinct points are separated by continuous functions. This last result can be understood as a cute proof of the corresponding property for Hausdorff Berkovich spaces, that is, that they are completely Hausdorff. 7.5 Shells Let X be an adic space. The universal skeleton of X is a spine, so that any function with an admissible lower bound is invertible. If, for example, the skeleton is adic over Z∨, then this is the same as every bounded function being invertible. Intuitively, this means that we have not defined a good notion of convexity for functions on skX . We obtain a more restrictive notion of convexity by embedding skX into a shell, that is, a skeleton B inside which skX is a subdivision - in fact, the intersection of all subdivisions. At the level of the Berkovich spectrum X (R∨), this is akin to choosing a kind of 'pro-affine structure' (a concept that I do not define here). Suppose that X is qcqs, and let j : X → X + be a formal model of X . Write sk(X ; X +) := Spec B c( j∗OX ; j∗O + X ), for the X +-shell of X . It is an affine skeleton whose integral model is the universal skeleton skX + of X +. More generally, if X is any adic space admitting a formal model X +, then a qcqs cover U + • ։ X + with generisation U• = X ×X + U + • gives rise to an X +-shell sk(X ; X +) := sk(U•;U + • ), 50 which is a cell complex whose integral model, again, is skX +. The blow-up formula 6.22 shows that the colimit 7.23, for each qcqs U ,→ X , is in fact over all possible free localisations of Bc( j∗OU; j∗O + U). In other words, 7.29 Proposition. Let X be an adic space, j : X → X + a formal model; skX ⊆ sk(X ; X +) is the intersection of all subdivisions of sk(X ; X +). Any open subset of the X +-shell is induced by a blow-up X + i → X + followed by a Zariski- i . I do not know of any easily-checked necessary criterion to deter- open immersion U + ,→ X + • ) ։ sk(X ; X +) of mine when a family of blow-ups {X + the corresponding shells; it is certainly sufficient that the blow-up centres have no common point. i → X +}k i=1 gives rise to a cover sk(X ; X + Note that the formal model X + can be recovered from the data of X and the shell skX ,→ sk(X ; X +). Indeed, one obtains from these data the continuous map j : Xf→skX → skX + to the integral model of sk(X ; X +), X + is the formal scheme with the same underlying space X . as skX + and structure sheaf j∗O + Finally, the fact that any two models of X are dominated by a third means that any two shells of skX have a common open subshell; the shells can therefore be glued together to create a universal shell skX . In abstract terms, the functor sk is obtained by left Kan extension along the inclusion Adaff ,→ Ad of Spec O pre : Adaff → Sk, where O pre, as before, denotes the presheaf Spa A 7→ Bc(A; A+). Again, the universal shell skX contains the spine skX as the intersection of all subdivisions. The universal shell is a universal way of defining a 'pro-affine structure' on X (R∨) with respect to which the valuations of sections of OX are convex. It also supports convex poten- tials for semipositive metrics on X . 8 Examples & applications I conclude this paper with some abstract constructions of skeleta which are already well- known via combinatorial means in their respective fields. 8.1 Polytopes and fans Let N be a lattice with dual M, and let ∆ ⊂ N ⊗ R be a rational polytope with supporting half-spaces {〈−, f i〉 ≤ λi}k i=1, λi ∈ Q. We will allow ∆ to be non-compact, as long as it has at least one vertex; this means that the submonoid M∆ ⊆ M of functions bounded above on ∆ separates its points. In this case, we can compactify ∆ ⊆ ∆ in, for example, the real projective space RP(N ⊕ Z). The semiring of 'tropical functions' on ∆ is presented Z∨{∆} := Z∨[M∆]/(f i ≤ λi)k i=1; its elements have the formWd j=1 X i + ni, with X i ∈ M∆ and ni ∈ Z. 51 8.1 Definition. The semiring Z∨{∆} is the polytope semiring associated to ∆. Its spectrum sk∆ is the corresponding polytope skeleton, or just polytope if the skeletal structure is im- plied by the context. The construction sk is functorial for morphisms φ : M1 → M2, φ(∆1) ⊆ ∆2 of polytopes. In particular, every sub-polytope ∆′ ⊆ ∆ (with N fixed) induces an open immersion of skeleta sk∆′ ,→ sk∆. The morphism induced by a refinement N → 1 d N can be thought of as a degree d 'base extension' sk∆ → skd∆. The polytope skeleton sk∆ is a skeletal enhancement of ∆, in the sense that there is a canonical homeomorphism sk∆(R∨) ≃ ∆ of the real points, and surjective homomorphism Z∨{∆} → CPAZ(∆, Z∨) onto the semiring of integral, convex piecewise-affine functions on ∆ (that is, the semiring of integral, convex piecewise-affine, and bounded above functions on ∆). We can produce a continuous map sk∆ → ∆, right inverse to the natural inclusion, whose inverse image functor sends an open U ⊆ ∆ to the union skσ ,→ sk∆, [σ⊆U ranging over all polytopes σ contained in U. This map presents ∆ as a Hausdorff quotient of sk∆ (cf. thm. 7.25). Polytope semirings admit an alternate presentation, related to the theory of toric de- generations. Let N ′ = N ⊕ Z, with dual M′ ∼= M ⊕ Z, and take the closed cone σ := [λ>0 λ∆ × {λ} ⊂ N ′ ⊗ R over the polytope placed at height one. The inclusion of the factor Z induces a homomor- phism i : N → σ∨ ∩ M′ of monoids; we topologise N linearly with ideal of definition 1, and the cone monoid adically with respect to i. In other words, a fundamental system of open ideals of σ∨ ∩ M′ is given by the subsets σ∨ ∩ M′ + i(n) for n ∈ N. We find that Z◦ ∨{∆} = B{σ∨ ∩ M′} = B c(σ∨ ∩ M′) (see definitions 5.16 and 3.15 for notation) is the semiring of integers (def. 5.10) in Z∨{∆}. i=1 X i with X i ∈ σ∨ ∩ M′, subject to X i ≤ 0. Note ∨ corresponds, perhaps somewhat confusingly, to (0,1) ∈ M′. i=1(X = Its elements are idempotent expressionsWk An element S =Wk ∨{∆} corresponds to a finite union of subcones σS =Sk that under this notation −1 ∈ Z◦ i=1 X i ∈ Z◦ 0) ⊆ σ and hence of faces ∆S of ∆, and the induced restriction Z◦ ∨{∆} → Z◦ ∨{∆S} is a localisation at S. The topology of the integral model sk∆◦ = Spec Z◦ ∨{∆} of sk∆ is there- fore equal, as a partially ordered set, to the set of unions of faces of ∆. In particular, sk∆◦ is a finite topological space. A refinement of the lattice N 7→ 1 k N commutes with base extension Spec 1 k Z∨ → Spec Z∨: 1 k Z∨{∆} := 1 k Z∨ ⊕Z∨ Z∨{∆}f→Z∨{k∆}. 52 The dual morphism 1 k of integral skeleta is a homeomorphism. skk∆◦ ∼= Z◦ ∨ ×Z◦ ∨ sk∆◦ → sk∆◦ Let k[[σ∨ ∩ M′]] denote the completed monoid algebra of σ∨ ∩ M′, and write zm for the monomial corresponding to an element m ∈ σ∨ ∩ M′. I introduce the special notation t := z(0,1) for the uniformiser; the completion is with respect to the t-adic topology. The monoid inclusion σ∨ ∩ M′ ⊂ k[[σ∨ ∩ M′]] induces a continuous embedding Z◦ ∨{∆} ,→ B c¡k[[σ∨ ∩ M′]]¢ into the ideal semiring of k[[σ∨ ∩ M′]], matching −1 ∈ Z◦ ∨ with the ideal of definition (t). The formal spectrum D+ k[[t]]∆ of k[[σ∨ ∩ M′]] is an affine toric degeneration in the sense of Mumford. That is, it is a flat degeneration D+ k[[t]]∆ Spf k[[t]] of varieties over the formal disc arising as the formal completion of a toric morphism of toric varieties. 8.2 Definition. Let k be a ring. The polyhedral algebra of functions convergent over ∆ is the finitely presented k((t))-algebra k((t)){∆} := k[[σ∨ ∩ M′]][t−1]. Its (analytic) spectrum Dk((t))∆ is called the polyhedral domain over k((t)) associated to ∆. For example, if ∆ is the negative orthant in Rn, then Dk((t))∆ is just the ordinary unit polydisc Dn k((t)) over k((t)). Note that the polyhedral algebra has relative dimension equal to the rank of N, while the polyhedral semiring depends only on the lattice points of ∆ and not on the ambient lattice. A similar construction is possible in mixed characteristic. In light of the main result 7.21, there is a commuting diagram Dk((t))∆ D+ k[[t]]∆ µ∆ sk∆ / sk∆◦ in which the top and bottom horizontal arrows are morphisms of adic spaces and of skeleta, respectively, and the vertical arrows are continuous maps. If ∆ spans M, then I would like to call the leftmost arrow µ∆ a standard non-Archimedean torus fibration over ∆. This construction can be globalised to obtain torus fibrations on toric varieties and on certain possibly non-compact analytic subsets, in analogy with (and, more precisely, mirror 53   / /     / to) the symplectic theory. Let Σ be a fan in a lattice N, and let X = XΣ be the associated toric variety over a non-Archimedean field K, considered as an analytic space. Each cone ∆ of Σ corresponds to a Zariski-affine subset U∆ ⊆ X . Considering the cone as a polytope embed it in a filtered family of expansions ∆ = k\i=1 {〈−, f i〉 ≤ λi}k i=1, ∆r = k\i=1 {〈−, f i〉 ≤ λi + r i}k i=1 for r = (r i) ∈ Rk ≥0; the analytic version of the subset U∆ fits into the increasing union Dk((t))∆r Uσ of standard non-Archimedean torus fibrations. Note that it is not quasi-compact unless N = 0. By glueing, we obtain a skeleton skΣ and torus fibration sk∆r /Sr→∞ sk∆r XΣ µΣ skΣ which is covered by the standard fibrations over affine polyhedral domains sk∆r ⊆ skΣ, where ∆r ranges over all expansions of cones of Σ. 8.2 Dual intersection skeleta Let X + be a reduced, Noetherian formal scheme, and let X be the analytic space obtained by puncturing X + along its reduction X + 0 . 8.3 Definitions. The dual intersection or Clemens semiring of X + is the subring Cl(X ; X +) ,→ B c( j∗OX ; j∗O + X ) generated by the additive units of Bc( j∗OX , j∗O + X ), that is, the invertible fractional ideals of X in j∗OX . It is a sheaf of semirings on X +, and it is functorial in both X and X +. The j∗O + elements of the semiring of integers Cl(X ; X +)◦ correspond to monomial subschemes of X +. The dual intersection or Clemens skeleton of X + is sk∆(X , X +) := Spec Γ(X +;Cl(X ; X +)). It comes equipped with a collapse map X → skX → sk∆(X ; X +). 54 / /     /   It is possible, where confusion cannot occur, to drop X and/or sk from the notation. I also write Cl◦(X ; X +) and sk∆◦(X ; X +) for the semiring of integers and integral model, respectively. The flattening stratification decomposes X + =`i∈I E i into locally closed, irreducible subsets such that the restriction of the normalisation ν : eX + particular, the set underlying each monomial subscheme appears in this stratification. 0 to each E i is flat. In 0 → X + 8.4 Lemma. If E ,→ X + is a monomial subscheme with complement V +, then ClX + → ClV + is a cellular localisation at E. Proof. For this we may repeat the argument of 6.18 with 'ideal' replaced by 'monomial ideal' throughout. 8.5 Example. If X + is an affine toric degeneration associated to some polytope ∆, with general fibre X = DOK ∆, then the Clemens skeleton is sk∆(X , X +) = sk∆ and the collapse map µ is a standard torus fibration µ∆. Its real points sk∆(R∨) are the dual intersection complex of X + in the classical sense: its n-dimensional faces correspond to codimension n toric strata of X +. The commuting diagram D+ OK ∆′ D+ OK ∆ ∆′ / ∆ coming from the open inclusion of a face ∆′ of ∆ can be seen as an instance of lemma 8.4. 8.6 Definition. A normal formal scheme X + over a field k is said to have toroidal crossings is isomorphic to an if it admits a cover {f i : U + open subset of some affine toric degeneration D+ i=1 by open strata such that each U + i i → X +}k k[[t]]∆i. One can choose whether to consider étale or Zariski-open subsets for the covering, with the former being the usual choice. Zariski-local toroidal crossings is a very restrictive notion - for instance, it forces the irreducible components of X + 0 to be rational. For simplicity, I will nonetheless work with this latter notion in this section, though the arguments may be generalised with some additional work. as in the definition. Write Ui = U + given inclusions U + i Let X + be a formal scheme with Zariski-local toroidal crossings, and select model data i ×X + X . Assume, without loss of generality, that the k[[t]]∆i induce a bijection on the sets of strata, and hence isomor- i )). They identify ∆i with the dual intersection complex of identifies its dual ,→ D+ i ;Cl(Ui;U + i . By lemma 8.4, the inclusion of the open substratum U + U + intersection complex with a face ∆i j common to ∆i and ∆ j. phisms Z∨{∆i}f→Γ(U + i j := U + i ∩U + j It follows from this and the cellular cover formula 7.9 that {∆(Ui;U + i ) → ∆(X ; X +)}k i=1 is a (cellular) open cover. X µ ∆(X ; X +) i=1 Ui `k i ) i=1 ∆(Ui;U + `k 55 i=1 Di∆i µ∆i i=1 ∆i `k `k / /     /   o o o o   / /     o o o o Intuitively, ∆(X ; X +) is constructed by glueing together the dual intersection polytopes ∆i of the affine pieces U + i j. i along their faces ∆i j corresponding to the intersections U + 8.7 Proposition. Let X + be a locally toric formal scheme. Then is a cellular-open cover. In particular, ∆(X ; X +) is a cell complex (def. 7.7). ∆i j ⇒ ∆i → ∆(X ; X +) kai, j=1 kai=1 The collapse map µ is affine in the sense that ∆X + admits an open cover that pulls back X , where µ◦ : to an affine open cover of X +. It follows that X = Spaµ∗OX and X + = Spf µ◦ ∗ X + → ∆◦X + is the integral model of µ. It is locally isomorphic to standard torus fibrations µ∆i. O + blow-up with monomial centre Z ⊆ X + position Suppose that X + = D+∆ is an affine toric degeneration, and let p : eX + → X + be a toric 0 . The toric affine open cover of eX + induces a decom- of the dual intersection skeleton of eX + into polyhedral cells. The map induced by the blow-up is a subdivision at the function Z ∈ Cl(X ; X +). ∆i j ⇒ kai=1 kai, j=1 → ∆(X ;eX +) ∆(X ;eX +) → ∆(X ; X +) More geometrically, the Clemens semiring of a monomial blow-up is the strict transform semiring of the Clemens semiring of X + (cf. 6.25). It follows that monomial blow-ups induce subdivisions of the dual intersection skeleta. 8.3 Tropicalisation Let X be a toric variety, so that following §8.1 it comes with a canonical 'tropicalisation' X → skΣ. Let f : C ,→ X be a closed subspace of X . We would like to complete the composite skC ,→ skX → skΣ to a commuting square C trop X Trop(C/X /Σ)  / skΣ and to call C → Trop(C/X /Σ) the amoeba or tropicalisation of C in skΣ, after (in chronologi- cal order) [EKL07] and [Pay09]. Let us begin in the affine setting: let X = DK ∆ be a polyhedral domain, and let IC be the ideal defining C in OX . There is an associated toric degeneration j : X → X + over OK , and X . Let us we may close the subspace C to obtain an integral model C+ with ideal IC ∩ j∗O + set α∆ to be the image in Bc( j∗OC; j∗O + C) of Z∨{∆}, so that Bc( j∗OC; j∗O + C) Bc( j∗OX ; j∗O + X ) α∆ Z∨{∆} 56 / /      / o o O O o o O O commutes (here we confuse the sheaves Bc(O; O +) with their global sections). The elements of α◦ ∆ are subschemes of C+ monomial in the sense that they are defined by monomials from OX . We set Trop(C/X /∆) := Spec α∆. 8.8 Example (Plane tropical curves). Let ∆ be the lower quadrant 2 {r = (r1, r2) ∈ R r1, r2 ≤ λ} with 0 ≪ λ ∈ Z. The polyhedral domain DK ∆ = Spa K{tr1 x, tr2 y} is an arbitrarily large poly- disc in the affine plane X = A2 K . Let X + be the corresponding formal model (which is iso- morphic to A2 OK Let f =Pi, j,k ci jktkxi y j ∈ K{tr1 x, tr2 y} be some series, where (ci jk) is a matrix of con- stants in k. The 'tropicalisation' F of the function f in the polytope semiring Z∨{∆} = Z∨{X − r1,Y − r2} is ). ι †(f ) = _i, j,k iX + jY − k where ι : Z∨{∆} → Bc( j∗OX ; j∗O + X ) is the inclusion. Note that ι† is a norm, but not a valuation. Suppose that C ,→ DK ∆ is a plane curve. Let J be a monomial ideal of C+, {tkxi y j} a finite list of generators. A generator tkxi y j may be removed from the list if and only if it is expressible in terms of the other generators, which occurs exactly when the coefficient ci jk of that monomial in some f ∈ IC is non-zero. In other words, the relations of the quotient Z∨{∆} → α∆ are generated by those of the form F = ι †(f ) = _ (i, j,k)6=(i0, j0,k0) iX + jY − k where F is the tropicalisation of f and ci0 j0k0 6= 0. There are in general infinitely many such relations. The image of Trop(C/X /∆) in the Hausdorff quotient sk∆ → ∆ ⊂ R2 is the non-differentiability locus of the convex piecewise-affine function on ∆ defined by F. These relations were also obtained by different means in [GG13]. In order to globalise this procedure, we need to check the functoriality of the amoeba under inclusion ∆′ ⊆ ∆ of lattice polytopes. The corresponding open immersion sk∆′ ,→ sk∆ may be factored into a subdivision at some element Z ∈ Z◦ ∨{∆} followed by a cell inclusion. along Z, and then restricting to an affine subset. 8.9 Lemma. Let C ,→ X be a closed embedding of adic spaces, j : X → X + a formal model This is the combinatorial shadow of the operation of taking the toric blow-up eX + → X + of X , C+ ⊆ X + the closure of C in X +. Let eX + → X + be an admissible blow-up with ideal J. Then the closure eC+ of C in eX + is the blow-up of C+ along OC+ J. Proof. The definitions directly imply the following identity RJ IC ∩ RJ of the Rees algebras on C+. ∼= Ln∈N Jn tn Ln∈N IC ∩ Jn tn ∼=Mn∈N Jn IC ∩ Jn tn ∼= ROC+ J 57 As we observed in the previous section, the Clemens semiring Cl(X ;eX +) is the strict transform semiring of Z∨{∆} under the monomial blow-up eX + → X + (def. 6.24). Further- more, the formation of the strict transform semiring commutes with the tropicalisation of ideals on C: j∗O + CMn∈N Jntn ∼=Mn∈N Jn IC ∩ Jn tn ∼=Mn∈N j∗O + C Jn. By corollary 6.25, the image of Bc( j∗OC; j∗O + Now writing X ′ = DK ∆′ and C′ = C ×X X ′, we obtain a natural morphism of skeleta Trop(C′/X ′/∆′) → Trop(C/X /∆). The above arguments, together with lemma 8.4 show: C) in Cl(X ;eX +) is a free localisation of α∆. 8.10 Proposition. Trop(C′/X ′/∆′) → Trop(C/X /∆) is an open immersion. We can therefore glue tropicalisations as we glue polytopes. In particular, we can con- struct the amoeba Trop(C/X /Σ) = [σr⊂Σ Trop(C ×X Dkσr/DK σr/σr) of any subscheme of a toric variety, as promised above. 8.4 Circle Returning to the situation of 8.2, let us specialise to the case of an elliptic curve. Let K be a DVF with residue field k, E/K an elliptic curve; write E/K for the base change to the algebraic closure. Let Ω = Ω1,0 ∈ Γ(E; ωE/K ) \ {0} be a holomorphic volume form. 8.11 Definition. A formal model E+ of E is crepant if it is Q-Gorenstein and one of the following equivalent conditions are true: i) there exists a log resolution f : (E+)′ → E+ on which (E+)′ + f ∗Ω = tk as Q-divisors on (E+)′, where k ∈ Z and (E+)′ denotes the reduction of (E+)′; ii) The log canonical threshold is equal to a constant k on E+ (in equal characteristic zero); iii) The canonical bundle ωE+/O K over the algebraic closure is trivial. A formal model of E is crepant if it is finitely presented with trivial canonical bundle over K , or equivalently, it is obtained by flat base extension from a crepant formal model over O some finite extension of OK . A simple normal crossings model E+ of E is crepant if and only if its reduction is a cycle of projective lines. The multiplicity of a line in the central fibre E+ 0 := k ×OK E+ is one more than its multiplicity in the canonical divisor. One can make the multiplicities all one, and hence trivialise the canonical bundle, by effecting a finite base change followed by a normalisation. In particular, E+ is semistable if and only if ωE+/OK is trivial, that is, if and only if it is a minimal model. On the other hand, a formal model of E is locally toric if and only if it has at worst monomial cyclic quotient singularities and the components of its central fibre are smooth rational curves. It is automatically Q-Gorenstein. Such a model exists only if E has bad, but semistable reduction; let us assume this. 58 Let Mdlclt(E) denote the category of crepant, locally toric models of E. Let E+ be an object of this category. Its singularities occur at the intersections of components, and they have the form k[[t]]∆ → ∆◦ D+ where ∆ = [a, b] is an interval with rational endpoints. They may be resolved explicitly, and crepantly, by subdividing the interval at all its integer points. Since, by assumption, a crepant resolution of E+ exists, (E+)′ must be a crepant model. ks. Since ∆(E+)′ → ∆E+ is a subdivision, it follows that Its reduction is therefore a cycle of P1 both are cycles of intervals; ∆E+(R∨) has the topology of a circle. The Clemens functor ∆ : Mdlclt(E) → Skaff is defined, and its image is a diagram of subdivisions. It may therefore be glued to obtain the Kontsevich-Soibelman (or KS) skeleton sk(E; Ω) := colim∆. It is a shell of any crepant Clemens skeleton, and hence comes with a collapse map µ : E → sk(E; Ω), which is a torus fibration: every point of sk(E; Ω) has an overconvergent neighbourhood over which µ is isomorphic over Z∨ to a standard torus fibration on an interval. In the introduction we introduced an explicit 'atlas' for sk(E; Ω) under the assumption (which may be lifted) that the minimal model of E consist of at least three reduced lines.6 If, more generally, E has only bad reduction, we can still define sk(E; Ω) as the colimit of the Clemens functor on Mdlclt(E). If L ⊇ K is a finite extension over which EL := L ×K E has semistable reduction, then sk(E; Ω) ∼= Q∨ ×Z∨ sk(EL; Ω) by the base change property for polytopes. In particular, the collapse map E → sk(E; Ω) is again a torus fibration. The KS skeleton is of finite presentation over Q∨. There is a continuous projection π : sk(E; Ω) → B := sk(E; Ω)(R∨) ≃ S1. The local models for µ induce a canonical smooth structure on B with respect to which the affine functions, that is, invertible sections AffZ(B, Q) of CPAZ(B, Q∨) := π∗O canc, are smooth. It therefore attains an affine structure in the sense of [KS06, §2.1] defined by the exact sequence 0 −→ Q −→ AffZ(B, Q) −→ Λ∨ −→ 0 of Abelian sheaves on B and the induced embedding Λ∨ ,→ T ∨B. Let E+ L be a semistable minimal model of EL. By writing Ω locally in the form λd log x E and λ ∈ L×, we can think of Ω as a non-zero section of L ⊗Z Λ∨. It for some monomial x ∈ O × induces a K-orientation of B that does not depend on the choice of L or x. 6I insert the word 'atlas' between inverted commas because I do not prove that these open sets really cover sk(E; Ω). That statement would be equivalent to the conjecture raised in example 7.4. 59 References [AGV70] M. Artin, A. Grothendieck, and J. L. Verdier. SGA 4; Théorie de topos et cohomolo- gie étale des schémas. Springer-Verlag, 1970. [BBT13] O. Ben-Bassat and M. Temkin. Berkovich spaces and tubular descent. Advances in Mathematics, 234, 2013. arXiv 1201.4227. [Ber93] V. Berkovich. Étale cohomology for non-Archimedean analytic spaces. Publica- tions mathématiques de l'IHÉS, 78, 1993. Numdam. [Ber99] V. Berkovich. Smooth p-adic analytic spaces are locally contractible. Inventiones Mathematicae, 137(1), 1999. online. [BGR84] S. Bosch, U. Güntzer, and R. Remmert. Non-Archimedean analysis: a systematic approach to rigid analytic geometry. Springer, 1984. [Bou62] N. Bourbaki. Seminaire Banach. Springer, 1962. [BPR11] M. Baker, S. Payne, and J. Rabinoff. Nonarchimedean geometry, tropicalization, and metrics on curves. 1104.0320, 2011. [Dur07] N. Durov. A new approach to Arakelov geometry. arXiv 0704.2030, 2007. [EKL07] M. Einsiedler, M. Kapranov, and D. Lind. Non-Archimedean amoebas and tropical varieties. Journal für die reine und angewandte Mathematik, 601, 2007. arXiv 0408311. [FK13] K. Fujiwara and F. Kato. Foundations of rigid geometry I. arXiv 1308.4734, 2013. [GG13] J. Giansiracusa and N. Giansiracusa. Equations of tropical varieties. arXiv 1308.0042, 2013. [Gro60] A. Grothendieck. Éléments de géometrie algébrique. IHES, 1960. [Gro11] M. Gross. Mirror symmetry and tropical geometry. AMS, 2011. online. [GS07] M. Gross and B. Siebert. From real affine geometry to complex geometry. Annals of mathematics, 2007. arXiv 0703822. [Hub96] R. Huber. Étale cohomology of rigid analytic varieties and adic spaces. Aspects of mathematics. Springer Vieweg, 1996. [KS00] M. Kontsevich and Y. Soibelman. Homological mirror symmetry and torus fi- brations. In Symplectic geometry and mirror symmetry. World Scientific, 2000. 0011041. [KS06] M. Kontsevich and Y. Soibelman. Affine structures and non-Archimedean analytic spaces. In The unity of mathematics. Springer, 2006. arXiv 0406564. [Mac14] A. W. Macpherson. Skeleta. PhD thesis, Imperial College London, 2014. in prepa- ration. 60 [Mik06] G. Mikhalkin. Tropical geometry and its applications. In Proceedings of the ICM, volume 2, 2006. arXiv 0601041. [MM92] S. Maclane and I. Moerdijk. Sheaves in geometry and logic: a first introduction to topos theory. Springer, 1992. [NX13] J. Nicaise and C. Xu. The essential skeleton of a degeneration of algebraic vari- eties. arXiv 1307.4041, 2013. [Pay09] S. Payne. Analytification is the limit of all tropicalisations. Math. Res. Lett., 16(3), 2009. arXiv 0805.1916. [TV09] B. Toën and M. Vaquié. Au-dessous de spec Z. Journal of K-theory, 3(03), 2009. arXiv 0509684. 61
1211.4897
3
1211
2016-02-17T15:59:25
Lipschitz geometry of complex surfaces: analytic invariants and equisingularity
[ "math.AG", "math.CV" ]
We prove that the outer Lipschitz geometry of a germ $(X,0)$ of a normal complex surface singularity determines a large amount of its analytic structure. In particular, it follows that any analytic family of normal surface singularities with constant Lipschitz geometry is Zariski equisingular. We also prove a strong converse for families of normal complex hypersurface singularities in $\mathbb C^3$: Zariski equisingularity implies Lipschitz triviality. So for such a family Lipschitz triviality, constant Lipschitz geometry and Zariski equisingularity are equivalent to each other.
math.AG
math
LIPSCHITZ GEOMETRY OF COMPLEX SURFACES: ANALYTIC INVARIANTS AND EQUISINGULARITY. WALTER D NEUMANN AND ANNE PICHON Abstract. We prove that the outer Lipschitz geometry of the germ of a nor- mal complex surface singularity determines a large amount of its analytic struc- ture. In particular, it follows that any analytic family of normal surface singu- larities with constant Lipschitz geometry is Zariski equisingular. We also prove a strong converse for families of normal complex hypersurface singularities in C3: Zariski equisingularity implies Lipschitz triviality. So for such a family Lipschitz triviality, constant Lipschitz geometry and Zariski equisingularity are equivalent to each other. 1. Introduction This paper has two aims. One is to prove the equivalence of Zariski and bilips- chitz equisingularity for families of normal complex surface singularities. The other, on which the first partly depends, is to describe which analytic invariants are de- termined by the Lipschitz geometry of a normal complex surface singularity. In [1] the richness of the Lipschitz geometry of a normal surface singularity was demonstrated in a classification in terms of discrete invariants associated to a refined JSJ decomposition of the singularity link. That paper addressed the inner metric. The present paper concerns the outer metric, and we show it is even richer. Equisingularity. The question of defining a good notion of equisingularity of a reduced hypersurface X Ă Cn along a non-singular complex subspace Y Ă X in a neighbourhood of a point 0 P X started in 1965 with two papers of Zariski ([31, 32]). This problem has been extensively studied with different approaches and by many authors such as Zariski himself, Abhyankar, Brian¸con, Gaffney, Hironaka, Le, Lejeune-Jalabert, Lipman, Mostowski, Parusi´nski, Pham, Speder, Teissier, Thom, Trotman, Varchenko, Wahl, Whitney and many others. One of the central concepts introduced by Zariski is the algebro-geometric equi- singularity, called nowadays Zariski equisingularity. The idea is that the equisingu- larity of X along Y is defined inductively on the codimension of Y in X by requiring that the reduced discriminant locus of a suitably general projection p : X Ñ Cn´1 be itself equisingular along ppY q. The induction starts by requiring that in codi- mension one the discriminant locus is nonsingular. (This is essentially Speder's definition in [22]; Zariski later gave a variant in [33], but for surface singularities the variants are equivalent to each other.) When Y has codimension one in X, it is well known that Zariski equisingularity is equivalent to all main notions of equisingularity, such as Whitney conditions for 2010 Mathematics Subject Classification. 14B05, 32S25, 32S05, 57M99. Key words and phrases. bilipschitz, Lipschitz geometry, normal surface singularity, Zariski equisingularity, Lipschitz equisingularity. 1 2 WALTER D NEUMANN AND ANNE PICHON the pair pX r Y, Y q and topological triviality. However these properties fail to be equivalent in higher codimension: Zariski equisingularity still implies topological triviality ([28, 29]) and Whitney conditions ([22]), but the converse statements are false ([4, 5, 27]) and a global theory of equisingularity is still far from being established. For good surveys of equisingularity questions, see [16, 25]. The first main result of this paper states the equivalence between Zariski equi- singularity, constancy of Lipschitz geometry and triviality of Lipschitz geometry in the case Y is the singular locus of X and has codimension 2 in X. We must say what we mean by "Lipschitz geometry". If pX, 0q is a germ of a complex variety, then any embedding φ : pX, 0q ãÑ pCn, 0q determines two metrics on pX, 0q: the outer metric doutpx1, x2q :" φpx1q ´ φpx2q p i.e., distance in Cnq and the inner metric dinnpx1, x2q :" inftlengthpφ γq : γ is a rectifyable path in X from x1 to x2u . The outer metric determines the inner metric, and up to bilipschitz equivalence both these metrics are independent of the choice of complex embedding. We speak of the (inner or outer) Lipschitz geometry of pX, 0q when considering these metrics up to bilipschitz equivalence. If we work up to semi-algebraic or semi-analytic bilipschitz equivalence, we speak of semi-algebraic or semi-analytic Lipschitz geometry. As already mentioned, the present paper is devoted to outer Lipschitz geometry. We consider here the case of normal complex surface singularities. Let pX, 0q Ă pCn, 0q be a germ of hypersurface at the origin of Cn with smooth codimension 2 singular set pY, 0q Ă pX, 0q. The germ pX, 0q has constant Lipschitz geometry along Y if there exists a smooth retraction r : pX, 0q Ñ pY, 0q whose fibers are transverse to Y and there is a neigh- bourhood U of 0 in Y such that for all y P U there exists a bilipschitz homeomor- phism hy : pr´1pyq, yq Ñ pr´1p0q X X, 0q. The germ pX, 0q is Lipschitz trivial along Y if there exists a germ at 0 of a bilipschitz homeomorphism Φ : pX, Y q Ñ pX, 0q Y with ΦY " idY , where pX, 0q is a normal complex surface germ. We say pX, 0q has constant semi-analytic Lipschitz geometry or is semi-analytic Lipschitz trivial if the maps r and hy, resp. Φ above are semi-analytic. Theorem 1.1. The following are equivalent: (1) pX, 0q is Zariski equisingular along Y ; (2) pX, 0q has constant Lipschitz geometry along Y ; (3) pX, 0q has constant semi-analytic Lipschitz geometry along Y ; (4) pX, 0q is semi-analytic Lipschitz trivial along Y The equivalence between (1) and (4) has been conjectured by Mostowski in a talk for which written notes are available ([17]). He also gave there brief hints to prove the result using his theory of Lipschitz stratifications. Our approach is different and we construct a decomposition of the pair pX, X r Y q using the theory of carrousel decompositions, introduced by Le in [12], on the family of discriminant curves. In both approaches, polar curves play a central role. The implications (4)ñ(3)ñ(2) are trivial and the implication (2)ñ(1) will be a consequence of one of our other main results, part (5) of Theorem 1.2 below. The implication (1)ñ(4) will be proved in the final sections of the paper. LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 3 It was known already that the inner Lipschitz geometry is not sufficient to under- stand Zariski equisingularity. Indeed, the family of hypersurfaces pXt, 0q Ă pC3, 0q with equation z3 ` tx4z ` x6 ` y6 " 0 is not Zariski equisingular (see [6]; at t " 0, the discriminant curve has 6 branches, while it has 12 branches when t ‰ 0). But it follows from [1] that it has constant inner geometry (in fact pXt, 0q is metrically conical for the inner metric for all t). Invariants from Lipschitz geometry. The other main results of this paper are stated in the next theorem. Theorem 1.2. If pX, 0q is a normal complex surface singularity then the outer Lipschitz geometry on X determines: (1) the decorated resolution graph of the minimal good resolution of pX, 0q which resolves the basepoints of a general linear system of hyperplane sections1; (2) the multiplicity of pX, 0q and the maximal ideal cycle in its resolution; (3) for a generic hyperplane H, the outer Lipschitz geometry of the curve pX X H, 0q; (4) the decorated resolution graph of the minimal good resolution of pX, 0q which resolves the basepoints of the family of polar curves of plane projections2; (5) the topology of the discriminant curve of a generic plane projection; (6) the outer Lipschitz geometry of the polar curve of a generic plane projection. By "decorated resolution graph" we mean the resolution graph decorated with ar- rows corresponding to the components of the strict transform of the resolved curve. As a consequence of this theorem, the outer Lipschitz geometry of a hypersurface pX, 0q Ă pC3, 0q determines other invariants such as its extended Milnor number µpX, 0q and the invariants kpX, 0q and φpX, 0q (number of vanishing double folds resp. cusp-folds, see [2]). The analytic invariants determined by the Lipschitz geometry in Theorem 1.2 are of two natures: (1) -- (3) are related to the general hyperplane sections and the maximal ideal of OX,0 while (4) -- (6) are related to polar curves and the jacobian ideal. The techniques we use for these two sets of invariants differ. We prove (1) -- (3) in section 11, building in part from the classification theorem for inner bilipschitz geometry of [1]. The more difficult part is (4) -- (6). These polar invariants are determined in Sections 12 and 14 from a geometric decomposition of pX, 0q defined in Sections 8 and 9 which refines the decomposition used in [1]. This paper has four parts, of which Parts 1 and 2 (Sections 2 to 9) introduce concepts and techniques that are needed later, Part 3 (Sections 11 to 14) proves Theorem 1.2 and Part 4 (Sections 15 to 21) proves Theorem 1.1. Acknowledgments. We are especially grateful to Lev Birbrair for many con- versations which have contributed to this work. We are also grateful to Terry Gaffney and Bernard Teissier for useful conversations and to Joseph Lipman for helpful comments. Neumann was supported by NSF grants DMS-0905770 and DMS-1206760. Pichon was supported by ANR-12-JS01-0002-01 SUSI and FP7- Irses 230844 DynEurBraz. We are also grateful for the hospitality and support of the following institutions: Columbia University (P), Institut de Math´ematiques de Marseille and Aix Marseille Universit´e (N), Universidad Federal de Ceara (N,P), CIRM recherche en binome (N,P) and IAS Princeton (N,P). 1 This is the minimal good resolution which factors through the blow-up of the maximal ideal. 2 This is the minimal good resolution which factors through the Nash modification. 4 WALTER D NEUMANN AND ANNE PICHON Part 1: Carrousel and geometry of curves 2. Preliminary: geometric pieces and rates In [1] we defined some metric spaces (with inner metric) called A-, B- and D- pieces, which will be used extensively in the present paper. We give here the definition and basic facts about these pieces (see [1, Sections 11 and 13] for more details). First examples will appear in the next section. The pieces are topologically conical, but usually with metrics that make them shrink non-linearly towards the cone point. We will consider these pieces as germs at their cone-points, but for the moment, to simplify notation, we suppress this. D2 denotes the standard unit disc in C, S1 is its boundary, and I denotes the interval r0, 1s. Definition 2.1 (Apq, q1q-pieces). Let q, q1 be rational numbers such that 1 ď q ă q1. Let A be the euclidean annulus tpρ, ψq : 1 ď ρ ď 2, 0 ď ψ ď 2πu in polar coordinates and for 0 ă r ď 1 let gprq q,q1 be the metric on A: q,q1 :" prq ´ rq1 gprq q2dρ2 ` ppρ ´ 1qrq ` p2 ´ ρqrq1 q2dψ2 . So A with this metric is isometric to the euclidean annulus with inner and outer radii rq1 and rq. The metric completion of p0, 1s S1 A with the metric dr2 ` r2dθ2 ` gprq q,q1 compactifies it by adding a single point at r " 0. We call a metric space which is bilipschitz homeomorphic to this completion an Apq, q1q-piece or simply A-piece. Definition 2.2 (Bpqq-pieces). Let F be a compact oriented 2-manifold, φ : F Ñ F an orientation preserving diffeomorphism, and Mφ the mapping torus of φ, de- fined as: Mφ :" pr0, 2πs F q{pp2π, xq „ p0, φpxqqq . Given a rational number q ą 1, we will define a metric space BpF, φ, qq which is topologically the cone on the mapping torus Mφ. For each 0 ď θ ď 2π choose a Riemannian metric gθ on F , varying smoothly with θ, such that for some small δ ą 0: gθ "#g0 φg0 for θ P r0, δs , for θ P r2π ´ δ, 2πs . Then for any r P p0, 1s the metric r2dθ2 ` r2qgθ on r0, 2πs F induces a smooth metric on Mφ. Thus dr2 ` r2dθ2 ` r2qgθ defines a smooth metric on p0, 1s Mφ. The metric completion of p0, 1s Mφ adds a single point at r " 0. Denote this completion by BpF, φ, qq. We call a metric space which is bilipschitz homeomorphic to BpF, φ, qq a Bpqq-piece or simply B-piece. If F is an annulus we may speak of a B-piece of type Apq, qq. A Bpqq-piece such that F is a disc and q ě 1 is called a Dpqq-piece or simply D-piece. Definition 2.3 (Conical pieces). Given a compact smooth 3-manifold M , choose a Riemannian metric g on M and consider the metric dr2 ` r2g on p0, 1s M . The completion of this adds a point at r " 0, giving a metric cone on M . We call a metric space which is bilipschitz homeomorphic to a metric cone a conical piece. LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 5 Notice that Bp1q- and Ap1, 1q-pieces are conical. For short we will call any conical piece a Bp1q-piece, even if its 3-dimensional link is not a mapping-torus. (Conical- pieces were called CM -pieces in [1].) The diameter of the image in BpF, φ, qq of a fiber Fr,θ " tru tθu F is Oprqq. Therefore q describes a rate of shrink of the surfaces Fr,θ in BpF, φ, qq with respect to the distance r to the point at r " 0. Similarly, the inner and outer boundary components of any tru ttu A in an Apq, q1q have rate of shrink respectively q1 and q with respect to r. Definition 2.4 (Rate). The rational number q is called the rate of Bpqq or Dpqq. The rationals q and q1 are the two rates of Apq, q1q. In [1, Section 11] we explain how to glue two A- or B-pieces by an isometry along cone-boundary components having the same rate. Some of these gluings give again A- or B-pieces and then could simplify by some rules as described in [1, Section 13]. We recall here this result, which will be used later in amalgamation processes. Lemma 2.5. In this lemma -- means bilipschitz equivalence and Y represents gluing along appropriate boundary components by an isometry. (1) BpD2, φ, qq -- BpD2, id, qq; BpS1 I, φ, qq -- BpS1 I, id, qq. (2) Apq, q1q Y Apq1, q2q -- Apq, q2q. (3) If F is the result of gluing a surface F 1 to a disk D2 along boundary com- ponents then BpF 1, φF 1 , qq Y BpD2, φD2 , qq -- BpF, φ, qq. (4) Apq, q1q Y BpD2, id, q1q -- BpD2, id, qq. (5) A union of conical pieces glued along boundary components is a conical (cid:3) piece. 3. The plain carrousel decomposition for a plane curve germ A carrousel decomposition for a reduced curve germ pC, 0q Ă pC2, 0q is con- structed in two steps. The first one consists in truncating the Puiseux series expan- sions of the branches of C at suitable exponents and then constructing a decom- position of pC2, 0q into A-, B- and D-pieces with respect to the truncated Puiseux series. We call this first decomposition the unamalgamated carrousel decomposition. It is the carrousel decomposition used in [1], based on ideas of Le in [12]. The second step consists in amalgamating certain pieces of the unamalgamated carrousel decomposition using the techniques of Lemma 2.5. There are two choices in the construction: where we truncate the Puiseux series expansions to form a carrousel decomposition, and how we amalgamate pieces of the carrousel decomposition to form the amalgamated carrousel decomposition. Amalgamated carrousel decompositions will be a key tool in the present paper and we will in fact use several different ones based on different choices of truncation and amalgamation. In this section, we will construct an amalgamated carrousel decomposition which we call the plain carrousel decomposition of C. We start by constructing the ap- propriate unamalgamated carrousel decomposition. The tangent cone of C at 0 is a unionŤm j"1 Lpjq of lines. For each j we denote the union of components of C which are tangent to Lpjq by Cpjq. We can assume our coordinates px, yq in C2 are chosen so that the y-axis is transverse to each Lpjq. 6 WALTER D NEUMANN AND ANNE PICHON We choose ǫ0 ą 0 sufficiently small that the set tpx, yq : x " ǫu is transverse to C for all ǫ ď ǫ0. We define conical sets V pjq of the form V pjq :" tpx, yq : y ´ apjq 1 x ď ηx, x ď ǫ0u Ă C2 , where the equation of the line Lpjq is y " apjq 1 x and η ą 0 is small enough that the cones are disjoint except at 0. If ǫ0 is small enough Cpjq X tx ď ǫ0u will lie completely in V pjq. There is then an R ą 0 such that for any ǫ ď ǫ0 the sets V pjq meet the boundary of the "square ball" Bǫ :" tpx, yq P C2 : x ď ǫ, y ď Rǫu only in the part x " ǫ of the boundary. We will use these balls as a system of Milnor balls. We first define the carrousel decomposition inside each V pjq with respect to the branches of Cpjq. It will consist of closures of regions between successively smaller neighbourhoods of the successive Puiseux approximations of the branches of Cpjq. As such, it is finer than the one of [12], which only needed the first Puiseux exponents of the branches of Cpjq. We will fix j for the moment and therefore drop the superscripts, so our tangent line L has equation y " a1x. The collection of coefficients and exponents appearing in the following description depends, of course, on j " 1, . . . , m. Truncation. We first truncate the Puiseux series for each component of C at a point where truncation does not affect the topology of C. Then for each pair i"1 aixpi and an exponent pk i"1 aixpi ` . . . describing some component of C, we consider all components of C which fit this data. If ak1, . . . , akmκ are the coefficients of xpk which occur in these Puiseux polynomials we define κ " pf, pkq consisting of a Puiseux polynomial f "řk´1 for which there is a Puiseux series y "řk k´1ÿi"1 Bκ :"!px, yq : ακxpk ď y ´ aixpi ď βκxpk y ´ p k´1ÿi"1 aixpi ` akjxpk q ě γκxpk for j " 1, . . . , mκ) . Here ακ, βκ, γκ are chosen so that ακ ă akj ´ γκ ă akj ` γκ ă βκ for each j " 1, . . . , mκ. If ǫ is small enough, the sets Bκ will be disjoint for different κ. It is easy to see that Bκ is a Bppkq-piece: the intersection Bκ X tx " tu is a finite collection of disks with some smaller disks removed. The diameter of each of them is Optpk q. The closure of the complement in V " V pjq of the union of the Bκ's is a union of A- and D-pieces. Finally, Bǫ rŤ V pjq is a Bp1q-piece. We have then decomposed each cone V pjq and the whole of Bǫ as a union of B-, A- and D-pieces. Definition 3.1 (Carrousel sections). A carrousel section is the picture of the intersection of a carrousel decomposition with a line x " t. Amalgamation. For the study of plane curves a much simpler decomposition suffices, which we obtain by amalgamating any D-piece that does not contain part of the curve C and any A-piece with the piece just outside it. LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 7 Definition 3.2. We call this amalgamated carrousel decomposition the plain car- rousel decomposition of the curve C. Example 3.3. We give examples for the carrousel section and plain carrousel section in Figure 1 below. On the left the figure shows a carrousel section for a curve C having two branches with Puiseux expansions respectively y " ax4{3 `bx13{6 `. . . and y " cx7{4 ` . . . truncated after the terms bx13{6 and cx7{4 respectively. The D-pieces are gray. Note that the intersection of a piece of the decomposition of V with the disk V X tx " ǫu will usually have several components. Note also that the rates in A- and D-pieces are determined by the rates in the neighbouring B-pieces. On the right is the corresponding plain carrousel section. Bp1q Ap4{3, 13{6q Bp13{6q Bp1q Bp13{6q Bp7{4q Bp7{4q Bp4{3q Bp4{3q Ap4{3, 7{4q Ap1, 4{3q Figure 1. Unamalgamated and plain carrousel sections for C " ty " ax4{3 ` bx13{6 ` . . .u Y ty " cx7{4 ` . . .u. The combinatorics of the plain carrousel section can be encoded by a rooted tree, with vertices corresponding to pieces, edges corresponding to pieces which intersect along a circle, and with rational weights (the rates of the pieces) associated to the nodes of the tree. It is easy to recover a picture of the carrousel section, and hence the embedded topology of the plane curve, from this weighted tree. For a careful description of how to do this in terms of either the Eggers tree or the Eisenbud-Neumann splice diagram of the curve see [19]. Proposition 3.4. The combinatorics of the plain carrousel section for a plane curve germ C determines the embedded topology of C. (cid:3) 4. Lipschitz geometry and topology of a plane curve In [19] we proved the following strong version of a result of Pham and Teissier [21] and Fernandes [9] about plane curve germs. Proposition 4.1. The outer Lipschitz geometry of a plane curve germ pC, 0q Ă pC2, 0q determines its embedded topology. More generally, if pC, 0q Ă pCn, 0q is a curve germ and ℓ : Cn Ñ C2 is a generic plane projection then the outer Lipschitz geometry of pC, 0q determines the embedded topology of the plane projection pℓpCq, 0q Ă pC2, 0q. 8 WALTER D NEUMANN AND ANNE PICHON What was new in this proposition is that the germ is considered just as a metric space germ up to bilipschitz equivalence, without the analytic restrictions of [21] and [9]. The converse result, that the embedded topology of a plane curve determines the outer Lipschitz geometry, is easier and is proved in [21]. In this section we briefly recall the proof of Proposition 4.1, since we use the last part of it in Section 11 (proof of Theorem 1.2(3)) and we also mildly adapt its proof (as given in [19]) to illustrate a technique we will use in Section 14. Proof. The case n ą 2 of Proposition 4.1 follows immediately from the case n " 2 since Pham and Teissier prove in [21] that for a generic plane projection ℓ the restriction ℓC : pC, 0q Ñ pℓpCq, 0q is bilipschitz for the outer geometry. So we assume from now on that n " 2, so pC, 0q Ă pC2, 0q is a plane curve. We first describe how to recover the combinatorics of the plain carrousel section of C using the analytic structure and the outer geometry. We assume, as in the previous section, that the tangent cone of C at 0 is a union of lines transverse to the y-axis. We use again the family of Milnor balls Bǫ, ǫ ď ǫ0, of the previous section. We put Sǫ " BBǫ. Let µ be the multiplicity of C. The lines x " t for t P p0, ǫ0s intersect C in a finite set of points p1ptq, . . . , pµptq which depends continuously on t. For each 0 ă j ă k ď µ the distance dppjptq, pkptqq has the form Optqjk q, where qjk is either an essential Puiseux exponent for a branch of the plane curve C or a coincidence exponent between two branches of C. Lemma 4.2. The map tpj, kq 1 ď j ă k ď µu ÞÑ qjk determines the embedded topology of C. Proof. By Proposition 3.4 it suffices to prove that the map tpj, kq 1 ď j ă k ď µu ÞÑ qjk determines the combinatorics of the plain carrousel section of the curve C. Let q1 ą q2 ą . . . ą qs be the images of the map pj, kq ÞÑ qjk. The proof consists in reconstructing a topological version of the carrousel section of C from the innermost pieces to the outermost ones by an inductive process starting with q1 and ending with qs. We start with µ discs Dp0q 1 , . . . , Dp0q µ , which will be the innermost pieces of the carrousel decomposition. We consider the graph Gp1q whose vertices are in bijection with these µ disks and with an edge between vertices vj and vk if and only if qjk " q1. Let Gp1q ν1 be the connected components of Gp1q and denote by m the number of vertices of Gp1q αp1q m ą 1 we consider a disc Bp1q m , into the inner boundary components of Bp1q one vertex, vjm say, we rename Dp0q m . For each Gp1q m holes, and we glue the discs Dp0q m . We call the resulting disc Dp1q , vj P vert Gp1q m . For a Gp1q 1 , . . . , Gp1q m with αp1q m with αp1q j jm as Dp1q m . m with just The numbers qjk have the property that qjl ě minpqjk, qklq for any triple j, k, l. So for each distinct m, n the number qjmkn does not depend on the choice of vertices vjm in Gp1q m and vkn in Gp1q n . We iterate the above process as follows: we consider the graph Gp2q whose ver- ν1 and with an n q if and only if qjmkn equals q2 (with ν2 be the con- m be the number of its vertices. tices are in bijection with the connected components Gp1q edge between the vertices pGp1q vertices vjm and vjn in Gp1q nected components of G2. For each Gp2q n respectively). Let Gp2q m q and pGp1q 1 , . . . , Gp1q 1 , . . . , Gp2q m and Gp1q m let αp2q LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 9 m ą 1 we take a disc Bp2q into these holes. We call the resulting piece Dp2q m holes and glue the corresponding disks m " 1 we rename the m with αp2q m . If αp2q If αp2q Dp1q corresponding Dp1q l m disk to Dp2q m . By construction, repeating this process for s steps gives a topological version of the carrousel section of the curve C, and hence, by Proposition 3.4, its embedded topology. (cid:3) To complete the proof of Proposition 4.1 we must show that we can find the numbers qjk without using the complex structure, and after a bilipschitz change to the outer metric. The tangent cone of C at 0 is a union of lines Lpiq, i " 1, . . . , m, transverse to the y-axis. Denote by Cpiq the part of C tangent to the line Lpiq. It suffices to discover the qjk's belonging to each Cpiq independently, since the Cpiq's are distinguished by the fact that the distance between any two of them outside a ball of radius ǫ around 0 is Opǫq, even after bilipschitz change to the metric. We will therefore assume from now on that the tangent cone of C is a single complex line. Our points p1ptq, . . . , pµptq that we used to find the numbers qjk were obtained by intersecting C with the line x " t. The arc p1ptq, t P r0, ǫ0s satisfies dp0, p1ptqq " Optq. Moreover, for any small δ ą 0 we can choose ǫ0 sufficiently small that the other points p2ptq, . . . , pµptq are always in the transverse disk of radius δt centered at p1ptq in the plane x " t. Instead of a transverse disk of radius δt, we now use a ball Bpp1ptq, δtq of radius δt centered at p1ptq. This Bpp1ptq, δtq intersects C in µ disks D1ptq, . . . , Dµptq, and we have dpDjptq, Dkptqq " Optqjk q, so we still recover the numbers qjk. 1ptq, . . . , D1 µptq with dpD1 1ptq, δtq still consists of µ disks D1 We now replace the arc p1ptq by any continuous arc p1 Such a change may make the components of BC pp1 1ptqq " Optq. If δ is sufficiently small, the intersection BC pp1 1ptq on C with the property that dp0, p1 1ptq, δtq :" C X Bpp1 kptqq " Optqjk q. So we have gotten rid of the dependence on analytic structure in discovering the topology. But we must consider what a K-bilipschitz change to the metric does. 1ptq, δtq disintegrate into many pieces, so we can no longer simply use distance between pieces. To resolve this, K 4 tq where B1 means we are us- we consider both B1 ing the modified metric. Then only µ components of B1 C pp1ptq, δtq will intersect 1ptq, . . . , D1 B1 µptq again, we still have dpD1 (cid:3) Cpp1ptq, δ jptq, D1 We end this section with a remark which introduces a key concept for the proof K 4 tq. Naming these components D1 kptqq " Optqjk q so the qjk are determined as before. 1ptq, δtq and B1 jptq, D1 C pp1 C pp1 1ptq, δ of Proposition 14.1. Remark 4.3. Assume first that C is irreducible. Fix q P Q and replace in the arguments above the balls BC pp1ptq, δtq of radius δt by balls of radius δtq. If δ is big enough then BC pp1ptq, δtqq consists of η discs D2 ηptq for some η ď µ and the rates q2 j,k q coincide with the above rates qj,k such that qj,k ě q. These rates determine the carrousel sections of C inside pieces of the carrousel with rates ě q. Definition 4.4. We say that the rates q2 Lipschitz geometry (resp. the carrousel section) of C beyond rate q. j,k with 1 ď j ă k ď η determine the outer j,k given by dpD2 kptqq " Optq2 1ptq, . . . , D2 j ptq, D2 If C is not irreducible, take p1 inside a component C0 of C. Then the rates q2 j,k recover the outer Lipschitz geometry beyond q of the union C1 of components of C 10 WALTER D NEUMANN AND ANNE PICHON having exponent of coincidence with C0 greater than or equal to q. If C contains a component which is not in C1, we iterate the process by choosing an arc in it. After iterating this process to catch all the components of C, we say we have determined the outer Lipschitz geometry of the whole C (resp. its carrousel section) beyond rate q. If C is the generic projection of a curve C1 Ă Cn, we speak of the outer Lipschitz geometry of C1 beyond rate q. Part 2: Geometric decompositions of a normal complex surface singularity 5. Introduction to geometric decomposition Birbrair and the authors proved in [1] the existence and unicity of a decompo- sition pX, 0q " pY, 0q Y pZ, 0q of a normal complex surface singularity pX, 0q into a thick part pY, 0q and a thin part pZ, 0q. The thick part is essentially the metrically conical part of pX, 0q with respect to the inner metric while the thin part shrinks faster than linearly in size as it approaches the origin. The thick-thin decomposi- tion was then refined further by dividing the thin part into A- and B-pieces, giving a classification of the inner Lipschitz geometry in terms of discrete data recording rates and directions of shrink of the refined pieces (see [1, Theorem 1.9]). This "classifying decomposition" decomposes the 3-manifold link X pǫq :" X X Sǫ of X into Seifert fibered pieces glued along torus boundary components. In general this decomposition is a refinement of the JSJ decomposition of the 3-manifold X pǫq (minimal decomposition into Seifert fibered pieces). In this Part 2 of the paper we refine the decomposition further, in a way that implicitly sees the influence of the outer Lipschitz geometry, by taking the position of polar curves of generic plane projections into account. We call this decomposition the geometric decomposition (see Definition 8.4) and write it as pX, 0q " νďi"1 pXqi , 0q Yďiąj pAqi,qj , 0q , q1 ą ą qν " 1 . Each Xqi is a union of pieces of type Bpqiq and each Aqi,qj is a (possibly empty) union of Apqi, qjq-pieces glued to Xqi and Xqj along boundary components. This refinement may also decompose the thick part by excising from it some D-pieces (which may themselves be further decomposed) which are "thin for the outer met- ric". The importance of the geometric decomposition is that it can be recovered using the outer Lipschitz geometry of pX, 0q. We will show this in Part 3 of the paper. We will need two different constructions of the geometric decomposition, one in terms of a carrousel, and one in terms of resolution. The first is done in Sections 6 to 8 and the second in Section 9. It is worth remarking that the geometric decomposition has a topological descrip- tion in terms of a relative JSJ decomposition of the link X pǫq. Consider the link K Ă X pǫq consisting of the intersection with X pǫq of the polar curves of a generic pair of plane projections X Ñ C2 union a generic pair of hyperplane sections of C. The decomposition of X pǫq as the union of the components of the X pǫq qi 's and the intermediate Apqi, qjq-pieces is topologically the relative JSJ decomposition for the pair pX pǫq, Kq, i.e., the minimal decomposition of X pǫq into Seifert fibered pieces LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 11 separated by annular pieces such that the components of K are Seifert fibers in the decomposition. Actually, the geometric decomposition that we will use is slightly stronger, in that it may refine an annular piece by cutting it into a sequence of annular pieces. We come back to this in Definition 9.1. 6. Polar wedges We denote by Gpk, nq the grassmannian of k-dimensional subspaces of Cn. Definition 6.1 (Generic linear projection). Let pX, 0q Ă pCn, 0q be a normal surface germ. For D P Gpn ´ 2, nq let ℓD : Cn Ñ C2 be the linear projection Cn Ñ C2 with kernel D. Let ΠD Ă X be the polar of this projection, i.e., the closure in pX, 0q of the singular locus of the restriction of ℓD to X r t0u, and let ∆D " ℓDpΠDq be the discriminant curve. There exists an open dense subset Ω Ă Gpn ´ 2, nq such that tpΠD, 0q : D P Ωu forms an equisingular family of curve germs in terms of strong simultaneous resolution and such that the discriminant curves ∆D are reduced and no tangent line to ΠD at 0 is contained in D ([14, (2.2.2)] and [26, V. (1.2.2)]). The projection ℓD : Cn Ñ C2 is generic for pX, 0q if D P Ω. The condition ∆D reduced means that any p P ∆D r t0u has a neighbourhood U in C2 such that one component of pℓDX q´1pU q maps by a two-fold branched cover to U and the other components map bijectively. Let Bǫ be a Milnor ball for pX, 0q (in Section 8 we will specify a family of Milnor balls). Fix a D P Ω and a component Π0 of the polar curve of ℓ " ℓD. We now recall the main result of [1, Section 3], which defines a suitable region A0 containing Π0 in X X Bǫ, outside of which ℓ is a local bilipschitz homeomorphism. Let us consider the branch ∆0 " ℓpΠ0q of the discriminant curve of ℓ. Let V be a small neighbourhood of D in Ω. For each D1 in V let ΠD1,0 be the component of ΠD1 close to Π0. Then the curve ℓpΠD1,0q has Puiseux expansion ajpD1qxpj P Ctx 1 N u, with pj P Q, 1 ď p1 ă p2 ă , y " ÿjě1 where ajpD1q P C. Here N " lcmjě1 denomppjq, where "denom" means denomina- tor. (In fact, if V is sufficiently small, the family of curves ℓpΠD1,0q, for D1 P V is equisingular, see [26, p. 462].) Definition 6.2 (Wedges, contact and polar rates). Let s be the first exponent pj for which the coefficient ajpD1q is non-constant, i.e., it depends on D1 P V . For α ą 0, define B0 :" px, yq :y ´ÿjě1 ajxpj ď αxs( . We call B0 a ∆-wedge (about ∆0). Let A0 be the germ of the closure of the connected component of ℓ´1pB0q r t0u which contains Π0. We call A0 a polar wedge (about Π0). We call s the polar rate of A0 (resp. B0). As described in [1], instead of B0 one can use B1 truncating higher-order terms does not change the bilipschitz geometry. 0 :" px, yq : y ´řjě1,pj ďs ajxpj ď αxs(, since We say two irreducible complex curves C1 and C2 through the origin in Cn have contact q if dpC1 X Sǫ, C2 X Sǫq " Opǫqq. The contact between ΠD1,0 and Π0 is the polar rate of A0 and was called "contact exponent" in [1]. 12 WALTER D NEUMANN AND ANNE PICHON Clearly B0 is a Dpsq-piece. Moreover: Proposition 6.3 (Proposition 3.4(2) of [1]). A polar wedge A0 with rate s is a Dpsq-piece. (cid:3) The proof in [1] shows that A0 can be approximated to high order by a set 0 "Ťtďβ ΠDt,0 for some β ą 0, where t P C, t ď β, parametrizes a piece of a A1 suitable line through D in Gpn ´ 2, nq. We will need some of the details later, so we describe this here. We can choose coordinates pz1, . . . , znq in Cn and px, yq in C2 so that ℓ is the projection px, yq " pz1, z2q and that the family of plane projections ℓDt used in [1] is ℓDt : pz1, . . . , znq Ñ pz1, z2 ´ tz3q. In [1] it is shown that A1 0 can be parametrized as a union of the curves ΠDt,0 in terms of parameters pu, vq as follows: z1 " uN z2 " uN f2,0puq ` v2uN sh2pu, vq zj " uN fj,0puq ` vuN shjpu, vq , ΠDt,0 " tpz1, . . . , znq : v " tu , j " 3, . . . , n where h2pu, vq is a unit. to order ą s with A1 We then have A0 " tpz1pu, vq, . . . , znpu, vqq : v2h2pu, vq ď αu, which agrees up 0 " tpz1pu, vq, . . . , znpu, vqq : v ď βu with β "aα{h2p0, 0q. In [1] it was also pointed out that at least one hjpu, vq with j ě 3 is a unit, by a modification of the argument of Teissier [26, p. 464, lines 7 -- 11]. We will show this using our explicit choice of coordinates above. Lemma 6.4. h3pu, vq is a unit. More specifically, writing h2pu, vq "řiě0 vifi`2puq, we have h3pu, vq "řiě0 vi i`2 Proof. Make the change of coordinates z1 2 :" z2 ´ tz3, so ℓDt is the projection to the pz1, z1 2q-plane. Since ΠDt,0 is given by v " t in our pu, vq coordinates, we change coordinates to pu, wq with w :" v ´ t, so that ΠDt,0 has equation w " 0. We then know from above that z1 i`1 fi`2puq. 2 must have the form z1 2 " uN f 1 2,0puq ` w2uN sh1 2pu, wq with h1 explicitly as 2pu, wq a unit. On the other hand, we rewrite the expressions for z2, z3 vifipuq vigipuq . z2 " uN f2,0puq ` uN sÿiě2 z3 " uN f3,0puq ` uN sÿiě1 z2 " uN f2,0puq ` uN sÿiě2 z3 " uN f3,0puq ` uN sÿiě1 Then direct calculation from pw ` tqifipuq pw ` tqigipuq LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 13 gives: z1 2 "z2 ´ tz3 ti`1pfi`1puq ´ gipuqq "uN pf2,0puq ´ tf3,0puqq ` uN s"ÿiě1 ` wÿiě1 ` w2pf2puq ` tp3f3puq ´ g2puqq ` t2p6f4puq ´ 3g3puqqq ` . . .‰ . tippi ` 1qfi`1puq ´ igipuqq The part of this expression of degree 1 in w must be zero, so gipuq " i`1 for each i. Thus i fi`1puq z3 " uN f3,0puq ` uN sÿiě1 " uN f3,0puq ` vuN sÿiě0 vi i ` 1 i fi`1puq vi i ` 2 i ` 1 fi`2puq , (cid:3) completing the proof. For x P X, we define the local bilipschitz constant Kpxq of the projection ℓD : X Ñ C2 as follows: Kpxq is infinite if x belongs to the polar curve ΠD, and at a point x P X r ΠD it is the reciprocal of the shortest length among images of unit vectors in TxX under the projection dℓD : TxX Ñ C2. For K0 ě 1, set BK0 :" p P X X pBǫ r t0uq : Kppq ě K0( , and let BK0pΠ0q denote the closure of the connected component of BK0 rt0u which contains Π0 rt0u. As a consequence of [1, 3.3, 3.4(1)], we obtain that for K0 ą 1 sufficiently large, the set BK0 can be approximated by a polar wedge about Π0. Precisely: Proposition 6.5. There exist K0, K1 P R with 1 ă K1 ă K0 such that BK0pΠ0q Ă A0 X Bǫ Ă BK1pΠ0q (cid:3) Let us consider again a component Π0 of Π and a polar wedge A0 around it, the corresponding component ∆0 of ∆ and ∆-wedge B0, and let s be their polar rate. Let γ be an irreducible curve having contact r ě s with ∆0. The following Lemma establishes a relation between r and the geometry of the components of the lifting ℓ´1pγq in the polar wedge A0 about Π0. This will be a key argument in Part 4 of the paper (see proof of Lemma 20.3). It will also be used in section 13 to explicitly compute some polar rates. We choose coordinates in Cn as before, with ℓ " pz1, z2q our generic projection. Recall the Puiseux expansion of a component ∆0 of the discriminant: z2 " ÿiěN z2 " ÿiěN aizi{N 1 P Ctz1{N 1 u . aizi{N 1 ` λzr 1 , Lemma 6.6. Let pγ, 0q be an irreducible germ of curve in pC2, 0q with Puiseux expansion: with r ě s, rN P Z and λ P C, 0 ă λ ă 1. Let L1 γ be the intersection of the lifting Lγ " ℓ´1pγq with a polar wedge A0 about Π0 and let ℓ1 another generic plane projection for X which is also generic for the curve ℓ´1pγq. Then the rational number qprq :" s`r 2 is determined by the topological type of the curve ℓ1pL1 γq. 14 WALTER D NEUMANN AND ANNE PICHON Proof. We first consider the case r ą s. Using again the parametrization of A1 0 given after the statement of Proposition 6.3, the curve L1 γ has z2 coordinate satisfying z2 " uN f2,0puq ` v2uN sh2pu, vq " uN f2,0puq ` λuN r, so v2uN sh2pu, vq " λuN r. Write gpu, vq " v2h2pu, vq ´ λuN pr´sq. Since h2pu, vq is a unit, we have gp0, vq ‰ 0, so we can write v as a Puiseux expansion in terms of u as follows: (1) If N pr ´ sq is odd then we have one branch N pr´sq u aiui{2 P Ctu1{2u . (2) If N pr ´ sq is even we have two branches h2p0, 0q v "c λ v " c λ u h2p0, 0q 2 ` ÿiąN pr´sq 2 ` ÿią N pr´sq N pr´sq 2 b i ui P Ctuu . Inserting into the parametrization of A1 0 we get that L1 γ is given by: z1 " uN z2 " uN f2,0puq ` λuN r zj " uN fj,0puq `c λ h2p0, 0q hjp0, 0qu N pr`sq 2 ` h.o. , j " 3, . . . , n , where "h.o." means higher order terms in u. Notice that L1 depending on the parity of N pr ` sq. γ has one or two branches Taking ℓ1 " ℓDt for some t, the curve ℓ1pL1 γq is given by (since r ą r`s 2 ): z1 " uN z1 2 " uN pf2,0puq ´ tf3,0puqq ´ tc λ h2p0, 0q hjp0, 0qu N pr`sq 2 ` h.o. . If N pr ` sq is odd, then ℓ1pL1 is an essential Puiseux exponent of its Puiseux expansion. Otherwise, ℓ1pL1 γq has two components and qprq is the coincidence exponent between their Puiseux expansions. In both cases, qprq is determined by the topological type of ℓ1pL1 γq has one component and qprq " s`r 2 γq. Finally if r " s then L1 A0 by polars, so they have contact qprq " s with each other. γ consists to high order of two fibers of the saturation of (cid:3) 7. Intermediate and complete carrousel decompositions of the discriminant curve In this section we define two carrousel decompositions of pC2, 0q with respect to the discriminant curve ∆ of a generic plane projection ℓ : pX, 0q Ñ pC2, 0q. Intermediate carrousel decomposition. This is obtained by truncating the Puiseux series expansions of the branches of ∆ as follows: if ∆0 is a branch of ∆ with Puiseux expansion y " řiě1 aixpi and if s " pk is the rate of a ∆-wedge about ∆0, then we consider the truncated Puiseux series y "řk i"1 aixpi and form the carrousel decomposition for these truncations (see the construction in Section 3). We call this the unamalgamated intermediate carrousel decomposition. Notice that a piece of this carrousel decomposition which contains a branch of ∆ is in fact a ∆-wedge and is also a Dpsq-piece, where s is the rate of this ∆-wedge. Using Lemma 2.5, we then amalgamate according to the following rules: LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 15 Amalgamation 7.1 (Intermediate carrousel decomposition). (1) We amalgamate any ∆-wedge piece with the piece outside it. We call the resulting pieces ∆-pieces, (2) We then amalgamate any D-piece which is not a ∆-piece with the piece outside it. This may create new D-pieces and we repeat this amalgamation iteratively until no further amalgamation is possible. We call the result the intermediate carrousel decomposition of ∆. It may happen that the rate s of the ∆-wedge piece about a branch ∆0 of ∆ is strictly less that the last characteristic exponent of ∆0. We give an example in [1]. This is why we call this carrousel decomposition "intermediate". Complete carrousel decomposition. This is obtained by truncating each Puiseux series expansion at the first term which has exponent greater than or equal to the polar rate and where truncation does not affect the topology of ∆. By definition, ∆-wedge pieces are not amalgamated, so this carrousel decomposition refines the ∆-pieces of the intermediate carrousel decomposition. We then amalgamate itera- tively any D-pieces which do not contain components of ∆. We call the result the complete carrousel decomposition of ∆. The complete carrousel decomposition is a refinement of both the plain and the intermediate carrousel decompositions. In particular, according to Proposition 3.4, its combinatorics determine the topology of the curve ∆. In Section 14, we complete the proof of part (5) of Theorem 1.2 by proving that the outer Lipschitz geometry of pX, 0q determines the combinatorics of a complete carrousel section of ∆. We close this section with two examples illustrating carrousels. Example 7.2. Let pX, 0q be the D5 singularity with equation x2y`y4`z2 " 0. The discriminant curve of the generic projection ℓ " px, yq has two branches y " 0 and x2 ` y3 " 0, giving us plain carrousel rates of 1 and 3{2. In Example 13.1, we will compute the corresponding polar rates, which equal 2 and 5{2 respectively, giving the additional rates which show up in the intermediate and complete carrousels. See also Example 9.5. Figure 2 shows the sections of three different carrousels for the discriminant of the generic plane projection of the singularity D5: the plain carrousel and the com- plete and intermediate carrousel decompositions. The ∆-pieces of the intermediate carrousel section are in gray, and the pieces they contain in the complete carrousel are D-pieces which are in fact ∆-wedge pieces since for both branches of ∆ the polar rate is greater than the last characteristic exponent. Example 7.3. Our next example was already partially studied in [1]: the surface singularity pX, 0q with equation pzx2 ` y3qpx3 ` zy2q ` z7 " 0. In this case ∆ has 14 branches with 12 distinct tangent lines L1, . . . , L12. ∆ decomposes as follows: (1) Six branches, each lifting to a component of the polar in one of the two thick parts of X. Their tangent lines L1, . . . , L6 move as the linear projection is changed, so their polar rates are 1. (2) Five branches, each tangent to one of L7, . . . , L11, and with 3{2 as single characteristic Puiseux exponent. So their Puiseux expansions have the form u " aiv ` biv3{2 ` . Their polar rates are 3{2. 16 WALTER D NEUMANN AND ANNE PICHON 3{2 1 1 Plain carrousel 3{2 1 5{2 3{2 1 5{2 2 2 Complete carrousel Intermediate carrousel Figure 2. Carrousel sections for D5 1 5{4 3{2 3{2 3{2 6{5 3{2 3{2 Complete carrousel 7{5 1 3{2 3{2 3{2 5{4 6{5 3{2 3{2 Intermediate carrousel 7{5 Figure 3. Carrousel sections for pzx2 ` y3qpx3 ` zy2q ` z7 " 0 LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 17 (3) Three branches δ, δ1, δ2 tangent to the same line L12, each with single char- acteristic Puiseux exponent, respectively 6{5, 6{5, 5{4. Their Puiseux ex- pansions have the form: δ : u " av`bv6{5` , δ1 : u " av`b1v6{5 ` and δ2 : u " av ` b2v5{4 ` . Their polar rates are respectively 7{5, 7{5, 5{4. The polar rates were given in [1, Example 15.2] except for the two with value 7{5, whose computation is explained in Section 13. Figure 3 shows the complete and intermediate carrousel sections. The gray regions in the intermediate carrousel section represent ∆-pieces with rates ą 1 (the Bp1q-piece is also a ∆-piece). 8. Geometric decomposition of pX, 0q In this section we describe the geometric decomposition of pX, 0q as a union of semi-algebraic subgerms by starting with the unamalgamated intermediate car- rousel decomposition of the discriminant curve of a generic plane projection, lifting it to pX, 0q, and then performing an amalgamation of some pieces (see Lemma 2.5). Setup. From now on we assume pX, 0q Ă pCn, 0q and our coordinates pz1 . . . , znq in Cn are chosen so that z1 and z2 are generic linear forms and ℓ :" pz1, z2q : X Ñ C2 is a generic linear projection for X. We denote by Π the polar curve of ℓ and by ∆ " ℓpΠq its discriminant curve. Instead of considering a standard ǫ-ball Bǫ as Milnor ball for pX, 0q, we will use, as in [1], a standard "Milnor tube" associated with the Milnor-Le fibration for the map h :" z1X : X Ñ C. Namely, for some sufficiently small ǫ0 and some R ą 1 we define for ǫ ď ǫ0: Bǫ :" tpz1, . . . , znq : z1 ď ǫ, pz1, . . . , znq ď Rǫu and Sǫ " BBǫ , where ǫ0 and R are chosen so that for ǫ ď ǫ0: (1) h´1ptq intersects the standard sphere SRǫ transversely for t ď ǫ; (2) the polar curve Π and its tangent cone meet Sǫ only in the part z1 " ǫ. The existence of such ǫ0 and R is proved in [1, Section 4]. We will now always work inside the Milnor balls Bǫ just defined. Definition 8.1. A component of a semi-algebraic germ pA, 0q means the closure of a connected component of pA X Bǫq r t0u for sufficiently small ǫ (i.e., the family of Bǫ1 with ǫ1 ď ǫ should be a family of Milnor balls for A). We lift the unamalgamated intermediate carrousel decomposition of ∆ to pX, 0q by ℓ. Any component of the inverse image of a piece of any one of the types Bpqq, Apq, q1q or Dpqq is a piece of the same type from the point of view of its inner geometry (see [1] for details). By a "piece" of the decomposition of pX, 0q we will always mean a component of the inverse image of a piece of pC2, 0q. Amalgamation 8.2 (Amalgamation in X). We now simplify this decomposition of pX, 0q by amalgamating pieces in the two following steps. (1) Amalgamating polar wedge pieces. We amalgamate any polar wedge piece with the piece outside it. We call the resulting pieces polar pieces. (2) Amalgamating empty D-pieces. Whenever a piece of X is a Dpqq-piece con- taining no part of the polar curve (we speak of an empty piece) we amal- gamate it with the piece which has a common boundary with it. These 18 WALTER D NEUMANN AND ANNE PICHON amalgamations may form new empty D-pieces. We continue this amalga- mation iteratively until the only remaining D-pieces contain components of the polar curve. Polar pieces are then all the D-pieces and some of the Apq, qq- and B-pieces. Remark 8.3. The inverse image of a ∆-wedge B0 may consist of several pieces. All of them are D-pieces and at least one of them is a polar wedge in the sense of Definition 6.2. In fact a polar piece contains at most one polar wedge over B0. The argument is as follows. Assume B0 is a ∆-wedge about the component ∆0 of ∆. Let Π0 be the component of Π such that ∆0 " ℓpΠ0q and let N be the polar piece containing it. Assume Π1 0 is another component of Π inside N . According to Section 6, Π X N consists of equisingular components having pairwise contact s. Since the projection ℓΠ : Π Ñ ∆ is generic ([26, Lemme 1.2.2 ii)]), we also have dpℓpΠ0 X Sǫq, ℓpΠ1 0 X Sǫqq " Opǫsq. Therefore, if B0 is small enough, we get B0 X ℓpΠ1 0q " t0u. Except for the Bp1q-piece, each piece of the decomposition of pC2, 0q has one "outer boundary" and some number (possibly zero) of "inner boundaries." When we lift pieces to pX, 0q we use the same terminology outer boundary or inner bound- ary for the boundary components of components of lifted pieces. After performing Amalgamation 8.2, if q is the rate of some piece of pX, 0q we denote by Xq the union of all pieces of X with this rate q. There is a finite collection of such rates q1 ą q2 ą ą qν. In fact, qν " 1 since Bp1q-pieces always exist, and X1 is the union of the Bp1q-pieces. Definition 8.4 (Geometric decomposition of pX, 0q). We say geometric de- composition of pX, 0q for the union pX, 0q " pAqi,qj , 0q , νďi"1 pXqi , 0q Yďiąj where Aqi,qj is a union of intermediate Apqi, qjq-pieces between Xqi and Xqj and the semi-algebraic sets Xqi and Aqi,qj are pasted along their boundary components. Note that the construction of pieces via a carrousel involves choices which imply that, even after fixing a generic plane projection, the pieces are only well defined up to adding or removing collars of type Apq, qq at their boundaries. Definition 8.5 (Equivalence of pieces). We say that two pieces with same rate q are equivalent if they can be made equal by attaching collars of type Apq, qq at their boundaries. Similarly, two Apq, q1q-pieces are equivalent if they can be made equal by removing Apq, qq collars at their outer boundaries and Apq1, q1q collars at their inner boundaries. The following is immediate from Propositions 9.4 and 10.1 below: Proposition 8.6. The geometric decomposition is unique up to equivalence of the pieces. (cid:3) 9. Geometric decomposition through resolution In this section, we describe the geometric decomposition using a suitable resolu- tion of pX, 0q. We will do this in two steps, first considering the following coarser decomposition: LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 19 Definition 9.1 (Inner geometric decomposition). The inner geometric de- composition of pX, 0q is obtained from the geometric decomposition 8.4 by Amal- gamations 8.2 and adding: (3) we amalgamate B-pieces which are Apq, qq-pieces (but not polar pieces) with the A-pieces adjacent to them. We remark that, other than that polar pieces are never amalgamated, the inner geometric decomposition is the decomposition used in [1] to classify inner bilipschitz geometry of X (see Theorem 1.7 and Sec. 15 of [1]). We first need to describe the Nash modification of pX, 0q. Definition 9.2 (Nash modification). Let λ : X r t0u Ñ Gp2, nq be the Gauss graph of λ in X Gp2, nq is a reduced analytic surface. The Nash modification of map which maps x P X r t0u to the tangent plane TxX. The closure qX of the pX, 0q is the induced morphism ν : qX Ñ X. In this section we consider the minimal good resolution π : prX, Eq Ñ pX, 0q with According to [23, Part III, Theorem 1.2], a resolution of pX, 0q factors through the Nash modification if and only if it has no basepoints for the family of polar curves ΠD parametrized by D P Ω. the following three properties: (1) it resolves the basepoints of a general linear system of hyperplane sections of pX, 0q (i.e., it factors through the normalized blow-up of the maximal ideal of X); (2) it resolves the basepoints of the family of polar curves of generic plane projections (i.e., it factors through the Nash modification of X); (3) there are no adjacent nodes in the resolution graph (nodes are defined in Definition 9.3 below). This resolution is obtained from the minimal good resolution of pX, 0q by blowing up further until the basepoints of the two kinds are resolved and then by blowing up intersection points of exceptional curves corresponding to nodes. We denote by Γ the dual resolution graph and by E1, . . . , Er the exceptional curves in E. We denote by vk the vertex of Γ corresponding to Ek. Definition 9.3. An L-curve is an exceptional curve in π´1p0q which intersects the strict transform of a generic hyperplane section. The vertex of Γ representing an L-curve is an L-node. A P-curve (P for "polar") is an exceptional curve in π´1p0q which intersects the strict transform of the polar curve of any generic linear projection. The vertex of Γ representing this curve is a P-node. A vertex of Γ is called a node if it is an L- or P-node or has valency ě 3 or represents an exceptional curve of genus ą 0. A string of a resolution graph is a connected subgraph whose vertices have valency 2 and are not nodes, and a bamboo is a string attached to a non-node vertex of valency 1. For each k " 1, . . . , r, let N pEkq be a small closed tubular neighbourhood of Ek and let N pEkq " N pEkq r ďk1‰k N pEk1 q. 20 WALTER D NEUMANN AND ANNE PICHON For any subgraph Γ1 of Γ define: N pΓ1q :" ďvkPΓ1 N pEkq and N pΓ1q :" N pΓq r ďvkRΓ1 N pEkq . Proposition 9.4. The pieces of the inner geometric decomposition of pX, 0q (Def- inition 9.1) can be described as follows: For each node vi of Γ, let Γi be vi union any attached bamboos. Then, with "equivalence" defined as in Definition 8.5, (1) the Bp1q-pieces are equivalent to the sets πpN pΓj qq where vj is an L-node; (2) the polar pieces are equivalent to the sets πpN pΓj qq where vj is a P-node; (3) the remaining Bpqq are equivalent to the sets πpN pΓj qq where vj is a node which is neither an L-node nor a P-node; (4) the Apq, q1q-pieces with q ‰ q1 are equivalent to the πpN pσqq where σ is a maximal string between two nodes. We describe later how the geometric decomposition results from the inner one as nodes (after possibly extending the strings by blowing up). A straightforward by modifying Γ if necessary to a graphpΓ given by labeling some vertices on strings consequence will be that to each node vj of pΓ one can associate the rate qj of the corresponding piece of the decomposition, and the graph pΓ with nodes weighted by their rates determines the geometric decomposition pX, 0q " Ťν Ťiąj pAqi,qj , 0q up to equivalence, proving the unicity stated in Proposition 8.6. Before proving Proposition 9.4 we give two examples. They each have geometric decomposition the same as inner geometric decomposition; an example where the decompositions differ is Example 14.4. i"1pXqi , 0q Y Example 9.5. Let pX, 0q be the D5 singularity with equation x2y ` y4 ` z2 " 0. Carrousel decompositions for its discriminant ∆ are described in Example 7.2. Let v1, . . . , v5 be the vertices of its minimal resolution graph indexed as follows (all Euler weights are ´2): v5 v1 v2 v3 v4 The multiplicities of the restriction h to X of a generic linear form C3 Ñ C are given by the minimum of the compact part of the three divisors pxq, pyq and pzq: ph πq " E1 ` 2E2 ` 2E3 ` E4 ` E5 ` h, where means strict transform. The minimal resolution resolves a general linear system of hyperplane sections and the strict transform of h is one curve intersecting E3. So v3 is the single L-node. But as we shall see, this is not yet the resolution which resolves the family of polars. The total transform by π of the coordinate functions x, y and z are: px πq " 2E1 ` 3E2 ` 2E3 ` E4 ` 2E5 ` x py πq " E1 ` 2E2 ` 2E3 ` 2E4 ` E5 ` y pz πq " 2E1 ` 4E2 ` 3E3 ` 2E4 ` 2E5 ` z . Set f px, y, zq " x2y ` y4 ` z2. The polar curve Π of a generic linear projection ℓ : pX, 0q Ñ pC2, 0q has equation g " 0 where g is a generic linear combination of LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 21 the partial derivatives fx " 2xy, fy " x2 ` 4y3 and fz " 2z. The multiplicities of g are given by the minimum of the compact part of the three divisors pfx πq " 3E1 ` 5E2 ` 4E3 ` 3E4 ` 3E5 ` f x pfy πq " 3E1 ` 6E2 ` 4E3 ` 2E4 ` 3E5 ` f y pfz πq " 2E1 ` 4E2 ` 3E3 ` 2E4 ` 2E5 ` f z . So the total transform of g is equal to: pg πq " 2E1 ` 4E2 ` 3E3 ` 2E4 ` 2E5 ` Π . 2 , which intersect respectively E2 and E4. In particular, Π is resolved by π and its strict transform Π has two components Π 1 and Π Since the multiplicities m2pfxq " 5, m2pfyq " 4 and m2pzq " 6 along E2 are distinct, the family of polar curves of generic plane projections has a basepoint on E2. One must blow up once to resolve the basepoint, creating a new exceptional curve E6 and a new vertex v6 in the graph. So we obtain two P-nodes v4 and v6 as in the resolution graph below (omitted Euler weights are ´2): Π 1 v6 ´1 v4 Π 2 ´3 We will compute in Example 13.1 that the two polar rates are 5{2 and 2. It follows that the geometric decomposition of X consists of the pieces X5{2 " πpN pE6qq, X2 " πpN pE4qq, X3{2 " πpN pE1YE2YE5qq and X1 " πpN pE3qq , plus intermediate A-pieces, and that X5{2 and X2 are the two polar pieces. The geometric decomposition for D5 is described by Γ with vertices vi weighted with the corresponding rates qi: 3{2 5{2 ´1 ´3 3{2 3{2 1 2 Example 9.6. We return to example 7.3 with equation pzx2`y3qpx3`zy2q`z7 " 0, to clarify how the carrousels described there arise. We first compute some invariants from the equation. The following picture shows the resolution graph of a general linear system of hyperplane sections. The negative numbers are self-intersection of the exceptional curves while numbers in parentheses are the multiplicities of a generic linear form. 22 WALTER D NEUMANN AND ANNE PICHON p1q ´23 p10q ´1 p5q ´2 ´1 p2q ´1 p2q p4q ´5 p1q ´23 p10q ´1 p5q ´2 Set f px, y, zq " pzx2 ` y3qpx3 ` zy2q ` z7. The multiplicities of the partial derivatives pmpfxq, mpfyq, mpfzqq are given by: pě 10, ě 10, ě 10q pě 5, ě 5, ě 5q pě 5, ě 5, ě 5q p57, ě 58, ě 60q pě 58, 57, ě 60q p29, 29, ě 30q p23, 23, ě 25q p29, 29, ě 30q The basepoints of the family of polar curves of generic plane projections are resolved by this resolution and the strict transform of the generic polar curve Πℓ is given by the following graph. The multiplicities are those of a generic linear combination g " afx ` bfy ` cfz. p11q p5q p5q p11q p57q p57q p29q p23q p29q We now indicate on the resolution graph the rate corresponding to each vertex. LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 23 3{2 3{2 1 1 6{5 6{5 7{5 5{4 7{5 These rates were computed in [1] except for the rates 7{5 which are explained in Section 13 Proof of Proposition 9.4. Let ℓ : pX, 0q Ñ pC2, 0q be again a generic plane projec- tion. Let ρ : Y Ñ C2 be the minimal sequence of blow-ups starting with the blow-up of 0 P C2 which resolves the basepoints of the family of images ℓpΠDq by ℓ of the polar curves of generic plane projections and let ∆ be the discriminant curve of ℓ. We set ρ´1p0q "Ťm k"1 Ck, where C1 is the first curve blown up. Denote by R the dual graph of ρ, so v1 is its root vertex. We say ∆-curve for an exceptional curve in ρ´1p0q intersecting the strict transform of ∆, and ∆-node for a vertex of R which represents a ∆-curve. We call any vertex of R which is either v1 or a ∆-node or a vertex with valency ě 3 a node of R. If two nodes are adjacent, we blow up the intersection points of the two corre- sponding curves in order to create a string between them. Consider the intermediate carrousel decomposition of ∆ after amalgamation of the ∆-wedge pieces (so we don't amalgamate any other piece here). This carrousel decomposition can be described as follows: Lemma 9.7. (1) The Bp1q-piece is equivalent to the set ρpN pC1qq; (2) the ∆-pieces are equivalent to the sets ρpN pCkqq where vk is a ∆-node; (3) the remaining B-pieces are equivalent to the sets ρpN pCkqq where vk is a node which is neither v1 nor a ∆-node; (4) the sets ρpN pβqq where β is a bamboo of R are equivalent to empty D-pieces; (5) the A-pieces are equivalent to the sets ρpN pσqq where σ is a maximal string between two nodes. Proof of Lemma 9.7. Let Bκ be a B-piece as defined in section 3: Bκ :"!px, yq : ακxpk ď y ´ k´1ÿi"1 aixpi ď βκxpk y ´ p k´1ÿi"1 aixpi ` akj xpk q ě γκxpk for j " 1, . . . , mκ) with ακ, βκ, γκ satisfying ακ ă akj ´ γκ ă akj ` γκ ă βκ for each j " 1, . . . , mκ. Then Bκ is foliated by the complex curves Cλ, λ P Λ with Puiseux expansions: y " k´1ÿi"1 aixpi ` λxpk , 24 WALTER D NEUMANN AND ANNE PICHON where Λ " tλ P C; ακ ď λ ď βκ and λ ´ akj ě γκ, j " 1, . . . , mκu. The curves Cλ are irreducible and equisingular in terms of strong simultaneous resolution, and Bκ " ρpN pCqq where C is the exceptional component of ρ´1p0q which intersects the strict transforms C λ . We use again the notations of Section 3. The Bp1q-piece is the closure of the cones V pjq, j " 1, . . . , m, so it is foliated by the complex lines with equations y " ax, where a P C is such that apjq 1 ´ a ě η for all j " 1, . . . , m. This implies (1) since C1 is the exceptional curve which intersects the strict transform of those lines. Recall that a ∆-piece is a B-piece obtained by amalgamating ∆-wedges with a B-piece outside them. By Definition 6.2, a ∆-wedge is foliated by curves y " řpj ďs ajxpj ` axs, which are resolved in family by the resolution ρ, their strict transforms intersecting a ∆-node. This proves (2) and then (3). (4) and (5) immediately follow from the fact that the D- and A-pieces are the (cid:3) closures of the complement of the B-pieces. We now complete the proof of Proposition 9.4. Let πhj : Xhj Ñ X be the Hirzebruch-Jung resolution of pX, 0q obtained by pulling back the morphism ρ by the cover ℓ and then normalizing and resolving the remaining quasi-ordinary sin- gularities. We denote its dual graph by Γhj. Denote by ℓ1 : Xhj Ñ Y the morphism defined by ℓ πhj " ρ ℓ1. Now πhj factors through π by a morphism µ : Xhj Ñ rX. Let us choose a bundle neighborhood N pCkq for each component Ck of ρ´1p0q. Then ℓ1´1pN pCkqq is a union of bundle neighbourhoods N pE1 j of pρ ℓ1q´1p0q. For each component Ei of π´1p0q we can define the bundle neighbourhood N pEiq by j q of curves E1 N pEiq " µ` ďE1 jõ´1pEiq N pE1 jq. By construction, we have the following two correspondences between the resolu- tion graphs R associated with ρ´1p0q and Γ associated with π´1p0q Ă rX. (1) If R0 is a subgraph of R, let ΓR0 be the subgraph of Γ which represents Ckqq; if R0 Ă R is a maximal the irreducible components of µpℓ1´1pŤvkPR0 j be the irreducible jq. string between two nodes of R, then ΓR0 is a union of strings in Γ; (2) For each node (resp. P-node, resp. L-node) vi of Γ, let E1 component of µ´1pEiq which maps surjectively to Ei and set Ck " ℓ1pE1 Then vk is a node (resp. ∆-node, resp. the root v1) of R. We then obtain that for each node (resp. P-node, resp. L-node) vi of Γ, the set πpN pEiqq is a connected component of pℓ πq´1pN pCkqq where vk is a node (resp. a P-node, resp. the root v1) of R, and Proposition 9.4 is now a straightforward consequence of Lemma 9.7. (cid:3) Finally, we return to the geometric decomposition (Definition 8.4). Note that it differs from the "inner geometric decomposition" (Definition 9.1) only in that some annular pieces may have been decomposed by Apq, qq-pieces into sequences of annular pieces. Proposition 9.8. There exists a resolution pπ obtained from π by blowing-up fur- ther, if necessary, intersection points of exceptional curves (so creating only new valency 2-vertices) such that the Apq, qq-pieces of the geometric decomposition of LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 25 pX, 0q (Definition 8.4) are equivalent to some pieces πpN pvqq where v are valency 2 nodes of the resolution graph pΓ of pπ which are not P- or L-nodes. section, so we omit the (easy) proof of this proposition. We give a more detailed description of the location of Apq, qq-pieces in the next 10. The Apq, qq-pieces of the geometric decomposition. One goal of this section is to provide tools which will be used in Sections 12 and 13 for detecting the geometric decomposition and the carrousel using only the outer geometry of pX, 0q. The following proposition, which is a consequence of Lemmas 10.8 and 10.9, describes which B-pieces of type Apqj, qjq of the geometric decomposition appear in annular Apqi, qkq-pieces (with qi ă qj ă qk) of the inner geometric decomposition. Note that for any such piece A of the geometric decomposition the map ℓA : A Ñ ℓpAq is an unbranched covering. Proposition 10.1. An Apqj, qjq-piece occurs inside an Apqi, qkq-piece A of the inner geometric decomposition (Definition 9.1) with qi ă qj ă qk if and only if either: (1) there is a non-annular Bpqj q-piece of the geometric decomposition with ℓpBpqj qq Ă ℓpAq and the outer boundary of ℓpBpqj qq is isotopic to the inner and outer boundaries of ℓpAq, or: (2) there is an Apqi1 , qk1 q-piece A1 of the inner geometric decomposition with qi1 ă qj ă qk1 and (a) the inner boundaries of ℓpAq and ℓpA1q are inside ℓpA1q resp. ℓpAq; (b) dpBipℓpAqq X S3 ǫ , BipℓpA1qq X S3 means inner boundary and S3 ǫ q " Opǫqj q, where d is outer distance, Bi ǫ is the sphere of radius ǫ about 0 in C2. Remarks. Note that the Apqj , qjq-piece inside an Apqi, qkq-piece with qi ă qj ă qk is unique up to equivalence (Definition 8.5). In case (2) by symmetry there is an Apqj, qjq-piece inside each of A and A1; we have no example of this, although we expect that it can occur. Closeness and proximity classes. For pY, 0q a subgerm of pX, 0q or pC2, 0q, we use the notation Nq,apY q to mean the union of components of tx P X : dpx, Y q ď axqu which intersect Y r t0u. Here again, d refers to outer distance. For i " 2, . . . , ν let X pi´1q be the union of all Xqj with j ď i ´ 1 and all Apqj, qkq-pieces with k ă j ď i ´ 1 connecting them. Remark 10.2. The sets Np,βpX pi´1qq can be used to describe the components of Xqi with non-empty inner boundaries in the following way. First, for 1 ă p ď qi´1 we can describe Np,βpX pi´1qq up to equivalence in terms of the carrousel as follows. Let Bpi´1q be the union of all carrousel pieces which are Bpqj q-pieces with j ď i ´ 1 and Apqj, qkq-pieces with k ă j ď i ´ 1. For β ą 0, let N pi´1q be the union of the components of ℓ´1pNp,βpBpi´1qqq which contain components of X pi´1q. Then for p R tqi, . . . , qν´1u and any β, β1 ą 0, Np,βpX pi´1qq is equivalent to N pi´1q . Moreover, the topology of Np,βpX pi´1qq does not depend on p if qj ă p ă qj´1, and it has the same topology also if p " qj and β is sufficiently large or p " qj´1 and β is sufficiently small, and then again Np,βpX pi´1qq is equivalent to N pi´1q if β1 is also sufficiently large resp. small. p,β1 p,β p,β1 26 WALTER D NEUMANN AND ANNE PICHON As a consequence, the components of Xqi with non-empty boundary are, up to equivalence, included among the components of Nqi,βpX pi´1qq r Nqi,αpX pi´1qq with α sufficiently small and β sufficiently large. We will use this in subsection 12.5 Definition 10.3. Let Y1 and Y2 be two components of Nq,βpX pi´1qq and let q1 be the rational number such that dpY1 X Sǫ, Y2 X Sǫq " Opǫq1 q. We say that Y1 and Y2 are close if q1 ě q. Closeness defines an equivalence relation on the components of Nq,βpX pi´1qq. We call the equivalence classes the proximity classes. Note that the notions "closeness" and "proximity classes" are invariant under replacing pieces by equivalent pieces (Definition 8.5). Moreover, closeness can be measured just on outer boundaries, i.e., dpY1 X Sǫ, Y2 X Sǫq " Opǫq1 q if and only if dpBoY1 X Sǫ, BoY2 X Sǫq " Opǫq1 q, where Bo means outer boundary. Therefore the definition of closeness extends to the components of Nq,βpX pi´1qq r Nq,αpX pi´1qq for α ă β, and depends on q but not on α and β. i be the infimum of q ă qi´1 for which the topology of Nq,1pX i´1q Lemma 10.4. Let q1 has not changed. So we have q1 i ă q ď qi´1 decreases, the number of proximity classes of Nq,1pX pi´1qq is non-increasing, so the number of elements in each proximity class is non-decreasing. i ď qi As q with q1 Proof. Let q ă q1 ď qi´1. Let Y1 and Y2 be two components of Nq,1pX pi´1qq j :" Yj X Nq1,1pX pi´1qq, j " 1, 2, the components of Nq1,1pX pi´1qq contained and Y 1 respectively in Y1 and Y2. If Y 1 (cid:3) 2 are close, then Y1 and Y2 are close. 1 and Y 1 Definition 10.5. Let Y be a component of X pi´1q and q ă qi´1. We say that the cardinality of the proximity class of Y grows at q if for any p with q ă p ď qi´1, the proximity class of the component of Np,1pX pi´1qq containing Y has constant cardinality while for any p ă q, it has cardinality strictly larger. Lemma 10.6. For 1 ă p ď qi´1, let N1 and N2 be two components of N pi´1q (see Remark 10.2) and N 1 2 pieces equivalent to them (e.g., components of ℓ´1pNp,βpBpi´1qqq). Then ℓpN1q and ℓpN2q have the same outer boundaries if and only if N 1 1 and N 1 p,β 2 are close. 1 and N 1 Proof. Let q be such that the outer distance dpN 1 1 X Sǫ, N 1 2 X Sǫq is Opǫqq. Assume N 1 1 and N 1 1 and N 1 1 and N 2 2 be N 1 2 are close, i.e., q ě p. Depending on p, choose j and q1 with qj ă q1 ă p ď qj´1 and let N 2 2 union Apq1, pq-pieces added on their outer boundaries. Since ℓ is a linear projection, dpℓpxq, ℓpyqq ď dpx, yq for any x, y P Cn. Therefore dpℓpN 1 q with q ď q2. Assume ℓpN1q and ℓpN2q do not have equal outer boundaries. Then ℓpN 2 2 q cannot have equal outer boundaries, and since ℓpN 2 2 q are subgerms in C2 with Apq1, pq-pieces at their outer boundaries, we obtain that dpℓpN 2 2 X Sǫqq is Opǫq0 q with q0 ď q1. Finally, q2 ď q0, as dpℓpN 2 1 X Sǫq, ℓpN 2 1 X Sǫq, ℓpN 1 1 X Sǫq, ℓpN 2 2 X Sǫqq ď dpℓpN 1 2 X Sǫqq is Opǫq2 2 X Sǫqq. Summarizing: 1 X Sǫq, ℓpN 1 1 q and ℓpN 2 1 q and ℓpN 2 which is a contradiction. q ď q2 ď q0 ď q1 ă p ď q , Conversely, suppose p ą q. Since the outer distance dpN 1 1 X Sǫ, N 1 Opǫqq ąą Opǫpq, if ℓD ‰ ℓ is a generic plane projection for X then ℓDpN 1 2 X Sǫq is 1q X ℓDpN 1 2q LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 27 is empty. Since N 1 ℓDpN2q, so ℓpN1q and ℓpN2q could not have equal outer boundaries. j is equivalent to Nj for j " 1, 2 the same holds for ℓDpN1q and (cid:3) Let ℓ : pX, 0q Ñ pC2, 0q be a generic projection and let ∆ be its discriminant curve. Recall that for all i " 1, . . . , ν ´ 1, qi is one or more of: a polar rate, an essential Puiseux exponent of ∆ or a coincidence exponent between two branches of ∆. Let ∆0 be a component of ∆ and s0 its polar rate and let qi ď s0 be a carrousel rate which is either equal to s0 or is an essential Puiseux exponent of ∆0 or a coincidence exponent of ∆0 with another component of ∆. Let B be the Bpqiq- piece of the carrousel such that the Dpqiq-piece D formed of a union of carrousel pieces and with common outer boundary with B, contains ∆0. Now let D1 be the component of ℓ´1pDq which contains the component of the polar curve Π0 over ∆0 Lemma 10.8. Let i P t1, . . . , ν ´ 1u. Let p∆0, qiq be a pair as above with qi ‰ s0. not an Apqi, qiq-piece; and let rB be the union of components of Xqi contained in D1. Definition 10.7. rB is called the bunch of components associated with p∆0, qiq. Note that each outer boundary of D1 is an outer boundary of a component of rB and by Lemma 10.6 the components of rB are close to each other. Then the associated components of rB have non-empty inner boundary. Moreover, (1) if qi is an essential Puiseux exponent of ∆0, then some Bpqiq-piece in rB is (2) if qi is a coincidence exponent between two branches ∆0 and ∆1 of ∆ which is strictly less than their polar rates s0 and s1, then the corresponding two components B0 and B1 of Xs0 and Xs1 satisfy dpB0 XSǫ, B1 XSǫq " Opǫqi q; Conversely, for all j, k P t1 . . . , ν ´ 1u, if there exist two components B0 and B1 of Xqj and Xqk respectively such that dpB0 X Sǫ, B1 X Sǫq " Opǫqq with q ă qj, qk, then q " qi for some i ą j, k, and qi is a coincidence coefficient between two components of ∆ corresponding to the components B0 and B1. Proof. The first part of the Lemma is obvious since any piece of the geometric decomposition with empty inner boundary is a polar piece. Now assume qi is an essential Puiseux exponent of ∆0 and is not its polar rate. Let D, B and rB be as in the paragraph before Definition 10.7 and let D0 be the component of D r B which contains ∆0. Denote by Ft the set tz1 " tu X X where ℓ " pz1, z2q and z1 is a generic linear function for X. Since qi is an essential expo- nent, the section tz1 " tuXD0 consists of several discs whose boundary components are inner boundaries of the section B X tz1 " tu. These boundaries lift by ℓ into the inner boundaries of some components of rB. Let B1 be such a component. Since qi is not a polar rate, no component of ℓ´1pD0q amalgamates with B1. So the section Ft X B1 has several inner boundary components, and B1 cannot be an annular piece. This proves (1). Assume now that qi is a coincidence exponent between two branches ∆0 and ∆1 of ∆ which is strictly less than their polar rates s0 and s1. Let Π0 and Π1 be the two branches of the polar curve such that ℓpΠiq " ∆i. According to [26, page 462 Lemme 1.2.2 ii)], the restriction ℓΠ : Π Ñ ∆ is a generic plane projection of the curve Π, so dp∆0 X Sǫ, ∆1 X Sǫq " dpΠ0 X Sǫ, Π1 X Sǫq " Opǫqi q. Let B0 (resp. B1) be the Bps0q-piece of Xs0 (resp. Bps1q-piece of Xs1 ) which contains the component Π0 (resp. Π1) of the polar curve over ∆0 (resp. ∆1). Since qi ă s0, s1, we have dpB0 X Sǫ, B1 X Sǫq " dpΠ0 X Sǫ, Π1 X Sǫq " Opǫqi q. This proves (2). 28 WALTER D NEUMANN AND ANNE PICHON Conversely, assume there exist two components B0 and B1 of Xqj and Xqk respectively such that dpB0 X Sǫ, B1 X Sǫq " Opǫqq with q ă qj, qk. Let B1 1 and B1 2 be the components of X pjq resp. X pkq with empty inner boundaries and which have common outer boundaries with B0 resp. B1. Let Π0 and Π1 be components of the polar curve Π inside B1 2 are unions of B and A-pieces with rates ě q, then dpΠ0 X Sǫ, Π1 X Sǫq " Opǫqq. Again by [26, page 462 Lemme 1.2.2 ii)], we have dp∆0 X Sǫ, ∆1 X Sǫq " Opǫqq where ℓpΠ0q " ∆0 and ℓpΠ1q " ∆1. Therefore q is the coincidence coefficient between ∆0 and ∆1. (cid:3) 2. Since B1 1 resp. B1 1 and B1 Lemma 10.9. Let i P t2, . . . , ν ´ 1u. Let X 2 qi consist of the components of Xqi with nonempty inner boundary. Then X 2 qi is a union of proximity classes, each of which has at least one component of Xqi corresponding to a pair p∆0, qiq as above. More precisely, if B is a Bpqiq-piece of the carrousel, the components of X 2 qi whose images by ℓ have same outer boundary as B are the components of a proximity qi if and only if either class, and the components of a proximity class pB are in X 2 pB contains a Bpqiq-piece which is not an Apqi, qiq-piece or the cardinality of the proximity class of X pi´1q X pB grows at qi. Proof. X 2 qi is non-empty if and only if there exists a component ∆0 as above such that qi ‰ s0. Let B be a Bpqiq-piece of the carrousel corresponding to such a pair p∆0, qiq. By Lemma 10.6, any two components of Xqi projecting onto B are close, empty inner boundary belongs to X 2 of X 2 so they belong to the same proximity class pB, and any component of pB with non- qi is inside such a proximity class pB. Moreover, according to Lemma 10.8, any such pB contains a component which is not an Apqi, qiq-piece or the cardinality of the proximity class of X pi´1q X pB grows at qi (this gives cases (2a) and (2b) of Proposition 10.1). qi. Conversely, it is clear that any component (cid:3) Part 3: Analytic invariants from Lipschitz geometry 11. General hyperplane sections and maximal ideal cycle In this section we prove parts (1) to (3) of Theorem 1.2 in the Introduction. We need some preliminary results about the relationship between thick-thin decompo- sition and resolution graph and about the usual and metric tangent cones. The geometric decompositions (see, e.g., Propositions 9.4 and 9.8), which are strong refinements of the thick-thin decomposition, are not needed in this section. Thick-thin decomposition and resolution graph. According to [1, Theorem 1.6], the thick-thin decomposition of (X,0) is determined up to homeomorphism close to the identity by the inner Lipschitz geometry, and hence by the outer Lipschitz geometry (specifically, the homeomorphism φ : pX, 0q Ñ pX, 0q satisfies dpp, φppqq ď pq for some q ą 1). It is described through resolution as follows. The thick part pY, 0q is the union of semi-algebraic sets which are in bijection with the L-nodes of any resolution which resolves the basepoints of a general linear system of hyperplane sections. More precisely, let us consider the minimal good resolution π1 : pX 1, Eq Ñ pX, 0q of this type and let Γ be its resolution graph. Let v1, . . . , vr be the L-nodes of Γ. For each i " 1, . . . , r consider the subgraph Γi of Γ consisting of the L-node vi plus any attached bamboos (ignoring P-nodes). Then the thick LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 29 part is given by: pY, 0q " pYi, 0q where Yi " π1pN pΓiqq . Let Γ1 1, . . . , Γ1 s be the connected components of Γ rŤr pZ, 0q " pZj, 0q where Zj " π1pN pΓ1 j qq . i"1 Γi. Then the thin part is: rďi"1 sďj"1 Recall that we denote by Apǫq :" A X Sǫ (with Sǫ " BBǫ) the link of any semi- algebraic subgerm pA, 0q Ă pX, 0q for ǫ sufficiently small. We call the links Y pǫq and Z pǫq thick and thin zones respectively. i j Inner geometry and L-nodes. Since the graph Γi is star-shaped, the thick zone Y pǫq is a Seifert piece in a graph decomposition of the link X pǫq. Therefore the i inner Lipschitz geometry already tells us a lot about the location of L-nodes: for a thick zone Y pǫq i with unique Seifert fibration (i.e., not an S1-bundle over a disk or annulus) the corresponding L-node is determined in any negative definite plumbing graph for the pair pX pǫq, Y pǫq q. However, a thick zone may be a solid torus D2 S1 or toral annulus I S1 S1. Such a zone corresponds to a vertex on a chain in the resolution graph (i.e., a subgraph whose vertices have valency 1 or 2 and represent curves of genus 0) and different vertices along the chain correspond to topologically equivalent solid tori or toral annuli in the link X pǫq. Thus, in general inner Lipschitz geometry is insufficient to determine the L-nodes and we need to appeal to the outer metric. i Tangent cones. We will use the Bernig-Lytchak map φ : T0pXq Ñ T0pXq between the metric tangent cone T0pXq and the usual tangent cone T0pXq ([3]). We will need its description as given in [1, Section 9]. j Ñ S1 Consider the restriction h " z1X : X Ñ C of a generic linear form. The map h restricts to a fibration ζj : Z pǫq ǫ , and, as described in [1, Theorem 1.7], the components of the fibers have inner diameter opǫqq for some q ą 1. If one scales the inner metric on X pǫq by 1 ǫ then in the Gromov-Hausdorff limit as ǫ Ñ 0 the components of the fibers of each thin zone collapse to points, so each thin zone collapses to a circle. On the other hand, the rescaled thick zones are basically stable, except that their boundaries collapse to the circle limits of the rescaled thin zones. The result is the link T p1qX of the so-called metric tangent cone T0X (see [1, Section 9]), decomposed as T p1qX " lim ǫÑ0 GH 1 ǫ X pǫq "ď T p1qYi , , and these are glued along circles. where T p1qYi " limGH ǫÑ0 i One can also consider 1 1 ǫ Y pǫq ǫ X pǫq as a subset of the unit p2n ´ 1q-sphere S1 " BB1 Ă Cn and form the Hausdorff limit in S1 to get the link T p1qX of the usual tangent cone T0X (this is the same as taking the Gromov-Hausdorff limit for the outer metric). One thus sees a natural branched covering map T p1qX Ñ T p1qX which extends to a map of cones φ : T0pXq Ñ T0pXq (first described in [3]). We denote by T p1qYi the piece of T p1qX corresponding to Yi (but note that two different Yi's can have the same T p1qYi). 30 WALTER D NEUMANN AND ANNE PICHON j : 1 j Proof of (1) of Theorem 1.2. Let Lj be the tangent line to Zj at 0 and hj the map hj :" z1BpLjXBǫq : BpLj X Bǫq -- Ñ S1 ζj to a fibration ζ1 j moves points distance opǫq´1q, so the fibers of ζ1 j are shrinking at this rate. In particular, once ǫ is sufficiently small the outer Lipschitz geometry of 1 j determines this fibration up to homotopy, and hence also up to isotopy, since homotopic fibrations of a 3-manifold to S1 are isotopic (see e.g., [8, p. 34]). j Ñ BpLj X B1q, and written in this form ζ1 ǫ . We can rescale the fibration h´1 ǫ Z pǫq ǫ Z pǫq ǫ Y pǫq i Consider now a rescaled thick piece Mi " 1 . The intersection Ki Ă Mi of Mi with the rescaled link of the curve th " 0u Ă pX, 0q is a union of fibers of the Seifert fibration of Mi. The intersection of a Milnor fiber of h with Mi gives a homology between Ki and the union of the curves along which a Milnor fiber meets BMi, and by the previous paragraph these curves are discernible from the outer Lipschitz geometry, so the homology class of Ki in Mi is known. It follows that the number of components of Ki is known and Ki is therefore known up to isotopy, at least in the case that Mi has unique Seifert fibration. If Mi is a toral annulus the argument still works, but if Mi is a solid torus we need a little more care. If Mi is a solid torus it corresponds to an L-node which is a vertex of a bamboo. If it is the extremity of this bamboo then the map T p1qYi Ñ T p1qYi is a covering. Otherwise it is a covering branched along its central circle. Both the branching degree pi and the degree di of the map T p1qYi Ñ T p1qYi are determined by the Lipschitz geometry, so we can compute di{pi, which is the number of times the Milnor fiber meets the central curve of the solid torus Mi. A tubular neighbourhood of this curve meets the Milnor fiber in di{pi disks, and removing it gives us a toral annulus for which we know the intersection of the Milnor fibers with its boundary, so we find the topology of Ki Ă Mi as before. link Ť Ki of the strict transform of h in the link X pǫq. Denote K 1 pX pǫq,Ť K 1 We have thus shown that the Lipschitz geometry determines the topology of the i " Ki unless pMi, Kiq is a knot in a solid torus, i.e., Ki is connected and Mi a solid torus, in which case put K 1 i " 2Ki (two parallel copies of Ki). The resolution graph we are seeking represents a minimal negative definite plumbing graph for the pair iq, for pX, 0q. By [18] such a plumbing graph is uniquely determined by the topology. When decorated with arrows for the Ki only, rather than the K 1 i, it gives the desired decorated resolution graph Γ. So Γ is determined by pX, 0q and its Lipschitz geometry. (cid:3) Proof of (2) of Theorem 1.2. Recall that π1 : X 1 Ñ X denotes the minimal good resolution of pX, 0q which resolves a general linear system of hyperplane sections. k"1 Ek the decomposition of the exceptional divisor π1´1p0q into its irreducible components. By point (1) of the theorem, the Lipschitz geometry of pX, 0q determines the reso- lution graph of π1 and also determines h Ek for each k. We therefore recover the k"1 mkphqEk ` h of h (since El phq " 0 for all l " 1, . . . , d Denote by h the strict transform by π1 of the generic linear form h and letŤd total transform phq :"řd In particular, the maximal ideal cycleřd řd k"1 mkphqEk is determined by the geom- etry, and the multiplicity of pX, 0q also, since it is given by the intersection number (cid:3) and the intersection matrix pEk Elq is negative definite). k"1 mkphqEk h. LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 31 Proof of (3) of Theorem 1.2. Let H be the kernel of a linear form Cn Ñ C which is generic for X and K " X X H the hyperplane section. So K Ă tπ1pN pEiqq : Ei an L-curveu . rďi"1 For i " 1, . . . , r the tangent cone of K X π1pN pEiqq at 0 is a union of lines, say Li1Y YLiαi. For ν " 1, . . . , αi denote by Kiν the union of components of K which are tangent to Liν . Note the curves Ei, i " 1, . . . , r are the result of blowing up the maximal ideal mX,0 and normalizing. If two of the curves Ei and Ej coincided before normalization, then we will have equalities of the form Liν " Ljν1 with i ‰ j (whence also Kiν " Kjν1 ). For fixed i and varying ν the curves Kiν are in an equisingular family and hence have the same outer Lipschitz geometry. We must show that we can recover the outer Lipschitz geometry of Kiν from the outer Lipschitz geometry of X. The argument is similar to that of the proof of Proposition 4.1. We consider a continuous arc γi inside π1pN pEiqq with the property that dp0, γptqq " Optq. Then for all k sufficiently small, the intersection X X Bpγiptq, ktq consists of some number µi of 4- kptqq " Optqjk q before we change the metric balls D4 by a bilipschitz homeomorphism. The qjk are determined by the outer Lipschitz geometry of pX, 0q and are still determined after a bilipschitz change of the metric by the same argument as the last part of the proof of 4.1. Moreover, they determine the outer Lipschitz geometry of Kiν , as in Section 4. They also determine the number mi of components of Kiν which are in π1pN pEiqq. µiptq with dpD4 1ptq, . . . , D4 j ptq, D4 By the above proof of (2) of Theorem 1.2, the number αi of Kiν 's for fixed i is pEi hq{mi, and hence determined by the outer Lipschitz geometry of X. Since different Kiν's have different tangent lines, the outer geometry of their union is determined, completing the proof. (cid:3) 12. Detecting the decomposition The aim of this section is to prove that the geometric decomposition pX, 0q " i"1pXqi , 0qYŤiąj pAqi,qj , 0q introduced in Section 8 can be recovered up to equiv- Ťν alence (Definition 8.5) using the outer Lipschitz geometry of X. Specifically, we will prove: Proposition 12.1. The outer Lipschitz geometry of pX, 0q determines a decompo- , 0q r Aqi,qj qi ,qj , 0q glued along their boundaries, where each X 1 qi , 0q into semi-algebraic subgerms pX 1 qi sition pX, 0q "Ťν , 0q YŤiąj pA1 and pA1 is a union of collars of type Apqi, qiq. i"1pX 1 qi qi,qj r Xqi and A1 qi,qj So X 1 qi is obtained from Xqi by adding an Apqi, qiq collar on each outer bound- ary component of Xqi and removing one at each inner boundary component, while A1 qi,qj is obtained from Aqi,qj by adding an Apqi, qiq collar on each outer bound- ary component of Aqi,qj and removing an Apqj , qjq collar at each inner boundary component. An important part of the proof of Proposition 12.1 consists in discovering the polar pieces in the germ pX, 0q by exploring them with small balls. Let us first introduce notation and sketch this method. 32 WALTER D NEUMANN AND ANNE PICHON We will use the coordinates and the family of Milnor balls Bǫ, 0 ă ǫ ď ǫ0 introduced in Section 8. We always work inside the Milnor ball Bǫ0 and we reduce ǫ0 when necessary. Let q1 ą q2 ą ą qν " 1 be the series of rates for the geometric decomposition of X. In this section we will assume that ν ą 1 so q1 ą 1. For x P X define Bpx, rq to be the component containing x of the intersection Brpxq X X, where Brpxq is the ball of radius r about x in Cn. Definition 12.2. For a subset B of X, define the abnormality αpBq of B to be maximum ratio of inner to outer distance in X for pairs of distinct points in B. The idea of abnormality was already used by Henry and Parusi´nski to prove the existence of moduli in bilipschitz equivalence of map germs (see [11, Section 2]). Definition 12.3. We say a subset of X r t0u has trivial topology if it is contained in a topological ball which is a subset of X r t0u. Otherwise, we say the subset has essential topology. 12.1. Sketch. Our method is based on the fact that a non-polar piece is "asymp- totically flat" so abnormality of small balls inside it is close to 1, while a polar piece is "asymptotically curved" (Lemmas 12.5 and 12.8). It consists in detect- ing the polar pieces using sets Bpx, axqq as "searchlights". Figure 4 represents schematically a polar piece (in gray). ΠD1 ΠD 0 Figure 4. A polar piece In order to get a first flavour of what will happen later, we will now visualize some pieces and some sets Bpx, axqq by drawing their real slices. We call real slice of a real algebraic set Z Ă Cn the intersection of Z with tz1 " tu X P where z1 is a generic linear form and P a general p2n ´ 1q-plane in Cn -- R2n. We use again the notations of the previous sections: ℓ " pz1, z2q is a generic plane projection for pX, 0q and h " z1X . We first consider a component M Ă Xqi of the geometric decomposition of pX, 0q which is not a polar piece. We assume that qi ą 1 (so i P t1, . . . , ν ´ 1u) and that the sheets of the cover ℓX : X Ñ C2 inside this zone have pairwise contact ą qi, i.e., there exists q1 0 ‰ γ0 in i ą qi such that for an arc γ0 in M , all the arcs γ1 LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 33 ℓ´1pℓpγ0qq have contact at most q1 two of these arcs in M . i with γ0 and the contact q1 i is reached for at least In Figure 5 the dotted circles represent the boundaries of the real slices of balls Bpx, axqq Ă Cn for some x P M . The real slices of the corresponding sets Bpx, axqq Ă X are the thickened arcs. q1 i ă q qi ă q ă q1 i q ă qi x x x Figure 5. Balls through a non-polar piece Next, we consider a polar piece N Ď Xqi. If x P N and a is large enough then Bpx, axqi q either has essential topology or high abnormality (or both). Figure 6 represents the real slice of a ball Bpx, axqi q with x P N when Bpx, axqi q has high abnormality. It includes the real slice of N which is the gray arc. x N Figure 6. The rest of the section is organized as follows. The aim of Subsections 12.2 and 12.3 is to specify the set pX 1 q1 , 0q of Proposition 12.1. This will be the first step of an induction carried out in Subsections 12.4 to 12.6 leading to the construction of subgerms pX 1 , 0q Ă pX, 0q having the properties stated in Propo- qi sition 12.1. Finally, the bilipschitz invariance of the germs pX 1 qi,qj is qi proved in Proposition 12.13, completing the proof of Proposition 12.1. , 0q and pA1 , 0q and A1 qi,qj 12.2. Finding the rate q1. 34 WALTER D NEUMANN AND ANNE PICHON Lemma 12.4. Let ℓ : X Ñ C2 be a generic plane projection, and A Ă X a union of polar wedges about the components of the polar curve for ℓ. If x P X and r ă x are such that Bpx, rq X A " H, then there exists K ą 1 such that ℓ restricted to Bpx, r K 2 q is an inner K-bilipschitz homeomorphism onto its image. Moreover, Bpx, r K 2 q has trivial topology and abnormality ď K. Proof. There exists K such that ℓXrA is locally a K-bilipschitz map (see Section 6). Let Bpy, rq Ă C2 denote the ball of radius r about y P C2. Then Bpℓpxq, r K q Ď ℓpBpx, rqq. Let B be the component of ℓ´1pBpℓpxq, r K qq which is contained in Bpx, rq. So ℓB is a K-bilipschitz homeomorphism of B onto its image Bpℓpxq, r K q. Now Bpx, r K 2 q has trivial topology. Moreover, for each pair of points x1, x2 P Bpx, r K 2 q Ď B, so Bpx, r K 2 q, we have: dinnerpx1, x2q ď KdC2pℓpx1q, ℓpx2qq . On the other hand, dC2pℓpx1q, ℓpx2qq ď douterpx1, x2q, so dinnerpx1, x2q{douterpx1, x2q ď K . Thus αpBpx, r K 2 qq ď K. (cid:3) Lemma 12.5. (1) There exist a ą 0, K1 ą 1 and ǫ0 ą 0 such that for all x P X X Bǫ0 the set Bpx, axq1 q has trivial topology and abnormality at most K1. (2) For all q ą q1 and a ą 0 there exist K1 ą 1 and ǫ1 with ǫ1 ď ǫ0 such that for all x P X X Bǫ1 the set Bpx, axqq has trivial topology and abnormality at most K1. (3) For all q ă q1, K2 ą 1 and a ą 0 there exist ǫ1 with ǫ1 ď ǫ0 such that for all x P Xq1 X Bǫ1 we have either αpBpx, axq qq ą K2 or Bpx, axqq has essential topology. Before proving the lemma we note the following immediate corollary: Proposition 12.6 (Finding q1). q1 is the infimum of all q satisfying: there exists K1 ą 0 such that for all a ą 0 there exists ǫ1 ą 0 such that all sets Bpx, axqq with x P X X Bǫ1 have trivial topology and abnormality at most K1. (cid:3) Proof of Lemma 12.5 (1) and (2). We just prove (1), since (2) follows immediately from (1), using that Bpx, axqq " Bpx, a1xq1 q with a1 " axq´q1 , which can be made as small as one wants by reducing ǫ1. Let ℓD1 : pX, 0q Ñ pC2, 0q and ℓD2 : pX, 0q Ñ pC2, 0q be a generic pair of generic plane projections. For i " 1, 2 let Ai be a union of polar wedges about the compo- nents of the polar curve ΠDi. to equivalence, the component containing Π0 of a set of the formŤxPΠ0 Note that a polar wedge of rate s about a component Π0 of the polar curve is, up Bpx, cxsq, so we may take our polar wedges to be of this form, and we may choose c and ǫ0 small enough that none of the polar wedges in A1 and A2 intersect each other in B2ǫ0 r t0u. Now replace c by c{4 in the construction of the polar wedges. Since q1 ě s for every polar rate s, for any x P pX X Bǫ0q r t0u the set Bpx, cxq1 q is then disjoint from one of A1 and A2. Thus part (1) of the lemma follows from Lemma 12.4, with K1 chosen such that each restriction ℓDi to pX r Aiq X B2ǫ is a local K1-bilipschitz homeomorphism, and with a " c (cid:3) K 2 . 1 LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 35 Proof of Lemma 12.5 (3). Fix q ă q1, K2 ą 1 and a ą 0. We will deal with Xq1 component by component, so we assume for simplicity that Xq1 consists of a single component (see Definition 8.1). Assume first that Xq1 is not a Dpq1q-piece. Then it is a B-piece and there exist ǫ ď ǫ0 and K ą 1 such that Xq1 XBǫ is K-bilipschitz equivalent to a standard model as in Definition 2.2. Let x P Xq1 X Bǫ. If d is the diameter of the fiber F used in constructing the model then Bpx, 2Kdxq1 q contains a shrunk copy of the fiber F . Then, if ǫ is small enough, Bpx, axq q also contains a copy of F , since q ă q1. In [1] it is proved that F contains closed curves which are not null-homotopic in X r t0u, so Bpx, axq q has essential topology. We now assume Xq1 is a Dpq1q-piece. Consider the resolution π : prX, Eq Ñ pX, 0q with dual graph Γ defined in Section 9. We will use again the notations of Section 9. For each irreducible component of E, let N pEiq be a small closed tubular neighbourhood of Ei. Let Γq1 be the subgraph of Γ which consists of the P-node corresponding to Xq1 plus attached bamboos. After adjusting N pEiq if necessary, one can assume that the strict transform π´1pXq1 r t0uq of Xq1 by π is the set N pΓq1 q (see Proposition 9.4). We set E1 "ŤviPΓqi Let pαq : pX r t0uq p0, 8q Ñ r1, 8q be the map which sends px, aq to the abnormality αpBpx, axq qq and let rαq : prX r Eq p0, 8q Ñ r1, 8q be the lifting of pαq by π. The intersection E X π´1pXq1 r t0uq is a compact set inside E1 r E2, so to prove Lemma 12.5 (3), it suffices to prove that for each y P E1 r E2, Ei and E2 " E r E1. lim xÑy xPĂXrE rαqpx, aq " 8 . This follows from the following more general Lemma, which will be used again (cid:3) later. Lemma 12.7. Let N be a polar piece of rate qi i.e., N " πpN pΓ1 qi qq where Γqi is a subgraph of Γ consisting of a P-node plus any attached bamboo. Assume that the Ej and E2 " E r E1. Then for all q with q ă qi and all y P E1 r E2, outer boundary of N is connected. Set E1 "Ťvj PΓqi xPĂXrE rαqpx, aq " 8 . lim xÑy The proof of Lemma 12.7 needs a preparatory Lemma 12.8. Recall that the resolution π factors through the Nash modification ν : qX Ñ X. Let σ : rX Ñ Gp2, nq be the map induced by the projection p2 : qX Ă X Gp2, nq Ñ Gp2, nq. The map σ is well defined on E and according to [10, Section 2] (see also [23, Part III, Theorem 1.2]), its restriction to E is constant on any connected component of the complement of P-curves in E. The following lemma about limits of tangent planes follows from this: Lemma 12.8. (1) Let Γ1 be a maximal connected component of the graph Γ with its P-nodes removed (we call this a P-Tjurina component of Γ). There exists P Γ1 P Gp2, nq such that limtÑ0 TγptqX " P Γ1 for any real analytic arc γ : pr0, ǫs, 0q Ñ pπpN pΓ1qq, 0q with γptq " Optq and whose strict transform meetsŤvPΓ1 Ev. (2) Let Ek Ă E be a P-curve and x P Ek be a smooth point of the exceptional divisor E. There exists a plane Px P Gp2, nq such that limtÑ0 TγptqX " Px for any 36 WALTER D NEUMANN AND ANNE PICHON real analytic arc γ : pr0, ǫs, 0q Ñ pπpN pEkqq, 0q with γptq " Optq and whose strict transform meets E at x. (cid:3) Proof of Lemma 12.7. Let y P E1 r E2 and let γ : r0, ǫs Ñ X be a real analytic γptq " Optq. We then have to prove that arc inside N X Bǫ whose strict transform rγ by π meets E at y and such that lim tÑ0rαqprγptq, aq " 8 . Let B be the component of the unamalgamated intermediate carrousel such that the outer boundary of a component B1 of ℓ´1pBq is the outer boundary of N . Let N 1 Ď N be B1 with its polar wedges amalgamated, so that ℓN 1 : N 1 Ñ ℓpN 1q is a branched cover of degree ą 1. Then ℓpN q has an Apq2, qiq annulus outside it for some q2, which we will simply call Apq2, qiq. The lift of Apq2, qiq by ℓ is a covering space. Denote by Apq2, qiq the component of this lift that intersects N ; it is connected since the outer boundary of N is connected, and the degree of the covering is ą 1 since its inner boundary is the outer boundary of N 1. Apq2, qiq is contained in a N pΓ1q where Γ1 is a P-Tjurina component of the resolution graph Γ so we can apply part (1) of Lemma 12.8 to any suitable arc inside it. This will be the key argument later in the proof. We will prove the lemma for q1 with q2 ă q1 ă qi since it is then certainly true for smaller q1. Choose p1 with q1 ă p1 ă qi and consider the arc γ0 : r0, ǫs Ñ C2 defined by γ0 " ℓ γ and the function γsptq :" γ0ptq ` p0, stp1 q for ps, tq P r0, 1s r0, ǫs . We can think of this as a family, parametrized by s, of arcs t ÞÑ γsptq, or as a family, parametrized by t, of real curves s ÞÑ γsptq. For t sufficiently small γ1ptq lies in ℓpBpγptq, a1γptqq1 qq and also lies in the Apq2, qiq mentioned above. Note that for any s the point γsptq is distance Optq from the origin. We now take two different continuous lifts γp1q s ptq and γp2q of arcs γsptq, for 0 ď s ď 1, with γp1q covering degree of Apq2, qiq Ñ Apq2, qiq is ą 1). Since q ă p1, γp1q in Bpγptq, a1γptqqq for each t, s. 0 " γ and γp2q 0 s ptq by ℓ of the family also in N (possible since the s ptq lie s ptq and γp2q 1 and P2 " γp2q To make notation simpler we set P1 " γp1q 1 . Since the points P1ptq and P2ptq are on different sheets of the covering of Apq2, qiq, a shortest path between them will have to travel through N , so its length linnptq satisfies linnptq " Optp1 q. We now give a rough estimate of the outer distance loutptq " P1ptq ´ P2ptq which will be sufficient to show limtÑ0plinnptq{loutptqq " 8, completing the proof. For this, we choose p2 with p1 ă p2 ă qi and consider the arc p : r0, ǫs Ñ C2 defined by: pptq :" γst ptq " γ0ptq ` p0, tp2 q with st :" tp2´p1 , and its two liftings p1ptq :" γp1q st ptq, belonging to the same sheets of the cover ℓ as the arcs P1 and P2. A real slice of the situation is represented in Figure 7. st ptq and p2ptq :" γp2q The points p1ptq and p2ptq are inner distance Optp2 q apart by the same argument as before, so their outer distance is at most Optp2 q. By Lemma 12.8 the line from pptq to γ1ptq lifts to almost straight lines which are almost parallel, from p1ptq to LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 37 P1ptq p1ptq loutptq P2ptq p2ptq Figure 7. γp1q 0 ptq N Πℓ γp2q 0 ptq P1ptq and from p2ptq to P2ptq respectively, with degree of parallelism increasing as t Ñ 0. Thus as we move from the pair piptq, i " 1, 2 to the pair Piptq the distance changes by f ptqptp1 q where f ptq Ñ 0 as t Ñ 0. Thus the outer distance loutptq between the pair is at most Optp2 q. Dividing by linnptq " Optp1 q gives loutptq{linnptq " Opf ptqq, so limtÑ0ploutptq{linnptqq " 0. Thus limtÑ0plinnptq{loutptqq " 8, completing the proof of Lemma 12.7. (cid:3) q ` f ptqptp1 ´ tp2 ´ tp2 12.3. Constructing X 1 q1. Definition 12.9 (q-neighbourhood). Let pU, 0q Ă ppX, 0q Ă pX, 0q be semi- algebraic sub-germs. A q-neighbourhood of pU, 0q in ppX, 0q is a germ pN, 0q Ě pU, 0q with N Ď tx P pX dpx, U q ď Kdpx, 0qqu for some K, using inner metric in ppX, 0q. For a ą 0 and L ą 1, define: Zq1,a,L :"ď Bpx, axq1 q : αpBpx, axq1 qq ą L or Bpx, axq1 q has essential topology( Lemma 12.10. For sufficiently large a and L and sufficiently small ǫ1 ď ǫ0, the set Zq1,a,L X Bǫ1 is a q1-neighbourhood of Xq1 X Bǫ1 in X X Bǫ1 . Proof. As before, to simplify notation we assume that Xq1 r t0u is connected, since otherwise we can argue component by component. The following three steps prove the lemma. (1) If Xq1 is not a D-piece there exists a1 sufficiently large and ǫ sufficiently small that for each a ě a1 and x P Xq1 XBǫ the set Bpx, axq1 q has essential topology. (2) If Xq1 is a D-piece then for any L ą 1 there exists a1 sufficiently large and ǫ sufficiently small that for each a ě a1 and x P Xq1 X Bǫ the set Bpx, axq1 q has abnormality ą L. (3) There exists a K ą 0 such that for all a ą 0 there exists a q1-neighbourhood Zq1 paq of Xq1 and ǫ ą 0 such that any set Bpx, axq1 q not intersecting Zq1 paq has trivial topology and abnormality ď K. Item (1) is the same proof as the first part of the proof of Lemma 12.5 (3). For Item (2) we use again the resolution π : prX, Eq Ñ pX, 0q with dual graph Γ defined in section 9 and the notations introduced in the proof of Lemma 12.5 (3). Denote by Eq1 the union of components of E corresponding to vertices of Γ which are nodes of rate q1 and attached bamboos. We claim that for each y P Eq1 r E1 38 WALTER D NEUMANN AND ANNE PICHON we have: lim aÑ8 limxÑy xRE rαq1 px, aq " 8 . Indeed, consider a real analytic arc γ in X such that γptq " Optq andrγp0q " y. Choose q2 ă q ă q1. For each t, set aptq " γptqq´q1 , so we have limtÑ0 aptq " 8. Since q2 ă q ă q1, Bpγptq, γptqqq has trivial topology for all t. By Item (3) of Lemma 12.5 we have that for any K2 ą 1 and a " 1, there is ǫ ă ǫ0 such that for each t ď ǫ, αpBpγptq, γptqqqq ě K2. As Bpγptq, γptqqq " Bpγptq, aptqγptqq1 q, we then obtain lim tÑ0rαq1 prγptq, aptqq " lim tÑ0 αpBpγptq, aptqγptqq1 qq " 8 . Choose L ą 1. The set N pΓqi intersects E in a compact set inside Eq1 r pEq1 X L. This completes the proof of Item (2) E1q. Therefore, by continuity of rαq1 , there exists a1 ą 0 and ǫ ą 0 such that for all a ě a1 and x P pN pΓq1 qrEqXπ´1pBǫq we haverαq1 px, aq " αpBpπpxq, aπpxqq1 qq ą To prove Item (3) we pick ℓ1, ℓ2, K and c as in the proof of Lemma 12.5 (1). So with A1 and A2 as in that proof, any Bpx, cxq1 q is disjoint from one of A1 and A2. We choose ǫ ą 0 which we may decrease later. 1 Y A1 1 and A1 We now choose a ą 0. Denote by A1 2 the union of components of A1 and A2 with rate ă q1. Note that the distance between any pair of different components of A1 2 is at least Oprq2 q at distance r from the origin, so after decreasing ǫ if necessary, any Bpx, aK 2xq1 q which intersects one of A1 2 is disjoint from the other. Thus if Bpx, aK 2xq1 q is also disjoint from Xq1 , then the argument of the proof of Lemma 12.5 (1) shows that Bpx, axq1 q has trivial topology and abnormality at most K. Since the union of those Bpx, aK 2xq1 q which are not disjoint from Xq1 is a q1-neighbourhood of Xq1, Item (3) is proved. (cid:3) 1 and A1 The germ pX 1 q1 , 0q of Proposition 12.1 will be defined as a smoothing of Zq1,a,L for L and a sufficiently large. Let us first define what we mean by smoothing. The outer boundary of Xq1 attaches to Apq1, q1q-pieces of the (non-amalgamated) carrousel decomposition with q1 ă q1, so we can add Apq1, q1q collars to the outer boundary of Xq1 to obtain a q1-neighbourhood of Xq1 in X. We use X ` q1 to denote such an enlarged version of Xq1 . An arbitrary q1-neighbourhood of Xq1 in X can be embedded in one of the form X ` q1, and we call the process of replacing such a neighbourhood by X ` q1 a smoothing of Xq1 . This smoothing process can be applied more generally as follows: Definition 12.11 (Smoothing). Let pY, 0q be a subgerm of pX, 0q which is a union of B- and A-pieces lifted from a plane projection of X. So pY, 0q has (possibly empty) inner and outer boundaries. Assume that the outer boundary components of Y are inner boundary components of Apq1, qq-pieces with q1 ă q. Let W be a q-neighbourhood of Y in pX, 0q and let V be the union of the com- ponents of W r Y which are inside Apq1, qq-pieces attached to the outer boundaries of Y . We call the union Z " V Y Y an outer q-neighbourhood of Y . We call a smoothing of Z any outer q-neighbourhood Z ` of Z obtained by adding to Y some Apq, qq-pieces to its outer boundaries. Remark 12.12. The smoothing Z ` of Z is uniquely determined from Z up to homeomorphism since a q-neighbourhood of the form Z ` is characterized among all outer q-neighbourhoods of Z by the fact that the links of its outer boundary LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 39 components are tori and the number of outer boundary components is minimal among outer q-neighbourhoods of Z in X. Definition of the germ pX 1 Lemma 12.10. We define X 1 q1 q1 as the smoothing , 0q. Let a ą 0 and L ą 1 sufficiently large as in X 1 q1 :" Z ` q1,a,L . Since Zq1,a,L is a q1 neighbourhood of Xq1 inside a Milnor ball Bǫ, the germ pX 1 is equivalent to pXq1 , 0q. q1 , 0q 12.4. Discovering q2. Using the results of Section 8 and the beginning of Section 12, we discover q2 in two steps. Step 1. We consider Nq,1pX 1 to the boundary of X 1 of q for which the topology and the number of proximity classes of Nq,1pX 1 not changed. Then for α ą 0 sufficiently small, the topology Nq1 not changed. q1 q` with q ď q1. If q is close to q1 this just adds collars 1 be the infimum q1 q` has q1 q` has also q1 so it does not change the topology. Let q1 1,αpX 1 Step 2. Using essentially the same argument as we used to discover q1(Proposition 12.6) we discover q2 as follows. q2 is the infimum of all q with q1 1 ă q ă q1 such that there exists K1 ą 0 such that for all a ą 0 there exists ǫ1 ą 0 such that all q1 q` have trivial topology and abnormality sets Bpx, axqq with x P X r Nq1 at most K1. If there are no such q we set q2 " q1 1. 1,αpX 1 q2 and A1 q2 Y q2 ) is the union of components which have nonempty (resp. q2 decomposes as the union X 1 q2,q1. X 1 q2 " X 2 q2 where X 2 12.5. Constructing X 1 q2 (resp. X 3 X 3 empty) inner boundary. We first discover X 3 ing now inside X r Nq,apX 1 Subsection 12.4). Namely, choose q slightly smaller than q2 and set: q2 . We use again the procedure of Subsection 12.3, work- q1 q with q slightly smaller than q2 (see also Step 2 of Zq2,a,L :"ď Bpx, axq2 q : x P X r Nq,apX 1 q1 q and αpBpx, axq2 qq ą L or Bpx, axq2 q has essential topology( q2,a,L is a q2-neighbourhood of the q2 :" Z ` with a and L sufficiently large. Then X 3 union of components of Xq2 having empty inner boundary. 1 then X 2 1, then X 2 q2 . If q2 ą q1 We now describe X 2 q1 q` r Nq2,αpX 1 q2 is the union of the components A q2 X Sǫq " Opǫqq and q ě q2. If q2 has some extra components which are described as follows. There q1 and there q1 q` is no longer that of Nq2,βpX 1 q2 " q1 exist α sufficiently small that Nq2,αpX 1 exists β ą 0 such that for all a ě β the topology of Nq2,apX 1 of X 1 q1 and increasing a does not change the topology. q1 q` has the same topology as X 1 q1 q` with dpA X Sǫ, X 3 q1 q` r X 1 Note that Nq2,βpX 1 q1 consists of at least one piece which is not an Apq2, q1q-piece and maybe some Apq2, q1q-pieces. Then the extra components of X 2 q1 q` which are not Apq2, q2q- pieces or such that the cardinality of their proximity class is strictly greater than that of the corresponding components of Nq1,1pX 1 q2 are the components of Nq2,βpX 1 q1 q` with q2 ă q1 ă q1. q1 q` r Nq2,αpX 1 Now Nq2,αpX 1 q1 q` r X 1 q1 consists of Apq2, q1q-pieces. We define A1 q2,q1 to be the union of these Apq2, q1q-pieces which intersect Nq2,αpX 1 q1 q`. 40 WALTER D NEUMANN AND ANNE PICHON 12.6. Constructing X 1 done the construction for smaller i. Let X pi´1q be the union of all X 1 and all Apqj , qkq-pieces connecting an X 1 for i ą 2. We now assume we have already qj with j ď i´1 qk with k ă j ď i ´ 1. qj with X 1 and A1 qi ,qj qi We use the same arguments as in the construction for i " 2, using X pi´1q in place of X 1 q1 in the discovery of qi and construction of the sets X 1 qi and A1 qi,qj . This iterative procedure ends when steps 1 and 2 of Subsection 12.4 adapted to qν " X 1 the discovery of qi leads to qi " 1, in which case i " ν. We then define X 1 1 as the closure of the complement of the result of gluing a piece of type Ap1, qjq on each outer boundary of X pν´1q. This completes the construction from the outer metric of the decomposition of pX, 0q in Proposition 12.1. Finally, we show that the decomposition can still be recovered after a bilipschitz change to the metric. Proposition 12.13. Let pX, 0q Ă pCn, 0q and pX 1, 0q Ă pCn1 , 0q be two germs of normal complex surfaces endowed with their outer metrics. Assume that there is a bilipschitz map Φ : pX, 0q Ñ pX 1, 0q. Then the inductive process described in Subsections 12.2 to 12.6 leads to the same sequence of rates qi for both pX, 0q and pX 1, 0q and for a ą 0 and L ą 1 sufficiently large, the corresponding sequences of subgerms Zqi,a,L in X and Z 1 qi,a,Lq and Z 1` qi,a,L in X 1 have the property that ΦpZ ` qi,a,L are equivalent. Proof. Let K be the bilipschitz constant of Φ in a fixed neighbourhood V of the origin. The proof follows from the following three observations. (1). Given x P V , a ą 0 and q ě 1, there exists a1 ą 0 such that ΦpBpx, axq qq Ă BpΦpxq, a1Φpxqqq . Indeed, ΦpBpx, axq qq Ă BpΦpxq, aKxq q Ă BpΦpxq, aK q`1Φpxqqq. So any a1 ě aK q`1 works. (2). Let N be a subset of V . Then the abnormality of ΦpN q is controlled by that of N : 1 K 2 αpN q ď αpΦpN qq ď K 2αpN q . (3). Let pU, 0q Ă ppX, 0q Ă pX, 0q be semi-algebraic sub-germs. If pN, 0q is a q- neighbourhood of pU, 0q in ppX, 0q then pΦpN q, 0q is a q-neighbourhood of pΦpU q, 0q in pΦppXq, 0q. 13. Explicit computation of polar rates (cid:3) The argument of the proof of Lemma 12.7 enables one to compute the rate of a polar piece in simple examples such as the following. Assume pX, 0q is a hypersurface with equation z2 " f px, yq where f is reduced. The projection ℓ " px, yq is generic and its discriminant curve ∆ has equation f px, yq " 0. Consider a branch ∆0 of ∆ which lifts to a polar piece N in X. We consider the Puiseux expansion of ∆0: y "ÿiě1 aixqi P Ctx1{mu . The polar rate s of N is the minimal r P 1 γ: y "řiě1 aixqi ` λxr is in ℓpN q. In order to compute s, we set r1 " rm and we m N such that for any small λ, the curve LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 41 parametrize γ as: x " wm, y " ÿiě1 aiwmqi ` wr1 . Replacing in the equation of X, and approximating by elimination of the monomials with higher order in w, we obtain z2 „ awr1`b1 for some a ‰ 0 and some positive integer b1. We then have an outer distance of Opx 2 q between the two sheets of the cover ℓ, where b " b1 m . So the optimal s such that the curve γ is in ℓpN q is given by s`b 2 " s, i.e., s " b. r`b Example 13.1. We apply this to the singularity D5 with equation z2 " ´px2y`y4q (see Example 9.5). The discriminant curve of ℓ " px, yq has equation ypx2 `y3q " 0. For the polar piece πpN pE6qq, which projects on a neighbourhood of the cusp δ2 " tx2 ` y3 " 0u, we use y " w2, x " iw3 ` wr1 , so b1 " 5 and this polar piece has rate s " 5{2. Similarly one computes that the polar piece πpN pE4qq, which projects on a neighbourhood of δ1 " ty " 0u, has rate 2. , so z2 „ ´2iw5`r1 Notice that in the above computation, we have recovered the relation qprq " r`s 2 established for r ě s in the proof of Lemma 6.6. In fact, combining Lemma 6.6 and Lemma 12.7, we can compute polar rates in a more general setting, as we explain now, using e.g., Maple. Let Y be a polar piece in pX, 0q which is a Dpsq-piece. Let q be the rate of the B- piece outside it, so there is an intermediate Apq, sq-piece A between them. Assume we know the rate q. Let Π0 be a component of Π inside Y and let ∆0 " ℓpΠ0q. Consider a Puiseux expansion of ∆0 as before and an irreducible curve γ with Puiseux expansion y "ÿiě1 aixqi ` axr . having contact r ą q with ∆0. Let Lγ " ℓ´1pγq and let ℓ1 : X Ñ C2 be a generic plane projection of pX, 0q which is also generic for Lγ. Let L1 γ be the intersection of Lγ with Y Y A. Let qprq be the greatest characteristic exponent of the curve ℓ1pL1 γq. Given r one can compute qprq from the equations of X. If q ă r ă s, then, according to Lemma 12.7, qprq ą r. If r ě s, then, according to Lemma 6.6, qprq ď s. Therefore, testing the inequality qprq ą r, one can choose an r0 big enough so that qpr0q ď r0, i.e., r0 ě s. Now, we have qpr0q " r0`s (again by Lemma 6.6). Therefore s " 2qpr0q ` r0. 2 Example 13.2. This method gives rise to the polar rate s " 7 7.3, using a Maple computation. 5 claimed in Example 14. Carrousel decomposition from Lipschitz geometry The aim of this section is to complete the proof of Theorem 1.2. We will first prove that the outer Lipschitz geometry determines the sections of the complete carrousel decomposition of the discriminant curve ∆ of a generic plane projection of pX, 0q. Let ℓ " pz1, z2q : X Ñ C2 be the generic plane projection and h " z1X the generic linear form chosen in Sections 8 and 11. We denote by F ptq :" h´1ptq the Milnor fibre of h for t P p0, ǫs. We will continue to use the terminology Xqi and Aqi,qj for the pieces of the geometric decomposition as constructed from the 42 WALTER D NEUMANN AND ANNE PICHON carrousel in Section 8, and X 1 8.5) as recovered using outer geometry (Section 12). qi and A1 qi,qj for pieces equivalent to them (Definition Proposition 14.1. The outer Lipschitz geometry of pX, 0q determines: (1) the combinatorics of the complete carrousel section of ∆. (2) the number of components of the polar curve Π in each B- or D-piece of pX, 0q. Proof of Proposition 14.1. By Proposition 12.1 we may assume that we have re- covered the pieces Xqi of X for i " 1, . . . , ν and the intermediate A-pieces up to equivalence (Definition 8.5). We will recover the combinatorics of the carrousel sections by an inductive pro- cedure on the rates qi starting with q1. We define AXqi to be AXqi :" Xqi Yďkăi Aqi,qk . Note that AXq1 " Xq1 . We first recover the pieces of ℓpAXq1 X F pǫqq and in- side them, the section of the complete carrousel of ∆ beyond rate q1 (Definition 4.4). Then we glue to some of their outer boundaries the pieces corresponding to ℓppAXq1 Y AXq2 q X F pǫqq r ℓpAXq1 X F pǫqq, and inside them, we determine the com- plete carrousel beyond q2, and so on. At each step, the outer Lipschitz geometry will determine the shape of the new pieces we have to glue, how they are glued, and inside them, the complete carrousel section beyond the corresponding rate. For any i ă ν we will denote by AXqi` the result of adding Apq1, qiq-pieces to all components of the outer boundary BoAXqi of AXqi for some q1 with qi`1 ă q1 ă qi. We can assume that the added A-pieces are chosen so that the map ℓ maps each by a branched covering map to an Apq1, qiq-piece in C2. We will denote by BoAXqi` the outer boundary of AXqi`, so it is a horn-shaped cone on a family of tori with rate q1. We denote Fqi ptq :" F ptq X AXqi , Fqi`ptq :" F ptq X AXqi` , and their outer boundaries by BoFqi ptq " F ptq X BoAXqi , BoFqi`ptq " F ptq X BoAXqi` . A straightforward consequence of the proof of Theorem 1.2(1) is that the outer Lipschitz geometry determines the isotopy class of the Milnor fibre F ptq and how the Milnor fiber lies in a neighbourhood of the exceptional divisor of the resolution described there. In particular it determines the intersection Fqi ptq " F ptq X AXqi for i ă ν and t P p0, ǫs up to small deformation caused by any bilipschitz change of metric. Set Bq1 " ℓpAXq1 q and Bq1 pǫq :" ℓpFq1 pǫqq, so Bq1 pǫq consists of the innermost disks of the intermediate carrousel section of ∆. Notice that Bq1 pǫq is the union of all the sections of the complete carrousel beyond rate q1. Consider the restriction ℓ : Fq1 pǫq Ñ Bq1 pǫq. We want to show that this branched cover can be seen in terms of the outer Lipschitz geometry of a neighbourhood of AXq1 . Since Bq1 pǫq is a union of disks, the same is true for Bq1`pǫq :" ℓpFq1`pǫqq. Consider a component Fq1,0ptq of Fq1 ptq and let AXq1,0 be the component of AXq1 such that Fq1,0ptq Ď F ptq X AXq1,0. Consider a continuous arc p : r0, ǫs Ñ BoAXq1,0 such that pptq P BoFq1,0ptq for t P p0, ǫs. By the argument of Lemma 10.6, if δ is big LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 43 enough then for all t small enough a component of Fq1 ptq maps to the component Bq1,0ptq :" ℓpFq1,0ptqq of Bq1 ptq if and only if it intersects the ball Bppptq, δtq1 q. If there is a component of Fq1 ptq which does not map on Bq1,0ptq, we iterate the process until we have determined the number of discs in Bq1 ptq and how ℓ maps the components of Fq1 ptq onto them. Using similar arcs on the outer boundaries of the components of AXq1` r AXq1 , we obtain by the same argument that the outer Lipschitz geometry determines the degree of the restriction of ℓ on each annular component of Fq1`ptq r Fq1 ptq, and then, the degree of ℓ restricted to each component of BoFq1`ptq, resp. BoFq1 ptq (alternatively, this follows from the proof of Lemma 12.7). We hence obtain the degree n of ℓFq1 ,0ptq : Fq1,0ptq Ñ Bq1,0ptq. We wish to determine the number β0 of branch points of this map. Since the cover is general, the inverse image by ℓFq1 ,0ptq of a branch point consists of n ´ 1 points in Fq1,0ptq and the Hurwitz formula applied to ℓFq1,0ptq gives: β0 " n ´ χpFq1,0ptqq . Denote by β the product of β0 by the number of components of AXq1,0 X F ptq. We now determine the number of components of the polar inside the component AXq1,0 of AXq1 which contains Fq1 ,0ptq. We use a similar argument as in the proof of Proposition 4.1, taking account of Remark 4.3. We denote by Dptq the connected component containing pptq of the intersection of AXq1 with a ball Bppptq, ηtq, η ăă 1. There is a µ ą 0 such that for all q ą q1 sufficiently close to q1 the ball Bppptq, tqq intersects ℓ´1pℓpDptqqq in µ connected components D1ptq, . . . , Dµptq for all small t. For each j, k with 1 ď j ă k ď µ, let qjk be defined by dpDj ptq, Dkptqq " Optqjk q. Let A be the set of indices j P t1, . . . , µu with Djptq Ă AXq1`,0. The outermost piece of AXq1,0 before amalgamation is a union of equisingular curves ΠD X AXq1,0 with D generic and each curve ΠD X AXq1,0 consists of equisingular components having pairwise contact q1 (see Section 6). Therefore the collection of rates qj,k indexed by j, k P A reflects the outer Lipschitz geometry beyond rate q (Definition 4.4) of each irreducible component ΠD,0 of ΠD XAXq1,0. In particular, it determines the number r of points in the intersection ΠD,0 X F pǫq. Then Π X AXq1,0 consists of β r equisingular irreducible curves with pairwise contact q1 and such that each component has the outer Lipschitz geometry of ΠD,0. As in the proof of Proposition 4.1, the components we use to determine the numbers qjk may disintegrate into several pieces with a bilipschitz change of metric, but this can be dealt with as in that proof. To complete the initial step of our induction, we will now recover the section of the complete carrousel of ∆ beyond rate q1, i.e., inside each component of Bq1 pǫq. As we have just seen, the collection of rates qj,k indexed by j, k P A determines the number of components of the polar inside each component AXq1,0 of AXq1 and their outer geometry. Since the projection ℓ " ℓD is a generic projection for its polar curve Π " ΠD [26, Lemme 1.2.2 ii)], we have then showed that the outer Lipschitz geometry recovers the outer Lipschitz geometry of ℓpΠq beyond rate q1 inside Bq1,0, or equivalently, the section of the complete carrousel of the discriminant curve ∆ " ℓpΠq inside Bq1,0pǫq. Doing this for each component Bq1,0pǫq of Bq1 pǫq we then reconstruct the com- plete carrousel section of ∆ beyond rate q1. 44 WALTER D NEUMANN AND ANNE PICHON This completes the initial step of our induction. Overlapping. In the next steps i ě 2, we give special attention to the following phenomenon. It can happen that ℓpAXqi q contains connected components of some ℓpAXqj q for j ă i. We say that AXqi overlaps AXqj . This occurs when some components of Xqi are obtained by amalgamation of D-pieces to Bpqiq-pieces lifted from Bqi -pieces of the carrousel. This phenomenon is illustrated further in Example 14.3 where the thick part AX1 overlaps the components of a thin part. We now consider Fq2 pǫq. As before, we know from the outer Lipschitz metric for Fq2`pǫq how the outer boundary BoFq2 pǫq covers its image. We will focus first on a single component Fq2,0pǫq of Fq2 pǫq. The map ℓFq2 ,0pǫq : Fq2,0pǫq Ñ ℓpFq2 ,0pǫqq is a covering map in a neighbourhood of its outer boundary, whose degree, m say, is determined from the outer Lipschitz geometry. The image BoℓpFq2,0pǫqq of the outer boundary is a circle which bounds a disk Bq2,0pǫq in the plane tz1 " ǫu. The image ℓpFq2,0pǫqq is this Bq2,0pǫq, possibly with some smaller disks removed, depending on whether and how AXq2 overlaps AXq1 . The image of the inner boundaries of Fq2,0pǫq will consist of disjoint circles inside Bq2,0pǫq of size proportional to ǫq1. Consider the components of Fq1 pǫq which fit into the inner boundary components of Fq2,0pǫq. Their images form a collection of disjoint disks Bq1,1pǫq, . . . , Bq1,spǫq. For each j denote by Fq1,jpǫq the union of the components of Fq1 pǫq which meet Fq2,0pǫq and map to Bq1,jpǫq by ℓ. By the first step of the induction we know the degree mj of the map BoFq1,jpǫq Ñ BoBq1,jpǫq. This degree may be less than m, in which case ℓ´1pBq1,jpǫqq X Fq2,0pǫq must consist of m ´ mj disks. Thus, after j"1 Bq1,jpǫq by a branched covering. Moreover, the j"1 Bq1,jpǫq are the intersection points of Fq2,0pǫq with the polar Π. We again apply the Hurwitz formula it is mp1 ´ sq ´ χp Fq2 ,0pǫqq " to discover the number of these branch points: removing řjpm ´ mjq disks from Fq2 ,0pǫq, we have a subset Fq2 ,0pǫq of Fq2,0pǫq which maps to Bq2,0pǫq rŤs branch points of the branched cover Fq2,0pǫq Ñ Bq2,0pǫq rŤs mp1 ´ sq ´ χpFq2 ,0pǫqq `řs j"1pm ´ mjq. Then we use balls along arcs as before to find the number of branches of the polar in each component of AXq2 , and since outer Lipschitz geometry tells us which components of Fq2 pǫq lie over which components of ℓpFq2 pǫqq, we recover how many components of the discriminant meet each component of ℓpFq2 pǫqq. Different components Fq2,0pǫq of Fq2 pǫq may correspond to the same Bq2,0pǫq, but as already described, this is detected using outer Lipschitz geometry (Lemma 10.6). We write Bq2 pǫq for the union of the disks Bq2,0pǫq as we run through the components of Fq2 . The resulting embedding of Bq1 in Bq2 is the next approxima- tion to the carrousel decomposition. Moreover, by the same procedure as in the initial step, we determine the complete carrousel section of ∆ beyond rate q2, i.e., within Bq2 " ℓpAXq1 Y AXq2 q. Iterating this procedure for Fqi pǫq with 2 ă i ă ν, we build up for i " 1, . . . , ν ´1 the picture of the complete carrousel section for ∆ beyond rate qi while finding the number of components of the polar in each component of AXqi and therefore the number of components of the discriminant in the pieces of the carrousel decompo- sition. Finally for a component AX1,0 of AXqν " AX1 the degree of the map to ℓpAX1,0q can be discovered by the same ball procedure as before; alternatively it is the multiplicity of the maximal ideal cycle at the corresponding L-node, which was LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 45 determined in Item (2) of Theorem 1.2. The same Euler characteristic calculation as before gives the number of components of the polar in each piece, and the ball procedure as before determines the complete carrousel section of ∆ beyond rate qν " 1, i.e., the complete carrousel section. (cid:3) Completion of proof of Theorem 1.2. Parts (1) to (3) of Theorem 1.2 were proved in Section 11 and part (5) is proved above (Proposition 14.1). (5)ñ(6) is straight- forward: By Pham-Teissier [21] (see also Fernandes [9], Neumann-Pichon [19]), the outer Lipschitz geometry of a complex curve in Cn is determined by the topology of a generic plane projection of this curve. By Teissier [26, page 462 Lemme 1.2.2 ii)], if one takes a generic plane projection ℓ : pX, 0q Ñ pC2, 0q, then this projection is generic for its polar curve. Therefore the topology of the discriminant curve determines the outer Lipschitz geometry of the polar curve. It remains to prove part (4). We consider the decomposition of the link X pǫq as the union of the Seifert fibered qi and show first that we know the Seifert fibration for each X pǫq manifolds X pǫq qi . Indeed, the Seifert fibration for a component of a X pǫq is unique up to isotopy unless qi that component is the link of a D- or A-piece. Moreover, on any torus connecting two pieces X pǫq qj we know the relative slopes of the Seifert fibrations, since this is given by the rates qi and qj. So if there are pieces of the decomposition into Seifert fibered pieces where the Seifert fibration is unique up to isotopy, the Seifert fibration is determined for all pieces. This fails only if the link X pǫq is a lens space or torus bundle over S1, in which case pX, 0q is a cyclic quotient singularity or a cusp singularity. These are taut, so the theorem is trivial for them. qi and X pǫq We have shown that the outer Lipschitz geometry determines the decomposition of pX, 0q as the union of subgerms pXqi , 0q and intermediate A-pieces as well as the number of components of the polar curve in each piece. The link L of the polar curve is a union of Seifert fibers in the links of some of the pieces, so we know the topology of the polar curve. By taking two parallel copies of L (we will call it 2L), as in the proof of (1) of Theorem 1.2, we fix the Seifert fibrations and can therefore recover the dual graph Γ we are seeking as the minimal negative definite plumbing diagram for the pair pX pǫq, 2Lq. Since we know the number of components of the polar curve in each piece, the proof is complete. (cid:3) Example 14.2. Let us show how the procedures described in Section 12 and in proof of (1) of Proposition 14.1 work for the singularity D5, i.e., recover the decom- position X " X5{2 Y X2 Y X3{2 Y X1 plus intermediate A-pieces, and the complete carrousel sections of the discriminant curve ∆. It starts by considering high q, and then, by decreasing gradually q, one reconstructs the sequence Xq1 , . . . , Xqν up to equivalence as described in Proposition 12.1, using the procedure of Section 12. At each step we will do the following: (1) On the left in the figures below we draw the dual resolution graph with black vertices corresponding to the piece Xqi just discovered, while the vertices corresponding to the previous steps are in gray, and the remaining ones in white. We weight each vertex by the corresponding rate qi. (2) We describe the cover ℓ : Fqi Ñ Bqi and Bqi and on the right we draw the new pieces of the carrousel section. To simplify the figures (but not the data), we will rather draw the sections of the AXqi -pieces. The ∆-pieces are 46 WALTER D NEUMANN AND ANNE PICHON in gray. We add arrows on the graph corresponding to the strict transform of the corresponding branches of the polar. We now carry out these steps. We use the indexing v1, . . . , v6 of the vertices of Γ introduced in Example 9.5. ‚ The characterization of Proposition 12.6 leads to q1 " 5{2. We get X5{2 " πpN pE6qq up to collars. Using the multiplicities of the generic linear form (determined by (2) of Theorem 1.2) we obtain that F5{2 " F X πpN pE6qq consists of two discs. Then B5{2 " ℓpF5{2q also consists of two discs. By the Hurwitz formula, each of them contains one branch point of ℓ. 5{2 5{2 5{2 ‚ q2 " 2 is the next rate appearing, and X2 " πpN pE4qq. F2 is a disk as well as B2. Then it creates another carrousel section disk. There is one branching point inside B2. 5{2 2 5{2 5{2 2 ‚ Next is q3 " 3{2. F3{2 is a sphere with 4 holes. Two of the boundary circles are common boundary with F1, the two others map on the same circle BB5{2. A simple observation of the multiplicities of the generic linear form shows that there X3{2 does not overlap X5{2 so B3{2 " ℓpF3{2q is connected with one outer boundary component and inner boundary glued to B5{2. 3{2 5{2 3{2 3{2 3{2 2 5{2 5{2 ‚ Last is q4 " 1, which corresponds to the thick piece. 3{2 5{2 3{2 1 2 2 3{2 3{2 1 2 5{2 We read in the carrousel section the topology of the discriminant curve ∆ from this carrousel sections: one smooth branch and one transversal cusp with Puiseux exponent 3/2. Example 14.3. We now describe how one would reconstruct the decomposition X " Ť Xqi plus intermediate A-pieces and the carrousel section from the outer Lipschitz geometry for Example 7.3 (see also Example 9.6). We will again simplify LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 47 the figures by drawing only the sections of AXqi -pieces. The sequence of rates qi appears in the following order: q1 " 7{5, q2 " 3{2, q3 " 5{4, q4 " 6{5, q5 " 1 . Overlapping. The inverse image by ℓ of the carrousel section of ℓpX5{4q consists of two annuli (according to the multiplicities of the generic linear form) and each of them double-covers the disk image. Then the restriction of ℓ to X5{4 has degree 4 while the total degree of the cover ℓX is 6. Therefore either X6{5 or X1 overlaps X5{4. Moreover, the restriction of ℓ to X6{5 also has degree 4 according to the multiplicity 10 at the corresponding vertices. Therefore X1 overlaps X5{4. We now show step by step the construction of the carrousel section. Here again we represent the sections of the AXqi -pieces rather than that of the Xqi and inter- mediate annular pieces. ‚ q1 " 7{5. X7{5 consists of two polar pieces. We represent the two ∆-pieces with two different gray colors in the carrousel section. ‚ q2 " 3{2 ‚ q3 " 5{4 7{5 7{5 3{2 7{5 7{5 3{2 7{5 7{5 7{5 7{5 3{2 7{5 7{5 3{2 7{5 7{5 7{5 7{5 3{2 7{5 7{5 3{2 7{5 7{5 7{5 7{5 3{2 7{5 7{5 3{2 7{5 7{5 7{5 7{5 3{2 7{5 7{5 3{2 7{5 7{5 5{4 ‚ q4 " 6{5 and q5 " 1 1 5{4 3{2 3{2 3{2 6{5 3{2 3{2 7{5 This completes the recovery of the carrousel decomposition of Example 7.3. Example 14.4. Our final example exhibits some B-pieces of type Apq, qq. It is obtained by modifying the previous example (discussed also in 9.6, 7.3 and in [1]), 48 WALTER D NEUMANN AND ANNE PICHON which had equation pzx2 ` y3qpx3 ` zy2q ` z7 " 0. We use instead the equation pzx2 ` y3qpx5 ` z3y2q ` z9 " 0 which replaces one the two Bp6{5q-pieces of the geometric decomposition of the previous example by a Bp8{7q-piece. Each of these B-pieces has non-trivial topol- ogy, and by Proposition 10.1 each of X6{5 and X8{7 has a second piece which is a B-pieces of type Ap6{5, 6{5q resp. Ap8{7, 8{7q. The first figure shows the resolution graph of a general linear system of hy- perplane sections for this example, followed by the graph of the resolution which resolves the base points of the generic polar, for which an additional blow-up was needed (the intersection numbers at the bottom right are ´3 and ´1). The rates associated with each vertex are also shown. The second figure gives a picture of the complete carrousel section. We use the complete carrousel here to make the A-pieces evident. p1q ´35 p10q ´1 p5q ´2 p7q 1 p77q 6 5 p39q 7 5 ´1 p2q ´1 p2q p6q ´3 p4q ´4 3 2 p15q p1q ´41 p14q ´1 p7q ´2 p7q 1 p107q 8 7 3 2 p15q p46q p31q 5 4 p54q 10 7 p55q Figure 8. Example 14.4: Resolutions of hyperplane sections and basepoints of the polar. LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 49 1 1 5{4 6{5 8{7 3{2 3{2 3{2 3{2 3{2 3{2 3{2 3{2 3{2 3{2 3{2 3{2 10{7 7{5 3{2 3{2 Figure 9. Example 14.4: Complete carrousel section. Part 4: Zariski equisingularity implies semi-analytic Lipschitz triviality 15. Equisingularity In this section, we define Zariski equisingularity as in Speder [22] and we specify it in codimension 2. For a family of hypersurface germs in pC3, 0q with isolated singularities, the definition is equivalent to the last definition of equisingularity stated by Zariski in [33] using equidimensionality type. We also define Lipschitz triviality in this codimension and we fix notations for the rest of the paper. Definition 15.1. Let pX, 0q Ă pCn, 0q be a reduced hypersurface germ and pY, 0q Ă pX, 0q a nonsingular germ. X is Zariski equisingular along Y near 0 if for a generic linear projection L : Cn Ñ Cn´1 for the pair pX, Y q at 0, the branch locus ∆ Ă Cn´1 of the restriction L X : X Ñ Cn´1 is equisingular along L pY q near 0 (in particular, pL pY q, 0q Ă p∆, 0q). When Y has codimension one in X, Zariski equisingularity means that ∆ is nonsingular. The notion of a generic linear projection is defined in [22, definition 4]. When Y has codimension one in X, Zariski equisingularity is equivalent to Whit- ney conditions for the pair pX r Y, Y q and also to topological triviality of X along Y . From now on we consider the case where pY, 0q is the singular locus of X and Y has codimension 2 in X (i.e., dimension n ´ 3). Then any slice of X by a smooth 3-space transversal to Y is a normal surface singularity. If L is a generic projection for pX, Y q at 0 and H a hyperplane through y P Y close to 0 which contains the line ker L , then the restriction of L to H X X is a generic projection of the normal surface germ pH X X, 0q in the sense of Definition 6.1. We have the following characterization of Zariski equisingularity for families of isolated singularities in pC3, 0q which will be used throughout the rest of the paper: Proposition 15.2. Let pX, 0q Ă pCn, 0q be a reduced hypersurface germ at the origin of Cn with 2-codimension smooth singular locus pY, 0q. Then X is Zariski 50 WALTER D NEUMANN AND ANNE PICHON equisingular along Y near 0 if for a generic linear projection L : Cn Ñ Cn´1, the branch locus ∆ Ă Cn´1 of L X : X Ñ Cn´1 is topologically equisingular along L pY q near 0. (cid:3) Remark 15.3. In [33], Zariski introduced an alternative definition of equisingu- larity based on the concept of dimensionality. The definition is also by induction on the codimension and involves the discriminant locus of successive projections. At each step, a number called dimensionality is defined by induction, and it is the same for "almost all" projections. This defines a notion of genericity for such pro- jections. The definition says X is equisingular along Y if the dimensionality equals 1 at the last step of the induction i.e., when Y has codimension 1 in X. This is equivalent to saying that the family of discriminants is equisingular as a family of curves. For details see [33] or [13]. In [2], Brian¸con and Henry proved that when Y has codimension 2 in X, i.e., for a family of complex hypersurfaces in C3, the dimensionality can be computed using only linear projections. As a consequence, in codimension 2, Zariski equisingularity as defined in 15.1 coincides with Zariski's equisingularity concept using dimensionality. Definition 15.4. (1) pX, 0q has constant (semi-analytic) Lipschitz geometry along Y if there exists a smooth (semi-analytic) retraction r : pX, 0q Ñ pY, 0q whose fibers are transverse to Y and a neighbourhood U of 0 in Y such that for all y P U , there exists a (semi-analytic) bilipschitz homeomorphism hy : pr´1pyq, yq Ñ pr´1p0q X X, 0q. (2) The germ pX, 0q is (semi-analytic) Lipschitz trivial along Y if there exists a germ at 0 of a (semi-analytic) bilipschitz homeomorphism Φ : pX, Y q Ñ pX, 0q Y with ΦY " idY , where pX, 0q is a normal complex surface germ. The aim of this last part is to prove Theorem 1.1 stated in the introduction: Theorem. The following are equivalent: (1) pX, 0q is Zariski equisingular along Y ; (2) pX, 0q has constant Lipschitz geometry along Y ; (3) pX, 0q has constant semi-analytic Lipschitz geometry along Y ; (4) pX, 0q is semi-analytic Lipschitz trivial along Y The implications (4) ñ (3) and (3) ñ (2) are trivial. (2) ñ (1) is an easy consequence of part (5) of Theorem 1.2: Proof of (2) ñ (1). Assume pX, 0q has constant semi-analytic Lipschitz geometry along Y . Let L : Cn Ñ Cn´1 be a generic linear projection for X. Let r : L pXq Ñ L pY q be a smooth semi-analytic retraction whose fibers are transversal to L pY q. Its lift by L is a retractionrr : X Ñ Y whose fibers are transversal to Y . For any t P Y sufficiently close to 0, Xt " prq´1ptq is semi-analytically bilipschitz equivalent to X0 " prq´1p0q. Then, according to part (5) of Theorem 1.2, the discriminants ∆t of the restrictions L Xt have same embedded topology. This proves that X is Zariski is equisingular along Y . (cid:3) The last sections of the paper are devoted to the proof of (1) ñ (4). Notations. Assume X is Zariski equisingular along Y at 0. Since Zariski equi- singularity is a stable property under analytic isomorphism pCn, 0q Ñ pCn, 0q, we will assume without lost of generality that Y " t0u Cn´3 Ă C3 Cn´3 " Cn LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 51 where 0 is the origin in C3. We will denote by px, tq the coordinates in C3 Cn´3, with x " px, y, zq P C3 and t P Cn´3. For each t, we set Xt " X X pC3 ttuq and consider X as the n ´ 3-parameter family of isolated hypersurface singularities pXt, p0, tqq Ă pC3 ttu, p0, tqq. For simplicity, we will write pXt, 0q for pXt, p0, tqq. Let L : Cn Ñ Cn´1 be a generic linear projection for X. We choose the coordi- nates px, tq in C3 Cn´3 so that L is given by L px, y, z, tq " px, y, tq. For each t, the restriction L C3ttu : C3 ttu Ñ C2 ttu is a generic linear projection for Xt. We denote by Πt Ă Xt the polar curve of the restriction L Xt : Xt Ñ C2 ttu and by ∆t its discriminant curve ∆t " L pΠtq. Throughout Part 4, we will use the Milnor balls defined as follows. By Speder [22], Zariski equisingularity implies Whitney conditions. Therefore one can choose a constant size ǫ ą 0 for Milnor balls of the pXt, 0q as t varies in some small ball Dδ about 0 P Cn´3 and the same is true for the family of discriminants ∆t. So we can actually use a "rectangular" Milnor ball as introduced in Section 8: B6 ǫ " tpx, y, zq : x ď ǫ, py, zq ď Rǫu , and its projection B4 ǫ " tpx, yq : x ď ǫ, y ď Rǫ, u , where R and ǫ0 are chosen so that for ǫ ď ǫ0 and t in a small ball Dδ, (1) B6 ǫ ttu is a Milnor ball for pXt, 0q; (2) for each α such that α ď ǫ, h´1 t pαq intersects the standard sphere S6 Rǫ transversely, where ht denotes the restriction ht " xXt ; (3) the curve Πt and its tangent cone T0Πt meet BB6 ǫ ttu only in the part x " ǫ. 16. Proof outline that Zariski implies bilipschitz equisingularity We start by outlining the proof that Zariski equisingularity implies semi-analytic Lipschitz triviality. We assume therefore that we have a Zariski equisingular family X as above. We choose B6 ǫ Ă C3 and Dδ Ă Cn´3 as in the previous section. We want to ǫ Dδq Ñ pX0 ǫ Dδq which preserves the t-parameter. Our homeomorphism will be construct a semi-analytic bilipschitz homeomorphism Φ : X X pB6 Cn´3q X pB6 a piecewise diffeomorphism. In Section 8 we used a carrousel decomposition of B4 ǫ for the discriminant curve ∆0 of the generic linear projection L X0 : X0 Ñ B4 ǫ and we lifted it to obtain the geometric decomposition of pX, 0q (Definition 8.4). Here, we will consider this construction for each L Xt : Xt Ñ B4 ǫ ttu. Next, we construct a semi-analytic bilipschitz map φ : B4 which restricts for each t to a map φt : B4 carrousel decomposition of B4 for X0. We also arrange that φt preserves a foliation of B4 decompositions. A first approximation to the desired map Φ : X X pB6 pX0 Cn´3q X pB6 ǫ Dδ ǫ ttu which takes the ǫ for Xt to the corresponding carrousel decomposition ǫ adapted to the carrousel ǫ Dδq Ñ ǫ Dδq is then obtained by simply lifting the map φ via the ǫ ttu Ñ B4 ǫ Dδ Ñ B4 52 WALTER D NEUMANN AND ANNE PICHON branched covers L and L X0 idDδ : X X pB6 ǫ Dδq ΦÝÝÝÝÑ pX0 X B6 ǫ q Dδ L§§đ B4 ǫ Dδ φ ÝÝÝÝÑ L X0 idDδ§§đ B4 ǫ Dδ with aiptq P Cttu. For each t, the tangent cone of ∆t is a union of tangent lines Lp1q and these are distinct lines for each t. In Section 3 we described carrousel decompositions when there is no parameter t. They now decompose a conical neighbourhood of , . . . , Lpmq t t Now, fix t in Dδ. For x P Xt XB6 of the projection L Xt : Xt Ñ B4 and consider the set BK0,t " tx P Xt X pB6 ǫ , denote by Kpx, tq the local bilipschitz constant ǫ ttu (see Section 6). Let us fix a large K0 ą 0 ǫ Dδq : Kpx, tq ě K0u. zone CK0 " ŤtPDδ We then show that Φ is bilipschitz for the inner metric except possibly in the BK0,t, where the bilipschitz constant for the linear projection L becomes large. This is all done in Section 17. Using the approximation Proposition 6.5 of BK0,t by polar wedges, we then prove that Φ can be adjusted if necessary to be bilipschitz on CK0. The construction just explained is based on the key Lemma 20.3 which states constancy of the polar rates of the components of p∆t, 0q as t varies. The proof of this lemma is based on the use of families of plane curves pγt, 0q such that p∆tYγt, 0q has constant topological type as t varies, and their liftings by L , spreadings and projected liftings defined in Section 18. The key argument is the restriction formula of Teissier [24] that we recall in section 19. Similar techniques were used by Casas- Alvero in [7] in his study of equisingularity of inverse images of plane curves by an analytic morphism pC2, 0q Ñ pC2, 0q. Once Φ is inner bilipschitz, we must show that the outer Lipschitz geometry is also preserved. This is done in Section 20 using again families of plane curves pγt, 0q as just defined and the restriction formula. Namely we prove the invariance of bilipschitz type of the liftings by L of such families, which enables one to test invariance of the outer Lipschitz geometry. The pieces of the proof are finally put together in Section 21. 17. Foliated Carrousel Decomposition ǫ Dδ Ñ B4 We use again the notations of Section 15. In this section, we construct a self-map φ : B4 ǫ Dδ of the image of the generic linear projection which removes the dependence on t of the carrousel decompositions and we introduce a foliation of the carrousels by 1-dimensional complex leaves which will be preserved by φ. We first present a parametrized version of the carrousel construction of section 3 for any choice of truncation of the Puiseux series expansions of ∆. Carrousel depending on the parameter t. Since the family of discriminants ∆t is Zariski equisingular, then by [30, Theorem 7] (see also [20]), it admits a Puiseux parametrization with parameter, i.e., the branches of ∆t admits Puiseux parametrizations of the form y " ÿiě1 aiptqxpi , LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 53 t each line Lpjq the j-th line be y " apjq cones . These conical neighbourhoods are as follows. Let the equation of 1 ptqx. We choose a small enough η ą 0 and δ ą 0 such that V pjq t :" tpx, yq : y ´ apjq 1 ptqx ď ηx, x ď ǫu Ă C2 t are disjoint for all t P Dδ, and then shrink ǫ if necessary so ∆pjq completely in V pjq for all t. t t X tx ď ǫu will lie We now describe how a carrousel decomposition is modified to the parametrized case. We fix j " 1 for the moment and therefore drop the superscripts, so our tangent line L has equation y " a1ptqx. for each pair κ " pf, pkq consisting of a Puiseux polynomial f " řk´1 and an exponent pk for which there is a Puiseux series y " řk We first choose a truncation of the Puiseux series for each component of ∆t. Then i"1 aiptqxpi i"1 aiptqxpi ` . . . describing some component of ∆t, we consider all components of ∆t which fit this data. If ak1ptq, . . . , akmκ ptq are the coefficients of xpk which occur in these Puiseux polynomials we define Bκ,t :"!px, yq : ακxpk ď y ´ k´1ÿi"1 y ´ p aiptqxpi ď βκxpk , k´1ÿi"1 aiptqxpi ` akj ptqxpk q ě γκxpk for j " 1, . . . , mκ) . Again, ακ, βκ, γκ are chosen so that ακ ă akj ptq ´ γκ ă akj ptq ` γκ ă βκ for each j " 1, . . . , mκ and all small t. If ǫ is small enough, the sets Bκ,t will be disjoint for different κ. The closure of the complement in Vt of the union of the Bκ,t's is a union of A- and D-pieces. Foliated carrousel. We now refine our carrousel decomposition by adding a piece- wise smooth foliation of Vt by complex curves compatible with the carrousel. We do this as follows: A piece Bκ,t as above is foliated with closed leaves given by curves of the form Cα :" tpx, yq : y " aiptqxpi ` αxpk u k´1ÿi"1 for α P C satisfying ακ ď α ď βκ and akjptq ´ α ě γκ for j " 1, . . . , mκ and all small t. We foliate D-pieces similarly, including the pieces which correspond to ∆-wedges about ∆t. An A-piece has the form A " tpx, yq : β1xpk`1 ď y ´ p kÿi"1 aiptqxpi q ď β0xpk u , where akptq may be 0, pk`1 ą pk and β0, β1 are ą 0. We foliate with leaves the immersed curves of the following form Cr,θ :" tpx, yq : y " aiptqxpi ` βprqeiθ xru kÿi"1 54 WALTER D NEUMANN AND ANNE PICHON with pk ď r ď pk`1, θ P R and βprq " β1 ` pk`1´r pβ0 ´ β1q. Note that these pk`1´pk leaves may not be closed; for irrational r the topological closure is homeomorphic to the cone on a torus. Definition 17.1. We call a carrousel decomposition equipped with such a foliation by curves a foliated carrousel decomposition. Trivialization of the family of foliated carrousels. We set V :" ďtPDδ Vt ttu . Proposition 17.2. If δ is sufficiently small, there exists a semi-analytic bilipschitz map φV : V Ñ V0 Dδ such that: (1) φV preserves the x and t-coordinates, (2) for each t P Dδ, φV preserves the foliated carrousel decomposition of Vt i.e., it maps the carrousel decomposition of Vt to that of V0, preserving the respective foliations. (3) φV maps complex lines to complex lines on the portion x ă ǫ of BV . Proof. We first construct the map on the slice V X tx " ǫu. We start by extending the identity map V0 X tx " ǫu Ñ V0 X tx " ǫu to a family of piecewise smooth maps Vt X tx " ǫu Ñ V0 X tx " ǫu which map carrousel sections to carrousel sections. For fixed t we are looking at a carrousel section Vt X tx " ǫu as exemplified in Figure 1. The various regions are bounded by circles. Each disk or annulus in Vt X tx " ǫu is isometric to its counterpart in V0 X tx " ǫu so we map it by a translation. To extend over a piece of the form Bκ,t X tx " ǫu we will subdivide Bκ,t further. Assume first, for simplicity, that the coefficients ak1ptq, . . . , akmκ ptq in the description of Bκ,t satisfy ak1ptq ă ak2ptq ă ă akmκ ptq for small t. For each j " 1, . . . , mκ we define Bκj,t :"!px, yq : α1 k´1ÿi"1 κakj ptqxpk ď y ´ aiptqxpi ď β1 κakj ptqxpk , y ´ p aiptqxpi ` akj ptqxpk q ě γ1 k´1ÿi"1 κakj ptqxpk ) , κak1ptq ă β1 κ ă 1 ´ γ1 κ ă 1 ` γ1 κ, β1 κakjptq ă α1 with α1 κak,j`1 for j " 1, . . . , mκ ´ 1, γ1 κakjptq ď γk, ακ ă α1 κakmκ ptq ă βκ. This subdivides Bκ,t into pieces Bκj,t with Appk, pkq-pieces between each Bκj,t and Bκ j`1,t and between the Bκj,t's and the boundary components of Bκ,t. We can map Bκj,tXtx " ǫu to Bκj,0X i"1 aiptqǫpi q followed by a similarity i"1 aip0qǫpi. We map the annuli of i"1 aip0qǫpi ´řk´1 tx " ǫu by a translation by přk´1 which multiplies by akj p0q{akjptq centered atřk´1 Bκ,0 rŤ Bκj,0 X tx " ǫu to the corresponding annuli of Bκ,t rŤ Bκj,t X tx " ǫu by maps which agree with the already constructed maps on their boundaries and which extend over each annulus by a "twist map" of the form in polar coordinates: pr, θq ÞÑ par ` b, θ ` cr ` dq for some a, b, c, d. We assumed above that the akj ptq are pairwise unequal, but if akj ptq " akj1 ptq as t varies, then akjptq{akj1 ptq is a complex analytic function of t with values in the unit circle, so it must be constant. So in this situation we use a single Bκj piece for all j1 with akj ptq " akj1 ptq and the construction still works. LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 55 This defines our map φV on tx " ǫu Dδ for δ sufficiently small, and the requirement that φV preserve x-coordinate and foliation extends it uniquely to all of V Dδ. The map φV is constructed to be semi-analytic, and it remains to prove that it is bilipschitz. It is bilipschitz on pV X tx " ǫuq Dδ since it is piecewise smooth on a compact set. Depending what leaf of the foliation one is on, in a section x " ǫ1 with 0 ă ǫ1 ď ǫ, a neighbourhood of a point p scales from a neighbourhood of a point p1 with x " ǫ by a factor of pǫ1{ǫqr in the y-direction (and 1 in the x direction) as one moves in along a leaf, and the same for φV ppq. So to high order the local bilipschitz constant of φV at p is the same as for p1 and hence bounded. Thus the bilipschitz constant is globally bounded on pV r t0uq Dδ, and hence on V Dδ. (cid:3) The V of the above proposition was any one of the m sets V piq :"ŤtPDδ foliation structure by Bǫ,t. We finally extend φ to the whole of ŤtPDδ V piq , t so the proposition extends φ to all these sets. We extend the carrousel foliation on the union of these sets to all of B4 ǫ ttu by foliating the complement of the union of V piq ǫ with this carrousel decomposition and Bǫ,t by a diffeomorphism which takes the complex lines of the foliation linearly to complex lines of the foliation on Bǫ,0 Dδ, preserving the x coordinate. The resulting map remains obviously semi-analytic and bilipschitz. 's by complex lines. We denote B4 t We then have shown: Proposition 17.3. There exists a map φ : ŤtPDδ Bǫ,t Ñ Bǫ,0 Dδ which is semi- analytic and bilipschitz and such that each φt : Bǫ,t Ñ Bǫ,0 ttu preserves the carrousel decompositions and foliations. (cid:3) 18. Liftings and spreadings Let pX, 0q Ă pC3, 0q be an isolated hypersurface singularity and let ℓ : C3 Ñ C2 be a generic linear projection for pX, 0q. Consider an irreducible plane curve pγ, 0q Ă pC2, 0q. Definition 18.1. The lifting of γ is the curve Lγ :" pℓX q´1pγq . Let ℓ1 : C3 Ñ C2 be another generic linear projection for pX, 0q which is also generic for Lγ. We call the plane curve a projected lifting of γ. Pγ " ℓ1pLγq Notice that the topological type of pPγ , 0q does not depend on the choice of ℓ1. Let us choose the coordinates px, y, zq of C3 in such a way that ℓ " px, yq and γ is tangent to the x -- axis. We consider a Puiseux expansion of γ: xpwq " wq ypwq " a1wq1 ` a2wq2 ` . . . ` an´1wqn´1 ` anwqn ` . . . Let F px, y, zq " 0 be an equation for X Ă C3. 56 WALTER D NEUMANN AND ANNE PICHON Definition 18.2. We call the plane curve pFγ , 0q Ă pC2, 0q with equation a spreading of the lifting Lγ. F pxpwq, ypwq, zq " 0 Notice that the topological type of pFγ , 0q does not depend on the choice of the parametrization. Lemma 18.3. The topological types of the spreading pFγ , 0q and of the projected lifting pPγ , 0q determine each other. Proof. Assume that the coordinates of C3 are chosen in such a way that ℓ1 " px, zq. Then Pγ " ρpFγq where ρ : C2 Ñ C2 denotes the morphism ρpw, zq " pwq, zq, which is a cyclic q-fold cover branched on the line w " 0. Thus Pγ determines Fγ. The link of Pγ Y tz-axisu consists of an iterated torus link braided around an axis and it has a Z{q-action fixing the axis. Such a Z{q-action is unique up to isotopy. Thus the link of Fγ can be recovered from the link of Pγ by quotienting by this Z{q -- action. (cid:3) 19. Restriction formula In this section, we apply the restriction formula to compute the Milnor number of a spreading Fγ. Let us first recall the formula. Proposition 19.1. ([24, 1.2]) Let pY, 0q Ă pCn`1, 0q be a germ of hypersurface with isolated singularity, let H be a hyperplane of Cn`1 such that H X Y has isolated singularity. Assume that pz0, . . . , znq is a system of coordinates of Cn`1 such that H is given by z0 " 0. Let proj: pY, 0q Ñ pC, 0q be the restriction of the function z0. Then mH " µpY q ` µpY X Hq , where µpY q and µpY X Hq denote the Milnor numbers respectively of pY, 0q Ă pCn`1, 0q and pY X H, 0q Ă pH, 0q, and where mH is the multiplicity of the ori- gin 0 as the discriminant of the morphism proj. (cid:3) The multiplicity mH is defined as follows ([24]). Assume that an equation of pY, 0q is f pz0, z1, . . . , znq " 0 and let pΓ, 0q be the curve in pCn`1, 0q defined by Bf " 0. Then mH " pY, Γq0, the intersection at the origin of the two Bz1 germs pY, 0q and pΓ, 0q. " " Bf Bzn Remark 19.2. Applying this when n " 1, i.e., in the case of a plane curve pY, 0q Ă pC2, 0q, we obtain: where multpY q denotes the multiplicity of pY, 0q. mH " µpY q ` multpY q ´ 1 , Proposition 19.3. Let pX, 0q Ă pC3, 0q be an isolated hypersurface singularity. Let ℓ : C3 Ñ C2 be a generic linear projection for pX, 0q and let pγ, 0q Ă pC2, 0q be an irreducible plane curve which is not a branch of the discriminant curve of ℓX . The Milnor number at 0 of the spreading pFγ , 0q can be computed in terms of the following data: (1) the multiplicity of the surface pX, 0q; (2) the topological type of the triple pX, Π, Lγq where pΠ, 0q Ă pX, 0q denotes the polar curve of ℓX. LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 57 Proof. Choose coordinates in C3 such that ℓ " px, yq, and consider a Puiseux expansion of γ as in Section 18. Applying the restriction formula 19.1 to the projection proj : pFγ , 0q Ñ pC, 0q defined as the restriction of the function w, we obtain: µpFγq " pFγ , Γq0 ´ multpFγq ` 1 , where pΓ, 0q has equation BF Bz pxpwq, ypwq, zq " 0 . We have easily: multpFγq " multpX, 0q. Moreover, the polar curve Π is the set tpx, y, zq P X : BF Bz px, y, zq " 0u. Since the map θ : Fγ Ñ Lγ defined by θpw, zq " pxpwq, ypwq, zq is a bijection, the intersection multiplicity pFγ , Γq0 in C2 equals the intersection multiplicity at 0 on the surface pX, 0q of the two curves Lγ and Π. (cid:3) 20. Constancy of projected liftings and of polar rates We use again the notations of Section 16 and we consider a Zariski equisingular family pXt, 0q of 2-dimensional complex hypersurfaces. Proposition 20.1. The following data is constant through the family pXt, 0q: (1) the multiplicity of the surface pXt, 0q, (2) the topological type of the pair pXt, Πtq. Proof. (1) follows (in all codimension) from the chain of implications: Zariski equi- singularity ñ Whitney conditions (Speder [22]) ñ µ-constancy (e.g., [6]; see also [28] in codimension 2). (2) follows immediately since pXt, Πtq is a continuous family of branched covers of C2 of constant covering degree and with constant topology of the branch locus. (cid:3) Corollary 20.2. Consider a family of irreducible plane curves pγt, 0q Ă pC2, 0q such that γt is not a branch of ∆t and the topological type of pγt Y ∆t, 0q is constant as t varies. Let L 1 : C3 Cn´3 Ñ C2 Cn´3 be a generic linear projection for pX, 0q which is also generic for the family of liftings pLγt qt. Then the family of projected liftings Pγt " L 1pLγtq has constant topological type. Proof. According to Propositions 19.3 and 20.1, the Milnor number µpFγt q is con- stant. The result follows by the Le-Ramanujan theorem for a family of plane curves ([15]) and Lemma 18.3. (cid:3) Let us now consider a branch p∆1 tqt of the family of discriminant curves p∆tqt. For each t, let sp∆1 tq denote the polar rate of ∆1 t. Lemma 20.3. The polar rate sp∆1 tq does not depend on t P Dδ. Proof. Let r P 1 curves defined by N N and λ P C and let pγt, 0q Ă pC2, 0q be the family of irreducible y " ÿiěN aiptqxi{N ` λxr , where y "řiěN aiptqxi{N is a Puiseux expansion of ∆1 N N, the curves p∆t Y γt, 0q have constant topological type as t varies. Then, by Corollary 20.2, the projected liftings Pγt have constant topological type. We take r big enough to be sure that the curve γt is in a ∆-wedge about ∆1 γt be the union of t. Since r P 1 t. Let P 1 58 WALTER D NEUMANN AND ANNE PICHON components of Pγt which are in a polar wedge about Π1 γt qt also has constant topological type. By Lemma 6.6, this implies the constancy of the rates qtprq " sp∆1 tq and qtprq ă r for each r ą sp∆1 tq. Since qtprq is constant through the family, the inequality qtprq ă r holds for each t or for none. This proves the constancy of the polar rate sp∆1 tq as t varies in Dδ. (cid:3) t. Then the family pP 1 tq`r 2 . Obviously qtpsp∆1 tqq " sp∆1 21. Proof that Zariski equisingularity implies Lipschitz triviality We use the notations of Sections 15 and 16. For t fixed in Dδ and K0 ą 0 sufficiently large, recall (Section 16) that BK0,t denotes the neighbourhood of Πt in Xt where the local bilipschitz constant Kpx, tq of L Xt is bigger than K0. The aim of this section is to complete the proof that Zariski equisingularity implies Lipschitz triviality. According to Lemma 20.3, the polar rate of each branch of p∆t, 0q is constant as t varies. Then, we can consider the carrousel decomposition for p∆t, 0q obtained by truncating the Puiseux expansion of each branch at the first term which has exponent greater or equal to the corresponding polar rate and where truncation does not affect the topology of p∆t, 0q. In other words, we consider the complete carrousel decomposition for p∆t, 0q as defined in Section 7, obtaining a carrousel decomposition depending on the parameter t, as introduced in Section 17. In particular, each component of p∆t, 0q is contained in a D-piece which is a ∆-wedge about it. We consider the corresponding foliated carrousels as described in Section 17 and we denote by Bǫ,t the ball B4 ttu equipped with this foliated carrousel. Then, we consider a trivialization of this family of carrousels as in Propositions 17.2 and 17.3. The following result is a corollary of Proposition 17.3: Lemma 21.1. There exists a commutative diagram pX, 0q X pC3 Dδq ΦÝÝÝÝÑ pX0, 0q Dδ L X§§đ pC2, 0q Dδ φ ÝÝÝÝÑ pC2, 0q Dδ L X0 id§§đ BK0,t. such that φ is the map of the above proposition and Φ is semi-analytic, and inner bilipschitz except possibly in the set CK0 :"ŤtPDδ Proof. Φ is simply the lift of φ over the branched cover L X. The map L X is a local diffeomorphism with Lipschitz constant bounded above by K0 outside CK0 so Φ has Lipschitz coefficient bounded by K 2 0 times the Lipschitz bound for φ outside CK0. The semi-analyticity of Φ is because φ, L X and L X0 id are semi-analytic. (cid:3) We will show that the map Φ of Lemma 21.1 is bilipschitz with respect to the outer metric after modifying it as necessary within the set CK0. Proof. We first make the modification in BK0,0. Recall that according to Propo- sition 6.5, BK0,t can be approximated for each t by a polar wedge about Πt, and then its image NK0,t " L Xt pBK0,tq can be approximated by a ∆-wedge about the polar curve ∆t, i.e., by a union of D-pieces of the foliated carrousel decomposition Bǫ,t. LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 59 Let N 1 K0,0 be one of these D-pieces, and let q be its rate. By [1, Lemma 12.1.3], any component B1 K0,0 of BK0,0 over N 1 K0,0 is also a D(q)-piece in X0. By Proposition 17.3, the inverse image under the map φt of N 1 K0,0 is N 1 K0,t. Then the inverse image under the lifting Φt of B1 K0,t in Xt with the same polar rate. So we can adjust Φt as necessary in this zone to be a bilipschitz equivalence for the inner metric and to remain semi-analytic. K0,0 Ă X0 will be the polar wedge B1 We can thus assume now that Φ is semi-analytic and bilipschitz for the inner metric, with Lipschitz bound K say. We will now show it is also bilipschitz for the outer metric. We first consider a pair of points p1, p2 of Xt X B6 ǫ which lie over the same point p P Bǫ,t (we will say they are "vertically aligned"). We may assume, by moving them slightly if necessary, that p is on a closed curve γ of our foliation of Bǫ,t with Puiseux expansion xpwq " wq ypwq " a1wq1 ` a2wq2 ` . . . ` an´1wqn´1 ` anwqn ` . . . Let L 1 be a different generic linear projection as in Corollary 20.2. If w " w0 is the parameter of the point p P γ, consider the arc Cpsq " pxpsw0q, ypsw0qq, s P r0, 1q and lift it to Xt to obtain a pair of arcs p1psq, p2psq, s P r0, 1q with p1p1q " p1 and p2p1q " p2. We assume that the distance dpp1psq, p2psqq shrinks faster than linearly as s Ñ 0, since otherwise the pair is uninteresting from the point of view of bilipschitz geometry. Note that the distance dpp1psq, p2psqq is a multiple kdpL 1pp1psqq, L 1pp2psqqq (with k depending only on the projections L and L 1). Now dpL 1pp1psqq, L 1pp2psqqq is to high order srdpL 1pp1p1qq, L 1pp2p1qqq for some rational r ą 1. Moreover, by Corollary 20.2, if we consider the corresponding picture in X0, starting with Φpp1q and Φpp2q we get the same situation with the same exponent r. So to high order we see that dpΦpp1psqq, Φpp2psqqq{dpp1psq, p2psqq is constant. There is certainly an overall bound on the factor by which Φ scales distance between vertically aligned points p1 and p2 so long as we restrict ourselves to the compact complement in X X pB6 ǫ Dδq of an open neighbourhood of t0u Dδ. The above argument then shows that this bound continues to hold as we move towards 0. Thus Φ distorts distance by at most a constant factor for all vertically aligned pairs of points. Finally, consider any two points p1, p2 P X0 X B6 ǫ and their images q1, q2 in C2. We will assume for the moment that neither p1 or p2 is in the approximated polar wedge BK0,0. The outer distance between p1 and p2 is the length of the straight segment joining them. Let γ be the image of this segment, so γ connects q1 to q2. If γ intersects the approximated ∆-wedge NK0,0, we can modify it to a curve which avoids NK0,0 and which has length less than π times the original length of γ. Lift γ to a curve γ1 starting at p1. Then γ1 ends at a point p1 2 which is vertically aligned with p2. Let γ2 be the curve obtained by appending the vertical segment p1 2p2 to γ1. We then have: lenpp1p2q ď lenpγ2q , where len denotes length. On the other hand, lenpγ1q ď K0lenpγq, lenpγq ď πlenpq1q2q, and since the segment q1q2 is the projection of p1p2, lenpq1q2q ď lenpp1p2q. Thus lenpγ1q ď πK0lenpp1p2q. If we connect the segment p1p2 to γ1 at p1 we get a curve from p2 to p1 2p2q ď p1 ` πK0qlenpp1p2q and as 2, so lenpp1 60 WALTER D NEUMANN AND ANNE PICHON lenpγ2q " lenpγ1q ` lenpp1 2p2q, we then obtain: lenpγ2q ď p1 ` 2K0πqlenpp1p2q . We have thus shown that up to a bounded constant the outer distance between two points can be achieved by following a path in X followed by a vertical segment. We have proved this under the assumption that we do not start or end in the approximated polar wedge BK0,0. Now, take two other projections L 1 and L 2 such that for K0 sufficiently large, the corresponding approximated polar wedges BK0,0 " BK0,0pL q, BK0,0pL 1q and BK0,0pL 2q are pairwise disjoint outside the origin. Then if p1 and p2 are any two points in X0 X B6 ǫ , they are outside the polar wedge for at least one of L , L 1 or L 2 and we conclude as before. The same argument applies to X0 Dδ. Now paths in X are modified by at most a bounded factor by Φ since Φ is inner bilipschitz, while vertical segments are also modified by at most a bounded factor by the previous argument. Thus outer metric is modified by at most a bounded factor. (cid:3) References [1] Lev Birbrair, Walter D Neumann and Anne Pichon, The thick-thin decomposition and the bilipschitz classification of normal surface singularities, Acta Math. 212 (2014), 199 -- 256. [2] Joel Brian¸con and Jean-Pierre Henry, ´Equisingularit´e g´en´erique des familles de surfaces `a singularit´es isol´ees, Bull. Soc. Math. France 108 (1980), 260 -- 284. [3] Andreas Bernig and Alexander Lytchak, Tangent spaces and Gromov-Hausdorff limits of subanalytic spaces, J. Reine Angew. Math. 608 (2007), 1 -- 15. [4] Joel Brian¸con and Jean-Paul Speder, La trivialit´e topologique n'implique pas les conditions de Whitney, C.R. Acad. Sc. Paris, 280 (1975), 365 -- 367. [5] Joel Brian¸con and Jean-Paul Speder, Familles ´equisinguli`eres de surfaces `a singularit´e isol´ee, C.R. Acad. Sc. Paris, t.280, 1975, 1013 -- 1016. [6] Joel Brian¸con and Jean-Paul Speder, Les conditions de Whitney impliquent µpq constant, Annales de l'Inst. Fourier, tome 26, 2 (1976), 153 -- 163. [7] Eduardo Casas-Alvero, Discriminant of a morphism and inverse images of plane curve singu- larities, Math. Proc. Camb. Phil. Soc. (2003), 135, 385 -- 394. [8] David Eisenbud and Walter D Neumann, Three dimensional link theory and invariants of plane curves singularities, Annals of Mathematics Studies 110, Princeton University Press, 1985. [9] Alexandre Fernandes, Topological equivalence of complex curves and bi-Lipschitz maps, Michigan Math. J. 51 (2003), 593 -- 606. [10] G´erard Gonzalez-Springberg, R´esolution de Nash des points doubles rationnels, Ann. Inst. Fourier, Grenoble 32 (1982), 111 -- 178. [11] Jean-Pierre Henry and Adam Parusi´nski, Existence of moduli for bi-Lipschitz equivalence of analytic functions, Compositio Math., 136 (2003), 217 -- 235. [12] Le Dung Tr´ang, The geometry of the monodromy theorem. C. P. Ramanujam -- a tribute, Tata Inst. Fund. Res. Studies in Math., 8 (Springer, Berlin-New York, 1978), 157 -- 173. [13] Joseph Lipman, Equisingularity and simultaneous resolution of singularities, Resolution of singularities (Obergurgl, 1997), Progr. Math., 181 (Birkhauser, Basel, 2000), 485 -- 505. [14] Le Dung Tr´ang and Bernard Teissier, Vari´et´es polaires locales et classes de Chern des vari´et´es singuli`eres, Ann. Math., 2nd Ser. 114 (3) (1981), 457 -- 491. [15] Le Dung Tr´ang and C.P. Ramanujam, The invariance of Milnor's number implies the invari- ance of the topological type, Amer. J. Math. 98 (1976), 67 -- 78. [16] Joseph Lipman and Bernard Teissier, Zariski's papers on equisingularity, in The unreal life of Oscar Zariski by Carol Parikh, Springer, 1991, 171 -- 179. [17] Tadeusz Mostowski, Lipschitz equisingularity problems (Several topics theory), Departmental Bulletin Paper larity http://hdl.handle.net/2433/43241 (Kyoto University, in singu- 73 -- 113, 2003) LIPSCHITZ GEOMETRY, ANALYTIC INVARIANTS AND EQUISINGULARITY 61 [18] Walter D Neumann, A calculus for plumbing applied to the topology of complex surface singularities and degenerating complex curves, Trans. Amer. Math. Soc. 268 (1981), 299 -- 343. [19] Walter D Neumann and Anne Pichon, Lipschitz geometry of complex curves, Journal of Singularities 10 (2014), 225 -- 234. [20] Wies law Paw lucki, Le th´eor`eme de Puiseux pour une application sous-analytique Bull. Polish Acad. Sci. Math. 32 (1984), no. 9-10, 555 -- 560. [21] Fr´ed´eric Pham and Bernard Teissier, Fractions Lipschitziennes d'une alg`ebre analytique com- plexe et saturation de Zariski. Pr´epublications ´Ecole Polytechnique No. M17.0669 (1969). [22] Jean-Paul Speder, ´Equisingularit´e et conditions de Whitney, Amer. J. Math. 97 (1975) 571 -- 588. [23] Mark Spivakovsky, Sandwiched singularities and desingularization of surfaces by normalized Nash transformations, Ann. Math. 131 (1990), 411 -- 491. [24] Bernard Teissier, Cycles ´evanescents, sections planes et conditions de Whitney, Ast´erisque 7-8 (SMF 1973). [25] Bernard Teissier, Introduction to equisingularity problems, Proc. of Symposia in pure Math- ematics, Vol. 29 (1975), 593 -- 632. [26] Bernard Teissier, Vari´et´es polaires II. Multiplicit´es polaires, sections planes, et conditions de Whitney, Algebraic geometry (La R`abida, 1981) 314 -- 491, Lecture Notes in Math. 961 (Springer, Berlin, 1982). [27] Ren´e Thom, Ensembles et morphismes stratifi´es, Bull. Amer. Math. Soc., 75 (1969), 240 -- 284. [28] Alexander Varchenko, The relations between topological and algebro-geometric equi- singularities according to Zariski, Functional Anal. Appl. 7 (1973), 87 -- 90. [29] Alexander Varchenko, Algebro-geometrical equisingularity and local topological classification of smooth mapping, Proc. Internat. Congress of Mathematicians, Vancouver 1974, Vol. 1, 427 -- 431. [30] Oscar Zariski, Studies in equisingularity. I. Equivalent singularities of plane algebroid curves. Amer. J. Math. 87 (1965) 507 -- 536; Oscar Zariski: Collected Papers, Volume IV, MIT Press, 31 -- 60. [31] Oscar Zariski, Some open questions in the theory of singularities, Bull. of the Amer. Math. Soc., 77 (1971), 481 -- 491. [32] Oscar Zariski, The elusive concept of equisingularity and related questions, Johns Hopkins Centennial Lectures (supplement to the American Journal of Mathematics) (1977), 9 -- 22. [33] Oscar Zariski, Foundations of a general theory of equisingularity on r-dimensional algebroid and algebraic varieties, of embedding dimension r ` 1, Amer. J. of Math, Vol. 101 (1979), 453 -- 514. Department of Mathematics, Barnard College, Columbia University, 2009 Broadway MC4429, New York, NY 10027, USA E-mail address: [email protected] Aix Marseille Universit´e, CNRS, Centrale Marseille, I2M, UMR 7373, 13453 Mar- seille, FRANCE E-mail address: [email protected]
1309.0596
1
1309
2013-09-03T07:13:34
Multiple Derived Lagrangian Intersections
[ "math.AG", "math.AT", "math.KT" ]
We give a new way to produce examples of Lagrangians in shifted symplectic derived stacks, based on multiple intersections. Specifically, we show that an m-fold fiber product of Lagrangians in a shifted symplectic derived stack its itself Lagrangian in a certain cyclic product of pairwise homotopy fiber products of the Lagrangians.
math.AG
math
MULTIPLE DERIVED LAGRANGIAN INTERSECTIONS OREN BEN-BASSAT Abstract. We give a new way to produce examples of Lagrangians in shifted symplectic derived stacks, based on multiple intersections. Specifically, we show that an m-fold homotopy fiber product of Lagrangians in a shifted symplectic derived stack its itself Lagrangian in a certain cyclic product of pairwise homotopy fiber products of the Lagrangians. 1. Introduction In a recent article [15] of Pantev, Toen, Vaqui´e, and Vezzosi, shifted symplectic structures were defined and studied from the point of view of algebraic geometry. The objects of study in this area are shifted symplectic derived stacks. As is pervasive in symplectic geometry, the study of Lagrangians is central. The notion of a Lagrangian in this context was also introduced in [15]. In [15] an interesting new structure was produced on the (homotopy) fiber product of Lagrangians in a shifted symplectic derived stack. Namely, that given two such Lagrangians inside an n-shifted symplectic derived stack, their fiber product has the natural structure of an n − 1 shifted symplectic derived stack. Our main result was a conjecture of Dominic Joyce. The proof of this result (Theorem 3.2) uses techniques directly from the article [15]. Recently, an interesting preprint [2] of Calaque appeared on categories build out of shifted symplectic derived stacks and Lagrangians in them (also Lagrangian correspondences) where some interesting new techniques were introduced. The topics in [2] are closely related to the result which we present. We pursue these relations in separate work [1] where we look at a certain infinity category of shifted symplectic derived stacks and relate this to permutahedra inside the spaces of symplectic forms on various iterated intersections. 2. Review of some parts the article [15] of Pantev, Toen, Vaqui´e, and Vezzosi As we are following closely the notation and techniques of [15] we refer to that work and the references contained in it for background material on derived Artin stacks F , tangent and cotangent complexes on them, their homotopy fiber products, and their sheaf theory, in particular the infinity categories Lqcoh(F ). The ∞-category of derived stacks over k for the ´etale topology is denoted dStk. We also need the ∞-categories dggr k and ǫ − dgk of graded mixed complexes and of graded complexes which were described in the first section of [15]. We use the convention that we do not denote homotopy fiber products in any special way because all of our fiber products are homotopy fiber products of derived stacks. Let F be a derived Artin stack. Given a distinguished triangle S1 → S2 → S3 in Lqcoh(F ), then S3[−1] → S1 → S2 is also a distinguished triangle. Let f : F → F ′ be a morphism of derived Artin stacks. We will use the cotangent complex LF/k of a derived Artin stack over k, we will simply denote this as LF . More information about this can be found in HAG II [18] and Section 7.3 of Higher Algebra [12]. There is a distinguished triangle in Lqcoh(F ) of the form f ∗LF ′ → LF → Lf I would like to thank Dominic Joyce for his conjecture on the existence of this Lagrangian structure and many key conversa- tions. I would also like to thank Lino Amorim, Chris Brav, Dennis Borisov, Vittoria Bussi, Damien Calaque, Kobi Kremnizer and Tony Pantev for helpful discussions. I acknowledge the support of the European Commission under the Marie Curie Pro- gramme for the IEF grant which enabled this research to take place. The contents of this article reflect my own views and not the views of the European Commission. 1 2 OREN BEN-BASSAT which defines the relative cotangent complex Lf . For the derived Artin stacks we will be dealing with, these complexes will be perfect and there is a dual distinguished triangle of tangent complexes Tf → TF → f ∗TF ′ . The ∞-category of simplicial sets is denoted S and these are sometimes called spaces. Given a derived Artin stack F , the author's of [15] define a space of n-shifted p-forms Ap(F, n) ∈ S and similarly a space of n-shifted closed p-forms Ap,cl(F, n) ∈ S. We briefly outline some key steps in their definition. Recall that there is an ∞-functor defined as the composition of the ∞-functors N Cw : dStop k −→ dggr k DR : dStop k −→ ǫ − dgop k (as defined in [17] or [13]) and a weighted negative cyclic complex functor (defined on page 15 of [15]) denoted We have by Proposition 1.14 of [15] N Cw : ǫ − dgop k −→ dggr k . Ap(F, n) ∼= M apLqcoh(F )(OF , ∧pLF [n]) ∼= DR(F )[n − p](p) and as ∞-functors dStop k → S. There is a natural transformation Ap,cl(F, n) ∼= N Cw(F )[n − p](p) N Cw(F )[n − p](p) → ∧pLF [n] which gives a natural transformation of topological functors Ap,cl(F, n) → Ap(F, n). In Section 1.2 of [15] it is explained that 2-form ω ∈ A2(F, n) gives rise [15] to a certain morphism Θω : TF → LF [n]. The connected components of the space A2(F, n) for which this morphism is a quasi-isomorphism are called non-degenerate. These components form a full subspace A2(F, n)nd inside A2(F, n). The space of n-shifted symplectic structures on the derived stack F is defined as the homotopy pullback of A2(F, n)nd and A2,cl(F, n) over A2(F, n). Let f : X → F be a morphism of derived Artin stacks. An isotropic structure h on an n- shifted symplectic form ω on F is a path from 0 to f ∗ω in X. In Section 2.2 of [15] it is explained that an isotropic structure h gives rise to a certain morphism Θh : Tf → LX [n − 1] and h is said to be Lagrangian if Θh is a quasi-isomorphism. Recall from [15] that given Lagrangians f : X → F and g : Y → F in an n-shifted symplectic derived stack F , corresponding to paths h from 0 to f ∗ω and k from 0 to g ∗ω they define an n − 1 shifted symplectic form R(ω, h, k) on X ×F Y whose underlying morphism of complexes ΘR(ω,h,k) fits into a diagram with exact rows (2.1) TX×F Y pr∗ X TX ⊕ pr∗ Y TY ΘR(ω,h,k) (Θh,Θk) π∗TF Θω LX×F Y [n − 1] / pr∗ X Lf [n − 1] ⊕ pr∗ Y Lg[n − 1] / π∗LF [n] / /   / /     / / in which the bottom row is (a rotation of) the distinguished triangle coming from (2.2) pr∗ X Lf [n − 1] / 0 3 pr∗ Y Lg[n − 1] / π∗LF [n] / pr∗ Y LY [n] where the bottom square is a homotopy pushout (see [18], Lemma 1.4.1.12). 0 / pr∗ X LX[n] / LX×F Y [n] 3. Multiple Derived Lagrangian Intersections Let (S, ω) be a n-shifted symplectic derived stack. We will consider three Lagrangians L, M, N for S. Consider the homotopy cartesian diagram (3.1) L ×S M ×S N L ×S M N ×S L M ×S N qM qN ♥♥♥♥♥ w♦♦♦♦♦♦♦♦♦♦♦♦ 'PPPPPPPPPPPPP w♥♥♥♥♥♥♥ (PPPPPPPPPPPPPPP pN pL fL fM M S qL PPPPP 'PPPPPPPPPPPP v♥♥♥♥♥♥♥♥♥♥♥♥♥ (PPPPPPP v♥♥♥♥♥♥♥♥♥♥♥♥♥♥♥ pN fN pL pM L pM N Note that the term L ×S M ×S N is the homotopy limit of the rest of the diagram. The following theorem was a conjecture of Dominic Joyce, specialized to the case of three Lagrangians. We prove the case of three Lagrangians here and then after that we will present a more general Theorem 3.2 for m Lagrangians. Theorem 3.1. Let S be an n-shifted symplectic derived stack and let fL : L → S, fM : M → S, fN : N → S be Lagrangian morphisms. Then (qN , qL, qM ) : L ×S M ×S N → (L ×S M ) × (M ×S N ) × (N ×S L) is Lagrangian with respect to the symplectic structure q∗ N ωLM , where the morphisms qN , qL, qM are the projection morphisms and ωN L, ωMN , ωLM are the n − 1 shifted symplectic structures provided by [15]. M ωN L + q∗ LωMN + q∗ Proof. The given Lagrangian structures for fL, fM , fN include isotropic structures, hL, hM , hN . These are • a path hL between 0 and f ∗ • a path hM between 0 and f ∗ • a path hN between 0 and f ∗ Lω in A2,cl(L, n) M ω in A2,cl(M, n) N ω in A2,cl(N, n). Consider the homotopy commutativity of the lower three squares of 3.1. This gives • a natural homotopy uLM between p∗ • a natural homotopy uN L between p∗ M f ∗ Lf ∗ L and p∗ N and p∗ Lf ∗ N f ∗ M as maps from A2,cl(S, n) to A2,cl(L ×S M, n) L as maps from A2,cl(S, n) to A2,cl(N ×S L, n)   /     /   /   / / w   '   ' ( w v   (   v 4 OREN BEN-BASSAT • a natural homotopy uMN between p∗ N f ∗ M and p∗ M f ∗ N as maps from A2,cl(S, n) to A2,cl(M ×S N, n). If we apply these homotopies to ω we get Lω to p∗ N ω to p∗ M ω to p∗ • a path uLM ω from p∗ • a path uN Lω from p∗ • a path uMN ω from p∗ M f ∗ Lf ∗ N f ∗ Lf ∗ N f ∗ M f ∗ M ω in A2,cl(L ×S M, n) Lω in A2,cl(N ×S L, n) N ω in A2,cl(M ×S N, n) Similarly, the homotopy commutativity of the upper three squares of (3.1) gives us • a natural homotopy vL between q∗ • a natural homotopy vM between q∗ • a natural homotopy vN between q∗ M p∗ N p∗ M p∗ N and q∗ L and q∗ L and q∗ N p∗ Lp∗ Lp∗ M as maps from A2,cl(L, n) to A2,cl(L ×S M ×S N, n) N as maps from A2,cl(M, n) to A2,cl(L ×S M ×S N, n) M as maps from A2,cl(N, n) to A2,cl(L×S M ×S N, n). If we apply these homotopies to the paths hL, hM , hN we get paths of paths • kL from q∗ • kM from q∗ • kN from q∗ N p∗ N p∗ Lp∗ M hL to q∗ LhM to q∗ M hN to q∗ M p∗ Lp∗ M p∗ N hL N hM LhN in the space of paths starting at the origin in A2,cl(L ×S M ×S N, n). Consider the loops at zero • ωLM = p∗ • ωN L = p∗ • ωMN = p∗ Lh−1 N h−1 M h−1 M ◦ uLM ω ◦ p∗ L ◦ uN Lω ◦ p∗ N ◦ uMN ω ◦ p∗ M hL in the space A2,cl(L ×S M, n)) LhN in the space A2,cl(N ×S L, n)) N hM in the space A2,cl(M ×S N, n)). In the space A2,cl(L ×S M ×S N, n) we have the figure (3.2) q∗ N p∗ M f ∗ Lω z✉✉✉✉✉✉✉✉✉✉✉ q∗ N uLM ω ❉❉❉❉❉❉❉❉❉❉ q∗ N p∗ M hL q∗ M p∗ N f ∗ Lω ③③③③③③③③③③ q∗ M p∗ N h−1 L ■■■■■■■■■■■ q∗ M uN Lω q∗ N p∗ Lf ∗ M ω q∗ N p∗ Lh−1 M q∗ M p∗ LhN q∗ M p∗ Lf ∗ N ω kM ■■■■■■■■■■■ ③③③③③③③③③③ q∗ Lp∗ N f ∗ M ω b❉❉❉❉❉❉❉❉❉❉ kN :✉✉✉✉✉✉✉✉✉✉✉ q∗ Lp∗ M f ∗ N ω where we can recognize the concatenation (3.3) M ωN L ◦ q∗ q∗ LωMN ◦ q∗ N ωLM of loops in A2,cl(L ×S M ×S N, n). In this figure, using kL, kM , and kN we filled in three triangles with 2-simplices which have been denoted kL, kM , and kN . The fact that (3.1) is homotopy cartesian implies that the path around the outside of (3.2) is the boundary of a filled in hexagon in the space A2,cl(L ×S M ×S N, n). Therefore we have produced a homotopy from the constant loop at 0 to the the concatenation of loops (3.3). This can be interpreted as a homotopy from the N ωLM at 0 in the space A2,cl(L ×S M ×S N, n). Because constant loop at 0 to the loop q∗ concatenation and sum agree up to homotopy this can be interpreted as a homotopy from the constant loop at 0 to the loop q∗ LωMN ◦ q∗ M ωN L ◦ q∗ M ωN L + q∗ LωMN + q∗ This gives us a path h from 0 to the corresponding element of the space A2,cl(L ×S M ×S N, n − 1). But this element is precisely the pullback of the product symplectic structure on (L×S M )×(M ×S N )×(L×S N ) to L ×S M ×S N given by the construction from [15] applied to each of the three components and described N ωLM . ✉✉✉✉✉✉✉✉✉✉✉ $■■■■■■■■■■■ v∗ M f ∗ M ω Lf ∗ v∗ Lω kL 0 b❉❉❉❉❉❉❉❉❉❉ ③③③③③③③③③③ q∗ Lp∗ N hM ③③③③③③③③③③ ❉❉❉❉❉❉❉❉❉❉ q∗ Lp∗ M h−1 N q∗ LuM N ω d■■■■■■■■■■■ ✉✉✉✉✉✉✉✉✉✉✉ v∗ N f ∗ N ω z o o $ / / / / b d / / b : in Section 2. Therefore, we have produced an isotropic structure h on the morphism L ×S M ×S N → (L ×S M ) × (M ×S N ) × (N ×S L). 5 Let and rL : L ×S M ×S N → L rM : L ×S M ×S N → M rN : L ×S M ×S N → N r : L ×S M ×S N → S be the projection morphisms. In order to show that the isotropic structure h is Lagrangian, we need to consider a pair of 4 term rows which correspond to diagram (3.1). By definition of the isotropic structure h we have a commutative diagram where the top row is the homotopy limit of the rest of the diagram: (3.4) TL×SM ×S N Θh / L(qN ,qM ,qL)[n − 2] q∗ N TL×SM ⊕ q∗ M TN ×SL ⊕ q∗ L TM ×S N Θq∗ M ωN L+q∗ L ωM N +q∗ N ωLM / (q∗ N LL×SM ⊕ q∗ M LN ×S L ⊕ q∗ L LM ×S N )[n − 1] r∗ L TL ⊕ r∗ M TM ⊕ r∗ N TN (ΘhL ,ΘhM ,ΘhN ) / (r∗ L LfL ⊕ r∗ M LfM ⊕ r∗ N LfN )[n − 1] r∗TS Θω / r∗LS[n] where Θq∗ N ωLM = (ΘR(ωLM ,hL,hM ), ΘR(ωN L,hN ,hL), ΘR(ωM N ,hM ,hN )). M ωN L+q∗ LωM N +q∗ Due to the fact that the lower three horizontal arrows are quasi-isomorphisms, the top horizontal arrow is as well. Therefore, the isotropic structure h which we constructed endows the morphism (qN , qL, qM ) with a Lagrangian structure. (cid:3) Similarly, we have the following Theorem (conjectured by Dominic Joyce) Theorem 3.2. Let m be an integer greater than equal to 2. Let S be an n-shifted symplectic derived stack and let Li → S be Lagrangian morphisms for i = 1, . . . , m. Then the morphism L1 ×S · · · ×S Lm → (L1 ×S L2) × (L2 ×S L3) × · · · × (Lm−1 ×S Lm) × (Lm ×S L1) is Lagrangian with respect to the product n − 1-shifted symplectic structure on the right hand side provided by [15]. Proof. When m ≥ 3, the proof follows precisely along the lines of the proof of Theorem 3.1. The only difference in the proof is notational, for instance, the analogue of the figure in (3.2) now must be divided into 2m triangles. When m = 2 this is just the diagonal morphism to the product of L1 ×S L2 and its opposite. It is easy to see that this is Lagrangian. (cid:3)   /     /     /   / 6 OREN BEN-BASSAT 4. Examples In this section we mention some explicit examples of Lagrangians in shifted symplectic derived stack. The idea of these Lagrangians goes back to Tyurin (see for example [21]) and was recently explained in [2] by Calaque. Tyurin's construction is in fact a holomorphic (or algebraic) version of an older construction by Casson relating to moduli of local systems. This construction became popular because of its role in Donaldson-Thomas theory ([19], [20]). Derived stacks are expected to play an important role in a deeper understanding of Donaldson -- Thomas theory, see for example [3, 4, 5, 8, 16, 7, 10, 11, 9] and the references therein. One type of examples come from Calabi-Yau 3-folds which degenerate into a pair of Fano 3-folds intersect- ing in a K3 surface. Sometimes, one passes to the stack of expanded degenerations [14]. These degenerations provide other examples of the same type. One would like to extract invariants from the derived moduli stacks of complexes of sheaves on the generic (smooth) fiber and argue that these invariants are constant in this degenerating family. Therefore, these invariants can be extracted from some derived moduli stacks of com- plexes of sheaves on the singular (special) fiber. One would like to relate these invariants to the so-called relative invariants of the Fano 3-folds relative to the K3 surface. For more details see the article [14] by J. Li and B. Wu and the references therein. The key feature of this example (discussed in [21] and [2]) is that the inclusion of the K3 surface inside the Fano (or similar manifold) induces a Lagrangian strucutre on the derived stack of perfect complexes. One can ask for which other types of morphisms of derived Artin stacks induce Lagrangian morphisms on their derived stacks of perfect complexes. This article can be used to construct many other examples of this phenomenon. For example, mapping to the homotopy colimit of (F1 a Z F2) a(F2 a F3) a(F3 a F1) Z Z (4.1) F1 `Z F2 F1 `Z F3 F2 `Z F3 F1 F3 8qqqq f▼▼▼▼▼▼▼▼▼▼▼ qqqqqq g◆◆◆◆◆◆◆◆◆◆◆◆◆ f▼▼▼▼ 8qqqqqqqqqqq ▼▼▼▼▼▼ 7♣♣♣♣♣♣♣♣♣♣♣♣♣ F2 Z gives a Lagrangian structure after passing to derived stacks of perfect complexes whenever the morphisms Z → Fi have this same property for i = 1, 2, 3. In otherwords, we only need to assume that there is a shifted symplectic structure on the stack of perfect complexes on Z and that the morphisms Z → Fi induce Lagrangian morphisms on derived stacks of perfect complexes. References [1] L. Amorim, O. Ben-Bassat, The space of shifted symplectic stacks, preprint. [2] D. Calaque, Lagrangian structures on mapping stacks and semi-classical TFTs, arXiv:1306.3235. [3] C. Brav, V. Bussi, D. Joyce, A Darboux theorem for derived schemes with shifted symplectic structure, arXiv:1305.6302. [4] C. Brav, V. Bussi, D. Dupont, D. Joyce, B. Szendroi, Symmetries and stabilization for sheaves of vanishing cycles, arXiv:1211.3259. [5] O. Ben-Bassat, C. Brav, V. Bussi, D. Joyce, A Darboux Theorem for shifted symplectic structures on derived Artin stacks, with applications, preprint. [6] R. Hartshorne, Algebraic Geometry Springer-Verlag, 1977. [7] Z. Hua, Spin structure on moduli space of sheaves on Calabi-Yau threefold, arXiv:1212.3790. [8] D. Joyce, A classical model for derived critical loci, arXiv:1304.4508. [9] Y.-H. Kiem, J. Li, A categorification of Donaldson-Thomsas Invariants Via Perverse Sheaves, arXiv:1212.6444. [10] M. Kontsevich and Y. Soibelman, Stability structures, motivic Donaldson -- Thomas invariants and cluster transformations, arXiv:0811.2435. O O 8 8 f f O O 7 g O O 7 [11] M. Kontsevich and Y. Soibelman, Cohomological Hall algebra, exponential Hodge structures and motivic Donaldson -- Thomas invariants, Commun. Number Theory Phys. 5 (2011), 231 -- 352. arXiv:1006.2706. [12] J. Lurie, Higher Algebra, available from the author's homepage. [13] J. Lurie, Higher Topos Theory, Annals of Mathematics Studies, 170, Princeton University Press, Princeton, NJ, 2009. [14] J. Li, B. Wu, Good degeneration of quot-schemes and coherent systems, arXiv:1110.0390. [15] T. Pantev, B. Toen, M. Vaqui´e, and G. Vezzosi, Shifted Symplectic Structures, Publications math´ematiques de l'IH´ES, June 2013, Volume 117, Issue 1, 271-328. [16] B. Toen, Higher and derived stacks: a global overview, Proceedings of the Seattle 2005 Algebraic Geometry meeting, AMS D. Abramovich, A. Bertram, L. Katzarkov, R. Pandharipande, and M. Thaddeus, editors. [17] B. Toen, G. Vezzosi, Caract`eres de Chern, traces ´equivariantes et g´eom´etrie algebrique d´eriv´ee, arXiv:0903.3292 [18] B. Toen, G. Vezzosi, Homotopical algebraic geometry II: Geometric stacks and applications, Mem. Amer. Math. Soc. 193 (2008) [19] R. Thomas, A holomorphic Casson invariant for Calabi-Yau 3-folds, and bundles on K3 fibrations, Journal of Differential Geometry, Vol. 54, 367-438. [20] R. Thomas, Gauge Theory on Calabi-Yau manifolds, University of Oxford thesis, available from the author's home page. [21] A. Tyurin, Vector Bundles, edited by F. Bogomolov, A. Gorodentsev, V. Pidstrigach, M. Reid and N. Tyurin, published by Universitatsverlag Gottingen, 2008. Mathematical Institute, Radcliffe Observatory Quarter, Woodstock Road, Oxford, OX2 6GG, UK E-mail address: [email protected]
1006.5248
4
1006
2011-06-21T02:42:54
Syzygies of Segre embeddings and Delta-modules
[ "math.AG" ]
We study syzygies of the Segre embedding of P(V_1) x ... x P(V_n), and prove two finiteness results. First, for fixed p but varying n and V_i, there is a finite list of "master p-syzygies" from which all other p-syzygies can be derived by simple substitutions. Second, we define a power series f_p with coefficients in something like the Schur algebra, which contains essentially all the information of p-syzygies of Segre embeddings (for all n and V_i), and show that it is a rational function. The list of master p-syzygies and the numerator and denominator of f_p can be computed algorithmically (in theory). The central observation of this paper is that by considering all Segre embeddings at once (i.e., letting n and the V_i vary) certain structure on the space of p-syzygies emerges. We formalize this structure in the concept of a Delta-module. Many of our results on syzygies are specializations of general results on Delta-modules that we establish. Our theory also applies to certain other families of varieties, such as tangent and secant varieties of Segre embeddings.
math.AG
math
SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES ANDREW SNOWDEN Abstract. We study syzygies of the Segre embedding of P(V1) × · · · × P(Vn), and prove two finiteness results. First, for fixed p but varying n and Vi, there is a finite list of "master p-syzygies" from which all other p-syzygies can be derived by simple substitutions. Second, we define a power series fp with coefficients in something like the Schur algebra, which contains essentially all the information of p-syzygies of Segre embeddings (for all n and Vi), and show that it is a rational function. The list of master p-syzygies and the numerator and denominator of fp can be computed algorithmically (in theory). The central observation of this paper is that by considering all Segre embeddings at once (i.e., letting n and the Vi vary) certain structure on the space of p-syzygies emerges. We formalize this structure in the concept of a ∆-module. Many of our results on syzygies are specializations of general results on ∆-modules that we establish. Our theory also applies to certain other families of varieties, such as tangent and secant varieties of Segre embeddings. 1 1 0 2 n u J 1 2 ] . G A h t a m [ 4 v 8 4 2 5 . 6 0 0 1 : v i X r a Contents Introduction 1. 2. ∆-modules 3. Hilbert series 4. Syzygies of ∆-schemes 5. Application to Segre varieties 6. Questions and problems References 1 7 15 26 29 34 34 1. Introduction Let V1, . . . , Vn be finite dimensional complex vector spaces. The Segre embedding is the map P(V1) × ··· × P(Vn) → P(V1 ⊗ ··· ⊗ Vn) taking a tuple of lines to their tensor product. Despite its fundamental nature, the syzygies of this embedding are poorly understood (excluding some special cases, such as when n = 2). The purpose of this paper is to establish a general structural theory of these syzygies. We show that for p fixed, but n and the Vi variable, the collection of all p-syzygies can be given an algebraic structure, and that this structure has favorable finiteness properties. One consequence of this result is that almost any question about p-syzygies -- such as: is the module of 13-syzygies of every Segre embedding generated in degrees at most 20? -- can be resolved by a finite computation (in theory). In the rest of the introduction, we describe this structure and present our two main theorems about it. As the precise statements of these theorems are somewhat technical, we begin with an informal discussion that attempts to elucidate the essential content of the first theorem. Date: June 20, 2011. The author was partially supported by NSF fellowship DMS-0902661. 1 2 ANDREW SNOWDEN 1.1. The first theorem (informal). The equations of the Segre embedding have been known for as long as the embeddings themselves have been known, and are familiar to most anyone who has studied algebraic geometry. We briefly recall them. First suppose that n = 2, i.e., consider the embedding of P(V1) × P(V2) into P(V1 ⊗ V2). Let {xi}i∈I1 be coordinates on P(V1), let {xj}j∈I2 be coordinates on P(V2) and let {xij}(i,j)∈I1×I2 be coordinates on P(V1 ⊗ V2). Then the equations xi1j1xi2j2 − xi1j2xi2j1 = 0 cut out P(V1) × P(V2) inside P(V1 ⊗ V2). Now suppose that n = 3. Then, in the obvious notation, we have the equation xi1j1k1xi2j2k2 − xi1j1k2xi2j2k1 = 0. This is not symmetrical in the i, j and k indices, and so we get two other families of equations by changing the roles of the indices. The resulting collection of equations cuts out P(V1) × P(V2) × P(V3) inside of P(V1 ⊗ V2 ⊗ V3). The situation for n > 3 is similar. The equations we have just described are quite similar to each other. Indeed, they can all be recovered from the single "master equation" x12x34 − x14x32 = 0 by certain substitution rules. When n = 2 we obtain all the equations by replacing the numbers 1 and 3 (resp. 2 and 4) in the master equation by elements of I1 (resp. I2). When n = 3 we obtain all the equations by replacing the numbers 1 and 3 by elements of I1 × I2 and 2 and 4 by elements of I3, or by doing this with the roles of I1, I2 and I3 interchanged. For general n, we obtain the equations for P(V1)×···× P(Vn) by partitioning the set {1, . . . , n} into two disjoint subsets A and B and then replacing 1 and 3 by elements of Qk∈A Ik and 2 and 4 by elements of Qk∈B Ik. We now examine the syzygies between equations (2-syzygies). We again begin with the case n = 2. Consider the determinant xi1j1 xi1j2 xi1j3 xi1j1 xi1j2 xi1j3 xi2j1 xi2j2 xi2j3 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) The first two rows are repeated, and so the determinant vanishes when considered as a polynomial in the variables xij. Now, the 2× 2 minors along the bottom are exactly the previously constructed equations. Thus by taking the Laplace expansion of the determinant along the top row we obtain a linear relation between equations, i.e., a syzygy. This construction can be modified to produce 2-syzygies for n > 2, in the same way that the n = 2 equations were modified to get equations for n > 2. These syzygies generate the module of 2-syzygies, for any n. As was the case for equations, all the 2-syzygies just produced can be recovered from a "master syzygy" by certain substitution rules. The master syzygy is just the one obtained from the Laplace expansion of (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) The substitution rules are similar: partition {1, . . . , n} into two sets A and B and replace 1 and 2 by elements of Qk∈A Ik and 3, 4 and 5 by elements of Qk∈B Ik. Given these observations, one might ask: are there master higher syzygies? An affirmative x13 x14 x15 x13 x14 x15 x23 x24 x25 answer is provided by the following result, the first of our two main theorems: (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Theorem A (Informal version). Given p, there is a finite list of "master p-syzygies" from which all p-syzygies can be obtained through the substitution procedures sketched above. This result is slightly different than what we have seen for equations and syzygies of equations in that there is no longer a single master syzygy, but rather a finite list of them. The theorem does not provide any canonical list of master syzygies, it just asserts the existence of such a list. The proof of the theorem can easily be adapted to yield an algorithm which, given p, produces a SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 3 list of master p-syzygies. Unfortunately, this algorithm is fantastically impractical, even for small p -- it involves linear algebra over a polynomial ring in p2p variables! Note that the substitution rules do not change the general form of the syzygy (e.g., all equations are the difference of two monomials involving two variables), and so the above theorem implies that there are only finitely many "forms" of syzygies. 1.2. The first theorem. We now give a more formal version of the above discussion. We begin by explaining precisely what the "master equation" is in coordinate-free language. Once this is accomplished, generalizing to higher syzygies will be straightforward. Let In(V1, . . . , Vn) be the ideal defining the Segre embedding of P(V1) × ··· × P(Vn). The n subscript simply indicates that we are dealing with n vector spaces; it is redundant but provides clarity. The ideal is generated by its degree two piece, so we focus on it; we denote it by I (2) n (V1, . . . , Vn). We now observe that I (2) has the following pieces of structure: (A1) Given maps Vi → V ′ (A2) Given σ ∈ Sn, there is a natural map σ∗ : I (2) i , there is a natural map I (2) n (V1, . . . , Vn) → I (2) n (V1, . . . , Vn) → I (2) n (Vσ(1), . . . , Vσ(n)). In other n (V ′ 1, . . . , V ′ words, I (2) n n). In other is a functor. n words, the functor I (2) n (A3) There is a natural map is Sn-equivariant. n (V1, . . . , Vn−1, Vn ⊗ Vn+1) → I (2) I (2) n+1(V1, . . . , Vn−1, Vn, Vn+1). Geometrically, this map comes from the commutative diagram of Segre embeddings: P(V1) × ··· × P(Vn−1) × P(Vn) × P(Vn+1) Y Y Y Y P(V1) × ··· × P(Vn−1) × P(Vn ⊗ Vn+1) Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y ,Y P(V1 ⊗ ··· ⊗ Vn−1 ⊗ Vn ⊗ Vn−1) Note that (A3) seems to favor the nth parameter of In; however, due to (A2) symmetry is not actually broken. The Segre embedding of P1 × P1 into P3 is defined by a single quadratic equation, and so the space I (2) 2 (C2, C2) is one dimensional. The "master equation" is simply a non-zero element of this space. As we now explain, any quadratic equation for any Segre embedding can be obtained from the master equation through the operations (A1) -- (A3). Let {x1, x3} be a basis for the first C2 and {x2, x4} a basis for the second C2, so that {x12, x14, x32, x34} is a basis for C2 ⊗ C2. Then the quadratic polynomial (1) x12x34 − x14x32 is a non-zero element of I (2) 2 (C2, C2), and can be taken as the master equation. Let V1, . . . , Vn be vector spaces and let {x(k) j }j∈Ik be a basis for Vk. Let A∐ B be a partition of {1, . . . , n}, let α and γ be elements of Qk∈A Ik and let β and δ be elements of Qk∈B Ik. Then, in the obvious notation, (2) xαβxγδ − xαδxγβ is an element of I (2) map n (V1, . . . , Vn), and such elements span I (2) n (V1, . . . , Vn). Consider the composite (3) I (2) 2 (C2, C2) → I (2) 2 (cid:18)Ok∈A Vk,Ok∈B Vk(cid:19) → I (2) n (V1, . . . , Vn). / / ,   4 ANDREW SNOWDEN The first map in (3) is the one coming form (A1) applied to the maps f : C2 →Ok∈A g : C2 → Ok∈B Vk, f (x1) = xα, f (x3) = xγ Vk, g(x2) = xβ, g(x4) = xδ. The second map in (3) comes from repeated applications of (A2) and (A3). The composite map in (3) simply takes an equation in the variables x12, x14, x32, x34 and changes the 1 and 3 to α and γ and the 2 and 4 to β and δ. Thus the operations (A1) -- (A3) truly do provide a coordinate-free replacement for the substitution rules of §1.1. In particular, the master equation (1) is taken to the equation (2). This shows that any quadratic equation can be obtained from the master equation through the operations (A1) -- (A3). mild technical condition. Here, Vec denotes the category of complex vector spaces. As stated, generalizing the above discussion on equations to higher syzygies is straightforward. However, it will be convenient to first introduce a definition. A ∆-module is an object F consisting of: (B1) For each non-negative integer n, an Sn-equivariant functor Fn : Vecn → Vec satisfying a (B2) For each n, a natural transformation Fn(V1, . . . , Vn−1, Vn ⊗ Vn+1) → Fn+1(V1, . . . , Vn+1). There are certain conditions placed on the maps in (B2) which we do not state here. It is clear how the above definition relates to the previous discussion: I (2) is a ∆-module. (Of course, (B1) takes the place of both (A1) and (A2), while (B2) corresponds to (A3).) However, this definition is inelegant, and thus technically inconvenient. An equivalent (but much nicer) definition is the following. Let Vec∆ be the following category: We think of (V, L) as a family of vector spaces. • An object is a pair (V, L) consisting of a finite set L and, for each x ∈ L, a vector space Vx. • A morphism (V, L) → (V ′, L′) consists of a surjection L′ → L together with a linear map Vx →Ny7→x V ′ A ∆-module is then simply a functor F : Vec∆ → Vec (satisfying a mild technical condition). The functor Fn in the previous formulation is obtained by restricting F to families with n elements. We now precisely define our spaces of syzygies and explain how they admit the structure of a ∆-module. Let (V, L) be a family of vector spaces. Let R(V, L) be the projective coordinate ring y , for each x ∈ L. of Qx∈L P(Vx) and let P (V, L) be the projective coordinate ring of P(Nx∈L Vx). Precisely, ∞ P (V, L) = P (n)(V, L), ∞ Mn=0 Mn=0 R(V, L) = R(n)(V, L), Vx(cid:19) P (n)(V, L) = Symn(cid:18)Ox∈L R(n)(V, L) =Ox∈L Symn(Vx). The Segre embedding corresponds to a surjection of rings P (V, L) → R(V, L). Put Fp(V, L) = TorP (V,L) p (R(V, L), C), where C is given the structure of a P (V, L)-module by letting all positive degree elements act by 0. We call Fp(V, L) the space of p-syzygies. This terminology is justified for the following reason: if ··· M1 → M0 → R(V, L) → 0 is a minimal free resolution of R(V, L) by finite free P (V, L)-modules, then Mp is isomorphic to P (V, L) ⊗C Fp(V, L). (This is a general fact about minimal free resolutions and has nothing to do with the specifics of the Segre embedding.) The functoriality of Fp with respect to maps in Vec∆ stems from a corresponding functoriality of R and P (as mentioned in the discussion of the SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 5 properties (A1) -- (A3)), and the basic properties of Tor. Thus each Fp is a ∆-module. Note that F1 is identified with I (2). Before precisely stating our first theorem, we must make one more definition. Let F be a ∆- module. An element of F is an element of F (V, L) for some (V, L). Given a set S of elements of F , there is a minimal ∆-submodule F ′ of F containing S. We call F ′ the submodule of F generated by S. We say that F is finitely generated if it is generated by some finite set of elements. We now come to our first main theorem: Theorem A. The ∆-module Fp is finitely generated. Of course, the "master syzygies" are just generators for Fp. As previously mentioned, our proof of this theorem provides an algorithm to compute a set of generators. 1.3. The second theorem. Theorem A tells us that all p-syzygies can be derived from a finite list of p-syzygies via some simple operations. However, this by itself tells us very little about the structure of the space Fp(V, L) for particular values of (V, L). Our second theorem attempts to close this gap. The full version of the theorem requires some set-up, so we begin by stating a simple, elementary version of it: Theorem B (Preliminary version). Fix non-negative integers p and d. Then the series is a rational function. ∞ Xn=1 dim Fp,n(Cd, . . . , Cd) · tn The series in the above theorem counts the number of p-syzygies of the Segre embedding of (Pd−1)n. Thus the result states that these numbers vary in a predictable manner with n. The above preliminary version of the theorem falls short of what we would like, in three ways. First, it only deals with syzygies of (Pd−1)n, rather than more general products. Second, it only gives us the dimension of Fp,n, rather than the dimensions of its graded pieces. And third, and most importantly, it gives no information on the action of the general linear groups on the spaces of syzygies. The full version of the theorem addresses all of these issues. Before properly stating the theorem, we must introduce a generating function. For a partition λ, write Sλ for the corresponding Schur functor. (See §2.1 for a review of this theory.) By general theory, we have a decomposition Fp,n(V1, . . . , Vn) = Mi∈Ip,n Sλ1,i (V1) ⊗ ··· ⊗ Sλn,i(Vn) for some index set Ip,n and partitions λi,j (depending on n and p). It is not difficult to show that Ip,n is finite. The above decomposition holds as functors Vecn → Vec, i.e, the set Ip,n and the partitions λi,j do not depend on (V1, . . . , Vn). Define f ∗ p,n = Xi∈Ip,n sλ1,i ··· sλn,i, regarded as a polynomial in (commuting) formal variables sλ. The functor Fp,n can be recovered from f ∗ p,n -- allowing the variables to commute does not lose information since the functor Fp,n has an Sn-equivariance. (Note, however, that the data of the Sn-equivariance is not recorded in f ∗ p,n.) Now define ∞ a power series in the variables sλ. A priori, there could be infinitely many variables occurring in f ∗ p ; we will show that this is not the case. Our second theorem is then: Xn=1 f ∗ p = f ∗ p,n, 6 ANDREW SNOWDEN Theorem B. The series f ∗ p is a rational function of the sλ. an example to illustrate how to extract information from f ∗ obtain This theorem completely encompasses the preliminary version, but is much stronger. Let us give p . From the computations of §5.5, we 1 1 − s (1 − s)2 − w2 − 1 − s f ∗ 1 = , where s = s(2) and w = s(1,1). Developing the above into a power series, we obtain f ∗ 1 = w2 + 3sw2 + (6w2s2 + w4) + ··· The second order term of this series tells us that F1,2(V1, V2) is isomorphic to V2V1 ⊗V2V2. The third order terms tells us that F1,3(V1, V2, V3) is isomorphic to (cid:0) Sym2 V1 ⊗V2V2 ⊗V2V3(cid:1) ⊕(cid:0)V2V1 ⊗ Sym2 V2 ⊗V2V3(cid:1) ⊕(cid:0)V2V1 ⊗V2V2 ⊗ Sym2 V3(cid:1). In general, we can recover the isomorphism class of Fp,n(V1, . . . , Vn) as a representation of GL(V1)× ··· × GL(Vn) from the order n term of f ∗ p,n. In particular, we can recover the dimensions of the graded pieces of Fp,n(V1, . . . , Vn) from f ∗ p,n, as these pieces can be detected by the group action. As with Theorem A, the proof of Theorem B provides an algorithm to compute the numerator and denominator of f ∗ p . Thus, essentially all information about all p-syzygies of Segre embeddings can be determined from a single straightforward calculation. Unfortunately, this algorithm has complexity comparable to the previous one and is worthless in practice. The leading term of f ∗ p is known by Lascoux's work (§5.2). We will compute a certain Euler characteristic involving the f ∗ p 's (§5.4). By known vanishing results, this allows us to compute f ∗ 1 , 2 , f ∗ f ∗ 4 (§5.5). In particular, this gives an essentially complete description of the module of p-syzygies for p = 1, 2, 3. We have not been able to compute beyond these examples, however. 3 and part of f ∗ 1.4. Syzygies of ∆-schemes. Given a finite family of vector spaces (V1, . . . , Vn) we have the Segre variety Xn(V1, . . . , Vn) ⊂ P(V1 ⊗ ··· ⊗ Vn). The variety Xn is functorial in the Vi's and Sn-equivariant. Furthermore, we have inclusions (4) Xn+1(V1, . . . , Vn+1) ⊂ Xn(V1, . . . , Vn−1, Vn ⊗ Vn+1). These are in fact the only properties of the Segre embedding we need for our theory to apply, as we now explain. A ∆-scheme X is a rule assigning to each finite family of vector spaces (V1, . . . , Vn) a closed subscheme Xn(V1, . . . , Vn) of P(V1 ⊗ ··· ⊗ Vn) which is functorial, Sn-equivariant and such that the inclusion (4) holds. The Segre embeddings constitute an example of a ∆-scheme, but there are many others: for instance, the secant varieties of the Segre are ∆-schemes. Let X be a ∆-scheme. Let Fp(V, L) be the space of p-syzygies of the embedding X(V, L) → P(Nx∈L Vx). Then Fp forms a ∆-module. Let F (d) p )(d) denote the degree d piece of Fp and let (f ∗ denote the series defined from F (d) . Then we have the following result: p p Theorem C. The ∆-module F (d) p is finitely generated, and the series (f ∗ p )(d) is rational. This result is slightly weaker than the two main theorems for Segre embeddings, since it only applies to the graded pieces of Fp. To get comparable results, one would need to know that Fp is supported in only finitely many degrees. This is true in the case of Segres, but probably not true for a general ∆-scheme. In §4.5 we define a class of ∆-schemes, those of finite level, for which it seems plausible that Fp is supported in finitely many degrees. SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 7 1.5. Outline of proofs. We begin with a general study of ∆-modules. The heart of the paper is occupied with the proof of two theorems: (a) a small ∆-module is noetherian; and (b) the Hilbert series of a small ∆-module is rational. "Small" ∆-modules form a certain subclass of ∆- modules. The class of small ∆-modules is closed under subquotients. Theorem C is then a formal consequence of (a) and (b) and the following observation: there is a complex of small ∆-modules (a Koszul complex) whose homology is Fp. The two stronger theorems for Segre embeddings are obtained by applying the well-known result that the p-syzygies of the Segre embedding are supported in only finitely many degrees. We have not been able to find a direct method of studying ∆-modules. Instead, we access them through the following "ladder": {graded modules} ↔ {modules in Sym(Vec)} ↔ {modules in Sym(S)} ↔ {∆-modules} We define the middle two categories below. The arrows here do not mean anything precise, only that the two categories are related. We prove our results about ∆-modules by starting with the cor- responding results for graded modules (which are easy and well-known) and moving up the ladder. This process is actually fairly explicit: for example, the Hilbert series of a ∆-module is identified with the (G-equivariant) Hilbert series of a module over a graded ring, which is constructed in an explicit manner from the ∆-module. This identification is the reason that the Hilbert series of a ∆-module -- at least one which is described in a reasonable manner -- is algorithmically computable. 1.6. Future directions. We prove a number of purely algebraic results (about ∆-module, twisted commutative algebras, etc.) which seem to scratch the surface of a much larger theory. We hope to pursue some of this theory in the future. We also hope to refine our results on syzygies -- for instance, what can one say about the rational functions f ∗ p ? A list of some questions and problems along these lines appears in §6. Some of the results and methods we develop seem to apply to slightly different settings. For example, if X is a projective variety with an action of a reductive group G, then the syzygies of the GIT quotients X n//G (as n varies) have some of the algebraic structure that we observe here for the Segre varieties. One can hope that there are finiteness results in this setting. We will return to this topic in a future paper. Acknowledgments. I would like to thank Aldo Conca, Tony Geramita, Ben Howard, Sarah Kitchen, Rob Lazarsfeld, Diane Maclagan, John Millson, Lawrence O'Neil, Steven Sam and Ravi Vakil for helpful discussions. In particular, I would like to thank Sarah Kitchen, Rob Lazarsfeld and Ravi Vakil for their comments on drafts of this paper. 2. ∆-modules The purpose of this section is to introduce the algebraic objects that we will use in the rest of the paper, most notably ∆-modules. We begin by quickly reviewing the theory of Schur functors and define the Schur algebra S, which we regard as a category. We then discuss symmetric powers of semi-simple abelian categories. This operation is not strictly necessary for our purposes, but clarifies some later definitions. We then introduce the categories Sym(Vec) and Sym(S) and give explicit models for them which do not use the symmetric power construction. Finally we discuss ∆-modules, which are objects of Sym(S) with an additional piece of structure. The most important results of this section are various statements that certain types of objects are noetherian. Algebras in Sym(Vec) are known as "twisted commutative algebras," and there is some literature on them (see [B], [J, Ch. 4], [GS]); closely related is Joyal's theory of tensorial species. However, the existing literature -- at least that which we are aware of -- studies these objects from a perspective different from our own. In particular, we have not encountered the noetherian result 8 ANDREW SNOWDEN we prove about them in the literature. The categories Sym(S) and Mod∆ are a bit more esoteric, and we do not know of any occurrence of them in the literature. 2.1. The Schur algebra. We quickly review what we need about the Schur algebra. Further discussion can be found in [FH, §6.1]. Let Vec denote the category of complex vector spaces. A functor Vec → Vec is "nice" if it appears as a constituent of a functor of the form V 7→ V ⊗n, or a (possibly infinite) direct sum of such functors. Essentially every functor Vec → Vec one encounters naturally that does not use duality is nice; an example of a non-nice functor is the double dual. Let S denote the category of all nice functors. It is a semi-simple abelian tensor category which we call the Schur algebra. (By "semi-simple" here we mean that every object is a possibly infinite direct sum of simple objects.) The tensor product is the point-wise one: (F ⊗ G)(V ) = F (V ) ⊗ G(V ). Let λ be a partition of n; we denote this by λ = n or λ ⊢ n. Let Mλ be the irreducible complex representation of the symmetric group Sn associated to λ. For a vector space V put Sλ(V ) = (V ⊗n⊗Mλ)Sn, where the subscript denotes coinvariants and Sn acts on V ⊗n by permuting tensor factors. Then Sλ belongs to S and is a simple object; furthermore, every simple object is isomorphic to Sλ for a unique λ. We use the convention that the partition λ = (n) gives the functor Sλ = Symn while the partition λ = (1, . . . , 1) gives the functor Sλ = Vn. We identify partitions with Young diagrams by the convention that λ = (n) has one row and λ = (1, . . . , 1) has n rows. We will also need multivariate functors of vector spaces. A functor Veck → Vec is "nice" if it appears as a constituent of a functor of the form (V1, . . . , Vk) 7→ V ⊗n1 , or a (possibly infinite) direct sum of such functors. The category of all nice functors is again a semi-simple abelian tensor category, and is naturally equivalent to S ⊗k. This means that any nice functor F : Veck → Vec can be written as 1 ⊗ ··· ⊗ V ⊗nk k F (V1, . . . , Vk) =Mi∈I Sλ1,i(V1) ⊗ ··· ⊗ Sλk,i(Vk) for some index set I and partitions λi,j. This expression is unique. We say that a partition λ occurs in F if it is amongst the λi,j. If F : Vec → Vec is a nice functor then the functors Vec2 → Vec given by mapping (V, W ) to F (V ⊕ W ) and F (V ⊗ W ) are both nice. We thus get a co-addition map a∗ and a co-multiplication map m∗ from S to S ⊗2. Let A be a C-linear abelian tensor category. (Our tensor categories are always symmetric, i.e., there is a given involution of functors A ⊗ B → B ⊗ A.) There is then a natural action of S on A. Given a partition λ of n and an object A of A, the object Sλ(A) is given by (A⊗n ⊗ Mλ)Sn. 2.2. Symmetric powers of abelian categories. We now describe how one can form the sym- metric algebra on semi-simple abelian categories. The reader who is not comfortable with this discussion need not worry: all categories that we eventually use will admit concrete descriptions. We provide this discussion only to give some context and motivation for later constructions. Let A be a semi-simple C-linear abelian category. One can then make sense of the tensor power A⊗n (see [D, §5] for a discussion of tensor products of abelian categories). We define Symn(A) to be the category (A⊗n)Sn, where the superscript denotes homotopy invariants (equivariant objects). Thus an object of Symn(A) is an object A of A⊗n together with an isomorphism σ∗A → A for each σ ∈ Sn, satisfying the obvious compatibility conditions. We define Sym(A) to be the sum of the categories Symn(A) over n ≥ 0. The category Sym(A) has a natural tensor structure. Multiplication involves averaging (induc- tion). Precisely, let ⊗ denote the usual concatenation tensor product A⊗n ⊗ A⊗m → A⊗(n+m). If A is an object of (A⊗n)Sn and B of (A⊗m)Sm then A⊗ B naturally has an Sn × Sm equivariance. One can then form the induction IndSn+m (A⊗ B), which is an object of (A⊗(n+m))Sn+m. This is Sn×Sm the product of A and B in Sym(A), which we denote by A ⊗ B. Note that if A itself has a tensor structure then (A⊗n)Sn does as well, which gives an alternate tensor structure on Sym(A). When present, we will denote this tensor product by ⊠ and call it the point-wise tensor product. SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 9 Remark 2.1. The above definition looks more like the divided power algebra than symmetric alge- bra. However, one can verify it has the correct universal property. We believe that in the setting of C-linear abelian tensor categories, the right analogue of divided powers is an action of the Schur algebra, with Symn taking the place of γn. Since all such categories have a natural action of the Schur algebra, they can all be considered to have divided powers. Thus the symmetric and divided power algebras on A coincide. This is a conceptual reason explaining why we can use invariants (rather than coinvariants) to form the symmetric algebra. As Sym(A) is an abelian tensor category, we can speak of algebras in it and modules over algebras. (Algebras will always be commutative, associative and unital.) Let A be an algebra. We say that A is finitely generated (as an algebra) if it is a quotient of Sym(F ) for some finite length object F of Sym(A). Similarly, we say that an A-module M is finitely generated if it is a quotient of A ⊗ F for some finite length F . We say that M is noetherian if every ascending chain of submodules stabilizes. We say that A is noetherian (as an algebra) if every finitely generated A-module is noetherian. The category Sym(A) is again semi-simple abelian, and its simple objects can be described easily. Let A be a simple object of A and let Mλ be an irreducible representation of Sn. Then Mλ ⊗ A ⊗ n has a natural Sn-equivariance and so defines an object of Symn(A); in fact, this is nothing other than Sλ(A) in the category Sym(A). This object is simple, and all simple objects are of the form N Sλi(Ai) where the Ai are mutually non-isomorphic simple objects of A. Let K(−) denote the Grothendieck group of an abelian category, tensored with Q. We have a map K(Symn(A)) → Symn(K(A)) defined as follows: first apply the functor Symn(A) → A⊗n, then use the identification K(A⊗n) = K(A)⊗n, then project K(A)⊗n → Symn(K(A)) and finally divide by n!. For A ∈ Symn(A) we let [A] denote the corresponding class in Symn(K(A)). We also put [A]∗ = n![A]. The above map if [A] = [B] then A and B have is not injective, but only forgets the Sn-equivariant structure: isomorphic images in A⊗n. Summing the above maps over n, we get a map K(Sym(A)) → Sym(K(A)). Again, for A ∈ Sym(A) we let [A] denote the corresponding class in Sym(K(A)); we define [A]∗ in the obvious manner. We have [A ⊗ B] = [A][B]; the modified class [−]∗ does not satisfy this. As an example, if A is an object of A then [Symn(A)] = 1 n! [A]n (which leads to the beautiful formula [Sym(A)] = exp([A])). We will often need to use the completion of Sym, both in the setting of Z-modules and abelian categories. Whereas Sym(Zn) is the ring of polynomials in n variables, the completion is the ring of power series in n variables. We will not bother to introduce extra notation for the completion, as there should be no confusion. 2.3. The category Sym(Vec). The following abelian tensor categories are equivalent: (a) The category Sym(Vec). (b) The category of functors (fs) → Vec, where (fs) is the category whose objects are finite sets and whose morphisms are bijections. Multiplication is given by convolution, using the monoidal structure on (fs) given by disjoint union. Precisely, if V and W are two functors (fs) → Vec then (V ⊗ W )L = ML=A∐B VA ⊗ WB, where the sum is over all partitions of L into two disjoint subsets A and B. subscripts to indicate the value of the functor on a set). (We use 10 ANDREW SNOWDEN (c) The category of sequences (Vn)n≥0, where Vn is a vector space with an action of Sn. Mul- tiplication is given by the formula (V ⊗ W )n = Mn=i+j IndSn Si×Sj (Vi ⊗ Wj). (d) The Schur algebra S. Multiplication is the point-wise tensor product. (e) The full subcategory of the representation category of GL(∞) on objects which appear as a constituent of a direct sum of tensor powers of the standard representation C∞. Here GL(∞) is the union of GL(n, C) over n ≥ 1. Multiplication is the usual tensor product of representations. We briefly describe the various equivalences. The equivalence between (b) and (c) is clear. Since Vec⊗n = Vec, the category of Sn-equivariant objects in Vec⊗n is just the representation category of Sn; this gives the equivalence between (a) and (c). The equivalence between (c) and (d) is through Schur-Weyl duality. Precisely, given a sequence (Vn) in the category (c), let S : Vec → Vec be the functor taking a vector space W to S(W ) =Mn≥0 (W ⊗n ⊗ Vn)Sn. Then (Vn) 7→ S is the equivalence. Regarding Mλ as an object of (c) supported at n = λ, this equivalence maps Mλ to Sλ. Finally, the equivalence of (d) and (e) is given by evaluating a Schur functor on C∞. We regard (b) -- (e) as "models" for the category Sym(Vec). We name them the "standard," "sequence," "Schur," and "GL" models, respectively. (There are even more models for this category, but they are not relevant for our purposes.) We now come to an important definition: Definition 2.2. A twisted commutative algebra is an algebra in the category Sym(Vec). As always, "algebra" means commutative, associative and unital. In the standard model, a twisted commutative algebra is a functor A : (fs) → Vec together with a multiplication map AL ⊗ AL′ → AL∐L′ satisfying the appropriate identity. In the GL-model, a twisted commutative algebra is just a commutative C-algebra, in the usual sense, with an action of GL(∞) by algebra homomorphisms (under which the algebra forms an appropriate kind of representation). The notion of a module over a twisted commutative algebra is evident. Let V be an object of Sym(Vec), in the standard model. By an element of V we mean an element of VL for some L; we then call #L its order. (It would be more natural to use the term "degree" here, but we want to reserve that term for future use.) Given a collection S of elements of V there is a unique minimal subobject of V containing S; we call it the subspace of V generated by S. Similarly, if A is a twisted commutative algebra and S is a collection of elements of A then one can speak of the subalgebra of A generated by S. One can do the same for modules over A. We therefore have a notion of "finitely generated" for algebras and modules; this agrees with the one defined in §2.2 using finite length objects. Let U be a vector space. We let Uh1i be the object of Sym(Vec) which is U in order 1 and 0 in other orders. We put A = Sym(Uh1i); this is the most important twisted commutative algebra in this paper. Here are precise descriptions of Uh1i and A in the various models: • In the standard model, Uh1i assigns to a finite set L the vector space U if #L = 1 and the vector space 0 otherwise. The algebra A assigns to a finite set L the vector space U ⊗L. The multiplication map AL ⊗ AL′ → AL∐L′ is concatenation of tensors. (Note: U ⊗L is isomorphic as a vector space to U ⊗n, where n = #L, but is functorial in L. It can be defined as the universal vector space equipped with a multi-linear map from U × L. We think of the factors of a pure tensor in U ⊗L as being labeled with elements of L.) • In the sequence model, Uh1i is the vector space U in order 1 and 0 in other orders. The algebra A is given by An = U ⊗n, with its usual Sn action; the multiplication map Ai⊗Aj → Ai+j is again concatenation of tensors. SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 11 • In the Schur model, Uh1i is the functor U⊗Sym1. The algebra A is given by Sym(U⊗Sym1). If V is a vector space then (Uh1i)(V ) = U ⊗ V and A(V ) = Sym(U ⊗ V ). • In the GL-model, Uh1i is U ⊗ C∞. The algebra A is Sym(U ⊗ C∞). Thus, in this model, A is a polynomial ring in infinitely many variables with GL(∞) acting by linear substitutions. The following result underlies everything else in the paper. We learned the key ideas of its proof from Harm Derksen and Ben Howard. Theorem 2.3. A twisted commutative algebra finitely generated in order 1 is noetherian. Proof. A twisted commutative algebra which is finitely generated in order 1 is a quotient of Sym(Uh1i) for some finite dimensional vector space U . Thus it suffices to show that such algebras are noetherian. Fix U and put A = Sym(Uh1i). In the GL-model, A is given by Sym(U ⊗C∞). Let d = dim U . The following lemma shows that contraction from GL(∞)-stable ideals of Sym(U⊗C∞) to ideals of Sym(U ⊗ Cd) is injective. As Sym(U ⊗ Cd) is a polynomial algebra in finitely many variables, it is noetherian. We conclude that A is a noetherian module over itself. A slight modi- fication of this argument shows that any finitely generated A-module is noetherian, which proves that A is noetherian as an algebra. (cid:3) Lemma 2.4 (Weyl). Let U be a vector space of dimension d. GL(∞)-stable subspace then W is generated as a GL(∞)-module by W ∩ Symk(U ⊗ Cd). Proof. Using the formula for the symmetric power of a tensor product (see [FH, Exercise 6.11(b)]), we obtain a diagram If W ⊂ Symk(U ⊗ C∞) is a Symk(U ⊗ Cd) Symk(U ⊗ C∞) L Sλ(U ) ⊗ Sλ(Cd) L Sλ(U ) ⊗ Sλ(C∞) The sums are taken over the partitions λ of k. Now, let W be a GL(∞)-stable subspace of Symk(U ⊗ C∞). Since the Sλ(C∞) are mutually non-isomorphic irreducible GL(∞)-modules, we can write W =L Wλ ⊗ Sλ(C∞), where Wλ is a subspace of Sλ(U ). Note that if λ has more than d rows then Sλ(U ) = 0 and so Wλ = 0. The space W ∩ Symk(U ⊗ Cd) is equal to L Wλ ⊗ Sλ(Cd). Since Sλ(Cd) generates Sλ(C∞) whenever λ has at most d rows, we find that W ∩ Symk(U ⊗ Cd) generates W . (cid:3) Any twisted commutative algebra finitely generated in order 1 has the property that the parti- tions appearing in it have a bounded number of rows. Most of the results we prove about algebras generated in order 1, such as the above theorem, can easily be extended to the larger class of alge- bras which have a bounded number of rows. Going beyond this class of algebras is harder though. For instance, we do not know how much the order 1 hypothesis in Theorem 2.3 can be relaxed; see §6. We need one more general finiteness result about twisted commutative algebras. Proposition 2.5. Let A be a noetherian twisted commutative algebra on which a reductive group G acts. Then AG is noetherian. If M is a finitely generated A-module with a compatible action of G then M G is a finitely generated AG-module. Proof. The usual proof works. To show how it carries over, we prove that AG is noetherian. We must therefore show that AG ⊗ F is a noetherian AG-module for any finite length object F of Sym(Vec). We have an inclusion AG⊗ F ⊂ A⊗ F . Let G act on A⊗ F by acting trivially on F . Let M be an AG-submodule of AG ⊗ F and let M ′ be the A-submodule of A⊗ F it generates. Let y be an element of M ′ which is G-invariant. We can then write y =P aixi where the ai are elements of A and the xi are elements of M . As y and the xi are invariant, we have y = avg(y) =P avg(ai)xi. Thus y belongs to M . This shows that M = (M ′)G. Therefore, the map {AG-submodules of AG ⊗ F} → {A-submodules of A ⊗ F}  _    _   12 ANDREW SNOWDEN which takes an AG-submodule to the A-submodule it generates is inclusion preserving and injective. As any ascending chain on the right side stabilizes, so too does any ascending chain on the left side. This proves that AG is noetherian. We leave the statement about modules to the reader. (cid:3) 2.4. The category Sym(S). We now investigate the category Sym(S), which will feature promi- nently in what follows. We begin by giving a useful description of it. Let Vecf denote the category of finite families of vector spaces. An object of Vecf is a pair (V, L) where L is a finite set and V assigns to each element x of L a vector space Vx. A morphism (V, L) → (V ′, L′) consists of a bijection i : L → L′ and a linear map Vx → V ′ i(x) for each x ∈ L. We now claim that the following three categories are equivalent: (a) The category Sym(S). (b) The category of nice functors Vecf → Vec. (c) The category of sequences (Fn) where Fn is a nice Sn-equivariant functor Vecn → Vec. The equivalences, as well as the definition of "nice" in (b), should be clear. Each of these categories is an abelian tensor category and the various equivalences respect this structure. The tensor structure in (b) is given by (F ⊗ G)(V, L) = ML=A∐B F (V A, A) ⊗ G(V B, B). The tensor structure in (c) is like that for the sequence model of Sym(Vec). Note that since S is itself a tensor category we have a point-wise tensor product ⊠ on Sym(S). In (b) this product is given by hence the name "point-wise tensor product." This tensor product will not be used in the rest of this section but will show up later. (F ⊠ G)(V, L) = F (V, L) ⊗ G(V, L), As Sym(S) is an abelian tensor category, we have the notions of algebras and modules over algebras in Sym(S). We note that for A ∈ Sym(S) giving a multiplication map A ⊗ A → A amounts to giving a natural map A(V, L) ⊗ A(V ′, L′) → A(V ∐ V ′, L ∐ L′), where V ∐ V ′ denotes the natural map L′ ∐ L → Vec built out of V and V ′. Define an element of F ∈ Sym(S) to be an element of F (V, L) for some (V, L). We call #L the order of the element. (We will use the term "degree" later to reference the grading on S.) As in the twisted commutative setting, one can then give an elemental definition for "finitely generated" and this agrees with the more general definition in terms of finite length objects given in §2.2. We now give examples of some of the above definitions to give the reader some sense of their nature. Let Fλ be the object of Sym(S) which assigns to (V, L) the space 0 if #L 6= 1 and the space Sλ(Vx) if L = {x}. Then Fλ ⊗ Fµ assigns to (V, L) the space 0 if #L 6= 2 and the space [Sλ(Vx) ⊗ Sµ(Vy)] ⊕ [Sλ(Vy) ⊗ Sµ(Vx)] if L = {x, y}. As a second example, Sym(Fλ) is the object of Sym(S) which assigns to (V, L) the spaceNx∈L Sλ(Vx). Note in particular that the only partition appearing in the symmetric algebra Sym(Fλ) is λ itself. These examples underline the fact that the product in Sym(S) is formal: the Littlewood -- Richardson rule does intervene in any way. (The Littlewood -- Richardson rule is used in the point-wise product ⊠ for Sym(S).) The following is the main result we need on algebras in Sym(S). Theorem 2.6. An algebra in Sym(S) which is finitely generated in order 1 is noetherian. Proof. Let A be an algebra finitely generated in order 1. Let F be a finite length object of Sym(S). We must show that A ⊗ F is a noetherian A-module. As with any finitely generated algebra in Sym(S), the number of rows in any partition appearing in A is bounded; the same holds for A⊗ F . SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 13 Let d be an integer such that any partition appearing in A ⊗ F has at most d rows, and let U be a vector space of dimension d. For a finite set L let UL be the constant family on U , i.e., the object (V, L) of Vecf with Vx = U for all x. Then L 7→ UL defines a functor i : (fs) → Vecf , which respects the monoidal structure (disjoint union) on each category. The induced functor i∗ : Sym(S) → Sym(Vec) is a tensor functor. One easily sees that if M ′ and M are two sub-objects of A ⊗ F then M = M ′ if and only if i∗M = i∗M ′. We have thus shown that the map {A-submodules of A ⊗ F} → {(i∗A)-submodules of i∗(A ⊗ F )} is injective; it is obviously inclusion preserving. Now, i∗A is a twisted commutative algebra finitely generated in order 1; to see this, write A as a quotient of Sym(S) with S ∈ S, so that i∗A is a quotient of Sym(S(U )h1i). We thus find that i∗A is noetherian by Theorem 2.3. As i∗(A ⊗ F ) = (i∗A) ⊗ (i∗F ) is a finitely generated (i∗A)-module, it is noetherian. We thus find that every ascending chain of A-submodules of A ⊗ F stabilizes, and so A is noetherian. (cid:3) The following proposition is proved just like Proposition 2.5. Proposition 2.7. Let A be a noetherian algebra in Sym(S) on which a reductive group G acts. Then AG is again noetherian. If M is a finitely generated A-module with a compatible action of G then M G is a finitely generated AG-module. y for each x ∈ L. We now come to a central concept in this paper: 2.5. ∆-modules. Let Vec∆ be the category whose objects are pairs (V, L) as in Vecf , but where now a morphism (V, L) → (V ′, L′) consists of a surjection L′ → L together with a map Vx → Ny7→x V ′ Definition 2.8. A ∆-module is a nice functor Vec∆ → Vec. In the above definition, we say that a functor Vec∆ → Vec is nice if it is so when restricted to Vecf . We denote the category of these functors by Mod∆. It is abelian, though not semi-simple. Since Vecf is a sub-category of Vec∆, every ∆-module defines an object of Sym(S). For a ∆-module F we let [F ] be the class in Sym(K(S)) obtained by regarding F as an object of Sym(S). The forgetful functor Mod∆ → Sym(S) has a left adjoint, which we denote by Φ. To describe this functor explicitly, we must introduce a piece of notation. Let (V, L) be an object of Vecf and let U be a partition of L. For a set S ∈ U put VS =Nx∈S Vx. Then (V, U ) is an object of Vecf . There is a natural map (V, U ) → (V, L) in Vec∆, the surjection L → U mapping an element of L to the part of U to which it belongs. With this notation in hand, we can give the following formula for Φ: (ΦF )(V, L) =M F (V, U ), where the sum is over all partitions U of L. We leave it to the reader to work out the ∆-structure on Φ(F ) and verify its universal property. We call Φ(F ) the free ∆-module on F . By a (finite) free ∆-module we mean one isomorphic to Φ(F ), where F is a (finite length) object in Sym(S). Free modules are projective since Φ is a left adjoint and Sym(S) is semi-simple. We define an element of a ∆-module F to be an element of F (V, L) for some (V, L). Given a collection of elements of F one can speak of the ∆-submodule that it generates. We say that F is finitely generated if there is a finite set of elements of F that generate it. This is equivalent to F being a quotient of a finite free ∆-module. We say that a ∆-module is noetherian if every ascending chain of ∆-submodules stabilizes. Note that noetherian implies finitely generated. Let n be a positive integer. Let Tn be the object of Sym(S) which assigns to (V, L) the space if L = {x} and 0 otherwise. Let Wn be the symmetric algebra on Tn in the category Sym(S). x V ⊗n It is given by Wn(V, L) =Ox∈L V ⊗n x 14 ANDREW SNOWDEN The multiplication map in Wn is given by concatenation of tensors. Note that Sn acts on Wn by algebra homomorphisms. We also consider the free ∆-module on Tn. It is given by (ΦTn)(V, L) =Ox∈L V ⊗n x if #L ≥ 1, while (ΦTn)(V, L) = 0 for #L = 0. Observe that Φ(Tn) is naturally a subobject of Wn in the category Sym(S), and is in fact an ideal. The following is the key result connecting ∆-modules to objects we have previously studied, the final rung of the ladder of §1.5. Proposition 2.9. Any ∆-submodule of Φ(Tn) is a W Sn n -submodule of Φ(Tn). y , so that Wn(V ′, L′) = Proof. Let (V, L) and (V ′, L′) be two objects of Vecf . Let W = Ny∈L′ V ′ W ⊗n. Let v ∈ (ΦTn)(V, L), and let v′ ∈ W ⊗n be Sn-invariant. We must show that the image of v ⊗ v′ under the multiplication map (ΦTn)(V, L) ⊗ W ⊗n → (ΦTn)(V ∐ V ′, L ∐ L′) belongs to the ∆-submodule of ΦTn generated by v. Now, (W ⊗n)Sn is spanned by nth tensor powers of elements of W . It thus suffices to treat the case v′ = w⊗n for some w ∈ W . Pick an element x0 of L. Define a map f : (V, L) → (V ∐ V ′, L ∐ L′) in Vec∆, as follows. The map L ∐ L′ → L is the identity on L and collapses all of L′ to x0. For x 6= x0, the map fx : Vx → Vx is the identity. The map fx0 : Vx0 → Vx0 ⊗ W is given by id ⊗ w. The map f induces a map (ΦTn)(V, L) → (ΦTn)(V ∐ V ′, L ∐ L′), which one easily verifies is the map induced by multiplication by w⊗n on Wn. Thus if v is an element of (ΦTn)(V, L), then its product with w⊗n in Wn can be computed by taking its image under (ΦTn)(f ). This shows that the product of v and v′ belongs to the ∆-module generated by v, which completes the proof. (cid:3) Theorem 2.10. The ∆-module Φ(Tn) is noetherian. Proof. The algebra Wn is noetherian by Theorem 2.6, and so W Sn As Wn is a finite W Sn Φ(Tn). W Sn ∆-module. n is noetherian by Proposition 2.7. n -module. The same holds for the submodule If Mi is an ascending chain of ∆-submodules of Φ(Tn) then it is an ascending chain of n -submodules by the previous proposition, and therefore stabilizes. Thus Φ(Tn) is a noetherian (cid:3) n -module, it is a noetherian W Sn Call a ∆-module small if it is a subquotient of a finite direct sum of Φ(Tn)'s (with n allowed to vary). The above theorem implies that small ∆-modules are noetherian, and in particular finitely generated. We record the following result, which follows immediately from the definitions and Proposition 2.9. Proposition 2.11. Let F be a small ∆-module. Then there exists a finite chain 0 = F0 ⊂ ··· ⊂ Fr = F of ∆-submodules of F and integers ni such that Fi/Fi−1, regarded as an object of Sym(S), can be given the structure of a finitely generated module over W . Sni ni Remark 2.12. We can in fact show that all finitely generated ∆-modules are noetherian. The argument in the general case is by a Grobner degeneration, and is much different than our argument for Φ(Tn) presented above. However, the above argument for Φ(Tn), which relates ∆-submodules to modules over W Sn n , is important for our later arguments with Hilbert series. 2.6. More on ∆-modules. Let F be a ∆-module. We define F old(V, L) to be the space spanned by the images of the maps F (V, U ) → F (V, L), as U varies over all non-discrete partitions of L. One easily verifies that F old is a ∆-submodule of F . We define a functor Ψ : Mod∆ → Sym(S), Ψ(F ) = F/F old. Note that Ψ(F ) is naturally a ∆-module. However, if (V, L) → (V ′, L′) is a map in Vec∆ and L′ → L is not an isomorphism, then Ψ(F ) applied to this map is zero; this is why we regard Ψ(F ) as an object of Sym(S). In fact, Ψ(F ) is the universal quotient of F with this property. One may thus regard Ψ(F ) as the maximal semi-simple quotient (i.e., cosocle) of F . SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 15 A ∆-module M is finitely generated if and only if Ψ(M ) is a finite length object of Sym(S); this is a version of Nakayama's lemma. In fact, M is a quotient of Φ(Ψ(M )), though non-canonically. We have Ψ(Φ(F )) = F . The functor Ψ is right exact, but not exact. Its left derived functors LiΨ exist. If M is a finitely generated ∆-module then LiΨ(M ) is a finite length object of Sym(S); this can be deduced easily from the fact that finitely generated ∆-modules are noetherian. One can recover [M ] from [LΨM ] by applying Φ. Thus the sequence of polynomials [LiΨM ] contains more information than the series [M ]. We now give an alternative definition of ∆-modules. Recall that for an object (V, L) of Vecf and a partition U of L, there is a natural map (V, U ) → (V, L) in Vec∆. One easily verifies that every map in Vec∆ can be factored as one of these maps followed by a map in Vecf . In fact, we can even say a bit more. Call a partition little if all its parts are singletons, except one which has two elements. Call a map (V, U ) → (V, L) little if U is. One then verifies that any map (V, U ) → (V, L) can be factored into a sequence of little maps. Thus all morphisms in Vec∆ can be factored into little maps and maps in Vecf . Therefore, a ∆-module can be thought of as an object of Sym(S) together with the extra data of functoriality with respect to little maps. This extra data can be recorded in an elegant manner. Let m∗ : S → S ⊗2 be the co-multiplication map. It takes F ∈ S to the functor m∗F ∈ S ⊗2 given by (V, W ) 7→ F (V ⊗ W ). The functor m∗F has a natural S2-equivariant structure and so defines an object of Sym2(S). There is a unique extension of m∗ to a derivation ∆ : Sym(S) → Sym(S). Here by "derivation" we mean ∆ satisfies the Leibniz rule and interacts correctly with divided powers (Schur functors). A ∆-module is then just an object M of Sym(S) together with a map ∆M → M which satisfies a certain associativity condition, which we do not write out. The map ∆M → M precisely records the functoriality of M with respect to little maps in Vec∆, and the associativity condition ensures that M extends to a functor with respect to all maps in Vec∆. The image of the map ∆M → M is M old, and so its cokernel is Ψ(M ). There is an analogy between ∆-modules and graded C[t]-modules. The category Sym(S) is analogous to the category of graded vector spaces. The functor Φ is analogous to the functor which takes a graded vector space V to the free graded C[t]-module C[t] ⊗ V , while the functor Ψ is analogous to the functor which takes a graded C[t]-module M to the graded vector space M ⊗C[t] C. The map ∆M → M is analogous to multiplication by t, while the space M old is analogous to the image of t. One might hope that M 7→ LΨ(M ) provides an equivalence between the derived category of Mod∆ and some other natural derived category, in analogy with Koszul duality; we have not worked this out. 3. Hilbert series In this section we develop the theory of Hilbert series for certain objects of Sym(Vec), Sym(S) and Mod∆. The main results are rationality theorems. If A is a finitely generated graded ring, in the usual sense, one can prove the rationality of its Hilbert series by picking a surjection P → A, where P is a polynomial ring, resolving A by free P -modules and then explicitly computing the Hilbert series of a free P -module. The key fact that makes this work is that P has finite global dimension. In the setting of twisted commutative algebras, this approach is no longer viable: the twisted commutative algebra Sym(Uh1i) has infinite global dimension for any non-zero U . The reason for this is that no wedge power of Uh1i vanishes, so the Koszul complex does not terminate! We get around this problem by relating Hilbert series of twisted commutative algebras to equivariant Hilbert series of usual rings, where we can use the usual methods. To study Hilbert series of objects in Sym(S), we relate them to Hilbert series of twisted commutative algebras. Finally, to study Hilbert series of objects in Mod∆ (what we ultimately care about), we relate them to Hilbert series of objects in Sym(S). 16 ANDREW SNOWDEN 3.1. Hilbert series in Sym(Vec). Let M be an object of Sym(Vec), taken in the sequence model. We assume each Mn is finite dimensional. We define the Hilbert series HM of M by: HM (t) = 1 n! (dim Mn) tn. ∞ Xn=0 Of course, HM (t) is simply the element [M ] of Sym(K(Vec)) = QJtK. Our goal in this section is to demonstrate the following theorem: Theorem 3.1. Let A be a twisted commutative algebra finitely generated in order 1 and let M be a finitely generated A-module. Then HM (t) is a polynomial in t and et. Define H ∗ M (t) similarly to HM (t), but without the factorial. The theorem is equivalent to the following statement, which is what we actually prove: we have H ∗ M (t) = d Xk=0 pk(t) (1 − kt)ak for some integer d, polynomials pk(t) and non-negative integers ak. Note that for a module over a graded ring, in the usual sense, the Hilbert series only has a pole at t = 1, while our Hilbert series for modules over twisted commutative algebras can have poles at t = 1/k for any non-negative integer k. The above theorem only applies to modules over twisted commutative algebras generated in order 1, and is false more generally. For example, let M = A = Sym((C∞)⊗2), a twisted commutative algebra generated in order 2. Then HM (t) = et2 . Although this is not a polynomial in t and et, it is a very reasonable function, and one can hope that there is a nice generalization of Theorem 3.1. Before getting into the proof of Theorem 3.1 we introduce equivariant Hilbert series. Say a group G acts on an object M of Sym(Vec). We define its G-equivariant Hilbert series H ∗ M,G by: ∞ H ∗ M,G(t) = [Mn]tn Xn=0 where [Mn] denotes the class of Mn in the Grothendieck group K(G). Thus H ∗ M,G is a power series with coefficients in the ring K(G). We will need to use these Hilbert series in our proof of Theorem 3.1 and we will also need a generalization of Theorem 3.1 to the equivariant setting. We now begin the proof of Theorem 3.1. Thus let A and M be given. Since A is finitely generated in order one, it is a quotient of Sym(Uh1i) for some finite dimensional vector space U . Of course, M is a finitely generated module over Sym(Uh1i). It thus suffices to consider the case where A = Sym(Uh1i). Now, regard A and M in the Schur model. If Sλ occurs in A then λ has at most dim U rows. Since M is finitely generated, it is a quotient of A ⊗ S for some finite length object S of S; it follows that there is an integer d such that only those Sλ for which λ has at most d rows appear in M . We therefore do not lose information by considering M (Cd) with its GL(d) action. In fact, we can even consider M (Cd) with its T action without losing information, where T is the diagonal torus in GL(d). The main idea of the proof of Theorem 3.1 is to relate H ∗ M to H ∗ M (Cd),T , prove that the latter is of a specific form and then deduce from this the rationality of H ∗ M . (One can regard any graded C-algebra as a twisted commutative algebra. Thus H ∗ M (Cd),T makes sense. In fact, it agrees with the usual T -equivariant Hilbert series of M (Cd).) i We need to introduce a bit of notation related to T . We let α1, . . . , αd be the standard projections T → Gm. We define an involution of the coordinate ring of T , denoted with an overline, by αi = α−1 , and we write x2 for xx. We let ∆(α) be the discriminantQi<j(αi−αj). For a character χ of T we define RT χ(α)dα to be 1 if χ is trivial and 0 otherwise, and we extend RT dα linearly to all functions on T . (The symbol RT dα is just notation and does not indicate actual integration.) SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 17 If χ1 and χ2 are characters of irreducible representations of GL(d) then Weyl's integration formula (see [FH, §26.2]), stated in our language, reads 1 d!ZT χ1(α)χ2(α)∆(α)2dα =(1 if χ1 = χ2 0 if χ1 6= χ2 We identify K(T ) with Q[αi, α−1 ] so that a T -equivariant Hilbert series can be identified with a power series in t whose coefficients are Laurent polynomials in the αi. The following is the key step in our understanding of H ∗ i M : Lemma 3.2. We have Proof. Write M =L S⊕mλ λ On the other hand 1 H ∗ M (t) = d!ZT M (Cd),T (t; α) ∆(α)2 H ∗ 1 −P αi , the sum taken over λ. We then have: mλ · (dim Mλ) · tλ. H ∗ M (t) =Xλ dα. Put H ∗ M (Cd),T (t; α) =Xλ f (α) =Xλ mλ · (the character of Sλ(Cd)) · tλ. (the character of Sλ(Cd)) · dim Mλ. Weyl's integration formula now gives us H ∗ M (t) = We must compute f (α). Observe: ∞ 1 d!ZT ′ H ∗ M (Cd),T (t; α)f (α)∆(α)2dα. The character of the right side is f (α). The character of the left side is Sλ(Cd) ⊗ Mλ. (Cd)⊗k =Mλ Mk=0 αi!k Xk=0 d Xi=1 ∞ = . 1 1 −P αi (cid:3) This yields the stated formula. We have thus related H ∗ M (Cd),T , which should be easier to understand since M (Cd) is a finitely generated module over the polynomial ring A(Cd). We now see that H ∗ M , what we care about, to H ∗ M (Cd),T is indeed easy to understand: Lemma 3.3. We have H ∗ M (Cd),T (t; α) = p(t; α) Qd i=1(1 − αit)n where p(t; α) is a polynomial and n = dim U . Proof. The terms of the minimal resolution for M (Cd) over A(Cd) are A(Cd) ⊗ E• where Ei = TorA(Cd) i (M (Cd), C). Since Tor is functorial, each Ei carries an action of GL(d) (and therefore T ), and the minimal resolution is equivariant (or rather, can be taken to be so). Thus the T -equivariant Hilbert series 18 ANDREW SNOWDEN for M is the alternating sum of those for A(Cd) ⊗ Ei; of course, each of these is the product of those for A(Cd) and Ei. Since Ei is a finite dimensional representation of T its Hilbert series is a polynomial. Thus the lemma is reduced to the case M = A. Now, A(Cd) = Sym(U ⊗ Cd) = Sym(Cd ⊕ ··· ⊕ Cd) = Sym(Cd) ⊗ ··· ⊗ Sym(Cd) where Cd is summed with itself n = dim U times. We thus find that HA(Cd),T is the nth power of H ∗ Sym(Cd),T . Similarly, Sym(Cd) = Sym(C ⊕ ··· ⊕ C), where there are d copies of C and T acts on the ith one by the character αi. Thus H ∗ (cid:3) Sym(Cd),T =Q(1 − αit)−1. This proves the lemma. Combining the two lemmas, we obtain an expression where p(t; α) is a polynomial in t, the αi and the α−1 into p.) Expanding the integrand into a power series, we find i H ∗ M (t) =ZT p(t; α) 1 dα Q(1 − αit)n 1 −P αi . (We have absorbed the 1/d! and ∆(α)2 H ∗ n(cid:21) αk(cid:16)X αi(cid:17)ℓ M (t) =ZT  Xk,ℓ (cid:20) k where the sum is taken over k ∈ Zd ≥0 and ℓ ∈ Z≥0. Here n(cid:21) =(cid:18)k1 + n − 1 (cid:20) k n − 1 (cid:19)···(cid:18)kd + n − 1 n − 1 (cid:19), αk = αk1 1 ··· αed p(t; α)tk  dα 1 ··· αkd d and k = k1 + ··· + kd. We must show that this is a rational function in t. It suffices, by linearity, to treat the case where p(t; α) = te0αe1 d where the ei are integers. Of course, the te0 factor does not really affect anything, so we leave it out. We are thus reduced to showing that ZT  Xk,ℓ (cid:20) k n(cid:21) αk+e(cid:16)X αi(cid:17)ℓ tk  dα is rational. By degree considerations, the (k, ℓ) term in the above sum integrates to zero unless ℓ = k + e. Furthermore, when ℓ = k + e only one monomial in (P αi)ℓ contributes a non-zero quantity, namely the one where αi has exponent ki + ei. We therefore find that the above is equal to where Xk (cid:20) k n(cid:21) Ck+etk (k)! (k1)!··· (kd)! Ck = is the multinomial coefficient. (We use the convention that Ck = 0 if any of the coordinates of k are negative.) Theorem 3.1 now follows from the following lemma: Lemma 3.4. Let d be a positive integer, let e ∈ Zd and let p be a polynomial of d variables. Then is a rational function of t, with poles only at t = 1/a where 1 ≤ a ≤ d is an integer. (The sum is taken over k ∈ Zd ≥0.) X p(k)Ck+etk SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 19 Proof. Observe that the formula k1Ck1,k2,...,kd = kCk1−1,k2,...,kd is valid for any tuple of integers k ∈ Zd. To prove the lemma, it suffices to treat the case where p is a monomial. Thus, if p is not a constant, we can write p = kip′ for some index i and some smaller monomial p′. We therefore have X p(k)Ck+etk =X kip′(k)Ck+etk =X p′(k)(ki + ei − ei)Ck+etk =X p′(k)k + eCk+e′tk − eiX p′(k)Ck+etk where e′ is obtained from e by replacing ei with ei − 1. In the right term we have replaced p with a lower degree polynomial. In the left term we have replaced p with an equal degree polynomial, but one that is of the form p′(k)k where p′ has smaller degree. (Note that k + e = k + e.) It follows that by repeatedly applying the above process, we can reduce to the case where p is a function of k. Now, note that X Ck+etk. It thus suffices to show that P Ck+etk is a rational function. We have X knCk+etk =(cid:18)t d dt(cid:19)n Cktk Xk≥0 Ck+etk =Xk≥e Cktk−e = t−eXk≥e where k ≥ e means ki ≥ ei for each i. Now, the terms in the right sum for which some ki is negative are zero and therefore do not contribute. We can thus assume that each ei is non-negative. We can also ignore the t−e factor. Now, write k = (k1, k′) where k′ is a d − 1 tuple, and do similarly for e. Then Xk≥e Cktk = Xk1≥e1,k′≥e′ Cktk = Xk1≥0,k′≥e′ Cktk − e1−1 Xk1=0 Xk′≥e′ Cktk Now, for k1 fixed, we have Cktk = Xk′≥e′ tk1 k1! Xk′≥e′ (k′ + k1)··· (k′ + 1)Ck′tk′. Thus each of the sums on the right in the previous expression is of the general form that we are considering in this lemma but with a smaller d. We can therefore assume that they are each rational by induction. By repeating this procedure, we can thus reduce to the case e = 0. The identity now gives Ck = dn Xk=n Cktk = Xk≥0 1 1 − dt . This completes the proof. (cid:3) 20 ANDREW SNOWDEN 3.2. Equivariant Hilbert series in Sym(Vec). Unfortunately, Theorem 3.1 is too weak for our eventual applications. Before stating the result we need, we make a definition for the sake of clarity: Definition 3.5. Let A be a ring. A series f ∈ AJtK is rational if there exists a polynomial q ∈ A[t] with q(0) = 1 such that qf is a polynomial. Note that it could be that f ∈ AJtK is not rational, but that there is an extension A ⊂ B so that f is rational when regarded as an element of BJtK. Thus a bit of care needs to be taken with the definition. However, we do have the following simple result, the proof of which is left to the reader: Lemma 3.6. Let A ⊂ B be an inclusion of rings and let f be an element of AJtK such that f is rational when regarded as an element of BJtK. Then f itself is rational in the following cases: (1) there is a finite group G acting on B such that A = BG; (2) A and B are fields. The main result of this section is the following: Theorem 3.7. Let G be a connected reductive group, let Γ be a finite group, let A be a twisted commutative algebra finitely generated in order 1 on which G × Γ acts and let M be a finitely generated A-module with a compatible action of G × Γ. Then the G-equivariant Hilbert series M Γ,G of M Γ, regarded as an element of the power series ring K(G)JtK, is a rational function. H ∗ Most likely, this proposition could be generalized by replacing Γ ⊂ G × Γ with an arbitrary normal reductive subgroup of an arbitrary reductive group. We do not need this more general result, and so only prove the special one, which allows for some simplifications in the proof. With some book-keeping, one can also show that the denominator of H ∗ M Γ,G has a particular form, but we do not do this. The theorem implies a certain result about HM Γ,G, but one that is not so elegant: exponentials of algebraic functions (roots of polynomials over K(G)) appear. We prove this proposition following the same plan as the proof of last one, after some preliminary reductions. M,Γ×G, an element of K(Γ × G)JtK, is a rational function. To see this, assume we have an equation (1+tq)H ∗ M,Γ×G = p with p and q in K(Γ×G)JtK. Now, observe that K(Γ) can be thought of as the ring of class functions on Γ (at least, after an extension of scalars, which does not affect rationality by Lemma 3.6) and so we may write First, we observe that it suffices to prove that H ∗ H ∗ M,Γ×G =X Hiδi, q =X qiδi, p =X piδi where Hi, qi and ti belong to K(G)JtK and the δi are characteristic functions of conjugacy classes in Γ. Since the δi are orthogonal idempotents, we have (1 + tq)H ∗ M,Γ×G =X(1 + tqi)Hiδi =X piδi Now, we can think of the coefficients of H ∗ M,Γ×G as class functions on Γ. The series H ∗ and so (1 + tqi)Hi = pi holds for each i. Thus each Hi is a rational function in K(G)JtK. Since H ∗ M Γ,G = (#Γ)−1P Hi, it follows that it too is a rational function. This establishes our claim. M,Γ×G defines a rational element of K(Γ × G)JtK if and only if the series H ∗ M,γ,G obtained by evaluating on the element γ ∈ Γ is a rational element of K(G)JtK, for each γ. Let Γ′ be the cyclic subgroup generated by some γ ∈ Γ. By the same reasoning, H ∗ M,γ,G will be a rational element of K(G)JtK M,Γ′×G is a rational element of K(Γ′ × G)JtK. Thus it suffices to show that for each cyclic if H ∗ M Γ′×G is a rational element of K(Γ′ × G)JtK. In other words, we may subgroup Γ′ of Γ, the series H ∗ assume from the outset that Γ is cyclic. We make another reduction. We have K(G) = K(H)W where H is a maximal torus in G and W is its Weyl group. By Lemma 3.6, a power series with coefficients in K(G) is rational if and only if it is so when regarded with K(H) coefficients. Thus it suffices to show that H ∗ M,Γ×H is rational. In other words, we may as well assume from the outset that G is a torus. Finally, as in the proof of the Theorem 3.1, we may assume A = Sym(Uh1i), where U is a finite dimensional representation of Γ × G. Since Γ × G is a commutative reductive group, we can write SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 21 j=1 Cψj where the ψj are characters of Γ × G. Pick d large compared to the number of U = Ln rows appearing in M and let T ⊂ GL(d) be the diagonal torus. We now have: Lemma 3.8. We have H ∗ M (Cd),Γ×G×T (t; α) = p(t; α) . Qij(1 − αiψjt) Here p belongs to K(Γ × G × T )[t] = K(Γ × G)[t, αi]. The product is taken over 1 ≤ i ≤ d and 1 ≤ j ≤ dim U . Proof. As before, we can reduce to the case M = A by considering the minimal resolution of M . An easy computation, similar to the previous one, gives H ∗ (cid:3) A(Cd),Γ×G×T =Qij(1 − αiψjt)−1. Lemma 3.2 carries over exactly to the present situation. Combining it with the previous lemma dα. p(t; α) 1 Qij(1 − αiψjt) =X αa 1 − (P αi) i ψata Qj(1 − αiψjt) ≥0, with n = dim U , a is defined as a1 + ··· + an and ψa is (cid:20) k n(cid:21)ψ ψa, = Xa=k yields (5) For i fixed, we have H ∗ M,Γ×G(t) =ZT 1 where the sum is taken over a ∈ Zn defined as ψa1 n . Define 1 ··· ψan where a belongs to Zn ≥0, so that 1 = ∞ Xk=0(cid:20) k n(cid:21)ψ αk i tk. H ∗ αk(cid:16)X αi(cid:17)ℓ The sum is taken over k ∈ Zd M,Γ×G(t) =ZT  With this notation in hand, we now expand the integrand of (5) into a series. We find Qj(1 − αiψjt) Xk,ℓ (cid:20) k n(cid:21)ψ (cid:20) k =(cid:20) k1 n(cid:21)ψ We must show that the previous equation is rational in t. By linearity, it suffices to treat the case 1 ··· αed where p(t; α) is of the form xte0αe1 d where x belongs to K(Γ× G) and the ei are integers. Of course, xte0 pulls out of the integral, and can thus be safely ignored. Hence, it suffices to consider the case where p is αe, with e = (e1, . . . , ed). As before, the (k, ℓ) term only contributes if ℓ = k +e and then only one term of (P αi)ℓ contributes. The previous integral thus evaluates to: ≥0 and ℓ ∈ Z≥0 and our notation is as before; we mention p(t; α)tk  dα. n (cid:21)ψ ···(cid:20) kd n (cid:21)ψ . Xk (cid:20) k n(cid:21)ψ Now, for a single integer k the expression (cid:20) k n(cid:21)ψ ≥0 the expression (cid:20) k n(cid:21)ψ function of the ψ. It follows that for k ∈ Zd Ck+etk. is of the form Pi aiψk i where ai is a rational is of the form Pi aiψk i where 22 ANDREW SNOWDEN the sum is taken over tuples i ∈ {1, . . . , n}d and ψk functions in the ψ. It thus suffices to show that for each such tuple i, the expression i denotes ψk1 i1 ··· ψkd id ; again the ai are rational i Ck+etk ψk Xk is rational in t. This is accomplished in the following lemma, which completes the proof of Theo- rem 3.7: Lemma 3.9. Keep the above notation and let p be a polynomial. Then i Ck+etk is a rational function of t. (The sum is taken over k ∈ Zd which appear are of the form 1 − at where a is a sum of at most d of the ψ's. Proof. As in the proof of Lemma 3.4, we reduce to the case where p = 1. We then change k to k − e and pull monomials out of the sum, to obtain an expression of the form ≥0.) Furthermore, the only denominators X p(k)ψk The difference i Cktk. ψk Xk≥e i Cktk −Xk≥e ψk Xk≥0 i Cktk ψk is a finite sum of sums of the form considered in the lemma, but with a smaller value of d. (The terms which appear may no longer have p = 1; this is the reason for including p in the general form of the sum we consider.) It thus suffices to show that the first sum above is rational in t. We have ψk i Ck = (ψi1 + ··· + ψid)n = an Xk=n by the multinomial theorem. Thus the first sum in the previous expression is (1 − at)−1, which completes the proof. (cid:3) 3.3. Hilbert series in Sym(S). Let M be an object of Sym(S). We can regard M as a sequence of equivariant functors Mn : Vecn → Vec. Write Mn(V1, . . . , Vn) = Mi∈In Sλi,1(V1) ⊗ ··· ⊗ Sλi,n(Vn) for some index set In and partitions λi,j (both depending on n). We assume In is finite for each n. Put ∞ m∗ n = Xi∈In sλi,1 ··· sλi,n, H ∗ M (t) = m∗ ntn. Xn=0 n as an element of the polynomial ring Q[sλ] and H ∗ We regard m∗ M (t) as an element of the power series ring Q[sλ]JtK. The variable t is basically superfluous, since the power of t can be obtained M agrees with [M ]∗. One from the order of the polynomial m∗ can also define HM (t), which is analogous to [M ], by replacing m∗ n. Our main result concerning these series is the following: n. When t is omitted (or set to 1), H ∗ n with mn = 1 n! m∗ Theorem 3.10. Let A be an algebra in Sym(S) finitely generated in order 1 on which a finite group Γ acts and let M be a finitely generated A-module with a compatible action of Γ. Then H ∗ is a rational function of the sλ. M Γ SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 23 As with Theorem 3.7, this result does not translate to an elegant statement about HM Γ. We now explain the basic strategy of our proof, in the case where Γ is trivial. Let M be a finitely generated A-module. We would like to relate H ∗ M to the Hilbert series of an object of Sym(Vec), so that we can apply the results we have established in that case. The most obvious way to obtain an object of Sym(Vec) is to pick a vector space U , let i : (fs) → Vecf be the functor assigning to L the constant family UL and then consider i∗M . Unfortunately, H ∗ M cannot be recovered from H ∗ i∗M , or even H ∗ i∗M,GL(U ). Thus the most obvious approach fails. However, a slight modification works: instead of picking just one vector space U we pick finitely many U1, . . . , Ur and build from M an object of Sym(Vec) with an action of GL(U1) × ··· × GL(Ur). It turns out that, due to the form of M , we can take r large enough so that information is not lost -- this is essentially the content of Proposition 3.14 below. Before proving that result, we need some lemmas, the first two of which are left to the reader. Lemma 3.11. Let f : Cn → Cm be a polynomial map whose components are homogeneous of positive degree and whose image is not contained in any linear subspace of Cm. Then for r ≫ 0 any element of Cm can be expressed as a sum of r elements of the image of f . Lemma 3.12. Let K be a field and let (fi) be a sequence of elements in K. Assume that there exists m > 0 such that for all k1, . . . , km sufficiently large, the m×m matrix (fki−j) has determinant zero. Then Pi≥0 fiti can be expressed in the form a/b where a and b belong to K[t] and deg b ≤ m − 1. Lemma 3.13. Let A be a UFD and let (fi) be a sequence of elements of A. Assume that there exists an integer m and elements α1, . . . , αm of A such that m fk = αifk−i Xi=1 holds for all k ≫ 0. Amongst all such expressions choose one with m minimal. Then given any N there exist k1, . . . , km > N such that the determinant of the matrix (fki−j) is non-zero. Proof. Put f = P fiti, an element of AJtK, and q = 1 −Pm i=1 αiti, an element of A[t]. We have that qf belongs to A[t]; write p = qf so that f = p/q. Note that A[t] is also a UFD and that the minimality assumption on m implies that p and q are coprime. Indeed, say p and q are both divisible by some non-unit r ∈ A[t]. Then we can write p = rp′ and q = rq′. Evaluating the second expression at t = 0 gives 1 = r(0)q′(0) so that r(0) and q′(0) both belong to A×. We can therefore scale r so that r(0) = 1, in which case q′(0) = 1 as well. Since r is assumed to be a non-unit and r(0) is a unit, it follows that r has degree at least 1. We then have f = p′/q′ with q′(0) = 1 and deg q′ < m, contradicting the minimality of m. Now, assume for the sake of contradiction that det(fki−j) = 0 for all sufficiently large ki. By Lemma 3.12 we have f = g/h where g and h belong to K[t] and deg h < m. Here K is the field of fractions of A. Pick a ∈ A non-zero so that ag and ah belong to A[t]. We then have (ah)p = (ag)q. Since p is coprime to q it follows that ah is divisible by q. However, this contradicts h having smaller degree than q. We thus conclude that det(fki−j) cannot vanish for all sufficiently large ki. (cid:3) Proposition 3.14. Let W be a finite dimensional vector space over a subfield C and let V be a finite dimensional subspace of Sym(W ) spanned by homogeneous elements. For a positive integer r, let ir : Sym(V ) → Sym(W )⊗r be the ring homomorphism which on V is given by x 7→ P{x}i, where {−}i : Sym(W ) → Sym(W )⊗r is the ring homomorphism given by inclusion into the ith factor. Then for r ≫ 0 we have: (a) The map ir is injective. (b) If x ∈ Frac(Sym(V )) and ir(x) belongs to Sym(W )⊗r then x belongs to Sym(V ). (c) A series f ∈ Sym(V )JtK is rational if and only if ir(f ) is. 24 ANDREW SNOWDEN Proof. We can check the conclusions of the proposition by tensoring up to C, and so we may thus assume we are working with complex vector spaces. The map Sym(V ) → Sym(W ) corresponds to an algebraic map f : W ∗ → V ∗ whose components are homogeneous of positive degree. Since V → Sym(W ) is injective, the image of f is not contained in any linear subspace of V ∗. It thus follows from Lemma 3.11 that for r ≫ 0, any element of V ∗ is a sum of r elements of the image of f . Now, the map ir corresponds to the map i∗ r : (W ∗)r → V ∗ given by (w1, . . . , wn) 7→ P f (wi). We thus see that i∗ r is surjective for all r ≫ 0. Fix r ≫ 0 and put i = ir. We now show that (a) and (b) hold. The equation i(x) = 0 is equivalent to x ◦ i∗ = 0, where x is thought of as a function V ∗ → C. Since i∗ is surjective, this equation implies x = 0. This shows that i is injective. Now say that x belongs to Frac(Sym(V )) and i(x) is a polynomial. Then x defines a rational function on V ∗ such that x ◦ i∗ is a regular function on (W ∗)r. Since i∗ is surjective, x must be regular on V ∗ and thus it belongs to Sym(V ). We now prove (c). It is clear that if f is rational then i(f ) is as well. Thus let f ∈ Sym(V )JtK be given and assume that i(f ) is rational. Write f =P fiti with fi ∈ Sym(V ). The rationality of i(f ) means that we can find a polynomial q = 1 −Pm i=1 αiti with αi ∈ Sym(W )⊗r such that qf is a polynomial. Choose q with m minimal. We then have m fk = αifk−i Xi=1 for all sufficiently large k. Thus for all large k1, . . . , km we have the equation Ax = y where A is the m × m matrix (fki−j), x is the column vector (αi) and y is the column vector (fki). By the minimality of m and Lemma 3.13 we can pick ki so that det A is non-zero. We then find x = A−1y, which shows that αi belongs to i(Frac(Sym(V ))). Since αi also belongs to Sym(W )⊗r, statement (b) of the lemma implies that each αi belongs to i(Sym(V )). Thus q = i(q′) for a unique q′ in Sym(V )[t] with the required properties to establish that f is rational. (cid:3) We now prove the theorem: Proof of Theorem 3.10. Let Γ, A and M be given. Let P be the set of partitions appearing in M . The set P is finite; indeed, only finitely many partitions appear in A (see the discussion preceding Theorem 2.6) and M is a quotient of A ⊗ F for some finite length object F of Sym(S). Let V be the subspace of K(S) spanned by the sλ with λ ∈ P . Thus H ∗ Let U be a finite dimensional vector space whose dimension exceeds the number of rows of any partition appearing in P and let G = SL(U ). Then K(G) is a polynomial ring. (We use SL(U ) instead of GL(U ) so that the Grothendieck group is a polynomial ring; it makes the argument a bit cleaner.) Evaluation on U gives a map K(S) → K(G) which is injective when restricted to V . Let r be a large integer and let M Γ belongs to Sym(V )JtK. be the ring map which is given by φ : Sym(K(S)) → K(G)⊗r r φ([S]) = {[S(U )]}i Xi=1 for [S] in K(S), where {−}i : K(G) → K(G)⊗r is the inclusion in the ith factor. By Proposi- tion 3.14(c), a power series f ∈ Sym(V )JtK is a rational function if and only if φ(f ) is. It thus suffices to show that φ(H ∗ To understand the map φ we lift it to a functor Φ, as follows. First, identify K(G)⊗r with K(Gr). Let U1, . . . , Ur be copies of U and put Gi = SL(Ui); we think of Gr as G1 × ··· × Gr. We regard φ as a map M Γ) is rational. φ : Sym(K(S)) → K(G1 × ··· × Gr). SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 25 It can be described explicitly on K(S) as follows: Xi=1 φ([S]) = r Now define Φ to be the functor [S(Ui)]. Φ : Sym(S) → Sym(Vec), Φ(F )(L) = ML=L1∐···∐Lr F ((U1)L1 ∐ ··· ∐ (Ur)Lr ). Here the sum is over all partitions of L into r parts and (Ui)Li denotes the family (V, Li) where Vx = Ui for all x ∈ Li. (A more conceptual description of Φ is as follows. Let i : (fs)r → Vecf take (L1, . . . , Lr) to (U1)L1 ∐ ··· ∐ (Ur)Lr and let j : (fs)r → (fs) be the addition map. Then Φ(F ) = j∗i∗F .) One readily verifies that Φ is a tensor functor. We now claim that Φ lifts φ, that is, we have φ([N ]) = [Φ(N )] for all N in Sym(S). Both sides above are additive in N so it suffices to treat the case where N is a simple object of Sym(S). As we have previously stated (§2.2), the simple objects are of the form N Sλi(Si) where the λi are partitions and Si are distinct simple objects of S. Since Φ is a tensor functor and φ is a ring homomorphism, we are reduced to the case of considering N = Sλ(S). Put k = λ. Then N is the equivariant functor Veck → Vec given by We thus find that Φ(N ) ∈ Sym(Vec) is supported in order k and assigns to the set {1, . . . , k} the space (V1, . . . , Vk) 7→ Mλ ⊗ S(V1) ⊗ ··· ⊗ S(Vk). where the sum is over all (i1, . . . , ik) in {1, . . . , r}k. We therefore have Mλ ⊗ Mi1,...,ik S(Ui1 ) ⊗ ··· ⊗ S(Uik ) On the other hand, [Φ(N )] = dim Mλ k! r Xi=1 [S(Ui)]!k . [N ] = dim Mλ k! [S]k. Applying φ to the above gives exactly the previous formula for [Φ(N )]. This proves the claim. The above discussion, and the fact that Φ commutes with the formation of Γ invariants, shows that φ(H ∗ M Γ) = H ∗ Φ(M )Γ,G1×···×Gr . As Φ is a tensor functor, Φ(A) is a twisted commutative algebra finitely generated in order 1 and Φ(M ) is a finitely generated module over it. Thus H ∗ is a rational function in K(G1 × ··· × Gr)JtK by Theorem 3.7. We thus find that φ(H ∗ M Γ) is rational, which completes the proof. (cid:3) Φ(M )Γ,G1×···×Gr We note the following corollary of the proposition. Corollary 3.15. Let A be an algebra in Sym(S) finitely generated in order 1 on which a finite group Γ acts and let M be a finitely generated AΓ-module. Then H ∗ M is a rational function of the sλ. Proof. Let N = A ⊗AΓ M . Then N is a finitely generated A-module and N Γ = M , where Γ acts on N by acting trivially on M . We thus have that H ∗ N Γ is rational by the proposition. (cid:3) M = H ∗ 26 ANDREW SNOWDEN 3.4. Hilbert series of ∆-modules. Let M be a ∆-module. We define its Hilbert series to be the Hilbert series of the underlying object of Sym(S). Our main result on such Hilbert series is the following, which follows immediately from Corollary 3.15 and Proposition 2.11. Theorem 3.16. Let M be a small ∆-module. Then H ∗ M is a rational function of the sλ. Remark 3.17. We expect this result to hold for all finitely generated ∆-modules. We can prove it for a much larger class than the class of small ∆-modules, but we have not been able to prove it for all finitely generated ∆-modules. 4. Syzygies of ∆-schemes We now apply the theory we have developed to the study of the syzygies of certain families of schemes, which we call ∆-schemes. 4.1. ∆-schemes. An abstract ∆-scheme is a functor X from the category Vec∆ to the category of schemes over C. The notion of a morphism of abstract ∆-schemes is evident. In this way we have a category of abstract ∆-schemes. We say that a morphism X → Y of abstract ∆-schemes is a closed immersion if X(V, L) → Y (V, L) is a closed immersion for all (V, L). The category of abstract ∆-schemes is too large for our purposes; we now introduce a more manageable category. For an object (V, L) of Vec∆, let V(V, L) be the vector space Nx∈L V ∗ x , regarded as a scheme. Then V is an abstract ∆-scheme, in an obvious way. A ∆-scheme is a pair (X, i) consisting of an abstract ∆-scheme X such that X(V, L) is non-empty for all (V, L) and a closed immersion i : X → V. Note that if X is a ∆-scheme, then X(V, L) ⊂ V(V, L) is closed under scaling (by functoriality), and is thus a cone. A morphism of ∆-schemes is a morphism of abstract ∆-schemes which commutes with the embeddings into V. There is at most one morphism between two ∆-schemes, and so the category of ∆-schemes is partially ordered. We can think of ∆-schemes in terms of their ideals of definitions, as follows. Let P (V, L) be the affine coordinate ring of V(V, L). We have P (V, L) = Sym(cid:18)Ox∈L Vx(cid:19). A ∆-ideal is a rule which assigns to each object (V, L) of Vec∆ an ideal I(V, L) of P (V, L) such that if (V, L) → (V ′, L′) is a morphism in Vec∆ then under the map P (V, L) → P (V ′, L′) the ideal I(V, L) maps into I(V ′, L′). Suppose I is a ∆-ideal. Let X(V, L) be the subscheme of V(V, L) cut out by I(V, L). Then X is naturally a ∆-scheme. The association I 7→ X is an anti-equivalence of categories. By an element of P we mean an element of P (V, L) for some (V, L). Given any collection S of elements of P there is a unique minimal ∆-ideal I containing S; we call I the ∆-ideal generated by S. In more concrete terms, a ∆-scheme can be thought of as a sequence of rules (Xn)n≥0, where Xn assigns to each n-tuple of vector spaces (V1, . . . , Vn) a closed subscheme Xn(V1, . . . , Vn) of i , such that: Xn is functorial in (V1, . . . , Vn); Xn is Sn-equivariant; and Xn+1(V1, . . . , Vn+1) is N V ∗ contained in Xn(V1, . . . , Vn−1, Vn ⊗ Vn+1). Of course, we can describe ∆-ideals similarly. 4.2. Examples of ∆-schemes. We now give some examples of ∆-schemes. There are two rather trivial examples: V itself is a ∆-scheme, and the final object in the category; and the rule (V, L) 7→ Spec(C) (embedded in V as the origin) is a ∆-scheme, and the initial object in the category. A more interesting example, and the one which motivated the whole theory, is the Segre embedding: take X(V, L) to be the set of pure tensors in V(V, L) =Nx∈L V ∗ x , with its usual scheme structure. In fact, the Segre example is just the first in a family. Let d ≥ 0 be an integer. As a vector space, the algebra P (V, L) breaks up into isotypic pieces for the action of the group Qx∈L GL(Vx); each such piece corresponds to a family of partitions indexed by L. Let Id(V, L) be the sum of all pieces in which some partition has more than d rows. One easily checks that Id(V, L) is an ideal SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 27 of P (V, L), and so cuts out a closed subscheme Subd(V, L) of V(V, L). Geometrically, a point v in V(V, L) belongs to Subd(V, L) if and only if for each x ∈ L there exists a quotient Ux of Vx of dimension at most d such that v belongs to Nx∈L U ∗ x . For further discussion of these subspace varieties, see [LW2, §3]. Now, it is clear that Subd(V, L) is functorial for maps in Vecf . However, for d ≥ 2, it is not functorial for maps in Vec∆, and is therefore not a ∆-scheme. Nonetheless, Subd contains a unique maximal ∆-scheme, which we call ∆Subd. Geometrically, we have ∆Subd(V, L) =\ Subd(V, U ), where the scheme-theoretic intersection is taken over all partitions U of L. Algebraically, ∆Subd corresponds to the ∆-ideal generated by the Id(V, L). It follows from [LW2, Thm 3.1] that the ∆-ideal corresponding to ∆Subd−1 (for d > 1) is generated by any element of the one dimensional space One can describe ∆Subd as the largest ∆-scheme whose coordinate ring has the property that every partition appearing in it has at most d rows. VdCd ⊗VdCd ⊂ Symd(Cd ⊗ Cd) ⊂ P2(Cd, Cd). The ∆-scheme ∆Sub0 is just the initial ∆-scheme, i.e., the origin in V. The ∆-scheme ∆Sub1 coincides with Sub1, and is the Segre embedding. The ∆-scheme ∆Sub2 (which does not coincide with Sub2) is equal to the secant variety of the Segre embedding -- this assertion is essentially equivalent to the recently proved GSS conjecture (see [Ra]). We do not know an elegant geometric description of ∆Subd for d > 2. The category of ∆-schemes is closed under several natural operations. Let X and Y be two ∆-schemes. Then the union X ∪ Y and scheme-theoretic intersection X ∩ Y of X and Y inside of V are ∆-schemes. In this way, the poset of ∆-schemes is a lattice. Let X + Y be the scheme-theoretic image of the map X × Y → V given by (x, y) 7→ x + y. Then X + Y is a ∆-scheme. In particular, the secant schemes to a ∆-scheme are again ∆-schemes. The same holds for tangent schemes. The functor Xred which attaches to (V, L) the reduced subscheme Xred(V, L) of X(V, L) is a ∆-scheme. Starting with the ∆Subd and using these operations, one obtains a large number of examples of ∆-schemes. 4.3. Syzygies of ∆-schemes. Fix a ∆-scheme X. Let R(V, L) be the affine coordinate ring of X(V, L), a quotient of the polynomial ring P (V, L). Put Fp(V, L) = TorP (V,L) p (R(V, L), C). As discussed in the introduction, the Fp(V, L) record the syzygies of R(V, L) as a P (V, L)-module. The functorial properties of R, P and Tor imply that Fp is a ∆-module; this will be made more clear in the proof of the following theorem. Let F (d) be its degree d piece; it too is a ∆-module. Our main result on syzygies is the following theorem: p Theorem 4.1. The ∆-module F (d) p is small. Proof. Let W (V, L) =Nx∈L Vx be the degree one piece of P (V, L). The P (V, L)-module C admits a Koszul resolution, the terms of which are P (V, L)⊗ViW (V, L). Tensoring this over P (V, L) with R(V, L), we find that there is a complex computing Fp(V, L) whose terms are R(V, L)⊗ViW (V, L). Put Mi,j(V, L) = R(i)(V, L) ⊗^j W (V, L). We then find that F (d) p (V, L) is the homology of a complex Md−p−1,p+1(V, L) → Md−p,p(V, L) → Md+1−p,p−1(V, L) 28 ANDREW SNOWDEN Now, the functor (V, L) 7→ Mi,j(V, L) is easily seen to be a ∆-functor. A short calculation with the Koszul complex shows that the differentials in the above complex respect the ∆-module structure on the Mi,j. Thus F (d) , as a ∆-module, is the homology of the complex of ∆-modules p We now claim that Mi,j is a small ∆-functor. Indeed, R(i) is a quotient of P (i) = Φ(Symi), which Md−p−1,p+1 → Md−p,p → Md+1−p,p−1. is a quotient of Φ(Ti). Of course, (V, L) 7→VjW (V, L) is the ∆-module Φ(Vj), and thus a quotient of Φ(Tj). Thus Mi,j = R(i) ⊠ Φ(Vj) a quotient of Φ(Ti) ⊠ Φ(Tj) = Φ(Ti+j), and is thus small. Now, a subquotient of a small ∆-module is small; thus F (d) is small. This completes the proof. (cid:3) p Corollary 4.2. The ∆-module F (d) p is finitely generated. Thus there exists a finite list of p-syzygies of degree d for the schemes X(V, L) which give rise to all p-syzygies of degree d under the basic operations on the ∆-module F (d) In view of this corollary, the statement "Fp is finitely generated as a ∆-module" is equivalent to the statement "F (d) p = 0 for d ≫ 0." This probably does not hold in full generality, but in §4.5 we single out a large class of ∆-schemes for which it might hold. Corollary 4.3. The series [F (d) ]∗ is a rational function in the sλ. . p p This essentially shows that the information content of the p-syzygies of degree d of X is finite. As we have said, this series can be computed algorithmically (this is discussed below). Thus, essentially all the information about p-syzygies of degree d of X can be computed in finite time! 4.4. Computational aspects. We now elaborate on the remark we have made that our proofs give algorithms to calculate the relevant objects. Keep the same notation as the previous section. Say we would like to compute generators for the ∆-module F (d) of p-syzygies of degree d. (The algorithm for computing the rational function [F (d) ]∗ proceeds along similar lines.) We proceed as follows. First, F (d) is the homology of the sequence p p p Md−p−1,p+1 → Md−p,p → Md+1−p,p−1. As explained, R(i) is a quotient of Φ(Ti) while (V, L) 7→ VjW (V, L) is a quotient of Φ(Tj), and so Mi,j is a quotient of Φ(Ti+j). In particular, each module in the above complex is a quotient of Φ(Td). This shows that each is canonically a W Sd d -module and the differentials respect this structure. Now, Wd only has partitions with at most d rows; the same is true for the Mi,j above. Thus it suffices to see what happens when we evaluate on Cd. Precisely, let A be the twisted commutative algebra L 7→ W Sd d ((Cd)L) and let Ni,j be the A-module given by L 7→ Mi,j((Cd)L), where (Cd)L denotes the constant family on Cd. Let E be the homology of the complex Nd−p−1,p+1 → Nd−p,p → Nd+1−p,p−1; p thus EL = F (d) generators for F (d) A-module are generators for F (d) as a W Sd ((Ck)L). The row bounds then imply that generators for E as an A-module are d -module. Proposition 2.9 thus implies that generators for E as an p p as a ∆-module. Now, A is equal to Sym(Uh1i)Sd where U = (Cd)⊗d. It follows that any partition in A has at most dim U = dd rows. Since Md−i,i is a subquotient of Wd, it follows that Nd−i,i is a subquotient of Sym(Uh1i). Thus any partition appearing in Nd−i,i has at most dd rows as well. Thus we do not loose information by evaluating on Cdd (regarding everything in the Schur model). That is, generators for E(Cdd )-module give generators for E. ) as an A(Cdd Finally, A(Cdd of the modules Nd−i,i(Cdd ) is the subring of the polynomial ring in d2d variables which are Sd-invariant. Each ) is the homology of the ) is a finite module over this ring. And E(Cdd SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 29 complex N• at i = p − 1. We have thus reduced the problem to a computation involving explicitly described finitely generated rings and modules. These computations can be done algorithmically. 4.5. ∆-schemes of finite level. A ∆-scheme X (resp. ∆-ideal I) has level ≤ d if every partition appearing in its coordinate ring has at most d rows. One easily sees that X has level ≤ d if and only if it is contained in ∆Subd. We say that X has finite level if it has level ≤ d for some d. All the operations we have discussed (union, intersection, sum, formation of secant, tangent and reduced schemes) preserve the finite level condition. We thus have many examples of ∆-schemes of finite level. Note, however, that the final ∆-scheme V does not have finite level. The following three statements are equivalent: (a) Every ascending chain of ∆-ideals of finite level stabilizes. (b) Every ∆-ideal of finite level is finitely generated. (c) Every ∆-ideal I of finite level is "uniformly generated," i.e., there exists an integer d such that I(V, L) is generated in degrees ≤ d for all (V, L). It is clear that (a) and (b) are equivalent, and clear that each implies (c). That (c) implies (b) follows from Corollary 4.2 and the discussion following it. It seems reasonable to hope that statements (a) -- (c) are in fact true. Indeed, there has been some recent work establishing that particular finite level ∆-ideals are finitely generated: variety to the Segre embedding is finitely generated as a ∆-ideal. • Raicu [Ra] has proved the GSS conjecture, which implies that the ideal of the first secant • Draisma and Kuttler [DK] have shown that the secant varieties to the Segre varieties are (In fact, in [DK, §7], a • Landsberg and Weyman [LW] have obtained results about the ideal of the tangent variety all cut out set-theoretically by equations of a bounded degree. conjecture concerning something similar to statements (a) -- (c) above is posed.) to the Segre, which imply that it is finitely generated as a ∆-ideal. A proof of the statements (a) -- (c) would have great consequences: for instance, it would essentially encompass all of the above results. Assuming these statements are true, it would be reasonable to believe that the pth syzygy module of a finite level ∆-scheme is supported in finitely many degrees. We now apply the theory we have developed to the study of syzygies of Segre embeddings. 5. Application to Segre varieties 5.1. Theorems A and B. We begin by establishing the theorems stated in the introduction. Let Fp be the ∆-module of p-syzygies of the Segre embedding, as defined in §4.3. By Theorem 4.1 and its corollaries, Theorems A and B follow from the following result, which states that Fp has only finitely many non-zero graded pieces. This result is well-known to the experts, so we only give a brief proof. We thank Aldo Conca for showing it to us. Proposition 5.1. For p ≥ 1, the space Fp is supported in degrees p + 1, . . . , 2p. Proof. This follows from the fact that the ideal of the Segre variety is generated by a Grobner basis of degree 2 [ERT, Prop. 17], together with general facts about Grobner bases (such as the Taylor resolution, see [E, Exercise 17.11]). (cid:3) Remark 5.2. The bound in the proposition is not optimal. Indeed, it is known [Ru] that F2 and F3 are supported in degrees 3 and 4, while the upper bounds provided by the proposition are 4 and 6. The optimal upper bound is not known. However, our proof of Theorem A provides an algorithm for finding it for any particular value of p. 30 ANDREW SNOWDEN 5.2. The work of Lascoux. Lascoux determined the entire minimal resolution of certain deter- minantal varieties (see [L], and also [PW], where a gap in [L] is resolved). The rank 1 case of his result exactly gives the leading term of our series fp. We recall his results in our language. First, we discuss some terminology. Let m = sλ1 ··· sλn be a monomial in the variables sλ. We say that m has order n. We say that m has degree d if each λi is a partition of d. Every term in the series [F (d) ] has degree d, while the orders of the terms are unbounded. The degree d, order n terms in fp give information about the p-syzygies of degree d for the Segre embedding of an n-fold product of projective spaces. p p,2 . Write d = p + h. Of course, if h ≤ 0 then f (d) p,2 = 0 for h > p. Lascoux gives a much better bound: f (d) Let fp,2 be the order two two term of fp. This is the leading order term of fp. We consider its degree d piece f (d) p,2 = 0. Proposition 5.1 implies p,2 = 0 for h > √p. Assume now that f (d) 1 ≤ h ≤ √p. Let S be the set of pairs of partitions (α, β) such that α has at most h columns, β has at most h rows and α + β = p − h2. Associate to (α, β) a new pair of partitions (µ, ν) as follows. Start with a rectangle with h columns and h + 1 rows. To get µ, append α to the bottom and β to the right. To get ν, append the dual of β to the bottom and the dual of α to the right. Lascoux's result is then f (d) p,2 = 1 2 X(α,β)∈S sµsν. 1,2 = 1 2 s2 For example, say p = 1 and d = 2. Then h = 1. Since p − h2 = 0 the set S consists of the single pair (α, β) where α = β is the zero partition. The partitions µ and ν are both (1, 1) and so we find f (2) (1,1), i.e., the quadratic piece of the ideal of the embedding of P(V1) × P(V2) is V2V1 ⊗V2V2. 5.3. The polynomial gp. Let Gp = Ψ(Fp) be the cokernel of the map ∆Fp → Fp. Since Fp is a finitely generated ∆-module, Gp is a finite length object of the category Sym(S). The object Gp records precisely those syzygies that cannot be built out of syzygies on a product of fewer p = [Gp]∗). The objects LiΨFp for projective spaces. We let gp be the polynomial i ≥ 1 are important as well -- indeed, [Fp] can be recovered from them -- though they are bit less accessible. We remark that our two main theorems can be rephrased using fp and gp so as to look more p is a polynomial, while Theorem B is exactly [Gp] (and g∗ similar: Theorem A is exactly the statement that g∗ the statement that f ∗ p is a rational function. 5.4. An Euler characteristic. Define χ =Xp≥0 (−1)pfp. There are only finitely many terms of a given degree and order in the sum, and so it makes sense. We remark that f0 = 1 -- the first term in the resolution of R is always P = P ⊗ C and so F0 = C for any (V, L). The main result of this section is an explicit computation of χ. The notation used in the following proposition is defined below it. Proposition 5.3. We have χ =" ∞ Xk=0 exp(s(k))# ⊠"Xλ κλ exp(s′ λ)# , where κλ is the rational number (−1)λ sgn(cλ)z−1 -- including λ = 0, where the term is 1. λ . The second sum is taken over all partitions λ Extracting the degree k piece of the above formula yields: SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 31 Corollary 5.4. For k > 0 we have χ(k) = k Xp=0   (−1)p p! Xλ⊢p (#cλ) sgn(cλ) exp(s(k−p) The p = 0 term of the above sum is exp(s(k)). We have χ(0) = 1. ⊠ s′ λ)  . We now define notation that will be in place for the rest of the section (and is used in the above proposition). Let λ be a partition of p. We let cλ denote the conjugacy class of Sp corresponding to λ, normalized so that λ = (1, . . . , 1) corresponds to the identity element, and we let zλ = p!/#cλ be the order of the centralizer of any element of cλ. We let χλ denote the character of Sp corresponding to λ, normalized so that λ = (1, . . . , 1) corresponds to the sign character sgn. The notation sλ means what is has meant previously, namely the element [Sλ] of K(S); in particular, s(k) = [Symk]. We define s′ λ to be the element of K(S) of degree p which corresponds to the class function on Sp supported on cλ and taking value zλ there. Explicitly, χµ(cλ)sµ. s′ λ =Xµ⊢p The symbol ⊠ is the point-wise tensor product: sλ ⊠ sµ is computed using the Littlewood -- Richardson rule. As usual, W = W1 is the object of Sym(S) which assigns to (V, L) the tensor product of the V 's. Throughout this section ViW and Sλ(W ) refer to point-wise operations in Sym(S). For instance, ViW is the object of Sym(S) given by (ViW )(V, L) =Vi(W (V, L)) =^i Ox∈L Vx! . We now begin proving Proposition 5.3. We begin with the following. Lemma 5.5. We have ∞ χ = [R] ⊠ Xp=0  (−1)p[VpW ]  . Proof. We have Proof. As discussed in the proof of Theorem 4.1, Fp is the homology of the complex R ⊠ViW at i = p − 1. The formula follows from standard facts about Euler characteristics. Lemma 5.6. We have [R(d)] = exp(s(d)), and so [R] =Pd≥0 exp(s(d)). R(d) n (V1, . . . , Vn) = Symd(V1) ⊗ ··· ⊗ Symd(Vn). n! [Symd]n in Symn(K(S)). The result follows. Thus [R(d) Lemma 5.7. Let λ be a partition of p > 0. Let sλ,n be the class in Symn(K(S)) of the functor (V1, . . . , Vn) 7→ Sλ(V1 ⊗ ··· ⊗ Vn). Then sλ,n = (#cµ)χλ(cµ) exp(s′ n ] is equal to 1 ∞ (cid:3) (cid:3) µ). Xn=0 1 p!Xµ⊢p Note that the left side above is nothing other than [Sλ(W )]. Proof. A simple manipulation shows that for any vector spaces U and V we have where the sum is over all partitions µ and ν of p, and Sλ(U ⊗ V ) =M Cλµν Sµ(U ) ⊗ Sν(V ), Cλµν = dim(Mλ ⊗ Mµ ⊗ Mν )Sp = dim HomSp(Mλ, Mµ ⊗ Mν ). 32 ANDREW SNOWDEN (This appears as Exercise 6.11(b) in [FH].) We therefore have Sλ(V1 ⊗ ··· ⊗ Vn) =Mµ,ν Cλµν Sµ(V1) ⊗ Sν(V2 ⊗ ··· ⊗ Vn). We thus have a recurrence sλ,n = Cλµν sµsν,n−1. 1 nXµ,ν vn = 1 n Avn−1. X vn = exp(A)v0. It will now be convenient to switch from working in the degree p piece of K(S) to working in K(Sp). The two are in isomorphism via sλ = [Sλ] ↔ [Mλ]. Thus sλ,n can be regarded as an element of Symn(K(Sp)). Note that the sumPµ Cλµν sµ is equal to [Mλ⊗Mν]. We can thus rephrase our last expression as follows. Let vn be the column vector (sλ,n)λ and let A be the matrix ([Mλ ⊗ Mµ])λ,µ. Then We thus have Note that v0 has a 1 in the entry λ = (p) and a 0 in all other entries. Indeed, an empty tensor product is equal to C, so sλ,0 is the class of Sλ(C) in Sym0(K(Sp)) = Q; in other words, aλ,0 = dim Sλ(C). This is 1 if λ = (p) and 0 otherwise. We therefore find that the initial vector v0 in the above recurrence is quite simple. The problem is to determine the exponential of the matrix A. We will achieve this by diagonalizing A. Let B be the matrix (χλ(cµ))λ,µ. We index by rows first, then columns. Thus the rows of B are indexed by irreducible characters and the columns by conjugacy classes; B is the character table of Sp. Let D be the diagonal matrix given by Dλλ = zλδλ where zλ = p!/#cλ is the cardinality of the centralizer of cλ and δλ is the class function on Sp which assigns cλ the value 1 and all other conjugacy classes 0. We then have the following fundamental identity (6) AB = BD. We now explain this identity. First, we regard the entries of A as class functions, so Aλµ is the character of Mλ ⊗ Mµ. The entry of the product AB at (λ, µ) evaluated at cζ is thus given by Xν AλνBνµ! (cζ) =Xν χλ(cζ)χν (cζ)χν(cµ) = χλ(cζ )Xν χν(cζ)χν (cµ). Now, for g and h in Sp the sum P χν(g)χν (h) is the trace of (g, h) acting on the representation C[Sp] of Sp × Sp. Since this is a permutation representation, the trace is given by the number of fixed points. An element x ∈ Sp is a fixed point if gxh−1 = x, or equivalently, if g = xhx−1. Thus the number of fixed points is 0 if g and h are not conjugate, and is otherwise the size of the centralizer of g. We therefore find χλ(cζ)Xν This proves (6). χν(cζ)χν (cµ) = χλ(cµ)zµδµ(cζ) = BλµDµµ(cζ). The equation (6) diagonalizes A. However, for it to be useful we need to compute B−1. This is λ . Then the orthonormality of straightforward. Let C be the diagonal matrix given by Cλλ = z−1 characters is precisely the identity BCBt = 1 and so B−1 = CBt. (This again uses the fact that all representations of symmetric groups are self-dual, which is equivalent to their characters being real valued.) We now find exp(A)v0 = B exp(B−1AB)B−1v0 = B exp(D)CBtv0. Simple matrix multiplication now gives the stated formula. (cid:3) SYZYGIES OF SEGRE EMBEDDINGS AND ∆-MODULES 33 s = [Sym2], w = [V2] 2 es−w − es f1 = 1 g1 = 1 2 es+w + 1 2 w2 s = [Sym3], w = [V3], 3 es+w+2t − 1 t = [S(2,1)] 3 es+w−t − es+t + es f2 = 1 g2 = wt a = [S(3,1)], f3 = 1 s = [Sym4], w = [V4], 8 es+w+3a+2b+3c − 1 + 1 2es+b−c − 1 g3 = aw + 1 2 c2 8 es+w−a+2b−c + 1 2 es+2a+b+c + es+a − es. (+?) b = [S(2,2)], 4 es−w−a+c − 1 c = [S(2,1,1)] 4 es−w+a−c Figure 1. Values of fp and gp for p small. Lemma 5.8. Let x and y belong to K(S). Then exp(x) ⊠ exp(y) = exp(x ⊠ y). Proof. For an object F of S let F ′ be the object of Sym(S) given by (V, L) 7→Nx∈L F (Vx). Then exp([F ]) = [F ′]. Now, let x and y in K(S) be given. Since ⊠ is additive and exp is multiplicative, it suffices to treat the case where x = [F ] and y = [G], with F and G in S. We then have exp(x) ⊠ exp(y) = [F ′] ⊠ [G′] = [F ′ ⊠ G′] = [(F ⊠ G)′] = exp([F ⊠ G]) = exp(x ⊠ y). The key fact is the obvious formula F ′ ⊠ G′ = (F ⊠ G)′. (cid:3) The proposition and corollary follow easily from the above lemmas. 5.5. Examples for small p. The main result of [Ru] states that Segre embeddings satisfy the Green -- Lazarsfeld property N3 but not N4. This means that F1, F2 and F3 are supported exactly in degrees 2, 3 and 4 respectively (and that F4 has support outside degree 5). From this, we deduce the following equalities: f1 = −χ(2), f2 = χ(3), f3 = −χ(4), f (5) 4 = χ(5). These values have been computed in Proposition 5.3. They are listed explicitly, and in simplified form, in Figure 1 (other than f (5) 4 ). We explain how the value for f1 given in the figure was derived, the values of f2 and f3 being gotten in a similar fashion. Proposition 5.3 gives (1) = s(1), while s′ f1 = − exp(s(2)) + exp(s(1) ⊠ s′ 2(cid:16)exp(s′ We have s′ (2) = s(2) − s(1,1). Now, the product s(1) ⊠ s(1) is just the usual product in K(S), i.e., it is the class of the functor V 7→ Sym1(V ) ⊗ Sym1(V ). This, of course, is equal to s(2) + s(1,1). We thus find 2 exp(s(2) + s(1,1)) + 1 (1,1) = s(2) + s(1,1) and s′ (1,1)) − exp(s′ (2))(cid:17) . (1)) − 1 2 exp(s(2) − s(1,1)) − exp(s(2)). f1 = 1 This is the value given in the figure. We have previously stated (without proof) that the equation for P1 × P1 in P3 generates F1 as a ∆-functor. This implies that g1 is the order two piece of f1. Similarly, we have stated (without proof) that any non-zero syzygy for P1 × P2 in P5 generates F2. This implies that g2 is the order two piece of f2. Finally, the order two piece of g3 is the same as that of f3; we are unaware if g3 has any terms of higher order. These remarks explain the values of g1, g2 and g3 given in the figure. 34 ANDREW SNOWDEN 6. Questions and problems (1) Are finitely generated twisted commutative algebras noetherian? We proved this for algebras generated in order 1, and can also prove it for certain algebras in order 2. (2) Are finite level ∆-ideals finitely generated? More generally, are the ∆-modules Fp for a finite level ∆-scheme concentrated in finitely many degrees? (3) Let M be a finitely generated module over a twisted commutative algebra finitely generated in order 1. We have shown that HM (t) is a polynomial of t and et. Our proof shows that the maximal power of et is related to the number of rows in M . How else does the form of HM (t) relate to the structure of M ? (4) Our series fp forgets the Sn-equivariance on Fp,n. Can one modify fp to retain this information, and still have something resembling a rationality result? (5) Is the series fp a polynomial in the sλ and the e±sλ? This does not follow from what we have proved, but one might hope that it is true based on some of our results. In fact, based on the computations of f1, f2 and f3 for the Segre, one might hope for a stronger statement: fp is a polynomial in only the e±sλ. We suspect this is false, but do not know. (6) Compute fp and gp for more values of p or for ∆-schemes other than the Segre. We have given an algorithm to do this, but it is too inefficient to use. It would be particularly interesting to compute f4 for the Segre since this is the first place where the Green -- Lazarsfeld property fails and the value is not given by the Euler characteristic formula. (7) Is the series Pi≥0(−1)i[LiΨFp]∗qi a rational function of q and the sλ? This series contains more information than fp and gp, since one can recover f ∗ one can recover g∗ p by applying Φ and setting q = 1, and p by setting q = 0. References [B] [D] M.G. Barratt, Twisted Lie algebras, in Geometric applications of homotopy theory (Proc. Conf., Evanston, Ill., 1977), II, 9 -- 15, Lecture Notes in Math., 658, Springer, Berlin, 1978. P. Deligne, Cat´egories tannakiennes, The Grothendieck Festschrift, Vol. II, 111 -- 195, Progr. Math., 87, Birkhauser Boston, Boston, MA, 1990. J. Draisma and J. Kuttler, Bounded-rank tensors are defined in bounded degree, arXiv:1103.5336, 2011. [DK] [E] D. Eisenbud, Commutative algebra, Graduate Texts in Mathematics, 150, Springer -- Verlag, New York, 1995. [ERT] D. Eisenbud, A. Reeves and B. Totaro, Initial ideals, Veronese subrings, and rates of algebras, Adv. Math. 109 (1994), no. 2, 168 -- 187. [FH] W. Fulton and J. Harris, Representation Theory: A First Course, Graduate Texts in Mathematics, 129, [GS] [J] [L] [LW] [LW2] [PW] [Ra] [Ru] in Combinatoire ´enum´erative (Montreal, Que., Springer -- Verlag, New York, 1991. V. Ginzburg and T. Schedler, Differential operators and BV structures in noncommutative geometry, Selecta Math. (N.S.) 16 (2010), no. 4, 673 -- 730. A. Joyal, Foncteurs analytiques et esp´eces de structures, 1985), 126 -- 159, Lecture Notes in Math., 1234, Springer, Berlin, 1986. A. Lascoux, Syzygies des vari´et´es d´eterminantales, Adv. in Math. 30 (1978), 202 -- 237. J.M. Landsberg and J. Weyman, On tangential varieties of rational homogeneous varieties, J. Lond. Math. Soc. 76 (2007) (2), 513 -- 530. J.M. Landsberg and J. Weyman, On the ideals and singularities of secant varieties of Segre varieties, Bull. London Math. Soc. 39 (2007) no. 4, 685 -- 697. P. Pragacz and J. Weyman, Complexes associated with trace and evaluation. Another approach to Lascoux's resolution, Adv. in Math. 57 (1985), 163 -- 207. C. Raicu, The GSS conjecture, arXiv:1011.5867, 2010. E. Rubei, Resolutions of Segre embeddings of projective spaces of any dimension, J. of Pure and Applied Algebra 208 (2007), 29 -- 37.
1706.02232
3
1706
2018-06-29T17:38:23
Vinberg's X_4 Revisited
[ "math.AG" ]
The article covers the unique complex K3 surface with maximal Picard rank and discriminant four. We discuss smooth, rational curves and identify generators of its automorphism group with certain Cremona transformations of $\mathbb{P}^2$. This gives a geometric perspective of Vinberg's results.
math.AG
math
VINBERG'S X4 REVISITED ŁUKASZ SIENKIEWICZ ABSTRACT. The article covers the unique complex K3 surface with maximal Picard rank and dis- criminant four. We discuss smooth, rational curves and identify generators of its automorphism group with certain Cremona transformations of P2. This gives a geometric perspective of Vinberg's results [Vin83]. INTRODUCTION In [SI77] Shioda and Inose proved a classification theorem for complex K3 surfaces with maximal Picard rank in terms of their transcendental lattices. In the course of the proof they discussed two K3 surfaces with maximal Picard rank which are the simplest in the sense that their tran- scendental lattices have the smallest possible discriminants equal to 3 and 4. Then Vinberg in his article [Vin83] called these surfaces the most algebraic K3 surfaces. He gave a complete descrip- tion of automorphism groups of these two surfaces as well as several examples of their birational models. In the article [OZ96] the authors classified these surfaces in terms of ramification lo- cus of some special automorphisms of these surfaces and related them to extremal log Enriques surfaces. Recently in [DBvGK+17] the authors used the Hilbert scheme of two points on the K3 surface having discriminant equal to four to solve the question concerning the cardinality of com- plete families of incident planes in P5 originally posed by Morin in [Mor30]. This is related to O'Grady's research [O'G06] on Hyperkähler manifolds. This article is devoted to study of a unique K3 surface X4 with Picard rank 20 and discriminant equal to four. We construct X4 as a double covering π ∶ X4 → Y4 of a smooth, rational surface Y4. Surface Y4 is defined as a blow up m ∶ Y4 → S5 of the Del Pezzo surface S5 of degree 5 in all intersection points of configuration of (−1)-curves on S5. In the book [AN06] Alexeev and Nikulin describe certain log Del Pezzo surfaces as quotients of K3 surfaces by non-symplectic in- volutions. In particular, Y4 is a log Del Pezzo surface. So the above construction is an important special case of their procedure. In this paper we describe geometrically generators of the group Aut(X4). The following theorem relates automorphisms of X4 and Y4. Here σ is a unique nontrivial automorphism of the covering π ∶ X4 → Y4. Theorem A (Theorem 5.4). There exists a short exact sequence of groups 1 ⟨σ⟩ Aut(X4) θ Aut(Y4) 1 and⟨σ⟩ is a central subgroup of Aut(X4). The covering morphism π has ten branch curves Fij for 0 ≤ i < j ≤ 4 and these are all (−4)-curves b be the group of bijections of the set b of branch curves of π. The on Y4 (Proposition 5.1). Let Σ action of Aut(Y4) on (−4)-curves of Y4 gives rise to a homomorphism of groups Aut(Y4) → Σ b. We define a subgroup G ⊆ Σb as the image of an injective homomorphism Σ5 → Σb sending τ ∈ Σ5 to the bijection given by Fij ↦ Fτ(i)τ(j). Group G can be also identified with the group of automorphisms of the Petersen graph (Figure 1). The author was supported by Polish National Science Center project 2013/08/A/ST1/00804. 1 2 ŁUKASZ SIENKIEWICZ b is G. Moreover, the epi- Theorem B (Theorem 5.6). The image of the homomorphism Aut(Y4) → Σ morphism Aut(Y4)→ G admits a section given by homomorphism G ≅ Aut(S5)→ Aut(Y4) that lifts an automorphism of S5 along birational contraction m ∶ Y4 → S5. Let Q be the quadro-quadric Cremona transformation of P2. Clearly Q is a birational involution of P2. It turns out (Proposition 5.7) that Q induces a regular involution fQ of Y4. Its lift fQ to an automorphism of X4 induces a hyperbolic reflection of the hyperbolic space associated with R (X4) (Corollary 5.8). Our last result identifies the hyperbolic reflection induced by fQ with a H1,1 reflection described in [Vin83]. R (X4) Theorem C (Corollary 5.9). Reflection induced by f ∗ is conjugate to reflections contained in S1 according to Vinberg's notation [Vin83, Section 2.2]. Let SX4 be the lattice of algebraic cycles of X4 and let O+(SX4) be the group of orthogonal auto- morphisms of SX4 that preserve the ample cone. According to Vinberg [Vin83, Theorem 2.4] we have an isomorphism of groups Q on the hyperbolic space associated with H1,1 O+(SX4)≅(Z~2Z⋆ ...⋆ Z~2Z) ´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶ 5 times ⋊Σ5 where five copies of Z~2Z in the free product are generated by reflections in S1 and Σ5 transi- tively permutes the factors of the free product. Thus Theorems B and C identify factors of this semidirect product in terms of autmorphisms of Del Pezzo surface S5 and quadratic transforma- tions of the plane P2. 1. PRELIMINARIES IN THE THEORY OF K3 SURFACES AND HYPERBOLIC GEOMETRY X ≅OX and H1(X,OX) = 0. In this paper a K3 surface is a smooth, projective surface X over C such that Ω2 Exhaustive presentation of the theory of K3 surfaces is [Huy16]. Fix a K3 surface X. It follows that H2(X, Z) is a free Z-module of rank 22 and cup product yields to intersection pairing on H2(X, Z). The transcendental lattice TX is the sublattice of H2(X, Z) orthogonal to the Neron- Severi lattice SX = NS(X) with respect to the cup product. The rank of SX is called the Picard number of X and is denoted by ρ(X). Let O+(SX) be the group of isometries of SX that preserve the ample cone. Suppose that ωX is a nontrivial holomorphic two-form on X. We define UX = {α ∈ O(TX)(1C⊗Z α)(ωX) ∈ CωX}. For every lattice L we denote by L∨ its dual. The groups are canonically identified. We call DX = H2(X, Z)~(SX+ TX) the discriminant group of X and we call its orderDX the discriminant of X. Note that every automorphism of X yields in an obvious way an element of O+(SX) as well as an element of UX. Moreover, every orthogonal transforma- X~TX respectively i.e. an automorphism tion of SX or TX yields an automorphism of S∨ of DX. Proposition 1.1 ([Vin83, Section 1.5, Formula (7)]). We have a cartesian square of abstract groups X~SX, H2(X, Z)~(SX+ TX), T∨ S∨ X~SX or T∨ X~TX Aut(X) O+(SX) UX Aut(DX) Our results significantly use the theory of elliptic fibrations on K3 surfaces. Recall that a proper and flat morphism with connected fibers p ∶ X → C defined on a surface X is an elliptic fibration if and only if its general fiber is an elliptic curve. We extensively use Kodaira classification of singular fibers of elliptic fibrations cf. [BHPVdV04, Chapter V, Section 7]. Next results describe elliptic fibrations on K3 surfaces. VINBERG'S X4 REVISITED 3 Theorem 1.2 ([PŠŠ71, Section 3, Theorem 1]). Let X be a projective K3 surface and L a line bun- dle which is nef and L2 = 0. Then L is base point free and the corresponding morphism φL ∶ X → PH0(X,L) factors as an elliptic fibration p ∶ X → P1 followed by a finite morphism P1 → PH0(X,L). Corollary 1.3 ([SI77, Lemma 1.1]). Let D be an effective divisor on a K3 surface X. Assume that D is not equal to a multiple of any other divisor. Suppose that D as a scheme has an isomorphism type of a singular fiber of some elliptic fibration. Then there exists an elliptic fibration p ∶ X → P1 such that D is a Theorem 1.4 ([Shi72], [SI77, Lemma 1.3]). Let p ∶ X → P1 be an elliptic fibration on a K3 surface and Di for 1 ≤ i ≤ k its singular fibers. Let mi and m(1) components of Di and the number of irreducible components of Di having multiplicity one. (1) Let r(p) be the torsion-free rank of the group of sections of p. Then the following formula holds. denote respectively the number of irreducible singular fiber of p. i ρ(X) = 2+ r(p)+ k Q i=1(mi− 1) (2) Moreover, if r(p) = 0 and n(p) denotes the order of the group of sections of p, then the following formula holds. Proposition 1.5 ([Keu00, Theorem 2.3]). Every elliptic fibration on a K3 surface X with ρ(X) = 20 and with discriminant equal to four or three admits a section. det(TX) = i=1 m(1) ∏k n(p)2 i In the last part of section 5 we use certain results concerning explicit models of hyperbolic geom- etry. For the reference cf. [Dol08, Section 2.2] or [Vin83, Section 1.3]. Pick n ∈ N and let E be a real vector space of dimension n+ 1 equipped with bilinear pairing (−,−) of signature (1, n). Then the set {x ∈ E (x, x) > 0} has two connected components and let C+ be one of them. Then we define Hn = C+~R>0 i.e. we consider vectors in C+ up to positive multiplicative constant. Proposition 1.6 ([Dol08, Section 2.2]). Bilinear form(−,−) induces a Riemannian metric on Hn. This construction gives rise to a Riemannian manifold with constant negative curvature. We call Hn an n-dimensional hyperbolic space. Corollary 1.7. Let X be a K3 surface. Pick E = H1,1(X) and choose C+ to be the connected component of {c ∈ H1,1(X) (c, c) > 0} that contains the ample class of X. Then C+~R>0 yields a model of a hyperbolic space. A reflection in a Riemannian manifold of constant curvature is a nontrivial order two isometry preserving every point inside some totally geodesic hypersurface [Dol08, Section 2.2]. Proposition 1.8 ([Vin83, Section 1.3]). For every vector e ∈ E such that(e, e) < 0 linear map E ∋ x ↦ x− 2(e, x) (e, e) e ∈ E induces a reflection of Hn. Moreover, every reflection of Hn is induced in such a way from a linear map on E. 2. CYCLIC COVERINGS AND n-TH ROOT OF A SECTION In this section we present material leading to important result concerning lifting of automor- phisms. The first result of this section is a part of the folklore and can be extracted from presen- tation of cyclic coverings in [Laz04, Section 4.1B]. 4 ŁUKASZ SIENKIEWICZ (1) tn E = q∗s morphism h ∶ X → Wn(E, s) in the category of schemes over Y such that t = h∗tE . Proposition 2.1. Let Y be a scheme, E be a locally free sheaf on Y and s ∈ Γ(Y, Symn(E)) be a global section for some n ∈ N. Then there exists a scheme q ∶ Wn(E, s) → Y over Y and a section tE ∈ Γ(Wn(E, s), q∗E) such that (2) For every morphism g ∶ X → Y and a section t ∈ Γ(X, g∗E) such that tn = g∗s there exists a unique If L is a line bundle and D is a divisor of zeros of some section s ∈ Γ(Y,L⊗n), then we call q ∶ Wn(L, s) → Y a cyclic covering of Y corresponding to L branched along D. Note that if Y is global section s ∈ Γ(Y,L⊗n) having D as the divisor of zeros. We use the notion of cyclic covering in the following special case. Definition 2.2. Let Y be a smooth and proper variety over C. We denote by ωY the sheaf of algebraic differential forms of the highest possible degree on Y i.e ωY = OY(KY), where KY is the canonical divisor on Y. Fix integer n ∈ N. Let D be an effective divisor linearly equivalent to −nKY. Then the cyclic covering of Y branched along D and corresponding to ω∨ Y is called the a complete variety over C, then the notion of cyclic covering does not depend on the choice of a anticanonical cyclic covering of Y. The following proposition is used in the section 5. Proposition 2.3. Let Y be a smooth, proper variety over C. Let D be an effective divisor such that D ∼ −nKY for some n ∈ N. Denote by q ∶ X → Y the anticanonical cyclic covering branched along D. Suppose that f is an automorphism of Y such that f ∗D = D. Then there exists an automorphism f of X such that the following square is commutative. X q Y f f X q Y Proof. Let s be a global section of (ω∨ X = Wm(ω∨ Y)⊗n ≅ OY(−nKY) having D as its divisor of zeros. Then Y, s). It follows from the universal property described in Proposition 2.1 that we have a base change diagram Wn( f ∗ω∨ S , f ∗s) Y f ′ f Wn(ω∨ S , s) Y ing to the fact that f ∗D = D we derive that f ∗s has D as the divisor of zeros. Hence there Y induces an isomorphism f ∗ωY → ωY. Y → Ω1 Next observe that the cotangent morphism f ∗Ω1 Dualizing we derive that there exists an isomorphism φ ∶ ω∨ Y. Since φ⊗n is an isomor- Y)⊗n also has D as the divisor of zeros. Accord- phism, we derive that section φ⊗n(s) of f ∗(ω∨ exists α ∈ C∗ such that f ∗s = αφ⊗n(s) = φ⊗n(αs). Therefore, again by universal property of Proposition 2.1 map φ induces an isomorphism f ′′ ∶ Wm(ω∨ Y, f ∗s) of schemes Y, αs) of schemes over Y. Now the composition f = f ′ ⋅ f ′′ ⋅ f ′′′ is a f ′′′ ∶ Wm(ω∨ over Y. Finally, since αs and s have the same divisor of zeros, there exists an isomorphism Y, αs) → Wm( f ∗ω∨ Y, s) → Wm(ω∨ Y → f ∗ω∨ lift of f . (cid:3) VINBERG'S X4 REVISITED 5 3. CONSTRUCTION OF X4 In this section we construct X4 explicitly as a double covering of some rational surface. Consider four points P = {p1, p2, p3, p4} of P2 such that no three of them are on the same line. Blow them up to get a Del Pezzo surface S5 = BlP(P2) = Blp1,p2,p3,p4(P2) of degree 5. For 1 ≤ i < j ≤ 4 denote by Eij the strict transform on BlP(P2) of a line on P2 passing through points P ∖{pi, pj} = {p1, p2, p3, p4} ∖{pi, pj}. For 1 ≤ i ≤ 4 denote by E0i the exceptional divisor of BlP(P2) over pi. Proposition 3.1 ([Dol12, Section 8.5.1]). The following assertions hold. (1) Curves Eij for 0 ≤ i < j ≤ 4 are all irreducible (−1)-curves on S5. (2) The divisor E = ∑0≤i<j≤4 Eij is linearly equivalent to −2KS5. (3) The incidence graph of curves{Eij}0≤i<j≤4 is the Petersen graph in the Figure 1. (01) (23) (34) (02) (04) (24) (13) (14) (Figure 1) (12) (03) Three thick edges in the Figure 1 describe three linearly equivalent divisors whose complete linear system defines a fibration S5 → P1 with general fiber being smooth and rational curve. It is usually called a conic fibration due to the fact that its fibers are of degree two with respect to −KS5. Let m ∶ Y4 → S5 be the blowing up of intersection points of curves {Eij}0≤i<j≤4. Let Fij be a strict transform of Eij in Y4 for every 0 ≤ i < j ≤ 4. For any {i, j}, {k, l} ⊆ {0, 1, 2, 3, 4} such that {i, j} ∩{k, l} = ∅ we denote by F(ij)(kl) a curve on Y4 which is the preimage of the intersection point Eij ∩ Ekl. Next result is a consequence of the Proposition 3.1. Proposition 3.2. The following statements hold. (1) Fij are pairwise disjoint smooth rational(−4)-curves on Y4 for 0 ≤ i < j ≤ 4. (2) The divisor ∑0≤i<j≤4 Fij is linearly equivalent to −2KY. 6 ŁUKASZ SIENKIEWICZ (3) The incidence graph of curves Fij and F(ij)(kl) for 0 ≤ i < j ≤ 4, 0 ≤ k < l ≤ 4 and {i, j} ∩{k, l} = ∅ is the extended Petersen graph in the Figure 2. (01) (01)(23) (23) (02)(13) (23)(04) (23)(14) (01)(24) (24)(13) (13) (24) (01)(34) (34) (34)(02) (02) (34)(12) (02)(14) (13)(04) (24)(03) (04) (04)(12) (14) (14)(03) (Figure 2) (12) (12)(03) (03) the unique automorphism of π having order two. Definition 3.3. We define X4 to be the anticanonical degree two covering of Y4 branched along Bπ = ∑0≤i<j≤4 Fij. We denote by π ∶ X4 → Y4 the anticanonical covering map and by σ ∶ X4 → X4 Since Fij are branch curves of π, we derive that π∗Fij = 2Lij for some (−2)-curve Lij on X4 for 0 ≤ i < j ≤ 4. These curves are pairwise disjoint and ∐0≤i<j≤4 Lij is a fixed locus of σ on X4. Proposition 3.4. X4 is a K3 surface with ρ(X4) = 20. The automorphism σ acts as identity on the Picard group of X4 and every line bundle L on X4 is linearizable with respect to the group{1X4, σ}. Proof. Note that the branch divisor of π is smooth. Hence X4 is a smooth surface. Clearly it is projective as a finite covering of a projective surface Y4. First by Riemann-Hurwitz formula and Proposition 3.2 we derive that 1 2 Q 0≤i<j≤4 Ω2 ⎝ Q 0≤i<j≤4 X4 = π∗Ω2 Y4 π∗OY4⎛ ⎝ Lij⎞ ⎠ ≅ π∗OY4(KY4) ⊗OX4 Fij⎞ ⊗OX4 OX4⎛ ⎠ ≅ π∗OY4 ≅OX4 Next π∗OX4 = OY4 ⊕OY4(KY4) and H1(Y4,OY4(KY4)) = H1(Y4,OY4) = 0 by Serre duality. Hence H1(X4, π∗OX4) = 0. Now by Leray spectral sequence we have H1(X4,OX4) ≅ H1(π∗OX4) = 0. Therefore, X4 is a K3 surface. Note that rankZ(Cl(Y4)) = 20 and we have morphisms of abelian groups π∗ ∶ Cl(Y4) → Cl(X4) and π∗ ∶ Cl(X4) → Cl(Y4) such that π∗π∗ = deg(π)1Cl(Y4) = 2 ⋅ 1Cl(Y4). Thus π∗ is a monomor- phism. Since Y4 and X4 are smooth, we derive that π∗(Pic(Y4)) = π∗(Cl(Y4)) ⊆ Cl(X4) = Pic(X4) is a subgroup of rank 20. Thus ρ(X4) = 20. For every line bundle L on Y4 we have σ∗(π∗L) =(πσ)∗(L) = π∗L Thus σ acts as identity on π∗(Pic(Y4)). Since this is a subgroup of maximal rank in the torsion free group Pic(X4), we derive that σ∗ = 1Pic(X4). Denote by ⟨σ⟩ the subgroup of Aut(X4) generated by σ. Let Pic⟨σ⟩(X4) be the group of ⟨σ⟩- linearized line bundles on X4 and let Pic(X4)⟨σ⟩ be the group of line bundles on X4 invariant with respect to the action of ⟨σ⟩. Since σ∗ = 1Pic(X4), we derive that Pic(X4)⟨σ⟩ = Pic(X4). Now according to [Dol03, Remark 7.2] there is an exact sequence VINBERG'S X4 REVISITED 7 0 Hom(⟨σ⟩ , C∗) Pic⟨σ⟩(X4) Pic(X4)⟨σ⟩ H2(⟨σ⟩ , C∗) where the arrow Pic⟨σ⟩(X4) → Pic(X4)⟨σ⟩ = Pic(X4) forgets about the ⟨σ⟩-linearization and C∗ is a trivial ⟨σ⟩-module. According to ⟨σ⟩ ≅ Z~2Z and [Wei94, Theorem 6.2.2] we derive that if C∗ admits trivial action of Z~2Z, then H2(Z~2Z, C∗) = 0. This proves that arrow Pic⟨σ⟩(X4) → Pic(X4)⟨σ⟩ = Pic(X4) is surjective and hence every line bundle on X4 admits a ⟨σ⟩-linearization. Lemma 3.5. Let F be a curve on Y4 with negative self intersection. Then either π∗F = 2L or π∗F = L for some(−2)-curve L on X4. The first case holds if and only if F is a branch curve of π. (cid:3) Proof. There are three possibilities. (1) π∗F = 2L where L is a curve on X4 and the map L → F induced by π is birational. (2) π∗F = L where L is a curve on X4 and the map L → F induced by π is of degree two. (3) π∗F = L1 + L2 where L1, L2 are distinct curves on X4. 2 + 2L1.L2 = (π∗F)2 < 0. Observe that σ∗L1 = L2 as σ acts transi- Note that in (3) we have L2 tively on fibers of π. According to the fact that σ∗ = 1Pic(X4), we derive that L1 ∼ L2 and hence L2 1 + L2 2 + 2L1.L2 = 4L1.L2 > 0 a contradiction. So the only possibilities remaining are (1) and (2). In both cases L2 < 0 and since X4 is a K3 surface, we derive that L is a(−2)-curve on X4. Moreover, (1) holds if the ramification index of π at the generic point of L is 2 and this implies that F is a branch curve of π. (cid:3) 1 + L2 Corollary 3.6. Every curve L(ij)(kl) is a smooth rational curve on X4. For {i, j},{k, l} ⊆{0, 1, 2, 3, 4} and{i, j} ∩{k, l} = ∅ we define L(ij)(kl) = π∗F(ij)(kl). The incidence graph of curves Lij for 0 ≤ i < j ≤ 4 and L(ij)(kl) for {i, j}, {k, l} ⊆ {0, 1, 2, 3, 4} and {i, j} ∩{k, l} = ∅ is the extended Petersen graph. Proof. The first assertion is a direct consequence of Lemma 3.5 and the fact that F2 (ij)(kl) = −1. The second assertion follows from Proposition 3.2 and the fact that π∗ preserves intersection pairing up to multiplication by degree of π. (cid:3) Table 3.7. The following table collects information about all curves defined so far. S5 (Prop. 3.1) Y4 (Prop. 3.2) X4 (Cor. 3.6) Eij for 0 ≤ i < j ≤ 4. These are all (−1)- curves on S5. The strict transform Fij of Eij for 0 ≤ i < j ≤ 4 . These are smooth, ra- tional(−4)-curves. Lij = 1 2 π∗Fij for 0 ≤ i < j ≤ 4. These are (−2)- curves. The Petersen graph Figure 1 describes their intersection. extended The Pe- tersen graph Figure 2 describes their inter- section. extended The Pe- tersen graph Figure 2 describes their inter- section. Blow up F(ij)(kl) of in- tersection point Eij ∩ Ekl for {i, j}, {k, l} ⊆ {0, 1, 2, 3, 4} and{i, j} ∩ {k, l} = ∅. These are (−1)-curves. L(ij)(kl) = π∗F(ij)(kl) for {i, j}, ⊆ {0, 1, 2, 3, 4} and {i, j} ∩ {k, l} ∅. These are(−2)-curves. {k, l} = 8 ŁUKASZ SIENKIEWICZ Proposition 3.8. The cup product pairing restricted to the transcendental lattice TX4 is given by the matrix  2 0 0 2  with respect to some basis of TX4. Proof. Let us come back to the Del Pezzo surface S5 = BlP(P2). The following divisors: E01 + E23, E02 + E13, E12 + E03 are linearly equivalent. Their linear system is base point free and its members consist of degree two divisors with respect to very ample divisor −KS5 and gives rise to the morphism f ∶ S5 → P1 whose fibers are curves of degree two with respect to −KS5. This conic bundle has exactly three singular fibers given by the three divisors listed above. The morphism p = f ⋅ m ⋅ π is an elliptic fibration. Recall that by Proposition 3.6 the incidence graph of curves Lij and L(ij)(kl) for 0 ≤ i < j ≤ 4, 0 ≤ k < l ≤ 4 and {i, j} ∩{k, l} = ∅ is the extended Petersen graph. There are exactly three singular fibers of p given by pulling back along mπ singular fibers of f . They are depicted in the Figure 3. 1 2 s 1 2 2 2 2 1 1 1 1 1 2 1 s 1 1 s 1 s 1 2 2 (Figure 3) Here colored and thick subgraphs correspond to three singular fibers. Each colored vertex is labeled by its multiplicity in the corresponding fiber. Moreover, there are four black vertices that are connected with each colored subgraph by precisely one edge. These vertices correspond to sections of the fibration and are labeled by letters "s". Hence p has precisely three singular fibers each of type D6 according to Kodaira classification [BHPVdV04, Chapter V, Section 7] and it has at least four distinct sections. Using Theorem 1.4, we deduce that 2 where r(p) is the torsion free rank of the Mordell-Weil group of p. Hence r(p) = 0. Let n(p) be the order of the Mordell-Weil group of p. Then 4 ≤ n(p) due to existence of four distinct sections of p. Again using Theorem 1.4, we derive that 20 = r(p) + 2 + 3 ⋅ 6 and det(TX4) is a divisor of 43. Hence det(TX4) = 1, 2, 4. Now the fact that TX4 is a rank two, even, positive definite integral lattice, implies that det(TX4) = 4. There exists precisely one rank two, even, positive definite integral lattice with discriminant equal to 4 and it has a basis in which intersection form has the matrix det(TX4) = 43 n(p)2 ≤ 43 42 = 4  2 0 0 2  (cid:3) VINBERG'S X4 REVISITED 9 Remark 3.9. Since σ∗ acts as the identity on the algebraic lattice SX4, this implies that it also acts as the identity on the discriminant group DX4. On the other hand σ ∶ X4 → X4 is a non-symplectic involution of X4. Hence σ∗ induces multiplication by −1 on TX4. Thus it acts as multiplication by −1 on DX4. Therefore, we derive that DX4 is a direct sum of copies of Z~2Z. Now one can use Nikulin's formula in [Nik81, Theorem 4.2.2] (see also [Nik79]) to give an alternative proof of Proposition 3.8. Corollary 3.10. The K3 surface X4 constructed in this section is isomorphic to the unique K3 surface with maximal Picard rank and with discriminant equal to four. Proof. This follows from Proposition 3.8 and [SI77, Theorem 4]. (cid:3) 4. ELLIPTIC FIBRATIONS ON X4 morphism. Corollary 4.1. Let p ∶ X4 → P1 be an elliptic fibration. We consider X4 as a variety equipped with the action of ⟨σ⟩ = {1X4, σ}. Then there exists an action of Z~2Z on P1 such that p is an equivariant Proof. Let L = p∗OP1(1). Since p is a morphism with connected fibers we derive that H0(X4,L) is isomorphic to H0(P1,OP1(1)). According to Proposition 3.4 the line bundle L admits a ⟨σ⟩- linearization. Thus there exists a linear action of ⟨σ⟩ on global sections of L and there is a mor- phism H0(X4,L) ⊗C OX4 → L of sheaves with ⟨σ⟩-linearizations. Since P1 is the projectivization PH0(X4,L), we deduce that there exists an action of Z~2Z on P1 such that p is equivariant. (cid:3) Let p ∶ X4 → P1 be an elliptic fibration and τ ∶ P1 → P1 the, possibly trivial, involution inducing the Z~2Z action on P1 which makes p equivariant. Then there exists a fibration q ∶ Y4 → P1 such that the following diagram is commutative X4 p P1 π r Y4 q P1 where r ∶ P1 → P1 is the quotient morphism with respect to the action of τ on P1. We will call q the fibration induced by p. Clearly q is either an elliptic fibration or a fibration with general fiber being smooth, rational curve. Proposition 4.2. Let p ∶ X4 → P1 be an elliptic fibration and q ∶ Y4 → P1 the induced fibration. Then the following hold. (1) τ = 1P1 if and only if q is a fibration with general fiber being smooth rational curve. In this case every section of p is a ramification curve of π. (2) τ ≠ 1P1 if and only if q is an elliptic fibration. In this case there are at most two reducible fibers of p and they are preimages of fixed points of τ. Proof. According to Proposition 3.8 and Proposition 1.5 we deduce that every elliptic fibration on X4 admits a section. Suppose that the action of Z~2Z on P1 is trivial. In particular, we have We derive that p(σ(Fu)) = p(Fu) where u ∈ P1 is a point and Fu = p−1(u) is a fiber. This implies that σ(Fu) = Fu. Consider now a section s ∶ P1 → X4 of p. If v = s(u), then {v} = Fu ∩ s(P1). Hence σ(v) ∈ σ(Fu) = Fu. On the other hand s(P1) is a smooth, rational curve on a K3 hence p ⋅ σ = q ⋅ π ⋅ σ = q ⋅ π = p 10 ŁUKASZ SIENKIEWICZ smooth, rational curve. This implies that q is a fibration with general fiber being smooth, rational curve. a (−2)-curve. Since σ∗ = 1Pic(X4), we derive that σ(s(P1)) = s(P1). According to v ∈ s(P1), we deduce that σ(v) ∈ σ(s(P1)) = s(P1). Thus σ(v) ∈ Fu ∩ s(P1) = {v} and hence v is a fixed point of σ. Therefore, every point of s(P1) is a fixed point of the action of σ. This implies that s(P1) is a ramification curve of π. Next note that Fu = p−1(u) and Cu = q−1(r(u)) = q−1(u) are smooth curves if one chooses u ∈ P1 to be sufficiently general. Hence the morphism Fu → Cu induced by π is a ramified morphism of smooth curves. Thus we have g(Cu) < g(Fu) = 1. Hence Cu is a Suppose now that the action of Z~2Z on P1 is nontrivial. Then it is given by some nontrivial involution τ of P1. Pick u ∈ P1 such that Fu = p−1(u) is smooth and τ(u) ≠ u. We deduce that Fτ(u) = σ(Fu) is smooth. Moreover, we have where Cr(u) = q−1(r(u)) is some curve on Y4. It is clear that if one chooses u to be sufficiently general, then one may assume that Cr(u) is smooth. Let eu, eτ(u) and fu, fτ(u) be ramification indexes and inertia indexes of π at generic points of Fu and Fτ(u). Then we have formula [Liu02, 7.4.2, Formula 4.8] π(Fu) = Cr(u) = π(Fτ(u)) fueu + fτ(u)eτ(u) = 2 elliptic fibration. Finally note that by Kodaira classification [BHPVdV04, Chapter V, Section 7] Hence eu = eτ(u) = fu = fτ(u) = 1 and morphisms Fu → Cr(u) and Fτ(u) → Cr(u) induced by π are isomorphisms. This implies that Cr(u) = q−1(r(u)) is a smooth elliptic curve. Hence q is an reducible fibers of p correspond to unions of (−2)-curves on X4. In particular, reducible fibers of p are invariant under the action of σ. Hence every reducible fiber of p is contracted by p to a fixed point of τ. According to the fact that τ is a nontrivial involution of P1, it has two fixed points. Thus there are at most two reducible fibers of p. (cid:3) Both types of elliptic fibrations described in the previous proposition are realized on X4. For this observe that the elliptic fibration described in Proposition 3.8 is of the first type. Now consider divisors D1 and D2 corresponding to subgraphs of the configuration described in Proposition 3.6 and depicted by colored and thick parts in the Figure 4. (01) (01)(23) (23) (02)(13) (23)(04) (23)(14) (01)(24) (24)(13) (13) (24) (01)(34) (34) (34)(02) (02) (34)(12) (02)(14) (13)(04) (24)(03) (04) (04)(12) (14) (14)(03) (12) (03) (12)(03) (Figure 4) We have that D1.D2 = 0 and both these divisors are of Kodaira [BHPVdV04, Chapter V, Section 7] type A9. According to Corollary 1.3 there exists an elliptic fibration p ∶ X4 → P1 such that two of its fibers are precisely divisors D1 and D2. Since D1 ∪ D2 contains all ramification curves of π, we deduce that this elliptic fibration is of type (2) with respect to Proposition 4.2. Corollary 4.3. Y4 is an elliptic rational surface. VINBERG'S X4 REVISITED 11 Proof. According to Proposition 4.2 fibration of type (2) induces an elliptic fibration on Y4. Since by previous discussion fibrations of type (2) exist, we derive the assertion. (cid:3) 5. AUTOMORPHISM GROUP OF X4 AND CREMONA GROUP In order to describe automorphism group of X4 in geometric terms we need to prove first certain result on negative curves on Y4 and X4. Recall that π ∶ X4 → Y4 denotes the anticanonical covering of degree two. Proposition 5.1. Let F be a curve on Y4 such that F2 < 0. Then F is a smooth rational curve and F2 ∈{−1, −4}. Moreover, these curves fit into the following classes. (b) F2 = −4 and F is a branch curve of π i.e F = Fij for some 0 ≤ i < j ≤ 4. (f1) F2 = −1 and F intersects exactly one of the curves in the set {Fij}0≤i<j≤4 at two distinct points and transversally F Fij (f2) F2 = −1 and F intersects exactly two curves in the set{Fij}0≤i<j≤4 each at one point and transversally Fij Fkl F Let L be a curve on X4 such that, L2 < 0. Then L is a smooth rational curve and L2 = −2. Moreover, these curves fit into the following classes. (r) L is a ramification curve of π i.e L = Lij for some 0 ≤ i < j ≤ 4. (l1) L intersects exactly one of the curves in the set {Lij}0≤i<j≤4 at two distinct points and transversally (l2) L intersects exactly two curves in the set {Lij}0≤i<j≤4 each at one point and transversally L Lij We have bijections Lij Lkl L 1 2 b ∋ F ↦ L = π∗F ∈ r f1 ∋ F ↦ L = π∗F ∈ l1 f2 ∋ F ↦ L = π∗F ∈ l2 Proof. According to Lemma 3.5 we have two cases: (1) π∗F = 2L where L is a curve on X4 and the map L → F induced by π is birational. (2) π∗F = L where L is a curve on X4 and the map L → F induced by π is of degree two. In case (1) we have F2 = 2L2 = −4 and F is a component of branch divisor Bπ. Thus F = Fij for some 0 ≤ i < j ≤ 4. This gives the class b and the corresponding class r. Let us study (2). In this case we derive F2 = 1 2 L2 = −1. Since π∗F = L, the ramification index of a local parameter at generic point of F is equal to 1. This means that F is not contained in branch 12 ŁUKASZ SIENKIEWICZ locus of π and hence 0 ≤ F.Bπ = −2F.KY4. Hence F.KY4 ≤ 0. By formula on arithmetic genus we have 2pa(F) − 2 = F2 + F.KY4 ≤ −1 Thus pa(F) = 0 and F.KY4 = −1. This implies that F.Bπ = 2. Moreover, morphism L → F induced by π is a degree two cyclic covering of F branched along divisor BπF. Since L is smooth, we derive that BπF must be smooth. Hence BπF is a union of two points each with multiplicity one. This gives classes f1 and f2. Note that every(−2)-curve L on X4 is either the pullback or half the pullback of a negative curve on Y4. This gives classes l1 and l2 corresponding to f1 and f2. (cid:3) Remark 5.2. All these classes are nonempty. For this note that class r is clearly nonempty. More- over, L(ij)(kl) is a curve of type l2 that intersects Lij and Lkl for disjoint {i, j}, {k, l} ⊆ {0, 1, 2, 3, 4}. We now prove that for 0 ≤ i < j ≤ 4 there exists a curve L ∈ l1 that intersects Lij. Without loss of generality we may assume that i = 3 and j = 4. Consider the extended Petersen graph with some extra decorations as shown in the Figure 5. (01) (01)(23) (23) (01)(34) Red (01)(24) L (34) (34)(02) (02) (02)(13) (24)(13) (13) (24) (23)(04) (23)(14) (34)(12) (02)(14) (13)(04) (24)(03) (04) (04)(12) (14) (14)(03) (12) (03) (12)(03) (Figure 5) The red thick subgraph corresponds to a divisor D of type A17. According to Corollary 1.3, D is a singular fiber of some elliptic fibration p ∶ X4 → P1. Note that p is of type (2) with respect to Proposition 4.2. Indeed, ramification curve L34 of π does not intersect with D hence it could not be a section of p. On the other hand any other ramification curve Lij of π is contained in D. Thus sections of D are not ramification curves of π and this shows that p is of type (2). Since p is of type (2) in Proposition 4.2, we derive that there exists reducible fiber D′ of p distinct from D. Using notation as in discussion preceding Proposition 4.2, we derive that p(D) and p(D′) are fixed points of τ. Since p is equivariant, we derive that p(L34) is a fixed point of τ. Thus L34 ⊆ D′. According to the fact that D contains all ramification curves of π except L34 and due to Proposition 5.1, fiber D′ must be of type A1 with respect to Kodaira classification [BHPVdV04, Chapter V, Section 7]. Hence there exists a smooth, rational curve L such that L ∪ L34 is of type A1. We denote L ∪ L(34) by blue dotted subgraph on the picture above. Clearly L ∈ l1. We are ready to prove main results of this work. Fact 5.3. The nontrivial automorphism σ ∶ X4 → X4 of the covering π is a central element of Aut(X4). Proof. Recall that according to [Huy16, Chapter 3, Corollary 3.4] the group of Hodge isometries of the transcendental lattice of a projective K3 surface is finite and cyclic. In particular, the image , we derive of σ under Aut(X4) → UX4 is central. According to the fact that σ∗ = 1Pic(X4) = 1SX4 that σ∗ ∈ O+(SX4) is also central. According to Proposition 1.1 there is an isomorphism Aut(X4) ≅ UX4 ×Aut(DX4 ) O+(SX4) VINBERG'S X4 REVISITED 13 It follows that σ is a central element of Aut(X4). Let f ∈ Aut(X4) be an automorphism. Then there exists a unique automorphism θ( f) ∈ Aut(Y4) such that π ⋅ θ( f) = f ⋅ π. Indeed, Y4 is a quotient of X4 with respect to ⟨σ⟩ = {1X4, σ} ⊆ Aut(X4) and this is a central subgroup of Aut(X4) by Fact 5.3. Hence every f ∈ Aut(X4) is equivariant with respect to action of this subgroup. Finally every equivariant automorphism induces a unique automorphism of a quotient. Theorem 5.4. There exists a short exact sequence of groups (cid:3) 1 ⟨σ⟩ Aut(X4) θ Aut(Y4) 1 Proof. By the discussion above it remains to prove that θ is onto. Recall that Bπ denotes the branch and⟨σ⟩ is a central subgroup of Aut(X4). divisor of π. Since Bπ = ∑0≤i<j≤4 Fij and Fij for 0 ≤ i < j ≤ 4 are the only (−4)-curves on Y4 and any two among them do not intersect, we derive that Bπ is preserved by every automorphism of Y4. Application of Proposition 2.3 to the anticanonical cyclic covering π shows that every automorphism h of Y4 admits a lift h ∶ X4 → X4 i.e there exists an automorphism h ∶ X4 → X4 such that π ⋅ h = h ⋅ π. Obviously θ(h) = h. This shows that θ is surjective. Finally note that the kernel of θ consists of automorphisms of X4 over Y4. Hence it is equal to ⟨σ⟩. (cid:3) Remark 5.5. [Vin83, Theorem 2.4] implies that exact sequence in Theorem 5.4 does not admit a section. Let Σ G ⊆ Σ b is G. Moreover, the epimorphism b be the group of bijections of the set b of branch curves of π. The action of Aut(Y4) on (−4)-curves of Y4 gives rise to a homomorphism of groups Aut(Y4)→ Σb. We define a subgroup b as the image of an injective homomorphism Σ5 → Σ b sending τ ∈ Σ5 to the bijection given by Fij ↦ Fτ(i)τ(j). Theorem 5.6. The image of the homomorphism Aut(Y4) → Σ Aut(Y4) → G admits a section given by homomorphism G ≅ Aut(S5) → Aut(Y4) that lifts an auto- morphism of S5 along birational contraction m ∶ Y4 → S5. Proof. Recall that Y4 is constructed as the blowing up m ∶ Y4 → S5 of intersection points of all(−1)- curves on S5. Since every automorphism of S5 permutes these intersection points, we derive that every automorphism of S5 can be uniquely lifted to an automorphism of Y4. This gives a homomorphism s ∶ G ≅ Aut(S5)→ Aut(Y4). We define a graph Γ. Vertices of Γ are primitive elements of the lattice SX4 with negative self- intersection. Two vertices v1, v2 of Γ are adjacent if ⟨v1, v2⟩ ≠ 0. The set of ramification curves r of π is a subset of the set of vertices of Γ. We define a subgroup Aut(r, Γ) of Aut(Γ) consisting of these automorphisms of Γ that preserve vertices in r. Pick an automorphism φ ∈ Aut(X4). Then φ induces an orthogonal transformation of SX4 and hence it induces an automorphism φ∗ of the graph Γ. Now according to the short exact sequence of Theorem 5.4 and the fact that π sends curves in the class r to curves in the class b of all (−4)-curves on Y4(Proposition 5.1), we derive that φ∗ preserves r. Thus there exists a homomorphism of groups Φ ∶ Aut(X4)→ Aut(r, Γ) given by φ↦ φ∗. We have the following diagram Aut(X4) Φ Aut(r, Γ) Aut(Γ) Ξ Σr ≅ Σ b 14 ŁUKASZ SIENKIEWICZ Here the second vertical arrow is the canonical monomorphism of groups and Ξ is restriction to r. Since σ induces identity on SX4, we deduce that Φ(⟨σ⟩) ={1Γ}. Thus we have a factorization Ξ ⋅ Φ = Ψ ⋅ θ, where Ψ ∶ Aut(Y4) → Σr ≅ Σ b is a homomorphism given by restricting automorphisms of Y4 to the set of (−4)-curves and θ is defined in Theorem 5.4. In [Vin83, Section 2.2] it is stated that Aut(Γ) ≅ Σ5. Therefore, Φ(Aut(X4)) ⊆ Aut(Γ) ≅ Σ5 and hence Ψ(Aut(Y4)) ≤ 5! = G. Moreover, by construction of s we have (Ψ ⋅ s)(G) = G. Thus Ψ has G as its image and induces an epimorphism Aut(Y4)→ G that has s as its section. (cid:3) Now we come back to the construction of X4 from section 3. In that section we constructed X4 starting from four points P ={p1, p2, p3, p4} on P2 such that no three of them are on the same line. Hence there exists a system of homogeneous coordinates [x0, x1, x2] on P2 such that p1 = [−1, 1, 1], p2 = [1, −1, 1], p3 = [1, 1, −1], p4 = [1, 1, 1]. For given {i, j} ∈ {1, 2, 3, 4} denote by Nij the line through pk and pl for {k, l} = {1, 2, 3, 4} ∖ {i, j}. Observe that [1, 1, −1], [1, −1, 1], [−1, 1, 1] and [1, 1, 1] are points of multiplicity three of this configuration of lines and there are three additional points of multiplicity two. One can easily calculate that these points are q1 = [0, 0, 1], q2 = [0, 1, 0], q3 = [1, 0, 0]. The whole configuration is showed in the following picture. N34 ∶ x0 + x1 = 0 N12 ∶ x0 − x1 = 0 N14 ∶ x1 + x2 = 0 N24 ∶ x0 + x2 = 0 p1 = [−1, 1, 1] p3 = [1, 1, −1] q3 = [0, 1, 0] q1 = [1, 0, 0] p2 = [1, −1, 1] p4 = [1, 1, 1] N13 ∶ x0 − x2 = 0 q2 = [0, 0, 1] N23 ∶ x1 − x2 = 0 In the picture every point and line is described in terms of the homogeneous coordinates [x0, x1, x2]. For given {i, j} ⊆ {1, 2, 3} denote by Kij the line through points qi and qj. Now consider the stan- dard quadratic transformation of P2 given by formula Q([x0, x1, x2]) = [x1x2, x0x2, x0x1]. One verifies that (i) Q is undefined precisely at points q1, q2, q3 and blowing up of points {q1, q2, q3} resolves the indeterminacy of Q. (ii) Q(Kij ∖ {q1, q2, q3}) = qk for {i, j} ⊆ {1, 2, 3} and {k} = {1, 2, 3} ∖ {i, j}. (iii) Q(Nij ∖ {q1, q2, q3}) ⊆ Nij for {i, j} ⊆ {1, 2, 3, 4} and Q(pi) = pi for i ∈ {1, 2, 3, 4}. Proposition 5.7. Quadratic tranformation Q of P2 gives rise to an automorphism fQ of order two of the surface Y4. This automorphism has the following properties. (a) f ∗ Q ∶ Pic(Y4)→ Pic(Y4) acts as identity on some sublattice of Pic(Y4) of rank 19. (b) fQ stabilizes all curves {Fij}0≤i<j≤4 and stabilizes all curves {F(ij)(kl)} for {i, j}, {k, l} ⊆ {0, 1, 2, 3, 4}, {i, j} ∩ {k, l} = ∅ except F(12)(34), F(13)(24), F(14)(23). This is denoted in the Figure 6, where filled vertices correspond to curves of the configuration that are stabilized by fQ. VINBERG'S X4 REVISITED 15 (01) (01)(23) (23) (01)(34) (01)(24) (34) (34)(02) (02) (02)(13) (24)(13) (13) (24) (23)(04) (23)(14) (34)(12) (24)(03) (02)(14) (13)(04) (04) (14) (04)(12) (14)(03) (Figure 6) (12) (12)(03) (03) Proof. Recall that S5 = BlP(P2) for P = {p1, p2, p3, p4}. Consider the birational morphism S5 → P2. As in the beginning of Section 3, denote by Eij the strict transform of Nij for {i, j} ∈ {1, 2, 3, 4} and let E0i be the exceptional curve mapping to the point pi for i ∈ {1, 2, 3, 4}. Denote also by q1, q2, q3 preimages under S5 → P2 of points q1, q2, q3. Since {q1} = N14 ∩ N23, {q2} = N12 ∩ N34, {q3} = N13 ∩ N24 we derive that {q1} = E14 ∩ E23, {q2} = E12 ∩ E34, {q3} = E13 ∩ E24 Finally let Kij for every {i, j} ∈ {1, 2, 3} be the strict transform of Kij to S5. Clearly Q lifts to a birational automorphism Q of S5 and Q has the following properties (1)-(3) corresponding to properties (i)-(iii) of Q. (1) Q is undefined precisely at points q1, q2, q3 and blow up of q1, q2, q3 resolves the indetermi- nacy of Q. (2) Q(Kij ∖ {q1, q2, q3}) = qk for {i, j} ⊆ {1, 2, 3} and {k} = {1, 2, 3} ∖ {i, j}. (3) Q(Eij ∖ {q1, q2, q3}) ⊆ Eij for every {i, j} ∈ {0, 1, 2, 3, 4}. Since Y4 is obtained from S5 via blowing up intersection points of configuration {Eij} for 0 ≤ i < j ≤ 4 and due to we derive using (1) that there exists a morphism rQ ∶ Y4 → S5 such that the following diagram {q1} = E14 ∩ E23, {q2} = E12 ∩ E34, {q3} = E13 ∩ E24 Y4 m S5 rQ S5 Q is commutative. Obviously rQ contracts all curves F(ij)(kl) for {i, j}, {k, l} ⊆ {0, 1, 2, 3, 4}, {i, j} ∩ {k, l} = ∅ except F(12)(34), F(13)(24), F(14)(23) and according to (2) preimage under rQ of each of points q1, q2, q3 is one-dimensional. Thus preimage under rQ of each intersection point of configuration {Eij} for 0 ≤ i < j ≤ 4 is one-dimensional. Therefore, using [Har77, Chapter 4, 16 ŁUKASZ SIENKIEWICZ Proposition 5.3], we derive that rQ factors through blow up of S5 at all intersection points of configuration {Eij}0≤i<j≤4. Since this blow up is Y4, we derive that there exists a morphism fQ such that the following diagram Y4 m S5 fQ Q Y4 m S5 is commutative. Next according to the fact that as a rational map Q ⋅ Q is equal to 1P2, we derive that as a rational map fQ ⋅ fQ is equal to 1Y4. Since fQ is a morphism, we deduce that fQ ⋅ fQ = 1Y4 i.e. fQ is an algebraic automorphism and f 2 Q = 1Y4. If fQ is trivial, then Q will be trivial and this is not the case. Hence fQ is a nontrivial involution of Y4. ∶ Pic(Y4) → Pic(Y4) acts as identity on some sublattice of Pic(Y4) of rank Now we show that f ∗ Q 19. According to (1) and (3), we derive that fQ stabilizes all curves {Fij}0≤i<j≤4 and stabilizes all curves {F(ij)(kl)} for {i, j}, {k, l} ⊆ {0, 1, 2, 3, 4}, {i, j} ∩ {k, l} = ∅ except F(12)(34), F(13)(24), F(14)(23). Thus f ∗ Q acts as identity on a sublattice of Pic(Y4) generated by {Fij}0≤i<j≤4, {F(ij)(kl) 0 ≤ i < j ≤ 4, 0 ≤ k < l ≤ 4, {i, j} ∩ {k, l} = ∅} ∖ {F(12)(34), F(13)(24), F(14)(23)} Next note that similarly to the situation in Proposition 3.8, there exists a fibration S5 → P1 with general fiber being smooth, rational curve having three singular fibers E14 + E23, E12 + E34, E13 + E24 Precomposing this fibration with m ∶ Y4 → S5, we obtain a fibration Y4 → P1 having precisely three singular fibers F(14) + F(23) + 2F(14)(23), F(12) + F(34) + 2F(12)(34), F(13) + F(24) + 2F(13)(24). Thus we have linear equivalences 2F(14)(23) − 2F(12)(34) ∼ F(14) + F(23) − F(12) + F(34) Since fQ stabilizes curves {Fij}0≤i<j≤4, we deduce that f ∗ 2F(14)(23) − 2F(13)(24) ∼ F(14) + F(23) − F(13) + F(24) Q acts as identity on F(14)(23) − F(12)(34), F(14)(23) − F(13)(24) Construction of Y4 via birational morphisms m ∶ Y4 → S5 and S5 → P2 shows that the sublattice of Pic(Y4) generated by {Fij}0≤i<j≤4, {F(ij)(kl) 0 ≤ i < j ≤ 4, 0 ≤ k < l ≤ 4, {i, j} ∩ {k, l} = ∅} ∖ {F(12)(34), F(13)(24), F(14)(23)} and by F(14)(23) − F(12)(34), F(14)(23) − F(13)(24) is of rank 19. (cid:3) Denote by fQ ∶ X4 → X4 any lift of fQ ∶ Y4 → Y4 along homomorphism described in Theorem 5.4. Corollary 5.8. Consider the model of a hyperbolic space described in Corollary 1.7 specified to X4. The linear map f ∗ Q ∶ H1,1 R (X4)→ H1,1 R (X4) induced by a lift fQ ∶ X4 → X4 of fQ ∶ Y4 → Y4 gives rise to a reflection of this hyperbolic space. Proof. According to Proposition 5.7, action of an automorphism fQ on H1,1 R (X4) leaves invariant vectors of some subspace of codimension one. Moreover, f ∗ R (X4) is a linear Q isometry. Thus f ∗ Q is an isometric linear endomorphism of H1,1 R (X4) leaving invariant vectors of R (X4) → H1,1 ∶ H1,1 VINBERG'S X4 REVISITED 17 a hyperplane and preserving the ample cone. This implies that there exists a vector e ∈ H1,1 such that (e, e) < 0 and R (X4) f ∗ Q(x) = x − 2(e, x) (e, e) e By Proposition 1.8, f ∗ Q induces a reflection of the hyperbolic space described in Corollary 1.7. (cid:3) Corollary 5.9. Reflection induced by f ∗ R (X4) is conjugate as an element of O(SX4) to reflections contained in S1 according to Vinberg's notation [Vin83, Section 2.2]. Q on the hyperbolic space associated with H1,1 Proof. In [Vin83, Section 2.2] author claims that there are three conjugacy classes of linear reflec- tions inside O(SX4) that induce reflections of the hyperbolic space associated with H1,1(X4). He chooses three sets of these reflections S ′ 2 and S1 each contained in precisely one conjugacy class inside O(SX4). Then he shows that only reflections conjugate to those in the set S1 induce automorphisms of X4. (cid:3) 2, S ′′ ACKNOWLEDGEMENTS I would like to express my gratitude to my advisor Jarosław Wi´sniewski for his constant sup- port and patient guidance. I wish to thank Joachim Jelisiejew who made valuable editorial and mathematical suggestions that vastly improve the presentation of the material. REFERENCES [AN06] Valery Alexeev and Viacheslav V. Nikulin, Del Pezzo and K3 surfaces, MSJ Memoirs, vol. 15, Mathematical Society of Japan, Tokyo, 2006. MR 2227002 [BHPVdV04] Wolf P. Barth, Klaus Hulek, Chris A. M. Peters, and Antonius Van de Ven, Compact complex surfaces, second ed., Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathe- matics, vol. 4, Springer-Verlag, Berlin, 2004. MR 2030225 [DBvGK+17] Maria Donten-Bury, Bert van Geemen, Grzegorz Kapustka, Michał Kapustka, and Jarosław Wi´sniewski, A [Dol03] [Dol08] [Dol12] [Har77] [Huy16] [Keu00] [Laz04] [Liu02] [Mor30] [Nik79] [Nik81] [O'G06] [OZ96] very special EPW sextic and two IHS fourfolds, Geom. Topol. 21 (2017), no. 2, 1179–1230. MR 3626600 Igor Dolgachev, Lectures on invariant theory, London Mathematical Society Lecture Note Series, vol. 296, Cambridge University Press, Cambridge, 2003. MR 2004511 Igor V. Dolgachev, Reflection groups in algebraic geometry, Bull. Amer. Math. Soc. (N.S.) 45 (2008), no. 1, 1–60. MR 2358376 , Classical algebraic geometry, Cambridge University Press, Cambridge, 2012, A modern view. MR 2964027 Robin Hartshorne, Algebraic geometry, Springer-Verlag, New York-Heidelberg, 1977, Graduate Texts in Mathematics, No. 52. MR 0463157 Daniel Huybrechts, Lectures on K3 surfaces, Cambridge Studies in Advanced Mathematics, vol. 158, Cam- bridge University Press, Cambridge, 2016. MR 3586372 JongHae Keum, A note on elliptic K3 surfaces, Trans. Amer. Math. Soc. 352 (2000), no. 5, 2077–2086. MR 1707196 Robert Lazarsfeld, Positivity in algebraic geometry. I, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics, vol. 48, Springer-Verlag, Berlin, 2004, Classical setting: line bundles and linear series. MR 2095471 Qing Liu, Algebraic geometry and arithmetic curves, Oxford Graduate Texts in Mathematics, vol. 6, Oxford University Press, Oxford, 2002, Translated from the French by Reinie Erné, Oxford Science Publications. MR 1917232 Ugo Morin, Sui sistemi di piani a due incidenti, Atti del Reale Istituto Venetodi Scienze Lettere ed Arti LXXXIX (1930), 907–926. V. V. Nikulin, Quotient-groups of groups of automorphisms of hyperbolic forms of subgroups generated by 2- reflections, Dokl. Akad. Nauk SSSR 248 (1979), no. 6, 1307–1309. MR 556762 , Quotient-groups of groups of automorphisms of hyperbolic forms by subgroups generated by 2-reflections. Algebro-geometric applications, 3–114. MR 633160 Kieran G. O'Grady, Irreducible symplectic 4-folds and Eisenbud-Popescu-Walter sextics, Duke Math. J. 134 (2006), no. 1, 99–137. MR 2239344 Keiji Oguiso and De-Qi Zhang, On the most algebraic K3 surfaces and the most extremal log Enriques surfaces, Amer. J. Math. 118 (1996), no. 6, 1277–1297. MR 1420924 18 [PŠŠ71] [Shi72] [SI77] [Vin83] [Wei94] ŁUKASZ SIENKIEWICZ I. I. Pjateckiı-Šapiro and I. R. Šafarevic, Torelli's theorem for algebraic surfaces of type K3, Izv. Akad. Nauk SSSR Ser. Mat. 35 (1971), 530–572. MR 0284440 Tetsuji Shioda, On elliptic modular surfaces, J. Math. Soc. Japan 24 (1972), 20–59. MR 0429918 T. Shioda and H. Inose, On singular K3 surfaces, Complex analysis and algebraic geometry, Iwanami Shoten, Tokyo, 1977, pp. 119–136. MR 0441982 È. B. Vinberg, The two most algebraic K3 surfaces, Math. Ann. 265 (1983), no. 1, 1–21. MR 719348 Charles A. Weibel, An introduction to homological algebra, Cambridge Studies in Advanced Mathematics, vol. 38, Cambridge University Press, Cambridge, 1994. MR 1269324 INSTYTUT MATEMATYKI UW, BANACHA 2, 02-097 WARSZAWA, POLAND E-mail address: [email protected]
1702.06752
1
1702
2017-02-22T11:05:12
Relative singular support and the semi-continuity of characteristic cycles for \'etale sheaves
[ "math.AG", "math.NT" ]
Recently, the singular support and the characteristic cycle of an \'etale sheaf on a smooth variety over a perfect field are constructed by Beilinson and Saito, respectively. In this article, we extend the singular support to a relative situation. As an application, we prove the generic constancy for singular supports and characteristic cycles of \'etale sheaves on a smooth fibration. Meanwhile, we show the failure of the lower semi-continuity of characteristic cycles in a higher relative dimension case, which is different from Deligne and Laumon's result in the relative curve case.
math.AG
math
RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES FOR ÉTALE SHEAVES HAOYU HU AND ENLIN YANG Abstract. Recently, the singular support and the characteristic cycle of an étale sheaf on a smooth variety over a perfect field are constructed by Beilinson and Saito, respectively. In this article, we extend the singular support to a relative situation. As an application, we prove the generic constancy for singular supports and characteristic cycles of étale sheaves on a smooth fibration. Meanwhile, we show the failure of the lower semi-continuity of characteristic cycles in a higher relative dimension case, which is different from Deligne and Laumon's result in the relative curve case. 7 1 0 2 b e F 2 2 ] . G A h t a m [ 1 v 2 5 7 6 0 . 2 0 7 1 : v i X r a Contents Introduction 1. 2. Notation and conventions 3. Radon Transforms and Legendre transforms 4. Relative singular supports 5. Existence of relative singular supports 6. Generic constancy of characteristic cycles 7. Failure of the lower semi-continuity for characteristic cycles References 1 5 7 11 17 22 27 27 1. Introduction 1.1. Let Y be a smooth complex algebraic variety of equidimension d, DY the sheaf of differential operators on Y and G a holonomic left DY -module. We can associate two geometric invariants to G. One is the singular support SSpGq, which is a closed subset of the cotangent bundle TY of equidimension d. The other one is the characteristic cycle CCpGq, which is a cycle of TY supported on SSpGq. They contain important information on the DY -module G. For example, assuming that Y is projective over C, we have [Du, Ka83] (1.1.1) χpY, DRY pGqq " degpT Y Y, CCpGqqTY , where χpY, DRY pGqq denotes the Euler-Poincaré characteristic of the de Rham complex DRY pGq " Ω1 Y Y the zero-section of TY . G and T Y {C bL DY 2000 Mathematics Subject Classification. Primary 14F05; Secondary 14C25. Key words and phrases. Semi-continuity, relative singular support, relative characteristic cycle. 1 2 HAOYU HU AND ENLIN YANG 1.2. It is well known that ℓ-adic étale sheaves are an analogue of algebraic D-modules. Let X be a smooth scheme over a field k, ℓ a prime number invertible in k, Λ a finite field of characteristic ℓ and K a bounded complex of constructible sheaves of Λ-modules on X. Recently, an important breakthrough is achieved by Beilinson. Using a Radon transform method initiated by Brylinski, he defined the singular support SSpKq of K and proved the existence of SSpKq [B15]. It is a closed conical subset of the cotangent bundle TX. When X is equidimensional, he proved that every irreducible component of SSpKq has the same dimension as X (loc. cit.). Assume that k is perfect. Based on Beilinsion's results, Saito constructed the characteristic cycle CCpKq of K as an algebraic cycle of TX supported on SSpKq [Sa16]. The coefficient of each component of CCpKq is obtain by the total dimension of the stalk of a vanishing cycles complex associated to K. The cycle CCpKq is positive if K is perverse. When X is projective over k, he proved the following index formula (loc. cit.) (1.2.1) χpX¯k, KX¯k q " degpT X X, CCpKqqTX , where ¯k denotes an algebraic closure of k and T one for D-modules (1.1.1). X X the zero-section of TX. It is similar to the Assume that X is a connected smooth curve over k and that K is the extension by zero of a locally constant and constructible sheaf of Λ-modules on an open dense subset U of X. Then, we have X Xs ´ ÿxPX´U (1.2.2) CCpKq " ´rkΛpKU q rT dimtotxpKq rT xXs, xX the fiber TX X x. The where dimtotxpKq denotes the total dimension of K at x (2.9) and T singular support SSpKq is the reduced closed subscheme of TX where CCpKq is supported. The index formula (1.2.1) is the well-known Grothendieck-Ogg-Shafarevich formula [SGA5]. In the rest of the introduction, let S be an excellent Noetherian scheme, f : X Ñ S a 1.3. separated and smooth morphism of purely relative dimension d, D an effective Cartier divisor of X relative to S, U the complement of D in X and j : U Ñ X the canonical injection. For any s P S, we denote by ¯s an algebraic geometric point above s and by X¯s and D¯s the fibers of f : X Ñ S and f D : D Ñ S at ¯s. We denote by TpX{Sq the relative cotangent bundle on X with respect to S and by T X pX{Sq the zero-section of TpX{Sq (2.3). Let ℓ be a prime number invertible in S, Λ a finite field of characteristic ℓ, G a locally constant and constructible sheaf of Λ-modules on U of constant rank and F " j!Grds. Theorem 1.4 ([L81, 2.1.1, 5.1.1]). Assume that f : X Ñ S has relative dimension d " 1. In this case, f D : D Ñ S is quasi-finite and flat. Then, (i) There exists a closed subscheme A of TpX{Sq which is open and of purely relative dimen- sion d over S such that, for any algebraic geometric point ¯s of S, we have SSpFX¯s q " pA¯sqred. (In fact, we take A " pTpX{Sq X DqŤ T (ii) There exists a finite and surjective morphism π : S1 Ñ S and a cycle B of TppX S S1q{S1q supported on A S S1 such that there exists an open dense subset W 1 of S1 and, for any algebraic geometric point ¯s1 of W 1, we have CCpF¯s1 q " B¯s1 (2.2); X pX{Sq). (iii) For any algebraic geometric point ¯s1 of S1 ´ W 1, the cycle B¯s1 ´ CCpF¯s1 q on TX¯s1 is effective; (iv) Let V be an open dense subset of S. Then, fV : X S V Ñ V is locally acyclic with respect to FXS V if and only if, for any algebraic geometric point ¯s1 of V S S1 , we have CCpF¯s1 q " B¯s1 . RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 3 Theorem 1.4 is the semi-continuity property for the Swan conductors of ℓ-adic sheaves on relative curves due to Deligne and Laumon. Here, we formulate it in terms of characteristic cycles of ℓ- adic sheaves. Saito asked whether the theorem can be generalized to the case where the relative dimension d ě 2. In this article, we answered his question for part (i)-(iii) of Theorem 1.4. Z{S Ñ Ω1 1.5. The main construction of this article is a relative version of singular supports of ℓ-adic sheaves which generalizes Beilinson's one in [B15]. Let Y and Z be two schemes smooth over S, g : Y Ñ X and h : Y Ñ Z two S-morphisms, and dg : TpX{Sq X Y Ñ TpY {Sq and dh : TpZ{Sq Z Y Ñ TpY {Sq two maps induced by canonical morphisms gΩ1 Y {S and hΩ1 Y {S, respectively. For a closed conical subset C of TpX{Sq, we say that g is C-transversal relative to S if pdg´1pT Y pY {Sqqq X pC X Y q is contained in the zero-section of TpX{Sq X Y . We denote by gC the scheme theoretic image dgpC X Y q in TpY {Sq. For a closed conical subset C1 of TpY {Sq, we say that h is C1-transversal relative to S if C1 X impdhq is contained in the zero-section of TpY {Sq. Let K be a bounded complex of constructible sheaves of Λ-modules on X. We say that K is micro-supported on a closed conical subset C of TpX{Sq if, for any pair of morphisms ph, gq : Z Ð Y Ñ X of schemes smooth over S such that g is C-transversal relative to S and that h is gC-transversal relative to S, the morphism h is locally acyclic relative to gK. If there is a smallest element among closed conical subsets of TpX{Sq on which K is micro-supported, we call it the singular support of K relative to S and we denote it by SSpF , X{Sq (cf. 4.4). X{S Ñ Ω1 The definition above was firstly introduced by Beilinson when S is spectrum of a field [B15]. Unlike the existence of singular support in loc. cit., the relative singular support SSpF , X{Sq does not exist in general. However, using a Radon transform method similar to Beilinson's, we obtain the following theorem: Theorem 1.6 (Theorem 5.4). The relative singular support SSpK, X{Sq exists after replacing S by a Zariski open dense subscheme. Let s be a point of S. The similarity of definitions of the singular support and its relative version allows us to compare SSpK, X{Sq S s with SSpKXsq. We prove the following theorem. Theorem 1.7 (Theorem 5.8 and Theorem 5.9). After replacing S by a Zariski open dense subset, the relative singular support SSpK, X{Sq is flat over Sred. Moreover, for any point s of S, we have (1.7.1) pSSpK, X{Sq S sqred " SSpKXs q. Theorem 1.7 implies that, in higher relative dimensional cases, we can take A " SSpK, X{Sq to generalize part (i) of Theorem 1.4, if allowing to shrink S. However, Example 7.2 implies that we cannot generalize part (i) in general. Part (ii) of Theorem 1.4 is generalized by the following theorem: Theorem 1.8 (Theorem 6.9). There exists a dominant and quasi-finite morphism π : S1 Ñ S and a cycle B on TppX S S1q{S1q such that, (i) B "řαPI mαrBαs pmα P Qq, where Bα (α P I) is flat over S1; (ii) For any algebraic geometric point ¯t Ñ S1, we have B¯t " CCpKX¯t q. 1.9. To prove Theorem 1.8, we may assume that S is integral. We denote by η the generic point of S and ¯η an algebraic geometric point above η. The characteristic cycle CCpKX ¯η q on TX¯η can be descent to a cycle C on TXη1 , where η1 is a finite extension of η. We denote by S1 the integral closure of S in η1 and put X 1 " X S S1. Let B be a cycle of TpX 1{S1q such that Yη1 " C. 4 HAOYU HU AND ENLIN YANG Using the semi-continuity of the total dimensions of vanishing cycles complexes due to Saito [Sa16], we proved that coefficients of characteristic cycles of K along fibers of f 1 : X 1 Ñ S1 are locally constant. Replacing S1 by a Zariski open dense subscheme, we obtain Theorem 1.8. In response to the notion of relative singular support, we call B the characteristic cycle of KX 1 relative to S1. 1.10. We show that part (iii) of Theorem 1.4 cannot be generalized to a higher relative dimension situation by Example 7.3. In other words, it says that the characteristic cycles is no longer lower semi-continuous in a higher relative dimension case, which is unlike the relative curve situation (part (2) of Theorem 1.4). The main reason of its failure is that, on a higher dimensional scheme, the singular support of an ℓ-adic sheaf with wild ramifications is not Lagrangian in general. In fact, we find an ℓ-adic sheaf F on a smooth fibration f : X Ñ S such that, for s P S, the family of singular supports SSpFXsq jumps from non-Lagrangian closed subset of TXs to Lagrangian ones at isolated closed points of S. 1.11. We expect in the future a generalization of part (iv) of Theorem 1.4 to higher dimension cases and its relation with vanishing cycles on general bases [O]. If we have a positive answer to this generalization, we could obtain that f : X Ñ S is locally acyclic with respect to F in 1.3 if and only if f : X Ñ S is universally locally acyclic with respect to F . Indeed, we only need to show that locally acyclicity implies universally acyclicity by the help of the generalization. To show the universally locally acyclicity, it is sufficient to show that the map fn : An S induced by f : X Ñ S is locally acyclic with respect to FAn If part (iv) of Theorem 1.4 is valid for higher dimensional situations, there exists a finite and surjective map π : S1 Ñ S and a cycle B1 of TppX S S1q{S1q such that, for any algebraic geometric point ¯s1 of S1, we have CCpF¯s1 q " B1 ¯s1. We put X 1 " X S S1 and we denote by B1 S1q. Then, for any algebraic geometric point ¯t1 of An nq¯t1. Using the generalization of part (iv) of Theorem 1.4 again for FAn S is locally acyclic with respect to FAn n the pull-back of B1 on TpAn S1, we have CCpF¯t1 q " B1 and fn : An X Ñ An S, we obtain that fn : An X Ñ An ¯t1 " pB1 X Ñ An X . X . X X 1{An 1.12. There have been several works on the generalization of Deligne and Laumon's semi-continuity of Swan conductors. Saito proved a semi-continuity property for the total dimension of the stalk of vanishing cycles complexes [Sa16]. It is an essential step to obtain characteristic cycles of ℓ-adic sheaves. For an ℓ-adic sheaf on a smooth variety, one can associate a divisor called the total di- mension divisor using Abbes and Saito's non-logarithmic ramification filtration of Galois group of local fields [AS02, Sa13]. In [HY], the authors proved the lower semi-continuity for total dimension divisors of ℓ-adic sheaves on a smooth fibration. In the theory of D-modules, a local invariant of an algebraic D-module on a smooth complex curve called irregularity is an analogue of Swan conductors. Deligne proved the semi-continuity for irregularities of relative differential systems on relative curves [De]. After that, André proved the same property for general differential systems [An]. We expect a generalization of their works in higher relative dimension situations for characteristic cycles of D-modules. 1.13. The article is organized as follows. After notation and geometric preliminaries, we give the notion of relative singular support of étale sheaves in §4. We prove the generic existence of relative singular support and study its fibers in §5. Based on results in §5, we prove the generic constancy of characteristic cycles in §6. Under certain geometric conditions, we also write down a proposition in §6 that compares the characteristic cycle and the relative characteristic cycle, which is given by Saito. In §7, we provide two examples which shows that part (i) and part (iii) cannot be generalized to higher dimensional situations. RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 5 Acknowledgement. The authors would like to express their gratitude to T. Saito for sharing ideas and for inspiring discussion. The authors would also like to thank A. Abbes and O. Gabber for their stimulating suggestions. The first author is supported by JSPS postdoctoral fellowship and Kakenhi Grant-in-Aid 15F15727 during his stay at the University of Tokyo and the second author is partially supported by Alexander von Humboldt Foundation for his research at Universität Regensburg and Freie Universität Berlin. Some part of the research was accomplished during their visit at IHES. The authors are grateful to these institutions. 2. Notation and conventions 2.1. Let X be a scheme and E a locally free sheaf of OX -modules of finite rank on X. We call the vector bundle of E and denote by VpEq the spectrum of the quasi-coherent OX -algebra SymOX pEq. Let E be such a vector bundle on X. We denote by PpEq the projectivization of E. Let C be a constructible subset of E. We say that C is conical, if C is invariant under the canonical Gm- action on E that fixes the zero-section. For a closed conical subset C of E, we denote by PpCq the projectivization of C, which is a closed subset of PpEq. We denote by BpCq the scheme theoretic image of C in X by the canonical projection π : E Ñ X and call it the base of C. Observe that the canonical map C Ñ BpCq is surjective. We say a conical subset C is strict if each irreducible component of C is not contained in the zero-section of E. We have a bijection tstrict closed conical subsets of Eu Ñ tclosed subsets of PpEqu, C ÞÑ PpCq. We always take reduced induced subscheme structures on a closed conical subset of a bundle and its projectivization. 2.2. Let f : X Ñ S be a morphism of schemes, s a point of S and ¯s Ñ S a geometric point above s. We denote by Xs (resp. X¯s) the fiber X S s (resp. X S ¯s). Let tZiuiPI be a finite set of closed subschemes of X and Z "ři mirZis a cycle supported on Z. We denote by Zs (resp. Z¯s) the cycle ři mirpZiqss (resp. ři mirpZiq¯ss) on Xs (resp. X¯s). 2.3. Let f : X Ñ S be a smooth morphism of Noetherian schemes. We denote by TpX{Sq the X{Sq_q on X and call it the relative cotangent bundle on X with respect to vector bundle VppΩ1 S. We denote by T X pX{Sq the zero-section of TpX{Sq. Let u be a point of X and ¯u Ñ X a geometric point of X above u. We put T upX{Sq " TpX{Sq X u and T ¯upX{Sq " TpX{Sq X ¯u. Let g : Y Ñ S be a smooth morphism and f : X Ñ S an S-morphism. Then, we have a canonical morphism (2.3.1) dh : TpX{Sq X Y Ñ TpY {Sq induced by the canonical map hΩ1 X{S Ñ Ω1 Y {S. 2.4. Let f : X Ñ S be a smooth morphism of Noetherian schemes and C a closed conical subset of TpX{Sq. Let Y be an S-scheme smooth over S and h : Y Ñ X an S-morphism. We say that h : Y Ñ X is C-transversal relative to S at a geometric point ¯y Ñ Y if for every non-zero vector µ P Chp¯yq " C X ¯y, the image dh¯ypµq P T hp¯yqpY {Sq is the canonical map. We say that h : Y Ñ X is C-transversal relative to S if it is C-transversal If h : Y Ñ X is C-transversal relative to S, the relative to S at every geometric point of Y . topological image of C X Y in TpY {Sq by the canonical map dh : TpX{Sq X Y Ñ TpY {Sq is closed, by the same argument of [B15, Lemma 1.1]. We put hC " dhpC X Y q. ¯y pY {Sq is not zero, where dh¯y : T ¯y pX{Sq Ñ T 6 HAOYU HU AND ENLIN YANG Let Z be an S-scheme smooth over S and g : X Ñ Z an S-morphism. We say that g : X Ñ Z is C-transversal relative to S at a geometric point ¯x Ñ X if for every non-zero vector ν P T gp¯xqpZ{Sq, we have dg¯xpνq R C¯x, where dg¯x : T ¯xpX{Sq is the canonical map. We say that g : X Ñ Z is C-transversal relative to S if it is C-transversal relative to S at all geometric point of X. If the base BpCq of C is proper over Z, we put gC :" pr1pdg´1pCqq, where pr1 : TpZ{Sq Z X Ñ TpZ{Sq denotes the first projection and dg : TpZ{Sq Z X Ñ TpX{Sq is the canonical map. It is a closed conical subset of TpX{Sq. gp¯xqpZ{Sq Ñ T When S is spectrum of a field, the definitions above are identical to those in [B15]. In this case, we follow the terminologies of loc. cit. and omit "relative to S". We observe that that g : X Ñ Z (resp. h : Y Ñ X) is C-transversal relative to S if and only if for every point s P S, the fiber gs : Xs Ñ Zs (resp. hs : Ys Ñ Xs) is Cs-transversal. Similarly to [B15, Lemma 1.1], if the map h : Y Ñ X (resp. g : X Ñ Z) is C-transversal relative to S at a geometric point ¯u Ñ Y (resp. ¯x Ñ X), then, there exists a Zariski open neighborhood Y0 of ¯u (resp. X0 of ¯x) such that hY0 : Y0 Ñ X (resp. gX0 : X0 Ñ Z) is C-transversal relative to S. 2.5. Let X be an irreducible excellent Noetherian scheme. We say that X is generically di- mensional if, for any open dense subscheme U of X, we have dim X " dim U . For example, an irreducible algebraic variety over a field is generically dimensional. But, the spectrum of an excel- lent integral local ring of dimension ě 1 is not. For any irreducible excellent Noetherian scheme Y , there exists an open dense subscheme V of Y such that V is generically dimensional. The con- dition of generic dimension is preserved by taking open dense subsechemes and irreducible closed subschemes. 2.6. We fix a prime number ℓ and a finite field Λ of characteristic ℓ. Let X be a Noetherian scheme over Zr1{ℓs. We denote by DpX, Λq the derived category of complexes of étale sheaves of Λ- modules on X and by Db cpX, Λq its full subcategory consisting of objects bounded with constructible cohomologies. 2.7. Let f : X Ñ S be a morphism of Noetherian schemes over Zr1{ℓs and F an object of Db cpX, Λq. We denote by Ef pFq the smallest closed subset of X such that f : X ´ Ef pFq Ñ S is universally locally acyclic with respect to F . Let S1 be a Noetherian S-scheme. We set X 1 " X S S1, set f 1 : X 1 Ñ S1 the base change of f : X Ñ S and set F 1 " FX 1. We have Ef 1 pF 1q Ď Ef pFq S S1. 2.8. Let X be an irreducible Noetherian scheme over Zr1{ℓs, ξ the generic point of X and F an object on Db cpX, Λq. We denote by C0pF , Xq the constructible subset C0pF , Xq " ts P X ; cosp : F¯s „ÝÑ F ¯ξu of X, where ¯s (resp. ¯ξ) denotes a geometric point above s (resp. ξ) and cosp is the canonical cospe- cialization map. We denote by N C0pF , Xq the complement of C0pF , Xq in X and by N CpF , Xq the closure of N C0pF , Xq in X and we call it the non-constant locus of F on X. We always take the reduced induced subscheme structure on the non-constant locus. 2.9. Let K be a complete discrete valuation field, OK its integer ring, F the residue field of OK, K a separable closure of K and GK the Galois group of K over K. We assume that F is perfect of characteristic p ą 0. Let M be a finite generated Λ-module with continuous GK -action. The total dimension of M is defined by where SwK M denotes the Swan conductor of M [Se97, 19.3]. dimtotK M " SwK M ` dimΛ M, RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 7 3. Radon Transforms and Legendre transforms 3.1. We fix an integer n ě 1. We denote by EZ " An`1 of rank n ` 1, by E_ projectivization of EZ (resp. E_ defined by the principal Cartier divisor associated to the section Z Z ). The universal hyperplane HZ of PZ Z P_ Z the dual bundle of EZ and by PZ " PpEZq (resp. P_ the trivial vector bundle over SpecpZq Z q) the Z is the hypersurface Z " PpE_ id P EndZpΓpPZ, OPZp1qqq " ΓpPZ Z P_ Z , pr 1 pOPZp1qq b pr 2 pOP_ Z p1qqq. s x P Hx1u, where Hx1 denotes For any point s P SpecpZq, we have pHZqs " tpx, x1q P Ps s P_ the hyperplane of Ps associated to x1 P P_ s . Let d be an integer ě 1 and N " `n`d Z the trivial vector bundle over Z of rank N , by rE_ d . We denote Z the dual bundle of rEZ and by by rEZ " AN rPZ " PprEZq (resp. rP_ Z q). LetriZ : PZ Ñ rPZ be a Veronese embedding of degree d. We denote by rHZ the universal hyperplane of rPZ ZrP_ Z , rEZ, ) above, we simply put A " AZ Z S. We denote byri : P Ñ rP the base change ofriZ : PZ Ñ rPZ and by pr : H Ñ P (resp. pr_ : H Ñ P_, rpr : rH ÑrP and rpr_ : rH ÑrP_) the canonical projection. Z q) the projectivization of rEZ (resp. rE_ 3.2. We have a canonical injection of locally free sheaves of OPZ -modules Let S be a scheme. For any AZ (AZ " EZ, E_ Z " PprE_ Z . with the cokernel OPZp1q. It induced an injection of vector bundles Ω1 PZ{Zp1q Ñ ΓpPZ, OPZp1qq bZ OPZ VppΩ1 PZ{Zq_p´1qq Ñ E_ Z Z PZ, hence an injection of their projectivizations which gives rise to an isomorphism PpTpPZ{Zqq Ñ P_ Z Z PZ, θZ : PpTpPZ{Zqq „ÝÑ HZ. Exchange PZ and P_ Z , we obtain an isomorphism θ_ Z : PpTpP_ Z {Zqq „ÝÑ HZ. The maps θZ and θ_ Z give rise to isomorphisms θ : PpTpP{Sqq „ÝÑ H and θ_ : PpTpP_{Sqq „ÝÑ H. The isomorphisms θZ, θ_ Z , θ and θ_ are called Legendre transforms. Let S be a Noetherian scheme and s a point of S. We have θs : PpTPsq „ÝÑ Hs, px, ¯λx,x1 q ÞÑ px, x1q, where px, x1q denotes a closed point of Hs and λx,x1 denotes the 1-dimensional vector space T which is orthogonal to the hyperplane Hx1 Ď Ps associated to x1 P P_ have canonical injections x Ps s . For any px, x1q P Hs, we dprx : T and dpr_ px,x1qHs, s in T xPs Ñ T x1 P_ dprxpλx,x1q " Lx,x1 " dpr_ x1 : T x1 P_ s Ñ T px,x1qHs. The intersection of T xPs and T px,x1qHs is a 1-dimensional vector space Lx,x1 and we have (3.2.1) Let C be a closed conical subset of TpP{Sq. We denote by C` the union of C and the zero-section of TpP{Sq. The projectivization PpCq of C can be considered as a closed subset of PpTpP_{Sqq by Legnedre transforms. We denote by C_ the strict closed conical subset of TpP_{Sq associated x1pλx1,xq. 8 HAOYU HU AND ENLIN YANG to PpCq. We take the same notation after exchanging P and P_. Obviously, we have C` " C__` and, if C is strict, we have C " C__. By (3.2.1), we have pr_ prpC`q " C_`. Lemma 3.3. Let X and Y be irreducible excellent Noetherian schemes and f : X Ñ Y a dominant morphism of finite type. Then, there exists a Zariski open dense subscheme V of Y such that XV " X Y V is generically dimensional. Proof. After replacing Y by a Zariski open dense subscheme, we may assume that Y is generically dimensional. By [EGA IV, II, 6.9.1], we may further assume that fred : Xred Ñ Yred is flat. Let η be the generic point of Y . By [EGA IV, III, 9.5.6], we may assume that, for any y P Y , each irreducible component Xv has dimension dimkpηq Xη. We have dim X " dim Y ` dimkpηq Xη. For any open dense subscheme U of X, the map f : Ured Ñ Yred is flat. Moreover, for any y P Y , the fiber Uy is empty or each irreducible component of Uy has dimension dimkpηq Xη. Hence, dim U " dim Y ` dimkpηq Xη " dim X. (cid:3) ([B15, 4.2]). Let k be a field and we assume that all k-schemes in this section are of finite 3.4. type over Specpkq. Let P be an irreducible k-scheme and π : H Ñ P a morphism of k-schemes. We denote by H p2q the complement of the diagonal in H P H. We say that two irreducible P P is closed subschemes Z1 and Z2 of H intersect well relatively to π : H Ñ P if pZ1 P Z2qŞ H p2q equidimensional and dimk´pZ1 P Z2q X H p2q P ¯ " dimk Z1 ` dimk Z2 ´ dimk P. Let Z be an irreducible excellent Noetherian scheme and g : Z Ñ P a generically surjective morphism. We may assume that Z is generically dimensional after replacing T by a Zariski open dense subset. The map g : Z Ñ P is called generically small if, after replacing T by a Zariski open dense subscheme, we have dimppZ P Zq ´ δpZqq ă dim Z, where δ : Z Ñ Z P Z denotes the diagonal map. A generically small map is generically radicial. We assume that π : H Ñ P is proper. Let Z1 and Z2 be irreducible closed subschemes of H which are dominant over T . We may assume that Z1 and Z2 are generically dimensional after Let Z be an irreducible k-scheme and g : Z Ñ P a generically surjective morphism. The map g : Z Ñ P is called small if dimkppZ P Zq ´ δpZqq ă dimk Z, where δ : Z Ñ Z P Z denotes the diagonal map. A small map is generically radicial. We assume that π : H Ñ P is proper. If irreducible closed subschemes Z1 and Z2 of H intersect well relatively to π : H Ñ P , if Z1 is generically finite over πpZ1q and if dimk Z2 ă dimk P , then Z1 Ď Z2 is equivalent to πpZ1q Ď πpZ2q ([B15, Lemma 4.2]). If an irreducible closed subscheme Z of H intersects well relatively to π : H Ñ P with itself and dimk Z ă dimk P , then the map πZ : Z Ñ πpZq is small and generically radicial (loc. cit.). 3.5. Let T be an irreducible excellent Noetherian scheme and P an irreducible T -scheme dominant and of finite type over T . We assume that P is generically dimensional. Let π : H Ñ P be a morphism of finite type. We denote by H p2q P the complement of the diagonal in H P H. Let Z1 and Z2 be irreducible closed subschemes of H which are dominant over T . After replacing T by a Zariski open dense subscheme, we may assume that all schemes above are generically dimensional (Lemma 3.3). We say that Z1 and Z2 generically intersect well relatively to π : H Ñ P if, after replacing T by a Zariski open dense subscheme, each irreducible component of pZ1 P Z2qŞ H p2q is generically dimensional, has same dimension, and P dim´pZ1 P Z2q X H p2q P ¯ " dim Z1 ` dim Z2 ´ dim P. RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 9 replacing T by a Zariski open dense subset. If Z1 and Z2 generically intersect well relatively to π : H Ñ P , if Z1 is generically finite over πpZ1q and if dim Z2 ă dim P , then Z1 Ď Z2 is equivalent to πpZ1q Ď πpZ2q. Let Z be an irreducible closed subscheme of H which is generically dimensional and flat over T . If Z generically intersects well relatively to π : H Ñ P with itself and dim Z ă dim P , then the map πZ : Z Ñ πpZq is generically small and generically radicial. Let η be the generic point of T . The two assertions above are deduced by using [B15, Lemma 4.2] to the generic fiber πη : Hη Ñ Pη, Lemma 3.6. Let T be an irreducible Noetherian schemes, X and Y irreducible T -schemes flat and of finite type over T and g : X Ñ Y a surjective and generically radicial T -morphism. For any s P T , we denote by tpXsqαuαPIpsq (resp. tpYsqβuβPJpsq) the irreducible components of Xs (resp. Ys). Then, there exists an open dense subset W of T such that, for any s P W , the map gs : Xs Ñ Ys induces an identity between Ipsq and Jpsq and, for any α P Ipsq, the map gs : pXsqα Ñ pYsqα is surjective and generically radicial. Proof. Since g : X Ñ Y is of finite type and generically radicial, we have a Cartesian diagram U j1 g1 l V j X g / Y where V is an open dense subset of Y , j : V Ñ Y is the canonical injection and g1 : U Ñ V is surjective and radicial. Hence, U is an open dense subset of X. By [EGA IV, III, 9.6.1], there exists an open dense subset W Ď T such that, for any s P W , we have U s " Xs and V s " Ys. Hence Us (resp. Vs) contains all generic points of irreducible components of Xs (resp. Ys). Since, for any s P W , g1 s : Us Ñ Vs is surjective and radicial, we have a one-to-one correspondence of the generic points of irreducible components of Xs and of Ys, i.e., Ipsq " Jpsq. For any s P W and any α P Ipsq, the map gs : pXsqα Ñ pYsqα is generically radicial since g1 is radicial. s : ppXsqαŞ Usq Ñ ppYsqαŞ Vsq (cid:3) Lemma 3.7. Let T be an irreducible Noetherian scheme, X and Y integral T -schemes dominant and of finite type over T , and g : X Ñ Y a T -morphism. We assume that g is dominant, generically finite, generically radicial and that the fibers Xt and Yt are geometrically integral for any t P T . We denote by ζ (resp. η) the generic points of X (resp. Y ) and, for any geometric point ¯t of T , by ζ¯t (resp. η¯t) the generic points of X¯t (resp. Y¯t). Then, there exists a Zariski open dense subset W of T such that, rζ¯t : η¯ts " rζ : ηs for any geometric point ¯t Ñ W . In particular, if the generic point of T is of characteristic 0, then the canonical map g¯t : X¯t Ñ Y¯t is birational for any ¯t Ñ W . Proof. Since g : X Ñ Y is dominant, generically finite and generically radicial, there is an open dense subset V of Y such that g1 : U " V Y X Ñ V is surjective, finite, flat and radicial, where g1 is the base change of g by V ãÑ Y . Let W be an open dense subset of T such that the fiber Vt is non-empty for any t P W . Then, for any geometric point ¯t of W , we have ζ¯t " η¯t V¯t U¯t " η¯t V U , hence rζ¯t : η¯ts " rU : V s " rζ : ηs. When the generic point of T is of characteristic 0, we have rζ : ηs " 1, i.e., g : V Ñ U is an (cid:3) isomorphism. Lemma 3.8 (cf. [B15, Lemma 4.3]). We take the notation and assumptions of 3.1, we assume that S is an irreducible excellent Noetherian scheme and that the Veronese embeddingri : P Ñ rP in 3.1 has degree d ě 3. Let C1 and C2 be irreducible closed conical subsets of TpP{Sq which   / /   / 10 HAOYU HU AND ENLIN YANG Proof. We may assume that S is integral. By Lemma 3.3, we may assume that all schemes are generically dimensional after shrinking S. By [EGA IV, II, 6.9.1], we may assume that C1 and C2 are dominant over S. We consider PpriC1q and PpriC2q as closed subschemes of rH by Legendre transform (3.2). Then, after replacing S by a Zariski open dense subset, PpriC1q and PpriC2q generically intersect well relatively to rpr_ : rH Ñ rP_. are flat over S after shrinking S. Since the canonical morphism dri : TprP{Sq rP P Ñ TpP{Sq is smooth, conical subschemesriC1 andriC2 are flat over S. Hence, PpriC1q and PpriC2q are flat over S. Let rδ : rH Ñ rH rP_ rH be the diagonal map and rHp2q rP_ the complement of rδprHq in rH rP_ rH. Shrinking S again, we may assume that each irreducible component of Y "´PpriC2q rP_ PpriC2q¯črHp2q rP_ is flat over S. Let η be the generic point of S. There is a one-to-one correspondence between the set of irreducible components of Y and that of Yη. By [B15, Lemma 4.3], pPpriC1qqη and pPpriC2qqη intersect well relatively to rprη : rHη ÑrP_ η . Hence, Yη is equidimensional and we have dimkpηq Yη " dimkpηqpPpriC1qqη ` dimkpηqpPpriC2qqη ´ dimkpηqrP_ Hence, Y is also equidimensional and (3.8.1) implies that (3.8.1) η . dim Y " dimpPpriC1qq ` dimpPpriC2qq ´ dimrP_. [B15, Proposition 4.5]). We take the notation and assumptions of 3.1, we Proposition 3.9 (cf. assume that S is an irreducible generically dimensional excellent Noetherian scheme and that the Hence, PpriC1q and PpriC2q generically intersect well relatively to rpr_ : rH Ñ rP_. Veronese embeddingri : P Ñ rP in 3.1 has degree d ě 3. Let C be a closed conical subset of TpP{Sq (1) If C is irreducible, then the map rpr_ : PpriCq Ñ rpr_pPpriCqq is generically small after such that Cη ‰ H is dominant over S and that dim C ď dim P. Then, (2) After replacing S by a Zariski open dense subscheme, there is a one-to-one correspondence replacing S by a Zariski open dense subscheme. (cid:3) between the sets of irreducible components of C and rpr_pPpriCqq. (2) Let tCvuvPI be set of the irreducible component of C. There is a one-to-one correspondence Proof. (1) By Lemma 3.3, we may assume that C and PpriCq are generically dimensional after shrinking S. We have dim PpriCq ď dim S ` N ´ 2. Hence dim PpriCq ă dimrP. After shrinking S, the irreducible scheme PpriCq Ď rH generically intersects well relatively to rP_ with itself (Lemma 3.8). Hence, rpr_ : PpriCq Ñ rpr_pPpriCqq is generically small, hence generically radicial (cf. 3.5). between tCvuvPI and tPpriCvquvPI . After shrinking S, we may assume that, for each v P I, we have pCvqη ‰ H and PpriCvq is generically dimensional and that, for each pair v, v1 P I (v and v1 can be the same), PpriCvq and PpriCv1 q intersect will relatively to rP_ (Lemma 3.8). Since PpriCvq pv P Iq are distinct irreducible subsets of rH, rpr_pPpriCvqq pv P Iq are distinct irreducible subsets of rP_ (3.5). Hence, we have a one-to-one correspondence between tPpriCvquvPI and trpr_pPpriCvqquvPI by the projection rpr_ : rH ÑrP_. We obtain (2). 3.10. We take the notation and assumptions of 3.1 and we assume that ℓ is invertible in S. We define the Radon transform by the functor [Bry] (cid:3) RS : Db cpP, Λq Ñ Db cpP_, Λq, RSpFq " Rpr_ pprFqrn ´ 1s, RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 11 and we define the dual Radon transform by the functor R_ S : Db cpP_, Λq Ñ Db cpP, Λq, R_ S pGq " Rprppr_Gqpn ´ 1qrn ´ 1s. Since pr : H Ñ P and pr_ : H Ñ P_ are projective and smooth, The functor RS is both left and right adjoint to R_ S by Poincaré duality and proper base change theorem. We denote by rRS and rR_ the Radon transform and dual Radon transform for the pair of morphisms prpr, rpr_q :rP Ð rH ÑrP_, respectively. Proposition 3.11 (cf. [B15, 1.6.1]). We take the notation and assumptions of 3.1 and we assume cpP_, Λq. Then, the that ℓ is invertible in S. Let F be an object of Db mapping cone of the adjunction map F Ñ pR_ S qpGq Ñ G) has locally constant cohomologies, after replacing S by a Zariski open dense subset. S RSqpFq (resp. pRS R_ cpP, Λq and G an object of Db S Proof. The proposition is valid when S is spectrum of a field [B15, 1.6.1]. Hence, (1) and (2) still hold since we allow shrinking S. (cid:3) Lemma 3.12. Let S and X be irreducible Noetherian schemes and f : X Ñ S a flat morphism of finite type with irreducible fibers. We assume that ℓ is invertible in X. Let F be an object of Db cpX, Λq. Then, there exists a Zariski open dense subset V of S such that, for any s P V , we have (2.8) pN CpF , Xq S sqred " N CpFXs , Xsq. Proof. We simply put Z0 " N C0pF , Xq and put Z " N CpF , Xq. The canonical map i : Z0 Ñ Z is dominant. By [EGA IV, III, 9.6.1], there exists a Zariski open dense subset V of S such that for any s P V , the map is : pZ0qs Ñ Zs is dominant. For any s P V , we have pZ0qs " N C0pFXs , Xsq and pZ0qs " N CpFXs , Xsq. Hence, for any s P V , we have pZsqred " pZ0qs " N CpFXs , Xsq. 4. Relative singular supports (cid:3) In this section, let S be a connected Noetherian scheme, f : X Ñ S a smooth morphism of finite type and C a closed conical subset in the relative cotangent bundle TpX{Sq. 4.1. A test pair of X relative to S is a pair of morphisms pg, hq : Y Ð U Ñ X such that U and Y are S-schemes smooth and of finite type over S and g : U Ñ Y and h : U Ñ X are S-morphisms. We say that pg, hq is C-transversal relative to S if h : U Ñ X is C-transversal relative to S and g : U Ñ Y is hC-transversal relative to S. We say that a test pair pg, hq : Y Ð U Ñ X relative to S is weak if it satisfies: (a) Y " A1 S is an affine line over S and (b) the morphism h : U Ñ X is ‚ a composition of pr1 : U " V S S1 Ñ V and h1 : V Ñ X, where S1 is finite and étale over S and h1 : V Ñ X is an open immersion, if Sred is a spectrum of a finite field; ‚ an open immersion in other situations. In the following of this section, we assume that all schemes are over Zr1{ℓs. Let F be an cpX, Λq. We say that a test pair pg, hq : Y Ð U Ñ X relative to S is F -acyclic if 4.2. object in Db g : U Ñ Y is locally acyclic with respect to hF . 12 HAOYU HU AND ENLIN YANG 4.3. We say that F is micro-supported on C relative to S if every C-transversal test pair of X relative to S is F -acyclic. We define CpF , X{Sq the set tC1 Ď TpX{Sq C1 is closed conical and F is micro-supported on C1 relative to Su. We will see that CpF , X{Sq is non-empty if f : X Ñ S is universally locally acyclic relative to F (cf. Proposition 4.9). We denote by CminpF , X{Sq the set of the minimal elements of CpF , X{Sq. We say that F is weakly micro-supported on C relative to S if every C-transversal weak test pair pg, hq relative to S which is C-transversal relative to S is F -acyclic. We define CwpF , X{Sq the set tC1 Ď TpX{Sq C1 is closed conical and F is weakly micro-supported on C1 relative to Su. Note that CpF , X{Sq Ď CwpF , X{Sq. Let C1 and C2 be two elements of CwpF , X{Sq. We choose a weak test pair pg, hq of X relative to S which is pC1 X C2q-transversal relative to S. Then g : U Ñ Y is hpC1 X C2q-transversal relative to S. For each geometric point ¯u of U , g is either hC1-transversal relative to S at ¯u or hC2-transversal relative to S at ¯u, since TpY {Sq is a line bundle on Y . Then, U can be covered by two Zariski open subsets U1 and U2 such that gUi : Ui Ñ Y is Ci-transversal relative to S pi " 1, 2q. Since C1, C2 P CwpF , X{Sq, the map gi pi " 1, 2q is locally acyclic with respect to F , i.e., the map g is locally acyclic with respect to F . It implies that the weak test pair pg, hq is F -acyclic. We deduce that C1 X C2 P CwpF , X{Sq. In conclusion, the set CwpF , X{Sq has a smallest element if it is non-empty. If CpF , X{Sq has a smallest element, we denote it by SSpF , X{Sq and call it the singular 4.4. support of F relative to S. We call the smallest element of CwpF , X{Sq the weak singular support of F relative to S and denote it by SSwpF , X{Sq. If SSpF , X{Sq exists, then SSωpF , X{Sq Ď SSpF , X{Sq. Notice that the relative singular support and the weak relative singular support are invariant after taking reduced induced subscheme for S and X. Example 4.5. Assume that f " idX . Then TpX{Xq " X and the X-test pair pid, idq : X Ð X Ñ X is TpX{Xq-transversal relative to X. Hence, for an object F of Db cpX, Λq, the relative singular support SSpF , X{Xq exists if and only if F has locally constant cohomologies. Remark 4.6. All definitions above were firstly introduced by Beilinson when S " Specpkq [B15]. In this case, we will omit the phase "relative to S" and we abbreviate the notion SSpF , X{Sq (resp. SSwpF , X{Sq) by SSpFq (resp. SSwpFq) to fit the notation in loc. cit.. Beilinson proved that SSpFq exists, that SSpFq " SSwpFq and that, when X is equidimensional, SSpFq is of equidimension dimk X (loc. cit.). Lemma 4.7. Let U and Y be S-schemes smooth and of finite type over S and let h : U Ñ X and g : X Ñ Y be S-morphisms. (1) If h : U Ñ X is smooth, then, for any closed conical subset C of TpX{Sq, the morphism h : U Ñ X is C-transversal relative to S. Moreover we have hC " C X U . (2) If g : X Ñ Y is C-transversal relative to S, then it is smooth on a Zariski neighborhood of the base BpCq. Proof. If h is smooth, then dh : TpX{SqX U Ñ TpU {Sq is injective. We obtain (1) by definition. Let x be a point of BpCq and ¯x Ñ BpCq a geometric point above x. If g : X Ñ Y is C-transversal gp¯xqpY {Sq Ñ T relative to S, then the canonical map dg¯x : T ¯xpX{Sq is injective. It is equivalent to that the canonical morphism of coherent OX -modules gΩ1 Y {S Ñ Ω1 X{S is injective at x. By [Fu, 2.5.7], we deduce that g is smooth on a Zariski neighborhood of x in X. Hence, we obtain (2). (cid:3) Lemma 4.8. Let pg, hq : Y Ð U Ñ X be a test pair of X relative to S. RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 13 (1) Assume that C " T X pX{Sq. Then pg, hq is C-transversal relative to S if and only if g is smooth. (2) Assume that C " TpX{Sq. Then pg, hq is C-transversal relative to S if and only if the canonical map h g : U Ñ X S Y is smooth. Proof. (1) When h : U Ñ X is C-transversal relative to S, the pull-back hC is the zero-section of TpU {Sq. By definition, g : U Ñ Y is hC-transversal relative to S if and only if, for any geometric point ¯u Ñ U , the canonical map dg ¯u : T ¯upU {Sq is injective. It is equivalent to that the canonical morphism of OX -modules OU -modules gΩ1 U{S is injective. It equals to that g : U Ñ Y is smooth [Fu, 2.5.7]. gp¯uqpY {Sq Ñ T Y {S Ñ Ω1 (2) If h g : U Ñ X S Y is smooth, then h and g are also smooth. By Lemma 4.7(1), the morphism h is C-transversal relative to S and hC " TpX{Sq X U . Since h g is smooth, the canonical morphism (4.8.1) dph gq : TppX S Y q{Sq pXS Y q U Ñ TpU {Sq is injective. Moreover, the canonical projections pr1 : X S Y Ñ X and pr2 : Y S X Ñ Y induces two injections dpr1 : TpX{Sq X pX S Y q Ñ TppX S Y q{Sq, dpr2 : TpY {Sq Y pX S Y q Ñ TppX S Y q{Sq. The intersection of the images of dpr1 and dpr2 is contained in the zero-section of T pX S Y {Sq. Combining (4.8.1), it implies that the intersection of the images of dh : TpX{SqX U Ñ TpU {Sq and dg : TpY {Sq X U Ñ TpU {Sq is contained in the zero section of TpT {Sq. Hence g is hC- transversal relative to S, i.e., the test pair pg, hq is C-transversal relative to S. Conversely, we assume that pg, hq is C-transversal relative to S. It implies that dh : TpX{SqX U Ñ TpU {Sq is injective and that g : U Ñ Y is hC-transversal relative to S. The map dh is injective implies that h is smooth. By Lemma 4.7(2), g is also smooth since the base BphCq is equal to U . Notice that hC " impdhq. Hence, g : U Ñ Y is hC-transversal relative to S implies that the intersection of impdhq and impdgq in TpU {Sq is T U pU {Sq. It implies that the canonical map dph gq : TpX S Y {Xq pXS Y q U Ñ TpU {Sq is injective, Hence, h g : U Ñ X S Y is smooth. (cid:3) Proposition 4.9. If f : X Ñ S is universally locally acyclic with respect to F , then TpX{Sq is an element of CpF , X{Sq. Proof. Let pg, hq : Y Ð U Ñ X be a test pair of X relative to S which is TpX{Sq-transversal relative to S. We need to show that it is F -acyclic. Consider the following diagram with Cartesian squares hg U / X S Y #❍❍❍❍❍❍❍❍❍❍ g pr2 pr1 / X l f Y / S Since f : X Ñ S is universally locally acyclic with respect to F , the second projection pr2 is also 1 F . By Lemma 4.8(2), the pair pg, hq is TpX{Sq- universally locally acyclic with respect to pr transversal implies that the morphism h g is smooth. By the smooth base change theorem, the composition g " pr2 ph gq is universally locally acyclic relatively to hF , i.e., the pair pg, hq is F -acyclic. (cid:3) # /   /   / 14 HAOYU HU AND ENLIN YANG Lemma 4.10. Let tViuiPI be a Zariski open covering of X. We assume that CpF , X{Sq is non- empty. An element C of CpF , X{Sq is minimal if and only if, for any i P I, CVi " C S Vi is a minimal element of CpF , Vi{Sq. Proof. We firstly prove the "only if" part. Let C P CminpF , X{Sq and V an open subscheme of X. It is easy to see that CV P CpFV , V {Sq. Let C1 be a minimal element of CpFV , V {Sq contained in CV . Let C1 be the closure of C1 in TpX{Sq. Then F is micro-supported on pC X pX ´ V qqŤ C1 relative to S. Since pC X pX ´ V qqŤ C1 Ď C, we have C " pC X pX ´ V qqŤ C1 since C is minimal. Hence CV " C1 is also minimal. For the "if" part, we suppose C2 is a minimal element of CpF , X{Sq contained in C. By the "only if" part, for any i P I, the restriction C2 Vi " C2 S Vi is a minimal element of CpFVi , Vi{Sq. Since CVi is also minimal, we have C2 Vi " CVi , i.e., C2 " C. (cid:3) Lemma 4.11 (cf. [B15, Lemma 2.1]). (i) We assume that CpF , X{Sq is nonempty. Then the base of SSwpF , X{Sq equals the support of F . Let C1 be an element of CminpF , X{Sq. Then the base of C1 equals the support of F . (ii) Assume that F is micro-supported on C relative to S. Then, for every test pair pg, hq : Y Ð U Ñ X relative to S which is C-transversal relative to S, the map g is universally locally acyclic with respect to hF . (iii) The complex F is micro-supported on the zero section of TpX{Sq if and only if each cohomology of F is locally constant on X. (iv) Let F 1 and F 2 be objects of Db cpX, Λq and F 1 Ñ F Ñ F 1 Ñ a distinguished triangle. If F 1 (resp. F 2) is micro-supported on C1 Ď TpX{Sq (resp. C2) relative to S, then F is micro-supported on C1 Y C2 relative to S. Proof. (i) Let Z be the support of F , B the base of SSwpF , X{Sq and B1 the base of C1. We have SSwpF , X{Sq Ď C1 and B Ď B1. By Lemma 4.10, we have C1 X pX ´ Zq " H since FX´Z " 0. Hence B Ď B1 Ď Z. We need to show that Z Ď B. Replacing X by X ´ B, we are reduced to show idÝÑ X, where that SSωpF , X{Sq is non-empty if F ‰ 0. Consider the test pair pr, idq : A1 S r : X Ñ A1 S. It is not F -acyclic if F ‰ 0. S is a composition of f : X Ñ S and the zero-section i : S Ñ A1 rÐÝ X (ii) We need to prove that, after any base change r : Y 1 Ñ Y , the morphism g1 : U 1 " U S S1 Ñ Y 1 is locally acyclic with respect to h1F , where h1 : U 1 Ñ X denotes the composition of the canonical morphism r1 : U 1 Ñ U and h : U Ñ X. By dévissage, we only need to treat the case where r : Y 1 Ñ Y is smooth. We can replace U by a Zariski neighborhood of the support of hF . By (i), the support of F is contained in the base of C. Hence, we may assume that h is smooth by Lemma 4.7. Then, the composition of the canonical maps r1 : U 1 Ñ U and h : U Ñ X is smooth, which implies that h1 " h r1 is C-transversal relative to S (Lemma 4.7). Since r : Y 1 Ñ Y is smooth, the morphism g : U Ñ Y is hC-transversal relative to S implies that g : U 1 Ñ Y 1 is h1C-transversal relative to S. Hence, pg1, h1q is also a test pair of X which is C-transversal relative to S. We deduce that g1 is locally acyclic with respect to h1F . (iii) If F is micro-supported on T X pX{Sq relative to S, then the test pair pid, idq : X Ð X Ñ X X pX{Sq-transversal. Hence id : X Ñ X is locally acyclic with respect to F . This implies that is T each cohomology of F is a locally constant sheaf on X. Conversely, If each cohomology of F is a locally constant sheaf on X, then any test pair pg, hq : Y Ð U Ñ X, where g is smooth, is F -acyclic. By Lemma 4.8, the sheaf F is micro-supported on T X pX{Sq relative to S. (iv) If a test pair pg, hq : Y Ð U Ñ X is pC1 Y C2q-transversal relative to S, then pg, hq is F 1-acyclic (resp. F 2-acyclic) relative to S, i.e., the morphism g : U Ñ Y is locally acyclic with RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 15 respect to hF 1 (resp. hF 2). Since F 1 Ñ F Ñ F 1 Ñ is a distinguished triangle, the morphism g : U Ñ Y is also locally acyclic with respect to hF , i.e., the pair pg, hq is F -acyclic relative to S. (cid:3) Lemma 4.12 (cf. [B15, Lemma 2.2]). We assume that C is an element in CpF , X{Sq. (1) Let Y be an S-scheme smooth and of finite type over S and h : Y Ñ X an S-morphism. If h is C-transversal relative to S, then, hC is contained in CphF , Y {Sq. (2) Let Y be an S-scheme smooth and of finite type over S and h : Y Ñ X an smooth and surjective S-morphism. If hC is contained in CminphF , Y {Sq, then C is contained in CminpF , X{Sq. (3) Let Z be an S-scheme smooth and of finite type over S and g : X Ñ Z an S-morphism. If the base BpCq is proper over Z, then gC is contained in CpRgF , Z{Sq. Proof. (1) Let U be a S-scheme smooth and of finite type over S and r : U Ñ Y an S-morphism which is hC-transversal relative to S. Since h : Y Ñ X is C-transveral relative to S, the composition h r : U Ñ X is C-transversal relative to S. Hence, a test pair pg, rq : Z Ð U Ñ Y relative to S is hC-transversal relative to S implies that the pair pg, h rq : Z Ð U Ñ X is C-transversal relative to S. Since C is an element of CpF , X{Sq, the latter implies that g is locally acyclic with respect to rphFq, i.e., the pair pg, rq is F -acyclic. Hence, hC is contained in CpF, X{Sq. (2) Let C1 be an element in CminpF , X{Sq that is contained in C. Then, by (1), hC1 is contained in CphF , X{Sq. Since hC is contained in CminphF , Y {Sq, the inclusion hC1 Ď hC is an identity. Since h is smooth, we have hC " C X Y and hC1 " C1 X Y (Lemma 4.7). Since h is surjective, the equality C X Y " C1 X Y implies that C " C1. (3) Let pr, sq : W Ð V Ñ Z be a gC-transversal test pair relative to S. We put X 1 " V Z X and denote by s1 : X 1 Ñ X and g1 : X 1 Ñ V the canonical projections. Since s : V Ñ Z is gC-transversal relative to S, there exists a Zariski open neighborhood U of s´1pBpgCqq in V which is smooth over Z. Hence U 1 " U Z X is a Zariski open neighborhood of s1´1pBpCqq in X 1 which is smooth over X. We have the following commutative diagram with Cartesian squares U 1 g1U 1 U j1 l j !❇❇❇❇❇❇❇❇ rU X 1 / V g1 r s1 l s X g Z W By Lemma 4.7(1), the restriction s1U 1 : U 1 Ñ X is C-transversal relative to S. By definition, rU : U Ñ W is spgCq-transversal relative to S implies that r g1U 1 : U 1 Ñ W is s1C-transversal relative to S. Hence, the test pair pr g1U 1 , s1U 1q : W Ð U 1 Ñ X is C-transversal relative to S. Since C is an element of CpF , X{Sq, the map r g1U 1 is universally locally acyclic with respect to ps1U 1 qF (Lemma 4.11(2)). Since BpCq contains the support of F (Lemma 4.11(1)), the scheme U 1 contains the support of s1F in X 1. Therefore, r g1 is universally locally acyclic with respect to s1F . Since the support of s1F , contained in s1´1pBpCqq is also proper over V , the morphism ps1Fq by proper base change theorem. By the r : V Ñ W is locally acyclic with respect to Rg1 ps1Fq „ÝÑ spRgFq. Hence, the test pair pr, sq : W Ð V Ñ Z is same theorem, we have Rg1 RgF -acyclic. In summary, each gC-transversal test pair relative to S is RgF -acyclic, i.e., RgF is micro-supported on gC relative to S.   / / / /     ! / / /   16 HAOYU HU AND ENLIN YANG (cid:3) Proposition 4.13 (cf. [B15, Lemma 2.5]). Suppose that Sred is not the spectrum of a field. Let P be an S-scheme smooth and of finite type over S and i : X Ñ P a closed immersion of S-schemes. (i) If SSwpiF , P {Sq and SSwpF , X{Sq exist, we have (4.13.1) SSwpiF , P {Sq " iSSwpF , X{Sq. (ii) If SSpiF , P {Sq exists, then SSpF , X{Sq also exists. Moreover, after replacing S by a Zariski open dense subscheme, we have (4.13.2) SSpiF , P {Sq " iSSpF , X{Sq. Please confer to [B15, Lemma 2.5] for the case where Sred " Specpkq is the spectrum of a field. We mimic the part of loc. cit. where k has infinite many elements to prove Proposition 4.13. Proof. (i). Since i : X Ñ P is a closed immersion, a weak test pair pg, hq : A1 S Ð U Ñ P relative to S is iF -acyclic if and only if pg1, h1q : A1 S Ð U P X Ñ X is F -acyclic, where g1 : U P X Ñ Y is the composition of the first projection pr1 : U P X Ñ U and g : U Ñ Y and h1 : U P X Ñ X is the base change of h : U Ñ P . It implies that iSSwpF , X{Sq is an element of CwpiF , P {Sq, i.e., SSwpiF , P {Sq Ď iSSwpF , X{Sq. Let x be a point of X, α a non-zero vector of T xpP {Sq. Let pu, vq : A1 S Ð V Ñ X be a weak test pair of X relative to S, such that V is a Zariski open neighborhood of x and that dux " α. Since P and X are smooth over S, we can find elements t1, , tm P OP,x such that xpX{Sq and β a pre-image of α in T 1. OX,x " OP,y{ptn`1, , tmq pn ă mq, and the element a P OX,x induced by u : V Ñ A1 S writes a "ř1ďjďn si¯ti psi P OSq; 2. dt1 b 1, , dtm b 1 is a base of kpxq-vector space T xpP {Sq " Ω1 P {S,x bOP,x kpxq and dt1 b 1, , dtn b 1 is a base of kpxq-vector space T xpX{Sq " Ω1 X{S,x bOX,x kpxq. xpX{Sq. We can choose an element b "ř1ďjďn sjtj `řn`1ďjďm rj tj P Notice that da b 1 " α P T OP,x prj P OSq, such that db b 1 " β. Observe that ¯b " a P OX,x. The element b induces a weak S Ð V 1 Ñ P , where V 1 is a Zariski neighborhood of x P P and the restriction test pair pu1, v1q : A1 of u1 to V 1 P X coincides with u : V Ñ A1 S in a neighborhood of x P X. This fact implies that, if α P SSw x piF , P {Sq. Hence, we get (4.13.1). (ii). By Lemma 4.10, this is a local problem for the Zariski topology on X. We may assume S . Since X is also smooth induced by the zero-section of that P is affine over S and that there exists an étale map χ : P Ñ Am over S, there is a canonical injection i0 : An Am´n S Ñ Am , such that the following diagram is Cartesian x pF , X{Sq, any vector β P di´1 x pαq is contained in SSw S S An´m S " An S S (4.13.3) X χ0 An S i l i0 P χ / Am S The canonical projection pr : Am S and also a section s : TpX{Sq ãÑ TpP {Sq P X of the canonical map di : TpP {Sq P X Ñ TpX{Sq. S induces a map χn " pr χ : P Ñ An S Ñ An S " An S S Am´n / /     / RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 17 Consider the following commutative diagram with Cartesian squares δ X / X An S X iid / P An S X pr2 / X pr1 1 l pr1 X / P i l χn χ0 / An S Since χ0 : X Ñ An S is étale, the scheme X An S X is a disjoint union of the diagonal δpXq and its complement Y . We denote by rP the complement of Y in P An X and by ρ : rP Ñ X and φ : rP Ñ P the restriction of pr1 and pr2 on rP , respectively. We observe that X " rP P X, which gives a section i : X Ñ rP of ρ : rP Ñ X. By Lemma 4.11(i), the base of SSpiF , P {Sq is contained in X. Hence, we have a closed conical subset ρφSSpiF , P {Sq " s´1pSSpiF , P {Sqq of TpX{Sq and we denote it by CpFq. Since F " RρφpiFq, we have CpFq P CpF , X{Sq (Lemma 4.12). For any C1 P CpF , X{Sq, we have iC1 P CpiF , P {Sq (loc. cit.). Hence SSpiF , P {Sq Ď iC1 and S CpFq " ρφSSpiF , P {Sq Ď ρφpiC1q " C1 Thus, CpFq is the smallest element of CpF , X{Sq, i.e., SSpF , X{Sq " CpFq. By loc. cit., we have SSpiF , P {Sq Ď iSSpF , X{Sq. Let θ : Am S Ñ Am composition pr χθ : P Ñ An TpP {Sq P X. Following the same argument above, we have S be an S-automorphism of Am S that satisfies θ i0 " i0. We put χθ " θ χ. The S induces a section sθ : TpX{Sq Ñ TpP {Sq P X of di : TpX{Sq Ñ (4.13.4) θ pSSpiF , P {Sqq " SSpF , X{Sq " s´1pSSpiF , P {Sqq. s´1 After shrinking S, we may assume that S is integral and we denote by η the generic point of S. To prove (4.13.2), it suffices to show (4.13.5) SSpiF , P {Sq S η " iSSpF , X{Sq S η The case where supppFq X Xη " H is trivial and we consider the case where supppFq X Xη ‰ H. Since iSSwpF , X{Sq " SSwpiF , P {Sq Ď SSpiF , P {Sq, the conical set SSpiF , P {Sq contains the restriction of the conormal bundle NXP to the support of iF . If (4.13.5) is false, we can fine a closed point z P Xη and a non-zero vector λ P T z pX{Sq, such that di´1 z pλq contains vectors in piSSpF , X{SqqzzpSSpiF , P {Sqqz. Replacing S by a Zariski open dense subset, we can take a sufficiently general χθ : P Ñ Am θ pSSpiF , P {Sqqqz. It contradicts to (4.13.4). (cid:3) S , such that λ R ps´1 5. Existence of relative singular supports In this section, let S be a connected Noetherian Zr1{ℓs-scheme. We take the notation and assumptions of 3.1 and 3.2. Lemma 5.1 (cf. [B15, Lemma 3.3]). Let C be a closed conical subset of TpP{Sq and F an object cpP, Λq. If F is micro-supported on C` relative to S, then the Radon transform RSpFq is of Db micro-supported on C_` relative to S. Conversely, if RSpFq is micro-supported on C_` relative to S, then, after replacing S by a Zariski open dense subset, F is micro-supported on C` relative to S. /   /   /   / / 18 HAOYU HU AND ENLIN YANG Proof. If F is micro-supported on C` relative to S, then the Radon transform RSpFq is micro- supported on pr_ prpC`q " C_` relative to S (Lemma 4.12). Conversely, if RSpFq is micro-supported on C_`, then the complex pR_ S RSqpFq is micro- supported on prpr_pC_`q " C__` " C`. After shrinking S, we may assume that the mapping cone of the adjunction F Ñ pR_ S RSqpFq has locally constant cohomologies (Proposition 3.11). By Lemma 4.11, F is also micro-supported on C` relative to S. (cid:3) Lemma 5.2 (cf. [B15, Lemma 3.4]). Let pg, hq : Y Ð U Ñ P be a test pair of P relative to S. We have the following diagram with Cartesian squares (3.2) (5.2.1) pr_ P_ H pr P hU HU l h prU U g Y Let C be a strict closed conical subset of TpP{Sq and E Ď H the image of PpCq in H by Legendre transform. Then, the test pair pg, hq is C-transversal relative to S if and only if the test pair pg prU , pr_ hU q of P_ is TP_-transversal relative to S at EU " E P U . In fact, Lemma 5.2 is deduced by [B15, Lemma 3.4] by taking fibers relative to S. Theorem 5.3 (cf. [B15, Theorem 3.2]). Let F be an object of Db by a Zariski open dense subset, the relative singular support SSpF , P{Sq exists and we have cpP, Λq. Then, after replacing S (5.3.1) PpSSpF , P{Sqq " Eprppr_RSpFqq Ď H by Legendre transform (3.2). Assume that S is irreducible and let η be the generic point of S. If supppFPη q ‰ Pη, then SSpF , P{Sq is the conical subset of TpP{Sq associated to Eprppr_RSpFqq. If, after shrinking S, we have supppFq " P, then SSpF , P{Sq is the union of conical subset of TpP{Sq associated to Eprppr_RSpFqq and the zero-section of TpP{Sq after shrinking S again. Proof. After shrinking S, we may assume that π : P Ñ S is universally locally acyclic with respect to F and that the mapping cone of the adjunction F Ñ pR_ S RSqpFq has locally constant cohomologies (Proposition 3.11). By smooth base change theorem, the composition π_pr : H Ñ S is universally locally acyclic with respect to prF . By proper base change theorem, the morphism π_ : P_ Ñ S is universally locally acyclic with respect to G " RSpFq. In the following, we replace F by R_ S pGq. We simply put E " Eprppr_RSpFqq Ď H and we denote by C Ď TpP{Sq the strict closed conical set associated to E by Legendre transform. Let C1 be an element of C minpF , P{Sq and E1 Ď H the Legendre transform of PpC1q. To prove the theorem, it is enough to show that C` " C1`. By the equality (3.2.1), the test pair ppr, pr_q : P Ð H ´ E1 Ñ P_ of P_ is C1_`-transversal relative to S. By Lemma 5.1, the complex F is micro-supported on C1 relative to S implies that the complex G is micro-supported on C1_` relative to S. Hence, the morphism pr : H ´ E1 Ñ P is universally locally acyclic with respect to pr_G, i.e., we have E Ď E1 and C` Ď C1`. To show C1` Ď C`, it suffices to verify that C` is an element of CpF , P{Sq. Let pg, hq : Y Ð U Ñ P be a test pair of P relative to S. We assume that pg, hq is C`-transversal relative to S. We take the notation of diagram (5.2.1) and we put EU " EPU . It is sufficient to show that g : U Ñ Y is locally acyclic with respect to hF " RprUph U ppr_Gqq. Since prU : HU Ñ U is proper, by / /   / /   / /   RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 19 proper base change theorem, we are reduced to prove that the morphism g prU : HU Ñ Y is locally acyclic with respect to h U ppr_Gq. Indeed, (i) By the definition of E, the morphism pr : H ´ E Ñ P is universally locally acyclic with respect to pr_G. Hence prU : HU ´ EU Ñ U is universally locally acyclic with respect U ppr_Gq. Since pg, hq is C`-transversal relative to S and C` contains the zero- to h section of TpP{Sq, the morphism g : U Ñ Y is smooth. By [SGA4 1 2 , 2.14], the morphism g prU : HU Ñ Y is locally acyclic with respect to h U ppr_Gq on HU ´ EU . (ii) By Lemma 5.2, the test pair pg prU , pr_ hU q of P_ is TP_-transversal relative to S at EU " E P U . Notice that π_ : P_ Ñ S is universally locally acyclic with respect to G. By Proposition 4.9, the morphism g prU : HU Ñ Y is locally acyclic with respect to h U ppr_Gq in a neighborhood of EU . (cid:3) Theorem 5.4 (cf. [B15, Theorem 1.3]). Let f : X Ñ S be a smooth morphism of finite type and F an object of Db cpX, Λq. Then, after replacing S by a Zariski open dense subscheme, the singular support SSpF , X{Sq of F relative to S exists. Proof. The case where Sred is the spectrum of a field is done by [B15, Theorem 1.3]. It suffices to consider the case where Sred is not the spectrum of a field. By Lemma 4.10, this is a local problem for the Zariski topology of X. We may assume that f : X Ñ S is affine. By Proposition 4.13, we are reduced to the case where X " An S. By Lemma 4.10 again, we are reduced to the case where X " Pn S. It is deduced by Theorem 5.3. (cid:3) Proposition 5.5 ([B15, Theorem 3.7]). Suppose that S is an irreducible excellent Noetherian cpP_, Λq. Then, after replacing S by a generically dimensional scheme. Let G be an object of Db Zariski open dense subscheme, we have dim Eprppr_Gq ď dim P ´ 1. The origin proof of [B15, Theorem 3.7] is for the case where S is an irreducible algebraic variety over a field. In fact, the same proof is also valid in our case. Corollary 5.6. Suppose that S is an irreducible excellent Noetherian scheme. Let F be an object of Db cpP, Λq. Then, after replacing S by a generically dimensional Zariski open dense subscheme, we have (5.6.1) dim SSpF , P{Sq ď dim P. Proof. By Proposition 5.5 (cf. [B15, Theorem 3.7]), after replacing S by a generically dimensional Zariski open dense subscheme, we have dimk Eprppr_RSpFqq ď dimk P ´ 1, i.e., dimk PpSSpF , P{Sqq ď dimk P ´ 1 (Theorem 5.3). Hence we obtain (5.6.1). (cid:3) Proposition 5.7 (cf. [B15, Theorem 1.7]). Suppose that S is an integral excellent Noetherian of Db dense subset, we have (i) For each irreducible component Dα (α P I) of D, there is a unique irreducible closed scheme and that the Veronese embeddingri : P Ñ rP in 3.1 has degree d ě 3. Let F be an object cpP, Λq and we put D " N CprRSpriFq,rP_q (2.8). Then, after replacing S by a Zariski open conical subset Cα Ď TpP{Sq such that Dα " rpr_pPpriCαqq and that the projection rpr_ : PpriCαq Ñ Dα is generically radicial. We have SSpF , P{Sq "ŤαPI Cα. (ii) The closed subscheme D is an effective Cartier divisor of rP_ relative to S. 20 HAOYU HU AND ENLIN YANG Proof. By Theorem 5.3, we may assume that SSpF , P{Sq and SSprRSpriFq,rP{Sq exist. We simply put C " SSpF , P{Sq and put rC " SSprRSpriFq,rP_{Sq. By Lemma 4.11(iii), we have D " BprCq. After shrinking S again, we have rC " priCq_ or rC " priCq_` (Proposition 4.13(ii) and Lemma 5.1). Hence, we get D " BprCq " rpr_pPpriCqq. We denote by tCαuαPI the set of irreducible dim PpriCq are generically dimensional. By Corollary 5.6, we have dim PpriCq ď dimrP_ ´ 1. components of C. By Lemma 3.3, we may assume that S and each irreducible component of Hence, after shrinking S, the map induces a one-to-one correspondence between the sets of irreducible components of C and D, and rpr_ : rH ÑrP_, PpriCαq ÞÑ Dα " rpr_pPpriCαqq s Ñ rPs and rpr_ moreover, rpr_ : PpriCαq Ñ Dα is generically small (Proposition 3.9). For any s P S, we denote byris : Ps Ñ rPs (resp. rprs : rH_ ofri : P Ñ rP (resp. rpr : rH_ Ñ rP and rpr_ : rH_ Ñ rP_) at s, and by rRs the Radon transform for the pair prprs, rpr_ s . By proper base change theorem, we have the canonical s q : rPs Ð rHs Ñ rP_ s Ñ rP_ s : rH_ s ) the fiber isomorphism Let η denotes the generic point of S. By Lemma 3.12, we have „ÝÑ rRsprispFPsqq. prRSpriFqqrP_ Dη " N CpprRSpriFqqrP_ η q " N CprRηpriηpFPη qq,rP_ ,rP_ η s η q. By [B15, Theorem 1.7], Dη " N CprRηpriηpFPη qq,rP_ D is a reduced closed subscheme of rP_ containing Dη, after shrinking S, the closed subscheme D is an effective Cartier divisor of rP_ relative to S. η q is an effective Cartier divisor of rP_ Theorem 5.8. Let S be an excellent Noetherian scheme, f : X Ñ S a smooth morphism of finite type and F an object of Db cpX, Λq. Then, after replacing S by a Zariski open dense subset, for any s P S, we have η . Since (cid:3) (5.8.1) SSpFXsq " pSSpF , X{Sq S sqred. Proof. By dévissage in the proof of Theorem 5.4, we are reduced to the case where X " P. We take the notation and assumptions of Proposition 5.7 and we assume that S is integral. We also omit the subscript "red" in the proof for simplicity. By Theorem 5.3, we may assume that C " SSpF , P{Sq exists. By (5.3.1), for any s P S, we have PpSSpFPsqq Ď PpCsq. After shrinking S, we may assume that supppFq " P if and only if, for any s P S, supppFPsq " Ps. Hence, we have SSpFPsq Ď Cs s surjective map for any s P S. It implies that PprispSSpFPsqqq is a closed subset of PpriCqs for any s P S. By Lemma 3.12, after shrinking S, for any s P S, we have Ds " N CpprRSpriFqqrP_ s q " N CprRsprispFPsqq,rP_ ,rP_ By [B15, Thorem 1.7] (Proposition 5.7 for S " s), the projection rpr_ rpr_ s : PprispSSpFPsqqq ։ Ds " N CprRsprispFPsqq,rP_ After shrinking S, the projection rpr_ : rH Ñ rP_ induces a one-to-one correspondence between the sets of irreducible components tPpripCαquαPI and tDαuαPI and the canonical surjection rpr_ : PpriCαq Ñ Dα is surjective and generically radicial (Proposition 5.7). Hence, after shrinking S, : rHs Ñ rP_ we may assume that (Lemma 3.6) induces a (5.8.2) s q. s q. s s (1) For each α P I, the components PpriCαq and Dα are flat over S; RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 21 (2) For any s P S, the projection rpr_ s : rHs ÑrP_ s induces a one-to-one correspondence between s : PpriCqsqβ Ñ pDsqβ is generically radicial. the irreducible components of PpriCqs and Ds. We denote by tpPpriCqsqβuβPJpsq (resp. tpDsqβuβPJpsq) the sets of irreducible components of PpriCqs (resp. Ds). For each β P Jpsq, the canonical projection rpr_ By (5.8.2) and (2), for any s P S, the closed subset PprispSSpFPsqqq contains all generic points of irreducible components of PpriCqs. Hence, for any s P S, we have Since bothrispSSpFPsqq andrispCsq are strict closed conical subsets of rPs, (5.8.3) implies that Since dris : TrPs rPs PprispSSpFPsqqq " PpriCqs. rispSSpFPsqq "rispCsq. By [EGA IV, II, 6.9.1], we have the following theorem. It implies that part (i) of Theorem 1.4 can be generalized to higher relative dimension cases if allowing to replace S by a Zariski open dense subset. Ps Ñ TPs is a surjective map, (5.8.4) implies that SSpFPsq " Cs. (cid:3) (5.8.3) (5.8.4) Theorem 5.9. Let S be an excellent Noetherian scheme, f : X Ñ S a smooth morphism of finite type and F an object of Db cpX, Λq. Then, after replacing S by a Zariski open dense subset, there exists a closed subscheme Z Ď TpX{Sq which is flat over Sred, such that, for any s P S, we have SSpFXsq " pZsqred. Corollary 5.10. Let X and S be irreducible excellent Noetherian schemes, f : X Ñ S a smooth morphism of finite type and F an object of Db cpX, Λq. Then, after replacing S by a generically dimensional Zariski open dense subset, we have X is generically dimensional, SSpF , X{Sq is equidimensional and has dimension dim X. Proof. It is deduced by Lemma 3.3, Theorem 5.8 and [B15, Theorem 1.3]. (cid:3) Corollary 5.11 (cf. [B15, Theorem 1.5]). Let S be an excellent Noetherian scheme, f : X Ñ S a smooth morphism of finite type and F an object of Db cpX, Λq. Then, after replacing S by a Zariski open dense subset, we have SSwpF , X{Sq " SSpF , X{Sq. Proof. We may assume that S is irreducible and let η be the generic point of S. We simply put B " SSwpF , X{Sq and put C " SSpF , X{Sq. It is sufficient to prove that Bη " Cη. A priori, we have B Ď C, hence Bη Ď Cη. Any Bη-transversal weak test pair pg0, h0q : A1 η Ð U0 Ñ Xη can be extended to a B-transversal week test pair pg, hq : A1 S Ð U Ñ X relative to S. Hence, any Bη-transversal weak test pair is FXη -acyclic, i.e., SSwpFXη q Ď Bη. By theorem 5.8, we have SSpFXη q " Cη. By [B15, Theorem 1.5], we have SSwpFXη q " SSpFXη q. In summary, we have Cη " SSpFXη q " SSwpFXη q Ď Bη Ď Cη. (cid:3) Corollary 5.12. Let S and S1 be excellent Noetherian schemes, f : S1 Ñ S a dominant morphism, f : X Ñ S a smooth morphism of finite type, X 1 " X S S1 and F an object of Db cpX, Λq. Then, after replacing S and S1 by Zariski open dense subsets, we have (5.12.1) SSpFX 1 , X 1{S1q " pSSpF , X{Sq S S1qred. 22 HAOYU HU AND ENLIN YANG Proof. We may assume that S and S1 are irreducible. Let η and η1 be the generic points of S and S1, respectively. By [B15, Theorem 1.4] and Theorem 5.8, we have SSpFX 1, X 1{S1q S1 η1 " SSpFη1q " pSSpFηq η η1qred " pSSpF , X{Sq S η1qred. Then, after shrinking S1 again, we get (5.12.1). (cid:3) 6. Generic constancy of characteristic cycles In this section, let k be a perfect field of characteristic p ą 0 (p ‰ ℓ). All k-schemes are assumed to be of finite type over Specpkq. 6.1. Let X be a smooth k-scheme of equidimension n, C a closed conical subset of TX and f : X Ñ A1 k a k-morphism. A closed point v P X is called an at most C-isolated characteristic point of f : X Ñ A1 k is C-transversal. A closed point v P X is called a C-isolated characteristic point if v is an at most C-isolated characteristic point of f : X Ñ A1 k if there is an open neighborhood V Ď X of v such that f : V ´ tvu Ñ A1 k is not C-transversal at v. k but f : X Ñ A1 We assume that C Ď TX is of equidimension n. Let tCαuαPI be the set of irreducible compo- k defines a section θ : X Ñ TA1 X, X Ñ TX is the canonical morphism. nents of C. We denote by o the origin of A1 hence a section df θ : X Ñ TX, where df : TA1 Let A " řαPI mαrCαs pmα P Zq be an n-cycle supported on C and v P X an at most C-isolated characteristic point of f : X Ñ A1 the intersection pA, rpdf θqpW qsq is a 0-cycle supported on a closed point of T is independent of the choice of the base θ P T pA, df qTX,v. k. Then, there is an open neighborhood V of v in X such that Its degree k and we denote this intersection number by k. A base θ P T k A1 kA1 o A1 k o A1 v V . k 6.2. Db ([Sa16, Theorem 5.9]) Let X be a smooth k-scheme of equidimension n, F an object of cpX, Λq and tCαuαPI the set of irreducible components of SSpFq. Then, there exists a unique n-cycle CCpFq " řαPI mαrCαs pmα P Zq of TX supported on SSpFq, satisfying the following Milnor type formula (6.2.1): For any étale morphsim g : V Ñ X, any morphism f : V Ñ A1 k, any isolated gSSpFq- characteristic point v P V of f : V Ñ A1 k and a geometric point ¯v of V above v, we have (6.2.1) p´1qidimtotpRiΦ¯vpgF , f qq " pgCCpFq, df qTV,v, ´ÿi where RΦ¯vpgF , f q denotes the stalk at ¯v of the vanishing cycle complex of gF relative to f , dimtotpRiΦ¯vpgF , f qq the total dimension of RiΦ¯vpgF , f q (2.9) and gCCpFq the pull-back of CCpFq to TV . We call CCpFq the characteristic cycle of F . Proposition 6.3 ([Sa16, Lemma 5.11, Lemma 5.13]). Let X be a smooth k-scheme of equidimen- sion n and F be an object of Db cpX, Λq. Then (i) For any étale morphism j : U Ñ X, we have jCCpFq " CCpjFq. (ii) For a smooth k-scheme P of equidimension m and a closed immersion h : X Ñ P , we have hpCCpFqq " CCphFq, where hpCCpFqq denotes the push-forward of the cycle p´1qm´ndhpCCpFqq on TP P X to TP . The index formula for ℓ-adic sheaves is the following: RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 23 Theorem 6.4 ([Sa16, Theorem 7.13]). Let ¯k be an algebraic closure of k, X a smooth and projective k-scheme and F an object of Db cpX, Λq. Then, we have (6.4.1) χpX¯k, FX¯k q " degpCCpFq, T X XqTX , where χpX¯k, FX¯k q denotes the Euler-Poincaré characteristic of FX¯k . Example 6.5. Let T be a smooth k-curve, E an effective Cartier divisor of T , T0 the complement of E in T , r : T0 Ñ T is the canonical injection and G a locally constant and constructible sheaf of Λ-modules on T0. Then, we have CCpr!Gq " ´rkΛpGq rT dimtottpGq rT t T s, T T s ´ ÿtPE where dimtottpGq denotes the total dimension of G at t (2.9). When T is proper, the index formula (6.4.1) for r!G is exactly the Grothendieck-Ogg-Shafarevich formula [SGA5]. 6.6. In the following of this section, we assume that all schemes are over Zr1{ℓs. Let D / X ❅❅❅❅❅❅❅❅ f g S Y ⑧⑧⑧⑧⑧⑧⑧⑧ h be a commutative diagram of morphisms of finite type of Noetherian schemes, where g : X Ñ Y is smooth, h : Y Ñ S is smooth of relative dimension 1, D is a closed subscheme of X such that f D : D Ñ S is qusai-finite. Denote by U the complement of D in X. Let F be an object of Db cpX, Λq. Assume that f : X Ñ S is locally acyclic with respect to F and that gU : U Ñ Y is locally acyclic with respect to FU . Let x be a point of D and ¯x a geometric point of X above x, ¯s a geometric point of S above s " f pxq and RΦ¯xpFX¯s , g¯sq the stalk of the vanishing cycles complex of FX¯s relative to g¯s : X¯s Ñ Y¯s. We define a function ϕF ,g : D Ñ Z by (6.6.1) p´1qidimtotpRiΦ¯xpFX¯s , g¯sqq, ϕF ,gpxq "ÿi which is independent of the choice of geometric point ¯x of X above x. The following proposition is the semi-continuity of total dimensions of vanishing cycles complex. Proposition 6.7 ([Sa16, Proposition 2.16]). We take the notation and assumptions of 6.6. Then, the function ϕF ,g : D Ñ Z is constructible. If f D : D Ñ S is étale, the function fpϕF ,gq : S Ñ Z, is locally constant on S. s ÞÑ ÿxPD¯s ϕF ,gpxq 6.8. Let S be an Noetherian scheme, f : X Ñ S a smooth morphism of finite type and F an object of Db relative to S if each Bi is open and equidimensional over S and if, for any algebraic geometric point ¯s of S, we have cpX, Λq. A cycle B " řiPI mirBis in TpX{Sq is called the characteristic cycle of F (6.8.1) mirpBiq¯ss " CCpFX¯s q B¯s "ÿiPI We denote by CCpF , X{Sq the characteristic cycle of F on X relative to S. Notice that relative characteristic cycles do not always exist. / / /   24 HAOYU HU AND ENLIN YANG Theorem 6.9. Let S be an excellent Noetherian scheme, f : X Ñ S a smooth morphism of finite type and F an object of Db cpX, Λq. Then, there exists a dominant and quasi-finite morphism π : S1 Ñ S such that the relative characteristic cycle CCpFXS S1 , pX S S1q{S1q exists. Theorem 6.9 generalizes part (ii) of Theorem 1.4 to higher relative dimensional cases. Proof. We may assume that S is affine and integral. By Proposition 6.3, we are reduced to the case where X is a projective space over S. We take the notation of 3.1 and 3.2, and we write X " P. By Theorem 5.8, we may assume that the relative singular support C " SSpF , P{Sq exists and that, for any s P S, we have SSpFPsq " pCsqred. s s rGs x P Ly) , s denotes the projective line associated to y P rGs. Step 1. We denote by rV the free OS-module ΓprP, OrPp1qq and by rV _ the free OS-module ΓprP_, OrP_p1qq. They are of rank N . Let rG be the Grassmannian of projective line in rP_, i.e., the Grassmannian GrprV _, 2q of projective quotient OS-module of rV _ of rank 2. Let rD be the universal line in rP_ S rG, i.e., for each s P S, rDs "!px, yq PrP_ where Ly Ď rP_ variety FlprV _, 2, 1q over S characterized by surjections rV _ ։ L2 ։ L1 of projective OS-modules where rankOS pL2q " 2 and rankOS pL1q " 1. Let rA Ď rP S rG be the universal axis, i.e., for any where Hz denotes the hyperplane of rPs associated to z P rP_ FlprV , N ´ 2, 1q over S characterized by the surjections rV ։ L1 1q " 1. Notice that rA is of codimensional 2 in rP S rG. We denote by ρ : rD Ñ rP_ and τ : rD Ñ rG the canonical projections and by π : P rP rH Ñ rP_ the composition ofri id : P rP rH Ñ rH and the projection rpr_ : rH Ñ rP_. We have a commutative rAs "!px, yq P rPs s rGs, x P Hz, for any z P Ly) , s . It is isomorphic to the Flag variety 1 of projective OS-modules N ´2q " N ´ 2 and rankOS pL1 It is isomorphic to the Flag diagram (cf. [Sa16, (5.2)]) where rankOS pL1 ։ L1 N ´1 s P S, (6.9.1) pP S rGq1 π1 ρ1 l ρ π P rP rH rP_ τ 1 τ pr2 / P S rG rG rD Step 2. In the rest of the proof, we simply put C " SSpF , P{Sq. We denote by tCαuαPI the set where pP S rGq1 denotes the blow-up of P S rG along rAŞpP S rGq and τ 1 is the canonical projection. Due to [Sa16, Lemma 5.2], the left square of (6.9.1) is Cartesian. We put pP S rGq " pP S rGqzprAŞpP S rGqq. of irreducible components of C, by rC (resp. rCα) the inverse image pdriq´1pCq (resp. pdriq´1pCαq) in P rP TprP{Sq (resp. pdriq´1pCαq Ď P rP TprP{Sq), by PprCq (resp. PprCαq) the projectivization of rC (resp. rCα) in P rP rH by Legendre transform and by rD (resp. rDα) the image πprCq (resp. πprCαq) in rP_. By [EGA IV, II, 6.9.1], we may assume that, for any α P I, Cα (resp. rCα and rDα) for any α P I, the fiber product Cη1 " C S η1 is geometrically irreducible (hence, PprCαqη1 and prDαqη1 are also geometrically irreducible). Let S1 be the normalization of S in η1. By Corollary Let η be the generic point of S. There exists a point η1 and finite morphism η1 Ñ η such that, is flat over S.     o o /   / / o o RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 25 5.12, after replacing S by a Zariski open dense subset of S1 (we still denote it by S for simplicity), we may assume that the relative singular support C exists and that, for any α P I, the generic Replacing S again by a Zariski open dense subscheme, we may assume that, for any s P S and and that pCαqs's are different from each other ([EGA IV, III, 9.7.8]). Assume that the Veronese fiber pCαqη is geometrically integral (hence, PprCαqη and prDαqη are also geometrically integral). any α P I, the fibers pCαqs, prDαqs and PprCαqs are geometrically integral ([EGA IV, III, 9.7.7]), embeddingri : P Ñ rP has degree d ě 3. By Lemma 3.6 and Proposition 3.9, we may assume that, for any s P S and any α P I, the canonical projection πs : PprCαqs Ñ prDαqs is generically radicial. Step 3. Let pP S rGq∇ be the largest open subset of pP S rGq such that the inverse image ZprCq " PprCq PrPrH pP S rGq∇ is quasi-finite over rG. Using [Sa16, Lemma 3.10] fiberwisely, we obtain that the complement pP rP rHqzPprCq is the largest open subset of P rP rH where the test pair pπ, pr1q : P_ Ð P rP rH Ñ P is C-transversal relative to S. Then, pP S rGq∇zZprCq is the largest open subset of pP S rGq∇ where the test pair pπ ρ1, pr1q : rD Ð pP S rGq∇ Ñ P is C-transversal relative to S. Hence, π1 : pP S rGq∇zZprCq Ñ rD is universally locally acyclic with respect to G " FpPSrGq∇. After shrinking S, we may assume that the canonical projection P Ñ S is universally locally acyclic with to F , hence, that pr2 : pP S rGq∇ Ñ rG is universally locally acyclic with respect to G. Consider the following commutative diagram (6.9.2) ZprCq / pP S rGq∇ $❍❍❍❍❍❍❍❍❍❍ pr2 / rD ✂✂✂✂✂✂✂✂ τ By [Sa16, Proposition 2.16] (cf. Proposition 6.7), the function (6.6.1) π1 rG ϕG,π1 : ZprCq Ñ Z, is constructible. For any α P I, we put ZprCαq " PprCαq PrPrH pP S rGq∇. Then, there exists an open dense subset Z 1prCq of ZprCq, such that ϕG,π1 is locally constant on Z 1prCq and that, for any α, β P I, the subsets Z 1prCαq " Z 1prCq X ZprCαq and Z 1prCβq " Z 1prCq X ZprCβq are disjoint. Notice that the projection ρ1 : pPSrGq∇ Ñ PrPrH is smooth and has geometrically integral fibers. Hence, we may assume that, after shrinking S, for any α P I, the scheme Z 1prCαq is irreducible and, for any s P S, the fiber Z 1prCαqs is non-empty and geometrically irreducible. Step 4. For any α P I, Let ξα (resp. ζα) be the generic point of PprCαq (resp. rDα) and ϕα the value of ϕG,π1 on Z 1prCαq. We put ϕα (6.9.3) B " ´ÿαPI rCαs. rξα : ζαs After replacing S by a Zariski open dense subscheme, we may assume that, for any α P I and any s P S, the fiber Z 1prCαqs is non-empty. Let ¯s be an algebraic geometric point of S. For any α P I, we denote by pξαq¯s (resp. pζαq¯s) the generic point of PprCαq¯s (resp. prDαq¯s). By [Sa16, Proposition 5.8, Theorem 5.9], the characteristic cycle CCpFP ¯s q is defined by (6.9.4) CCpFP ¯s q " ´ÿαPI ϕα rpξαq¯s : pζαq¯ss rpCαq¯ss. / $ / 26 HAOYU HU AND ENLIN YANG By Lemma 3.7, for any α P I and any algebraic geometric point ¯s of S, we have (6.9.5) rξα : ζαs " rpξαq¯s : pζαq¯ss. By (6.9.3), (6.9.4) and (6.9.5), we obtain that, for any algebraic geometric point ¯t of S, B¯t " CCpFP¯t q. Hence, B is the characteristic cycle of FXS S1 relative to S1. (cid:3) Proposition 6.10 (Saito). We assume that k is algebraically closed. Let T be a connected and smooth k-scheme of dimension n, g : Y Ñ T a smooth morphism of finite type and G an object of Db cpY, Λq. Assume that g : Y Ñ T is SSpGq-transversal and that each irreducible component of SSpGq is open and equidimensional over T . Then, the relative characteristic cycle CCpG, Y {Sq exists, and we have (6.10.1) CCpG, Y {Sq " p´1qnθpCCpGqq, where θ : TY Ñ TpY {T q denotes the projection induced by the canonical map Ω1 Y {k Ñ Ω1 Y {T . Proof. For any closed point t of T , we have the following Cartesian diagram TY θ TpY {T q i1 t l it TY Y Yt θt TYt where it denotes the canonical injection. We may assume that Y is of equidiemension d ` n. Since SSpGq is equidimensional over T , for any closed point t of T , the fiber product SSpGq Y Yt is of equidimension d. Since g : Y Ñ T is SSpGq-transversal, for any closed point t of T , the canonical injection ιt : Yt Ñ Y is SSpGq-transversal. Hence, for any closed point t P T , the canonical injection ιt : Yt Ñ Y is propoerly SSpGq-transversal ([Sa16, Definition 7.1]). Then, for any closed point t in T , we have ([Sa16, Theorem 7.6]) (6.10.2) CCpGYt q " p´1qnθtpi1 t pCCpGqqq. The restriction map θ : SSpGq Ñ θpSSpGqq is finite, since g : Y Ñ T is SSpGq-transversal ([Sa16, Lemma 3.1]). Then, each irreducible component of θpSSpGqq is also open and equidimensional over T . For any closed point t of T , we have an equality of cycles (6.10.3) θtpi1 t pCCpGqqq " i t pθpCCpGqqq. on TYt. We put B " p´1qnθpCCpGqq. By (6.10.2) and (6.10.3), for any closed point t of T , we have (6.10.4) CCpGYt q " i t B. By Theorem 6.9, there exists an integral k-scheme T 1 and a dominant and quasi-finite morphism π : T 1 Ñ T such that CCpGY T T 1 , pY T T 1q{T 1q exists. By (6.10.4) and the definition of relative characteristic cycles, we must have (6.10.5) π1B " CCpGY T T 1 , pY T T 1q{T 1q, where π1 : TpppY T T 1q{T 1q Ñ TpY {T q denotes the projection induced by π : T 1 Ñ T . Hence, B " CCpG, Y {T q. (cid:3)   o o   o o RELATIVE SINGULAR SUPPORT AND THE SEMI-CONTINUITY OF CHARACTERISTIC CYCLES 27 7. Failure of the lower semi-continuity for characteristic cycles In this section, let k denotes an algebraically closed field of characteristic p ě 3, S " 7.1. Specpkrusq, X " Specpkrx, y, zsq, D " pzq, U " X ´ D and f : X Ñ S the projection associated to krus Ñ krx, y, zs, u ÞÑ x. Example 7.2. Let G be a locally constant and constructible sheaf of Λ-modules of rank 1 on U associated to the Artin-Schreier covering defined by tp ´ t " pxyq{zp and we put F " j!Gr2s. For each s P Spkq, we have Xs " Specpkry, zsq and Us " Specpkry, z, z´1sq. The restriction GUs is a locally constant and constructible sheaf of Λ-modules of rank 1 on Us associated to the Artin- Schreier covering defined by tp ´ t " psyq{zp. Notice that GUs is the constant sheaf ΛUs when s " 0. We have ([Sa15, Example 2.2]) SSpFXsq "" T Xs T Xs XsŤ Ds xdzy, XsŤ Ds xdyy, if s " 0, if s ‰ 0, where Ds xdzy (resp. Ds xdyy) denotes the sub-line bundle of TXs Xs Ds spanned by the section dz (resp. dy). We see that part (i) of Theorem 1.4 cannot be generalized to F . Example 7.3. We take the sheaf G in Example 7.2. Let G1 be the constant sheaf ΛU and G2 the locally constant sheaf of Λ-modules of rank 1 on U associated to the Artin-Schreier covering defined by tp ´ t " y{zp. We put F 1 " j!pG1 ' G2 ' Gqr2s. For any closed point s P S, we have SSpF 1Xsq " T Xs Xsď Ds xdyyď Ds xdzy. We denote by D xdyy (resp. D xdzy) the sub-line bundle of TpX{SqX D spanned by the section dy (resp. dz). Hence, the closed subset A " T X pX{Sqď D xdyyď D xdzy of TpX{Sq satisfies part (i) of Theorem 1.4 for F 1 in the higher relative dimension situation. However, by [Sa13, Example 3.6], we have CCpF 1Xsq "" 3rT Xs 3rT Xs Xss ` 2rDs xdzys ` prDs xdyys, Xss ` rDs xdzys ` 2prDs xdyys, if s " 0, if s ‰ 0. It implies that F 1 does not have a lower semi-continuous property at the origin of S. Hence, we cannot find a finite and surjective map f : S1 Ñ S and a cycle B supported on A S S1 that generalize part (iii) of Theorem 1.4 for F 1. References [AS02] A. Abbes and T. Saito, Ramification of local fields with imperfect residue fields. Amer. J. Math. 124, (2002), 879 -- 920. [An] Y. Andre, Structure des connexions méromorphes formelles de plusieurs variables et semi-continuité de l'irrégularité, Invent. Math. 170, (2007), 147 -- 198. [B15] A. Beilinson, Constructible sheaves are holonomic. Selecta Math. 22, Issue 4, (2016), 1797 -- 1819. [Bry] J.-L. Brylinski, Transformations canoniques, dualité projective, théorie de Lefschetz, transformations de Fourier et sommes trigonométriques, Astérisque 140 -- 141, (1986), 3 -- 134. 2 ] P. Deligne et al., Cohomologie Étale. Lecture Notes in Math. 569, Springer-Verlag, (1977). [SGA4 1 [De] P. Deligne, Letter to Katz (Dec. 1, 1976), Singularités irrégulières: correspondance et documents, Documents Mathématiques 5, (2007). [Du] A. Dubson, Classes caractéristiques des varétés singulières, C. R. Acad. Sci. Paris 287, (1978), 237 -- 240. [HY] H. Hu and E. Yang, Semi-continuity for total dimension divisors of étale sheaves, Internat. J. Math. 28, no. 1, (2017), 21 pp. 28 HAOYU HU AND ENLIN YANG [L81] G. Laumon, Semi-continuité du conducteur de Swan (d'après Deligne). Séminaire E.N.S. (1978-1979) Exposé 9, Astérisque 82-83, (1981), 173-219. [Fu] L. Fu, Etale Cohomology Theory, Nankai Tracts in Mathematics Vol.14, revised Edition, 2015. [Ful98] W. Fulton, Intersection theory. Springer-Verlag, Berlin, second edition, 1998. [EGA III] A. Grothendieck and J. Dieudonné, Éléments de géométrie algébrique: III Étude cohomologique des faisceaux cohérents. Publ. Math. Inst. Hautes Sci. 11, (1961), 5 -- 167; 17, (1963). 5 -- 91. [EGA IV] A. Grothendieck and J. Dieudonné, Éléments de géométrie algébrique: IV Étude locale des schémas et des morphismes de schémas. Publ. Math. Inst. Hautes Sci. 20, (1964), 5 -- 259; 24, (1965), 5 -- 231; 28, (1966), 5 -- 255; 32, (1967), 5 -- 361. [SGA5] A. Grothendieck et al., Cohomologie ℓ-adique et fonctions L. Séminaire de Géométrie Algébrique du Bois- Marie 1965 -- 1966 (SGA 5). dirigé par A. Grothendieck avec la collaboration de I. Bucur, C. Houzel, L. Illusie, J.-P. Jouanolou et J-P. Serre. Lecture Notes in Mathematics 589, Springer-verlag, Berlin-New York, (1977). [Ka83] M. Kashiwara, Systems of microdifferential equations. Progress in Math. 34. Birkhauser, Boston, 1983. [O] F. Orgogozo, Modifications et cycles proches sur une base générale. Int. Math. Res. Not. 2006, Art. ID 25315, 38 pp. [Sa13] T. Saito, Wild ramification and the contangent bundle. J. Algebraic Geometry., Article electronically pub- lished on September 19, 2016. [Sa15] T. Saito, Characteristic cycle and the Euler number of a constructible sheaf on a surface. Kodaira Centennial issue of the Journal of Mathematical Sciences, the University of Tokyo, (2015) 22: 387-442. [Sa16] T. Saito, The characteristic cycle and the singular support of a constructible sheaf. Invent. Math. Published online 15 July 2016. [Se97] J.P. Serre, Linear representations of finite groups. Graduate Texts in Mathematics 42. Springer-Verlag, New York-Heidelberg, 1997. Graduate School of Mathematical Sciences, the University of Tokyo, 3-8-1 Komaba Meguro-ku Tokyo 153-8914, Japan E-mail address: [email protected], [email protected] Fakultät für Mathematik, Universität Regensburg, 93040 Regensburg, Germany E-mail address: [email protected], [email protected]
1709.07392
1
1709
2017-09-21T15:54:33
A mirror theorem for genus two Gromov-Witten invariants of quintic threefolds
[ "math.AG", "hep-th", "math.SG" ]
We derive a closed formula for the generating function of genus two Gromov-Witten invariants of quintic 3-folds and verify the corresponding mirror symmetry conjecture of Bershadsky, Cecotti, Ooguri and Vafa.
math.AG
math
A MIRROR THEOREM FOR GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS SHUAI GUO, FELIX JANDA, AND YONGBIN RUAN Abstract. We derive a closed formula for the generating function of genus two Gromov– Witten invariants of quintic 3-folds and verify the corresponding mirror symmetry conjecture of Bershadsky, Cecotti, Ooguri and Vafa. Contents Introduction 1. 2. A localization formula 2.1. Moduli space 2.2. Localization 2.3. Formula in genus two 2.4. Example 3. Proof of the Main Theorem 3.1. Genus zero mirror theorem for the twisted theory of P4 3.2. Quantum product and "extra" generators 3.3. A list of closed formulae for the contribution of localization graphs 3.4. Proof of the Main Theorem 3.5. Equivalence between our Main Theorem and the physicists' conjecture 4. Structures of the twisted invariants 4.1. S-matrix 4.2. Ψ-matrix and R-matrix: computations of Ψ1 and R∗1 4.3. Ψ-matrix and R-matrix: computations of the remaining entries 4.4. Generators and relations 5. Proof of key propositions 5.1. Proof of Proposition 3.7 5.2. Proof of Proposition 3.6 5.3. Proof of Proposition 3.5 References 1 5 5 7 9 10 13 13 14 16 18 19 20 20 23 27 29 32 33 35 38 42 1. Introduction The computation of the Gromov–Witten (GW) theory of compact Calabi–Yau 3-folds is a central problem in geometry and physics where mirror symmetry plays a key role. In the early 90's, the physicists Candelas and his collaborators [2] surprised the mathematical community to use the mirror symmetry to derive a conjectural formula of a certain generating function (the J-function, see Section 3 for its definition) of genus zero Gromov–Witten invariants of Date: September 2017. 1 2 S. GUO, F. JANDA, AND Y. RUAN a quintic 3-fold in terms of the period integral or the I-function of its B-model mirror. The effort to prove the formula directly leads to the birth of mirror symmetry as a mathematical subject. Its eventual resolution by Givental [17] and Liu–Lian–Yau [33] was considered to be a major event in mathematics during the 90's. Unfortunately, the computation of higher genus GW invariants of compact Calabi–Yau manifolds (such as quintic 3-folds) turns out to be a very difficult problem. For the last twenty years, many techniques have been developed. These techniques have been very successful for so-called semisimple cases such as Fano or toric Calabi–Yau 3-folds. In fact, they were understood thoroughly in several different ways. But these techniques have little effect on our original problem on compact Calabi–Yau 3-folds. For example, using B-model techniques, Bershadsky, Cecotti, Ooguri and Vafa (BCOV) have already proposed a conjectural formula for genus one and two Gromov–Witten invariants of quintic 3-folds as early as 1993 [1] (see also [42]). It took another ten years for Zinger to prove BCOV's conjecture in genus one [45]. During the last ten years, an effort has been made to push Zinger's technique to higher genus without success. Nevertheless, the problem inspires many developments in the subject such as the modularity problem [1, 27], FJRW-theory [15] and algebraic mathematical GLSM theory [16]. It was considered as one of guiding problems in the subject. The main purpose of this article is to prove BCOV's conjecture in genus two. To describe the conjecture explicitly, let us consider the so called I-function of the quintic 3-fold I(q, z) = zXd≥0 k=1(5H + kz) k=1(H + kz)5 qdQ5d Qd where H is a formal variable satisfying H 4 = 0. The I-function satisfies the following Picard–Fuchs equation 4 H − 5q (cid:16)D4 (5DH + kz)(cid:17)I(q, z) = 0, Yk=1 where DH := zq d dq + H. We separate I(q, z) into components: I(q, z) = zI0(q)1 + I1(q)H + z−1I2(q)H 2 + z−2I3(q)H 3 The genus zero mirror symmetry conjecture of quintic 3-fold can be phrased as a relation between the J- and the I-function up to a mirror map Q = qeτQ(q), where J(Q) = I(q) I0(q) τQ(q) = I1(q) I0(q) . The leading terms of I0 and τQ are I0(q) = 1 + 120 q + 113400 q2 + 168168000 q3 + O(q4), τQ(q) = 770 q + 717825 q2 + 3225308000 3 q3 + O(q4). Now we introduce the following degree k "basic" generators Xk := dk duk (cid:18)log I0 L(cid:19) , Yk := dk duk (cid:18)log I0I1,1 L2 (cid:19) , Zk := dk duk (cid:16)log(q 1 5 L)(cid:17) , GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 3 where I1,1 := 1 + q d generators are given in Remark 3.4. dq τQ, L := (1 − 55q)− 1 5 and du := L dq q . Some numerical data for these Let F GW g (Q) be the generating function of genus g Gromov–Witten invariants of a quin- tic 3-fold. The following is an equivalent formulation of the genus two mirror conjecture (Conjecture 3.10) of [42, 27] (see Section 3.5 for the argument). Conjecture 1.1. The genus 2 GW generating function F GW given by (Q) for the quintic threefold is 2 F GW 2 (Q) = I 2 0 L2 ·(cid:16) 70 X3 625 ZX2 9 575 X X2 18 175 ZYX + − 36 41 Z2Y 48 − 9 625 Z 3 144 + + + + 5YX2 6 + 1441 Z2X + 48 2233 ZZ2 72 − + 547 Z3 72 (cid:17), 128 are 557 X 3 − 629 YX 2 72 − 23 Y 2X 24 − Y 3 24 25 Z(X 2 + Y 2) 24 − 3125 Z 2(X + Y) 288 where Q = qeτQ(q). In particular, the leading terms of F GW 2 575 48 The following is our Main Theorem: (Q) = − 5 144 F GW + 2 Q + 5125 2 Q2 + 7930375 6 Q3 + O(Q4). g Theorem 1.2. The above genus two mirror conjecture of quintic 3-fold holds. Remark 1.3. A consequence of above conjecture is that (L/I0)2(q)F GW (Q) (more generally, (L/I0)2g−2(q)F GW (Q)) is a homogeneous polynomial of the generators Xk, Yk, Zk. We refer to this property as finite generation. On the other hand, the original conjecture (Conjec- ture 3.10) is an inhomogeneous polynomial of five generators. We found it easier to work with a homogeneous polynomial than an inhomogeneous polynomial. Since the Taylor expansions of the generators are known, we can easily compute numerical genus two GW-invariants for any degree. 2 As we mentioned previously, it has been ten years since Zinger proved the genus one BCOV mirror conjecture. A key new advancement during last ten years was the understanding of global mirror symmetry which was in the physics literature in the beginning but somehow lost in its translation into mathematics in the early 90's. The idea of the global mirror symmetry [2, 1, 27, 7] is to view GW theory as a particular limit (large complex structure limit) of the global B-model theory. Physicists use the results about the other limits such as the Gepner limit and conifold limit to yield the computation of the large complex structure limit/GW theory. Interestingly, one of the key pieces of information they used is the regu- larity of the Gepner limit. The latter can be interpreted as the existence of FJRW theory. A natural consequence of the above global mirror symmetry perspective is a prediction that the GW/FJRW generating functions are quasi-modular forms in some sense and hence are polynomials of certain finitely many canonical generators [1, 27, 42]. This imposes a strong structure for GW theory and we refer it as the finite generation property. For anyone with experience on the complexity of numerical GW invariants, it is not difficult to appreciate how amazing the finite generation property is! In fact, it immediately reduces an infinite computation for all degree to a finite computation of the coefficients of a polynomial. There- fore, it should be considered as one of the fundamental problems in the subject of higher genus GW theory. 4 S. GUO, F. JANDA, AND Y. RUAN The above global mirror symmetry framework was successfully carried out by the third au- thor and his collaborators [32, 36, 28] for certain maximal quotients of Calabi–Yau manifolds. These examples are interesting in their own right. Unfortunately, we understand very little about the relation between the GW-theory of a Calabi–Yau manifold and its quotient. Hence, the success on its quotient has only a limited impact on our original problem. Several years ago, an algebraic-geometric curve-counting theory was constructed by Fan–Jarvis–Ruan for so called gauged linear sigma model (GLSM) (see [41, 26] for its physical origin). One appli- cation of mathematical GLSM theory is to interpret the above global mirror symmetry as wall-crossing problem for a certain stability parameter ǫ of the GLSM-theory. It leads to a complete new approach to attack the problem without considering B-model at all. The first part of the new approach is to vary the stability parameter from ǫ = ∞ (stable map theory) to ǫ = 0+ (quasimap [9] or stable quotient [34] theory) and has been successfully carried out recently by several authors [8, 11, 12, 44]. Suppose that F SQ (q) is the the genus g generating function of stable quotient theory. Then, the above authors have proved g F GW g (Q) = I 2−2g 0 (q)F SQ g (q), which explains the appearance of I 2 0 at the right hand side of the conjecture. In the current paper, we take the next step to calculate the genus two generating function in quasimap theory and verify the conjectural formula in [1, 42]. The current paper relies on certain geometric input from [5] (see also [6]) which we now describe. Recall that the virtual cycle of the GLSM (stable map with p-field in this case) moduli space was constructed using cosection localization on an open moduli space [30]. The cosection is not C∗-invariant which prevents us from applying the localization technique. A naive idea is to construct a compactification of the GLSM moduli space such that the cosection localized class can be identified with the virtual cycle of the compactified moduli space. Hopefully, the latter carries C∗-action and the localization formula can be applied. Unfortunately, it is not easy to make naive idea work due to the difficulty of extending the cosection to the compactified moduli space. Working with Qile Chen, the last two authors solved the problem by introducing a certain "reduced virtual cycle" on an appropriate log compactification of the GLSM moduli space. In a sense, the current article and [5] belong together. Of course, [5] provides a general tool with applications beyond the current article. We will briefly describe [5] (specialized to a quintic 3-fold) in Section 2. Taking the localization formula of [5] as an input, we can express genus g Gromov–Witten invariants of a quintic 3-fold as a graph sum of twisted equivariant Gromov–Witten invariants of P4 and certain effective invariants. When g = 2, 3, the effective invariants can be computed from known degree zero Gromov–Witten invariants. The main content of the current article is to extract a closed formula for the generating functions. This is of course difficult in general. We solve it using Givental formalism. A subtle and yet interesting phenomenon is the choice of twisted theory. The general twisted theory naturally depends on six equivariant parameters, five for the base P4 and one for the twist. It is complicated to study the general twisted theory, and therefore Zagier–Zinger [43] specialize the equivariant parameters of the base to scalar multiples of (1, ξ, ξ2, ξ3, ξ4, 0), where ξ is a primitive fifth root of unity. Under this specialization, they show that the twisted theory is generated by the five generators predicted by physicists. Unfortunately, in our work we cannot set the equivariant parameter for the twist to zero. As a consequence, we have to introduce four extra generators. It was a miracle to us that the terms involving the four extra generators cancel and we have our GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 5 theorem! The appearance of four extra generators has a direct impact to our future work for g ≥ 3. For example, while no additional geometric input is needed to apply our method to genus 3 to prove the conjectural formula of Klemm-Katz-Vafa [29], and we could proceed with the methods developed in this paper by brutal force, the resulting proof would not be very illuminating. Recall that there is a conjectural formula up to genus 51 by A. Klemm and his collaborators [27]. Our eventual goal is to reach genus 51 and go beyond. To do so, we have to understand better the cancellation of terms involving extra generators. We will leave this to a future research [23]. The paper is organized as follows. In Section 2, we will summarize the relevant compacti- fied moduli space and its localization formula from [5]. The detailed analysis of contributions of the localization graphs and their closed formulae in terms of generators are stated in Sec- tion 3. The main theorem directly follows from these closed formulae. In Section 4, we derive important results about the twisted theory, which we then apply in Section 5 to yield a proof of the closed formulae. Finally, we would like to mention several independent ap- proaches to higher genus problem by Maulik and Pandharipande [35], Chang–Li–Li–Liu [4] and Guo–Ross [24, 25]. The last two authors would like to thank Qile Chen for the collaboration which provides the geometric input to the current work. The first author was partially supported by the NSFC grants 11431001 and 11501013. The second author was partially supported by a Simons Travel Grant. The third author was partially supported by NSF grant DMS 1405245 and NSF FRG grant DMS 1159265. 2. A localization formula The main geometric input is a formula computing GW-invariants of a quintic 3-fold in terms of a twisted theory and a certain "effective" theory. This formula is obtained by localization on a compactified moduli space of stable maps with a p-field. The proof of the formula and details about the moduli space can be found in [5] (see also [6]). In Section 2.1, we give an overview of the definition of the relevant moduli spaces, and in Section 2.2, we explain the localization formula in the general case. We then, in Sec- tion 2.3, specialize to genus two. Finally, in Section 2.4, we illustrate the formula by directly computing the genus two, degree one invariant. 2.1. Moduli space. Let Mg,0(Q5, d) denote the moduli space of genus-g, unpointed, degree- d stable maps f : C → Q5 to the quintic threefold Q5 ⊂ P4. This moduli space admits a perfect obstruction theory and hence a virtual class [Mg,0(Q5, d)]vir of virtual dimension zero, whose degree is defined to be the Gromov–Witten invariant Ng,d. For the computations in this paper, it will be much more convenient to work with the moduli space Qg,0(Q5, d) of genus-g, unpointed, degree-d stable quotients (or quasimaps) to the quintic threefold Q5 ⊂ P4 instead. We refer to [34, 9] for the definition of this moduli space, mentioning here just that it parameterizes prestable curves C together with a line bundle L and sections s ∈ H 0(L⊕5) satisfying the equation of Q5 and a stability condition. The moduli space Qg,0(Q5, d) also admits a perfect obstruction theory and virtual class [Qg,0(Q5, d)]vir, which after integration defines the stable quotient invariant N SQ g,d . By the wall-crossing formula [8, 11], the information of the N SQ g,d is equivalent to the information of Xd=0 qdN SQ g,d Xd=0 6 S. GUO, F. JANDA, AND Y. RUAN the Ng,d. In genus g ≥ 2, the wall-crossing formula says that, if ∞ ∞ F GW g (q) := qdNg,d, F SQ g (q) := are the generating series of stable map and stable quotient invariants, then F GW g (Q) = I 2g−2 0 F SQ g (q). It is difficult to directly access the stable quotient (or stable map) theory of Q5, in part because Q5 is a "non-linear" object. By results of Chang–Li [3]1, we can instead work on the more linear moduli space Qg,0(P4, d)p of stable quotients with a p-field, that is the cone π∗(ωπ ⊗ L−⊗5) over Qg,0(P4, d). Here, π denotes the universal curve and L denotes the universal line bundle L. The moduli space Qg,0(P4, d)p also has a perfect obstruction theory and hence a virtual class. However, because Qg,0(P4, d)p is in general not compact, this virtual class cannot directly be used to define invariants. To circumvent this problem, Chang–Li introduce a cosection σ of the obstruction sheaf, and show that N SQ g,d =Z[Qg,0(P4,d)p]vir σ (−1)1−g+5d, where [Qg,0(P4, d)p]vir on the compact moduli space Qg,0(Q5, d). σ is the cosection localized virtual class of Qg,0(P4, d)p, which is supported The main new idea of [5] is to define a modular compactification Xg,d of Qg,0(P4, d)p with a perfect obstruction theory extending the one of Qg,0(P4, d)p such that the cosection σ extends without acquiring additional degeneracy loci. It is then easy to see that Z[Qg,0(P4,d)p]vir σ (−1)1−g+5d =Z[Xg,d]vir (−1)1−g+5d. If Xg,d furthermore admits a non-trivial torus action (and the perfect obstruction theory is equivariant), we can apply virtual localization to the right hand side to express it in terms of hopefully simpler fixed loci. A suitable compactification Xg,d is given by a space of stable quotients (C, L, s) together with a log-section η : C → P := P(ωC ⊗ L−⊗5 ⊕ OC) where the target is equipped with the divisorial log-structure corresponding to the infinity section. The open locus where the log-section η does not touch the infinity section is clearly isomorphic to Qg,0(P4, d)p, and the canonical perfect obstruction theory of Xg,n,d extends the one of Qg,0(P4, d)p. Unfortunately, the cosection σ becomes singular on the complement of Qg,0(P4, d)p, Still, it can be extended to a homomorphism to a non-trivial line bundle L∨ N , which comes with a canonical section O → L∨ N , and we can define a new (reduced ) perfect obstruction theory by "removing" (taking the cone) under the induced map from the obstruction theory to the complex [O → L∨ N ]. With this reduced perfect obstruction theory, Xg,d satisfies all desired properties. Remark 2.1. The construction of the modular compactification and its reduced perfect obstruction theory can also be carried out in the setting of stable maps, and it satisfies all of the analogous properties. 1To be precise, we use the result of Chang–Li to rewrite Gromov–Witten invariants of the quintic in terms of stable maps with p-fields. After that, we apply the wall-crossing [12] to move to stable quotients with a p-field. GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 7 (7, 2) (1, 0) (0, 0) (g1, d1) (g2, d2) ∞ 0 Figure 1. A localization graph in genus g1 + g2 + 9 and degree d1 + d2 + 2. The bottom vertices lie in V0, the top vertices lie in V∞. The pair (g(v), d(v)) is specified at each vertex. An edge e is thick if δ(e) = 2. Otherwise, δ(e) = 1. 2.2. Localization. We now consider virtual localization [22] of the reduced perfect ob- struction theory of Xg,d with respect to the C∗-action of Xg,d that scales η. Let t be the corresponding equivariant parameter. The fixed loci of the C∗-action are indexed by bivalent graphs Γ with n legs. We let V (Γ), E(Γ) be the corresponding sets of vertices and edges. The vertices v ∈ V (Γ) correspond to components (or isolated points) of the curve C sent to either the zero or infinity section. We define the bivalent structure V (Γ) = V0(Γ) ⊔ V∞(Γ) accordingly. Furthermore, let g(v) (respectively, d(v)) be the genus of such a component (respectively, the degree of L on this component). A vertex v is unstable if and only if it corresponds to an isolated point of C, that is when g(v) = 0 and either n(v) = 1 or n(v) = 2 and d(v) = 0. The edges e ∈ E(Γ) correspond to the remaining components of C, which are rational, each contracted to a point in P4, and mapped via a degree-δ(e) Galois cover to the corresponding component of P (with a possible base point at the zero section). The fixed locus corresponding to a dual graph Γ is (up to a finite map) isomorphic to MΓ := Yv∈V0(Γ) Qg(v),n(v)(P4, d(v)) ×(P4)E(Γ) Yv∈V∞(Γ) Qg(v),n(v)(P, d(v), µ(v))∼, where unstable moduli spaces Qg(v),n(v)(P4, d(v)) are defined to be a copy of P4, and the rub- ber moduli space Qg(v),n(v)(P, d(v), µ(v))∼ generically parameterizes stable quotients (C, L, s) together with a nonzero holomorphic section of ω ⊗ L−⊗5 up to scaling with zeros prescribed by µ(v). Here, µ(v) is the n(v)-tuple of integers consisting of 0 for each leg, and δ(e) − 1 for each edge e at v. In analogy with [39], we will also refer to the rubber moduli space as the "effective" moduli space. In order for a fixed locus corresponding to a graph to be non-empty, there are many constraints on the decorated dual graph. First, we must have g(v) + h1(Γ), g = Xv∈V (Γ) d(v). d = Xv∈V (Γ) Second, there is a stability condition which says that for any vertex v ∈ V0(Γ) of genus zero and valence one, the corresponding unique edge e needs to satisfy δ(e) > 1 + 5d(v). Third, for every v ∈ V∞(Γ), the partition µ(v) must have size 2g(v) − 2 − 5d(v) ≥ 0. Note that the third condition implies that g(v) ≥ 1 for each v ∈ V∞(Γ). A localization graph satisfying all these conditions is depicted in Figure 1. 8 S. GUO, F. JANDA, AND Y. RUAN The contribution of a decorated graph Γ can then be written as (1) 1 AutZMΓ × Yv∈V∞(Γ) ∆!  Yv∈V0(Γ) t −t + 5H − ψ0 with the notation: e(−Rπv,∗(ωπv ⊗ L−⊗5) ⊗ [1]) e(H) e(H))( t−5ev∗ δ(e) − ψe) Qe at v(t − 5ev∗ ∩ [Qg(v),n(v)(P, d(v), µ(v))∼]red × Ye∈E(Γ) ∩ [Qg(v),n(v)(P4, d(v))]vir 1 i=1 δ(e)Qδ(e)−1 e (H) t−5ev∗ δ(e)   , e(H): pullback of H via any of the two evaluation maps corresponding to e • ∆ : (P4)E(Γ) → (P4)E(Γ) × (P4)E(Γ): diagonal map • πv: universal curve corresponding to v ∈ V0(Γ) • ev∗ • [1]: line bundle with Chern class t • 5H − ψ0: a universal divisor class on the effective moduli space • [Qg(v),n(v)(P, d(v), µ(v))∼]red: a reduced virtual class, which is discussed below In the unstable case that v ∈ V0(Γ) has valence 2 and is connected to two edges e1 and e2, we define (2) to be Qe at v(t − 5ev∗ e(−Rπv,∗(ωπv ⊗ L−⊗5) ⊗ [1]) e(H))( t−5ev∗ e (H) δ(e) − ψe) ∩ [Qg(v),n(v)(P4, d(v))]vir (t − 5H)(cid:16) t−5H 1 δ(e1) + t−5H δ(e2)(cid:17) , and when v has genus zero and is connected to a single edge e, we define (2) as (3) (t − 5H)5d(v)+1(5d(v) + 1)! δ(e)5d(v)+1 . Most parts of (1) are effectively computable, such as the integrals over the moduli spaces for v ∈ V0(Γ), which are twisted [13] invariants of P4. The most difficult part of the formula is related to the effective moduli spaces. Fortunately, these integrals are highly constrained. The reduced virtual class has dimension [Qg,n(P, d, µ)∼]red n − (2g − 2 − 5d), which is one more than the naive virtual dimension. Together with pullback properties of the reduced virtual class, this implies that the virtual class vanishes unless all parts of µ are 0 or 1. Furthermore, there is a dilaton equation which implies that Z[Qg,n(P,d,µ)∼]red ψn−2g+2 0 = (2g − 2 + n)! (4g − 4 − 5d)!Z[Qg,2g−2(P,d,(1,...,1))∼]red 1 as long as g ≥ 2. When g = 1, we need to have d = 0, and Q1,n(P, 0, (0, . . . , 0))∼ ∼= Q1,n × P4. The virtual class is also explicit, and given by [Q1,n(P, 0, (0, . . . , 0))∼]red = e((TP4 + O − OP4(5)) ⊗ E∨) = 205H 4 + 40H 3λ1, GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 9 (1, 0) (1, 0) (1, 0) (1, 0) (2, 0) (2, d) (1, d) Γ2 Γ1 (0, d) Γ0 b (0, d) (0, 0) (0, 0) Γ0 a Figure 2. The localization graphs in genus two. ∞ 0 where E denotes the Hodge (line) bundle, and λ1 its first Chern class. There are also explicit formulae for 5H − ψ0 in this case, for instance, when n = 1, we have 5H − ψ0 = 5H − λ1 = 5H − ψ1, and, when n = 2, we have 5H − ψ0 = 5H − ψ1 = 5H − ψ2. All in all, (1) gives an explicit computation of any Gromov–Witten invariant of the quintic, up to the determination of the constants cg,d :=Z[Qg,2g−2(P,d,(1,...,1))]red 1 ∈ Q, which are defined for every g ≥ 2, d ≥ 0 such that d ≤ 2g−2 . We call these constants "effective invariants". Note that for any particular genus, only finitely many of these invariants are needed. 5 Remark 2.2. All of the discussion of this section can also be carried out in the stable maps setting, with only the following essential differences: Vertices v ∈ V0(Γ) with (g(v), n(v)) = (0, 1) are stable unless d(v) = 0. Therefore, for such vertices v, the corresponding edge e only needs to have δ(e) > 1. Accordingly, we also need to replace 5d(v) + 1 by 1 in (3). Because of this, there are many more localization graphs in the stable maps setting than in the stable quotient setting. In fact, while in any case, there are only finitely many types of localization graphs for any fixed g and d, in the stable quotient setting, for fixed g, the number of localization graphs does not depend on d (as long as d is large enough). This is the main technical advantage of the stable quotient theory for our purpose. 2.3. Formula in genus two. We now specialize the localization formula to genus two, and apply it to the computation of the generating series F SQ 2 (q) = ∞ Xd=0 (−1)1−g+5dqdZ[X2,d]vir 1. There are 5 localization graphs, which are shown in Figure 2. Note that the fifth graph can only occur when d = 0, and that its contribution is given by the constant −c2,0. We label the first four graphs by Γ2, Γ1, Γ0 a, respectively. b and Γ0 10 S. GUO, F. JANDA, AND Y. RUAN We introduce the bracket notation hα1, . . . , αnit,SQ g,n = ∞ Xd=0 e(Rπ∗L⊗5 ⊗ [1]) ev∗ i (αi) n Yi=1 qdZ[Qg,n(P4,d)]vir Xd=0 ∞ = (−1)1−g+5dZ[Qg,n(P4,d)]vir e(−Rπ∗(ω ⊗ L⊗−5) ⊗ [1]) ev∗ i (αi). n Yi=1 The contribution of Γ2 is then simply given by For the contribution of Γ1, we first need the computation ContΓ2 := hit,SQ 2,0 . ZM1,1×P4 t −t + 5H − ψ0 (205H 4 + 40H 3λ1) = − 5 3 H 3 + 5 24 H 4t−1. Therefore, and we can also directly compute: 3 H 3 + 5 (t − 5H)(t − 5H − ψ)(cid:29)t,SQ 24H 4t−1 1,1 , ContΓ1 := −(cid:28) − 5 :=(cid:28) − 5 3H 3 + 5 24 H 4t−1 (t − 5H)(t − 5H − ψ1) ContΓ0 b , − 5 24H 4t−1 3 H 3 + 5 (t − 5H)(t − 5H − ψ2)(cid:29) t,SQ 0,2 Finally, for the contribution of Γ0 a, we need to know ZM1,2×P4 t −t + 5H − ψ0 (205H 4 + 40H 3λ1) = 5 3 H 3t−1 + 65 8 H 4t−2. Thus, ContΓ0 a :=(cid:28) 5 3(H 3 ⊗ H 4 + H 4 ⊗ H 3)t−1 + 65 (t − 5H)(t − 5H − ψ1) (t − 5H)(t − 5H − ψ2)(cid:29)t,SQ 8 H 4 ⊗ H 4t−2 0,2 . Summing all contributions gives (4) F SQ 2 (q) = ContΓ2 + ContΓ1 + 1 2 ContΓ0 a + 1 2 ContΓ0 b − c2,0. 2.4. Example. To illustrate our genus two formula, we compute explicitly the degree one invariant, that is the coefficient of q1 in F SQ defined by (4), and match it with the value given by Conjecture 1.1, that is the q1-coefficient of 2 F SQ 2 (q) = (I0(q))−2F GW 2 (Q) = − 5 144 + 325 16 q + 366875 24 q2 + 1030721125 48 q3 + O(q4). Note that the value of the constant c2,0 is irrelevant for this computation. To compute the q1-coefficients of the other terms on the right hand side of (4), we use localization for an additional diagonal (C∗)5-action on the base P4 with corresponding localization parameters λi. We refer to [34, Section 7] and [22] for the enumeration of fixed points and identification of localization contributions that we use below. It will be convenient for the computation to assume that λi = ξiλ where ξ is a primitive fifth root of unity. We index the fixed points of P4 by i ∈ {0, . . . , 4}. Note that the Euler class of the Poincar´e dual of fixed point i is given by Qj6=i(λi − λj). GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 11 2.4.1. Genus two contribution. We begin with the computation of ContΓ2 via localization. First, note that there are two types of fixed loci for localization on Q2,0(P4, 1). The first type of fixed locus is where, except for an order one base point, the entire source curve is contracted to a fixed point i. Such a fixed locus is isomorphic to M2,1. The contribution of the virtual class of this fixed locus is (5) Qj6=i((λi − λj)2 − λ1(λi − λj) + λ2) Qj6=i(λi − λj)(λi − λj − ψ1) , where, by abuse of notation, λ1 and λ2 denote the Chern classes of the Hodge bundle. The contribution of the twisting by e(Rπ∗L⊗5 ⊗ [1]) is (6) (t + 5λi − 5ψ1) · · · (t + 5λi − ψ1)(t + 5λi) (t + 5λi)2 − λ1(t + 5λi) + λ2 . Summing the product of (5) and (6) over all five fixed loci, taking the coefficient of t0 and integrating, gives the total resulting contribution of these 5 fixed loci ZM2,1 1370ψ4 1 − 3075ψ2 1λ2 1 + 2100ψ1λ3 1 + 75ψ2 1λ2 + 2925ψ1λ1λ2 = 1370 · 1 1152 − 3075 · 7 2880 + 2100 · 1 1440 + 75 · 7 5760 + 2925 · 1 2880 = − 4285 1152 , where we used a few well-known intersection numbers on M2,1. The second type of fixed locus corresponds to the locus of two genus one components contracted to fixed points i 6= j, and which are connected by a degree one cover of the torus fixed curve connecting i and j. This fixed locus is isomorphic to M1,1 × M1,1 (up to a Z/2Z-automorphism group that we will address later). The contribution of the virtual class to this fixed locus is Yk6=i λi − λk − λ1a λi − λk Yk6=j λj − λk − λ1b 1 λj − λk (λi − λj − ψ1a)(λj − λi − ψ1b) , where ψ1a, ψ1b, λ1a and λ1b denote the cotangent and Hodge classes on each M1,1-factor. The contribution of the twisting is (t + 5λi) · · · (t + 5λj) (t + 5λi − λ1a)(t + 5λj − λ1b) . Summing up the contribution from all of the 20 fixed loci gives 2965 288 . 2.4.2. Genus one contribution. We now compute ContΓ1. There are also two types of fixed loci for Q1,1(P4, 1). The first type of fixed locus is where, except for an order one base point, the entire source curve is contracted to a fixed point i. Such a fixed locus is isomorphic to M1,2. The contribution of the virtual class of this fixed locus is λi − λj − λ1 (λi − λj)(λi − λj − ψ2) , Yj6=i 12 S. GUO, F. JANDA, AND Y. RUAN and the contribution of the twisting is (t + 5λi − 5ψ2) · · · (t + 5λi − ψ2)(t + 5λi) t + 5λi − λ1 . In addition, we need to consider the insertion. For this, note that ev∗ descendent ψ-class is ψ1. Thus, the insertion gives a factor of 1(H) = λi, and that the − 5 3λ3 i + 5 24λ4 i t−1 Summing over i gives the contribution of (t − 5λi)(t − 5λi − ψ1) 975 64 . . The second type of fixed locus is where there is a genus one curve contracted to a fixed point i which is connected via a node to a rational component mapping isomorphically to the fixed line connecting fixed point i to another fixed point j such that the preimage of fixed point j is the marking. This locus is isomorphic to M1,1. The contribution of the virtual class is and the contribution of the twisting is Qk6=i(λi − λk − λ1) Qk6=i(λi − λk)Qk6=j(λj − λk) 1 λi − λj − ψ1 , (t + 5λi)(t + 4λi + λj) · · · (t + 5λj) . t + 5λi − λ1 Now, ev∗ factor of 1(H) = λj, and the descendent ψ-class is given by λi − λj. So, the insertion gives a − 5 3λ3 j + 5 24λ4 j t−1 . Summing over all i 6= j gives (t − λj)(t − 4λj − λi) − 3425 288 . 2.4.3. Genus zero contributions. We finally compute ContΓ0 combine the two insertions: a + ContΓ0 b . Note that we can H 3 + 5 3 (cid:18)− 5 24 H 4t−1(cid:19)⊗2 (H 3 ⊗ H 4 + H 4 ⊗ H 3)t−1 + 65 8 H 4 ⊗ H 4t−2 = H 3 ⊗ H 3 + 95 72 (H 3 ⊗ H 4 + H 4 ⊗ H 3)t−1 + 4705 576 H 4 ⊗ H 4t−2 + 5 3 25 9 There are also two types of fixed loci for Q0,2(P4, 1). The first type of fixed locus is where except for an order one base point the entire source curve is contracted to a fixed point i. Such a fixed locus is isomorphic to a point. The contribution of the virtual class for this fixed locus is given by and the contribution of the twisting is 1 Qj6=i(λi − λj)2 , (t + 5λi)6. GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 13 On this fixed locus, the descendent classes vanish, and we have ev∗ Thus, the insertion gives a factor of 9 λ6 i t−1 + 4705 i + 95 576 λ8 i t−2 25 36λ7 (t − 5λi)4 1(H) = ev∗ 2(H) = λi. In total, the contribution of the fixed loci of first type is 1967 192 . A fixed locus of second type is where the map is a degree one cover of a fixed line of P4 such that marking 1 is mapped to fixed point i and marking 2 is mapped to fixed point j. Such a locus is again just a point. The contribution of the virtual class is then given by and the contribution of the twisting is Qk6=i(λi − λk)Qk6=j(λj − λk) 1 , (t + 5λi)(t + 4λi + λj) · · · (t + 5λj). 2(H) = λj, the descendent class at marking one is λj − λi, while the 1(H) = λi, ev∗ We have ev∗ other descendent class is λi − λj. So, the insertion gives a factor of 9 λ3 j t−2 (t − 5λi)(t − 5λj)(t − 4λi − λj)(t − λi − 4λj) j t−1 + 4705 j t−1 + 95 j + 95 576 λ4 72 λ3 72 λ4 i λ3 i λ3 i λ4 i λ4 25 Thus, the total contribution is 3001 144 . 2.4.4. Final result. Collecting all the contributions gives 3425 288 which is the expected coefficient of q1 in F SQ 2965 288 4285 1152 975 64 1 2 − − + + (q). 2 + 1 2(cid:18)1967 192 + 3001 144 (cid:19) = 325 16 , 3. Proof of the Main Theorem In this section, we provide a list of closed formulae for the contribution of each graph in Section 2.3. Based on these closed formulae, we prove the main theorem. One subtlety are the extra generators appearing in the twisted theory. They mysteriously cancel each other when we sum up the contributions. We will come back to these cancellations in higher genus in [23]. The proof of these formulae will be presented in Section 5. 3.1. Genus zero mirror theorem for the twisted theory of P4. Let I(t, q, z) be the I-function of the twisted invariants, that is explicitly (7) I(t, q, z) = z Xd≥0 qdQ5d Qd j=1(5H + jz − t) , k=1(H + kz)5 which satisfies the following Picard–Fuchs equation (8) H − q (cid:16)D5 5 (5DH + kz − t)(cid:17)I(t, q, z) = 0, Yk=1 14 S. GUO, F. JANDA, AND Y. RUAN where DH := D + H := zq d z−1: dq + H. The I-function has the following form when expanded in I(t, q, z) = zI0(q)1 + I1(t, q) + z−1I2(t, q) + z−2I3(t, q) + · · · The genus zero mirror theorem [18] relates I(t, q, z) to the J-function, defined by J(t, z) := −1z + t(−z) +Xi ϕi(cid:28)(cid:28) ϕi z − ψ(cid:29)(cid:29)t 0,1 (t(ψ)), where we define the double bracket for the twisted Gromov–Witten invariants by hhγ1(ψ), · · · , γm(ψ)iit g,m (t(ψ)) = Xn = Xn,d 1 n! hγ1(ψ), · · · , γm(ψ), t(ψ), · · · , t(ψ)it g,m+n qd n!Z[Mg,m+n(P4,d)]vir e(R∗π∗f ∗O(5) ⊗ [1]) Yj=1 m m+n γj(ψj) t(ψk), Yk=m+1 where we recall that [1] is a line bundle with first Chern class t, and where {ϕi} is any basis of H ∗(P4) with dual basis {ϕi} under the inner product To state the precise relationship, we write (a, b)t :=ZP4 a ∪ b ∪ (5H − t). and define the mirror map by I1(t, q) = I1(q)H + I1;a(q)t, τ (q) = I1(t, q) I0(q) = H I1(q) I0(q) + t I1;a(q) I0(q) . Then Givental's mirror theorem states that the J-function of the twisted invariants can be computed from the I-function by J(τ (q), z) = I(t, q, z) I0(q) . 3.2. Quantum product and "extra" generators. We consider the quantum product in the twisted theory, which is defined for any point t ∈ H ∗ C∗(P4) by a ∗t b :=Xi ϕi hha, b, ϕiiit ϕi ha, b, ϕiit 0,3 . 0,3 (t) =Xi In the basis {H k}, the quantum product τ ∗τ , in which can be identified with a 5 × 5-matrix. It is not hard to see that it has the following form τ = H + q d dq τ, τ ∗τ = A := I1,1;at I2,2;bt2 I3,3;ct3 I4,4;dt4 I5,5;et5   I1,1 I2,2;at I3,3;bt2 I4,4;ct3 I5,5;dt4 I2,2 I3,3;at I4,4;bt2 I5,5;ct3 I3,3 I4,4;at I5,5;bt2 I4,4 I5,5;at , t   GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 15 where the Ii,i and Ii,i;∗ are certain power series in q. Recall that by the basic theory of Frobenius manifolds, the S-matrix S(t, z), which is defined by S(t, z)ϕi = ϕi +Xj ϕj (cid:28)(cid:28)ϕj, ϕi z − ψ(cid:29)(cid:29)t (t), is a solution of the quantum differential equation dS(t, z) = dt ∗t S(t, z), C∗(P4) (but not on the Novikov variable where the differential d acts on the coordinates t ∈ H ∗ q). By using the divisor equation and the matrix introduced above, we can write down the explicit form of the quantum differential equation at t = τ (q): DHS(τ (q), z) = A(t, q) · S(τ (q), z) Since the S-matrix can be obtained from the derivatives of the I-function by Birkhoff fac- torization (see e.g. [13] and see also Proposition 4.1 below), we see that all the entries in this matrix A can be written explicitly in terms of the derivatives of Ik. In particular, we have I1,1 = 1 + q d dq(cid:18)I1 I0(cid:19) , I1,1;a = q d dq(cid:18)I1,a I0 (cid:19) . Remark 3.1. One can check that our Ip,p coincide with the Ip in Theorem 1 of Zagier– Zinger's paper [43]. Recall that we have introduced the following degree k "basic" generators Xk := dk duk (cid:18)log I0 L(cid:19) , Yk := dk duk (cid:18)log I0I1,1 L2 (cid:19) , Zk := dk duk (cid:16)log(q 1 5 L)(cid:17) , By using the entries in the quantum product matrix, we define the following four "extra" generators Q = 1 L(cid:16)I1,1;a − 1 5(cid:17), P = Q + (X − Y)Q, P := I1,1I2,2;b L4 2 Q, L2 + Q2 + d du P + (X + Y)P. d du Their degrees are defined by Q := so that we have deg Q := 1, deg P := 2, deg d du := 1, deg Q = 2, deg P = 3. Remark 3.2. We will see that, in the genus 2 case, only linear terms of the following two extra generators are involved: P and Q . Remark 3.3. For any k, the generator Zk can be written as a (non-homogeneous) polyno- mial of L. This is because Z1 is a polynomial of L and d du L = 1 5 (L5 − 1). 16 S. GUO, F. JANDA, AND Y. RUAN For example, the first several of them are d dq Z1 = (q (log L) + 1 5 ) = L4 5 , 1 L 4 25 4 125 Z2 = Z3 = L3(L5 − 1), L2(L5 − 1)(8L5 − 3). Remark 3.4. Some leading terms of the basic generators are given by X = − 505q − 1425100q2 − 4155623250q3 + O(q4) Y = − 360q − 1190450q2 − 3759611500q3 + O(q4) X2 = − 505q − 2534575q2 − 10290963500q3 + O(q4) X3 = − 505q − 4753525q2 − 27310140500q3 + O(q4) L = 1 + 625q + 1171875q2 + 2685546875q3 + O(q4), and some leading terms of the extra generators are given by Q = − P = − 1 5 3 50 − 149q − 271030q2 − 591997100q3 + O(q4) − 399 10 q − 12732q2 + 131454705q3 + O(q4) Q = − 120q − 473525q2 − 1622526750q3 + O(q4) P = 12q + 331965 2 q2 + 984651825q3 + O(q4). In particular, from this data one can see the degree zero genus two Gromov–Witten invariant should be equal to the coefficient of the Z 3 in the homogenous polynomial 53 L2 F2 (see I0 Conjecture 1.1), which is −5 144 . 3.3. A list of closed formulae for the contribution of localization graphs. We now give a list of closed formulae for the contribution of each of the localization graphs in Sec- tion 2.3. We also rewrite each contribution in terms of Gromov–Witten double brackets using the wall-crossing formula2: The first proposition concerns the purely twisted contribution ContΓ2 = L−2Cont′ Γ2, where hit,SQ g,n = I 2g−2 0 hhiit g,n(τ (q)) Cont′ Γ2 := (L/I0)2 · hhiit 2,0 (τ (q)). Proposition 3.5. The contribution of Γ2 is a degree 3 homogeneous polynomial in the basic and extra generators, to be precise 1 48 31 X 576 Γ2 = − 205 Z1 Cont′ Y + + 5 P −(cid:0) 1152 19 X3 + 67 X 3 − − − 1152 19 Z1YX 64 − 3Z1Y 2 16 − 48 65 Z2X 1536 2304 (cid:1) · Q − Y(X2 + 5 YX ) 25 X2(X + Z1) 288 2X − 715 Z1 1152 − 2Y 45 Z1 128 − 1728 − 101 YX 2 1152 − Y 3 24 829 Z1Z2 107 Z1X 2 − + 384 349 Z3 13824 2While no proof exactly applies to our situation, the proofs [10], [8] and [12] can be easily adapted. GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 17 The leading terms of the genus two graph contribution are ContΓ2 = L−2Cont′ Γ2 = 1645 q 1152 + 1842665 q2 576 + 2419134175 q3 288 + O(q4). Note that the degree 1 term coincides with the one in the localization computation in Sec- tion 2.4: 1645 1152 = − 4285 1152 + 1 2 · 2965 288 . We next consider the contribution from the graph Γ1, which is a genus-one twisted theory with a special insertion. It can be rewritten as ContΓ1 = L−2Cont′ Γ1, where Cont′ Γ1 := L2 3H 3 + 5 I0 (cid:28)(cid:28) − 5 (t − 5H)(t − 5H − ψ)(cid:29)(cid:29)t 24 H 4t−1 1,1 (τ (q)). Proposition 3.6. The contribution of Γ1 is a degree 3 homogeneous polynomial in the basic and extra generators, to be precise −Cont′ Γ1 = 473 576 25 (X3 + X 3) P +(cid:0) 1 48 Y − + 72 X + 2093 1152 25 96 41 Y(X2 − YX − YZ1 + Z2) Z1(cid:1) · Q 48 + 1871 X X2 − 1271 YX 2 − 1471 Z1X 2 576 + + 1025 Z1X2 288 − 451 Z1YX 72 + 1267 Z2X 1152 2X + 1039 Z1 576 − 779 Z1 192 2Y + 4945 Z1Z2 864 − 155 Z3 3456 The leading terms of the genus one graph contribution are −ContΓ1 = −L−2Cont′ Γ1 = 1925 q 576 − 2344025 q2 288 − 4831529575 q3 144 + O(q4). Note that the degree 1 term coincides with the one in the localization computation in Sec- tion 2.4: 1925 576 = 975 64 − 3425 288 The remaining two (non-trivial) graphs involve a genus-zero two-pointed twisted theory. We rewrite them as ContΓ0 a = L−2Cont′ Γ0 a and ContΓ0 b = L−2Cont′ Γ0 b , where Cont′ Γ0 a Cont′ Γ0 b 3H 3 ⊗ H 4t−1 + 5 3H 4 ⊗ H 3t−1 + 65 :=L2(cid:28)(cid:28) 5 :=L2(cid:28)(cid:28) − 5 (t − 5H)(t − 5H − ψ1) (t − 5H)(t − 5H − ψ2)(cid:29)(cid:29)t (t − 5H)(t − 5H − ψ2)(cid:29)(cid:29)t (t − 5H)(t − 5H − ψ1) 8 H 4 ⊗ H 4t−2 3 H 3 + 5 3H 3 + 5 24H 4t−1 24H 4t−1 − 5 , 0,2 0,2 (τ (q)), (τ (q)). 18 S. GUO, F. JANDA, AND Y. RUAN Proposition 3.7. The contributions of Γ0 in the basic and extra generators, to be precise a and Γ0 b are both degree 3 homogeneous polynomials 1 12 X + 199 48 (X 3 + X 2Y + Z1 Z1(cid:1) · Q 2Y) + 77 Z1X 2 24 − 277 Z1 24 2X − 599 Z1Z2 72 + + 41 Z1YX 12 805 Z3 288 , 205X 288 + (X3 + X 3) + 265Z1 384 (cid:1) · Q 15595 X X2 288 − 8405 Y 576 (X 2 + 2X Z1 + Z 2 1 ) Cont′ Γ0 a = − + + Cont′ Γ0 b = − + P −(cid:0) 13 8 41 24 677 Z2X 96 P +(cid:0) 5 576 7595 576 250X2 + 9 +(cid:0) 215X 2 576 (cid:1)Z1 + 117215 Z2X 2304 − 2405 Z1 192 2X + 1585 Z1Z2 64 + 30325 Z3 2304 . Corollary 3.8. The summation Cont′ Γ0 of the two contributions is Cont′ Γ0 = − + − 941 576 8579 (X3 + X 3) P +(cid:0) 181 288 X − 1327 384 16579 X X2 Z1(cid:1) · Q 576 2X 4621 Z1 192 + + 288 133463 Z2X 2304 − 7421 YX (X + 2Z1) 576 + 250 Z1X2 9 − 7421 Z1 576 2Y + 4085 Z3 256 + 9473 Z1Z2 576 2063 Z1X 2 576 + . The leading terms of the genus zero graph contribution are ContΓ0 a + ContΓ0 b = L−2Cont′ Γ0 = 17905 576 q + 11650385 288 q2 + 13428251725 144 q3 + O(q4). Note that the degree 1 term coincides with the one in the localization computation in Sec- tion 2.4: 17905 576 = 1967 192 + 3001 144 3.4. Proof of the Main Theorem. Our main result now follows from the propositions in the last subsection. Theorem 3.9. The genus two Gromov–Witten free energy of a quintic Calabi–Yau threefold is given by F GW 2 (Q) = I 2 0 L2 ·(cid:16) 70 X3 625 ZX2 9 575 X X2 18 175 ZYX + − 36 41 Z2Y 48 9 625 Z 3 144 − + + + + 5YX2 6 + 1441 Z2X + 48 2233 ZZ2 72 − 128 + 547 Z3 72 (cid:17). 557 X 3 − 629 YX 2 72 − 23 Y 2X 24 − Y 3 24 25 Z(X 2 + Y 2) 24 − 3125 Z 2(X + Y) 288 Proof. By dimension considerations, the Cont′ value of c2,0 = −N2,0 can be read off from (see [37]) Γi have no constant term in q. Therefore, the N2,0 = 1 2ZQ5 (c3(Q5) − c1(Q5)c2(Q5)) ·ZM2,0 λ3 1 = 1 2 · (−200) · B4 4 · B2 2 · 1 2! = − 5 144 . GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 19 The rest is just a direct consequence of the following formula L2 · F SQ 2 (q) = −c2,0L2 + 1 2 the wall-crossing formula Cont′ Γ0 − Cont′ Γ1 + Cont′ Γ2, F GW 2 (Q) = I0(q)2 · F SQ 2 (q), and the formulae for Conti in Proposition 3.5, Proposition 3.6 and Corollary 3.8. (cid:3) 3.5. Equivalence between our Main Theorem and the physicists' conjecture. The closed formula for the genus two Gromov–Witten potential was first proposed by BCOV [1] and further clarified by Yamaguchi–Yau [42]. Later, Klemm–Huang–Quackenbush [27] extended the result to genus 51. We will follow the notation of [27]. In [42, 27], the authors introduce the following basic generators Ap := (−q d dq )p(cid:0)qI1,1(q)(cid:1) qI1,1(q) , Bp := (−q d dq )pI0(q) I0(q) , X := −55q 1 − 55q and the change of variables B1 = u, A1 = v1 − 1 − 2u B2 = v2 + uv1, B3 = v3 − uv2 + uv1X − 2 5 uX. The following is the original physical conjecture. Conjecture 3.10. Let Fg be the genus g Gromov–Witten potential and When g = 2, we have the following explicit formula −5 Pg(q) :=(cid:0) I 2 0 (1 − 55q)(cid:1)g−1Fg(Q). P2 = 25 144 − 625 v1 288 167 v1 X − 720 + v1 + 25 v1 24 2X 6 2 − 3 5 v1 24 − − 475 v2 X 12 + 625 v2 25 v1 v2 + 36 41 X 2 3600 − + 6 13 v1 X 2 288 350 v3 9 X 3 240 + − 5759 X 3600 . Proposition 3.11. Both Conjecture 3.10 and Conjecture 1.1 are equivalent to F GW 2 (Q) = (9) I 2 0 L2 ·(cid:16)70 X3 9 125 X2 L4 + + 18 − 36 + 31459 X L8 7200 − Proof. First notice that 575 X X2 557 X 3 5 YX2 629 YX 2 23 Y 2X + + 72 6 35 YX L4 5 X 2L4 − 24 2141 YL8 7200 − 9 29621 L12 12000 + − 72 − 5 Y 2L4 24 − 24 1441 X L3 300 − Y 3 24 41 YL3 300 − − 116369 L7 36000 + 547 L2 750 (cid:17). u = B1, v1 = A1 + 1 + 2B1, v2 = −A1B1 − 2B2 1 − B1 + B2, v3 = −A1B2 1 − B1A1X − 2B3 1 − 2B2 1X − B2 1 + B1B2 − 3 5 B1X + B3. 20 S. GUO, F. JANDA, AND Y. RUAN Therefore, Conjecture 3.10 is equivalent to 385 A1B1 1045 A1B2 1 5923 B1X P2 = − 2B1 36 65 A1 12 13 X 2A1 288 + + − − + + 18 25 A1B2 6 49 B1XA1 36 − 1X + 37 B2 18 475 B2 X 12 115 B2 1 6 − − + 425 B1B2 + + 565 B1 48 3 5 A1 24 X 3 240 − 27 X 2 800 9 + 205 A1 13 X 2B − + 288 2 5 A1 12 − 144 73 XA1 720 + 350 B3 9 − 333 X 200 − 475 B2 36 − 335 288 . 360 2X A1 6 865 B3 1 − 9 Next we define so that Ap :=(cid:16) − q d dq(cid:17)p log(q 1 5 I0), Bp :=(cid:16) − q d dq(cid:17)p log(q 1 5 I1,1), A1 = − B3 = 4 5 + A1, B1 = 1 125 B1 + 3 25 + 1 5 3 ( B2 + B2 5 + B1, B2 = 1 25 + 2 5 B1 + ( B2 + B2 1), 1) + ( B3 + 3 B1 B2 + B3). On the other hand, by definition of Xp, Yp and L, we have A = −L(Y + Z − X ), B1 = −L(X + Z), B3 = −L3(X3 + Z3) + 3 5 L2X(X2 + Z2) − L 25 B2 = L2(X2 + Z2) − L 5 (6X 2 − 5X)(X + Z). X(X + Z), Also X = 1 − L5 and the Zk are all polynomials of L (see Remark 3.3). Finally, a few direct computations show that, after the above change of variables, both conjectures are equivalent to equation (9). (cid:3) 4. Structures of the twisted invariants The twisted theory of P4 is semisimple, and can be computed by the Givental–Teleman formula using R-matrices. In this section, we write down the basic data and relations for the twisted invariants, and then we derive closed formulae for the entries of the R-matrix up to z3, which is all we need for the computation in genus two. 4.1. S-matrix. Recall that the S-matrix is a fundamental solution of the quantum differ- ential equation (10) DHS(τ (q), z) = A(t, q) · S(τ (q), z), where we recall that D := z q d dq and DH := D + H. Moreover, the S-matrix can be obtained from the derivatives of the I-function by Birkhoff factorization. We start from I-function. In the rest of this and next sections, we will fix the base point t = τ (q) = H I1(q) I0(q) + t I1;a(q) I0(q) . As a convention, we will omit the base point t in the double bracket hh−iit S-matrix S(τ (q), z) when t = τ . g,n (t) and the The following proposition is an application of Birkhoff factorization. GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 21 Proposition 4.1. We have the following formulae for the (·, ·)t-adjoint of the S-matrix S∗(z)(1) = I(z) zI0 DH −I3,3;at I1,1 I2,2 −1 S∗(z)(H) = (cid:18)DH − I1,1;at S∗(z)(H 2) = det DH −I2,2;at S∗(z)(H 3) = det    S∗(z)(H 4) = det I3,3 −1 I4,4 −1 DH −I4,4;at (cid:19) I(z) zI0 − I2,2;b t2 I2,2 DH −I1,1;at I1,1 ! I(z) zI0 I(z) zI0   − I3,3;b I3,3 DH −I2,2;at I2,2 −1 − I4,4;b I4,4 DH −I3,3;at t2 − I3,3;c t3 I3,3 − I2,2;b t2 I2,2 DH −I1,1;at I1,1 t2 − I4,4;c I4,4 − I3,3;b I3,3 DH −I2,2;at I3,3 −1 t3 − I4,4;d t4 I4,4 t2 − I3,3;c t3 I3,3 − I2,2;b t2 I2,2 DH −I1,1;at I1,1 I2,2 −1   I(z) zI0 . Note that there is a differential operator DH in the determinants. We define the differential operation from top to bottom. Proof. Noting that the S-matrix is a solution of equation (10), and that S(z)ϕi = ϕi+O(z−1), this proposition follows from a direct computation: S ∗(z)(1) = S ∗(z)(H) = S ∗(z)(H 2) = S ∗(z)(H 3) = S ∗(z)(H 4) = I(z) zI0 1 1 I1,1(cid:16)DH − I1,1;at(cid:17)S ∗(z)1 I2,2(cid:16)DH − I2,2;at(cid:17)S ∗(z)H − I3,3(cid:16)(DH − I3,3;at)S ∗(z)H 2 − I3,3;bt2S ∗(z)H − I3,3;ct3S ∗(z)1 I4,4(cid:16)(DH − I4,4;at)S ∗(z)H 3 − I4,4;bt2S ∗(z)H 2 − I4,4;ct3S ∗(z)H − I4,4;dt4S ∗(z)1(cid:17) I2,2;b I2,2 t2S ∗(z)1 1 1 Now we can write down all the entries of S-matrix. However, it soon becomes too compli- cated to write down all the explicit formulae for further computations. We want to establish an equation satisfied by (cid:3) S∗(z)(cid:16) 1 t + 5 H t2 + 25 H 2 t3 + 125 H 3 t4 + 625 H 4 t5 (cid:17). This equation will help us to simplify some computations involving the S-matrix (for example the computation of the modified V -matrix in Section 5.1). Also by using this equation we can deduce closed formulae for some special entries of the R-matrix and prove some important identities. 22 S. GUO, F. JANDA, AND Y. RUAN Lemma 4.2. Define Then, we have (11) Furthermore, H4 := 1 t + 5 H t2 + 25 H 2 t3 + 125 H 3 t4 + 625 H 4 t5 . I(z) = 0 1 5 1 5 (cid:18)DH − t(cid:19) S∗(z)( H4) + S∗(t − 5H)( H4) = S∗(cid:18)t − 5H 2 (cid:19) ( H4) = H4. Remark 4.3. The H4 in this lemma can be viewed an element of the dual basis, which satisfies ( H4, 1) = ( H4, H) = ( H4, H 2) = ( H4, H 3) = 0. Proof. We define formally S∗ 5 = det DH − I5,5;at −I5,5;bt2 DH −I4,4;at −1 I4,4 −1   −I5,5;ct3 −I5,5;dt4 −I5,5;et5 − I4,4;b − I4,4;d t2 t4 I4,4 I4,4 − I3,3;c DH −I3,3;at t3 I3,3 − I2,2;b t2 I2,2 DH −I1,1;at − I4,4;c t3 I4,4 − I3,3;b t2 I3,3 DH −I2,2;at I3,3 −1 I2,2 −1 I1,1 I(z) zI0   Since H 5 = 0, we have S∗ 5 = S∗(z)(H 5) = 0. On the other hand, by symmetry of the quantum product (see Section 4.4.1 for more details): (1 − I4,4) (1 − I3,3) − 1 5 I4,4;a 1 5 1 25 1 125 1 625 1 I5,5;a = I5,5;b = I5,5;c = I5,5;d = I5,5;e = (1 − I2,2) − (1 − I1,1) − (1 − I0) − 3125 1 25 1 125 1 625 I3,3;a − I2,2;a − I1,1;a − 1 5 I4,4;b 1 25 1 125 I3,3;b − 1 5 I4,4;c I2,2;b − 1 25 I3,3;c − 1 5 I4,4;d. Hence by replacing the first row of the matrix with the first row plus the sum of all the (k + 1)-th row multiplied by tk S∗ 5 = det DH − 1 5 t −1   t 5 DH − t2 25 DH −I4,4;at I4,4 −1 25 DH − t3 t2 125 − I4,4;b t2 I4,4 DH −I3,3;at 5k I5−k,5−k for k = 1, 2, 3, 4, we have 625 DH − t5 t4 − I4,4;d t4 I4,4 − I3,3;c t3 I3,3 − I2,2;b t2 I2,2 DH −I1,1;at t3 125 DH − t4 625 − I4,4;c t3 I4,4 − I3,3;b t2 I3,3 DH −I2,2;at I3,3 −1 I2,2 −1 I1,1 3125 (1 − I0) I(z) zI0   GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 23 Note that, in general, the determinant will change when we perform a row transformation for a matrix containing the operator DH. However in this case, it does not since there are only constant terms −1 under the diagonal. By Proposition 4.1, we obtain 0 = S∗ 5 =(cid:16)DH − 1 5 4 + t(cid:17)(cid:16)S∗ t 5 S∗ 3 + t2 25 S∗ 2 + t3 125 S∗ 1 + t4 625 t5 I(z) 3125 z = 0 S∗ 0(cid:17) + where Sk := S(z)∗(H k) for k = 0, 1, 2, 3, 4. This proves the first statement of the lemma. In particular, letting z = t − 5H, we have DH = H + (t − 5H)q d dq , and the equation becomes (t − 5H)(cid:18)− 1 5 + q d dq(cid:19) S∗(t − 5H)( H4) + We can solve this equation using the initial condition 1 5 = 0. S∗(z)(H k)q=0 = H k. It implies the second statement of the lemma. The third can be proved similarly. (cid:3) 4.2. Ψ-matrix and R-matrix: computations of Ψ1 and R∗1. The twisted theory of P4 is semisimple3, in the sense that there exist idempotents eα with respect to the quantum product (at t = τ (q)): In addition, we recall the definition of the normalized canonical basis eα ∗τ eβ = δαβeα ¯eα := ∆ αeα, 1 2 ∆−1 α := (eα, eα)t. By results of Dubrovin and Givental [14, 19, 21], there exists an asymptotic fundamental solution of the quantum differential equation (10) which has the following form (12) S(t, z) = Ψ−1(t)R(t, z)eU (t)/z, where Ψ−1 is the change of basis from a flat basis to the normalized canonical basis, U is a diagonal matrix with entries the canonical coordinates U = diag(u0, u1, . . . , u4) and R(t, z) = 1 + O(z) is a matrix of formal power series in z. While there is no direct relation between S and the S-matrix defined by two point correlators in Section 3, in the proof of Lemma 4.6 we will discuss a relation between their fully equivariant generalizations. We rewrite the fundamental solution in coordinates as (13) Siα(z) = Ri ¯α(z)euα/z =Xβ Ψi ¯βR ¯β ¯α(z)euα/z. where, viewing S as a linear transformation from H ∗ C∗(P4) with a flat basis, we write Siα(z) := (H i, S(z)¯eα), and where, viewing R as a linear transformation written in the normalized canonical basis, we set C∗(P4) with basis {¯eα} to H ∗ Ψi ¯β := (H i, ¯eβ)t, R ¯α ¯β(z) := (¯eβ, R(z)¯eα)t, Ri ¯β(z) := (H i, R(z)¯eβ)t. 3Semisimplicity will follow from the computations in this section. 24 S. GUO, F. JANDA, AND Y. RUAN Proposition 4.4. Let R(z) = 1 + R1z + R2z2 + · · · , and α := (H i, Rk eα)t. (Rk)i Then, we have where qα = −5ξαq 1 5 and Ψ0 ¯α = 1 + qα I0 · q2 α 2 5 (cid:19)− 3 ·(cid:18) −t (14) (R1)0 α = 12 qα 1 − 1 t · qα , (R2)0 α = 1 − 13 12 qα − 287 2 288 q α t2 · q2 α , (R3)0 α = Proof. By the following Lemma 4.6, the functions (15) Iα(q, z) = euα/zI0R0 ¯α(z) ∀α are solutions of the Picard–Fuchs equation 2 5 − 293 60 qα + 5347 α + 5039 2 1440 q q3 α 3 10368 q α . 5 (cid:16)D5 − q (5D + kz − t)(cid:17)I(q, z) = 0. Yk=1 The proposition then follows from the following asymptotic expansion of Iα (Lemma 4.5 (cid:3) and 4.6). Lemma 4.5. There exist constants Cα, c1α, c2α, . . . such that 1 + qα α (cid:16)1 + q2 5 + (c1α − 24 2 1 + c1αqα qα Iα(z) =Cα · + where uα satisfies 5 )qα + (c2α − c1α − 14 5 )q q3 α q d dq uα = Lα := t 5 qα 1 + qα . · t−1z + 1 + (c1α − 1)qα + c2αq 2 α · t−2z2 q2 α 2 α + c3αq 3 α · t−3z3 + · · ·(cid:17)euα/z Proof. By the arguments in the above proposition, applying the Picard–Fuchs equation to (15), we can see that R0 ¯α(z) satisfies the following equation (16) α − q (cid:16)D5 (5Dα + kz − t)(cid:17)I0R0 ¯α(z) = 0, Yk=1 5 dq uα. After writing down this equation carefully, we can first solve q d where Dα := D + q d dq uα by looking at the coefficient of z0 of (16). After that, we can determine the coefficients of zk in R0 ¯α(z) one by one. The coefficient of z1 in (16) determines the coefficient of z0 in R0 ¯α(z), and hence Ψ0 ¯α up to a constant Cα. At each step, we need to introduce a new undetermined constant ckα. (cid:3) We cannot directly fix the constant terms of the R-matrix because of the poles of the entries of the R-matrix. The idea to solve this problem is to consider a more general equivariant theory first, and then take the limit. We will do so in the following lemma. GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 25 Lemma 4.6. The functions Iα = euα/zI0R0 ¯α(z) are solutions of Picard–Fuchs equation 5 (cid:16)D5 − q (5D + kz − t)(cid:17)I(q, z) = 0. Yk=1 Furthermore, the constants Cα and ciα for i = 1, 2, 3 in Lemma 4.5 are independent of α, and have the following values Cα =(cid:18) −t 5 (cid:19)− 3 2 , c1α = − 1 12 , c2α = − 287 288 , c3α = 5039 10368 . Proof. There is a standard method of fixing the constants of the R-matrix in equivariant Gromov–Witten theory using an explicit formula for the R-matrix when the Novikov pa- rameters are sent to zero. Unfortunately, it does not directly apply to the twisted theory that we are considering since we do not work equivariantly on P4, and the theory hence becomes non-semisimple at q = 0. To solve this problem, we first introduce a more general equivariant theory, and then take a limit to recover the original theory. We introduce the (C∗)5-action which acts diagonally on the base P4, and we denote by λi the corresponding equivariant parameters. To simplify the computation, we set λi = ξiλ. We consider the corresponding twisted theory of P4, which has the I-function I(t, λ, q, z) = z Xd≥0 which satisfies the Picard–Fuchs equation j=1(5H + jz − t) k=1((H + kz)5 − λ5) qd Q5d Qd , 5 H − λ5 − q (cid:16)D5 (5DH + kz − t)(cid:17)I(t, λ, q, z) = 0. Yk=1 Similarly to the previous discussion, we introduce the mirror map τ (q), the S- and R-matrix S(t, λ, z), R(t, λ, z) at point τ (q), the normalized canonical basis, . . . . It is clear that by taking λ → 0 we recover the twisted theory considered in this paper. One main advantage of the more general twisted theory is that it stays semisimple at q = 0 because the classical equivariant cohomology of P4 has a basis of idempotents: (17) and (18) Hence, the normalized canonical basis stays well-defined at q = 0, and we have = Qβ6=α(H − ξβλ) 5ξ4αλ4 eαq=0 = Qβ6=α(H − ξβλ) Qβ6=α(ξαλ − ξβλ) ¯eαq=0 = Qβ6=α(H − ξβλ) p5(5λ5 − tξ4αλ4) p5(5λ5 − tξ4αλ4) −t + 5ξαλ Ψ0 ¯αq=0 = , . By uniqueness of fundamental solutions of the quantum differential equation (see also [21, 13]), we have (19) S(t, λ, z)(Ψ(t, λ)−1q=0)Γ−1(t, z)C −1(λ, z) = Ψ(t, λ)−1R(t, λ, z)eU (t,λ)/z 26 S. GUO, F. JANDA, AND Y. RUAN for constant matrices (with respect to q) given by (20) C(λ, z) = diag(cid:0)(cid:8)ePk>0,j6=i Γ(t, z) = ePk>0 2k(2k−1) B2k B2k 2k(2k−1) z2k−1 (t−5H)2k−1 , z2k−1 (λi−λj )2k−1(cid:9)i=0,1,2,3,4(cid:1), Indeed, these constant matrices together give exactly the constant term of the R-matrix. Notice that the ¯eαq=0 are eigenvectors for the eigenvalues hα of the classical multiplication by H. Hence, by the Picard–Fuchs equation (17) for I(t, λ, q, z) = zI0(q) · S(t, λ, z)∗1, the identity (19), and the fact that Γ(t, z) and C(λ, z) are diagonal, we see that satisfies the Picard–Fuchs equation Iα(t, λ, q, z) = euα(q)/zI0(q)R0 ¯α(z) (21) (cid:16)(D + hα)5 − λ5 − q (5D + 5hα + kz − t)(cid:17) Iα(t, λ, q, z) = 0. Yk=1 At the limit t = 0, this proves the first part of the lemma. Now we can use a similar method as in the proof of Lemma 4.5 to compute the R-matrix. By the definition of Iα(t, λ, q, z), we have the following Picard–Fuchs equation 5 5 1 1 (L4 (L4 t3L11 α + (R1)0 α = (R2)0 α = 13 L12 α t2 12 + (cid:16) − 15 t L8 Using this relation, we can solve R1, R2, R3 inductively as rational functions of Lα. The explicit formulae are: αt − 5λ5)3(cid:18) − αt − 5λ5)6(cid:18) L22 +(cid:16) 517 L12 +(cid:16) 209 L8 +(cid:16) 205 L4 (cid:17)λ10 (cid:17)λ15 (cid:17)λ20 +(cid:16) 5 t3 (cid:17)λ5 +(cid:16) 75 Lα 5 37 L23 α t5 60 3 915 L20 30 871 L18 313 L24 288 (cid:17)λ10(cid:19) (cid:17)λ25(cid:19) 47 L17 150 + 25 Lαt2 − α + 5 t2L7 α − 22855 L10 α t3 37255 L15 α t3 − 180 L5 αt3 + +(cid:16) − 5625 L11 α t2 1181 L19 α t4 6343 L14 α t4 67 L13 α t5 3 3379 L9 αt4 9375 L12 α t 10 L3 αt 3 t2L2 α αt5 αt4 1375 L2 αt 22875 L16 α t6 25 − α t6 − α t5 − α t6 + 72 α t2 4 αt + − 32 4675 L6 5625 L7 + + 2 α t3 5625 L8 α 24 αt2 + 4 + 4 11 t3L6 α (cid:17)λ5 20 8 72 2 − 18 − − + 12 24 36 α t4 + 2 2 16 4 − 4 1800 + (cid:16)(D + Lα)5 − λ5 − q (5D + 5Lα + kz − t)(cid:17)I0(q)R0 ¯α(t, q, z) = 0. Yk=1 where Lα = hα + q d dq uα. The coefficient of z0 gives us an equation for Lα: (22) q(5Lα − t)5 + (L5 α − λ5) = 0. We can choose the basis ¯eαq=0 such that the solution Lα satisfies the following Lα = ξαλ + O(q), for α = 0, 1, 2, 3, 4, and from here we see that hα = ξαλ. Then, the coefficient of z1 of the Picard–Fuchs equation and the initial condition (18) imply that Next, we look at the coefficients of z2, z3 and z4 of the equation. By equation (22) we have I0(q) · Ψ0 ¯α = q d dq Lα = − 5Lα − t . αt) p5(5λ5 − L4 (5 Lα − t)(cid:0)Lα 4t − 25 λ5 5 Lα 5 − λ5(cid:1) . GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 27 (R3)0 α = (L4 1 αt − 5λ5)9(cid:18) − α t7 1876667 L30 56201 t6L36 α 51840 109099 L29 α t8 + 9000 + 2089 t7L35 α 1440 − α t8 73 L34 300 + α t9 2 L33 625 163861 L32 α t5 96 + 324773 L31 α t6 288 + 161 L28 α t9 15000 (cid:17)λ5 +(cid:16) − 25916335 L28 10077151 L27 α t5 7283627 Lα 26t6 + (cid:16) − α t4 + 384 774907 L19 17474431 L21 − 216 α t8 + 576 α t6 9000 (cid:17)λ15 +(cid:16) − 236621 L14 540 α t8 6000 8640 1339837 L25 α t7 900 10554529 L20 α t7 3600 11464411 L15 α t7 5400 72617625 L20 α t2 128 39623875 L15 α t2 32 − + − + + − + − 152344375 L10 α t2 192 1370375 L5 αt2 12 − + 16 4296875 L6 αt 24 − 6305489 L24 α t8 108000 − − − 46053127 L22 α t5 288 10970767 L16 α t6 360 − + α t9 2023 L23 45000 12368975 L24 (cid:17)λ10 +(cid:16) 22253 L18 810000 α t9 + α t3 17130505 L23 α t4 48 6661633 L17 α t5 + − 48 (cid:17)λ20 +(cid:16) − 560791 L10 α t7 1200 + 1220625 L16 α t + 32 (cid:17)λ25 + (cid:16) − 40 67837 L6 αt6 80 + 1704375 L11 α t 36159375 L12 α − 128 (cid:17)λ30 +(cid:16) − + 393750 L7 α(cid:17)λ35 + (cid:16)22500 Lαt + 12 422734975 L18 α t4 28694675 L19 α t3 + 36 420693 L11 α t6 864 34497817 L12 α t5 − 96 15774785 L13 α t4 72 + + 8465875 L14 α t3 64 38689 L7 1706375 L8 αt4 33813575 L9 αt3 432 αt5 − 4 693 L2 αt5 − 4 14375 t2 − 738125 L4 162 αt3 96 αt4 175 L3 2 + − 209375 L2 α 2 24 (cid:17)λ40(cid:19) Notice that each time we have a constant to fix, and we fix these constants by (20). To be precise, the constant terms of the above (Rk)0 α are fixed the following initial condition R(z)0 αLα=ξαλ = R(z) ¯α ¯αq=0 = e− 1 12 ( where we have used 1 t−5ξα λ + 2 ξα λ )z+ 1 360 ( 1 (t−5ξα λ)3 + −1 (ξα λ)3 )z3 + O(z4) Xα=1,2,3,4 1 1 − ξα = 2, Xα=1,2,3,4 1 (1 − ξα)3 = −1. Finally, by taking λ = 0 in the above formulae of Rk(t, λ) (note that in this limit Lα becomes t 5 qα as in Lemma 4.5), we recover the results of Lemma 4.5, and obtain the constant terms 1+qα as well. (cid:3) 4.3. Ψ-matrix and R-matrix: computations of the remaining entries. From Proposi- tion 4.1, we are able to compute the Ψ-matrix and R-matrix using the asymptotic expansion of Proposition 4.4. Proposition 4.7. We have the following formula for the Ψ-matrix: Ψ0 ¯α = ∆ Ψ1 ¯α = 2 1 − 1 α = 2 1 + qα I0 · q2 α 5 (cid:19)− 3 ·(cid:18)−t 5 (cid:19)− 3 α (cid:18) −t I1,1(cid:16)Lα − I1,1;at(cid:17) 1 + qα I1,1 ! Ψ0 ¯α I0 · q2 − I2,2;b t2 I2,2 Lα−I1,1;at 2 I2,2 −1 Ψ2 ¯α = det Lα−I2,2;at Ψ3 ¯α = det  I3,3 −1 Lα−I3,3;at − I3,3;b I3,3 Lα−I2,2;at t2 − I3,3;c t3 I3,3 − I2,2;b t2 I2,2 Lα−I1,1;at I2,2 −1 I1,1 Ψ0 ¯α   = det(cid:18) Lα − I1,1;at I1,1 (cid:19) Ψ0 ¯α 28 S. GUO, F. JANDA, AND Y. RUAN Proof. Define Ψi β := (H i, eβ)t. Then since we have eα, 1 =Xα Ψ0 1 β = 1 = Ψ0 ¯β∆ 2 β − 1 β = Ψ0 ¯β. Recall that Ψ0 ¯β was computed in Proposition 4.4. Using Proposition 4.1, 2 i.e. ∆ we get 3 2 = det(cid:18) Lα − I1,1;at I1,1 (cid:19) Ψ0 ¯α 1 I1,1(cid:16)Lα − I1,1;at(cid:17) 1 + qα 5 (cid:19)− ·(cid:18) −t I1,1 ! Ψ0 ¯α I0 · q2 α − I2,2;b t2 I2,2 Lα−I1,1;at 1 + qα I0 · q2 α 3 2 5 (cid:19)− ·(cid:18) −t Ψ0 ¯α = Ψ1 ¯α = I2,2 −1 Ψ2 ¯α = det Lα−I2,2;at Ψ3 ¯α = det  I3,3 −1 Lα−I3,3;at − I3,3;b I3,3 Lα−I2,2;at t2 − I3,3;c t3 I3,3 − I2,2;b t2 I2,2 Lα−I1,1;at I2,2 −1 I1,1 Ψ0 ¯α   Ψj ¯αΨk ¯α = (H j, H k)t Xα The relations provide a consistency check of the constants Cα fixed in Lemma 4.6. (cid:3) We have already computed R0α. Next we compute the remaining entries of the R-matrix. By applying Proposition 4.1 to Equation (15), we have the following inductive formula: R(z)1 α = R(z)2 α = R(z)3 α = R(z)4 α = 1 2 α q 1 2 α q 1 2 α q 1 2 α q 1 1 I1,1 (cid:18)∆ I2,2 (cid:18)∆ I3,3 (cid:18)∆ I4,4(cid:16)∆ 1 1 d dq d dq 1 2 ∆− α R(z)0 − 1 2 ∆ α R(z)1 d dq d dq − 1 2 ∆ α R(z)2 α + Lα − I1,1;atR(z)0 α(cid:19) α + (Lα − I2,2;at)R(z)1 α − I2,2;bt2R(z)0 α(cid:19) α + (Lα − I3,3;at)R(z)2 α − I3,3;bt2R(z)1 α − I3,3;ct3R(z)0 α(cid:19) − 1 2 ∆ α R(z)3 α + (Lα − I4,4;at)R(z)3 α − I4,4;bt2R(z)2 α − I4,4;ct3R(z)1 α − I4,4;dt3R(z)0 α(cid:17) Together with Proposition 4.4 (more precisely Equation (14)), we can then write down all the necessary entries of the R-matrix. We will now compute some very explicit entries of the Ψ- and R-matrix. For this, let us introduce the notation R4 ¯α(z) := ( H4, R(z)¯eα)t, (Rk)4 ¯α := ( H4, Rk¯eα)t, (Rk)4 α := ( H4, Rkeα)t. By applying Lemma 4.2 to the asymptotic expansion formula (13), and by using the result in Lemma 4.6, we obtain (23) (cid:18)zq d dq + Lα − t 5(cid:19) R4 ¯α(z) + I0 5 R0 ¯α(z) = 0. GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 29 Then, we can use this equation to compute the following entries of the Ψ- and R-matrix. 4.3.1. The entries Ψ4 α. Note that Ψ4 α := ( H4, eα) = (R0)4 α since R0 is the identity matrix in any basis. Consider the coefficient of z0 of Equation (23). Since Lα − t 5 = − t 1+qα 5 , we obtain Ψ4 ¯α = − 1 + qα − t 5 I0 5 Ψ0 ¯α. Recall that ∆ 1 2 2 = I0·q α 1+qα 2 = (Ψ0 ¯α)−1, so that we have 5 (cid:1)− 3 ·(cid:0) −t Ψ4 α = 1 + qα t I0, Ψ4 ¯α = (1 + qα)2 q2 αt 3 2 . (cid:18)−t 5 (cid:19) 4.3.2. The entries (Rk)4 α. Considering the coefficient of z1 of equation (23), we obtain q d dq Ψ4 ¯α +(cid:18)Lα − t 5(cid:19) (R1)4 ¯α + I0 5 (R1)0 ¯α = 0 Since q d dq (1+qα)2 q2 α = − 2(1+qα) 5q2 α , we have Hence, by (14), we have −I0 2 5t − t 5 1 1 + qα (R1)4 α + I0 5 (R1)0 α = 0 (R1)4 α = (1 + qα)(12 − 25qα) 12 t2 · qα I0. Furthermore, by considering the coefficients of z2 and z3 of equation (23) one after another, we deduce (R2)4 α = (R3)4 α = (1 + qα)(288 − 1176qα + 625q 288 t3 · q2 α 2 α) I0, (1 + qα)(20736 − 460512qα + 338868q 2 α + 11875q 3 α) 51840 t4 · q3 α I0. 4.4. Generators and relations. Next, we derive some basic relations in order to be able to write down the closed formula for the S and R-matrices in terms of a minimal number of generators. 4.4.1. Quantum product and relations between I-functions. Recall that in the flat basis, the quantum product τ ∗τ can be written in the following form τ ∗τ H k−1 = Ik,kH k + Ik,k;aH k−1t + Ik,k;bH k−2t2 + · · · . By [43], the functions Ik,k have the following properties (24) I0,0I1,1 · · · I4,4 = (1 − 55q)−1, Recall A is the matrix for τ ∗ in the flat basis {H k}. Ip,p = I4−p,4−p. 30 S. GUO, F. JANDA, AND Y. RUAN Lemma 4.8. The characteristic polynomial of A is given by x5 − q(5x − t)5 det(x − A) = 1 − 55q . (25) In particular, we have (26) ( τ ∗)5 − q(5 τ ∗ −t)5 = 0. Proof. Note that the eigenvalues of the matrix A dq q are just duα. Recall that by Lemma 4.5, duα = Lα dq q , Lα = 1 5 −tξαq 1 − 5ξαq , 1 5 so that the characteristic polynomial of A is (x − Lα) = x5 − q(5x − t)5 1 − 55q . Yα In fact, one can see that the numerator is just the coefficient of z0 of (16) with q d by x. dq uα replaced (cid:3) The characteristic polynomial gives us many relations between the entries {Ik,k;a, Ik,k;b, · · · } in the matrix A. However, these do not cover all relations. For additional relations, we can use the symmetry of the quantum product (27) ( τ ∗τ H k, H j)t = ( τ ∗τ H j, H k)t. Example 4.9. By taking (k, j) = (0, 3) and (1, 2) in (27), we obtain the following relations between Ik,k;a: I1,1 − 5I1,1;a = I4,4 − 5I4,4;a, I2,2 − 5I2,2;a = I3,3 − 5I3,3;a Furthermore, we have 5 −5I5,5;a = I0 − 1, Ik,k;a = t−1TrA = − 55q 1 − 55q . Xk=1 Together with the symmetry of Ik,k: I3,3 = I1,1 and I4,4 = I0, we deduce (28) I1,1;a + I2,2;a = I2,2 − 1 10 − 1 2 · 55q 1 − 55q . In the end of this subsection, we conclude that by using the characteristic polynomial of A and the symmetry of quantum product, there are indeed only two independent functions in {Ik,k;a, Ik,k;b, · · · }: Lemma 4.10. Denoting the entries in A by ai,j, we have ai,j ∈ I −2 1,1 I −1 2,2 Q[I0, I1,1, I2,2, L, P, Q], ∀i, j = 0, 1, · · · , 4. Proof. Let us list all the relations mentioned in this subsection. First, the basic relations (Zagier–Zinger's relations) are Ik,k = I4−k,4−k, for k = 0, 1, 2, 3, 4, and I 2 0,0I 2 1,1I2,2 = X := 1 1 − 55q . GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 31 Next, we can use the symmetry (27) and the characteristic polynomial (25). For the extra I5,5;∗-functions, we have 1 5 I5,5;a = (1 − I0,0), I5,5;b = 1 625 (25 − 25I3,3 − 125I4,4;a), I5,5;c = I5,5;d = I5,5;e = 1 625 1 625 1 625 (5 − 5I2,2 − 25I3,3;a − 125I4,4;b), (1 − I1,1 − 5I2,2;a − 25I3,3;b − 125I4,4;c), (−I1,1;a − 5I2,2;b − 25I3,3;c − 125I4,4;d + 1 5 (1 − I0,0)). For the extra I4,4;∗-functions, we have I4,4;a = I4,4;b = I4,4;c = 1 5 1 5 1 5 (I0,0 + (5I1,1;a + 5I2,2;a − I2,2 − 5I3,3;a)), (−I2,2;a + 5I2,2;b + I4,4;a), (−I3,3;b + 5I3,3;c + I4,4;b). For the extra I3,3;∗, I2,2;∗-functions, we have I3,3;a = 1 5 (I1,1 + (5I2,2;a − I2,2)), I2,2;a = −I1,1;a − 1 10 (1 − I2,2) − 1 2 (X − 1) Moreover, for the other extra I-functions, we have I3,3;b = 1 100 I2,2(cid:0)25 L10 − 40 L5 − 200 L2P + 4 I1,1 I2,2 − I2,2 2(cid:1) 10 L5 − I1,1 2I2,2 I3,3;c = − 250 I1,1 I2,2 − L2(cid:0)5 L5 + I2,2(cid:1) P 10 I1,1 I2,2 + L6(cid:0)5 L5 + I2,2 − 8(cid:1) Q 20 I1,1 I2,2 + L2Q2 10 I1,1 − 2 QPL3 I1,1 I2,2 I4,4;d = 10 I1,1 L5 − 10 L5 + 2 I0,0 I 2 1,1I2,2 1250 I 2 1,1I2,2 − I 3 1,1I2,2 + L6(cid:0)−5 I1,1 L5 + I1,1 I2,2 + 8 I1,1 − 8(cid:1) 100 I1,1 2I2,2 Q + L2(cid:0)25 L10 − 40 L5 + 2 I1,1 I2,2(cid:1) 100 I1,1 2I2,2 Q2 + P L2(5 L5 − I2,2) 50 I1,1 I2,2 − L3(5L5 − 2I1,1) 5 I 2 1,1I2,2 Q − 2 L4 I 2 1,1I2,2 Q2! + L4P 2 I 2 1,1I2,2 The lemma follows from a careful examination of all these formulae. (cid:3) 4.4.2. Identities between derivatives of basic and extra generators. Lemma 4.11. We have the following identities between the basic generators and their deriva- tives: (29) (30) Y2 = − 3X2 − Y 2 − X 2 − 15 4 Z2 X4 = − 4X X3 − 3X 2 2 − 6X 2X2 − X 4 − 29 9 1 Z2 − 65 72 Z 2 Z1Z3 − 15 4 3 4 (Z2X 2 + Z2X2 + Z3X ) Z 2 2 − 23 24 Z4 + Proof. This is just a restatement of the identities in [42]. (cid:3) 32 S. GUO, F. JANDA, AND Y. RUAN Lemma 4.12. We have the following identities for the extra generators d2 du2 Q = − 2 X 5 d du Q −(cid:0)4 X2 + 2 X 2 + 2(cid:0)2Z1X2 + Z1X 2 + Z2X(cid:1) − 15 Z2 (cid:1)Q 4 125 Z1Z2 24 d2 du2 P = − (3 X + Y) d du − 5 Z3 24 , 15 Z2 4 P +(cid:0)2 X2 − X 2 − 2 YX + Y 2 + d du 1 d du 5 + X − Y)Q(cid:1)(cid:0)( 2(cid:0)( 2(cid:0)Z2(X2 + X 2) + Z1 + X − Y)(2Q + 5Z1)(cid:1) − 305 Z2 2 − 35 Z1 8 64 2X2 + Z3X(cid:1) + (cid:1)P 15 Z1 4 2Z2 − − + 2X 2 − 55 Z1Z2X 4 + 10 Z1 4. (31) (32) (33) (34) Proof. Recall that in Lemma 4.2, we have proved the following identity for two types of special S∗-matrices S∗(t − 5H)( H4) = H4, S∗(cid:16)1 2 (t − 5H)(cid:17)( H4) = H4 On the other hand, we have the explicit formula for S∗1 : S∗(t − 5H)1 = S∗(cid:16) t − 5H 2 (cid:17)1 = I −1 0 and all the other columns of the special S∗-matrices can be computed by Birkhoff factor- ization starting from S∗1 (see Proposition 4.1). Comparing with (33) and (34), we will get many identities between basic generators, extra generators P, Q and their derivatives. Let us introduce Pk := dk−2 duk−2 P, Qk := dk−1 duk−1 Q. In particular, we have Q = Q1, P = P2. By definition, the degrees of Pk and Qk are both k. By Lemma 4.11 and (33), we can solve Q4 and P4 as rational functions of the basic generators and P2, P3, Q1, Q2, Q3. Furthermore by using (34), we can solve Q3 as rational functions of the basic generators and P2, P3, Q1, Q2. Finally, by applying the formula for Q3 to the formula for P4, we also write down P4 as rational functions of the basic generators and the four extra generators P2, P3, Q1, Q2. Since the details of solving these equations are tedious, we omit them. The resulting two formulae for P4 and Q3 are precisely what we claim. (cid:3) Corollary 4.13. The ring generated by all basic generators, extra generators and their derivatives is isomorphic to Q[X , X2, X3, Y, Q, Q, P, P, L]. Proof. This follows from the preceding relations and Remark 3.3. (cid:3) Remark 4.14. Notice that here Q[X , X2, X3, Y, Q, Q, P, P, L] is defined as the ring gener- ated by these nine generators. The corollary does not imply that there are no other relations between the generators. 5. Proof of key propositions Recall that the key propositions 3.7, 3.6 and 3.5 directly imply the Main Theorem. In this section, we will finish their proofs. GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 33 5.1. Proof of Proposition 3.7. We want to compute the contributions of graphs with a genus 0 quasimap vertex. Recall that there are two graphs, with contributions given by the following correlators Cont′ Γ0 a Cont′ Γ0 b 3H 3 ⊗ H 4t−1 + 5 3H 4t−1 ⊗ H 3 + 65 :=L2(cid:28)(cid:28) 5 :=L2(cid:28)(cid:28) − 5 (t − 5H)(t − 5H − ψ1) (t − 5H)(t − 5H − ψ2) (cid:29)(cid:29) (t − 5H)(t − 5H − ψ2)(cid:29)(cid:29) (t − 5H)(t − 5H − ψ1) 8 H 4t−1 ⊗ H 4t−1 3 H 3 + 5 3H 3 + 5 24 H 4t−1 24H 4t−1 − 5 0,2 , t t , 0,2 Definition 5.1. We define the following modified S-matrix ∗ S and a modified V -matrix V (t)(γ1 ⊗ γ2) :=(cid:18)γ1 ⊗ γ2, ejhhej, (t)(γ) := γ +Xj 2t − 5(1 ⊗ H + H ⊗ 1)(cid:19)t 1 γ t − 5H − ψ iit 0,2, +(cid:28)(cid:28) γ1 t − 5H − ψ , γ2 t − 5H − ψ(cid:29)(cid:29)t 0,2 . By definition, we can write the modified matrix as a specialization of the original matrix, for example ∗ S (t) = Resz=t−5H S∗(z) 1 t − 5H − z = S∗(z)z=t−5H . Example 5.2. As pointed out in the proof of Lemma 4.12, by the definition of the I-function for the twisted theory (7), we have ∗ S (t)(1) = 1 I0 . By definition of V , the genus 0 contribution is just the modified V -matrix with given insertions. Since (see e.g. [19]) V (t, z, w)(γ1 ⊗ γ2) = 1 z + w (S(t, z)γ1, S(t, w)γ2)t the modified V -matrix can be computed as follows V (t)(γ1 ⊗ γ2) ∗ 2t − 5 (1 ⊗ H + H ⊗ 1)!t (eα) ⊗ S (eα) ∗ = γ1 ⊗ γ2, Pα S = γ1 ⊗ γ2,Pj=0,1,2,3 S ∗ ∗ (H j) ⊗ S (H 3−j) − t5 625 S 2t − 5 (1 ⊗ H + H ⊗ 1) ∗ ( H4) ⊗ S ∗ ( H4) !t (35) where we have used ( H4, H4)t = − 625 t5 . Denote by Sk := S where { Hi} is the dual basis of {H k}. ∗ (t)H k, Sk i := ( Hi, Sk)t, 34 S. GUO, F. JANDA, AND Y. RUAN Lemma 5.3. For i = 0, 1, 2, 3, 4; k = 0, 1, 2, 3, the entries of the modified S-matrix satisfy i Sk ∈ tk−i I0I1 · · · Ik Q[X , X2, X3, Y, L, P, Q, P3, Q2, I1, I2] where Ik := Ik,k L for k = 1, 2, 3 (notice that I3 = I1). Moreover, if we define deg Ik = 1 for k = 1, 2, 3 and rewrite Lj (j = 2, 3, 4, 7, 8, 12) in Sk as polynomials of Zk (k = 1, 2, 3) (see Remark 3.3), then for all i = 0, 1, · · · , 4 and k = 0, 1, · · · , 3 (ti−kI0I1 · · · Ik) · Sk i ∈ Q[X , X2, X3, Y, Z, Z2, Z3, P, Q, P, Q, I1, I2] are homogeneous polynomials of degree k. Proof. This lemma is a direct consequence of Proposition 4.1 and the fact S0 = I −1 more explicit, by a direct Birkhoff factorization computation we have 0 . To be S1 = S2 = S3 = L4 5 1 1 + Q + X(cid:1) · t(cid:17) I0I1(cid:16)(cid:0)L4 + 5X ) · H − ( I0I1I2(cid:16)(cid:0)4L3 − 3L8 + 5L4(X + Y) + 25X Y − 25X2(cid:1) · H 2 (cid:0) (X − 4Y) + 5 Q − 10X Y + 10 X2 − −16L3 + 17L8 L4 2 10 + (−3X + 2Y) − P − Q + X Y − X2 + 8L3 − 11L8 50 + L4 10 1 I0I 2 + h − + (4L3 − 3L8)Q + 5 L4Q(X + Y) + 25 Q(X Y − X2) + 211L12 − 282L7 + 67L2 + (33L3 − 31L8)X − 5 2 10 I2 10 (L4 + 5X )(cid:1) · tH 50 (cid:0)L4 + 5 Q + 5X(cid:1) I2 1 I2(cid:16)h36L12 − 47L7 + 12L2 + (60L8 − 55L3)X + 25L4(X 2 + X2) + 125(X 3 + 3X X2 + X3)i · H 3 L4X 2 − 15 L4X2 − 75(X 3 + 3X X2 + X3) + 5 Q(L4 + 5X ) I1 5 (cid:0)3 L8 − 4 L3 − 5 L4(X + Y) + 25(X2 − X Y)(cid:1)i · t H 2 + h 103L12 − 131L7 + 31L2 25 + (26L8 − 23L3)X 5 + 5 P + 5 Q(Q − 2X + L4 10 ) + (cid:0)L4 + 5X(cid:1) P + + L4(3X2 − 2X 2) + 15(X 3 + 3X X2 + X3) Q 10 (17L8 − 16L3 + 5L4(X − 4Y)) + 10Q(X2 − X Y) I1 − 50 (cid:16)50 Q + 17L8 − 16L3 + 5L4(X − 4Y) + 100(X2 − X Y)(cid:17) − I1I2 50 (cid:0)L4 + 5X(cid:1)i · t2H + h −67L12 + 74L7 − 9L2 250 + L4 10 (3X 2 − 2X2) − − (X 3 + 3X X2 + X3) (7L8 − L3)X 25 − P − ( 3 10 L4 + Q + X ) Q − (Q + X + L4 5 )P − Q(X2 − X Y) + Q 50 (8L3 − 11L8) + Q 10 L4(2Y − 3X ) + I1 250 (cid:16)11L8 − 8L3 + L4(15X − 10Y) + 50(X2 − X Y) + 50P + 50 Q(cid:17) + I1I2 250 (cid:0)L4 + 5Q + 5X(cid:1) − I 2 1 I2 250 i · t3(cid:17) where we have used the relations between {Ik,k;a, Ik,k;b, · · · } presented in the proof of Lemma 4.10 and the differential equations in Lemma 4.12 (cid:3) Finally, by applying the formulae in the proof of Lemma 5.1 and the formula in Lemma 4.2 to equation (35), with ∗ S (t)( H4) = H4 γ1 ⊗ γ2 = t−2 5 3H 3 ⊗ H 4t−1 + 5 3 H 4 ⊗ H 3t−1 + 65 8 H 4 ⊗ H 4t−2 1 − t−1 ⊗ 5H − 5H ⊗ t−1 + 25H ⊗ H or γ1 ⊗ γ2 = − 5 3 H 3 + 5 t − 5H 24 H 4t−1 ⊗ − 5 3H 3 + 5 t − 5H 24H 4t−1 , GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 35 we can write down the explicit formula of L2 · V (t) as a homogeneous polynomial of degree 3 in Q[X , X2, X3, Y, Z, Z2, Z3, P, Q, P, Q]. Note that here we have used 0 I 2 I 2 1 I2 = L2, deg L2 = 3. The explicit computations exactly give us the formulae presented in Proposition 3.7. 5.2. Proof of Proposition 3.6. There is only one graph with a genus 1 quasimap vertex, and it has the contribution Cont′ Γ1 := L2 3H 3 + 5 I0 (cid:28)(cid:28) − 5 (t − 5H)(t − 5H − ψ)(cid:29)(cid:29) 24 H 4t−1 t . 1,1 This type of twisted invariants can be computed by the following two steps. First, by using the modified S-matrix S(t), we can write the modified descendent correlators in terms of the S action on the ancestor genus one invariants (see [31] and [20]). Explicitly, for any α ∈ H ∗ C∗(P4), we have hh α t − 5H − ψ S( ¯ψ)α iit 1,1 = hh(cid:2) t − 5H − ¯ψ(cid:3)+iit 1,1 = hh S(t)α t − 5H − ¯ψ iit 1,1, where ¯ψ is the pullback psi class from M1,1, we expand t−5H− ¯ψ in terms of positive powers of H and ψ with S acting on the left, and the bracket [·]+ picks out the terms with a non-negative power of ψ. 1 Next, we evaluate the ancestor invariants of the following form hhγ( ¯ψ)iit 1,1 =ZM1,1 Ω1,1(γ(ψ1)) by using Givental–Teleman's reconstruction theorem [20, 40]. We recall here only the general shape of the reconstruction theorem and refer the reader to [38] for more details. For 2g − 2 + n > 0, the theorem says that Ωg,n(γ1, · · · , γn) = RT ωg,n(γ1, · · · , γn) = XΓ∈Gg,n 1 Aut(Γ) (ιΓ)∗T ωΓ(γ1, · · · , γn) where Gg,n is the set of genus g, n marked point stable graphs, ιΓ is the canonical morphism ιΓ : MΓ → Mg,n, the translation action T is given by T ωg,n(−) := 1 k! (pk)∗ωg,n+k(−, T (ψ)k) ∞ Xk=0 for the cohomological valued formal series (36) T (z) = T1z2 + T2z3 + T3z4 + · · · := z(1 − R−1(z)1), and below we will define ωΓ(γ1, · · · , γn) for each graph under discussion. In our case, the only insertion γ is given by − 5 γ(ψ) := S(t) 3 H 3 + 5 24H 4t−1 . (t − 5H)(t − 5H − ψ) 36 S. GUO, F. JANDA, AND Y. RUAN The contribution can be written as a summation over two stable graphs: Cont′ Γ1 = L2 I0 ZM1,1 Ω1,1(γ(ψ)) = where Γ1 a and Γ1 b are the following stable graphs L2 I0 (cid:16)ContΓ1 a + ContΓ1 b(cid:17) Γ1 a := γ •g=1 Γ1 b := γ WVUT PQRS • g=0 . We now compute the contribution of each graph separately. 5.2.1. Contribution of Γ1 a. We have T ωΓ1 a(γ) := T ω1,1(R−1(ψ) · γ) = ω1,1(γ) − ψ1 ω1,1(R1γ) + ψ1 ω1,2(γ, T1) = Xα (eα, γ)t + ψ1(cid:0) − (eα, R1γ)t + (eα, γ)t(eα, T1)t(cid:1) ωg,n(γ1, · · · , γn) = ∆g−1 n (eα, γi)t. Yi=1 where we have used In particular, we have T ωΓ1 a(eα) = 1 + ψ1(cid:0)(R∗ 1eα,X eβ)t + (R1)0 α(cid:1). α is in equation (14). After a direct computation using the α and the inductive formulae for the Ψ- and R-matrix of Section 4.3, we Note that the formula for (R1)0 formula for (R1)0 obtain (R∗ 1eα,X eβ)t = (37) 55 12t + 625 + 25 2t q2 5 αL2 (12(L5 − 1)L2 + 60L3X + 75L4 Q + 250 P) t qα αL3 (2(L5 − 1)L3 + L4(8X + Y) + 50L P + (15L5 − 5) Q) αL4 (8(L5 − 1)L4 + 10L5(2X + Y) t q3 625 2t q4 + + 5.2.2. Contribution of Γ1 b. Let + 10X + 250L2 P + 25L(3L5 − 2) Q) W (w, z) := Pα eα ⊗ eα − R−1(z)eα ⊗ R−1(w)eα w + z then we have W (w, z) = W0 + W1 + W2 + · · · where R1eα ⊗ eα, W1 =Xα (cid:16) − R2z − R∗ W0 =Xα W2 =Xα (cid:16)R3z2 + (R1R2 − R3)wz + R∗ 3w2(cid:17)eα ⊗ eα 2w(cid:17)eα ⊗ eα GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 37 The contribution from the loop type graph is given by T ωΓ1 b (γ) := T ω0,3(W (ψ), R−1(ψ) · γ) ω0,3(eα, R1eα, γ) = ω0,3(W0, γ) = Xα = Xα ∆−1 α (eα, R1eα)t(eα, γ)t = Xα (eα, γ)t(R1) ¯α ¯α In particular, T ωΓ1 b (eα) = (R1) ¯α ¯α. Again after a direct computation using the formula for (R1)0 Section 4.3, we obtain α and using the formulae from (R1) ¯α ¯α = 55 12t + 125 + 25 2t q2 αL2 (4(L5 − 1)L2 + 20L3X + 25L4 Q + 50 P) 5 t qα αL3 (2(L5 − 1)L3 + 10L4X + 50L P + (25L5 − 10) Q) αL4 (6X + 2Y + 50L2 P + 5L(4L5 − 3) Q) t q3 625 2t q4 (38) + + To summarize, given we have γ(ψ) = γ0 + γ1ψ + · · · ContΓ1 T ωΓ1 a(γ) = a = ZM1,1 2ZM0,3 a = 1 ContΓ2 T ωΓ1 a(γ) = 1 24Xα (cid:16)(γ1, eα)t + (γ0, eα)t(cid:16)(cid:0)R∗ 2Xα (γ0, eα)t(R1) ¯α ¯α. 1 1eα,X eβ(cid:1)t + (R1)0 α(cid:17)(cid:17) , Finally, by equation (14), (37) and (38), together with the explicit formula for the insertion (γ0, eα) = (γ1, eα) = t I0 t q 6000L2(cid:16)5381L12 − 7274L7 + 1893L2 + L8(8080X + 410Y) + 25L4(80X2 − 43X 2 + 82X Y) 15000(cid:16)3L5 + KL2 − 8850L3X + 20250(X 3 + 3X X2 + X3) − (1975L4 + 10250X ) Q + 250 P(cid:17) − + 15LX + 38(cid:17) − 6000L2(cid:16)(2941L12 − 4314L7 + 1373L2) + (4880X + 10Y)L8 + (1925X 2 + 50X Y + 2000X2)L4 − 6850L3X + 10250(X 3 + 3X2X + X3) − Q(250X + 2975L4) − 9750 P(cid:17) + 75000(cid:16)2L5 + KL2 + 10LX + 39(cid:17) − 375000(cid:16)L4 + KL + 5X(cid:17) − αL2K t q 1875000 3 αLI0 t qαI0 (∗) q 2 αI0 t q I0 k α , 4 4 Xk=1 where K := 123L8 − 164L3 − 205L4(X + Y) + 1025(X2 − X Y) + 25 Q, and where (∗) denote some α-independent functions which we have omitted because they are irrelevant for the further computations, we arrive at the following formulae. 38 S. GUO, F. JANDA, AND Y. RUAN ContΓ1 b are degree 3 homogeneous polynomials in the ContΓ1 a and L2 Lemma 5.4. Both L2 I0 basic and extra generators, to be precise L2 I0 576(cid:0)X + 2Z1(cid:1) Q + ContΓ1 a = 205 5 I0 3055 Z2X 5 Z1X 2 + + 2304 145X 576 − Y 48 − (3X 2 − X2 + X Y) − L2 I0 ContΓ1 b = − + − P +(cid:0) 384 473 576 41 Y 48 659 Z1 576 576(cid:0)X3 + 4X X2 + X 3 − YX 2 − 3Z1YX(cid:1) + 9275 Z1Z2 2X 2Y − 95 Z1 144 − 305 Z1X2 288 5285 Z3 13824 + 13824 + 205 Z1 288 45 64 2927 Z1X 2 + (X3 + X 3) − 299 X X2 64 2113Z1 1152 (cid:1) Q − 665 Z1X2 144 + 4223 Z1YX 576 − 621 Z2X 256 2X + 41 48 (Z1Y 2 − Z2Y) + 2747 Z1 576 29465 Z1Z2 4608 − 1555 Z3 4608 1152 2Y − The proposition is a direct consequence of the above lemma. 5.3. Proof of Proposition 3.5. Finally, we deal with the remaining graph with a single genus two quasimap vertex. It has the contribution Again using Givental–Teleman's theorem, we see that we need to compute Cont′ Γ2 := L2 I 2 0 · hhiit 2,0 . hhiit 2,0 =ZM2,0 There are 7 stable graphs in G2,0: Γ2 1 := •g=2 Γ2 2 := • g=1 • g=1 Ω2,0 = XΓ∈G2,0 1 Aut(Γ) ContΓ. Γ2 5 := g=0 WVUT WVUT PQRS PQRS • Γ2 6 := WVUT PQRS• Γ2 3 := •g=1 WVUT PQRS •g=0 WVUT PQRS Γ2 4 := • g=1 • WVUT PQRS g=0 Γ2 7 := • WVUT PQRS •g=0 Hence we can write the contribution as summation over contributions of the stable graphs: (39) Cont′ Γ2 = L2 I 2 0 (cid:18)ContΓ2 1 + ContΓ2 2 2 + ContΓ2 3 2 + ContΓ2 4 2 + ContΓ2 5 8 + ContΓ2 6 8 + ContΓ2 7 12 (cid:19) In the remaining part of this section, we compute each of these contributions. 5.3.1. Trivial graph. The first graph Γ2 its contribution: 1 is a single genus-two vertex. We directly compute ContΓ2 1 =ZM2,0 T ω2,0 = ZM2,0(cid:18)ω2,0 + (π1)∗ω2,1(T (ψ)) + = ω2,1(T3)ZM2,1 = Xβ ∆β(cid:16) 1 (eβ, T3)t + 29 5760 1152 1 + ψ4 2 2 1 2! (π2)∗ω2,2(T (ψ)2) + 1 3! ω2,2(T1, T2)ZM2,2 1ψ3 ψ2 2 + 1 6 ω2,3(T 3 (eβ, T1)t(eβ, T2)t + (π3)∗ω2,3(T (ψ)3)(cid:19) 1 )ZM2,3 1ψ2 ψ2 2ψ2 3 7 6 · 240(cid:16)(eβ, T1)t(cid:17)3(cid:17) GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 39 where we have used some known intersection numbers in Mg,n. 5.3.2. One-edge graphs. There are two graphs with one edge: Γ2 2 := • g=1 • g=1 Γ2 3 := •g=1 WVUT PQRS For the computation of ContΓ2 2 , it is useful to notice that T ωτ 1,1(eβ) = 1 + (eβ, T1)t · ψ. With this, we can compute: ContΓ2 2 (T ω1,1 ⊗ T ω1,1)(W (ψ1, ψ2)) = ZM1,1×M1,1 242 Xα,β (cid:16)(eβ, T1)t(eα, R1eβ)t(eα, T1)t − 2(eα, (R∗ = 1 2)eβ)t(eβ, T1)t We next compute Cont′(Γ2 3) using + (eα, (R1R2 − R3)eβ)t(cid:17) T ω1,2(−) = ω1,2(−) + (π1)∗ω1,3(−, T (ψ)) + 1 2! (π2)∗ω1,4(−, T (ψ)2) and R2 + R∗ 2 = R2 1: ContΓ2 3 T ω1,2(W (ψ1, ψ2)) = ZM1,2 = hω1,2(W2(ψ1, ψ2))i1,2 +(cid:10)ω1,3(W1(ψ1, ψ2), T1), ψ2 3(cid:11)1,3 1 )(cid:10)ψ2 4(cid:11)1,4 = Xβ + ω1,3(W0, T2)(cid:10)ψ3 ∆β(cid:16) 1 3 + R1R2)eβ)t − 3(cid:11)1,3 + (eβ, (R2 + R∗ ω1,4(W0, T 2 (eβ, (R∗ 1 12 3ψ2 1 2! 24 2)eβ)t(eβ, T1)t + 1 24 (eβ, R1eβ)t(eβ, T2)t + 1 2 · 6 (eβ, R1eβ)t(cid:16)(eβ, T1)t(cid:17)2(cid:17) 5.3.3. Two-edge graphs. There are two graphs with two edges: Γ2 4 := • g=1 We compute Γ2 4 as follows: • WVUT PQRS g=0 Γ2 5 := g=0 WVUT PQRS WVUT PQRS • (T ω1,1 ⊗ ω0,3)(W0 + W1, W0) ContΓ2 4 For Γ2 5, recalling 1 = = ZM1,1 24(cid:16)Xα,β 24Xα,β = 1 (eβ, T1)t(eβ, R1eα)t(eα, R1eα)t + (eβ, −R∗ 2eα)t(eα, R1eα)t(cid:17) ∆1/2 α ∆1/2 β (cid:16)(eβ, T1)t(R1)αβ(R1)αα − (R1)αα(R∗ 2)βα(cid:17) T ω0,4(−) = ω0,4(−) + ω0,5(−, T1) · ψ1, 40 S. GUO, F. JANDA, AND Y. RUAN we can write ContΓ2 5 T ω0,4(W ⊗ W ) = ZM0,4 = ω0,5(W 2 0 , T1) − 2Xα ∆β(cid:16)(cid:16)(eβ, R1eβ)t(cid:17)2 = Xβ ω0,4(W0, (R2 + R∗ 2)eα, eα) (eβ, T1)t − 2(eβ, R1eβ)t(eβ, (R2 + R∗ 2)eβ)t(cid:17) 5.3.4. Three-edge graphs. There are also two graphs with three edges: Γ2 6 := WVUT PQRS• •g=0 WVUT PQRS Γ2 7 := • WVUT PQRS •g=0 The computation of their contributions is similar to the previous ones: ContΓ2 6 = (ω0,3 ⊗ ω0,3)(W0 ⊗ W0 ⊗ W0) (R1eβ, eβ)t(eα ∗τ R1eα, R1eβ)t (∆α∆β)1/2(R1)αα(R1)ββ(R1)αβ = Xα,β = Xα,β = Xα,β ContΓ2 7 (∆α∆β)1/2(R1)3 αβ 5.3.5. Conclusion. For k = 1, 2, · · · , 7, we introduce Cont′ Γ2 k := L2 I 2 0 ContΓ2 k . After a long but direct computation using the R-matrix formulae in Section 4.3, we arrive at the following proposition. Proposition 5.5. Both Cont′ Γ2 1 basic generators: and Cont′ Γ2 2 are degree 3 homogeneous polynomials in the Cont′ Γ2 1 Cont′ Γ2 2 = = 1 21536 (24475Z3 − 17565Z1Z2) 5 576(cid:0)X3 + 4X X2 + X 3 − YX 2(cid:1) + 1 + 25 768 Z2X + 21436(cid:0)139157Z3 − 12531Z1Z2(cid:1) 5 144(cid:0)Z1X2 − Z1YX − Z 2 1 (X + Y)(cid:1) GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 41 Both Cont′ Γ2 3 erators: and Cont′ Γ2 4 are degree 3 homogeneous polynomials in the basic and extra gen- Cont′ Γ2 3 = 289 6912 P + 1345 13824 Z1 Q − − Cont′ Γ2 4 = − + − 1 24(cid:0)X3 + 2X X2 + X 3 + YX 2(cid:1) − 3347 Z2X + 390 Z1 Z1(cid:0)X 2 + YX(cid:1) 2Y 2X − 682 Z2Y − 1630 Z1 13824 1 6 + 21573Z1Z2 − 68155Z3 21235 5 X 288 − 865 Z1 13824(cid:1) Q (3X + Y)(X Y − X2) − P +(cid:0) 289 6912 1 24 205 Z2X 13824 + 1055 Z1 2304 2X − 235 Z2Y 6912 + 3455 Z1 6912 5 Z1X2 24 + 53 Z1X 2 192 + 2Y 1 12 + Z1Y 2 + 151 Z1YX 288 46085 Z1Z2 − 6875 Z3 331776 The sum Cont′ Γ2 5 extra generators: + Cont′ Γ2 6 + Cont′ Γ2 7 is a degree 3 homogeneous polynomial in the basic and Cont′ Γ2 5,6,7 := 1 8 Cont′ Γ2 5 + Cont′ 1 Γ2 8 6 3X + Y 48 + + Γ2 7 Cont′ 1 12 245Z1 2304 (cid:1) Q 5 1152 1 P −(cid:0) = − − + 24(cid:0)X 3 + 3 YX 2 + 3 Y 2X + Y 3 + 8 Z1X 2 + 11 Z1(YX + 15449 Z1Z2 161 Z2X 2X 2Y − 325 Z1 384 − 605 Z1 1152 − 27648 2304 Y 2 2 5225 Z3 82944 + ) + Z2Y(cid:1) Proposition 3.5 now follows directly from (39) and Proposition 5.5. For reference, we also provide individual formulae for Cont′ Γ2 5 are rational functions in the basic and extra generators. , Cont′ Γ2 6 and Cont′ Γ2 7 . They Lemma 5.6. Introduce the following inhomogeneous polynomial E := (cid:16)4 Q2 + 2 (3 X + Y) P(cid:17) L − 20 Q P L2 + 25 P 2 L3 Q(3 X + Y) L5 − 10 Q2 L6 + 25 Q P L7 + + 1 5 1 25 (3 X + Y)2 L9 + 25 4 Q2L11. 42 S. GUO, F. JANDA, AND Y. RUAN We have the following polynomiality for Cont′ Γ2 5 , Cont′ Γ2 6 and Cont′ Γ2 7 : Cont′ Γ2 5 = − + − 3E L5 − 1 1 + 1183 90 79 X 6 + P −(cid:0) 5(cid:0)6 X 3 − 2 YX 2 − 8 Y 2X(cid:1) + 119 Z2X 61 Z2Y 3221 Z1 − 71 Y 10 101 Z1X 2 360 + (cid:1) Q 431 Z1YX 43 Z1Y 2 + 30 20 176539 Z1Z2 2Y + 6912 − 17389 Z3 6912 − − 359 Z1YX 30 − 51 Z1Y 2 20 76795 Z1Z2 6912 + 6565 Z3 6912 4 + 2X 879 Z1 20 11 Y + 5 + 5 21 X − 72 E L5 − 1 1 + − 869 240 36 P +(cid:0) 5(cid:0)7 X 3 + 9 YX 2 + 3 Y 2X + Y 3(cid:1) − 33 Z2X 31 Z2Y 2X − 967 Z1 48 − 32 3E L5 − 1 1 − − 1147 80 72 P +(cid:0) 127 X 10 + 71 Y 10 − 557 Z1 180 439 Z1 288 (cid:1) Q 877 Z1X 2 2Y 60 1087 Z1 144 5957 Z1 480 (cid:1) Q Cont′ Γ2 6 = Cont′ Γ2 7 = − − + + 5(cid:0) − X 3 + 9 YX 2 + 9 Y 2X − Y 3(cid:1) − 467 Z2X 67 Z2Y 2X 96 − 48 − 3669 Z1 80 399 Z1X 2 20 2Y 91 Z1 240 − − 91 Z1YX 10 65321 Z1Z2 2304 − + 43 Z1Y 2 20 21461 Z3 6912 + The geometric meaning of this inhomogeneous polynomial is unclear. It has the following expansion E L5 − 1 = 45q + 227400q2 + 1195603370q3 + 5913833272300q4 + O(q5). All coefficients are integers. References [1] M. Bershadsky, S. Cecotti, H. Ooguri, and C. Vafa. Holomorphic anomalies in topological field theories. Nuclear Phys. B, 405(2-3):279–304, 1993. [2] P. Candelas, X. C. de la Ossa, P. S. Green, and L. Parkes. A pair of Calabi-Yau manifolds as an exactly soluble superconformal theory. Nuclear Phys. B, 359(1):21–74, 1991. [3] H.-L. Chang and J. Li. Gromov-Witten invariants of stable maps with fields. Int. Math. Res. Not. IMRN, (18):4163–4217, 2012. [4] H.-L. Chang, J. Li, W.-P. Li, and C.-C. Liu. An effective theory of GW and FJRW invariants of quintics Calabi-Yau manifolds. arXiv:1603.06184, 2016. [5] Q. Chen, F. Janda, and Y. Ruan. (in preparation). [6] Q. Chen, F. Janda, Y. Ruan, A. Sauvaget, and D. Zvonkine. Effective r-spin structures and a formula for Witten's r-spin class. (in preparation). [7] A. Chiodo and Y. Ruan. A global mirror symmetry framework for the Landau-Ginzburg/Calabi-Yau correspondence. Ann. Inst. Fourier (Grenoble), 61(7):2803–2864, 2011. [8] I. Ciocan-Fontanine and B. Kim. Quasimap wall-crossings and mirror symmetry. arXiv:1611.05023, 2016. [9] I. Ciocan-Fontanine, B. Kim, and D. Maulik. Stable quasimaps to GIT quotients. J. Geom. Phys., 75:17–47, 2014. [10] I. t. Ciocan-Fontanine and B. Kim. Higher genus quasimap wall-crossing for semipositive targets. J. Eur. Math. Soc. (JEMS), 19(7):2051–2102, 2017. [11] E. Clader, F. Janda, and Y. Ruan. Higher-genus quasimap wall-crossing via localization. arXiv:1702.03427, 2017. GENUS TWO GROMOV–WITTEN INVARIANTS OF QUINTIC THREEFOLDS 43 [12] E. Clader, F. Janda, and Y. Ruan. Higher-genus wall-crossing in the gauged linear sigma model. arXiv:1706.05038, 2017. [13] T. Coates and A. Givental. Quantum Riemann-Roch, Lefschetz and Serre. Ann. of Math. (2), 165(1):15– 53, 2007. [14] B. Dubrovin. Geometry of 2D topological filed theories. In Integrable Systems and Quantum Groups, volume 1620 of Springer Lecture Notes in Math., pages 120–348. 1996. [15] H. Fan, T. Jarvis, and Y. Ruan. The Witten equation, mirror symmetry, and quantum singularity theory. Ann. of Math. (2), 178(1):1–106, 2013. [16] H. Fan, T. Jarvis, and Y. Ruan. A mathematical theory of the gauged linear sigma model, 2015. [17] A. Givental. A mirror theorem for toric complete intersections. In Topological field theory, primitive forms and related topics (Kyoto, 1996), volume 160 of Progr. Math., pages 141–175. Birkhauser Boston, Boston, MA, 1998. [18] A. B. Givental. Equivariant Gromov-Witten invariants. Internat. Math. Res. Notices, (13):613–663, 1996. [19] A. B. Givental. Elliptic Gromov-Witten invariants and the generalized mirror conjecture. In Integrable Systems and Algebraic Geometry, pages 107–155. World Sci. Publ., River Edge, NJ, 1998. [20] A. B. Givental. Gromov-Witten invariants and quantization of quadratic Hamiltonians. Mosc. Math. J., 1(4):551–568, 645, 2001. Dedicated to the memory of I. G. Petrovskii on the occasion of his 100th anniversary. [21] A. B. Givental. Semisimple Frobenius structures at higher genus. Internat. Math. Res. Notices, (23):1265–1286, 2001. [22] T. Graber and R. Pandharipande. Localization of virtual classes. Invent. Math., 135(2):487–518, 1999. [23] S. Guo, F. Janda, and Y. Ruan. (in preparation). [24] S. Guo and D. Ross. Genus-one mirror symmetry in the Landau–Ginzburg model. arXiv:1611.08876, 2016. [25] S. Guo and D. Ross. The genus-one global mirror theorem for the quintic threefold. arXiv:1703.06955, 2017. [26] K. Hori. Duality in two-dimensional (2, 2) supersymmetric non-Abelian gauge theories. J. High Energy Phys., (10):2013:121, front matter+74, 2013. [27] M.-x. Huang, A. Klemm, and S. Quackenbush. Topological string theory on compact Calabi-Yau: mod- ularity and boundary conditions. In Homological mirror symmetry, volume 757 of Lecture Notes in Phys., pages 45–102. Springer, Berlin, 2009. [28] H. Iritani, T. Milanov, Y. Ruan, and Y. Shen. Gromov-witten theory of quotient of fermat calabi-yau varieties. arXiv:1605.08885, 2016. [29] S. Katz, A. Klemm, and C. Vafa. M-theory, topological strings and spinning black holes. Adv. Theor. Math. Phys., 3(5):1445–1537, 1999. [30] Y.-H. Kiem and J. Li. Localizing virtual cycles by cosections. J. Amer. Math. Soc., 26(4):1025–1050, 2013. [31] M. Kontsevich and Y. Manin. Gromov-Witten classes, quantum cohomology, and enumerative geometry. In Mirror symmetry, II, volume 1 of AMS/IP Stud. Adv. Math., pages 607–653. Amer. Math. Soc., Providence, RI, 1997. [32] M. Krawitz and Y. Shen. Landau-Ginzburg/Calabi-Yau correspondence of all genera for elliptic orbifold P1. arXiv:1106.6270, 2011. [33] B. H. Lian, K. Liu, and S.-T. Yau. Mirror principle. I. Asian J. Math., 1(4):729–763, 1997. [34] A. Marian, D. Oprea, and R. Pandharipande. The moduli space of stable quotients. Geom. Topol., 15(3):1651–1706, 2011. [35] D. Maulik and R. Pandharipande. A topological view of Gromov-Witten theory. Topology, 45(5):887– 918, 2006. [36] T. Milanov, Y. Ruan, and Y. Shen. Gromov-witten theory and cycle-valued modular forms. arXiv:1206.3879, 2012. [37] R. Pandharipande. Three questions in Gromov-Witten theory. In Proceedings of the International Con- gress of Mathematicians, Vol. II (Beijing, 2002), pages 503–512. Higher Ed. Press, Beijing, 2002. [38] R. Pandharipande, A. Pixton, and D. Zvonkine. Relations on M g,n via 3-spin structures. J. Amer. Math. Soc., 28(1):279–309, 2015. 44 S. GUO, F. JANDA, AND Y. RUAN [39] A. Polishchuk. Moduli spaces of curves with effective r-spin structures. In Gromov-Witten theory of spin curves and orbifolds, volume 403 of Contemp. Math., pages 1–20. Amer. Math. Soc., Providence, RI, 2006. [40] C. Teleman. The structure of 2D semi-simple field theories. Invent. Math., 188(3):525–588, 2012. [41] E. Witten. Phases of N = 2 theories in two dimensions. Nuclear Phys. B, 403(1-2):159–222, 1993. [42] S. Yamaguchi and S.-T. Yau. Topological string partition functions as polynomials. J. High Energy Phys., (7):047, 20, 2004. [43] D. Zagier and A. Zinger. Some properties of hypergeometric series associated with mirror symmetry. In Modular forms and string duality, volume 54 of Fields Inst. Commun., pages 163–177. Amer. Math. Soc., Providence, RI, 2008. [44] Y. Zhou. Higher-genus wall-crossing in Landau–Ginzburg theory, 2017. arXiv:1706.05109. [45] A. Zinger. The reduced genus 1 Gromov-Witten invariants of Calabi-Yau hypersurfaces. J. Amer. Math. Soc., 22(3):691–737, 2009. School of Mathematical Sciences and Beijing International Center for Mathematical Research, Peking University, Beijing, China E-mail address: [email protected] Department of Mathematics, University of Michigan, 2074 East Hall, 530 Church Street, Ann Arbor, MI 48109, USA E-mail address: [email protected] Department of Mathematics, University of Michigan, 2074 East Hall, 530 Church Street, Ann Arbor, MI 48109, USA E-mail address: [email protected]
1211.2450
1
1211
2012-11-11T19:19:26
Euler characteristics of universal cotangent line bundles on $\mbar_{1,n}$
[ "math.AG" ]
We give an effective algorithm to compute the Euler characteristics $\chi(\mbar_{1,n}, \otimes_{i=1}^n L_i^{d_i})$. In addition, we give a simple proof of Pandharipande's vanishing theorem $H^j (\mbar_{0,n}, \otimes_{i=1}^n L_i^{d_i})=0$ for $j \ge 1, d_i \ge0$.
math.AG
math
EULER CHARACTERISTICS OF UNIVERSAL COTANGENT LINE BUNDLES ON M1,n Y.P. LEE AND F. QU Abstract. We give an effective algorithm to compute the Euler characteris- tics χ(M1,n, ⊗n i ). This work is a sequel to [8]. i=1Ldi In addition, we give a simple proof of Pandharipande's vanishing theorem H j (M0,n, ⊗n i=1Ldi i ) = 0 for j ≥ 1, di ≥ 0. 0. Introduction Let M1,n be the moduli stack of n-pointed genus 1 stable curves, O its structure sheaf, H the Hodge bundle, and Li the universal cotangent line bundles at the i-th marked point, 1 ≤ i ≤ n. The main result of this paper is the following theorem. Theorem 0.1. There is an effective algorithm of computing the Euler character- istics χd,d1,...,dn := χ(M1,n,H−d ⊗ Ldi i ), d, di ≥ 0. nOi=1 The details of this algorithm are presented in Section 2. This work is a sequel to [8], where we calculated the Euler characteristics χ(M0,n,⊗iLdi i ) at genus zero. These are our preliminary attempts in search of a K-theoretic ver- sion of Witten–Kontsevich's theory of two-dimensional topological gravity. In the Witten-Kontsevich theory, the correlators are the intersection numbers of tautolog- ical classes on the moduli spaces of stable curves (0.1) ZMg,n ψd1 1 . . . ψdn n . The natural K-theoretic version of intersection numbers (i.e. pushforward to a point in cohomology theory) are the Euler characteristics (i.e. pushforward to a point in K-theory). As Witten–Kontsevich theory states that a generating function of (0.1) is the τ -function of the KdV hierarchy, it is reasonable to surmise that a similar generating function in K-theory could be a τ -function of a version of a discrete KdV hierarchy. Note that the phenomenon of replacing differential equations in quantum cohomology by finite difference equations in quantum K-theory were observed in earlier examples [3] and only very recently demonstrated to hold in general [4]. 2010 Mathematics Subject Classification. Primary 14H10; Secondary 14J15, 14D23, 14D22, 14H15. This project was partially supported by the NSF. 1 2 Y.P. LEE AND F. QU Furthermore, since Witten–Kontsevich theory is the Gromov–Witten theory for the target space being a point, it is natural to consider these calculations as basic ingredients for the quantum K-theory developed in [2] and [9]. In the calculation of quantum K-invariants at genus one via localization, the Hodge bundle will appear naturally. It is therefore reasonable to consider slightly more general correlators, which have the additional benefits of facilitating the induction process of our algo- rithm. Our strategy of proving Theorem 0.1 is to apply the orbifold Riemann-Roch theorem to χ′ := χ M1,n,H−d i − O)! , (Ldi d, di ≥ 0. nOi=1 We were able to determine χ′ by carefully examining and performing computa- tions on the twisted sectors of M1,n. In doing so, we find the use of generating functions essential. These functions can be found in Section 2, starting with Equa- It is then not difficult to see that one can determine χ on M1,n by tion (2.1). χ′ and χ on M1,n−1. Hence, we can reduce all calculations to M1,1, whose gen- erating function is calculated explicitly in Lemma 2.8 and Proposition 2.9. Note that when n = 1 the generating function is a rational function, as in the case of genus zero case. We expect the generating function of χd,d1,...,dn to be rational as well, but are not able to find the correct closed form. We did, however, perform a consistency check: Our algorithm produces χd,d1,...,dn as integers, even though the intermediate steps require rational numbers, which arise as a consequence of the orbifold Riemann–Roch formula and the stack structure of M1,n. Indeed, the n = 1 case is closely related to the theory of modular forms. Theo- rem 0.1 can be considered as a generalization of the following well-known fact. Proposition 0.2 (See Lemma 2.8 and Proposition 2.9). χ(cid:18)M1,1, 1 1 − qL1(cid:19) = 1 (1 − q4)(1 − q6) . Since M1,1 is the moduli stack of elliptic curves, and sections of Lk 1 are the modular forms of weight 2k, Proposition 0.2 can be considered as a rephrase of the classical result that the space of the modular forms is generated by a weight four and a weight six modular forms. Another result included in this paper, in Appendix A, is a new proof of Pand- haripande's vanishing theorem [11] at genus zero. Theorem 0.3 ([11]). for j ≥ 1 and di ≥ 0. H j(M0,n,⊗n i=1Ldi i ) = 0 Our proof is comparably simpler and shorter, and does not use M. Kapranov's results on M0,n [5]. Only basic definitions and elementary manipulation of spectral sequences are used. This paper is organized as follows. In Section 1 we recall the necessary back- ground. We then formulate a more precise version of the reduction algorithm in EULER CHARACTERISTICS OF COTANGENT LINE BUNDLES 3 Section 2, and prove Theorem 0.1 there. In Appendix A the (new) proof of Theo- rem 0.3 is given. We work over the ground field C. 1. Preliminaries 1.1. Twisted sectors of M1,n. We summarize the results we need concerning the inertia stack of M1,n in [10, section 3.b]. For a DM stack X , recall a Spec C point of its inertia stack IX is given by a pair (x, g) with x ∈ X (Spec C) and g ∈ AutX (x). X is naturally viewed as a component of IX consists of point (x, g) with g trivial, and we denote this component by (X , Id). A twisted sector is a connected component of IX disjoint from (X , Id). Theorem 1.1 ([10] Theorem 3.22, 3.24). The twisted sectors of IM1,n come from either of the following two sources: (1) the closure of a twisted sector of IM1,n in IM1,n, (2) a twisted sector of I(M0,K∪• × M1,K c∪•) via I∆K. Here ∆K : M0,K∪• × M1,K c∪• → M1,n is the closed immersion gluing the marked points •, K is a subset of [n] with K ≥ 2, and K c its complement. I∆K : I(M0,K∪• × M1,K c∪•) → IM1,n is the induced closed immersion between the cor- responding inertia stacks. built up from type (1) twisted sectors. As I(M0,K∪• × M1,K c∪•) ≃ M0,K∪• × IM1,K c∪•, type (2) twisted sectors are The analysis of type (1) twisted sectors in [10] starts with the following descrip- tion of M1,1. Proposition 1.2. M1,1 is equivalent to the weighted projective space P(4, 6). We briefly recall the equivalence appeared in [10, Theorem 3.8], as we will need it to do explicit calculations on M1,1, and it also serves to motivate the notations used in [10](Notation 3.9, 3.12; Definition 3.13, 3.16) that we follow. Let U = A2 − (0, 0) with C∗ action: λ · (a, b) = (λ4a, λ6b), where λ ∈ C∗, (a, b) ∈ U . The equivalence from P(4, 6) := [U/C∗] to M1,1 is induced from a C∗-equivariant family of 1 pointed genus one stable curves C → U . where C = {(a, b) × [x : y : z] ∈ U × P2 y2z = x3 + axz2 + bz3}, with the section the C∗ action is given by s : U → U × [0, 1, 0] ⊂ C, λ · ((a, b) × [x : y : z]) = (λ4a, λ6b) × [λ2x : λ3y : z]. Denote by • Ak: the component of IM1,k consisting of pairs (x, g) with g of order 2, here 1 ≤ k ≤ 4. • C4: the 1-pointed curve {[x : y : z] ∈ P2 y2z = x3 + xz2} with [0 : 1 : 0] marked. Its automorphism group is generated by i = √−1. • C6: 1-pointed curve {[x : y : z] ∈ P2 y2z = x3 + z3} with [0 : 1 : 0] marked. • C′ Its automorphism group is generated by ǫ = exp (2πi/6). 4: C4 with a 2nd marked point [0 : 0 : 1]. 4 Y.P. LEE AND F. QU • C′ • C′′ 6: C6 with a 2nd marked point [0 : 1 : 1]. 6 : C′ 6 with a 3rd marked point [0 : −1 : 1]. Theorem 1.3 ([10] Corollary 3.14). (1) IM1,5 = (M1,5, Id), n ≥ 5. (2) For k ≤ 4, the twisted sectors of IM1,k are of dimension 1 or 0. Ak is the only 1 dimensional twisted sector. Zero dimensional twisted sectors are of the form Bµr, and they are determined by • (C4, i), (C4,−i), (C6, ǫ), (C6, ǫ2), (C6, ǫ4), (C6, ǫ5) for M1,1. • (C′ • (C′′ 6, ǫ2), (C′ 4,−i), (C′ 6 , ǫ4) for M1,3. 6, ǫ4) for M1,2. 4, i), (C′ 6 , ǫ2), (C′′ Remark 1.4. Given x ∈ X (Spec C) and an order r element g ∈ AutX (x), the pair (x, g) determines a representable morphism from Bµr to X . (see [1] 3.2) As Bµr is proper, it is closed in IM1,k. Theorem 1.5 ([10] Corollary 3.11, lemma 3.17). Let Ak be the closure of Ak in IM1,k, then (1) • A1 is isomorphic to M1,1. • A2 ⊂ IM1,2 is isomorphic to P(2, 4). • A3 ⊂ IM1,3 is isomorphic to P(2, 2). • A4 ⊂ IM1,4 is isomorphic to P(2, 2). (2) When viewed as a closed substack of M1,k , Ak does not intersect with the boundary divisors ∆K for any K ⊂ [k], here 2 ≤ k ≤ 4. 1.2. Riemann-Roch formula for Stacks. We recall the Riemann-Roch formula in a version needed for this paper, adopted from Appendix A of [13]. Theorem 1.6 ([6],[12] Corollary 4.13). Let X be a smooth, proper Deligne-Mumford stack with quasi-projective coarse moduli space, E a vector bundle on X . Assume X has the resolution property, i.e. every coherent sheaf is a quotient of a vector bundle, then we have the following formula for the Euler characteristics of E : χ(X , E) =ZIX fCh(E)fT d(X ). Here X E). T d(IX ) Ch(ρ◦λ−1(N ∨ of F with eigenvalue ζ. • IX is the inertial stack of X , with projection pX : IX → X . IX /X )) , where T d and Ch are the usual Todd class and Chern character. NIX /X is the normal bundle for p, and N ∨ is the dual of N . • fCh(E) ∈ H ∗(IX ) is the Chern character of the bundle ρ(p∗ • ρ(F ) :=Pζ ζF (ζ) ∈ K 0(IX ), if F = ⊕F (ζ) with F (ζ) being the eigenbundle • fT d(X ) = • λ−1(V ) :=Pa≥0(−1)aΛaV is the λ−1 operation in K-theory. If V = ⊕iVi is direct sum of line bundles Vi, then λ−1(V ) =Qi(1 − Vi). Remark 1.7. M1,n satisfies the resolution property. See, e.g. [7] Proposition 5.1. Remark 1.8. For a vector bundle F over Y and a map f : X → Y inducing If : IX → IY, it is easy to see ρ(p∗ Y F )) in K 0(IX ). X f ∗(F )) = If ∗(ρ(p∗ EULER CHARACTERISTICS OF COTANGENT LINE BUNDLES 5 1.3. String Equation. Proposition 1.9 ([9] Section 4.4). Let π : Mg,n → Mg,n−1 be the forgetful map forgetting the n-th marked point, then in terms of generating functions with variables q, qi, 1 ≤ i < n, we have, when g = 0, 1 1 − qiLi 1 − qiLi · (1 +Xi<n qi 1 − qi ), and when g > 0, π∗(Yi<n 1 − qH−1 Yi<n 1 π∗( 1 ) = 1 − qiLi 1 ) =Yi<n 1 − qH−1 Yi<n 1 Here π∗ is the K-theoretic pushforward. 1 1 − qiLi · (1 − H−1 +Xi<n qi 1 − qi ). 2. Euler characteristics of universal cotangent line bundles In this section, we give an algorithm to compute It is more efficient to encode these numbers into a generating function χ M1,n,H−d Xn :=χ M1,n, nYi=1 = Xd,di≥0 qd nOi=1 1 Ldi i ! , d, di ≥ 0. 1 − qiLi! nOi=1 nYi=1 1 1 − qH−1 qdi i χ M1,n,H−d i ! . Ldi (2.1) (2.2) We will first show that the calculation of Xn can be reduced to that of Xn−1 if another generating function Φn in (2.2) can be calculated. We then explicitly determine Φn and X1. 2.1. Reduction from M1,n to M1,n−1. Let Φn be the generating function Φn := χ M1,n, 1 1 − qH−1 nYi=1(cid:18) 1 1 − qiLi − 1 1 − qiO(cid:19)! Lemma 2.1. When n > 1, Xn is determined by Φn and Xn−1. More precisely, we have Xn(q, q1,··· , qn) =Φn(q, q1,··· , qn) + XI⊂[n],I6=∅ (cid:16)Xn−1(q,{qj, j /∈ I}, 0,··· , 0 {z } {z } Xn−1(0,{qj, j /∈ I}, 0,··· , 0 I−1 I−1 1 q + )(cid:17). 1 (−1)I+1Yi∈I 1 − qi· +Xj /∈I )(1 − qj 1 − qj 1 q ) For the last line, note that Xn−1(q, q1,··· , qn−1) is a symmetric function of the variables q1, q2,··· , qn−1, and it is evaluated at {qj, j /∈ I} and I − 1 zeros. 6 Y.P. LEE AND F. QU Proof. This follows directly from the definition and the string equation. Expand By the string equation 1 i=1( the productQn Φn = Xn + XI⊂[n],I6=∅ χM1,n, = χM1,n−1, 1−qiLi − 1 (−1)IYi∈I 1−qiO ) in Φn we have 1 1 1 − qH−1Yj /∈I 1 − qi · χM1,n, 1 − qjLj 1 − qjLj · (1 − H−1 +Xj /∈I 1 − qH−1Yj /∈I 1 1 1 1 1 − qH−1Yj /∈I qj 1 − qj ) , 1 1 − qjLj . which is Xn−1(q,{qj, j /∈ I}, 0,··· , 0 )(1 − + 1 q Xn−1(0,{qj, j /∈ I}, 0,··· , 0 ). I−1 {z } {z } I−1 1 q +Xj /∈I qj 1 − qj ) (cid:3) From here it is easy to see the lemma holds. 2.2. Calculation of Φn. By the Riemann-Roch formula (Theorem 1.6), we have Φn =ZIM1,n fCh( 1 1 − qH−1 ( nYi=1 1 1 − qiLi − 1 1 − qiO ))fT d(M1,n). As the inertia stack IM1,n is the disjoint union of the distinguished component (M1,n, Id) and its twisted sectors, the integral is the sum of the contributions from these components. Proposition 2.2. The contribution to Φn from (M1,n, Id) is (n − 1)! 24(1 − q) nYi=1 qi (1 − qi)2 . Proof. On (M1,n, Id), fCh and fT d are the same as Ch and T d respectively, 1 Ch( 1 − qH−1 qi = 1 1 − q nYi=1 (1 − qi)2 Applying the dilaton equation ( nYi=1 1 1 − qiLi − nYi=1 1 ) 1 − qiO c1(Li) + higher degree terms. ZM1,n c1(L1)··· c1(Ln) = (n − 1)ZM1,n−1 c1(L1)··· c1(Ln−1), and we find that = EULER CHARACTERISTICS OF COTANGENT LINE BUNDLES 7 ZM1,1 c1(L1) = 1 24 , 1 1 − qiLi − 1 1 − qiO )! T d(M1,n) ( ZM1,n (n − 1)! 24(1 − q) Ch nYi=1 1 nYi=1 1 − qH−1 (1 − qi)2 . qi (cid:3) Proposition 2.3. The contribution to Φn from a twisted sector of type (2) in Theorem 1.1, i.e. from I∆K , is zero. Proof. Recall such a twisted sector is the product of M0,K∪• and a twisted sector I of M1,K c∪•. The natural map M0,K∪• × I → M1,n factors through ∆K : M0,K∪• × M1,K c∪• → M1,n. We quote some known results : • The dual of the normal bundle for ∆K is pr∗ • ∆∗ • ∆∗ 1(L•) ⊗ pr∗ the projection of M0,K∪• × M1,K c∪• onto its i-th factor. K (Li) is pr∗ K (H) = pr∗ 1(Li) for i ∈ K, and is pr∗ 2(H). 2(Li) for i /∈ K. 2(L•). Here pri is Using these results, it is then straightforward to see that pushing forward the inte- grand 1 1 − qH−1 ( nYi=1 fCh( 1 1 − qiLi − 1 1 − qiO ))fT d(M1,n) 1 to M0,K∪• gives us a class which has a factorQi∈K c1(Li) coming from fCh(Qi∈K ( 1−qiO )). As the degree ofQi∈K c1(Li) exceeds the dimension of M0,K∪•, the con- tribution is zero. (cid:3) 1 1−qiLi− Proposition 2.4. For 2 ≤ k ≤ 4, the contribution from Ak is (−1)k 1 24(1 + q) kYi=1 qi i · (11 + 1 − q2 2q 1 + q − nXi=1 2qi 1 + qi ) · dk, where dk is 6, 6, 3 for k = 4, 3, 2, respectively. The number dk is the degree of a maps Ak → M1,1 forgetting all but one marked point. 8 Y.P. LEE AND F. QU Proof. On Ak, we have 1 )) 1 ( 1 − qiLi − 1 + qiec1(Li) − 1 − qiO 1 1 ( ) 1 − qi 1 − qH−1 1 1 fCh( = 1 + qec1(H−1) = (−2)k 1 1 + q kYi=1 kYi=1 kYi=1 qi 1 − q2 i(cid:16)1 + q 1 + q c1(H) + kXi=1 1 − qi 2(1 + qi) c1(Li)(cid:17), and Ch(cid:0)ρ ◦ (Λ−1(N ∨ Ak/M1,k )(cid:1)(cid:1) = 2k−1(cid:0)1 + 1 2 c1(N ∨ Ak/M1,k )(cid:1). For the above equations, note that over Ak the eigenvalues involved in fCh must be −1 as the nontrivial automorphism is of order 2, also there is no higher degree terms as Ak is 1 dimensional. Thus ( 1 1 1 qi 1 + q 1 − qH−1 1 kYi=1 1 − qiLi − i · 1 − q2 ZAk fCh( kYi=1 = (−1)k 1 + qZAk 1 + qiZAk ·(cid:16) 2q ZAk c1(H) = dkZM1,1 c1(Li), 1 ≤ i ≤ k, and ZAk ZAk kXi=1 1 − qi c1(H) + It is easy to see 1 − qiO ))fT d(M1,k)) c1(Li) +ZAk c1(TM1,k)(cid:17). c1(H) = dk 24 , by considering a map Ak ⊂ M1,k → M1,1 forgetting all but one marked point, are determined by Corollary 2.6. c1(TM1,k). (cid:3) Lemma 2.5. Let π : M1,n+1 → M1,n be the forgetful map forgetting the (n + 1)-th marked point, then c1(Lj) = π∗c1(Lj) + ∆{j,n+1}, 1 ≤ j ≤ n; c1(TM1,n+1) = π∗c1(TM1,n) − c1(Ln+1) + X1≤j≤n Ln+1 = ωπ(cid:0) X1≤j≤n Lj = π∗Lj(cid:0)∆{j,n+1}(cid:1), 1 ≤ j ≤ n; Proof. Recall for π : M1,n+1 → M1,n we have ∆{j,n+1}. ∆{j,n+1}(cid:1), EULER CHARACTERISTICS OF COTANGENT LINE BUNDLES 9 where ωπ = ωM1,n+1 ⊗ π∗ω −1 M1,n is the relative dualizing sheaf for π. Taking the first Chern class of these equations proves the lemma. (cid:3) Corollary 2.6. ZAk c1(Lj) = dk 24 , 1 ≤ j ≤ k. ZAk c1(TM1,k) = (11 − k)dk 24 . Proof. This can be easily proved by applying the above lemma and Theorem 1.5 (2) to a forgetful map Ak → M1,1. (cid:3) Proposition 2.7. Over the zero dimensional twisted sectors, the contribution to Φn are: • the contribution from (C′ 4, i) ⊔ (C′ 4,−i) is qj (1 − qj) · 1 − q + q1 + q2 − q1q2 + qq1 + qq2 + qq1q2 ; (1 + q2)(1 + q2 1)(1 + q2 2) 1 4 Yj=1,2 3 Yj=1,2 1 • the contribution from (C′ 6, ǫ2) ⊔ (C′ 6, ǫ4) is qj (1 − qj) · 1 − q + (q + 2)(q1 + q2) + (2q + 1)q1q2 (1 + q + q2)(1 + q1 + q2 1)(1 + q2 + q2 2) ; • the contribution from (C′′ 3 Yj=1,2,3 1 (1 − qj)· qj − 6 , ǫ2) ⊔ (C′′ 6 , ǫ4) is 1 − q + (q + 2)(q1 + q2 + q3) + (2q + 1)(q1q2 + q1q3 + q2q3) + (q − 1)q1q2q3 . (1 + q + q2)(1 + q1 + q2 1)(1 + q2 + q2 2)(1 + q3 + q2 3) Proof. To simplify the notation, we will use (C, λ) to denote a twisted sector. The integrand are determined by the eigenvalues of the bundles involved in the Riemann Roch formula. For (C4, λ), (C6, λ) of IM1,1, explicit calculation will show that the eigenvalues are λ, λ, λ2 respectively. Consider the forgetful map π : of L1, H and ΩM1,1 M1,k → M1,k−1, as Li = π∗Li, i < k and H = π∗H, remark 1.8 implies that the eigenvalues of Li or H is λ for (C, λ). As π is smooth, we have a short exact sequence 0 → π∗ΩM1,k−1 → ΩM1,k → ωπ → 0 , and ωπ can be identified with Lk. It is then easy to see that the eigenvalues of ΩM1,2 6, λ), and the eigenvalues of ΩM1,3 are λ, λ, λ2 for (C′′ are λ, λ2 for (C′ 4, λ) or (C′ 6 , λ). From the analysis above, on the twisted sector (C, λ) of M1,n we have 1 1 − qH−1 ( nYi=1 1 1 − qiLi − 1 1 − qiO )) = (λ − 1)n 1 − qλ−1 nYi=1 qi (1 − qi)(1 − qiλ) , fCh( and fT d(M1,n) = 1 (1 − λ2)(1 − λ)n−1 = 1 (1 + λ)(1 − λ)n . 10 Y.P. LEE AND F. QU The sum of the integral on (C′ 4, i) ⊔ (C4,−i) is then 1 qi 1 4 Xλ=i,−i (1 − qλ−1)(1 + λ) Yi=1,2 (1 − qi)(1 − qiλ) , which equals 1 4 Yj=1,2 qj (1 − qj) · 1 − q + q1 + q2 − q1q2 + qq1 + qq2 + qq1q2 . (1 + q2)(1 + q2 1)(1 + q2 2) The remaining cases also follow directly from our formula of fCh,fT d. 2.3. Calculation for X1. Under the isomorphism M1,1 ≃ P(4, 6), the line bundles H and L1 all correspond to O(1), hence χ(M1,1,H−d ⊗ Ld1 1 ) = χ(P(4, 6),O(d1 − d)), (cid:3) and X1 is determined by χ(P(4, 6),O(k)), k ∈ Z. Lemma 2.8. Let h0(O(k)) = dimC H 0(P(4, 6),O(k)), then , 1 h0(O(k))qk = (1 − q4)(1 − q6) ∞Xk=0 and h0(O(k)) = 0 if k < 0. Proof. As a section of O(k) on P(4, 6) corresponds to a polynomial f (x, y) satisfying f (λ4x, λ6y) = λkf (x, y) for any λ ∈ C∗, monomials xayb such that 4a + 6b = k form a basis of H 0(P(4, 6),O(k)). Therefore, h0(O(k)) is given by the coefficient of qk in the power series (cid:3) 1 . Proposition 2.9. (1 − q4)(1 − q6) (1 − qq1)(1 − q4 − q6 − q2 1q6 − q2 1 + q6q6 1 + q8q8 1) . χ(M1,1, 1 1 1 − qH−1 1q8 − q4 ) = 1 − q1L1 1 + q4q4 1q8 + q2q2 1)(1 − q6 1) ∨ (1 − q4)(1 − q6)(1 − q4 Proof. We have H 1(P(4, 6),O(k)) ≃ H 0(P(4, 6),O(−10 − k)) by Serre duality, so χ(P(4, 6),O(k)) = h0(O(k)) − h0(O(−k − 10)), and the proposition now follows from the previous lemma. Appendix A. A simple proof of Pandharipande's vanishing theorem The purpose of this appendix is to give a very simple and self-contained proof of Theorem 0.3, first proved in [11]. Recall that the theorem states that at genus zero (cid:3) (A.1) for j ≥ 1 and di ≥ 0. 1 is a point. H j(M0,n,⊗n i=1Ldi i ) = 0 We will prove (A.1) by induction on n. Note that (A.1) holds for n = 3 as M0,3 1The method presented in this appendix can also be used to compute H 0(M0,n, ⊗Ldi i ). It is also hoped that this method can help to produce an Sn-equivariant version of our genus zero formula [8], which is needed in the quantum K-theory [9] computation of general target spaces. EULER CHARACTERISTICS OF COTANGENT LINE BUNDLES 11 For n > 3, we first treat the special case that one of the di is zero, then reduce the case that all di > 0 to that special case. If one of the di is zero, up to permutation of the marked points we can assume dn = 0. Consider the forgetful map π : M0,n → M0,n−1, as R1π∗(⊗n−1 i ) = 0 by cohomology and base change (for C rational and degree of OC (D) positive, H 1(C,OC (D)) = 0), we have a degenerated Leray spectral sequence which gives i=1 Ldi i=1 Ldi H j(M0,n,⊗n−1 =H j(cid:16)M0,n−1, (⊗n−1 i ) = H j(M0,n−1, R0π∗(⊗n−1 )(cid:17), i ) ⊗ (O + Xi,di6=0 L−m i=1 Ldi diXm=1 i=1 Ldi i i )) and this is zero by induction. Here we used the string equation (Prop. 1.9) that K-theoretically If all di > 0, consider V := ⊕n i=1Ldi sections si of Li such that the zero locus of the section ⊕n the remark below.) Then the Koszul complex di. Choose i=1Li of V is empty.(See π∗(⊗n−1 i=1 Ldi i ) = (⊗n−1 i=1 Ldi ). i L−m diXm=1 i ) ⊗ (O + Xi,di6=0 i , note that Vn V is ⊗n 2^ V → ··· d→ n^ V → 0, i=1s⊗di i 0→O d→ V d→ Form the double complex (Cp(U ,Kq); δ, d){p≥0,0≤q≤n−1}, where U is an affine is exact, and we can compute H ∗(M0,n,⊗Li covering of M0,n, Kq = Vq V , Cp are the Cech cochain groups, δ is the Cech di) from this resolution. differential. q d d C 0(U , O(V )) d C 0(U , O) δ δ C 1(U , O(V )) d C 1(U , O) δ δ p Using two canonical filtrations (by p and q respectively), we obtain two spectral sequences ′Ep,q r and ′′Ep,q r with ′Ep,q 1 = H p(M0,n,Kq), 2 = H p(M0,n,Hq d(K∗)). ′′Ep,q These two spectral sequences abut to the same hyper-cohomology H∗(M0,n,K∗). 12 Y.P. LEE AND F. QU r i i ) = ′′Ep,n−1 2 By induction, ′Ep,q 2 = 0 if q 6= n − 1, therefore ′′Ep,q = Hp+n−1(M0,n,K∗) = 0 when p ≥ 1. 1 = 0 if p 6= 0, since Kq is the direct sum of ⊗Ld′ i 's with some degenerates at r = 1, and by our construction Hq(M0,n,K∗) = degenerates at r = 2, and we i = 0 . So ′Ep,q d′ ′E0,q 1 = 0 when q ≥ n. Note that ′′Ep,q have H p(M0,n,⊗Ldi Remark A.1. The zero locus of the section ⊕n of the section ⊕n−2 is an open property for sections, it is easy to show that a generic section ofLn−2 on M0,n has empty zero locus inductively using a forgetful map as follows. The statement holds for n = 3, 4 obviously. For n > 4, consider the forgetful map π : M0,n → M0,n−1. Since Li = π∗Li(Di), 1 ≤ i ≤ n − 2, where Di is the image of the i-th section of π, a section ti of Li on M0,n−1 would induce a section si of Li on M0,n with support Supp si = π−1(Supp ti) ∪ Di. It is straightforward to check ∩n−2 i=1,i6=j Supp ti = ∅, and these conditions hold for generic ti by induction. i=1s⊗di is contained in the zero locus i=1 Li. Since having empty zero locus i=1 Li i=1 si of the vector bundleLn−2 i=1 Supp si = ∅ iff for all 1 ≤ j ≤ n − 2,∩n−2 r i References [1] D. Abramovich, T. Graber, A. Vistoli, Gromov-Witten theory of Deligne-Mumford stacks, Amer. J. Math, 130 (2008),no.5, 1337-1398. [2] A. Givental, On the WDVV equation in quantum K-theory, Michigan Math. J. 48 (2000), 295-304. [3] A. Givental, Y.-P. Lee, Quantum K-theory on flag manifolds, finite-difference Toda lattices and quantum groups, Invent. Math. 151 (2003), no. 1, 193-219. [4] A. Givental, V. Tonita, The Hirzebruch–Riemann–Roch theorem in true genus-0 quantum K-theory, arXiv:1106.3136. [5] M. Kapranov, Veronese curves and Grothendieck-Knudsen moduli space M 0,n, J. Algebraic Geom. 2 (1993), no. 2, 239-262. [6] T. Kawasaki, The Riemann-Roch theorem for complex V-manifolds, Osaka J. Math. 16 (1979) 151-159. [7] A. Kresch, On the geometry of Deligne-Mumford stacks. Algebraic geometry: Seattle 2005. Providence, Rhode Island, 259-271. [8] Y.-P. Lee, A formula for Euler Characteristics of Tautological on dles http://www.math.utah.edu/%7Eyplee/research/imrn.pdf. the Deligne-Mumford Moduli Spaces, IMRN, 1997, No.8, line Bun- 393-400. [9] Y.-P. Lee, Quantum K-theory I: foundations, Duke Math. J. 121 (2004), no. 3, 389-424. [10] N. Pagani, Chen-Ruan Cohomology of M1,n and M1,n , arXiv:0810.2744v3, to appear in Ann. Inst. Fourier (Grenoble) 63 (2013). [11] R. Pandharipande, The symmetric function h0(M0,n, Lx1 1 ⊗ Lx2 2 ⊗ · · · Lxn n ), J. Alg. Geom. 6 (1997), no. 4, 721-731. [12] B. Toen, Th´eor`emes de Riemann-Roch pour les champs de Deligne-Mumford, K-Theory 18 (1999), no. 1, 33-76. [13] H.-H. Tseng, Orbifold Quantum Riemann-Roch, Lefschetz and Serre, Geom. Topol. 14 (2010) 1-81. E-mail address: [email protected] E-mail address: [email protected] Department of Mathematics, University of Utah, Salt Lake City, Utah 84112-0090, U.S.A.
1501.01612
2
1501
2016-09-11T14:45:55
A quick proof of nonvanishing for asymptotic syzygies
[ "math.AG", "math.AC" ]
We give a quick new approach to the main cases of the nonvanishing theorems of first and third authors concerning the asymptotic behavior of the syzygies of a projective variety as the positivity of the embedding line bundle grows. Specifically, we present a surprisingly elementary and concrete proof of the asymptotic nonvanishing of Veronese syzygies, and we obtain effective results for arithmetically Cohen-Macaulay varieties. The idea is that one can reduce the statements to some simple computations with monomials.
math.AG
math
A QUICK PROOF OF NONVANISHING FOR ASYMPTOTIC SYZYGIES LAWRENCE EIN, DANIEL ERMAN, AND ROBERT LAZARSFELD Introduction The purpose of this note is to give a very quick new approach to the main cases of the nonvanishing theorems of [5] concerning the asymptotic behavior of the syzygies of a projective variety as the positivity of the embedding line bundle grows. In particular, we present a surprisingly elementary and concrete approach to the asymptotic nonvanishing of Veronese syzygies, and we obtain effective statements for arithmetically Cohen-Macaulay varieties. Let X be an irreducible projective variety of dimension n over an algebraically closed field k, and let L be a very ample divisor on X, defining an embedding X ⊆ PH 0(L) = Pr. Write S = Sym H 0(L) for the homogeneous coordinate ring of Pr, and for a fixed divisor B on X consider the S-module M = M(B; L) = Mm H 0(B + mL). We are interested in the minimal graded free resolution E• = E•(B; L) of M over S: 0 / Er / . . . / E1 / E0 / M / 0, with Ep = ⊕S(−ap,j). Denote by Kp,q(B; L) = Kp,q(X, B; L) the finite dimensional vector space of degree p + q minimal generators of the pth module of syzygies of M, so that Ep(B; L) = Mq Kp,q(B; L) ⊗k S(−p − q). (When B = OX , we write simply Kp,q(X; L) or Kp,q(L) if no confusion seems likely.) It is elementary that if L is very positive compared to B then non-zero syzygies can only appear in weights 0 ≤ q ≤ n + 1, and it turns out that the extremal cases q = 0 and q = n + 1 are easy to control. So the first interesting question is to fix B and 1 ≤ q ≤ n, and to ask which groups Kp,q(B; L) are nonvanishing when L becomes very positive. The main result of [5] asserts in effect that – contrary to what one might have expected by extrapolating from the case of curves – these groups are eventually non-zero for almost all values of p ∈ [1, r]. Research of the first author partially supported by NSF grant DMS-1001336. Research of the second author partially supported by NSF grant DMS-1302057. Research of the third author partially supported by NSF grant DMS-1439285. 1 / / / / / / Perhaps the most natural instance of these matters occurs when X = Pn, B = OPn(b) and L = Ld = OPn(d), so that one is looking at the syzygies of Veronese varieties. It was established in [5] that if one fixes q ∈ [1, n] and b ≥ 0, then for d ≫ 0 one has (1) Kp,q(Pn, B; Ld) 6= 0 for every value of p satisfying (2) (cid:18)d + q q (cid:19) − (cid:18)d − b − 1 q (cid:19) − q ≤ p ≤ (cid:18)d + n n (cid:19) − (cid:18)d + n − q n − q (cid:19) + (cid:18)n + b n − q(cid:19) − q − 1. For example, when n = 2 and b = 0, this asserts that (3) Kp,2(P2; OP2(d)) 6= 0 for 3d − 2 ≤ p ≤ (cid:18)d + 2 2 (cid:19) − 3, which was the main result of the interesting paper [8] of Ottaviani and Paoletti. The proof in [5] of the Veronese nonvanishing theorem involved a rather elaborate induction on n to show that certain well-chosen secant planes to the Veronese variety force the presence of non-zero syzygies. For b = 0 the same statement was obtained independently in characteristic zero by Weyman, who identified certain representations of SL(n + 1) that appear non-trivially in the Kp,q. Some other work concerning Veronese syzygies appears in [10], [1], [2], and [6], and a simplicial analogue of the results of [5] is given in [3]. The goal of the present paper is to present a much simpler and more elementary ap- proach to the nonvanishing of Veronese syzygies, and to use this method to establish effective statements for arithmetically Cohen-Macaulay varieties. The idea is that one can reduce the question to elementary computations with monomials by modding out by a suitable regular sequence. In order to explain how this goes, consider the problem of proving the first case of the Ottaviani-Paoletti statement (3), namely that if d ≥ 3 then (*) K3d−2,2(P2; OP2(d)) 6= 0. Writing Sk for the degree k piece of the polynomial ring S = k[x, y, z], it is well-known that the group in question can be computed as the cohomology at the middle term of the Koszul-type complex ... −→ Λ3d−1Sd ⊗ Sd −→ Λ3d−2Sd ⊗ S2d −→ Λ3d−3Sd ⊗ S3d −→ ... The most naive approach to (*) would be to write down explicitly a cocycle representing a non-zero element in K3d−2,2, but we do not know how to do this.1 On the other hand, consider the ring S = S/(xd, yd, zd). As xd, yd, zd form a regular sequence in S, the dimensions of the Koszul cohomolgy groups of S are the same as those of S, and hence the question is equivalent to proving the nonvanishing of the cohomology of (**) ... −→ Λ3d−1Sd ⊗ Sd −→ Λ3d−2Sd ⊗ S2d −→ Λ3d−3Sd ⊗ S3d −→ ... . Now view S as the ring spanned by monomials in which no variable appears with exponent ≥ d, with multiplication governed by the vanishing of the dth powers of each variable. The plentiful presence of zero-divisors in S means that one can write down by hand many monomial 1The argument in [8] proceeds by using duality to reformulate the question as the nonvanishing of a Kp ′ ,0, where one can exhibit directly the required class. 2 Koszul cycles: for instance if m1, . . . , m3d−2 are monomials of degree d each divisible by x or y, then c = m1 ∧ . . . ∧ m3d−2 ⊗ xd−1yd−1z2 gives a cycle for the complex (**). Note next that xd−1yd−1z2 has exactly 3d − 2 monomial divisors of degree d with exponents ≤ d − 1, viz: xd−1y , xd−2y2 , . . . , , x2yd−2 , xyd−1 xd−1z , xd−2yz , . . . , xyd−2z , yd−1z xd−2z2 , xd−3yz2 , . . . , xyd−3z2 , yd−2z2. Taking these as the mi, we claim that the resulting cycle c represents a non-zero Koszul cohomolgy class. In fact, suppose that c appears even as a term in the Koszul boundary of an element e = n0 ∧ n1 . . . ∧ n3d−2 ⊗ g, where the ni and g are monomials of degree d. After re-indexing and introducing a sign we can suppose that c = n1 ∧ . . . ∧ n3d−2 ⊗ n0g. Then the {nj} with j ≥ 1 must be a re-ordering of the monomials {mi} dividing xd−1yd−1z2. On the other hand n0g = xd−1yd−1z2, so n0 is also such a divisor. Therefore n0 coincides with one of n1, . . . , n3d−2, and hence e = 0, a contradiction. We show that this sort of argument gives the nonvanishing of Veronese syzygies appearing in equation (2), as well as a few further cases that were conjectured in [5]. Moreover, we obtain a new statement that subsumes the previous statement and includes all values of b, q, and d (Theorem 2.1). More interestingly, whereas the results of [5] for varieties other than Pn were ineffective, we are able here to give effective statements for a large class of general varieties. Specifically, consider an arithmetically Cohen-Macaulay variety X ⊆ Pm of dimension n, and for d > 0, b ≥ 0 write Ld = OX (d) , B = OX(b). Put c(X) = min(cid:8)k H n(X, OX (k − n)) = 0(cid:9), the Castelnuovo-Mumford regularity of OX , and write rd = dim H 0(X, OX (d)) , r′ d = rd − (deg X)(n + 1). We prove: Theorem. Assume that q ∈ [1, n − 1], and fix d ≥ b + q + c(X) + 1. Then for every value of p satisfying Kp,q(X, B; Ld) 6= 0 deg(X)(q + b + 1)(cid:18)d + q − 1 q − 1 (cid:19) ≤ p ≤ r′ d − deg(X)(d − q − b)(cid:18)d + n − q − 1 n − q − 1 (cid:19). Analogous statements hold, with slightly different numbers, when q = 0 and q = n; see Theorem 3.1 below. We note that Zhou [11] has given effective results for adjoint-type (and in particular, for very positive) line bundles B on an arbitrary smooth complex projective 3 variety. present techniques: see Remark 3.7. It would be interesting to know whether one could recover his statement by the We wish to thank Xin Zhou for valuable discussions, and the referee for some suggestions which significantly streamlined the statement of Theorem 2.1. 2. Nonvanishing Results for Pn This section is devoted to the nonvanishing results for Veronese syzygies. Let k be any field, and consider the polynomial ring S = k[x0, . . . , xn]. Given d ≥ 1 we denote by S(d) ⊆ S the Veronese subring S(d) = Mj∈Z Sjd ⊆ S of S. For an S-module M, we write M(b)(d) for the S(d)-module Lj∈Z Mb+jd. Note that M(b)(d) is also naturally a Sym(Sd)-module. We denote by Kp,q(n, b; d) = K Sym(Sd) p,q (S(b)(d)) the Koszul cohomology group of S(b)(d), where S(b)(d) is considered as a Sym(Sd)-module. Thus Kp,q(n, b; d) is the cohomolgy of the Koszul-type complex . . . −→ Λp+1Sd ⊗ S(q−1)d+b −→ ΛpSd ⊗ Sqd+b −→ Λp−1Sd ⊗ S(q+1)d+b −→ . . . and Since Kp,q(n, b; d) = Kp,q(Pn, OPn(b); OPn(d)). we will always assume that 0 ≤ b ≤ d − 1. Kp,q(n, b; d) = Kp,q+1(n, b − d; d), The following result is more precise than those in [5], since in that paper, b was always fixed and d ≥ n + 1. Theorem 2.1. Fix any d, any b ∈ [0, d − 1] and any q ∈ [0, n + 1 − n+b the quotient and remainder of qd + b by d − 1. Then: d ]. Define m and r as for all p in the range Kp,q(n, b; d) 6= 0 (cid:18)m + d m (cid:19) −(cid:18)m + d − r − 1 m (cid:19) − m ≤ p ≤ (cid:18)n + d n (cid:19) +(cid:18)n − m + r n − m (cid:19) −(cid:18)n − m + d n − m (cid:19) − m − 1. When q /∈ [0, n + 1 − n+b other hand, if d ≥ n + b, then the non-vanishing holds for all 0 ≤ q ≤ n. d ], then Kp,q(n, b; d) is automatically zero; see Remark 2.6. On the For the proofs, the idea is to mod out by a regular sequence to arrive at a situation where we can work by hand with monomials. Specifically, by the technique of Artinian reduction, 4 we can compute syzygies after modding out by a linear regular sequence. Having fixed d > 0, we put S =def S/(xd 0, . . . , xd n). Slightly abusively, we view S as the graded ring spanned by monomials in the xi in which no variable appears with exponent ≥ d, with multiplication determined by the vanishing of the dth power of each variable. Since xd 0, . . . , xd n is a linear regular sequence in Sym Sd, modding out by these powers does not affect the Koszul cohomology groups. In other words: It thus suffices to compute this last group, which is the homology at the middle of (cid:0)S(b)(d) ⊗Sym(Sd) Sym(Sd)(cid:1) ∼= K Sym(Sd) / Vp−1 Sd ⊗ S(q+1)d+b. / Vp Sd ⊗ Sqd+b p,q ∂p (S(b)(d)). K Sym(Sd) p,q (S(b)(d)) ∼= K Sym(Sd) p,q (2.1) Vp+1 Sd ⊗ S(q−1)d+b ∂p+1 In particular, Kp,q(n, b; d) 6= 0 if and only if this complex has non-trivial homology, and we are therefore reduced to studying cycles and boundaries in (2.1). We start with some notation that will prove useful. Fix a finite set of elements P ⊆ S (which in practice will be a collection of monomials). Definition 2.2. We write ζ ∈ V P (or ζ ∈ Vs P ) if ζ = m1 ∧ · · · ∧ ms with mi ∈ P for all i. We write ζ = det P if ζ is the wedge product of all elements in P (in some fixed order). We say that a wedge product m1 ∧ · · · ∧ ms is a monomial if each mi is a monomial. The following lemma guarantees the existence of many non-zero monomial classes in the cohomology of (2.1). It systematizes the computations worked out for a special case in the Introduction. Lemma 2.3. Fix a nonzero monomial f ∈ Sqd+b, and denote by Zf , Df ⊆ Sd respectively the set of degree d monomials that annihilate or divide f . (i). If ζ ∈ Vp Zf , then ζ ⊗ f ∈ ker ∂p. (ii). Let ζ ∈ Vs Sd be any monomial such that such that det Df ∧ ζ ⊗ f is nonzero. Then (det Df ∧ ζ) ⊗ f /∈ im ∂(Df +s). Proof. For (i), write ζ = m1 ∧ · · · ∧ ms with mi ∈ Zf . Since mif = 0 ∈ S for all i = 1, . . . , s, the assertion is immediate. Turning to (ii), assume that Then there must be some index j and some monomial appearing in ξj ⊗ gj that maps to the ∂(cid:0)X ξj ⊗ gj(cid:1) = (cid:0) det Df ∧ ζ(cid:1) ⊗ f monomial (cid:0) det Df ∧ ζ(cid:1) ⊗ f . In particular, ξj ⊗ gj must contain a non-zero monomial of the 5 / / form (m ∧ det Df ∧ ζ) ⊗ g where mg = f . But then m ∈ Df and hence m ∧ det Df = 0, a contradiction. (cid:3) Corollary 2.4. Given q, d and b, let f ∈ Sqd+b be a monomial such that Df ⊆ Zf . Then any non-zero monomial of the form where ζ ∈ V Zf , represents a nonzero element of the cohomology of (2.1). In particular, Kp,q(n, b; d) 6= 0 (cid:0) det Df ∧ ζ(cid:1) ⊗ f, for every p satisfying Df ≤ p ≤ Zf . (cid:3) Remark 2.5. The Koszul classes just constructed are linearly independent. In fact, keeping the notation of the corollary, and with an appropriate degree twist, there is a natural map from the Koszul complex on the linear forms in Zf to the minimal free resolution of S(b)(d) over Sym Sd given by sending 1 7→ f . This induced map yields an inclusion of the Koszul subcomplex on the linear forms Zf \ Df ⊆ Sym(Sd) In Conjecture B from [4], we spanning homological degrees p = Df , Df + 1, . . . , Zf . conjectured that each row of the Betti table of a high degree Veronese looks roughly like the Betti table of a Koszul complex. Although this result has a similar flavor, the lower bound on the size of the Koszul cohomology groups constructed via this method is far too small to verify the conjecture of [4]. Theorems 2.1 now follows from Corollary 2.4 by choosing a convenient monomial f and computing the number of elements in the resulting sets Zf and Df . Proof of Theorem 2.1. Put sd = dim Sd = (cid:18)n + d d (cid:19) − (n + 1). Let f be the "leftmost" monomial of S having degree dq + b; by our definition of m and r this is the monomial of the form: In order to establish the theorem, it suffices to prove three assertions: f = xd−1 0 · xd−1 1 · . . . · xd−1 m−1 · xr m. (i). sd − Zf = (cid:0)d+n−m (ii). Df = (cid:0)m+d m (cid:1) −(cid:0)m+d−r−1 (iii). Df ⊆ Zf . d r (cid:1) −(cid:0)n+r−m (cid:1) − m. m (cid:1) − n + m − 2. For (i), observe that Zf = (0 :S f )d contains all monomials of degree d that are divisible by any of x0, . . . , xm−1 as well as those divisible by xr m. Hence among the sd monomials in Sd, the ones not lying in Zf are the monomials of degree d appearing in the quotient S/(x0, . . . , xm−1, xd−r m ). 6 We can compute this via the resolution: · · · −→ S(−d) (x0, . . . , xm−1) ·xr m−→ S(−d + r) (x0, . . . , xm−1) ·xd−r m−→ S (x0, . . . , xm−1) −→ S (x0, . . . , xm−1, xd−r m ) . Therefore sd − Zf = dimk(cid:0)S/(x0, . . . , xm−1, xd−r m )(cid:1)d = dim(S/(x0, . . . , xm−1))d − dim(S/(x0, . . . , xm−1))r + dim(S/(x0, . . . , xm−1))0 = (cid:18)(cid:18)d + n − m d (cid:19) − n + m − 1(cid:19) −(cid:18)n + r − m r (cid:19) + 1. For (ii), note that Df can be identified with the degree d monomials of S/(xr+1 A similar computation yields m , xm+1, . . . , xn). Df = dim(cid:0)S/(xr+1 m , xm+1, . . . , xn)(cid:1)d = dim(cid:0)S/(xm+1, . . . , xn)(cid:1)d − dim(cid:0)S/(xm+1, . . . , xn)(cid:1)d−r−1 + dim(cid:0)S/(xm+1, . . . , xn)(cid:1)0 = (cid:18)(cid:18)m + d d (cid:19) − m − 1(cid:19) −(cid:18)d − r + m − 1 (cid:19) + 1. m Finally, since the exponent of xm in f is r ≤ d−1, it follows that any element of Df is divisible at least by one of x0, . . . , xq−1, and hence any such element annihilates f . (cid:3) If, Remark 2.6. If q < 0 then since b ≥ 0, we clearly have Kp,q(n, b; d) = 0 for all p. q > n + 1 − n+b then we claim that we also get vanishing for all p. We define q′ := n + 1 − q d and note that we the above inequality on q is equivalent to having q′d < n + b. We then use duality to compute: Kp,q(n, b; d)∗ = Krd−n−p,q′(n, −n − 1 − b; d). Since O(−n − 1 − b + q′d) has no sections when q′d < n + 1 + b, our assumptions imply that this group equals 0 for all rd − n − p ≥ 0 and hence for all p ≥ 0. Remark 2.7. Zhou [12] has recently established some results about the asymptotic distribu- tion of torus weights appearing in the Kp,q of toric varieties. It would be interesting to know if the present arguments can be used to give more refined information in the case X = Pn. Remark 2.8. It is conjectured in [5, Conjecture 7.5] that for d ≥ n + 1, the assertion of Theorem 2.1 is optimal in the sense that the Kp,q in question vanish outside the stated range, and we conjecture that the more general bounds in Theorem 2.1 are optimal as well. For instance, in the case d = 2 and b = 0, the full resolution is known in characteristic 0 by work of J´ozefiak, Pragacz and Weyman in [7]. Their theorem shows that, as long as n + 1 ≥ 2q, Kp,q(n, 0; 2) = 0 starting with p = 2q2 − q, and this value lines up with the lower bound in Theorem 2.1. It would be exceedingly interesting to know whether one can use the approach introduced here to make progress on this conjecture, at least in the case d ≫ 0 as in [5, Problem 7.7] . Unfortunately it seems that one can't work only with monomials – it's possible for instance that a monomial Koszul cocyle appears as the boundary of non-monomial elements. It is tempting to wonder whether there are Grobner-like techniques that could be used to study 7 the issue systematically. We note that Raicu [9] has reduced the general vanishing conjecture [5, Conjecture 7.1] to the case of Veronese syzygies. 3. Nonvanishing for arithmetically Cohen-Macaulay schemes In this section we extend the results of the previous section to the setting of arithmetically Cohen-Macaulay schemes. Consider an arithmetically Cohen-Macaulay scheme X ⊆ Pm of dimension n over the field k, and let R = ⊕ H 0(X, OX (k)) be the homogeneous coordinate ring of X. Setting Ld = OX (d) and B = OX(b), we are interested in the syzygies Kp,q(X, B; Ld) = Kp,q(R(b)(d)) of B with respect to Ld for d ≫ 0. Put and write c = c(X) = min(cid:8)k H n(X, OX (k − n)) = 0(cid:9), rd = dim H 0(X, OX(d)) = dim Rd , r′ d = rd − (deg X)(n + 1). Our first result holds when d ≥ b + q + c + 1. Theorem 3.1. Fix b ∈ [0, d − q − 1 − c]. (i). If q ∈ [1, n − 1], then Kp,q(X, B; Ld) 6= 0 for (deg X)(q + b + 1)(cid:18)d + q − 1 q − 1 (cid:19) ≤ p ≤ r′ d − (deg X)(d − q − b)(cid:18)d + n − q − 1 n − q − 1 (cid:19). (ii). When q = n, one has Kp,n(X, B; Ld) 6= 0 when (deg X)(n + b + 1)(cid:18)d + n − 1 n − 1 (cid:19) ≤ p ≤ r′ d − deg X. (iii). When q = 0 one has Kp,0(X, B; Ld) 6= 0 when 0 ≤ p ≤ r′ n − 1 (cid:19). d − (d − b)(cid:18)n − 1 + d A somewhat more complicated but sharper statement appears in Remark 3.4 below. Remark 3.2. If we fix b and q, we can interpret these bounds as asymptotic statements as d → ∞. Under these assumptions, we are saying that Kp,q(X, B; Ld) 6= 0 for all p in the range deg(X)(q + b + 1) (q − 1)! dq−1 + O(dq−2) ≤ p ≤ r′ d −(cid:18) deg(X) (n − q − 1)! dn−q + O(dn−q−1)(cid:19) Conjecture 7.1 in [5] states that one should have Kp,q = 0 for p ≤ O(dq−1); it would be interesting to understand the optimal leading coefficients as well. In the ACM case this 8 implies that Kp,q = 0 also for p > rd − O(dn−q), but in the non-ACM case the groups in question can be nonvanishing for p ≈ rd [5, Remark 5.3]. For the proofs of the theorem, let IX ⊆ k[x0, . . . , xm] be the defining ideal of X, so that R = k[x0, . . . , xm]/IX . The statement is independent of the ground field, so we may assume that k is infinite. Then, after a general change of coordinates, we may assume that x0, . . . , xn form a system of parameters for R. To help clarify the following arguments, we will relabel the variables xn+1, . . . , xm as yn+1, . . . , ym. Let S = k[x0, . . . , xn] ⊆ R, which is a Noether normalization since x0, . . . , xn is a system of parameters. As R is Cohen-Macaulay of dimension n + 1, it follows that it is a maximal Cohen-Macaulay S-module, and hence a free S-module. We may choose a set Λ of monomials of the form yβ ∈ R which form a basis for R as an S-module, so that R = Myβ ∈Λ S · yβ. We assume that 1 ∈ Λ. Thus deg(X) = #Λ and we observe that c(X) = max{deg yβ}. Fix q ∈ [0, n], d > 0 and b ≥ 0. Set R = R/(xd 0, . . . , xd n), and define S as in the previous section. Thus R = R ⊗S S, and R is a free S-module with basis Λ. Since R is Cohen-Macaulay, we have dim Kp,q(R(b)(d)) = dim Kp,q(R(b)(d)) for all p and q, where the group on the right is computed as the cohomology of the complex (3.1) Vp+1 Rd ⊗ R(q−1)d+b ∂ / Vp Rd ⊗ Rqd+b ∂ / Vp−1 Rd ⊗ R(q+1)d+b. In the natural way, we can speak of monomials in R: these are (the images in R of) elements of the form xαyβ where yβ ∈ Λ, and the degree of such a monomial is α + β. Given a monomial f ∈ S, we denote by respectively the set of degree d monomials that annihilate f and the collection of degree d monomials of the form xαyβ where xα divides f and yβ ∈ Λ. Zf , Ef ⊆ Rd We start with an analogue of Lemma 2.3. Lemma 3.3. Let f ∈ Sqd+b ⊆ Rqd+b be a monomial such that Ef ⊆ Zf . Then any non-zero monomial element with ζ ∈ V Zf represents a non-zero Koszul cohomology class. In particular for every p with m = (cid:0) det Ef ∧ ζ(cid:1) ⊗ f Kp,q(cid:0)X, OX (b); OX (d)(cid:1) 6= 0 Ef ≤ p ≤ Zf . 9 / / It remains to prove that it is not Proof. Since Ef ⊆ Zf , m is evidently a Koszul cycle. cohomologous to zero. In fact, we'll show that m cannot occur as a monomial appearing in the expansion (with respect to the chosen basis of R) of ∂(ξ ⊗g) for any monomials ξ ∈ Λp+1Rd and g ∈ R(q−1)d+b. Suppose to the contrary that m appears as a term in ∂(ξ0 ∧ . . . ∧ ξp ⊗ g). Then after possibly reindexing and introducing a sign, we can suppose and that f appears as a term in the expansion of ξ0g in terms of the basis Λ. Suppose that ξ1 ∧ . . . ∧ ξp = det(Ef ) ∧ ζ, where yβ, yδ ∈ Λ. Then in R we can rewrite ξ0 = xαyβ , g = xγyδ yβ+δ = h0 · 1 + Xyλ∈Λ hλ · yλ where hλ ∈ Sβ+δ−λ. Therefore f = xα+γh0, and consequently xαyβ ∈ Ef . In particular ξ0 also appears as one of ξ1, . . . , ξp, and hence m = 0. (cid:3) We now turn to the Proof of Theorem 3.1. As before, we take f to be the be the leftmost nonzero monomial of S of degree dq + b: f = xd−1 0 · xd−1 1 · . . . · xd−1 q−1 · xq+b q . We claim first of all that Ef ⊆ Zf provided that d ≥ b + q + c + 1. In fact, suppose that w = xa0 0 · . . . · xaq q · yβ ∈ Ef . Then aq ≤ q + b, and hence a0 + . . . + aq−1 = d − aq − β ≥ d − (q + b) − c > 0. Therefore at least one of a0, . . . , aq−1 is strictly positive, and consequently w ∈ Zf . In order to apply Lemma 3.3, it remains to estimate the sizes of Ef and Zf . Writing rd = dim Rd, we start by giving an upper bound on rd − Zf . Assume first that q ∈ [1, n − 1], and consider a monomial xα = xa0 n . Then a degree d monomial xαyβ lies in the complement of Zf if and only if 0 · . . . · xan a0 = . . . = aq−1 = 0 , aq ≤ d − b − q − 1. The number of possibilities for xα is (rather wastefully) bounded above simply by the number of degree d monomials in k[xq+1, . . . , xn], times the number of choices for aq, times the number of choices for yβ. Since Λ = deg X, this leads to the lower bound rd − (deg X)(d − q − b)(cid:18)d + n − q − 1 n − q − 1 (cid:19) ≤ Zf . Turning to an upper bound on Ef , observe that xαyβ ∈ Ef if and only if a0, . . . , aq−1 ≤ d − 1 , aq ≤ q + b and aq+1 = · · · = an = 0 10 We can bound this (again wastefully) by the number of monomials of degree d in k[x0, . . . , xq−1], times the number of choices for aq, times the number of choices for yβ. This leads to: (deg X)(q + b + 1)(cid:18)q − 1 + d q − 1 (cid:19) ≥ Ef . So to obtain assertion (i) of Theorem 3.1, it remains only to observe that rd = Xyβ ∈Λ dim Sd−β ≥ Xyβ ∈Λ(cid:0)dim Sd−β − (n + 1)(cid:1) = dim Rd − Λ(n + 1) = r′ d. When q = n we get the same bound on Ef , but now we find that rd − (deg X) ≤ Zf , and this yields statement (ii) of the Theorem. Finally, when q = 0 we get the same lower bound on Zf as above, and we can obtain nonvanishing starting with p = 0. (cid:3) Remark 3.4. By separating the estimates into two terms depending on whether β is equal to zero or not, one gets a slightly better upper bound on the size of Ef , when q ∈ [1, n − 1]: (deg X − 1)(q + b + 1)(cid:18)q − 1 + d − 1 q − 1 (cid:19) +(cid:18)q + d q (cid:19) −(cid:18)d − b − 1 q (cid:19) − q ≥ Ef . In particular, this reduces to the statements obtained for Pn in the previous sections. Remark 3.5. By defining m and r as the quotient and remainder of dq + b by d − 1, one can use an argument to the proof of Theorem 2.1 to extend this to some additional values of q, d, and b. However, we felt the asymptotic behavior was more clear when phrased in terms of q and b instead of r and m. Remark 3.6. The bounds for Ef and rd − Zf appearing in the proof of Theorem 3.1 could be improved by a more precise count of the relevant possibilities, in particular taking into account the degrees of the yβ. This amounts to computations involving the numerator of the Hilbert series of R (i.e. the Hilbert function of the Artinian reduction R), and confronted with a specific example, it is often quite easy to use directly the method of the proof to get stronger statements. For example, let X ⊆ P5 be the hypersurface Then Λ = {1, x5, x2 5}, so c = 2. We take (q, b, d) = (3, 0, 8) and x3 0 + . . . + x3 5 = 0. Then R = k[x0, . . . , x5]/(x3 0 +. . .+x3 4). The bounds from Theorem 3.1 and Remark 2.5 yield the nonvanishing result Kp,3(cid:0)X; OX(8)(cid:1) 6= 0 for p between 540 and 1005. 0x7 f = x7 2x3 3. 5) and R = R/(x8 0, . . . , x8 1x7 11 However, if we follow the method of the proof, we can compute the size of Ef directly. Let A := k[x0, . . . , xq]/(xd 0, . . . , xd q−1, xq+b+1 q ) = k[x0, . . . , x3]/(x8 0, x8 1, x8 2, x4 3). Then dim Ad−deg yβ = dim A8 + dim A7 + dim A6 = 301. Xyβ ∈Λ A similar computation shows that there are 14 monomials in the complement of Zf and so Zf = 1030 − 14 = 1016, and the nonvanishing statement can be extended to all values of p between 301 and 1016. Remark 3.7. Let X ⊆ Pm be an arbitrary variety of dimension n, and suppose that B is a line or vector bundle on X with the property that H i(cid:0)X, B ⊗ OX(k)(cid:1) = 0 for all k ∈ Z and 0 < i < n: in other words, M = ⊕H 0(cid:0)X, B ⊗ OX(k)(cid:1) is a Cohen-Macaulay module over the homogeneous coordinate ring of Pm. Replacing B by a twist, one can assume without loss of generality that M−1 = 0 but M0 6= 0. Then one can use the methods of this section to obtain effective nonvanishing statements for the syzygies Kp,q(X, B; OX(d)). In fact, the hypotheses on M imply that it has a generator in degree 0, and then in the arguments above one can replace R by M. We leave details to the interested reader. It would be interesting to compare the resulting statements with the results [11] of Zhou which fall under this rubric. Finally, we expect that nonvanishing statements similar to Theorem 3.1 hold for any finitely generated, graded k-algebra R. More precisely, we conjecture the following analogue of part (i) of Theorem 3.1. Conjecture 3.8. Fix b and R and q ∈ [1, n] where n := dim R−1. Then there exist constants c and C such that if d ≫ 0 then Kp,q(R(b)(d)) 6= 0 for all cdq−1 ≤ p ≤ rd − Cdn−q and for all d ≫ 0. We expect similar analogues of parts (ii) and (iii) of Theorem 3.1, as well as analogues of the cases where b is close to d, as in Remark 3.5. References [1] Winfried Bruns, Aldo Conca and Tim Romer, Koszul homology and syzygies of Veronese subalgebras, Math. Ann. 351 (2011), 761–779. [2] Winfried Bruns, Aldo Conca and Tim Romer, Koszul cycles, in Combinatorial aspects of commutative algebra and algebraic geometry, Abel Symposium, Springer 2011, 17–33. [3] Aldo Conca, Martina Juhnke-Kubitzke and Wolkmar Welker, Asymptotic syzygies of Stanley-Reisner rings of iterated subdivisions, preprint. [4] Lawrence Ein, Daniel Erman and Robert Lazarsfeld, Asymptotics of random Betti tables, Journal fur die reine und angew. Math., 702 (2015), 55–75. [5] Lawrence Ein and Robert Lazarsfeld, Asymptotic syzygies of algebraic varieties, Invent. Math. 190 (2012), 603–646. 12 [6] David Eisenbud, Mark Green, Klaus Hulek and Sorin Popescu, Restricting linear syzygies: algebra and geometry, Compos. Math. 141 (2005), 1460 –1478. [7] T. J´ozefiak, P. Pragacz, and J. Weyman, Resolution of determinantal varieties and tensor complexes associated with symmetric and antisymmetric matrices, Asterisque 87-88 (1981), 109-189. [8] Giorgio Ottaviani and Rafaella Paoletti, Syzygies of Veronese embeddings, Compos. Math. 125 (2001), 31–37. [9] Claudiu Raiciu, Representation stability for syzygies of line bundles on Segre-Veronese varieties, to appear. [10] Elena Rubei, A result on resolutions of Veronese embeddings, Ann. Univ. Ferrar Sez. VII 50 (2004), 151–165. [11] Xin Zhou, Effective nonvanishing of asymptotic adjoint syzygies, Proc. AMS 142 (2014), 2255–2264. [12] Xin Zhou, Asymptotic weights of syzygies of toric varieties, to appear. Department of Mathematics, University Illinois at Chicago, 851 South Morgan St., Chicago, IL 60607 Department of Mathematics, University of Wisconsin, Madison, WI 53706 Department of Mathematics, Stony Brook University, Stony Brook, NY 11794 13
1406.6041
4
1406
2015-09-02T17:01:09
The moduli scheme of affine spherical varieties with a free weight monoid
[ "math.AG", "math.RT" ]
We study Alexeev and Brion's moduli scheme $M_\Gamma$ of affine spherical varieties with weight monoid $\Gamma$ under the assumption that $\Gamma$ is free. We describe the tangent space to $M_\Gamma$ at its `most degenerate point' in terms of the combinatorial invariants of spherical varieties and deduce that the irreducible components of $M_\Gamma$, equipped with their reduced induced scheme structure, are affine spaces.
math.AG
math
THE MODULI SCHEME OF AFFINE SPHERICAL VARIETIES WITH A FREE WEIGHT MONOID PAOLO BRAVI AND BART VAN STEIRTEGHEM ABSTRACT. We study Alexeev and Brion's moduli scheme MΓ of affine spherical varieties with weight monoid Γ under the assumption that Γ is free. We describe the tangent space to MΓ at its 'most degenerate point' in terms of the combinatorial invariants of spherical varieties and deduce that the irreducible components of MΓ, equipped with their reduced induced scheme structure, are affine spaces. 1. INTRODUCTION As part of the classification problem of algebraic varieties equipped with a group action, spher- ical varieties, which include symmetric, toric and flag varieties, have received considerable atten- tion; see, e.g., [Bri90, Kno96, Lun01, Los09b]. In [AB05], V. Alexeev and M. Brion introduced an important new tool for the study of affine spherical varieties over an algebraically closed field k of characteristic 0. We recall that an affine variety X equipped with an action of a connected re- ductive group G is called spherical if it is normal and its coordinate ring k[X] is multiplicity-free as a G-module. For such a variety a natural invariant, which completely describes the G-module structure of k[X], is its weight monoid Γ(X). By definition, Γ(X) is the set of isomorphism classes of irreducible representations of G that occur in k[X]. In view of the classification problem, we have the following natural question: how 'good' an invariant is Γ(X), or more explicitly: to what extent does Γ(X) determine the multiplicative structure of k[X]? Alexeev and Brion brought geometry to this question as follows. After choosing a Borel sub- group B of G, and a maximal torus T in B, we can identify Γ(X) with a finitely generated sub- monoid of the monoid Λ+ of dominant weights. Let Γ be another such submonoid of Λ+ and put V(Γ) = ⊕λ∈ΓV(λ), where we used V(λ) for the irreducible G-module corresponding to λ ∈ Λ+. Let U be the unipo- tent radical of B and let V(Γ)U be the subspace of U-invariants, which is also the space of highest weight vectors in V(Γ). By choosing an isomorphism V(Γ)U → k[Γ] of T-modules, where k[Γ] is the semigroup ring associated to Γ, we equip V(Γ)U with a T-multiplication law. Alexeev and Brion's moduli scheme MΓ parametrizes the G-multiplication laws on V(Γ) which extend the mul- tiplication law on V(Γ)U. For an introduction to this moduli scheme, we refer the reader to [Bri13, §4.3]. Examples of MΓ have been computed in [Jan07, BCF08, PVS12]. Let Λ be the weight lattice of G, that is, Λ is the character group of T. Because X is normal, its weight monoid Γ(X) also satisfies the following equality in Λ ⊗Z Q Γ(X) = ZΓ(X) ∩ Q≥0Γ. (1.1) By definition, this makes Γ(X) a normal submonoid of Λ+. In [Bri13], Brion conjectured that the irreducible components of MΓ are affine spaces. A precise version of this conjecture is the following. Conjecture 1.1. If Γ is a normal submonoid of Λ+, then the irreducible components of MΓ, equipped with their reduced induced scheme structure, are affine spaces. 1 This conjecture was verified for free and G-saturated monoids of dominant weights in [BCF08]. In fact, Bravi and Cupit-Foutou proved that under these assumptions, MΓ is an affine space. In [PVS12, PVS15] it is shown that MΓ is an affine space when Γ is the weight monoid of a spherical G-module. Luna provided the first non-irreducible example (unpublished): for G = SL(2) × SL(2) and Γ = h2ω, 4ω + 2ω′i, where ω and ω′ are the fundamental weights of the two copies of SL(2), the scheme MΓ is the union of two lines meeting in a point. In this paper, we verify that Conjecture 1.1 holds when Γ is free. Theorem 1.2 (Corollary 5.3). If Γ is a free submonoid of Λ+, then the irreducible components of MΓ, equipped with their reduced induced scheme structure, are affine spaces. The bulk of this paper is devoted to the description of the tangent space to MΓ at its 'most degenerate point' X0 in terms of certain combinatorial invariants, called N-spherical roots. To be more precise, we introduce some more terminology and recall some facts. If X is an affine spherical G-variety X, then its root monoid MX is the submonoid of Λ generated by the set {λ + µ − ν λ, µ, ν ∈ Λ+ such that hk[X](λ) · k[X](µ)ik ∩ k[X](ν) 6= 0}. Here k[X](λ) is the isotypic component of k[X] of type λ ∈ Λ+. Loosely speaking, MX detects how far the decomposition k[X] = ⊕λ∈Γ(X)k[X](λ) is from being a grading by Γ(X). A deep result by Knop [Kno96, Theorem 1.3] says that the saturation of MX, which is the intersection in Λ ⊗Z Q of the cone Q≥0MX and the lattice ZMX, is a freely generated monoid. Its basis ΣN(X) is called the set of N-spherical roots of X. By [AB05, Proposition 2.13] a formal consequence of our theorem above is that if X is an affine spherical G-variety with a free weight monoid, then its root monoid MX is also free; see Corollary 5.2. In their seminal paper [AB05], Alexeev and Brion equipped MΓ with an action of the maximal torus T of G. For this action, MΓ has a unique closed orbit, which is a fixed point X0. Consequently, the tangent space TX0MΓ to MΓ at the point X0 is a finite-dimensional T-module. We describe this tangent space as follows. Theorem 1.3 (Theorem 4.1 and Corollary 4.2). If Γ is a free submonoid of Λ+ , then TX0MΓ is a multiplicity-free T-module, and γ ∈ Λ occurs as a weight in TX0MΓ if and only if there exists an affine spherical G-variety X−γ with weight monoid Γ and ΣN(X−γ) = {−γ}. To prove this we first use the combinatorial theory of spherical varieties [Kno91, Lun01, Los09b] to combinatorially characterize the weights γ for which such a variety X−γ exists; see Corol- lary 2.17. Such a characterization was sketched by Luna in 2005 in an unpublished note. To prove Theorem 1.2 we use Theorem 1.3: since it is known that the irreducible components of MΓ, equipped with their reduced induced scheme structure, are affine spaces after normalization (by [Kno96, Theorem 1.3] and [AB05, Corollary 2.14]), it is enough to show that they are smooth, and this follows from our description of the tangent space to MΓ at X0 (see Section 5). Notation. Except if explicitly stated otherwise, Γ will be a free submonoid of Λ+ with basis F = {λ1, λ2, . . . , λr}. We will use S for the set of simple roots of G (associated to B and T) and R+ for the set of positive roots. The irreducible representation of G associated to the dominant weight λ ∈ Λ+ is denoted by V(λ) and we use vλ for a highest weight vector in V(λ). We use g, b, t, n, etc. for the Lie algebra of G, B, T, U, etc., respectively. When α is a root, Xα ∈ gα is a root operator and α∨ the coroot. When g is simple, simple roots are denoted by α1, . . . , αn and numbered like by Bourbaki (see [Bou68]), the corresponding fundamental weights are denoted by ω1, . . . , ωn. Acknowledgement. The authors are grateful to the Institut Fourier for hosting them in the sum- mer of 2011, when work on this project began. They also thank the Centro Internazionale per la 2 Ricerca Matematica in Trento, as well as Friedrich Knop and the Emmy Noether Zentrum in Erlan- gen for their hospitality in the summers of 2012 and 2013, respectively. The second-named author received support from The City University of New York PSC-CUNY Research Award Program and from the National Science Foundation through grant DMS 1407394. The authors are grateful to Michel Brion for suggesting this problem and for helpful discussions. They also thank Domingo Luna for discussions and suggestions in the summer of 2011 and for sharing his working paper of 2005; they were particularly helpful for Section 2. They would like to thank Jarod Alper for a clarifying exchange, summarized in Remark 5.5, about scheme structures on irreducible components of affine schemes. It alerted them to a mistake in an earlier version of this paper. The authors thank the referee for several helpful suggestions, and in particular for providing an elementary proof of Proposition 2.10(b) and for correcting an error in an earlier version of the proof of (d) and (e) of the same proposition. As this paper was being completed, R. Avdeev and S. Cupit-Foutou announced that they had independently obtained similar results [ACF14]. Note added during review. While this paper was under review, a second version of the preprint [ACF14] was posted on the arXiv, in which Avdeev and Cupit-Foutou propose a proof of Conjec- ture 1.1 for all normal monoids Γ and an example of a non-reduced moduli scheme MΓ (cf. Re- mark 5.5). 2. SPHERICAL ROOTS ADAPTED TO Γ In this section Γ denotes a normal, but not necessarily free, submonoid of Λ+. By combining re- sults from [Lun01, Kno91, Los09b, BP11] we will describe when a set of spherical roots is 'adapted' or 'N-adapted' to Γ. In particular, in Corollaries 2.16 and 2.17 we give an explicit characterization for when an element σ of the root lattice is 'adapted' or 'N-adapted' to Γ. Definition 2.1. We say that a subset Σ of NS is N-adapted to Γ if there exists an affine spherical G-variety X such that Γ(X) = Γ and ΣN(X) = Σ. By slight abuse of language, we say that an element σ of NS is N-adapted to Γ if {σ} is N-adapted to Γ. We will give the definition of 'adapted', which requires some more notions from the theory of spherical varieties, in Definition 2.11 below. After recalling some basic definitions concerning spherical varieties, we briefly discuss, in Section 2.2, the notion of 'spherically closed spherical systems', and the role they play in classifiying spherically closed spherical subgroups of G. We then, in Section 2.3 review Luna's 'augmentations'. They classify the subgroups of G which have a given spherical closure K. Finally, after recalling some basic results from the Luna-Vust theory of spherical embeddings in Section 2.4, we deduce the combinatorial characterization of adapted and N-adapted spherical roots. 2.1. Basic definitions. In this section we briefly recall the basic definitions of the theory of spher- ical varieties by freely quoting from [Lun01]. For more details on these notions the reader can also consult [Pez10, Tim11]. We recall that a (not necessarily affine) G-variety X is called spherical if it is normal and contains an open dense orbit for B. If X is affine, this is equivalent to the definition given before in terms of k[X]. The complement of the open B-orbit in X consists of finitely many B-stable prime divisors. Among those, the ones that are not G-stable are called the colors of X. The set of colors of X is denoted by ∆X. 3 By the weight lattice Λ(X) of X we mean the subgroup of Λ made up of the B-weights in the field of rational functions k(X). Since X has a dense B-orbit two rational B-eigenfunctions on X of the same weight are scalar multiples of one another. Let PX be the stabilizer of the open B-orbit and denote by Sp sponding to PX, which is a parabolic subgroup of G containing B. X the subset of simple roots corre- Let VX ⊂ Hom(Λ(X), Q) be the so-called valuation cone of X, i.e. the set of Q-valued G- invariant valuations on k(X) seen as functionals on Λ(X). By [Bri90, Theorem 3.5] VX is a cosim- plicial cone. Let Σ(X) be the set of linearly independent primitive elements in Λ(X) such that VX = {v ∈ Hom(Λ(X), Q) : hv, σi ≤ 0 for all σ ∈ Σ(X)}, i.e. the set of spherical roots of X. Similarly, the discrete valuations on k(X) associated with colors give rise to functionals on Λ(X). This yields the so-called Cartan pairing of X, a Z-bilinear map denoted by cX : Z∆X × Λ(X) → Z. Since X has a dense B-orbit, it has a dense G-orbit. Let H be the stabilizer of a point in this orbit, which we can then identify with G/H. The group H is called a spherical subgroup of G because G/H is a spherical G-variety. To H, we can associate a larger group H, called the spherical closure of H: the normalizer of H in G acts by G-equivariant automorphisms on G/H and H is the kernel of the induced action of this normalizer on ∆X (see [Lun01, §6.1] or [BL11, §2.4.1]). We recall that it follows from [BL11, Lemma 2.4.2] that H = H (see [Pez13, Proposition 3.1] for a direct proof). 2.2. Spherical systems. Here we briefly recall the definition of spherical system and its role in the classification of spherical varieties, see [Lun01, BL11]. Wonderful varieties are special spherical varieties satisfying certain regularity properties. We do not need their definition here, we just recall that by [Los09b, BP11] wonderful G-varieties (or their open G-orbits) are classified by their so-called spherical systems. This was known as Luna's conjecture, another proof of which was proposed in [Cup10]. By [Kno96], spherical homogeneous spaces G/K with K spherically closed (that is, K = K) can be realized as the open G-orbit of a unique wonderful variety. Consequently, they correspond to spherically closed spherical G- systems (systems satisfying certain combinatorial conditions, as explained below): G/K 7−→ SG/K = (Sp G/K, Σ(G/K), AG/K). Let K be a spherically closed spherical subgroup of G. The set Σ(G/K) of spherical roots of G/K is included in the root lattice ZS (because K contains the center of G) and it is a basis of Λ(G/K). Let AG/K be the set of colors that are not stable under some minimal parabolic containing B and corresponding to a simple root belonging to Σ(G/K). The full Cartan pairing restricts to the Z- bilinear pairing cG/K : ZAG/K × ZΣ(G/K) → Z, also called restricted Cartan pairing. Definition 2.2. The set Σsc(G) of spherically closed spherical roots of G is defined as Σsc(G) := {σ ∈ ZS : σ ∈ Σ(G/K) for some spherically closed spherical subgroup K of G}. Let H be a spherical subgroup of G and let X be any spherical G-variety with open G-orbit G/H. Let H be the spherical closure of H. We define Σsc(X) := Σsc(G/H) := Σ(G/H); ΣN(X) := ΣN(G/H) := Σ(G/NG(H)). Remark 2.3. 1. It follows from [Kno96, Theorem 1.3] that for X affine, ΣN(X) given in Defini- tion 2.2 agrees with the description in Section 1 of the set of N-spherical roots of X. 4 TABLE 1. spherically closed spherical roots Type of support σ α A1 2α A1 α + α′ A1 × A1 An, n ≥ 2 α1 + . . . + αn α1 + 2α2 + α3 A3 Bn, n ≥ 2 α1 + . . . + αn 2(α1 + . . . + αn) α1 + 2α2 + 3α3 α1 + 2(α2 + . . . + αn−1) + αn 2(α1 + . . . + αn−2) + αn−1 + αn α1 + 2α2 + 3α3 + 2α4 4α1 + 2α2 α1 + α2 B3 Cn, n ≥ 3 Dn, n ≥ 4 F4 G2 2. Thanks to [Los09b, Theorem 2] one can precisely describe the relationship between the three sets Σ(X), Σsc(X) and ΣN(X); see Proposition 2.9 and [VS13] for more information. 3. While Σsc(X) and ΣN(X) are subsets of NS, there exist wonderful varieties X such that Σ(X) 6⊂ ZS (see [Was96]). 4. Σ(X) is not always a basis of Λ(X), but it is when X is wonderful. 5. The weight lattice, valuation cone and spherical roots are birational invariants of the spherical variety X since they only depend on its open G-orbit G/H. The same is true of the colors and the Cartan pairing once we (naturally) identify the colors of G/H with their closures in X. The set Σsc(G) is finite. More precisely, there is the next proposition, which follows from the classification of spherically closed spherical subgroups K of G with Λ(G/K) of rank 1 [Ahi83, Los09b], see also [BL11, § 1.1.6 and § 2.4.1]. We recall that the support supp(σ) of σ ∈ NS is the set of simple roots which have a nonzero coefficient in the unique expression of σ as a linear combination of the simple roots. Proposition 2.4. An element σ of NS belongs to Σsc(G) if and only if after numbering the simple roots in supp(σ) like Bourbaki (see [Bou68]) σ is listed in Table 1. Recall that K is a spherically closed spherical subgroup of G. Therefore, see [Lun01, §7.1], the G/K, Σ(G/K), AG/K) is a spherically closed Luna spherical system in the follow- triple SG/K = (Sp ing sense. Definition 2.5. Let (Sp, Σ, A) be a triple where Sp is a subset of S, Σ is a subset of Σsc(G) and A is a finite set endowed with a Z-bilinear pairing c : ZA × ZΣ → Z. For every α ∈ Σ ∩ S, let A(α) denote the set {D ∈ A : c(D, α) = 1}. Such a triple is called a spherically closed spherical G-system if all the following axioms hold: (A1) for every D ∈ A and every σ ∈ Σ, we have that c(D, σ) ≤ 1 and that if c(D, σ) = 1 then σ ∈ S; (A2) for every α ∈ Σ ∩ S, A(α) contains two elements, which we denote by D+ α and D− α , and for all σ ∈ Σ we have c(D+ α , σ) + c(D− α , σ) = hα∨, σi; (A3) the set A is the union of A(α) for all α ∈ Σ ∩ S; (Σ1) if 2α ∈ Σ ∩ 2S then 1 (Σ2) if α, β ∈ S are orthogonal and α + β belongs to Σ then hα∨, σi = hβ∨, σi for all σ ∈ Σ; 2 hα∨, σi is a non-positive integer for all σ ∈ Σ \ {2α}; 5 (S) every σ ∈ Σ is compatible with Sp, that is, for every σ ∈ Σ there exists a spherically closed spherical subgroup K of G with Sp G/K = Sp and Σ(G/K) = {σ}. Remark 2.6. 1. Condition (S) of Definition 2.5 can be stated in purely combinatorial terms as fol- lows (see [BL11, §1.1.6]). A spherically closed spherical root σ is compatible with Sp if and only if: • in case σ = α1 + . . . + αn with support of type Bn {α ∈ supp σ : hα∨, σi = 0} \ {αn} ⊆ Sp ⊆ {α ∈ S : hα∨, σi = 0} \ {αn}, • in case σ = α1 + 2(α2 + . . . + αn−1) + αn with support of type Cn {α ∈ supp σ : hα∨, σi = 0} \ {α1} ⊆ Sp ⊆ {α ∈ S : hα∨, σi = 0}, • in the other cases {α ∈ supp σ : hα∨, σi = 0} ⊆ Sp ⊆ {α ∈ S : hα∨, σi = 0}. 2. Definition 2.5 combines the standard definition of spherical system, see [Lun01, §2], with the requirement that it be spherically closed, see [Lun01, §7.1] and [BL11, §2.4]. As shown in [Lun01], the set ∆G/K of colors and the Cartan pairing c of G/K are uniquely determined by SG/K, in the sense that they can be naturally identified with the set of colors of and the full Cartan pairing of SG/K, defined as follows. Let S = (Sp, Σ, A) be a (spherically closed) spherical G-system. The set of colors of S is the finite set ∆ obtained as the disjoint union ∆ = ∆a ∪ ∆2a ∩ ∆b where: • ∆a = A, • ∆2a = {Dα : α ∈ S ∩ 1 2 • ∆b = {Dα : α ∈ S \ (Sp ∪ Σ ∪ 1 2 Σ}, α + β ∈ Σ. Σ)}/ ∼, where Dα ∼ Dβ if α and β are orthogonal and The full Cartan pairing of S is the Z-bilinear map c : Z∆ × ZΣ → Z defined as: (2.1) c′(D, γ) = c′(D, γ) 2 hα∨, γi 1 hα∨, γi 6 if D ∈ ∆a; if D = Dα ∈ ∆2a; if D = Dα ∈ ∆b. c(D, σ) = c(D, σ) 2 hα∨, σi 1 hα∨, σi if D ∈ ∆a; if D = Dα ∈ ∆2a; if D = Dα ∈ ∆b. 2.3. Augmentations. We continue to use K for a spherically closed spherical subgroup of G. By [Lun01, Proposition 6.4] spherical homogeneous spaces G/H such that H, the spherical closure of H, is equal to K are classified by their weight lattice, which is an augmentation of SG/K . Definition 2.7. Let S = (Sp, Σ, A) be a spherically closed spherical G-system with Cartan pairing c : ZA × ZΣ → Z. An augmentation of S is a lattice Λ′ ⊂ Λ endowed with a pairing c′ : ZA × Λ′ → Z such that Λ′ ⊃ Σ and (a1) c′ extends c; (a2) if α ∈ S ∩ Σ then c′(D+ (σ1) if 2α ∈ 2S ∩ Σ then α /∈ Λ′ and hα∨, ξi ∈ 2Z for all ξ ∈ Λ′; (σ2) if α and β are orthogonal elements of S with α + β ∈ Σ then hα∨, ξi = hβ∨, ξi for all ξ ∈ Λ′; α , ξ) = hα∨, ξi for all ξ ∈ Λ′; α , ξ) + c′(D− and (s) if α ∈ Sp then hα∨, ξi = 0 for all ξ ∈ Λ′. Let ∆ be the set of colors of S . The full Cartan pairing of the augmentation is the Z-bilinear map c′ : Z∆ × Λ′ → Z given by Remark 2.8. By the definition of spherical closure, ∆G/H and ∆ G/H are naturally identified and the full Cartan pairing Z∆G/H × Λ(G/H) → Z on G/H is the full Cartan pairing of the augmentation corresponding to H (see Proposition 6.4 and the proof of Theorem 3 in [Lun01]). We state here, for future reference, the following consequence of [Los09b, Theorem 2]. Proposition 2.9. Let G/H be a spherical homogeneous space with Σsc(G/H) = Σ. Then ΣN(G/H) = (Σ \ Σl) ∪ 2Σl, where Σl = {α ∈ Σ ∩ S : cG/H(D+ α , γ) = cG/H(D− α , γ) for all γ ∈ Λ(G/H)}. Proof. This follows immediately from comparing [Los09b, Theorem 2], which describes the re- lationship between Σ(G/H) and ΣN(G/H) with [Lun01, Lemma 7.1], which describes the rela- tionship between Σ(G/H) and Σsc(G/H). Note that [Lun01, Lemma 7.1] can be deduced from [Los09b] without appealing to Luna's conjecture. (cid:3) 2.4. Strictly convex colored cones and weight monoids of affine spherical varieties. An equi- variant embedding of a spherical homogeneous space G/H as a dense orbit in a spherical G- variety (an embedding of G/H, for short) is called simple if it has only one closed orbit. Affine spherical varieties are simple. If X is a simple embedding of the spherical homogeneous space G/H, let F (X) be the set of colors of X containing the closed orbit (identified with elements of ∆G/H), and let C(X) be the cone in Hom(Λ(G/H), Q) generated by the valuations associated with the G-stable divisors of X (identified with elements of VG/H) and by c(F (X), ·). The couple (C(X), F (X)) is a strictly convex colored cone in the sense of the following definition. A strictly convex colored cone is a couple (C, F ) where - F is a subset of ∆G/H such that the subset c(F , ·) of Hom(Λ(G/H), Q) does not contain 0, - C is a strictly convex polyhedral cone in Hom(Λ(G/H), Q) which is generated by c(F , ·) and finitely many elements of VG/H and whose relative interior intersects VG/H. We recall from [Kno91, Theorem 3.1] that simple embeddings X of the spherical homogoneous space G/H are classified by their strictly convex colored cones. By [Kno91, Theorem 6.7] the simple embedding X is affine if and only if there exists a character χ ∈ Λ(G/H) that is non- positive on VG/H, zero on C(X) and c(·, χ) is strictly positive on ∆G/H \ F (X). We gather some known results about the weight monoid of affine spherical varieties. Proposition 2.10. If X is an affine sperical G-variety with weight monoid Γ(X) and open orbit G/H, then (a) the weight lattice of X (or of G/H) is ZΓ(X); (b) the set Sp (c) the dual cone Γ∨(X) := {v ∈ Hom(ZΓ(X), Q) : hv, γi ≥ 0 for all γ ∈ Γ(X)} to Γ(X) is a strictly G/H) is equal to {α ∈ S : hα∨, γi = 0 for all γ ∈ Γ(X)}; X (which is the same as Sp convex polyhedral cone; (d) every ray of Γ∨(X) contains an element of c(∆G/H, ·) or of VG/H; (e) Γ∨(X) contains c(∆G/H, ·). Proof. These statements are well-known to experts and can be extracted from the results summa- rized in [Tim11, §15.1]. For the reader's convenience, we provide a proof. Assertion (a) follows from the fact that a rational B-eigenfunction on X is necessarily equal to the quotient of two reg- ular B-eigenfunctions; see e.g. [Bri10, Proposition 2.8(i)]. Assertion (b) is [Cam01, Lemme 10.2]. It follows from the fact that PX is the common stabilizer of the B-stable lines in k[X]. This is the case because PX is the common stabilizer of the B-stable prime divisors of X and the union of these divisors is the zero set of some B-eigenvector in k[X]. Assertion (c) is a standard fact in convex geometry. Parts (d) and (e) follow from the fact that a rational B-eigenfunction on X is regular if 7 and only if it does not have poles along the colors or G-stable prime divisors of X. This, in turn, is so because X is normal. (cid:3) 2.5. Adapted spherical roots. Recall that Γ is a normal submonoid of Λ+. Combining the results recalled above, one derives the condition on a set of spherical roots Σ for being adapted to Γ. Definition 2.11. We say that a subset Σ of Σsc(G) is adapted (or N-adapted) to Γ if there exists an affine spherical G-variety X such that Γ(X) = Γ and Σsc(X) = Σ (respectively, ΣN(X) = Σ). Remark 2.12. Let Σ be a subset of Σsc(G). Losev's Theorem [Los09a, Theorem 1.2] asserts that there is at most one affine spherical G-variety X with Γ(X) = Γ and ΣN(X) = Σ. Because Σsc(X) determines ΣN(X) (see Proposition 2.9) there is also at most one affine spherical G-variety Y with Σsc(Y) = Σ and Γ(Y) = Γ. The dual cone to Γ is Γ∨ := {v ∈ Hom(ZΓ, Q) : hv, γi ≥ 0 for all γ ∈ Γ}. It is a strictly convex polyhedral cone. We denote the set of primitive vectors on its rays by E(Γ): (2.2) E(Γ) := {δ ∈ (ZΓ)∗ : δ spans a ray of Γ∨ and δ is primitive}. Observe that (2.3) E(Γ) = {δ ∈ (ZΓ)∗ : δ is primitive, δ(Γ) ⊂ Z≥0, δ is the equation of a face of codim 1 of Q≥0Γ}. Moreover, for α ∈ S ∩ ZΓ, we define Finally, we put a(α) := {δ ∈ (ZΓ)∗ : δ(α) = 1 and(cid:0)δ ∈ E(Γ) or α∨ − δ ∈ E(Γ)(cid:1)}. Sp(Γ) := {α ∈ S : hα∨, γi = 0 for all γ ∈ Γ}. Proposition 2.13. Let Γ be a normal submonoid of Λ+. A subset Σ of Σsc(G) is adapted to Γ if and only if there exists a spherically closed spherical system S = (Sp, Σ, A) such that (1) Sp = Sp(Γ); and (2) ZΓ is an augmentation of ZΣ; and (3) if δ ∈ E(Γ), then hδ, σi ≤ 0 for all σ ∈ Σ or there exists D ∈ ∆ such that c(D, ·) is a positive multiple of δ; where ∆ is the set of colors of S and c : Z∆ × ZΓ → Z is the full Cartan pairing of the augmentation; and (4) c(D, ·) ∈ Γ∨ for all D ∈ ∆. Proof. This is a consequence of the results we reviewed in Sections 2.2 through 2.4. We begin with the necessity of the conditions. Let X be an affine spherical G-variety with Σsc(X) = Σ and Γ(X) = Γ. Let G/H be the open orbit of X and let H be the spherical closure of H. Then Σsc(X) = Σ(G/H) by definition, and Sp . It follows from §5.1 and Lemma 7.1 in [Lun01] that (Sp(Γ), Σ, AG/H) is a spherically closed spherical system. Since H has spherical closure H, (2) follows from [Lun01, Proposition 6.4]. Conditions (3) and (4) follow from (d) and (e) of Proposition 2.10. G/H = Sp(Γ) by Proposition 2.10(b). Moreover Sp G/H = Sp G/H We now show that the conditions are sufficient for Σ to be adapted to Γ. By [BP11] there exists a spherically closed spherical subgroup K of G with spherical system S . Condition (2) implies by [Lun01, Proposition 6.4] that there exists a spherical subgroup H of G with H = K and Λ(G/H) = ZΓ. What remains is to prove that G/H has an affine embedding X with weight monoid Γ. That is, by [Kno91, Theorems 3.1 and 6.7] we have to show that there exists a strictly convex colored cone (C, F ) in Hom(ZΓ, Q), with respect to V = {v ∈ Hom(ZΓ, Q) : hv, σi ≤ 0 for all σ ∈ Σ} and the set of colors ∆ of S , such that: 8 (i) there exists χ ∈ ZΓ that is non-positive on V, zero on C and strictly positive on ∆ \ F ; and (ii) Γ = {γ ∈ ZΓ : hv, γi ≥ 0 for all v ∈ C ∪ ∆}. We claim that if (1), (3) and (4) hold, then the desired strictly convex colored cone exists. Indeed, take C to be the maximal face of Γ∨ whose relative interior meets V with F the set of colors contained in C (such a maximal face exists since the zero face actually meets V). The set c(F , ·) does not contain 0. Indeed, a color D with c(D, ·) = 0 necessarily belongs to ∆b, whence c(D, ·) = hα∨, ·i for some α ∈ S but by (1) this implies α ∈ Sp. Moreover, C is contained in a hyperplane that separates V and ∆ \ F . This yields χ. The inclusion "⊂" of (ii) holds because C ⊂ Γ∨ and because c(∆, ·) ⊂ Γ∨ by (4). The other inclusion follows from (3) and the maximality of C. (cid:3) Remark 2.14. It follows from equation (2.4) below that the spherical system S and the Cartan pairing of the augmentation in Proposition 2.13 are uniquely determined by Γ and Σ. Corollary 2.15. Let Γ be a normal submonoid of Λ+. A subset Σ of Σsc(G) is N-adapted to Γ if and (2.4) a(α) = {c(D+ α , ·), c(D− α , ·)} as in Proposition 2.13, then Proof. This is a consequence of Proposition 2.13 and Proposition 2.9 once we show the following: if only if there exists a subset eΣ of Σsc(G) which is adapted to Γ and such that Σ = (eΣ \eΣl) ∪ 2eΣl, where eΣl = {α ∈eΣ ∩ S : a(α) has one element}. c is the full Cartan pairing of an augmentation ZeΣ ⊂ ZΓ of a spherical system S = (Sp(Γ),eΣ, A) for all α ∈eΣ ∩ S. To prove the inclusion "⊂" in (2.4), let δ ∈ a(α). Then, hδ, αi = hα∨ − δ, αi = 1 > 0 and at least one of δ and α∨ − δ is in E(Γ). By (3) in Proposition 2.13 it follows that {δ, α∨ − δ} con- tains a positive rational multiple of c(D, ·) for some color D. By axiom (A1) of the spherical system S , and the description (2.1) of c, the color D must be D+ α , α) = 1, this implies that the two sets {δ, α∨ − δ} and {c(D+ α , ·)} intersect, and so by axiom (a2) of the augmentation, they are equal. For the reverse inclusion in (2.4) we have to show that c(D+ α , ·) or c(D− α , ·) belongs to E(Γ). If neither belongs to E(Γ), then by (3) and (4) in Proposition 2.13 together with the description (2.1) of c and axiom (A1) in Definition 2.5, each of them is a linear combination with positive rational coefficients of elements of Hom(ZΓ, Q) which are nonpositive on α. This contradicts the fact that c(D+ (cid:3) α , α) = 1 and finishes the proof of equation (2.4). α or D− α . Since c(D+ α , α) = c(D− α , ·), c(D− As the next two corollaries show, one can characterize very explicitly whether a single spherical root is adapted (Corollary 2.16) or N-adapted (Corollary 2.17) to Γ. In a 2005 working document, Luna had proposed a statement like Corollary 2.16. We remark that while Proposition 2.13 and Corollary 2.15 depend on the full classification of wonderful varieties by spherical systems (Luna's conjecture), the next two results only use the combinatorial classification of rank 1 wonderful varieties, which was obtained in [Bri89] and also in [Ahi83]. Corollary 2.16. Let Γ be a normal submonoid of Λ+. If σ ∈ Σsc(G), then σ is adapted to Γ if and only if all of the following conditions hold: (1) σ ∈ ZΓ; (2) σ is compatible with Sp(Γ); (3) if σ /∈ S and δ ∈ E(Γ) such that hδ, σi > 0 then there exists β ∈ S \ Sp(Γ) such that β∨ is a positive multiple of δ; (4) if σ ∈ S then (a) a(σ) has one or two elements; and (b) hδ, γi ≥ 0 for all δ ∈ a(σ) and all γ ∈ Γ; and (c) hδ, σi ≤ 1 for all δ ∈ E(Γ); (5) if σ = 2α ∈ 2S, then α /∈ ZΓ and hα∨, γi ∈ 2Z for all γ ∈ Γ; 9 (6) if σ = α + β with α, β ∈ S and α ⊥ β, then α∨ = β∨ on Γ. Proof. Let σ ∈ Σsc(G). Define the triple S by S :=((Sp(Γ), {σ}, ∅) (Sp(Γ), {σ}, {D+ if σ /∈ S; if σ ∈ S. σ , D− σ }) Let ∆ be the set of colors of S (see Section 2.2) and let c : Z∆ × ZΓ be the bilinear pairing given by equation (2.1) if σ /∈ S and by (2.5) (2.6) c(D, γ) = hα∨, γi if D = Dα ∈ ∆b; {c(D+ σ , ·)} = a(σ), σ , ·), c(D− if σ ∈ S. By Remark 2.14, we have to show that the conditions of the corollary hold if and only if S is a spherically closed spherical system of which ZΓ together with c is an augmentation such that conditions (3) and (4) of Proposition 2.13 hold. We briefly describe the straightforward verification. We begin with the case σ /∈ S. Then we have that S is a spherically closed spherical G-system if and only if (2) holds. Then c gives an augmentation of S if and only if (1), (5) and (6) hold. Condition (4) of Proposition 2.13 is vacuous since Γ ⊂ Λ+ and every c(D, ·) is a positive multiple of a coroot. Condition (3) in the corollary is the same as condition (3) of Proposition 2.13 by the definition of c. We proceed to the case σ ∈ S. Now S is a spherically closed spherical G-system if and only if (2) and (4a) hold. Next, by construction, c gives an augmentation of S if and only if we have (1). Condition (4) of Proposition 2.13 is equivalent to (4b). Finally, condition (3) of Proposition 2.13 is equivalent to (4c), again by the definition of c. (cid:3) The combinatorial conditions that characterize N-adapted spherical roots are exactly the same except for conditions (4a) and (5). We report all of them again entirely in the next statement for later reference. Corollary 2.17. Let Γ be a normal submonoid of Λ+. If σ ∈ Σsc(G), then σ is N-adapted to Γ if and only if all of the following conditions hold: (1) σ ∈ ZΓ; (2) σ is compatible with Sp(Γ); (3) if σ /∈ S and δ ∈ E(Γ) such that hδ, σi > 0 then there exists β ∈ S \ Sp(Γ) such that β∨ is a positive multiple of δ; (4) if σ ∈ S then (a) a(σ) has two elements; and (b) hδ, γi ≥ 0 for all δ ∈ a(σ) and all γ ∈ Γ; and (c) hδ, σi ≤ 1 for all δ ∈ E(Γ); (5) if σ = 2α ∈ 2S, then hα∨, γi ∈ 2Z for all γ ∈ Γ; (6) if σ = α + β with α, β ∈ S and α ⊥ β, then α∨ = β∨ on Γ. Proof. By Corollary 2.15, if σ /∈ S ∪ 2S, then σ is adapted to Γ if and only if it is N-adapted to Γ. From the same corollary it follows that σ ∈ S is N-adapted to Γ if and only if it is adapted to Γ and a(σ) has two elements. The only remaining case is σ = 2α for some α ∈ S. Again by Corollary 2.15, 2α is N-adapted to Γ if and only if either (i) 2α is adapted to Γ; or (ii) α is adapted to Γ and a(α) has one element. We assume that (1) and (2) hold and claim that (3) and (5) hold if and only if (i) or (ii) is true. Indeed, it is clear from Corollary 2.16 that if 2α is adapted to Γ then we have (3) and (5). On the 10 2 α∨ and so (5) other hand, if α is adapted to Γ and a(α) has one element, then that element is 1 holds. Moreover, condition (4c) of Corollary 2.16 implies (3) of this corollary. Conversely, suppose that we have (3) and (5). Since the restricion of α∨ to ZΓ belongs to Γ∨ and hα∨, 2αi > 0, there exists δ ∈ E(Γ) such that hδ, 2αi > 0. It follows from (3) that δ = qβ∨ for some β ∈ S \ Sp(Γ) and q ∈ Q>0. Clearly, β = α, which proves that δ is the only element of E(Γ) that takes a positive value on 2α. Now, suppose that 2α is not adapted to Γ, i.e. that (i) does not hold. Then α must 2 α∨ takes integer values on ZΓ, and since it takes value 1 on α, it is be an element of ZΓ. By (5), 1 primitive in (ZΓ)∗ and therefore an element of E(Γ) and the only element of a(α). It follows from Corollary 2.16 that (ii) is true. This finishes the proof. (cid:3) 3. THE Tad-WEIGHTS IN (V/g · x0)Gx0 For the remainder of the paper, Γ will be a free monoid with basis F ⊂ Λ+. In this section, we begin by recalling that the moduli scheme MΓ is an open subscheme of a certain invariant Hilbert scheme HΓ. This allows one to realize the tangent space TX0MΓ as a T-submodule of a certain vector space (V/g · x0)Gx0 . In Section 3.2 we prove that if γ is a T-weight in (V/g · x0)Gx0 , then it is a spherical root of spherically closed type. In Section 3.3 we further show that γ is compatible Gx0 with Sp(Γ). We also show that if γ /∈ S, then the weight space (V/g · x0) (γ) has dimension at most 1. For notational and computational convenience, we actually work with the opposite of Alexeev and Brion's T-action on MΓ and with a twist of their action on HΓ (see Section 3.1). 3.1. The invariant Hilbert scheme and its tangent space. We briefly review some known facts regarding MΓ and its relation to a certain invariant Hilbert scheme HΓ. For more details we refer to [AB05], [Bri13, Section 4.3] and to [PVS12, §2.1 and §2.2]. Recall that Γ is a free monoid of dominant weights with basis F = {λ1, λ2, . . . , λr}, and put V := V(λ1) ⊕ V(λ2) ⊕ . . . ⊕ V(λr); x0 := vλ1 + vλ2 + . . . + vλr . We denote by HΓ the Hilbert scheme HilbG h (V) of [AB05], where h is the characteristic func- tion of Γ∗ := −w0Γ (where w0 is the longest element in the Weyl group of G). The scheme HΓ parametrizes the G-stable ideals I of k[V] such that k[V]/I ≃ ⊕λ∈Γ∗V(λ) as G-modules. We equip HΓ with the action of T described in [PVS12, §2.2]. This is the same action as in [BCF08], and is a 'twist' of the action in [AB05] and in [Bri13, p. 101]. We briefly recall its definition. Let GL(V)G be the group of linear automorphisms of V that commute with the action of G. Note that GL(V)G is a torus of dimension r. The natural action of GL(V)G on V (by G-equivariant automorphisms) induces an action on HΓ. Composing with the homomorphism (3.1) T → GL(V)G : t 7→ (λ1(t), λ2(t), . . . , λr(t)), we obtain our action of T on HΓ. The center Z(G) of G belongs to the kernel of this action, which therefore descends to an action of Tad := T/Z(G). We will refer to our action as the "Tad-action" on HΓ. For the reader's conve- nience, the corresponding Tad-action induced on the tangent space to HΓ at the Tad-fixed point is recalled below in (3.3). As was reviewed in [PVS12, §2.2] it follows from [AB05, Corollary 1.17 and Lemma 2.2] that since Γ∗ is free, we can view MΓ∗ as a Tad-stable open subscheme of HΓ. Under this identification, the Tad-fixed point X0 of MΓ∗ corresponds to a certain subvariety of V which we also denote by X0, namely (3.2) X0 = the closure of the G-orbit of x0 in V. The next proposition relates MΓ to HΓ. 11 Proposition 3.1. Let Γ be a free monoid of dominant weights. If we equip MΓ with the opposite of the Tad-action in [AB05] and HΓ with the Tad-action in [PVS12, §2.2], then there is a Tad-equivariant open embedding MΓ ֒→ HΓ which sends the unique Tad-fixed point of MΓ to the point X0 in equation (3.2). Proof. This a matter of "formal bookkeeping." Composing the action of G on V(Γ) with the Chevalley involution of G induces an isomorphism MΓ ≃ MΓ∗. Composing this isomorphism with the open Tad-equivariant embedding MΓ∗ ֒→ HΓ chosen above gives an open embedding MΓ ֒→ HΓ. Comparing the definition of the action in [AB05] with that of the action in [PVS12, §2.2] one shows that this open embedding is Tad-equivariant for the actions as given in the propo- sition. (cid:3) Remark 3.2. In what follows, MΓ and HΓ will always be equipped with the action given in Propo- sition 3.1. The action Alexeev and Brion defined on MΓ is conceptually the most natural, while we find the action we are using on HΓ computationally more convenient. By [AB05, Proposition 1.13], there is a canonical isomorphism TX0HΓ ≃ H0(X0, NX0V)G where H0(X0, NX0V)G is the space of G-invariant global sections of the normal sheaf NX0V of X0 in V. Moreover, by [Bri13, Proposition 3.10], there is an inclusion of Tad-modules H0(X0, NX0V)G ֒→ (V/g · x0)Gx0 ≃ H0(G · x0, NX0V)G, where the Tad-action on (V/g · x0)Gx0 is induced by the following action of Tad on V. For t ∈ Tad and v a T-weight vector of weight δ in V(λ) ⊂ V, we put t · v := λ(t)δ(t)−1v. (3.3) 3.2. The Tad-weights in (V/g · x0)Gx0 are spherical roots of G. In this section, we prove the fol- lowing theorem. Theorem 3.3. If γ is a Tad-weight in (V/g · x0)Gx0 then γ is a spherically closed spherical root of G. Proof. Corollary 3.8 and Corollary 3.14. (cid:3) For future use, we recall the following elementary and well-known facts regarding (V/g · x0)Gx0 . We include proofs for convenience. Before stating them we define F⊥ := {β ∈ R+ : hλ, β∨i = 0 for all λ ∈ F}. Proposition 3.4. (a) A basis of Tad-eigenvectors of g · x0 is given by {vλ : λ ∈ F} ∪ {X−β · x0 : β ∈ R+ \ F⊥}. (b) If [v] is a Tad-eigenvector in V/g · x0 of weight γ, then [v] ∈ (V/g · x0)Gx0 if and only if γ ∈ ZΓ and Xβ · v ∈ g · x0 for all β ∈ S ∪ −(S ∩ F⊥). Proof. Assertion (a) follows from the fact that g · x0 = b− · x0 = t · x0 + n− · x0 and that F is linearly independent. Assertion (b) follows from [PVS12, Lemma 2.16] and the fact that gx0 is generated as a Lie algebra by tx0 and the root spaces gβ with β ∈ S ∪ −(S ∩ F⊥) (see, e.g., [Hum75, Theorem 30.1]). (cid:3) In the remainder of this section, γ is a Tad-weight occuring in (V/g · x0)Gx0 and v ∈ V a Tad-eigenvector of weight γ such that [v] is a nonzero element of (V/g · x0)Gx0 . By Propostion 3.4 (and the choice of our Tad-action), the weight γ belongs to NS ∩ ZΓ. 12 Lemma 3.5 ([BCF08, Lemma 3.3]). (1) There exists at least one simple root α such that Xαv 6= 0. (2) If α is a simple root such that Xαv 6= 0 and γ 6= α, then γ − α is a positive root. (3) If α is a simple root such that γ − α is a root then there exists z ∈ k such that Xαv = z X−γ+αx0. Proof. The vector v cannot be a linear combination of the highest weight vectors vλi, otherwise (since the weights λi are linearly independent) it would belong to t · x0 ⊂ g · x0. Moreover, since Xα ∈ gx0 for all α ∈ S, Xαv is a Tad-eigenvector of weight γ − α in g · x0. (cid:3) We first deal with the case where γ is a root. Notice that since γ ∈ NS, it is then a positive root. As is well known, we then also have that supp(γ) is a connected subset of the Dynkin diagram of G. Lemma 3.6. If γ is a root, which is not simple, then there exist at least two distinct simple roots α such that γ − α is a root. Proof. Assume that there exists only one simple root α such that γ − α is a root. By Lemma 3.5, there exists z ∈ k such that Xαv = z X−γ+αx0. Moreover, there exists z′ ∈ k × such that [Xα, X−γ] = z′ X−γ+α. Therefore, if we put z′′ = z/z′ then Xα(v + z′′ X−γx0) = 0. Since [v] = [v + z′′ X−γx0] in V/g · x0 we can assume that Xαv = 0. Since γ − α′ is not a positive root for all α′ ∈ S \ {α}, it then follows that Xαv = 0 for all α ∈ S, which contradicts Lemma 3.5(1). (cid:3) Proposition 3.7. If γ is a root, of which the support is not of type G2, then it is a locally dominant short root, i.e. the dominant short root in the root subsystem generated by the simple roots of its support. Proof. I. Let α1 and α2 be two orthogonal simple roots such that γ − α1 and γ − α2 are roots. Notice that γ − α1 − α2 is also a root. We claim that if there exists λ ∈ F not orthogonal to γ − α1 − α2, then we can assume (3.4) Xα1 v = Xα2v = 0. Indeed, there exist z1, z2 ∈ k × such that [Xα1, X−γ] = z1 X−γ+α1; [Xα2, X−γ] = z2 X−γ+α2. Moreover, using the Jacobi identity and the fact that [Xα1, Xα2] = 0 one finds that [Xα2, X−γ+α1] = z2 z1 [Xα1, X−γ+α2]. By Lemma 3.5(3), there exist z′ 1, z′ 2 ∈ k such that Xα1v = z′ Xα2v = z′ 1 X−γ+α1 x0; 2 X−γ+α2 x0. Since Xα2 Xα1v = Xα1 Xα2v we obtain that ( z2 z1 1 − z′ z′ 2)[Xα1, X−γ+α2]x0 = 0. Using that there exists λ ∈ F not orthogonal to the root γ − α1 − α2 it follows that z2 z1 that is z′ 1 z1 = z′ 2 z2 . This implies that by replacing v by v − z′ 1 z1 X−γx0 = v − z′ 2 z2 X−γx0, we can assume (3.4). 13 1 − z′ z′ 2 = 0, II. The same can be done if we have α1, α2, . . . , αk simple roots with αj orthogonal to αj+1 for all j ∈ {1, 2, . . . , k − 1} and such that γ − αj is a root for all j ∈ {1, 2, . . . , k}. More precisely, we claim that if there exists λ ∈ F not orthogonal to γ − α1 − . . . − αk, then we can assume that for all j ≤ k (3.5) Xαjv = 0. j ∈ k such that [Xαj, X−γ] = Indeed, for every j ∈ {1, 2, . . . , k} there exists, as in part I, zj ∈ k zj X−γ+αj and Xαjv = z′ j X−γ+αjx0. Let λ be an element of F that is not orthogonal to γ − α1 − . . . − αk. Then λ is not orthogonal to γ − αj − αj+1 for all j ∈ {1, 2, . . . , k − 1}. By applying part I (k − 1) times to the pairs αj, αj+1 we obtain that × and z′ z′ 1 z1 This implies that by replacing v by v − z′ 1 z1 = z′ 2 z2 = . . . = z′ k zk . X−γx0, we can assume (3.5). III. Assume that there exist more than two simple roots, say α1, . . . , αk, such that γ − αj is a root for all j ∈ {1, 2, . . . , k}. We claim that they can be reordered such that αj is orthogonal to αj+1 for all j < k as in part II. This can be verified by making use of the classification of root systems, checking case-by-case all the positive roots, noticing along the way (although we will not need this) that k is at most 3. This is straightforward for the classical types. To avoid the large number of case-by-case checkings in the exceptional types E6, E7, E8 and F4 one can use for example the following argument. If it were not possible to reorder the simple roots α1, . . . , αk as required, then there would exist three roots among them, say αj1, αj2, αj3, such that αj2 is not orthogonal to both αj1 and αj3 . We will now show that this is impossible for each exceptional type using well-known properties of root systems of rank 2 and 3. Notice, in particular, that if the support of γ is not of type G2 and if γ − α is a root for some simple root α, then hα∨, γi ≥ 0 since otherwise there would exist a root string of length greater than 3. In types E6, E7 and E8 all the roots have the same length so we would necessarily have h(αjm )∨, γi = 1 for m ∈ {1, 2, 3}, but this is absurd since it would mean that h(αj1 + αj2 + αj3 )∨, γi = 3. In type F4 the three simple roots would generate a root subsytem of type B3 or of type C3. In the former case (type B3) we would necessarily have h(αj1 )∨, γi = h(αj2 )∨, γi = 1 assuming αj1 and αj2 are long, but this is absurd since it would mean h(αj1 + αj2 + αj3 )∨, γi ≥ 4. In the latter case (type C3) we would necessarily have h(αj1 )∨, γi = 1 assuming αj1 is long. If h(αj3 )∨, γi is positive, then h(αj1 + αj2 + αj3 )∨, γi is greater than 2, which is not possible in type F4. If h(αj3 )∨, γi = 0, then γ + αj3 is a root, and h(αj1 + αj2 + αj3 )∨, γ + α3i is greater than 2, which is again absurd. IV. We now want to prove that γ is locally dominant (if the support of γ is not of type G2). The fact that γ is locally short then follows. Indeed, if the support of γ is not simply laced, then the highest root in the root system generated by that support does not satisfy Lemma 3.6: - in type Bn, n ≥ 2, the highest root is α1 + 2(α2 + . . . + αn) = ω2; - in type Cn, n ≥ 3, the highest root is 2(α1 + . . . + αn−1) + αn = 2ω1; - in type F4 the highest root is 2α1 + 3α2 + 4α3 + 2α4 = ω1. To obtain a condradiction we assume that γ is not locally dominant, that is, we assume that there exists β ∈ supp(γ) such that hβ∨, γi < 0. Recall from part III that in type different from G2 if γ − α is a root for a simple root α, then hα∨, γi ≥ 0. Suppose first that there are exactly k > 2 simple roots, say α1, . . . , αk, such that γ − αj is a root for all j ≤ k. From the assumption that γ is not locally dominant, it follows that there exists λ ∈ F 14 not orthogonal to γ − α1 − . . . − αk. By parts II and III we can then assume that Xαjv = 0 for all j ≤ k. This contradicts Lemma 3.5(1). If there are exactly two simple roots α1 and α2 such that γ − α1 and γ − α2 are roots, and α1 and α2 are orthogonal, then by part I we get the same contradiction with Lemma 3.5(1). Furthermore, if the support of γ has cardinality ≤ 2, then the proposition follows by Lemma 3.6. Indeed, the only roots with support of cardinality ≤ 2 satisfying Lemma 3.6 are: - with support of type A1, α1, - with support of type A2, α1 + α2, - with support of type B2, α1 + α2. Therefore, we now restrict to the case of support of γ of cardinality > 2, and assume that there are only two simple roots α1 and α2, such that γ − α1 and γ − α2 are roots, and that α1 and α2 are not orthogonal. Notice that α1 + α2 is a root. Up to exchanging α1 and α2 we can assume that (3.6) Indeed, at least one of the two hα∨ 2 , γi must be positive (otherwise γ would be antidom- inant), and not both 2α1 + α2 and α1 + 2α2 can be roots. If say 2α1 + α2 is a root, then kα1k < kα2k, hence α2 is long and therefore hα∨ 2 , γi > 0 and α1 + 2α2 /∈ R. 1 , γi and hα∨ hα∨ Under (3.6) we have 2 , γi must be > 0. hα∨ 2 , γ − α1i ≥ 1 + 1 hence γ − α1 − α2 and γ − α1 − 2α2 are roots. Since γ is not locally dominant, there is an element λ of F such that h(γ − α1 − 2α2)∨, λi 6= 0. To conclude the proof of the proposition, we use once again an argument similar to that of part I. Indeed, we will show in part V that we can assume that Xα1v = Xα2v = 0, which contradicts Lemma 3.5(1). V. We finish by proving the following claim: if α1 and α2 are simple roots such that - α1 + 2α2 is not a root; - γ − α1, γ − α2, γ − α1 − α2, and γ − α1 − 2α2 are roots; and - h(γ − α1 − 2α2)∨, λi 6= 0 for some λ ∈ F; then we can assume that Xα1v = Xα2 v = 0. Since α1 + 2α2 is not a root we have that [Xα2, Xα1+α2] = 0. By the third assumption of the claim, (3.7) X−(γ−α1−2α2)x0 6= 0. We first show that we can assume that (3.8) There exist z′ 1, z′ 2 ∈ k such that Xα2v = Xα1+α2v = 0. Xα2v = z′ Xα1+α2v = z′ 1X−(γ−α2)x0; 2X−(γ−α1−α2)x0. Next, there exist z1, z2 ∈ k × such that [Xα2, X−γ] = z1X−(γ−α2); [Xα1+α2, X−γ] = z2X−(γ−α1−α2). As in part I, one deduces from Xα2 Xα1+α2v = Xα1+α2 Xα2v that ( z2 z1 1 − z′ z′ 2)[Xα2, X−(γ−α1−α2)]x0 = 0. 15 Using (3.7), it follows that (3.9) Hence, if we replace v by v − z′ 1 z1 z′ 1 z1 X−γx0 = v − z′ 2 z2 = z′ 2 z2 . X−γx0, then equations (3.8) hold. We now complete the proof by showing that (3.8) implies that (3.10) Xα1v = 0. There exists z ∈ k such that Xα1 v = zX−(γ−α1)x0. From (3.8) we have that 0 = Xα1+α2v = Xα2 Xα1v = zXα2 X−(γ−α1)x0 = zX−(γ−α1−α2)x0, where the second equality uses that Xα2v = 0 and the fourth one uses that Xα2 x0 = 0. Since equation (3.7) implies that Xα2 X−(γ−α1−α2)x0 6= 0, we have that X−(γ−α1−α2)x0 6= 0, and therefore that z = 0 which proves equation (3.10), the claim at the start of part V and the proposition. (cid:3) The following is Theorem 3.3 for the case that γ is a root. Corollary 3.8. Let γ be a Tad-weight in (V/g · x0)Gx0 . If γ is a root, then γ is a spherically closed spherical root of G. Proof. If the support of γ is not of type G2, then by Proposition 3.7 we have only to check the locally dominant short roots. The following roots do not satisfy Lemma 3.6. - With support of type Dn, n ≥ 4: α1 + 2(α2 + . . . + αn−2) + αn−1 + αn = ω2. - With support of type E6: α1 + 2α2 + 2α3 + 3α4 + 2α5 + α6 = ω2. - With support of type E7: 2α1 + 2α2 + 3α3 + 4α4 + 3α5 + 2α6 + α7 = ω1. - With support of type E8: 2α1 + 3α2 + 4α3 + 6α4 + 5α5 + 4α6 + 3α7 + 2α8 = ω8. Therefore, we are left with all spherically closed spherical roots. - With support of type An, n ≥ 1: α1 + . . . + αn. - With support of type Bn, n ≥ 2: α1 + . . . + αn. - With support of type Cn, n ≥ 3: α1 + 2(α2 + . . . + αn−1) + αn. - With support of type F4: α1 + 2α2 + 3α3 + 2α4. If the support of γ is of type G2 the only positive root satisfying Lemma 3.6 is α1 + α2, which is a spherically closed spherical root. (cid:3) Let us now consider the case where γ is not a root. In contrast to the root case, here we notice the following general fact. Proposition 3.9. Let α be a simple root and let β be a non-simple positive root such that α + β is not a root. Then there exists no simple root α′ 6= α such that (α + β) − α′ is a root. Proof. Assume that there exists a simple root α′ 6= α such that α + β − α′ is a root. Since β − α′ is nonzero, it is a root. This follows from the fact that α + β is not a root, whencehα∨, βi ≥ 0, and so hα∨, α + β − α′i > 0. Finally, to deduce that α + β is a root (i.e. a contradiction), one can use for example a saturation argument (see [Hum72, Lemma 13.4.B]) as follows. Restrict the adjoint representation to the Levi subalgebra associated with α and α′. Since β − α′ is a root, both β and α + β − α′ occur as weights in the same irreducible summand, say of highest weight λ. From hα∨, βi ≥ 0, we get that hα∨, α + βi > 0, and since α + β is not a root, h(α′)∨, α + β − α′i ≥ 0, and so h(α′)∨, α + βi > 0. Consequently, α + β is dominant with respect to α and α′. Moreover λ − α − β is a sum of simple roots, because λ − β and λ − (α + β − α′) both belong to spanN{α, α′}. This implies that α + β is a root. (cid:3) 16 Let γ be a Tad-weight in (V/g · x0)Gx0 which is not a root. Until Proposition 3.13, we assume that γ is not the sum of two orthogonal simple roots, so that we can speak of the unique simple root α such that γ − α is a root. Lemma 3.10. Let α be the simple root such that γ − α is a root. If γ 6= 2α then α is orthogonal to γ − α. Proof. We can choose a basis of g {Xβ : β root} ∪ {α∨ : α simple root} such that [Xβ, X−β] = β∨ for all positive roots β, and then for all roots β1, β2 denote by cβ1,β2 the scalar such that [Xβ1, Xβ2] = cβ1,β2Xβ1+β2. For example, a Chevalley basis does the job (see [Hum72, Theorem 25.2]). Since Xαv 6= 0, we can assume that Xαv = X−γ+αx0. Assume also, to obtain a contradiction, that hα∨, γ − αi > 0. Hence γ − 2α is a positive root. Since γ is not a root, we have that Xγ−αXαv = XαXγ−αv. From the following identities Xγ−αXαv = 1 cγ−2α,α [Xγ−2α, Xα]Xαv = 1 cγ−2α,α [Xγ−2α, Xα]X−γ+αx0 = cγ−2α,α(cid:0)Xγ−2α[Xα, X−γ+α] − Xα[Xγ−2α, X−γ+α](cid:1)x0 = cγ−2α,−γ+α [Xγ−2α, X−γ+2α]x0 − [Xα, X−α]x0 cγ−2α,α cα,−γ+α cγ−2α,α Xα[Xγ−2α, Xα]v = 1 cγ−2α,α Xα[Xγ−2α, X−γ+α]x0 = [Xα, X−α]x0 = = XαXγ−αv = = it then follows that (3.11) 1 1 cγ−2α,α cγ−2α,−γ+α cγ−2α,α cα,−γ+α cγ−2α,α (γ − 2α)∨ − 2 cγ−2α,−γ+α cγ−2α,α α∨ takes value zero on all λ ∈ F. Since γ ∈ ZF, the expression (3.11) takes value zero on γ, too. Actually, the linear combination (3.11) of coroots does not depend on the choice of the basis of g. Indeed, and cγ−2α,α(γ − α)∨ = [[Xγ−2α, Xα], X−γ+α] = = [Xγ−2α, [Xα, X−γ+α]] − [Xα, [Xγ−2α, X−γ+α]] = = cα,−γ+α(γ − 2α)∨ − cγ−2α,−γ+αα∨ (γ − α)∨ = kγ − 2αk2 kγ − αk2 (γ − 2α)∨ + kαk2 kγ − αk2 α∨. Therefore, since (γ − 2α)∨ and α∨ are linearly independent, (3.11) becomes (3.12) kγ − 2αk2 kγ − αk2 (γ − 2α)∨ + 2 kαk2 kγ − αk2 α∨ which is proportional to γ∨. Since kγk2 is not zero, the expression in (3.11) cannot take value zero on γ, and we have obtained the desired contradiction. (cid:3) Lemma 3.11 ([BCF08, Lemma 3.6]). Let α be the simple root such that γ − α is a positive root. γ − α = β1 + β2 with β1 and β2 positive roots then α + β1 or α + β2 is a root. If 17 Proof. Since Xαv 6= 0, we can assume that Xαv = X−γ+αx0. Next, we claim that if α + β1 6∈ R+ then Xβ2v = 0. Indeed, if Xβ2v were nonzero, then it would be a Tad-weight vector of weight α + β1. Since Xβ2v ∈ g · x0 it would follow by Proposition 3.4(a) that α + β1 ∈ R+. This proves the claim. Similarly, if α + β2 6∈ R+ then Xβ1v = 0. Therefore, if neither α + β1 nor α + β2 is a root, then Xγ−αv = 0. Since γ 6∈ R+, this implies 0 = XαXγ−αv = Xγ−αXαv = Xγ−αX−γ+αx0 which means X−γ+αx0 = 0, a contradiction. (cid:3) Lemma 3.12. Let α be the simple root such that γ − α is a root. Let δ be a simple root and k an integer 2 ≤ k ≤ 4 such that γ − jα − δ is a root for 1 ≤ j ≤ k, jα + δ is a root for 1 ≤ j < k, but kα + δ is not a root. Then γ − kα is orthogonal to every λ ∈ F; and in particular (3.13) kγ − αk2 = (k − 1)kαk2. Proof. We can choose a basis as in the proof of Lemma 3.10 and, since Xαv 6= 0, we can assume that Xαv = X−γ+αx0. First, let us assume also, for simplicity, that k = 2. Then one has the following identities. Xγ−αXαv = 1 cγ−α−δ,δ [Xγ−α−δ, Xδ]X−γ+αx0 = 1 cγ−α−δ,δ(cid:0)Xγ−α−δ[Xδ, X−γ+α] − Xδ[Xγ−α−δ, X−γ+α](cid:1)x0 = cγ−α−δ,−γ+α [Xγ−α−δ, X−γ+α+δ]x0 − [Xδ, X−δ]x0 = = cδ,−γ+α cγ−α−δ,δ cγ−α−δ,δ XαXγ−αv = 1 cγ−α−δ,δ Xα[Xγ−α−δ, Xδ]v = − 1 cγ−α−δ,δ XαXδXγ−α−δv = XαXδ[Xγ−2α−δ, Xα]v = XαXδ[Xγ−2α−δ, X−γ+α]x0 = = − = − = − = − 1 cγ−α−δ,δcγ−2α−δ,α 1 cγ−α−δ,δcγ−2α−δ,α cγ−2α−δ,−γ+α cγ−α−δ,δcγ−2α−δ,α cγ−2α−δ,−γ+αcδ,−α−δ cγ−α−δ,δcγ−2α−δ,α Xα[Xδ, X−α−δ]x0 = [Xα, X−α]x0 We thus find a linear combination of co-roots (3.14) cδ,−γ+α cγ−α−δ,δ (γ − α − δ)∨ − cγ−α−δ,−γ+α cγ−α−δ,δ δ∨ + cγ−2α−δ,−γ+αcδ,−α−δ cγ−α−δ,δcγ−2α−δ,α α∨ which must take value zero on all λ ∈ F. We now compute the coefficients in the above linear combination of coroots, showing they do not depend on the choice of the basis of g. Indeed, cγ−α−δ,δ(γ − α)∨ = [[Xγ−α−δ, Xδ], X−γ+α] = = [Xγ−α−δ, [Xδ, X−γ+α]] − [Xδ, [Xγ−α−δ, X−γ+α]] = = cδ,−γ+α(γ − α − δ)∨ − cγ−α−δ,−γ+αδ∨ 18 and, since also cγ−α−δ,δcγ−2α−δ,α(γ − α)∨ + cγ−α−δ,−γ+αcγ−2α−δ,αδ∨ = = [[[Xγ−2α−δ, Xα], Xδ], X−γ+α] − [[[Xγ−2α−δ, Xα], X−γ+α], Xδ] = = [[Xγ−2α−δ, Xα], [Xδ, X−γ+α]] = = [[Xγ−2α−δ, [Xδ, X−γ+α]], Xα] − [Xγ−2α−δ, [[Xδ, X−γ+α], Xα]] = = [[Xδ, [Xγ−2α−δ, X−γ+α]], Xα] − [Xγ−2α−δ, [[Xδ, X−γ+α], Xα]] = = −cδ,−α−δcγ−2α−δ,−γ+αα∨ − c−γ+α+δ,αcδ,−γ+α(γ − 2α − δ)∨, c−γ+α+δ,αcδ,−γ+α(γ − 2α − δ)∨ = −cγ−2α−δ,αcδ,−γ+α(γ − α − δ)∨ − cδ,−α−δcγ−2α−δ,−γ+αα∨. On the other hand, (γ − α)∨ = kγ − α − δk2 kγ − αk2 (γ − α − δ)∨ + kδk2 kγ − αk2 δ∨ and (γ − 2α − δ)∨ = kγ − α − δk2 kγ − 2α − δk2 (γ − α − δ)∨ − kαk2 kγ − 2α − δk2 α∨. Therefore, since (γ − α − δ)∨ is neither proportional to δ∨ nor to α∨, (3.14) becomes (3.15) kγ − α − δk2 kγ − αk2 (γ − α − δ)∨ + kδk2 kγ − αk2 δ∨ − kαk2 kγ − αk2 α∨ which is proportional to (γ − 2α)∨. For k > 2 the proof is similar. If k = 3, the analog of (3.14) is (γ − α − δ)∨ − cδ,−γ+α cγ−α−δ,δ cγ−2α−δ,−γ+αcδ,−α−δ cγ−α−δ,δcγ−2α−δ,α α∨ + cγ−α−δ,−γ+α cγ−α−δ,δ δ∨ + cγ−3α−δ,−γ+αcδ,−α−δcα,−2α−δ cγ−α−δ,δcγ−2α−δ,αcγ−3α−δ,α α∨ which is proportional to (γ − 3α)∨. If k = 4, we get + − + − + (γ − α − δ)∨ − cδ,−γ+α cγ−α−δ,δ cγ−2α−δ,−γ+αcδ,−α−δ cγ−α−δ,δcγ−2α−δ,α α∨ + cγ−α−δ,−γ+α cγ−α−δ,δ δ∨ + cγ−3α−δ,−γ+αcδ,−α−δcα,−2α−δ cγ−α−δ,δcγ−2α−δ,αcγ−3α−δ,α cγ−4α−δ,−γ+αcδ,−α−δcα,−2α−δcα,−3α−δ cγ−α−δ,δcγ−2α−δ,αcγ−3α−δ,αcγ−4α−δ,α α∨ + α∨ which is proportional to (γ − 4α)∨. Finally, since γ − kα is orthogonal to every λ ∈ F, we have (γ − kα, γ) = 0, which yields (3.13). Indeed, the assumption implies that γ 6= 2α, hence (α, γ − α) = 0 by Lemma 3.10, and 0 = (γ − kα, γ) = kγ − αk2 − (k − 1)kαk2. 19 (cid:3) Proposition 3.13. Suppose γ is not a root and let α be a simple root such that γ − α is a root. Then γ − α is locally the highest root, i.e. the highest root in the root subsystem generated by the simple roots of its support. Proof. I. First we want to prove that γ − α is locally dominant. We can assume that γ − α is not simple. Hence, by Lemma 3.10, α is orthogonal to γ − α. There exists a simple root δ (different from α) such that γ − α − δ is a root. By Proposition 3.9 and Lemma 3.11 α + δ is a root. Since α + δ is a root, hα∨, δi < 0. Therefore, hα∨, γ − α − δi > 0 hence γ − 2α − δ is a root. If moreover 2α + δ is a root, then by sl(2)-theory, hα∨, α + δi ≤ 0 and so hα∨, γ − 2α − δi ≥ 0, whence γ − 3α − δ is a root. If 3α + δ is also a root, then α and δ span a root system of type G2. Consequently, hα∨, γ − 3α − δi = −1 and γ − 4α − δ is a root. Therefore we can apply Lemma 3.12 and obtain that, for some k ≥ 1, γ − kα is orthogonal to every λ ∈ F. This implies that h(α′)∨, γi = 0 for all α′ ∈ supp(γ) \ {α}, whence h(α′)∨, γ − αi ≥ 0 for all such α′. Since α is orthogonal to γ − α, it follows that γ − α is locally dominant. II. To obtain a contradiction, we now assume that γ − α is not locally the highest root, that is, a locally short dominant root with support of non-simply-laced type: - in type Bn, n ≥ 2, the short dominant root is α1 + . . . + αn = ω1; - in type Cn, n ≥ 3, the short dominant root is α1 + 2(α2 + . . . + αn−1) + αn = ω2; - in type F4 the short dominant root is α1 + 2α2 + 3α3 + 2α4 = ω4; - in type G2 the short dominant root is 2α1 + α2 = ω1. By equation (3.13), α is also short and k = 2, in particular the support of γ is not of type G2. Moreover, by Lemma 3.10, α is orthogonal to γ − α. In type Bn and in type F4 this implies that γ is a root. We are left with the case where the support of γ − α is of type Cn. Since α is short, α is orthogonal to γ − α, γ is not a root, and moreover there exists a simple root δ 6= α satisfying the hypothesis of Lemma 3.12 for k = 2, we have that n > 3, δ = α2 and α = α3. This contradicts Lemma 3.11, because α1 and γ − α − α1 are roots, but neither α1 + α nor γ − α1 is a root. (cid:3) The following is Theorem 3.3 for the case that γ is not a root. Corollary 3.14. Let γ be a Tad-weight in (V/g · x0)Gx0 . If γ is not a root, then γ is a spherically closed spherical root of G. Proof. We list all the locally highest roots β and deduce which are the only possible non-roots γ (obtained by adding to β a simple root) satisfying Lemmas 3.10, 3.11 and 3.12. In general, hα∨, βi must be ≥ 0 otherwise α + β ∈ R+. If α is not in the support of β it must be orthogonal to β, and in this case, by Lemma 3.11, β must necessarily be simple. Let us start with β simple, i.e., with support of type A1: β = α1 = 2ω1 gives only or 2α1 α1 + α′ 1. Let us now pass to β not simple and recall that α must necessarily belong to the support of β, moreover by Lemma 3.10 hα∨, βi = 0 and by Lemma 3.12, for all α′ ∈ S \ {α}, h(α′)∨, α + βi = 0. With support of type An, n ≥ 2: β = α1 + . . . + αn = ω1 + ωn gives only, for n = 3, With support of type Bn, n ≥ 2: β = α1 + 2(α2 + . . . + αn) = ω2 if n ≥ 3 (it equals 2ω2 if n = 2) gives only α1 + 2α2 + α3. 2(α1 + . . . + αn) 20 or, for n = 3, α1 + 2α2 + 3α3. With support of type Dn, n ≥ 4: β = α1 + 2(α2 + . . . + αn−2) + αn−1 + αn = ω2 gives only or, for n = 4, and 2(α1 + . . . + αn−2) + αn−1 + αn α1 + 2α2 + 2α3 + α4 α1 + 2α2 + α3 + 2α4 which are equal to 2α1 + 2α2 + α3 + α4 up to an automorphism of the Dynkin diagram. With support of type G2: β = 3α1 + 2α2 = ω2 gives only The remaining cases give no other possibilities: 4α1 + 2α2. - with support of type Cn, n ≥ 3, β = 2(α1 + . . . + αn−1) + αn = 2ω1; - with support of type E6, β = α1 + 2α2 + 2α3 + 3α4 + 2α5 + α6 = ω2; - with support of type E7, β = 2α1 + 2α2 + 3α3 + 4α4 + 3α5 + 2α6 + α7 = ω1; - with support of type E8, β = 2α1 + 3α2 + 4α3 + 6α4 + 5α5 + 4α6 + 3α7 + 2α8 = ω8; - with support of type F4, β = 2α1 + 3α2 + 4α3 + 2α4 = ω1. (cid:3) 3.3. Further properties of Tad-weights in (V/g · x0)Gx0 . After Theorem 3.3 the only possible Tad- weights in (V/g · x0)Gx0 are spherically closed spherical roots of G, but each of them occur only under special conditions which we are going to describe. The first statement is indeed a refinement of Theorem 3.3. Recall the notion of compatibility with Sp (see axiom (S) of Definition 2.5 and Remark 2.6.1). Theorem 3.15. If γ is a Tad-weight in (V/g · x0)Gx0 then γ is a spherically closed spherical root of G compatible with Sp(Γ). Proof. If γ = α1 + α2 + . . . + αn with support of type An, then {α2, α3, . . . , αn−1} ⊂ Sp(Γ). This follows from part I of the proof of Proposition 3.7. If γ = α1 + 2α2 + α3 with support of type A3, then {α1, α3} ⊂ Sp(Γ). This follows by Lemma 3.12 (α = α2, δ = α1 and k = 2). If γ = α1 + α2 + . . . + αn with support of type Bn, then {α2, α3, . . . , αn−1} ⊂ Sp(Γ) and αn 6∈ Sp(Γ). The former follows from part I of the proof of Proposition 3.7. For the latter, we can assume that Xαn v = 0 and Xα1v = X−γ+αn x0 nonzero, which implies αn 6∈ Sp. If γ = 2(α1 + . . . + αn) with support of type Bn, then {α2, . . . , αn} ⊂ Sp(Γ). This follows by Lemma 3.12 (α = α1, δ = α2 and k = 2). If γ = α1 + 2α2 + 3α3 with support of type B3, then {α1, α2} ⊂ Sp(Γ). This follows by Lemma 3.12 (α = α3, δ = α2 and k = 3). If γ = α1 + 2(α2 + . . . + αn−1) + αn with support of type Cn, then {α3, α4, . . . , αn} ⊂ Sp(Γ). This follows from part V of the proof of Proposition 3.7. If γ = 2(α1 + . . . + αn−2) + αn−1 + αn with support of type Dn, then {α2, . . . , αn} ⊂ Sp(Γ). This follows by Lemma 3.12 (α = α1, δ = α2 and k = 2). If γ = α1 + 2α2 + 3α3 + 2α4 with support of type F4, then {α1, α2, α3} ⊂ Sp(Γ). This follows from part V of the proof of Proposition 3.7. If γ = 4α1 + 2α2 with support of type G2, then α2 ∈ Sp(Γ). This follows by Lemma 3.12 (α = α1, (cid:3) δ = α2 and k = 4). 21 Proposition 3.16. If γ is not a simple root then the Tad-eigenspace (V/g · x0) Gx0 (γ) has dimension ≤ 1. Proof. If γ is a root (not simple), recall that there exist two simple roots, say α1 and α2, such that γ − α1 and γ − α2 is a root, and γ − α is not a root for all α ∈ S \ {α1, α2}. In particular, for all α ∈ S \ {α1, α2}, we necessarily have Xαv = 0. By adding to v a suitable scalar multiple of X−γx0, we can assume that also Xα2v = 0. Moreover, by choosing a suitable scalar multiple, we can assume that Xα1 v = X−γ+α1x0. If γ is neither a root nor the sum of two orthogonal simple roots, recall that there exists a simple root α1 such that γ − α1 is a root, and γ − α is not a root for all α ∈ S \ {α1}. In particular, for all α ∈ S \ {α1}, we necessarily have Xαv = 0. Therefore, by choosing a suitable scalar multiple, we can assume that Xα1v = X−γ+α1 x0. In both cases we claim that under the above assumptions v is uniquely determined. Indeed, if v1 and v2 are two vectors in V of Tad-weight γ fulfilling the above conditions, then Xα(v1 − v2) = 0 for all α ∈ S, which implies v1 = v2. We are left with only one case: the spherical root γ = α + α′ with support of type A1 × A1. We can assume Xαv = X−α′ x0. For all i ∈ {1, . . . , r}, dim V(λi)(γ) ≤ 1, and the condition Xαv = X−α′ x0 uniquely determines every component vi ∈ V(λi) of v. (cid:3) In this section we prove the following. 4. THE WEIGHT SPACES OF TX0HΓ Theorem 4.1. If Γ is a free monoid of dominant weights, then TX0HΓ is a multiplicity-free Tad-module of which all the weights belong to Σsc(G). Moreover, if γ ∈ Σsc(G) occurs as a Tad-weight in TX0HΓ then γ is N-adapted to Γ. Proof. The assertion that all Tad-weights of TX0HΓ belong to Σsc(G) follows from the inclusion TX0HΓ ֒→ (V/g · x0)Gx0 and Theorem 3.3, while the assertion that the weight space (TX0HΓ)(γ) has dimension at most one follows from Proposition 3.16 if γ /∈ S, and from Proposition 4.6 below if γ ∈ S. The statement that if γ ∈ Σsc(G) is a Tad-weight in TX0HΓ, then γ is N-adapted to Γ, is contained in Proposition 4.6 for γ ∈ S and is shown in Section 4.3 for γ /∈ S. (cid:3) Recall from Proposition 3.1 that MΓ is Tad-equivariantly isomorphic to an open subscheme of HΓ. Because every Tad-weight in TX0MΓ ≃ TX0HΓ is an element of Σsc(G) (see Theorem 3.3) we obtain the following converse to the second statement in Theorem 4.1. Corollary 4.2. Let Γ be a free monoid of dominant weights and let σ ∈ Σsc(G). If σ is N-adapted to Γ, then σ is a Tad-weight in TX0MΓ. Proof. Let X be an affine spherical G-variety with Γ(X) = Γ and ΣN(X) = {σ}, and let MX be its root monoid. Recall that ΣN(X) is the basis of the saturation of MX. Let {a1, a2, . . . , ak} be a subset of N such that {a1σ, a2σ, . . . , akσ} is the minimal set of generators of MX. By [AB05, Proposition 2.13], the Tad-orbit closure of X, seen as a closed point of MΓ, is Spec(k[−MX]). A straightforward computation using the basic theory of semigroup rings (see, e.g., [MS05, §7.1]) shows that TX0(Tad · X) ≃ ka1σ ⊕ ka2σ ⊕ . . . ⊕ kakσ as Tad-modules, where we used kaiσ for the one-dimensional Tad-representation of weight aiσ. We claim that one of the ai is equal to 1 (and consequently that MX is generated by {σ}). We show this by contradiction. Suppose that all of the ai are at least 2. Then k ≥ 2, since otherwise σ would not be in ZMX. Since TX0(Tad · X) ⊂ TX0MΓ ⊂ (V/g · x0)Gx0 , it then follows from Theorem 3.3 that {σ, a1σ, a2σ} ⊂ Σsc(G). By the classification of spherically closed spherical roots (cf. Proposition 2.4) this is impossible: only the double or half of a spherically closed spherical root can be a spherically closed spherical root, and never both. (cid:3) 22 As before, Γ will be a free monoid of dominant weights with basis F = {λ1, λ2, . . . , λr}. If λ ∈ F, then we will write λ# for the corresponding element of the dual basis of (ZΓ)∗; in other words, for all µ ∈ F \ {λ} we have hλ#, µi = 0, whereas hλ#, λi = 1. Recall that E(Γ) is defined in (2.3). Because Γ is free, we have that E(Γ) is the dual basis to F: For λ ∈ F we put E(Γ) = {λ# ∈ (ZΓ)∗ : λ ∈ F}. zλ := x0 − vλ. 4.1. The extension criterion. We recall from [PVS15] a criterion which allows to decide whether a Tad-eigenvector [v] ∈ (V/g · x0)Gx0 ≃ H0(G · X0, NX0V)G belongs to the subspace TX0HΓ ≃ H0(X0, NX0V)G. We denote by X≤1 0 ⊂ X0 the union of G · x0 with all G-orbits of X0 that have codimension 1. By is an open subset of X0. The following proposition is a special case of [Bri10, Lemma 1.14] X≤1 0 [Bri13, Lemma 3.9]. Together with Theorem 4.5 it gives the aforementioned criterion. Proposition 4.3. A section s ∈ H0(G · X0, NX0V) extends to X0 if and only if it extends to X≤1 0 . We recall that the orbit structure of X0 is well-understood [VP72, Theorem 8]. describe the orbits of codimension 1 (see, e.g., [PVS12, Proposition 3.1] for details). It is easy to Proposition 4.4. The G-orbits of codimension 1 in X0 are exactly the orbits G · zλ where λ is an element of F that satisfies the following property: for every α ∈ S such that hα∨, λi 6= 0 there exists µ ∈ F \ {λ} such that hα∨, µi 6= 0. Theorem 4.5 ([PVS15, Theorem 2.5]). Let v ∈ V be a Tad-eigenvector of weight γ such that 0 6= [v] ∈ (V/g · x0)Gx0 . Let λ ∈ F. Recall that zλ = x0 − vλ. Assume that zλ ∈ X≤1 and put Z := G · x0 ∪ G · zλ. 0 Put a := hλ#, γi. Denote by s ∈ H0(G · x0, NX0V)G the G-equivariant section such that s(x0) = [v]. A) If a ≤ 0, then s extends to an element of H0(Z, NX0V)G. B) If a > 1, then s does not extend to an element of H0(Z, NX0V)G. C) If a = 1, then the following are equivalent: i) s extends to an element of H0(Z, NX0V)G; ii) there exist v ∈ V(λ) such that [v] = [ v] as elements of V/g · x0. 4.2. The spherical root γ = α ∈ S. In this section, we discuss the Tad-weight space (TX0HΓ)(α), where α is a simple root. Specifically, we will prove the following proposition, which is a special case of Theorem 4.1. Proposition 4.6. If α is a simple root then dim(TX0HΓ)(α) ≤ 1. Moreover, if dim(TX0HΓ)(α) = 1 then α is N-adapted to Γ. The proof of Proposition 4.6 will be given on page 24. We first need a few lemmas and introduce notation we will use for the remainder of this section. Put F(α) := {λ ∈ F : hα∨, λi 6= 0}. We order the elements of F such that for F(α) = {λ1, λ2, . . . , λp} for some p ≤ r. Then F \ F(α) = {λp+1, λp+2, . . . , λr}. Lemma 4.7. For every i ∈ {1, 2, . . . , p}, put vi = X−αvλi. Then v1 + v2 + . . . + vp spans the Tad-weight space of weight α in g · x0. If α ∈ ZΓ, then (V/g · x0) Gx0 (α) = h[v1], [v2], . . . , [vp−1]ik. 23 Proof. By elementary highest weight theory, the Tad-weight space in V of weight α is spanned by {v1, v2, . . . , vp}, and the intersection of this weight space with g · x0 is the line spanned by X−αx0 = v1 + v2 + . . . + vp. A straightforward application of Proposition 3.4 shows that [vi] ∈ (V/g · x0)Gx0 for every i ∈ {1, 2, . . . , p − 1}. (cid:3) Lemma 4.8. Suppose α ∈ ZΓ and F(α) ≥ 2. Let λ ∈ F. If hλ#, αi > 0, then G · zλ has codimension 1 in X0. Proof. We will apply Proposition 4.4. Since α ∈ ZΓ and Γ is free, there exists a partition F = F1 ∪ F2 of F and for every µ ∈ F a unique nonnegative integer aµ such that (4.1) α = ∑ µ∈F1 aµµ − ∑ µ∈F2 aµµ. By assumption λ ∈ F1 and aλ = hλ#, αi > 0. Let β ∈ S \ {α} such that hβ∨, λi 6= 0. Then, since F ⊂ Λ+ and hβ∨, αi ≤ 0, it follows from the expression (4.1) that aµhβ∨, µi ≥ aλhβ∨, λi > 0. ∑ µ∈F2 In particular, there exists µ ∈ F2 such that hβ∨, µi 6= 0. Furthermore, whether hα∨, λi is zero or not, by the assumption that F(α) ≥ 2, there exists µ ∈ F \ {λ} such that hα∨, µi 6= 0. This finishes the proof. (cid:3) Lemma 4.9. Let α be a simple root. Recall that F(α) = {λ ∈ F : hα∨, λi 6= 0} and put E(α) := {δ ∈ E(Γ) : hδ, αi = 1}. Then dim(TX0HΓ)(α) ≤ 1 and if dim(TX0HΓ)(α) = 1 then (i) α ∈ ZΓ; (ii) F(α) ≥ 2; (iii) hδ, αi ≤ 1 for all δ ∈ E(Γ); (iv) E(α) ≤ 2; (v) If E(α) = 2 then E(α) = {λ# ∈ E(Γ) : λ ∈ F(α)}. Gx0 Proof. Let us assume that dim(TX0HΓ)(α) ≥ 1. Let [v] be a nonzero element of (V/g · x0) (α) such that the G-equivariant section s ∈ H0(G · x0, N )G defined by s(x0) = [v] extends to X0. By Propo- sition 3.4 and Lemma 4.7, conditions (i) and (ii) hold. Lemma 4.8 and Theorem 4.5 then imply (iii). We now prove (iv). If E(α) ≥ 3, then by Theorem 4.5 and Lemma 4.8, there exist at least three elements λ, µ, ν ∈ F(α) such that there exist yλ ∈ V(λ), yµ ∈ V(µ) and yν ∈ V(ν) for which [v] = [yλ] = [yµ] = [yν] ∈ V/g · x0. This is impossible by Lemma 4.7 and (iv) is proved. We turn to (v). Suppose E(α) = {λ#, µ#}. By Lemma 4.8 and Theorem 4.5, there exist yλ ∈ V(λ) and yµ ∈ V(µ) such that [v] = [yλ] = [yµ] ∈ V/g · x0. Using Lemma 4.7 again, (v) follows. Finally, we show that dim(TX0HΓ)(α) ≤ 1. Since α ∈ ZΓ, there is at least one λ ∈ E(α). Lemma 4.8 and Theorem 4.5 again imply that [v] = [yλ] for some yλ ∈ V(λ), which finishes the proof. (cid:3) Remark 4.10. By Corollary 4.2 and the proof of Proposition 4.9 below, the preceding lemma gives alternative conditions for α to be N-adapted to Γ when Γ is free. We list them as a separate lemma, since they seem easier to check then those in Corollary 2.17. Proof of Proposition 4.6. Lemma 4.9 says that dim(TX0HΓ)(α) ≤ 1. We assume conditions (i) -- (v) in Lemma 4.9 and deduce conditions (1), (2), (4a), (4b) and (4c) in Corollary 2.17. For (1) and (4c), there is nothing to show. For the spherical root α, (2) follows from (1). To show (4a), we first claim that E(α) contains at least one element. Indeed, α ∈ ZΓ and hλ#, αi > 0 for at least one λ ∈ F, for otherwise −α would be a dominant weight. The claim now follows from (iii). Next, suppose 24 λ# ∈ E(α). Clearly λ# ∈ a(α). We claim that α∨ − λ# 6= λ#. Otherwise, we would have λ# = 1 2 α∨, which would contradict (ii). This shows a(α) ≥ 2. Now, if a(α) had a third element, then E(α) would have two elements, say λ# and µ#, with α∨ − λ# 6= µ#. But this yields a contradiction: by (v), we have that hα∨, λi = hα∨, µi = 1 and then that α∨ − λ# takes the same values as µ# on F. We have deduced (4a). Finally, (4b) is clear since a(α) = {λ#, α∨ − λ#} for some λ ∈ F(α). (cid:3) 4.3. The non-simple spherical roots. To complete the proof of Theorem 4.1, we show in this sec- tion that if γ is a spherically closed spherical root, which is not a simple root and which occurs as a Tad-weight in TX0HΓ, then γ is N-adapted to Γ. We recall that conditions (1) and (2) of Corollary 2.17 follow from Theorem 3.15. We now verify condition (3): if δ ∈ E(Γ) such that hδ, γi > 0 then there exists β ∈ S \ Sp(Γ) such that β∨ is a positive multiple of δ. The argument is the same for all the non-simple spherical roots γ. Let v ∈ V be a Tad-eigenvector of weight γ such that 0 6= [v] ∈ (V/g · x0)Gx0 . Let λ ∈ F. Recall that zλ = x0 − vλ and put a = hλ#, γi. Assume a > 0. We claim that under this assumption, codimX0 G · zλ ≥ 2. Indeed, if codimX0 G · zλ were 1, then by Theorem 4.5(B) a = 1 and by Theorem 4.5(C) there would exist v ∈ V(λ) such that [v] = [ v] as elements of V/g · x0. Therefore, there would exist α ∈ S such that γ − α ∈ R+, and such that Xα v is nonzero and is equal to X−γ+αx0 up to a nonzero scalar multiple. This would imply X−γ+αvλ 6= 0 and X−γ+αvµ = 0 for all µ ∈ F \ {λ}, and therefore that there exists α′ ∈ S such that h(α′)∨, λi > 0 and h(α′)∨, µi = 0 for all µ ∈ F \ {λ}, which gives a contradiction with Proposition 4.4 and proves the claim. The fact that codimX0 G · zλ ≥ 2 means that there exists β ∈ S such that hβ∨, λi > 0 and hβ∨, µi = 0 for all µ ∈ F \ {λ}. This says exactly that the restriction of β∨ to ZΓ is a positive multiple of λ#, which is condition (3). We continue with the remaining conditions of Corollary 2.17. Condition (4) does not apply to non-simple spherical roots. Condition (5) follows using the analysis of Section 3. Indeed, we have shown that if [v] is a nonzero Tad-eigenvector of weight 2α in (V/g · x0)Gx0 , with α ∈ S, then Xαv is a (nonzero) scalar multiple of X−αx0. Since 2α ∈ ZΓ, there exists λ ∈ F such that hα∨, λi > 0 and hλ#, 2αi > 0. By the argument we used for condition (3), λ is the unique element of F which is non-orthogonal to α. It follows that we actually have that Xαv is a nonzero scalar multiple of X−αvλ. This implies that the T-eigenspace of weight λ − 2α in V(λ) is nonzero, hence hα∨, λi ≥ 2. Consequently hα∨, λi ∈ {2, 4} and hα∨, µi = 0 for all µ ∈ F \ {λ}, hence α∨ takes an even value on every element of ZΓ. Condition (6) follows analogously from Section 3. Indeed, we have shown that if [v] is a nonzero Tad-eigenvector of weight α + α′ in (V/g · x0)Gx0 , with α and α′ orthogonal simple roots, then Xαv, if nonzero, is a scalar multiple of X−α′ x0, and Xα′ v, if nonzero, is a scalar multiple of Xαx0. Since α + α′ ∈ ZΓ, there exists λ ∈ F such that hα∨, λi > 0 and hλ#, α + α′i > 0. By the argument we used for condition (3), λ is the unique element of F which is non-orthogonal to α. Then Xαv 6= 0. Indeed if it were 0, then Xα′ v would be nonzero, hence scalar multiple of X−αvλ, which yields a contradiction: 0 = Xα′ Xαv = XαXα′ v = XαX−αvλ 6= 0, Therefore Xαv = X−α′ x0, and if h(α′)∨, µi 6= 0 then the T-eigenspace of weight µ − α − α′ in V(µ) is nonzero, hence also hα∨, µi 6= 0. This implies that α′ is non-orthogonal to λ and orthogonal to µ for all µ ∈ F \ {λ}. Therefore α∨ and (α′)∨ are equal on every element of ZΓ. This completes the proof of Theorem 4.1. 25 Remark 4.11. The information given in this remark is not needed for our results. We include it because it gives explicit conditions on F for each spherically closed spherical root γ, which is not a simple root, to occur as a Tad-weight in TX0HΓ, that is, to be N-adapted to Γ. For each spherically closed spherical root γ, there exists α ∈ S such that hα∨, γi > 0. If γ is a Tad-weight in TX0HΓ, then γ ∈ ZΓ, and so there exits λ ∈ F such that hα∨, λi > 0 and hλ#, γi > 0. If γ is not a simple root, then by the argument above showing that γ satifies condition (3) of Corollary 2.17, we have that λ is the only element of F which is not orthogonal to α, that is, bλ# = α∨ on ZΓ for some positive integer b. We now list, for each γ, the possibilities for λ#. (1) If γ = 2α, with α a simple root, then locally γ = 4ω. In this case α∨ = bλ# with b ∈ {2, 4}. (2) If γ = α + α′, with α and α′ two orthogonal simple roots, then locally γ = 2ω + 2ω′. In this case α∨ = (α′)∨ = bλ# with b ∈ {1, 2}. (3) If γ = α1 + α2 + . . . + αn with support of type An with n ≥ 2, then locally γ = ω1 + ωn. In this case, α∨ = λ# with α ∈ {α1, αn}. (4) If γ = α1 + 2α2 + α3 with support of type A3, then locally γ = 2ω2. In this case, we have (5) If γ = α1 + . . . + αn with support of type Bn with n ≥ 2, then locally γ = ω1. Here α∨ 1 = λ#. (6) If γ = 2α1 + 2α2 + . . . + 2αn with support of type Bn with n ≥ 2, then locally γ = 2ω1. (7) If γ = α1 + 2α2 + 3α3 with support of type B3, then locally γ = 2ω3. Here α∨ 3 = bλ# with (8) If γ = α1 + 2α2 + 2α3 + . . . + 2αn−1 + αn with support of type Cn with n ≥ 3, then locally α∨ 2 = bλ# with b ∈ {1, 2}. Here α∨ 1 = bλ#, with b ∈ {1, 2}. b ∈ {1, 2}. γ = ω2. Here α∨ 2 = λ#. (9) If γ = 2α1 + 2α2 + . . . + 2αn−2 + αn−1 + αn with support of type Dn with n ≥ 4, then locally γ = 2ω1. Here α∨ 1 = bλ# with b ∈ {1, 2}. (10) If γ = α1 + 2α2 + 3α3 + 2α4 with support of type F4, then locally γ = ω4. Here α∨ (11) If γ = 4α1 + 2α2 with support of type G2, then locally γ = 2ω1. Here α∨ 4 = λ#. 1 = bλ# with b ∈ {1, 2}. (12) If γ = α1 + α2 with support of type G2, then locally γ = −ω1 + ω2. Here α∨ 2 = λ#. In this section we prove the following. 5. THE IRREDUCIBLE COMPONENTS OF MΓ Theorem 5.1. Let Γ be a free monoid of dominant weights. Then the Tad-orbit closures in MΓ, equipped with their reduced induced scheme structure, are affine spaces. The proof is given below. By [AB05, Proposition 2.13] this theorem has the following formal consequence. Corollary 5.2. If X is an affine spherical variety with free weight monoid, then its root monoid MX is free too. Another consequence is that Conjecture 1.1 holds for free monoids. Corollary 5.3. If Γ is a free monoid of dominant weights then the irreducible components of MΓ, equipped with their reduced induced scheme structure, are affine spaces. Proof. Since the Tad-orbits in MΓ are in bijection with isomorphism classes of affine spherical G- varieties, by [AB05, Theorem 1.12] and there are only finitely many such isomorphism classes, by [AB05, Corollary 3.4], we have that every irreducible component Z of MΓ contains a dense Tad-orbit. It then follows from Theorem 5.1 that Z, equipped with its reduced induced scheme structure, is an affine space. (cid:3) 26 Proof of Theorem 5.1. Let X be an affine spherical G-variety of weight monoid Γ, seen as a (closed) point in MΓ. By [AB05, Corollary 2.14], we know that the normalization of Tad · X is an affine space. It is therefore enough to show that Tad · X is smooth at X0. We do this by showing that (5.1) dim TX0(Tad · X) = dim Tad · X. Recall that ΣN(X) is the basis of the monoid obtained by saturation of the root monoid MX. To deduce (5.1) we make use of Theorem 4.1: the Tad-weights in TX0(Tad · X) ⊆ TX0MΓ are spherical roots N-adapted to Γ, each one occurring with multiplicity 1. This, together with the fact that every Tad-weight in TX0(Tad · X) has to be an element of the root monoid MX, and hence a nonnegative integer linear combination of elements of ΣN(X), gives (5.1) once we prove Proposition 5.4 below. Indeed, applying this proposition with Σ = ΣN(X) yields that the Tad-weights in TX0(Tad · X) belong to ΣN(X), while dim Tad · X = ΣN(X) by [AB05, Proposition 2.13]. (cid:3) Proposition 5.4. Let Σ be a subset of Σsc(G) such that every γ ∈ Σ is N-adapted to Γ. If σ ∈ Σsc(G) ∩ NΣ is N-adapted to Γ, then σ ∈ Σ. Proof. First of all, σ (of spherically closed type) must be compatible with Sp(Γ) and is a nonneg- ative integer linear combination of other elements of Σsc(G) that satisfy the same compatibility condition. This gives strong restrictions. Indeed, σ can only be the sum of two simple roots (equal or not, orthogonal or not). All the other types of spherical roots have support that nontrivially intersects Sp(Γ), and they can be excluded by a straighforward if somewhat lengthy case-by-case verification. Moreover, σ cannot be the double of a simple root, say 2α, with α ∈ Σ, since α and 2α cannot both be N-adapted to Γ. Indeed, if 2α is N-adapted to Γ then, since hα∨, 2αi > 0 and α∨ ∈ Γ∨, there exists δ ∈ E(Γ) such that hδ, 2αi > 0. Condition (3) of Corollary 2.17 tells us that α∨ is a positive multiple of δ. By condition (5) of the same corollary, α∨ is not primitive in (ZΓ)∗. If now α ∈ ZΓ, then it follows from hα∨, αi = 2 that α∨ = 2δ on ZΓ. Hence δ is the only element of a(α) and α is not N-adapted to Γ. Analogously, σ cannot be the sum of two orthogonal simple roots, say α + α′, with α and α′ in Σ. Indeed, since α + α′ is adapted to Γ and hα∨, αi 6= h(α′)∨, αi, α cannot belong to ZΓ. Finally, let σ be the sum of two nonorthogonal simple roots, say α1 + α2, with α1 and α2 in Σ. Take δ ∈ E(Γ) with hδ, σi > 0. Such a δ exists because hα∨ 2 , σi is positive, σ ∈ ZΓ and Γ ⊂ Λ+. Then δ must be positive on at least one of the two simple roots α1 or α2. Suppose it is positive on α1. Then δ ∈ a(α1), since α1 is N-adapted to Γ, hence δ takes the value 1 on α1. By condition (3) of Corollary 2.17 it follows that α∨ 1 = 2δ, which is not possible if α1 is N-adapted to Γ. (cid:3) 1 , σi or hα∨ Remark 5.5. While the reduced induced scheme structure is the only natural scheme structure on the Tad-orbit closures of Theorem 5.1, there is at least one other natural scheme structure on the irreducible components of MΓ, namely the one given by the primary ideals of k[MΓ] associated to minimal primes. One can ask whether Conjecture 1.1 remains true for that scheme structure. Another natural question is whether or when MΓ is in fact a reduced scheme. We note that the tangent space TX0MΓ might fail to detect the "non-reducedness" of MΓ. For example, the two affine schemes Spec(k[x, y]/hxyi) and Spec(k[x, y]/hx2yi) have the same tangent space at the point corresponding to the maximal ideal hx, yi. REFERENCES [AB05] Valery Alexeev and Michel Brion, Moduli of affine schemes with reductive group action, J. Algebraic Geom. 14 (2005), no. 1, 83 -- 117. MR 2092127 (2006a:14017) [ACF14] Roman Avdeev and St´ephanie Cupit-Foutou, On the irreducible components of some moduli schemes for affine multiplicity-free varieties, arXiv:1406.1713v1 [math.AG], 2014. 27 [Ahi83] Dmitry Ahiezer, Equivariant completions of homogeneous algebraic varieties by homogeneous divisors, Ann. Global Anal. Geom. 1 (1983), no. 1, 49 -- 78. MR 739893 (85j:32052) [BCF08] Paolo Bravi and St´ephanie Cupit-Foutou, Equivariant deformations of the affine multicone over a flag variety, Adv. Math. 217 (2008), no. 6, 2800 -- 2821. MR 2397467 (2009a:14061) [BL11] Paolo Bravi and Domingo Luna, An introduction to wonderful varieties with many examples of type F4, J. Algebra 329 (2011), no. 1, 4 -- 51. MR 2769314 (2012f:14102) [Bou68] Nicolas Bourbaki, ´El´ements de math´ematique. Fasc. XXXIV. Groupes et alg`ebres de Lie. Chapitre IV: Groupes de Cox- eter et syst`emes de Tits. Chapitre V: Groupes engendr´es par des r´eflexions. Chapitre VI: syst`emes de racines, Actualit´es Scientifiques et Industrielles, No. 1337, Hermann, Paris, 1968. MR 0240238 (39 #1590) [BP11] Paolo Bravi and Guido Pezzini, Primitive wonderful varieties, arXiv:1106.3187v1 [math.AG], 2011. [Bri89] Michel Brion, On spherical varieties of rank one (after D. Ahiezer, A. Huckleberry, D. Snow), Group actions and invariant theory (Montreal, PQ, 1988), CMS Conf. Proc., vol. 10, Amer Math. Soc., Providence, RI, 1989, pp. 31 -- 41. MR 1021273 (91a:14028) [Bri90] [Bri10] [Bri13] , Vers une g´en´eralisation des espaces sym´etriques, J. Algebra 134 (1990), no. 1, 115 -- 143. MR 1068418 (91i:14039) , Introduction to actions of algebraic groups, Les cours du CIRM 1 (2010), no. 1, 1 -- 22. , Invariant Hilbert schemes, Handbook of moduli. Vol. I, Adv. Lect. Math. (ALM), vol. 24, Int. Press, Somerville, MA, 2013, pp. 64 -- 117. MR 3184162 [Cam01] Romain Camus, Vari´et´es sph´eriques affines lisses, Ph.D. thesis, Institut Fourier, Grenoble, 2001. [Cup10] St´ephanie Cupit-Foutou, Wonderful varieties: a geometrical realization, arXiv:0907.2852v3 [math.AG], 2010. [Hum72] James E. Humphreys, Introduction to Lie algebras and representation theory, Springer-Verlag, New York-Berlin, 1972, Graduate Texts in Mathematics, Vol. 9. MR 0323842 (48 #2197) [Hum75] , Linear algebraic groups, Springer-Verlag, New York, 1975, Graduate Texts in Mathematics, No. 21. MR 0396773 (53 #633) [Jan07] S´ebastien Jansou, D´eformations des cones de vecteurs primitifs, Math. Ann. 338 (2007), no. 3, 627 -- 667. MR 2317933 (2008d:14069) [Kno91] Friedrich Knop, The Luna-Vust theory of spherical embeddings, Proceedings of the Hyderabad Conference on Algebraic Groups (Hyderabad, 1989) (Madras), Manoj Prakashan, 1991, pp. 225 -- 249. MR 1131314 (92m:14065) , Automorphisms, root systems, and compactifications of homogeneous varieties, J. Amer. Math. Soc. 9 (1996), [Kno96] no. 1, 153 -- 174. MR 1311823 (96c:14037) [Los09a] Ivan V. Losev, Proof of the Knop conjecture, Ann. Inst. Fourier (Grenoble) 59 (2009), no. 3, 1105 -- 1134. MR 2543664 (2010j:14091) [Los09b] , Uniqueness property for spherical homogeneous spaces, Duke Math. J. 147 (2009), no. 2, 315 -- 343. MR MR2495078 (2010c:14055) [Lun01] Domingo Luna, Vari´et´es sph´eriques de type A, Publ. Math. Inst. Hautes ´Etudes Sci. (2001), no. 94, 161 -- 226. MR 1896179 (2003f:14056) [MS05] Ezra Miller and Bernd Sturmfels, Combinatorial commutative algebra, Graduate Texts in Mathematics, vol. 227, Springer-Verlag, New York, 2005. MR 2110098 (2006d:13001) [Pez10] Guido Pezzini, Lectures on spherical and wonderful varieties, Les cours du CIRM 1 (2010), no. 1, 33 -- 53. , On reductive automorphism groups of regular embeddings, arXiv:1206.0846v2 [math.AG], 2013. [Pez13] [PVS12] Stavros Argyrios Papadakis and Bart Van Steirteghem, Equivariant degenerations of spherical modules for groups of type A, Ann. Inst. Fourier (Grenoble) 62 (2012), no. 5, 1765 -- 1809, extended version at arXiv:1008.0911v3 [math.AG]. MR 3025153 , Equivariant degenerations of spherical modules: part II, arXiv:1505.07446v1 [math.AG], 2015 [PVS15] [Tim11] Dmitry A. Timashev, Homogeneous spaces and equivariant embeddings, Encyclopaedia of Mathematical Sciences, vol. 138, Springer, Heidelberg, 2011, Invariant Theory and Algebraic Transformation Groups, 8. MR 2797018 (2012e:14100) `Ernest B. Vinberg and Vladimir L. Popov, A certain class of quasihomogeneous affine varieties, Izv. Akad. Nauk SSSR Ser. Mat. 36 (1972), 749 -- 764, English translation in Math. USSR Izv. 6 (1972), 743 -- 758. MR 0313260 (47 #1815) [VP72] [VS13] Bart Van Steirteghem, Various interpretations of the root system(s) of a spherical variety, Oberwolfach Rep. 10 (2013), no. 2, 1464 -- 1467. [Was96] Benjamin Wasserman, Wonderful varieties of rank two, Transform. Groups 1 (1996), no. 4, 375 -- 403. MR 1424449 (97k:14051) 28 DIPARTIMENTO DI MATEMATICA "GUIDO CASTELNUOVO", "SAPIENZA" UNIVERSIT `A DI ROMA, PIAZZALE ALDO MORO 5, 00185 ROMA, ITALY E-mail address: [email protected] DEPARTMENT OF MATHEMATICS, MEDGAR EVERS COLLEGE - CITY UNIVERSITY OF NEW YORK, 1650 BEDFORD AVE., BROOKLYN, NY 11225, USA E-mail address: [email protected] 29
1111.1074
1
1111
2011-11-04T09:18:29
Some Further Remarks on the Local Fundamental Group Scheme
[ "math.AG" ]
We prove that the Local Fundamental Group Scheme satisfies the Lefschetz - Bott theorems in characteristic p. The proofs are standard applications of the Enriques-Severi -Zariski-Serre vanishing theorems and known facts about the p-curvature
math.AG
math
Some Further Remarks on the Local Fundamental Group Scheme V.B. Mehta∗ April 4, 2018 Abstract We prove that the Local Fundamental Group Scheme satisfies the Lefschetz- Bott theorems in characteristic p. The proofs are standard applications of the Enriques-Severi-Zariski-Serre vanishing theorems and known facts about the p-curvature . 1 Introduction The results of this paper were already proved by Indranil Biswas and Yogesh Holla ( arXiv.math /0603299 v1 [math AG], 13 March 2006, arXiv.math /0603299v1 [math AG], 1st May 2007, and finally published in the Journal of Algebraic Geometry, Vol. 16, No.3, 2007, pages [547-597]. The present work was done later ( around late 2006-early 2007) but independently. A preliminary version of the present work was put on the arXiv in January 2007. This has been withdrawn as soon as we were informed that Biswas and Holla had already put up their paper on the arXiv in March 2006. However since our proofs are shorter and more suited for future applica- tions ( "On the Grothendieck-Lefschetz Theorem for a family of Varieties " , with Marco Antei, in preparation ), we are putting a shortened version on the arXiv. The Fundamental Group Scheme was introduced by Madhav Nori in [N1,N2]. In order to prove the conjectures made in [loc.cit], S.Subramanian ∗email:[email protected] 1 and the present author had introduced the "Local Fundamental Group Scheme ", denoted by πloc(X)[MS1], which is the infinitesimal part of π(X),[MS2]. We had also introduced the notion of a F -trivial vector bundle on a variety in characteristic p[loc.cit.] One may ask whether the Lefschetz-Bott theorems hold for πloc(X). In other words, let X be a smooth projective variety over an algebraically closed filed of characteristic p,and let H be a very ample line bundle on X. The question is if πloc(Y ) → πloc(X) is a surjection if dim X ≥ 2 and degree Y ≥ n0, where n0 is an integer depending only on X. Sim- iliarly , if dim X ≥ 3,the question is if πloc(Y ) → πloc(X) is an isomorphism if deg Y ≥ n1, where n1 is an integer depending only on X. We give a positive answer to both these questions. The methods only in- volve applying the lemma of Enriques-Severi-Zariski-Serre ( E-S-Z-S) over and over again. We also heavily use the well-known facts about the p- curvature of integrable connections in characteristic p [K1]. The notation is the same as in [MS2],which is briefly recalled. We thank Vijaylaxmi Trivedi and Vittorio for helpful discussions. This work was completed while the au- thor was visiting the ICTP, Trieste as a Senior Research Associate. We would like to thank the ICTP for its support and hospitality during that period. Our method of proof is heavily influenced by a paper of Karen Smith [S], especially [Thm. 3.5]. All the facts about Tannaka Categories that we use may be found in [N1]. 2 The first theorem Theorem 2.1. Let X be a non-singular projective variety over an alge- braically closed field of characteristic p,of dimension ≥ 2. Let H be a very ample line bundle on X. Then there exists an integer n0, depending only on X, such that for any smooth Y ∈ nH, of deg n ≥ n0, the canonical map : πloc(Y ) → πloc(X) is a surjection. Before we begin the proof, we recall some facts from [MS2]. Let F : X → X be the Frobenius map. For any integer t ≥ 1, denote by Ct the category of all V ∈ V ect(X) such that F t∗(V ) is the trivial vector bundle on X. Let F T (X) denote the union of Ct(X) for all t ≥ 1. Fix a base point x0 ∈ X, Consider the functor S : Ct → V ect(X) given by V → Vx0 It is seen that Ct , with the fibre functor S, is a Tannaka category [MS1,MS2]. denote the 2 corresponding Tannaka group by Gt. It is also easily seen that πloc(X) ≃ lim ←t Gt , where πloc(X) is the Tannaka group associated to the category (V ∈ V ect(X) F t(V ) is trivial for some t. Now let t = 1, and consider G1(Y ) and G1(X). Let V ∈ C1(X). with V stable.V corresponds to a principal H bundle E → X ,where E is reduced in the sense of Nori [N1,page 87,Prop.3], or just N-reduced. To prove that G1(Y ) → G1(X) is a surjection , it is enough to prove that V /Y is stable, or better still that E/Y is N-reduced. Lemma 2.2. If X and E are as above , then there exists an integer n0, depending only on X ,such that for all n ≥ n0, and for all smooth Y ∈ nH, E/Y is N-reduced. Proof. Let f : E → X be given.Then f∗OE belongs to C1. Denote it by W for simplicity. It is easy to see that W ∈ F T (X). Also note that for any V ∈ V ect(X), V ∈ F T (X) if and only if V ∗ ∈ F T (X). Now consider 0 → OX → F∗OX → B1 → 0 (1) , Tensor (1) with W ∗(−n). There is an n0 such that for n ≥ n0, we have Hom(W (n), B1) = 0. In fact, we have that Hom(V (n), B1) = 0 for all V ∈ F T (X), for all n > n0, where n0 is independent of V in F T (X). So the canonical map H 1(W (−n) → H 1(F ∗(W (−n)) is injective. But the last space is just H 1(OX (−np))r, r = rankW . But this is 0 as soon as n ≥ n1, independent of V ∈ C1(X), as dim X ≥ 2. Hence H 0(Y, W/Y ) = 1, which proves that E restricted to Y is also N-reduced. Now assume t ≥ 2. We shall assume in fact that t = 2, because an 2 be identical proof works for t ≥ 3.Consider the map F 2 : X → X. Let B1 the cokernel.We have the exact sequence 0 → B1 → B1 2 → F∗B1 → 0 (2) Let W ∈ C2(X) and tensor (2) with W (−n). One sees immediately that if H 1(W (−n)⊗B1) = 0 for all n ≥ n0, then also H 1((W (−n)⊗B1 2) = 0 for all n ≥ n0, for the same integer n0. So if E is N-reduced on X, then E remains N-reduced on Y , if deg Y ≥ n0 , for the same integer n0, which worked for C1(X). The proof for bigger t goes the same way , by taking the cokernel of F t, where t ≥ 3. This concludes the proof of Theorem 2.1. 3 3 The second theorem Theorem 3.1. Let X be smooth and projective of dim ≥ 3. Then there exists an integer n0, depending only on X ,such that for any smooth Y ∈ nH, n ≥ n0, the canonical map πloc(Y ) → πloc(X) is an isomorphism. Remark 3.2. In the sequel we shall use the phrase " for a uniform n" or" there exists a uniform integer n" to denote a positive integer n ,which may depend on X , but not on V , for V ∈ F T (X). Before we begin the proof we need the following lemma: Lemma 3.3. With X as above, then there exists a uniform integer n0 such that we have H 1(Ω1 X(−n) ⊗ V ) = 0, for all V ∈ F T (X), for all n ≥ n0. Proof. Let TX be the tangent bundle of X. For some s , depending only on X, TX (s) is generated by global sections. So we have Dualizing,we get 0 → S ∗ → ON X → TX (s) → 0. 0 → Ω1(−s) → ON X → S → 0 .Tensor (1) with V (−t),to get 0 → V (−t) ⊗ Ω1(−s) → V (−t) ⊗ ON X → V (−t) ⊗ S → 0 . Applying F r to (2) we get (1) (2) 0 → [V (−t) ⊗ Ω1(−s)]pr → [V (−t) ⊗ ON X ]pr → [V (−t) ⊗ S]pr → 0 (3) . It is clear that H 0(V (−t) ⊗ S) = 0 for t >> 0, independent of V . Now look at H 1(V (−t) ⊗ Ω1(−s)) → H 1(V (−t) ⊗ ON X ) → H 1(V (−t) ⊗ ON X )pr → 0 (4) where the right arrow is the map induced by F r.But one knows that H 1(V (−t)⊗ ON H 1[V (−t) ⊗ ON X )pr vanishes for t >> 0, for a uniform t. And H 1[V (−t)) ⊗ Ω1(−s)] → X ] is injective for t >> 0, for a uniform t . Also H 1[V (−t) ⊗ ON X ]pr is injective for t >> 0, for a uniform t. Hence H 1(V (−t) ⊗ Ω1) vanishes for all t >> 0, for a uniform t. This concludes the proof of Lemma 3.3. X ] → H 1[V (−t) ⊗ ON 4 Remark 3.4. In fact ,the above proves the following : Let W be an arbitrary vector bundle on X. Then there exists a uniform t0, such that for all t ≥ t0, and for all V ∈ F T (X),we have H 1(V ⊗ W (−t)) = 0. This lemma and remark are the key points in the paper, and will be used over and over again, without explicit mention. Lemma 3.5. Any W ∈ Ct(Y ) lifts uniquely to an element V ∈ Ct(X) ,if degreeY = n >> 0, n is uniform. Proof. First ,we give an idea of the proof We assume that dimension X ≥ 3, and we pick an arbitrary smooth Y ∈ nH. First assume t = 1.Take a W ∈ C1(Y ) and we show that it lifts uniquely to V ∈ C1(X) . Such a W is trivialized by the Frobenius,so there exists M, an r × r matrix of 1- forms on Y ,which gives an integrable connection ∇ on Or Y , with p- curvature 0. Consider 0 → and I I 2 → Ω1 X /Y → Ω1 Y → 0 (1) (2) 0 → Ω1 X (−n) → Ω1 X → Ω1 X /Y → 0 X) → H 0(Y, Ω1 Here I is the idealsheaf of Y, I = OX (−n). We get an integer n0, such that for all n ≥ n0, H 0(X, Ω1 Y ) is an isomorphism. So M lifts to M1, a r × r matrix of 1 forms on X.Now ∇ has p-curvature 0 on Y .We now show that ∇1 defined by M1 has p-curvature 0 on X. The curvature of ∇ is an element of H 0(EndOr Y ), Again by E-S-Z-S, the curvature of ∇1 is 0 if degree Y ≥ n1, for some integer n1, depending only on X. The p- curvature of ∇1 is an element of H 0(F ∗Ω1 X ), which again vanishes if deg Y ≥ n2. So any element W ∈ C1(Y ) lifts uniquely to an element V ∈ C1(X). X ⊗ EndOr Y ⊗ Ω2 For t > 1,we use induction on t. Assume that for lower values of t, any element W ∈ Ct(Y ) has been lifted uniquely to an element V ∈ Ct(X) ,where degree Y = n, where n is uniform.We show in fact that there exists a uniform n with the property:if W ∈ Ct+1(Y ), then W lifts to X. This proceeds by showing that F ∗W lifts uniquely to X, and this lifted bundle, say Vt,on X has an integrable p-flat connection. So on X, Vt will descend under F to Vt+1. And Vt+1 restricts to Wt+1 on Y . We begin the proof by establishing a couple of claims: Claim (1) : There exists a uniform n such that for V ∈ F T (X) and for a Y ∈ nH , we have H 0(EndV ⊗ Ω1 Y (−n)) = 0. 5 Proof of Claim 1): Look at 0 → OY (−n) → Ω1 X /Y → Ω1 Y → 0 and 0 → Ω1 X (−n) → Ω1 X → Ω1 X /Y → 0 (1) (2) Tensor (2) with EndV (−n), we get H 0(EndV (−n) ⊗ Ω1 X /Y ) = 0 ,for a uniform n. From (1) after tensoring with EndV , one sees that H 1(EndV ⊗ OY (−2n)) = 0 implies that H 0(EndV (−n) ⊗ Ω1 Y )) = 0. All this is for a uniform n. Now we prove that H 1(EndV ⊗ OY (−2n)) = 0 , for a uniform n, in Claim (2); One has H 1(EndV ⊗ OY (−2n) = 0, for any V ∈ F T (X) and smooth Y ∈ nH, for a uniform n. Proof of Claim (2): Look at , , 0 → OX → F∗OX → B1 → 0 0 → B1 → Z 1 → Ω1 → 0 0 → Z 1 → F∗Ω1 → B2 → 0 (1) (2) (3) This are sequences obtained from the Cartier operator applied to the DeR- ham complex F.(Ω. X ). Tensor all the 3 sequences by EndV (−n), and take H 0. From (2), we get H 1(B1 ⊗ EndV (−n)) injects into H 1(Z 1 ⊗ EndV (−n)) for a uniform n. But by 3, we see that H 1(Z 1 ⊗ EndV (−n)) injects into H 1(F∗Ω1 ⊗ EndV (−n)) for a uniform n. But the last cohomology group vanishes for n >> 0, with n uniform (Lemma 2.2 and [S,Th.3.5]). Therefore ,H 1(B1 ⊗ EndV (−n)) vanishes for n >> 0, with n uniform. Now look at 0 → OX → F∗OX → B1 → 0 (4) Tensor with EndV (−n) and take cohomology.As H 1(B1 ⊗ EndV (−n)) van- ishes for n >> 0, one gets that H 2(EndV (−n)) injects into H 2(EndV (−n))pr for all r >> 0, and for n uniform. But this last group vanishes for n >> 0, and uniform, and for all r(Lemma 2.2 again). (This is where the hypothesis dimX ≥ 3 is used).Now finally look at 0 → OX (−3n) → OX(−2n) → OY (−2n) → 0 (5) 6 and tensor with EndV . We get H 1(OY (−2n) ⊗ EndV )) vanishes for n >> 0, and n uniform. Hence, we get H 0(EndV ⊗ Ω1 Y (−n)) = 0, with n uniform , which completes the proof of Claim 1. Claim (3): There is a uniform n with the property : let Y ∈ nH and assume that W ∈ F T (Y ) has been lifted to V ∈ F T (X). Then if W has a connection on Y ,then V on X has a connection. Proof of Claim (3):Look at 0 → OY (−n) → Ω1 X /Y → Ω1 Y → 0 (1) and 0 → Ω1 X (−n) → Ω1 X → Ω1 Tensor with EndV . From (2) we get H 1(EndV ⊗Ω1 Ω1 injects into H 1(EndV ⊗ Ω1 connection ,so does V . (2) X ) injects into H 1(EndV ⊗ X /Y ) Y ). But V restricted to Y is W . Hence if W has a X /Y ). And from (1) and the proof of claim (2), we get H 1(EndV ⊗ Ω1 X /Y → 0 Claim (4) : If this connection on W is integrable ,then the connection on V is also integrable. Proof of Claim (4) : Look at 0 → Ω1 Y (−n) → Ω2 X /Y → Ω2 Y → 0 and 0 → Ω2 X (−n) → Ω2 X → Ω2 X /Y → 0 (1) (2) Tensor with EndV and take H 0. One knows that H 0(EndV ⊗ Ω1 Y (−n) vanishes for n >> 0 with n uniform ,by the proof of Claim 1. Simil- iarly, H 0(EndV ⊗ Ω2 X(−n) also vanishes for n >> 0, and n uniform.So H 0(EndV ⊗ Ω2 Y ) is injective for n >> 0, n uniform. So if W has an integrable connection, so does V . Finally, the p-curvature of V is an element of H 0(EndV ⊗ F ∗Ω2 Y ). So if the p-curvature on W is 0, so is the p-curvature on V . X) which injects into H 0(EndV ⊗ F ∗Ω2 X ) → H 0(EndV ⊗ Ω2 Claims 1-4 imply that on X,if V restricts to W on Y , with degree Y = n with n uniform and W has a p-flat connection, then V also has a p-flat connection. So if W on Y descends to W1 , then V also descends under F to V1, and that V1 restricts to W1 on Y . This continues to hold for V ∈ Ct(X), restricting to W ∈ Ct(Y ), any t, for Y degree a uniform n, depending only on X. Hence the canonical map of Tannaka Categories : Ct(X) → Ct(Y ) , given by restriction from X to Y , induces an isomorphism: Gt(Y ) → Gt(X), 7 for all t.But πloc(X) ≃ lim ←t Gt(X) and similiarly for Y . Since there are only finitely many choices of n, depend- ing only on X, this completes the proof of Lemma 3.5 and hence of Theorem 3.1. Remark 3.6. It is also interesting to determine if the category of F-trivial vector bundles ,on a smooth X,is m0- regular. This is the case if dim X ≤ 3. Remark 3.7. The discerning reader will notice that the only property of F- trivial bundles which is used is the following : the set of isomorphism classes of stable bundles, which occur in a stable filtration of F n∗(V ), n = 0, 1, .... is only finite in number. If V is only assumed to be essentially finite[N1,p.82],then ∃ a Galois etale covering π : Z → X such that F m∗V is trivial on Z, for some m. Now one sees using [D1, Thm 2.3.2.4],that the set of stable components of F n∗(V ), n = 0, 1... is again finite. So all the proofs and propositions carry over with πlocX replaced by π(X),making use of Lemma 3.3. We leave the details as an exercise. References [BH1] Comparison of fundamental group schemes of a projective variety and an ample hypersurface, arXiv.math /0603299 v1 [math AG],13 March 2006. [BH2] Indranil Biswas, Yogesh Holla Comparison of fundamental group schemes of a projective variety and an ample hypersurface, arXiv.math /0603299vi [math AG], 1st May 2007. [BH3] Indrani Biswas, Yogesh Holla Comparison of fundamental group schemes of a projective variety and an ample hypersurface, Journal of Al- gebraic Geometry, 16 (2007),No.3, 547-597. [D1] Laurent Ducrohet Frobenius et Fibres sur les Courbes, Thesis,University of Paris 6, 2006. [K1] N.Katz Nilpotent Connections and the Monodromy Theorem ...... Publ.Math. I.H.E.S. No.39 (1970), 175-232. 8 [L1] Adrian Langer Semistable Sheaves in Positive Characteristic,Ann. of Math.,256 (2004), 251-276 [L2] Langer, Adrian Semistable principal G-bundles in positive characteris- tic, Duke Math. J. 128 (2005), no. 3, 511 -- 540. [MN] Semistable Sheaves on Homogeneous Spaces and Abelian Varieties, Proc. Indian Acad. Sci.(Math Sci.)Vol. 93 , No. 1, 1-12. [MS1] V. B. Mehta, S.Subramanian On the Fundamental group scheme In- vent. Math. 148,(2002),143-150. [MS2] V.B. Mehta, S.Subramanian Some Remarks on The Local Fun- damental Group Scheme Proc. of the Ind. Acad. of Sciences, (Math. Sci.),Vol.118,No.2, 2008,pp.207-211. [N1] M.V.Nori The Fundamental Group Scheme,Proc. of the Ind. Acad. of Sci.(math Sci.),91,1982. [N2] M.V.Nori Representations of the Fundamental Group,Compositio Math.33, 1976. [S-B] N.I. Shepherd-Barron Semi-Stability and Reduction mod p, Topology, Vol. 37,No.3, pp.659-664,1998. Smith [S] Karen Bounds ometry,Santa Math.,62,Part1,Amer.Math.Soc.,Providence,RI,1997 Vanishing,Singularities Characteristic Local Cruz,1995,289-325,Proc. Via Prime And Algebra,Alg. Sympos. Effective Ge- Pure [Su] S. Subramanian Strongly Semistable Bundles on a Curve over a Finite Field , Arch. Math. 89 (2007), 68-72 9
1312.1584
5
1312
2019-09-05T07:42:33
On the integral cohomology of quotients of manifolds by cyclic groups
[ "math.AG" ]
We propose new tools based on basic lattice theory to calculate the integral cohomology of the quotient of a manifold by an automorphism group of prime order. As examples of applications, we provide the Beauville--Bogomolov forms of some irreducible symplectic orbifolds; we also show a new expression for a basis of the integral cohomology of a Hilbert scheme of two points on a surface.
math.AG
math
On the integral cohomology of quotients of manifolds by cyclic groups Grégoire Menet September 6, 2019 Abstract We propose new tools based on basic lattice theory to calculate the integral cohomology of the quotient of a manifold by an automorphism group of prime order. As examples of applications, we provide the Beauville -- Bogomolov forms of some primitively symplectic orbifolds; we also show a new expression for a basis of the integral cohomology of a Hilbert scheme of two points on a surface. AMS Classification 2010: 55N10, 14F43, 53C261 1 Introduction Consider a compact connected orientable manifold (without boundary) X and a finite automorphism group G, of prime order p, acting on X. In this article, we study the integral cohomology of the quotient X/G. The group G acts naturally on the cohomology of X. In the case of cohomology with rational coefficients, the problem is much simpler, H ∗(X/G, Q) is isomorphic to the invariant space H ∗(X, Q)G. Nevertheless, for integral cohomology this property does not hold anymore. A fundamental tool for studying this question is given by the work of Smith in [27]. Let π : X → X/G be the quotient map. Smith has constructed a push-forward map π∗ : H ∗(X, Z) → H ∗(X/G, Z) such that: π∗ ◦ π∗ =Xg∈G π∗ ◦ π∗ = p idH∗(X/G,Z), g∗. (1) Let T be a Z-module, we denote by Tf := T sequence torsion the torsion-free part of T . Then, (1) provides the exact 0 / π∗(H k(X, Z))f / H k(X/G, Z)f / (Z /p Z)αk(X) / 0, (2) where αk(X) is a non-negative integer which we call the k-th coefficient of surjectivity of X (See Section 3.1). The objective of this paper is to provide tools for calculating αk(X). Using basic lattice theory, we provide an upper bound for the coefficients of surjectivity. For a finer result, one has to study the local action of G around the fixed points. A simple fixed point is a fixed point such that the local action of G around is given by a matrix of the form diag(1, ..., 1, ξd p ), where ξp is a p-th root of unity with d ∈ {1, ..., p − 1}. These fixed points have the interest to provide singularities on X/G that can be solved with only one blow-up. For instance, all fixed points are simple when p = 2; we will see other examples of such fixed points in Section 7. Our main theorem is: p, ..., ξd Theorem 1.1. Let X be a compact Kähler manifold of complex dimension n and G an automorphism group of prime order p ≤ 19. Let c := Codim Fix G. Assume that: (i) H ∗(X, Z) and H ∗(Fix G, Z) are p-torsion-free, (ii) the spectral sequence of equivariant cohomology with coefficients in Fp degenerates at the second page, (iii) all fixed points of G are simple, 1Keywords: Integral cohomology, group actions on lattices, compact orientable manifolds, Beauville -- Bogomolov forms. 1 / / / / (iv) c ≥ n 2 + 1. Then, for all 2n − 2c + 1 ≤ k ≤ 2c − 1, the k-th coefficient of surjectivity αk(X) of X vanishes. Condition (ii) is natural in the sense that there is already a theory to show the degeneration of spectral sequences at the second page (see for instance [8]). Sarti, Boissière and Niper-Wisskirchen, in [5], also developed theorems of degeneration of the equivariant spectral sequence that we will use in our applications. We will also see (Theorem 6.1) that condition (ii) can be replaced by a numerical condition involving invariants related to the action of the group and the cohomology of the fixed locus. This result can be applied to compute the Beauville -- Bogomolov form of some primitively symplectic orbifolds. A orbifold is a compact analytic complex space with at worst finite quotient singularities. A compact Kähler orbifold is called symplectic if its singularities are in codimension 4 and its smooth locus is endowed with an everywhere non-degenerate holomorphic 2-form. In addition, a symplectic orbifold is said primitively symplectic 2 if the holomorphic 2-form is unique up to scaling. Such varieties are good candidates to generalize the short list of known irreducible symplectic manifolds. Indeed, some aspects of the theory were already generalized in [22] and [16], for instance, the Beauville -- Bogomolov form, the local Torelli theorem, and the Fujiki formula. We recall that the Fujiki formula (45) establishes a propor- tionality between the cup-product and the Beauville -- Bogomolov form; the coefficient of proportionality is called the Fujiki constant. In this paper we compute the Beauville -- Bogomolov form of a manifold of K3[2]-type quotiented by symplectic automorphisms of order 3 and 11. There are two different examples of symplectic automor- phisms σ of order 11 on a manifold of K3[2]-type X (Example 4.5.1 and Example 4.5.2 in [19]). As explained in [19, Section 7.4.4], the lattice H 2(X, Z)G will be either  We denote the quotients X/σ, respectively, by M 1 11 and M 2 11. 2 2 8 −3 6 2 2 −3 8  or  2 1 1 6 0 0 0 0 22. Theorem 1.2. Let X be a manifold of K3[2]-type. Let G be a symplectic automorphism group of order 11 acting on X. Then, the Fujiki constant of both M 1 11 is 33 and the Beauville -- Bogomolov lattices are: 11 and M 2 H 2(M 1 11, Z) ≃ 3 8 −1 2 −1 −1 3 −1 6  , H 2(M 2 11, Z) ≃ 2 1 0 1 0 6 0 0 2 . In [19, Theorem 7.2.7], Mongardi distinguishes two kinds of symplectic automorphism of order 3 on a K3[2]-type manifold one with 27 isolated fixed points and another with a fixed abelian surface. Theorem 1.3. Let X be a manifold of K3[2]-type. Let G be a symplectic automorphisms group of order 3 of X with 27 isolated fixed points. We denote M3 = X/G. Then, the Beauville -- Bogomolov lattice H 2(M3, Z) is isomorphic to U (3) ⊕ U 2 ⊕ A2 2 ⊕ (−6), and the Fujiki constant of M3 is 9. 11 and M 2 These examples provide new small-dimensional moduli spaces of singular primitively symplectic va- rieties. In particular, M 1 11 provide two positive definite Beauville -- Bogomolov lattices of rank 3. It follows that the period domains of M i 11, i ∈ {1, 2} are given by the zero of the Beauville -- Bogomolov 11, C)) ≃ P2, hence are conics on P2. Moreover, none of their deformations quadratic forms on P(H 2(M i admit a Lagrangian fibration, indeed the existence of a Lagrangian fibration is in contradiction with having a definite positive Beauville -- Bogomolov form since, by the Fujiki formula, the form vanishes on the pull-back of an ample divisor in the base of the fibration. Hence, it seems that the varieties M 1 11 and M 2 11 and M 2 11 could be interesting examples to study in order to develop the theory of singular primitively symplectic varieties. 11 look very different from the known smooth irreducible symplectic varieties. So, M 1 As it is well known, the Hilbert scheme of two points A[2] on a surface A can be obtained as a quotient of Bl∆(A × A) the blow-up of A × A in the diagonal ∆. Therefore, studying the integral cohomology of A[2] is a possible application of our theory. As another application of our main theorem, we provide a new expression of a basis of the cohomology of A[2], already studied in [25] and [26]. This basis has the advantage to only require that the surface is Kähler. Moreover, the basis is expressed in terms of 2Here, we prefer the terminology primitively symplectic (introduced for the first time by Fujiki in [10]) rather than irreducible symplectic which we reserve for varieties appearing in a Bogomolov type decomposition theorem. 2 pull-backs and push-forwards of classes of the surface, allowing to calculate easily the ring structure of H ∗(A[2], Z) (see Lemma 7.9). This paper generalizes results of the author's Ph.D. thesis [18] (see also [17]), and it was already applied in [13] for finding the Beauville -- Bogomolov form of a partial resolution of the quotient by a symplectic involution of the generalized Kummer of dimension 4 which is the first example of odd Beauville -- Bogomolov form. The paper is structured as follows. Since the integral cohomology has a lattice structure, in Section 2, we study a general lattice T endowed with an isometry group G of prime order. In particular, using the Z[G]-module structure of T , we compute the discriminant of the invariant lattice T G (Proposition 2.15). In Section 3 we apply these results to find the discriminant of π∗(H ∗(X, Z)) (Corollary 3.8), and, using basic lattice theory, we provide an upper bound for the coefficients of surjectivity αk(X) (Proposition 3.10). Thanks to Section 2, we can calculate the cohomology of G in Section 4 (Corollary 4.2); this allows us to use equivariant cohomology to compute the cohomology of non-ramified quotients. In Section 5 we study resolution of singularities, which leads to the proof of the main Theorem 1.1 in Section 6. Section 7 is dedicated to some applications, namely the computation of the Beauville -- Bogomolov forms of certain orbifolds and the description of the integral cohomology of the Hilbert scheme of two points. Acknowledgements. I am very grateful to Samuel Boissière for helpful discussions, to Simon Kapfer for inspiring in the problem of Section 7.3 and to Patrick Popescu-Pampu for very useful advices about blow-ups. I also want to thank Emilio Franco, Dimitri Markushevich and Henrique Sá Earp for many helpful comments on some preliminary versions of this work. This work was supported by Fapesp grant 2014/05733-9. 2 The Z[G]-module structure of a lattice 2.1 Basic properties of lattices We recall the basic facts on lattices which are used in this paper (see for example [9, Chapter 8.2.1]). A lattice T is a free Z-module endowed with a non-degenerate bilinear form. We denote by discr T the discriminant of T , which is the absolute value of the determinant of the bilinear form of T . We say that T is unimodular if its discriminant is 1. A sublattice N of T is said primitive if T /N is torsion-free. If N is a sublattice of a lattice T of the same rank, we have the basic formula: If T is an unimodular lattice and N is a primitive sublattice, then [T : N ]2 = discr N discr T . More precisely, the projection discr N = discr N ⊥. T N ⊕ N ⊥ → AN (3) (4) (5) is an isomorphism, where AN := N ∨/N is the discriminant group. Moreover, this provides an isometry: ν : AN → AN ⊥, (6) between the discriminant groups. Let X be a compact connected orientable manifold of dimension n. We endowed H ∗(X, Z) with the natural bilinear form B defined as follows. Let x ∈ H k(X, Z) and y ∈ H q(X, Z), B(x, y) = 0 if k + q < n, B(x, y) = x · y if k + q = n, where · refers to the cup-product. 3 By Poincaré duality, it turns H ∗(X, Z) into an unimodular lattice. In our case, the lattice T will be endowed with an isometry group G of prime order p. We denote by T G the invariant lattice of T . We also denote by (T G)∨(p) the dual of T G with its bilinear form multiplied by p. Let G be an automorphism group of prime order p on X and T an unimodular sublattice of H ∗(X, Z); in Section 3.2, we will see that: (T G)∨(p) ≃ π∗(T ), where π : X → X/G is the quotient map. For this reason, we call (T G)∨(p) the push-forward lattice of (T, G); a key step for our purpose, which will be accomplished in Section 2.3, is the calculation of its discriminant. To do so, we will use the Z[G]-module structure of T given by g · x = g(x) for all g ∈ G and all x ∈ T . 2.2 The Boissière -- Nieper-Wisskirchen -- Sarti invariants Here we review some important notions introduced in [5]. Let p be a prime number, T a p-torsion-free Z-module of finite rank and G = hgi a group of prime order p acting linearly on T . Then, T ⊗ Fp is a Fp-vector space equipped with a linear action of G. The minimal polynomial of g, as an endomorphism of T ⊗ Fp, divides X p − 1 = (X − 1)p ∈ Fp[X], hence g admits a Jordan normal form over Fp. We can decompose T ⊗ Fp as a direct sum of some Fp[G]-modules Nq of dimension q for 1 ≤ q ≤ p, where g acts on Nq in a suitable basis by a matrix of the following form: 1  1 . . . 0 . . . 0 . . . . . . . . .  1 1 Definition 2.1. We define the integer ℓq(T ) as the number of blocks of size q in the Jordan decomposition of the Fp[G]-module T ⊗ Fp, so that T ⊗ Fp ≃ ⊕p q=1N ⊕ℓq(T ) . q When p ≤ 19, we can define the ℓq(T ) from another point of view. Let ξp be a primitive p-th root of unity, K := Q(ξp) and OK := Z[ξp] the ring of algebraic integers of K. By a classical theorem of Masley-Montgomery [15], OK is a PID if and only if p ≤ 19. The Z[G]-module structure of OK is defined by g ·x = ξpx for x ∈ OK. For any a ∈ OK, we denote by (OK, a) the module OK +Z whose Z[G]-module structure is defined by g · (x, k) = (ξpx + ka, k). In the proof of [5, Proposition 5.1], using [7, Theorem 74.3] of Diederichsen and Reiner, Boissière, Nieper-Wisskirchen and Sarti show the following. Proposition 2.2. Let T be a free Z-module of finite rank and let G = hgi be a group of prime order p ≤ 19 acting on T . Then, we have an isomorphism of Z[G]-module: (i) T ≃ ⊕r i=1(OK, ai) ⊕ O⊕s K ⊕ Z⊕t, where r, s, t are integers and ai /∈ (ξp − 1)OK. Moreover, the integers r, s and t are uniquely determined by the Z[G]-module structure of T . (ii) When p ≥ 3, (OK , ai) ⊗ Fp = Np, OK ⊗ Fp = Np−1 and Z ⊗ Fp = N1, so r = ℓp(T ), s = ℓp−1(T ) and t = ℓ1(T ). (iii) We have: with rk(OK , ai)G = 1. T G ≃ ⊕r i=1(OK, ai)G ⊕ Z⊕t, Remark 2.3. In the particular case p = 2, we have ℓ1(T ) = t + s and ℓ2(T ) = r. Hence, to distinguish the integers t and s in the case p = 2, we choose the following notation. 4 Definition 2.4. When p = 2, we define ℓ+(T ) := t, ℓ−(T ) := s. Remark 2.5. When p = 2, it follows: ℓ1(T ) = ℓ+(T ) + ℓ−(T ). Moreover, to avoid distinguishing the cases p = 2 and p ≥ 3 all over, we adopt the following notation. Notation 2.6. When p ≥ 3, we also denote ℓ+(T ) := ℓ1(T ) = t and ℓ−(T ) := ℓp−1(T ) = s. We can reformulate Proposition 2.2 in those terms: Corollary 2.7. Let T be a free Z-module and let G = hgi be a group of prime order p ≤ 19 acting on T . We have: (i) rk T = pℓp(T ) + (p − 1)ℓ−(T ) + ℓ+(T ), (ii) rk T G = ℓp(T ) + ℓ+(T ). As in [5], in the case of cohomology, we choose the following notation for our invariants: Notation 2.8. For all 0 ≤ k ≤ dim X, we denote: ℓk +(X) = ℓ+(H k f (X, Z)), ℓk −(X) = ℓ−(H k f (X, Z)) and for all q ∈ {1, ..., p} , ℓk q (X) = ℓq(H k f (X, Z)). 2.3 Discriminant of the push-forward lattice Proposition 2.9. Let T be an unimodular lattice and let G = hgi be a group of prime order p acting on T . Then, If moreover p ≤ 19, we have: discr(T G)∨(p) = prk T G−ℓp(T ). discr(T G)∨(p) = pℓ+(T ). This section is dedicated to proving this proposition. We need several lemmas. Lemma 2.10. Let T be a lattice and G = hgi be a group of prime order p acting on T . Let σ = id +g + ... + gp−1 then: ker σ = (T G)⊥. Proof. We have ker σ ⊂ (T G)⊥. Indeed, let x ∈ ker σ and y ∈ T G. For all 0 ≤ k ≤ p − 1, we have: gk(x) · y = x · y. Hence, 0 = σ(x) · y = px · y. So, x · y = 0. Conversely, let x ∈ (T G)⊥, let y = x + g(x) + ... + gp−1(x). Then, we have y ∈ (T G)⊥ ∩ T G. Since the bilinear form on T is non-degenerate, we have y = 0. Lemma 2.11. Let T be an unimodular lattice and G = hgi be a group of prime order p acting on T . (T G)∨ =(cid:26) y + g(y) + ... + gp−1(y) p Proof. For all x ∈ T G and all y ∈ T , (cid:12)(cid:12)(cid:12)(cid:12) y ∈ T(cid:27) . y + g(y) + ... + gp−1(y) p · x = y · x. Hence: (T G)∨ ⊃(cid:26) y + g(y) + ... + gp−1(y) p (cid:12)(cid:12)(cid:12)(cid:12) y ∈ T(cid:27) . Now, we prove the converse inclusion. Let x be a primitive element of T G and q ∈ N∗ such that q ∈ AT G . By (5), there is z ∈ T and y ∈ (T G)⊥ such that z = x+y q ∈ (T G)∨. Then, it provides x x q . q x + y+g(y)+...+gp−1(y) Then, z + g(z) + ... + gp−1(z) = p . But by Lemma 2.10, y + g(y) + ... + gp−1(y) = 0. Hence, z + g(z) + ... + gp−1(z) = p q x. Since x is primitive in T G, q divides p. Hence, q = 1 or q = p. If q = p, we get z + g(z) + ... + gp−1(z) = x so that x . If q = 1, we can write x = x+...+x q = z+g(z)+...+gp−1(z) = x+g(x)+...+gp−1(x) since x ∈ T G. p q p p 5 From a basic calculation (see proof of [5, Lemma 3.1]), we obtain: Lemma 2.12. Let T be a free Z-module and G = hgi be a group of prime order p acting on T . We denote τ = g − id and σ = id +g + g2 + ... + gp−1. Moreover, we denote by τ , σ ∈ Fp[G] respectively the reduction of τ and σ modulo p. Let (v1, ..., vq) be the canonical basis of the Fp[G]-module Nq defined in Section 2.2; that is g(v1) = v1 and g(vi) = vi + vi−1 for all i ≥ 2. Then the following hold: (i) ker(τ ) = hv1i and Im(τ ) = hv1, ..., vq−1i, for all q ≤ p. (ii) ker(σ) = Nq and Im(σ) = 0, for all q < p. Moreover, ker(σ) = hv1, ..., vq−1i and Im(σ) = hv1i, if q = p. Lemma 2.13. Let T be a free Z-module and G = hgi be a group of prime order p acting on T . Following notation of Section 2.2, we have: T ⊗ Fp ≃ ⊕p q=1N ⊕ℓq(T ) q . For all z ∈ T , we denote now by z ∈ T ⊗ Fp its reduction modulo p. Let x ∈ T G, then there exists y ∈ T such that x = y + g(y) + ... + gp−1(y) if and only if x ∈ N ⊕ℓp(T ) p . Proof. Assume that x = y + g(y) + ... + gp−1(y), with y ∈ T . If x = 0, then x ∈ N ⊕ℓp(T ) . Hence, x ∈ N ⊕ℓp(T ) assume that x 6= 0. Then, y /∈ ker σ, so by Lemma 2.12, (ii), y ∈ N ⊕ℓp(T ) p . Now we . p p Conversely, assume that x ∈ N ⊕ℓp(T ) the direct summands of N ⊕ℓp(T ) by Lemma 2.12, (ii). The result follows. p p (see Lemma 2.12, (i)). But, we have v1,i = vp,i + g(vp,i) + ... + gp−1(vp,i) , we can write x =Pi aiv1,i, where v1,i are invariant elements of Proposition 2.14. Let T be an unimodular lattice and let G = hgi be a group of prime order p acting on T . Then: Proof. Let T ⊗ Fp ≃ ⊕p Lemma 2.11 and 2.13, we can define a surjective morphism: q=1N ⊕ℓq(T ) q T T G ⊕ (T G)⊥ = (Z /p Z)ℓp(T ). be the isomorphism of Fp[G]-module provided in Section 2.2. By (T G)∨ x p / (N G p )⊕ℓp(T ) / x, where x denotes the reduction modulo p of x. Moreover, x morphism provides an isomorphism p ∈ T G ⇔ px ⇔ x = 0. Hence, the previous AT G ≃ (N G p )⊕ℓp(T ). Since N G p ≃ (Z /p Z), the result follows by (5). Now, by Proposition 2.14 and (3), we have So, from (4) and (7), we obtain the discriminant of T G: discr(cid:0)T G ⊕ (T G)⊥(cid:1) = p2ℓp(T ). (7) Proposition 2.15. Let T be an unimodular lattice and let G = hgi be a group of prime order p acting on T . Then: discr T G = pℓp(T ). Now, we conclude the proof of Proposition 2.9. Since T is unimodular, we know by Proposition 2.14 T G⊕(T G)⊥ = (Z /p Z)ℓp(T ). Hence, by of T G is isomorphic to T that the discriminant group AT G := (T G)∨ Proposition 2.15 and (3): T G It follows: discr(T G)∨ = p−ℓp(T ). discr(T G)∨(p) = prk T G−ℓp(T ). So, by Corollary 2.7, in the case p ≤ 19, we obtain: discr(T G)∨(p) = pℓ+(T ). 6 / / 3 The Z[G]-module structure of the cohomology lattice In all Section 3, X be a compact connected orientable manifold of dimension n and G = hgi be an automorphism group of prime order p. We denote by π : X → X/G the quotient map. 3.1 Coefficients of surjectivity The following proposition is straight-forward from [1, Theorem 5.4 and Corollary 5.8] (it is a generaliza- tion from homology to cohomology of original results of Smith [27]) Proposition 3.1. We can define a push-forward map π∗ : H ∗(X, Z) → H ∗(X/G, Z) such that: π∗ ◦ π∗ = p idH∗(X/G,Z), π∗ ◦ π∗ =Xg∈G g∗. From this proposition, let us examine the torsion of H ∗(X/G, Z). Notice that H ∗(X, Z) can be decomposed into three parts: the torsion-free part, the p-torsion part and the other torsion part. Remark 3.2. Denote by H ∗ Proposition 3.1 o−t(X, Z) the torsion part of H ∗(X, Z) without the p-torsion part. By using H ∗ o−t(X, Z)G ≃ H ∗ o−t(X/G, Z). Hence, the torsion which is not p-torsion does not provide any difficulties. On the other hand, the technique we propose does not say anything about the p-torsion part of H ∗(X/G, Z). So, in this paper, we will study only the torsion-free part of H ∗(X/G, Z). Notation 3.3. Let T be a Z-module. We denote: • Tf := T / torsion • Tp−f := T /(p − torsion) Let Y be a topological space and p a prime number. Set: • H ∗ • H ∗ f (Y, Z) := H ∗(Y, Z)/ torsion, p−f (Y, Z) := H ∗(Y, Z)/(p − torsion). We want to calculate the integer αk(X) provided by the exact sequence (2): 0 / π∗(H k(X, Z))f / H k f (X/G, Z) / (Z /p Z)αk(X) / 0. Definition 3.4. The integer αk(X) is called the k-th coefficient of surjectivity of (X, G). Hence, with our notation, αk(X) is expressed as follows: p−f (X/G, Z) H k π∗(H k(X, Z))p−f = f (X/G, Z) H k π∗(H k(X, Z))f = (Z /p Z)αk(X). Then, from (3), we obtain the following formula: αk(X) = nXk=0 logp discr π∗(H ∗(X, Z))f − logp discr H ∗ f (X/G, Z) 2 . (8) This formula is available under many different versions considering natural sublattices of H ∗(X, Z) as H k(X, Z) ⊕ H n−k(X, Z): αk(X) + αn−k(X) = logp discr(cid:16)π∗ (cid:0)H k(X, Z) ⊕ H n−k(X, Z)(cid:1)f(cid:17) − logp discr (cid:16)H k f (X/G, Z) ⊕ H n−k f (X/G, Z)(cid:17) . (9) In the rest of this section, we will see how this formula leads to an upper bound on the coefficients of surjectivity. 2 7 / / / / 3.2 The push-forward lattice Now in order to apply Proposition 2.9, in this section we prove the following proposition. Proposition 3.5. Let T be a unimodular sublattice of H ∗(X, Z) stable under the action of G. Then, we have an isometry: π∗(T )f ≃ (T G)∨(p). Before proving the proposition, the following lemma allows us to understand the cup-product form of the lattice π∗(H ∗(X, Z)G). Lemma 3.6. (i) Let 0 ≤ k ≤ n, q an integer such that kq ≤ n and x ∈ H k(X, Z)G. Then: π∗(x)q = pq−1π∗(xq) + (p − torsion). If moreover H kq(X, Z) is p-torsion-free, then the property that π∗(x) is divisible by p implies that π∗(xq) is divisible by p. (ii) Let (xi)1≤i≤q be elements of H ki (X, Z)G with (ki)1≤i≤q integers such that k1 + ... + kq ≤ n, then: π∗(x1) · ... · π∗(xq) = pq−1π∗(x1 · ... · xq) + (p − torsion). Proof. (i) By Proposition 3.1: π∗(π∗(x)q) = pqxq = π∗(pq−1π∗(xq)). The map π∗ is injective on the p-torsion-free part, which implies the claimed equality. If moreover π∗(x) is divisible by p, we can write π∗(x) = py with y ∈ H kq(X/G, Z). This gives: pqyq = pq−1π∗(xq) + (p − torsion). (10) We cannot divide by pq−1 because of the possible torsion of H kq(X/G, Z). We will use the fact that H kq(X, Z) is p-torsion-free to get around the problem. Applying π∗ to this equality, we obtain: Since H kq(X, Z) is p-torsion-free, π∗(p − torsion) = 0, and we have pqπ∗(yq) = pqxq + π∗(p − torsion). Pushing down by π∗, we obtain: (ii) Same proof as (10). π∗(yq) = xq. pyq = π∗(xq). f (X/G, Z), in this case, the torsion part Remark 3.7. Very often in this paper, we will be working in H ∗ in the previous equations can be omitted. In particular, when k1 + ... + kq = n, identifying H n(X, Z) and H n f (X/G, Z) with Z, we get: π∗(x1) · ... · π∗(xq) = pq−1π∗(x1 · ... · xq) = pq−1x1 · ... · xq. Now, we prove Proposition 3.5. By Lemma 2.11, we have: π∗(T )f ⊃ π∗((T G)∨). Indeed, let z ∈ (T G)∨, by Lemma 2.11 there exist y ∈ T such that z = y+g(y)+...+gp−1(y) π∗(z). Now let us prove the other inclusion: p . So, π∗(y) = π∗(T )f ⊂ π∗((T G)∨). 8 Let z ∈ T , by Proposition 2.14, we can write z = x+y 2.10, we obtain x = z + g(z) + ... + gp−1(z). Hence, π∗(z) = π∗( x p with x ∈ T G and y ∈ (T G)⊥. Hence, by Lemma p ) ∈ π∗((T G)∨). It follows Then, by Lemma 3.6 (ii), we have an isometry: π∗(T )f ≃ (T G)∨(p). π∗(T )f = π∗((T G)∨). That proves Proposition 3.5. Moreover, combining Proposition 2.9 and 3.5, we obtain the following corollary. Corollary 3.8. Let T be a unimodular sublattice of H ∗(X, Z) stable under the action of G; then If moreover p ≤ 19, we have: discr π∗(T )f = prk T G−ℓp(T ). discr π∗(T )f = pℓ+(T ). 3.3 Upper bound for the coefficients of surjectivity If we apply Corollary 3.8, to the lattice H k f (X, Z) ⊕ H n−k f (X, Z), we obtain: discr π∗(H k(X, Z) ⊕ H n−k(X, Z))f = prk(Hk f (X,Z)G⊕Hn−k f (X,Z)G)−ℓk p(X)−ℓn−k p (X). (11) Using (9), it follows that αk(X) + αn−k(X) ≤ rk(cid:16)H k f (X, Z)G ⊕ H n−k f (X, Z)G(cid:17) − ℓk 2 p(X) − ℓn−k p (X) . (12) Proposition 3.9. For all 0 ≤ k ≤ n: (i) rk H k f (X, Z)G = rk H n−k f (X, Z)G, (ii) ℓk ∗(X) = ℓn−k ∗ (X), where ∗ means +, − or q ∈ {1, ..., p}. Proof. By Poincaré duality, we have: H k f (X, Z) / H n−k f (X, Z)∨ x / fx := (y 7→ x · y), (13) which is an isomorphism. Moreover, we can endow H n−k setting g·f = f (g−1· ) for all f ∈ H n−k becomes an isomorphism of Z[G]-modules. So, it shows that: (X, Z)∨ with a structure of Z[G]-module by (X, Z)∨ and g ∈ G. With this structure, the Poincaré isomorphism f f ℓk ∗(X) = ℓ∗(H k f (X, Z)) = ℓ∗(H n−k f (X, Z)∨). Now, it remains to show that ℓ∗(H n−k f (X, Z)∨) = ℓ∗(H n−k f (X, Z)) = ℓn−k ∗ (X). At this stage of the proof, the case p = 2 differs from the case p ≥ 3. Let us first assume that p ≥ 3. We have to prove that, for 1 ≤ q ≤ p, the Fp[G]-modules Nq and N ∨ q q is given by are isomorphic (where Nq was defined in Section 2.2). The matrix of the action of g on N ∨ the inverse of the transpose matrix of the Jordan block of size q: 1   1 . . . 0 . . . 0 . . . . . . . . . 9 t −1  1 1 . / / q q )t)q = ((J q Denote by Jq the matrix of the Jordan block of size q. The minimal polynomial of Jq is (X − 1)q. q )−1)t = Iq, with Iq the identity matrix of size q. So, the minimal polynomial However, ((J −1 of (J −1 )t)d = Iq. However, this implies J d )t is given by Jq. Hence, Nq ≃ N ∨ q as Fp[G]-module. This proves the result for p ≥ 3. )t divides (X − 1)q. Let 0 ≤ d ≤ q minimal integer such that ((J −1 q = Iq. So, d = q. It follows that the Jordan form of (J −1 Now assume that p = 2. It is clear that the dual of Z endowed with the trivial Z[G]-module structure is again Z endowed with the trivial Z[G]-module structure. The Z[G]-module OK defined in Section 2.2, is only Z endowed with the Z[G]-module structure given by g · t = −t for all t ∈ Z. Hence, we also K endowed with the Z[G]-module structure have an isomorphism of Z[G]-module between OK and O∨ K. Let us consider now (OK , a)∨ endowed with the Z[G]-module given by g · f = f (g−1· ) for all f ∈ O∨ structure given by g · f = f (g−1· ) for all f ∈ (OK, a)∨. We can apply Proposition 2.2, (i) to (OK , a)∨. If ℓ−((OK, a)∨) and ℓ+((OK , a)∨) are not both zero, we obtain a contradiction by considering the double dual which is the dual of (OK , a)∨ and is canonically isomorphic to (OK, a). Hence, there exists b ∈ OK such that (OK, a)∨ ≃ (OK, b) as Z[G]-module. This concludes the proof of (ii). q q It remains to prove (i). By Poincaré duality isomorphism (13), we only have to show that rk H n−k (X, Z)G = f (X, Z)∨)G, where the action of G on H n−k rk(H n−k (X, Z)∨ is given by g · f = f (g−1· ) for all f ∈ f H n−k (X, Z)∨. This is equivalent to show that dim H n−k(X, Q)G = dim(H n−k(X, Q)∨)G, which is read- ily verified considering a basis B = {v1, ..., vd, vd+1, ..., vm} of H n−k(X, Q) where {v1, ..., vd} is a basis of H n−k(X, Q)G and taking the dual basis of B. f f Then, it follows from (12) (and from Corollary 2.7 in the case p ≤ 19): Proposition 3.10. For all 0 ≤ k ≤ n: αk(X) + αn−k(X) ≤ rk H k f (X, Z)G − ℓk p(X); if moreover p ≤ 19, then αk(X) + αn−k(X) ≤ ℓk +(X). Furthermore, these inequalities become equalities when X/G is smooth. Proof. The first inequality follows from (12) and Proposition 3.9. Then, we deduce the second one from (X/G, Z) is unimodular. Therefore, Corollary 2.7. When X/G is smooth, the lattice H k (9), (11) and Proposition 3.9 provide the equality. f (X/G, Z) ⊕ H n−k f Remark 3.11. When n = 2m is even and k = m, we calculated the discriminant of two copies of the lattice π∗(H m f (X, Z)) and the inequalities of Proposition 3.10 are: αm(X) ≤ rk H m f (X, Z)G − ℓm p (X) 2 ; and if p ≤ 19: 3.4 The H k-normality αm(X) ≤ ℓm + (X) 2 . A particular case of interest is when a coefficient of surjectivity αk(X) vanishes, for some k. In such a case, the push-forward map π∗ : H k p−f (X/G, Z) is surjective. p−f (X, Z) → H k Definition 3.12. Let 0 ≤ k ≤ n. When αk(X) = 0, we say that (X, G) is H k-normal. Proposition 3.13. Let 0 ≤ k ≤ n. The following statements are equivalent: (i) (X, G) is H k-normal. (ii) For x ∈ H k f (X, Z)G, π∗(x) is divisible by p if and only if there exists y ∈ H k f (X, Z) such that x = y + g(y) + ... + gp−1(y). 10 Proof. We assume (ii). Let z ∈ H k by (ii), there exists y ∈ H k f (X, Z) such that z = π∗(x) p . So f (X/G, Z), by (2), we can find x ∈ H k f (X, Z) such that x = y + g(y) + ... + (g)p−1(y). Hence z = π∗(y). Now, assume (i). Let x ∈ H k is surjective, there is y ∈ H k p(y + g(y) + ... + gp−1(y)) = px. f (X, Z)G such that p divides π∗(x). Since π∗ : H k f (X/G, Z) f (X, Z) such that pπ∗(y) = π∗(x). We apply π∗ and Proposition 3.1 to get f (X, Z) → H k We provide two important properties of H k-normality. First, in certain circumstances, it is possible to deduce the H k-normality from H kt-normality. Proposition 3.14. We assume p ≤ 19 and H ∗(X, Z) p-torsion-free (so H ∗(X, Z) ⊗ Fp = H ∗(X, Fp)). Let 0 ≤ k ≤ n, t an integer such that kt ≤ n. Assume that (X, G) is H kt-normal. For all x ∈ H ∗(X, Z), we denote by x ∈ H ∗(X, Fp) its reduction modulo p. If S : Symt H k(X, Fp) → H kt(X, Fp) x1 ⊗ ... ⊗ xt 7→ x1 · ... · xt is injective and S (Symt H k(X, Fp)) admits a complementary vector space, stable by the action of G, then (X, G) is H k-normal. Proof. We use the following notation for the Jordan decompositions of H k(X, Fp) and of H kt(X, Fp): H k(X, Fp) = q (X) ⊕lk q N := pXq=1 pXq=1 N k q and H kt(X, Fp) = q (X) ⊕lkt q N := pXq=1 pXq=1 N kt q . (14) Moreover, since p ≤ 19: H k(X, Fp) = N k 1 ⊕ N k p−1 ⊕ N k p and H kt(X, Fp) = N kt 1 ⊕ N kt p−1 ⊕ N kt p . Hence, by Proposition 2.2: H k(X, Fp)G = N k 1 ⊕ (N k p )G. (15) Let x ∈ H k(X, Z)G, we are going to use the characterization of Proposition 3.13. We assume that there is no y ∈ H k(X, Z) such that x = y + g(y) + ... + gp−1(y) and we show that π∗(x) is not divisible by p. By Lemma 2.13, x /∈ N k p . Again by Lemma 2.13, there is no z ∈ H kt(X, Z) such that xt = z + g(z) + ... + gp−1(z). Since (X, G) is H kt-normal, π∗(xt) is not divisible by p. Now, since H kt(X, Z) is p-torsion-free, by Lemma 3.6 (i), π∗(x) is not divisible by p. 1 = N1, by (15), we have xt /∈ N kt p . Since S is injective and N ⊗t We will apply this proposition to an example in Section 7.2. Now, we introduce the notion of the pullback of a pair (X, G). Definition 3.15. Let s : eX → X be a surjective morphism such that eX is a compact connected orientable manifold of dimension n and suppose that there is an automorphism eg of order p of eX which verifies s ◦eg = g ◦ s. Moreover, we assume that the degree of s is not divisible by p. We denote hegi by eG. The triple (eX, eG, s) will be called a pullback of (X, G). Lemma 3.16. Let (eX,eG, s) be a pullback of (X, G). We denote byeπ : eX → eX/eG the quotient map and by r : eX/eG → X/G the induced map. Let 0 ≤ k ≤ n and x ∈ H k(X, Z)G. Then: If moreover H k(eX, Z) is p-torsion-free, then the property that r∗(π∗(x)) is divisible by p implies that eπ∗(s∗(x)) is divisible by p. eπ∗(s∗(x)) = r∗(π∗(x)) + (p − torsion). Proof. The proof follows the same idea as the proof of Lemma 3.6. We have a Cartesian diagram 11 It induces a commutative diagram in cohomology: s eπ eX eX/eG r / X π / X/G. H k(X/G, Z) r∗ H k(eX/eG, Z) π∗ π∗ eπ∗ eπ∗ H k(X, Z) s∗ H k(eX, Z). It follows from Proposition 3.1 that eπ∗(r∗(π∗(x))) = s∗(π∗(π∗(x))) = p · s∗(x) =eπ∗(eπ∗(s∗(x))). The mapeπ∗ is injective on the torsion-free part, so we get the equality. If moreover r∗(π∗(x)) is divisible by p, we can write r∗(π∗(x)) = py with y ∈ H k(fM , Z). Thus gives: eπ∗(s∗(x)) + (p − torsion) = py. Applyingeπ∗, we get: Since H k(eX, Z) is p-torsion-free, this is also the case for the group s∗(H k(X, Z)), hence: Hence, by applyingeπ∗, we geteπ∗(s∗(x)) = py. Proposition 3.17. Let (eX, eG, s) be a pullback of (X, G). Assume that H ∗(X, Z) and H ∗(eX, Z) are p- of eG. Then, if (eX, eG) is H k-normal then (X, G) is H k-normal. s∗ : H k(X, Z) → H k(eX, Z) is injective too. Then, we can prove that s∗ : H k(X, Fp) → H k(eX, Fp) is injective. Since H ∗(X, Z) and H ∗(eX, Z) are p-torsion-free, we only have to verify that ps∗(x) implies f (X, Z). However, if s∗(x) is divisible by p then s∗(x) · s∗(y) = s∗(x · y) if divisible by p (X, Z). Since the degree of s is not divisible by p, it follows that p divides x · y for all ps∗(x) = peπ∗(y). eπ∗(y) = s∗(x). torsion-free. Assume that s∗(H k(X, Fp)) admits a complementary vector space, stable under the action Proof. We use the notation for the Jordan decomposition as in (14). First note that since s is surjective, px for all x ∈ H k for all y ∈ H n−k y ∈ H n−k f f (X, Z). So, by Poincaré duality, it means that x is divisible by p. Now, let x ∈ H k(X, Z)G, we use the characterization of Proposition 3.13. We assume that there is no y ∈ H k(X, Z) such that x = y + g(y) + ... + gp−1(y) and we show that π∗(x) is not divisible by p. By Lemma 2.13 x /∈ Np. Since s∗ : H k(X, Fp) → H k(eX, Fp) is injective and s∗(H k(X, Fp)) admits a complementary vector space, stable under the action of eG, s∗(x) /∈ Np. Hence, by Lemma 2.13, there is no z ∈ H k(eX, Z) such that s∗(x) = z + g(z) + ... + gp−1(z). Since (eX, eG) is H k-normal,eπ∗(s∗(x)) is not divisible by p. Hence, by Lemma 3.16, r∗(π∗(x)) is not divisible by p. It follows that π∗(x) is not divisible by p. We will see an example of the use of this proposition in Section 7.6. 12   /   / 1 1   p p   - - n n 4 Application to equivariant cohomology Let X be a CW-complex and G a group acting on X permuting the cells. Let EG → BG be a universal G-bundle in the category of CW-complexes. Denote by XG = EG ×G X the orbit space for the diagonal action of G on the product EG × X and f : XG → BG the map induced by the projection onto the first factor. The map f is a locally trivial fiber bundle with typical fiber X and structure group G. We define G(X, Z) := H ∗(EG ×G X, Z). Then, the Leray -- Serre spectral the G-equivariant cohomology of X by H ∗ sequence associated to the map f gives a spectral sequence converging to the equivariant cohomology: Ep,q 2 := H p(G; H q(X, Z)) ⇒ H p+q G (X, Z). Hence, we are interested in calculating the cohomology of our prime order group G = hgi, with coefficient in a Z-module. We have the following projective resolution of Z considered as a G-module: ... τ/ / Z[G] σ / / Z[G] τ / / Z[G] ǫ / Z , where τ = g − id, σ = id +g + ... + gp−1 and ǫ is the summation map: ǫ(Pp−1 j=0 αj. If H is a Z[G]-module of finite rank over Z, the cohomology of G with coefficients in H is computed as the cohomology of the complex: j=0 αjgj) =Pp−1 Proposition 4.1. Assume that H is a p-torsion-free finitely generated Z-module and a Z[G]-module with G = hgi group of prime order p ≤ 19. Then for all i ∈ N∗: 0 / H τ / / H σ / / H τ / / ... . (i) H 0(G, H) = H G, (ii) H 2i−1(G, H) = (Z /p Z)ℓ−(H), (iii) H 2i(G, H) = (Z /p Z)ℓ+(H). Proof. (i) By definition, H 0(G, H) = Ker τ . (ii) First, we remark that the cohomology of the torsion part is trivial. Since the torsion part Htorsion is p-torsion-free, p is always invertible in Htorsion. Hence, all x ∈ Htorsion ∩ Ker τ can be written x = 1 ) ∈ Im σ; and for x ∈ Htorsion such that x + g(x) + ... + gp−1(x) = 0, we can write p (x + ... + x So, henceforth, we assume that H is torsion-free. Using the notation of Section 2.2, in the proof of [7, Theorem 74.3], it is shown that: In odd degrees, H 2i−1(G, H) = Ker σ/ Im τ . p(cid:2)(p − 1)x + (p − 2)g(x) + ... + 2gp−3(x) + gp−2(x)(cid:3). p times {z x = g(y) − y with y = − 1 } Ker σ = OKb1 ⊕ ... ⊕ OKbr+s−1 ⊕ Abr+s, Im τ = E1b1 ⊕ ... ⊕ Er+s−1br+s−1 ⊕ Er+sAbr+s, with b1, ..., bn, OK-free elements in Ker σ, A an OK-ideal of K and E1 = ... = Er = OK, Er+1 = ... = Er+s = (ξp − 1)OK. The numbers r and s in the last equalities are the same as in Proposition 2.2. Moreover, we find in the proof of [7, Theorem 74.3] that OK/(ξp − 1)OK = A/(ξp − 1)A = Z /p Z. Hence, we get Ker σ/ Im τ = (Z /p Z)ℓk −(X). (iii) For i ≥ 1, H 2i(G, H) = Ker τ / Im σ. By (iii) of Proposition 2.2: H G ≃ ⊕r i=1(OK , ai)G ⊕ Z⊕t . By Lemma 2.13 and Proposition 2.2 (ii), all the elements in ⊕r y + g(y) + ... + gp−1(y) with y ∈ H. Then, the result follows. i=1(OK, ai)G can be written as 13 / / If we apply Proposition 4.1 to cohomology, we obtain: Corollary 4.2. Let X be a topological space and G be an automorphism group of prime order p. Assume that H ∗(X, Z) is p-torsion-free and finitely generated with p ≤ 19. Then, for 0 ≤ k ≤ 2 dim X we have: • H 0(G, H k(X, Z)) = H k(X, Z)G, • H 2i−1(G, H k(X, Z)) = (Z /p Z)ℓk −(X), • H 2i(G, H k(X, Z)) = (Z /p Z)ℓk +(X), for all i ∈ N∗. Remark 4.3. In [5, Section 3], a similar result is shown for cohomology with coefficients in Fp. When G acts freely, H ∗(X/G, Z) ≃ H ∗ G(X, Z), where the isomorphism is induced by the natural map f : EG ×G X → X/G, (see for instance [2]). Hence, when the action of G is free and the spectral sequence of equivariant cohomology is degenerate at the second page, Corollary 4.2 will provide the cohomology of X/G. 5 Resolution of singularities Let X be a compact connected orientable manifold of dimension n and G = hgi be an automorphism f (X/G, Z) is a lattice, we have only provided an group of prime order p. With the information that H ∗ upper bound for the coefficients of surjectivity (Proposition 3.9). To obtain more information, we need to consider also the geometry of X/G. One way to do so is to consider a resolution of singularities. 5.1 Exceptional lattice of a resolution Let r : ]X/G → X/G be a resolution of singularities and consider the following diagram: H ∗(X, Z) π∗ H ∗(X/G, Z) r∗ / H ∗(]X/G, Z). Remark first that since r is surjective r∗ is an injective map. Definition 5.1. The sublattice of H k f (]X/G, Z) ⊕ H n−k f Nk,r := r∗(cid:2)π∗(cid:0)H k(X, Z) ⊕ H n−k(X, Z)(cid:1)(cid:3)⊥ f (]X/G, Z) will be called the k-th exceptional lattice of r. It follows that N ⊥ following inclusion: k,r is a primitive over-lattice of r∗hπ∗(cid:0)H k(X, Z) ⊕ H n−k(X, Z)(cid:1)fi. We have the r∗hπ∗(cid:0)H k(X, Z) ⊕ H n−k(X, Z)(cid:1)fi ⊂ r∗hH k f (X/G, Z) ⊕ H n−k f (X/G, Z)i ⊂ N ⊥ k . Now, since r∗ is injective, it induces the following injective map of Z /p Z-vector space: f (X/G, Z) ⊕ H n−k (X/G, Z) H k π∗ (H k(X, Z) ⊕ H n−k(X, Z))f f ֒→ N ⊥ k r∗hπ∗ (H k(X, Z) ⊕ H n−k(X, Z))fi . 14   / Then, using (3) and taking the logarithm in base p, it follows that: logp discr π∗(cid:0)H k(X, Z) ⊕ H n−k(X, Z)(cid:1)f − logp discr H k ≤ logp discr r∗hπ∗(cid:0)H k(X, Z) ⊕ H n−k(X, Z)(cid:1)fi − logp discr N ⊥ k . f (X/G, Z) ⊕ H n−k f (X/G, Z) By (9): Moreover, since r∗ is injective: αk(X) + αn−k(X) ≤ logp discr r∗(cid:2)π∗(H k(X, Z) ⊕ H n−k(X, Z))f(cid:3) − logp discr N ⊥ discr r∗hπ∗(cid:0)H k(X, Z) ⊕ H n−k(X, Z)(cid:1)fi = discr π∗(cid:0)H k(X, Z) ⊕ H n−k(X, Z)(cid:1)f . 2 k . (16) f (]X/G, Z) ⊕ H n−k f (]X/G, Z) is unimodular. Hence, by (4), discr Nk = Furthermore, ]X/G is smooth, so H k discr N ⊥ k . It follows that αk + αn−k ≤ logp discr π∗(cid:0)H k(X, Z) ⊕ H n−k(X, Z)(cid:1)f − logp discr Nk 2 . Finally, by (11) and Proposition 3.9, we obtain: Proposition 5.2. Let X be a compact connected orientable manifold of dimension n and G an auto- morphism group of prime order p. Let r : ]X/G → X/G be a resolution of singularities. Let Nk,r be the k-th exceptional lattice of r. Then: αk(X) + αn−k(X) ≤ rk H k f (X, Z)G − ℓk p(X) − 1 2 logp discr Nk,r. If moreover, p ≤ 19: αk(X) + αn−k(X) ≤ ℓk +(X) − 1 2 logp discr Nk,r. Remark 5.3. When n = 2m is even and k = m, the discriminant of Nm is given by the discriminant of two copies of N f and the inequalities of Proposition 5.2 are m := r∗ [π∗ (H m(X, Z))]⊥ 1 2 and, when p ≤ 19, αm ≤ rk H m f (X, Z)G − ℓm p (X) − logp discr N 2 1 2 m ; αm ≤ ℓm + (X) − logp discr N 2 1 2 m . Now, the objective will be to express Nk in term of exceptional divisors and calculate its discriminant. 5.2 Using blow-ups 7.31]. In this section X is a complex manifold and G an automorphism group of prime order p. The simplest case that we can consider for applying Proposition 5.2 is the following. Let s : eX → X be the blow-up of X in the fixed locus Z of G, let eG be the automorphism group induced by G on eX. When X is Kähler, the cohomology of eX can be calculated using [29, Theorem Let s : eX → X be the blow-up of X in Z. Let E = s−1(Z) be the exceptional divisor and j : E ֒→ eX be the embedding. Let h = c1(OE(1)) ∈ H 2(E, Z). Then, we have an isomorphism of Hodge structures: Theorem 5.4. ([29]) Let X be a Kähler manifold and let Z ⊂ X be a submanifold of codimension d. H k(X, Z) ⊕(cid:16)Ld−2 i=0 H k−2i−2(Z, Z)(cid:17) τ ∗+Pi j∗◦hi◦τ ∗ E Here, hi is the morphism by the cup-product with hi ∈ H 2i(E, Z). / H k(eX, Z). 15 / Now if eX/eG is smooth, eX/eG provides a resolution of singularities of X/G: X, s eπ eX eX/eG r π / X/G where the exceptional lattice of r (introduced in Definition 5.1) could be expressed in terms of the smooth. cohomology of the fixed locus Z via Theorem 5.4. So, in this section, we will study when such eX/eG is Remark 5.5. The triple (eX, eG, s) is a pullback of (X, G) (cf. Definition 3.15). At each fixed point of G, by Cartan [6, Lemma 1] we can locally linearize the action of G. Thus, at a fixed point x ∈ X, the action of G on X is locally equivalent to the action of G = hgi on Cn via where ξp is a p-th root of unity. Without loss of generality, we can assume that k1 ≤ ... ≤ kn ≤ p − 1. g = diag(ξk1 p , ..., ξkn p ), Lemma 5.6. Let X be a complex manifold of dimension n and Z be a connected component of Fix G, the fixed locus of an automorphism group G of prime order p. Let m = dim Z and E → Z be the exceptional divisor of the blow-up eX → X of X in Z. Let eG be the automorphism group induced by G on eX. Assume that the local action of G around a point of Z is given by g = diag(1, ..., 1, ξkm+1 p , ..., ξkn p ), p p , ξki matrices: , ..., ξki−1−ki p , ξki+1−ki for all m + 1 ≤ i ≤ n. diag(1, ..., 1, ξk1+m−ki then the different local actions of eG around its fixed points contained in E are given by the diagonal Proof. Let G = (cid:10)diag(ξk1 p )(cid:11) acting on Cn. Since k1 = ... = km = 0, we have to consider the Let fCn be the blow-up of Cn in 0 and eG the automorphism group of fCn induced by G. We will describe the action of eG on blow-up of Cm × Cn−m in Cm × {0}, which is given by Cm × ^Cn−m where ^Cn−m is the blow-up of Cn−m in 0. So, without loss of generality, we can assume that m = 0. , ..., ξkn−ki p , ..., ξkn ) p p fCn = {((x1, ..., xn), (a1 : ... : an)) ∈ Cn × Pn−1 rk ((x1, ..., xn), (a1, ..., an)) = 1} . We denote by the chart ai 6= 0. We have Oi =(cid:8)((x1, ..., xn), (a1 : ... : an)) ∈ Cn × Pn−1 ai 6= 0(cid:9) fCn ∩ Oi = {((x1, ..., xn), (a1, ..., ai−1, ai+1, ..., an)) ∈ Cn × Cn−1 xj = xiaj, j ∈ {1, ..., n}} . Hence, we have an isomorphism: f : fCn ∩ Oi ((x1, ..., xn), (a1 : ... : an)) / Cn / (a1, ..., ai−1, xi, ai+1, ..., an) Thus, the action of eG on fCn ∩ Oi provides an action on Cn given by the diagonal matrix p , ξki+1−ki , ..., ξki−1−ki diag(ξk1−ki , ..., ξkn−ki , ξki ). p p p p 16 / /     / / / We recall the following property from the proof of [24, Proposition 6]: Proposition 5.7. ([24]) The quotient Cn/G is smooth if and only if rk g − id = 1; that is k1 = ... = kn−1 = 0. Hence, we deduce from Lemma 5.6 the following. Corollary 5.8. Let X be a complex manifold of dimension n and G an automorphism group of prime order p. Let x ∈ Fix G. (i) The variety X/G is smooth in π(x) if and only if the local action around x is given by diag(0, ..., 0, ξk p ). (ii) Let eX be the blow-up of X in the connected components of Fix G of codimension ≥ 2 and eG the automorphism group induced by G on eX. The quotient eX/eG is smooth if and only if the local action of G around all x ∈ Fix G, is given by a matrix of the form: diag(0, ..., 0, ξk p , ..., ξk p ). So, the local actions which are given by a diagonal matrix of the form diag(0, ..., 0, ξk p , ..., ξk particular interest for our purpose of using Proposition 5.2. Keeping the same notation, we set: p ) are of Definition 5.9. A fixed point is simple if there is i ∈ {1, ..., n} such that k1 = ... = ki = 0 and ki+1 = ... = kn. Let X be a complex manifold and G an automorphism group of prime order acting on X. We have studied the case of a single blow-up, we now consider a sequence of blow-ups: Gk Xk πk Xk/Gk sk rk s2 · · · G1 / X1 π1 s1 G X, π / · · · r2 / / X1/G1 r1 / X/G where each si+1 is the blow-up of Xi in Fix Gi where X = X0 and G = G0. However, in most cases, it is not possible to find k such that Xk/Gk is smooth. Proposition 5.10. There exists k ∈ N such that Xk/Gk is smooth if and only if either all the fixed points of G are simple or p = 3. Moreover, in the case where p = 3, X2/G2 is smooth. Proof. By [6, Lemma 1], we can assume that X = Cn and First, if 0 is a simple fixed point, by Corollary 5.8, X1/G1 is smooth. If p = 3, we will show that G =(cid:10)diag(ξk1 X2/G2 is smooth. Let G =Ddiag(ξk1 3 )E acting on Cn. Without loss of generality, we can assume that all ki are different from 0. Let fCn be the blow-up of Cn in 0. By Lemma 5.6, in the chart ai 6= 0 the action of G1 is given by the diagonal matrix p )(cid:11) . 3 , ..., ξkn p , ..., ξkn diag(ξk1−ki , ..., ξki−1−ki 3 3 , ξki 3 , ξki+1−ki 3 , ..., ξkn−ki ). 3 As p = 3, there is a j such that k1 = ... = kj = 1 and kj+1 = ... = kn = 2. So, if i ≤ j, by permuting the i-th and the j-th coordinates of the chart ai 6= 0, we reduce the action to the form If i > j, by a permutation of coordinates we obtain diag(1, ..., 1, ξ3, ..., ξ3). diag(ξ2 3 , ..., ξ2 3, 1, ..., 1). 17   / /   /   / /       / / In both cases, these are simple fixed points. Hence, all the points of Fix G1 are simple fixed points. More- over, as there are no fixed points with both eigenvalues ξ3, ξ2 3 present in the diagonal matrix of the action, we can conclude that the components of Fix G1 with spectra (1, ..., 1, ξ3, ..., ξ3) and (ξ2 3 , ..., ξ2 3, 1, ..., 1) are disjoint and can be blown up independently. Hence, by Corollary 5.8, (ii), X2/G2 is smooth. Secondly, we will show that in the other cases, Xk/Gk will never be smooth. We start with dim X = 2, so that by [6, Lemma 1], we can assume that X = C2 and G =(cid:10)diag(ξp, ξα Let ξp be a fixed non-trivial p-th root of the unity, for x ∈ Fix Gi, we can write: p )(cid:11), p > 3 and α not equal to 0 or 1. where ∼ means that the local actions are the same. Hence, we can define a sequence as follows: (Xi, Gi, x) ∼ (C2,(cid:10)diag(ξp, ξβ p )(cid:11) , 0), ui =(cid:8)β ∈ Z /p Z ∃x ∈ Xi : (Xi, Gi, x) ∼ (C2,(cid:10)diag(ξp, ξβ p )(cid:11) , 0)(cid:9) . such that Xi/Gi is smooth. Let i be the smallest integer such that Xi/Gi is smooth. By Corol- lary 5.8, (i), we have ui = {0} and we can write ui−1 = {α1, ..., αk}. Let x ∈ Fix Gi−1 such that For instance, u0 = (cid:8)α, 1 (Xi−1, Gi−1, x) ∼ (C2,(cid:10)diag(ξp, ξαj action of(cid:10)diag(ξp, ξαj α o,... Now assume that there is i ∈ N α(cid:9), u1 = nα − 1, p )(cid:11) , 0), with αj ∈ ui−1 \ {0}. Let fC2 be the blow-up of C2 in 0. The p )(cid:11) on fC2 has two fixed points a1 and a2 with (see Lemma 5.6): 1−α , 1−α α−1 , α 1 ), 0) and p p )(cid:11) , a1) ∼ (C2, diag(ξp, ξαj −1 (fC2,(cid:10)diag(ξp, ξαj p )(cid:11) , a2) ∼ (C2, diag(ξαj (fC2,(cid:10)diag(ξp, ξαj (Xi−2, Gi−2, x) ∼(cid:16)C2,Ddiag(ξp, ξ p , ξ1−αj 1, ..., α′ α′ j p ), 0). Hence, αj − 1 ∈ ui, but ui = {0}, so, necessarily, αj = 1 and so ui−1 = {1}. We do the same calculation with ui−2 = {α′ k}. Let x ∈ Fix Gi−2 be such that: with α′ a′ 1 and a′ 2 with (see Lemma 5.6): j ∈ ui−2 \ {0, 1}. We remark that ui−2 \ {0, 1} is not empty because Xi−1/Gi−1 is not smooth by p )E , 0(cid:17) . definition of i. Let fC2 be the blow-up of C2 in 0. The action ofDdiag(ξp, ξ )E on fC2 has 2 fixed points ), 0(cid:17) ), 0(cid:17) . ), 0 . p )E , a1(cid:17) ∼(cid:16)C2, diag(ξp, ξ p )E , a1(cid:17) ∼(cid:16)C2, diag(ξ ), 0(cid:17) ∼C2, diag(ξp, ξ (cid:16)fC2,Ddiag(ξp, ξ (cid:16)fC2,Ddiag(ξp, ξ (cid:16)C2, diag(ξ 1−α′ α′ j j p , ξ p 1−α′ α′ j j p , ξ p But and αj′ p 1−α′ j α′ p j −1 α′ j α′ j α′ j p j = 2 and α′ j = 1 2 . In Z /p Z, 2 = 1 2 if, and only if, p = 3. Hence, p = 3 and we are Hence, necessarily, α′ done. Without loss of generality, we can assume that all the ki are different from 0. Moreover, by assumption, p > 3 and the ki are not all equal. Without loss of generality, we can assume that k1 = 1. Since not all the ki are equal, there is j ∈ {1, ..., n} such that kj 6= 1. We also denote α = kj. We denote X ′ = C2 Now, we assume n > 2. By [6, Lemma 1], we can assume that X = Cn and G =(cid:10)diag(ξk1 and G′ =(cid:10)diag(ξp, ξα p )(cid:11). And we define as before the sequence: i =(cid:8)β ∈ Z /p Z ∃x ∈ X ′ i, x) ∼ (C2,(cid:10)diag(ξp, ξβ p )(cid:11) , 0)(cid:9) . p , ..., ξkn p )(cid:11). i : (X ′ p , ξk2 i, G′ u′ 18 We also define the following sequence: Ui = {β ∈ Z /p Z ∃x ∈ Xi : (Xi, Gi, x) ∼ (Cn,(cid:10)diag(ξp, ξβ p , ξt3 p , ..., ξtn p )(cid:11) , 0)(cid:9) . We have to show that Ui 6= {0} for all i ∈ N. Indeed, Ui ⊃ u′ i ∈ N. i, and we have seen that u′ i 6= {0} for all 6 Proof of Theorem 1.1 and outcomes Theorem 1.1 is a direct consequence of the following result which will be proven in this section. Theorem 6.1. Let X be a compact Kähler manifold of complex dimension n and G = hgi an automor- phism group of prime order p ≤ 19. Let c := Codim Fix G. Assume that (i) H ∗(X, Z) and H ∗(Fix G, Z) are p-torsion-free, (ii) h∗(Fix G, Q) ≥P1≤q<p ℓ∗ (iii) all fixed points of G are simple, q(X), (iv) c ≥ n 2 + 1. Then, for all 2n − 2c + 1 ≤ k ≤ 2c − 1, (X, G) is H k-normal. We first explain why Theorem 1.1 follows from Theorem 6.1. Boissière, Nieper-Wisskirchen and Sarti in [5, Corollary 4.3] have proven the following formula: Proposition 6.2. Let X be a compact connected orientable manifold of dimension n and G an auto- morphism group of prime order p. If the spectral sequence of equivariant cohomology with coefficients in Fp degenerates at the second page, then: h∗(Fix G, Fp) = X1≤q<p ℓ∗ q(X). Hence, since we assume that H ∗(Fix G, Z) is p-torsion-free, we obtain Theorem 1.1 from Theorem 6.1 and Proposition 6.2. Before proving Theorem 6.1 in Section 6.1, we discuss its hypotheses in few words. Remark 6.3. The condition p ≤ 19 is necessary to work with the spectral sequence of equivariant cohomology with coefficients in Z and use Corollary 4.2 (see Lemma 6.10). However, it is possible to avoid the condition p ≤ 19 working, under some additional assumptions, with the spectral sequence of equivariant cohomology with coefficients in Fp. I made the choice not to provide such a theorem, since in practice it is quite rare to have to work with p > 19; for instance, until now, none symplectic automorphism of such order is known on irreducible symplectic manifolds. Remark 6.4. The condition on the codimension of Fix G in Theorem 6.1 could be improved. However, hypothesis (ii) would have to be replaced by a more elaborate condition. Remark 6.5. As explained in Section 5.2, to avoid the condition that the fixed points of G are simple, we need to find another practical way of resolving the singularities of X/G. We could also use two blow-ups in the case p = 3. We provide such an example in Section 7.6. 6.1 Proof of Theorem 6.1 The idea of the proof is to calculate the discriminant of the exceptional lattice Nk,r, where r is obtained from a blow-up, and apply Proposition 5.2. 19 All the fixed points of G are simple, so we recall from Section 5.2 that we are in the configuration of the following commutative diagram: X π / X/G, (17) s eπ eX eX/eG r π∗ π∗ eπ∗ eπ∗ H k(M, Z) H k(X, Z) r∗ s∗ H k(fM , Z) H k(eX, Z). eG be the automorphism group induced by G and eX/eG is smooth. We also denote M = X/G, fM = eX/eG, V = X r Fix G, U = π(V ), F = s−1(Fix G), we use the same symbol F for its image byeπ and we adopt the notation introduced in Section 2.2 and Section 3.1. Diagram (17) induces a commutative diagram on cohomology: Lemma 6.6. Let 0 ≤ k ≤ 2n and x ∈ H k(X, Z). Then: (i) eπ∗(s∗(x)) = r∗(π∗(x)) + (p − torsion), (ii) Nk,r =eπ∗(s∗(H k(X, Z) ⊕ H 2n−k(X, Z)))⊥ Proof. We remark that (eX,eG, s) is a pull-back of (X, G), so (i) follows from Lemma 3.16. Applying (i) f (fM , Z), we obtain: to H k (18) f . eπ∗(s∗(H k(X, Z))f = r∗(π∗(H k(X, Z))f ). Then, it follows (ii) by definition of Nk,r. Lemma 6.7. (i) For all i ≤ 2c − 2, H i(X, Z) = H i(V, Z), (ii) H 2c−1(V, Z) is p-torsion-free. Proof. We have the following exact sequence: H i(X, V, Z) / H i(X, Z) / H i(V, Z) / H i+1(X, V, Z). Moreover, by Thom's isomorphism for each connected component Sm of Fix G: H i(X, X r Sm, Z) = H i−2 Codim Sm(Sm, Z). Hence: H i(X, V, Z) = XSm⊂Fix G H i(X, X r Sm, Z) = 0, for all i ≤ 2c − 1. So, we get (i). Moreover, we obtain the following exact sequence: 0 / H 2c−1(X, Z) / H 2c−1(V, Z) / H 2c(X, V, Z). (19) From the hypotheses, we know that H 2c−1(X, Z) and H ∗(Fix G, Z) are p-torsion-free; so, by Thom's isomorphism, H 2c(X, V, Z) is p-torsion-free. Then, (ii) follows from (19). We consider the following commutative diagram: g deπ∗ H k(N fM/F , N fM/F − 0, Z) = H k(fM , U, Z) H k(N eX/F , N eX/F − 0, Z) = H k(eX, V, Z) h / fM/F −0 are vector bundles minus the zero section. We denote Tk := h(H k(eX, V, Z)). / H k(fM , Z) / H k(eX, Z), (20) eπ∗ 20 where N eX/F −0 and N / /     / 1 1   q q   - - m m / / / / / /   /   Lemma 6.8. When 2n − 2c + 1 ≤ k ≤ 2c − 1, we have: Moreover: H k(eX, Z) = s∗(H k(X, Z)) ⊕ Tk. Tk⊥ s∗(H 2n−k(X, Z)). Proof. By Thom's isomorphism, H k(eX, V, Z) = H k−2(F, Z), and the map h can be identified with the morphism j∗ : H k−2(F, Z) → H k(eX, Z), where j is the inclusion in eX. As in the proof of [29, Theorem 7.31], the map is surjective and its kernel is the image of the map (s∗, j∗) : H k(X, Z) ⊕ H k−2(F, Z) → H k(eX, Z) MSd⊂Fix G H k−2rd (Sd, Z) → H k(X, Z) ⊕ H k−2(F, Z), where rk is the codimension of the component Sd of Fix G. But in our caseLSd⊂Fix G H k−2rd (Sd, Z) = 0. The orthogonality of this sum follows from the projection formula. Moreover, Theorem 5.4 also shows that: rk Tk = rk MSd⊂Fix G rd−2Mi=0 H k−2i−2(Sd, Z), where rd is the codimension of the component Sd of Fix G. Hence, when 2n − 2c + 1 ≤ k ≤ 2c − 1, we obtain: rk Tk = h2∗+ǫ(Fix G, Z), (21) where ǫ = 1 if k is odd and ǫ = 0 if k is even. Lemma 6.9. We have: (i) When 2n − 2c + 1 ≤ k ≤ 2c − 1: where ǫ = 1 if k is odd and ǫ = 0 if k is even. discreπ∗(Tk ⊕ T2n−k) = p2h2∗+ǫ(Fix G,Z), (ii) Nk,r is the primitive over-lattice of eπ∗(Tk ⊕ T2n−k). Proof. By Lemma 6.8, we have: H k(eX, Z) ⊕ H 2n−k(eX, Z) = s∗(cid:0)H k(X, Z) ⊕ H 2n−k(X, Z)(cid:1) ⊕⊥ (Tk ⊕ T2n−k) . Since eX and X are smooth, H k(eX, Z) ⊕ H 2n−k(eX, Z) and H k(X, Z) ⊕ H 2n−k(X, Z) are unimodular. It follows that Tk ⊕ T2n−k is unimodular. Moreover, by Thom's isomorphism H k(eX, V, Z) = H k−2(F, Z), and the map h can be identified with the morphism j∗ : H k−2(F, Z) → H k(eX, Z), where j is the inclusion in eX. It follows that Tk ⊂ H k(eX, Z) eG. Hence, by Lemma 3.6 (ii), we obtain: (22) From (21), when 2n − 2c + 1 ≤ k ≤ 2c − 1, discreπ∗(Tk ⊕ T2n−k) = prk Tk⊕T2n−k . rk Tk = h2∗+ǫ(Fix G, Z), where ǫ = 1 if k is odd and ǫ = 0 if k is even. Remark that if 2n − 2c + 1 ≤ k ≤ 2c − 1 then 2n − 2c + 1 ≤ 2n − k ≤ 2c − 1. Hence, rk Tk = rk T2n−k = h2∗+ǫ(Fix G, Z). So, it follows (i). Now taking the image of (22) byeπ∗, we obtaineπ∗(s∗(cid:0)H k(X, Z) ⊕ H 2n−k(X, Z)(cid:1))⊕⊥eπ∗((Tk ⊕ T2n−k)) as a sublattice of full rank of H k(fM , Z) ⊕ H 2n−k(fM , Z). Moreover, applying Proposition 3.1, the direct sum remains orthogonal. So, (ii) follows from Lemma 6.6 (ii). 21 The Thom class of N fM/F is sent by deπ∗ to p times the Thom class of N eX/F . This provides: : deπ∗(H k(N Then, by commutativity of diagram (20) and Proposition 3.1, we have g(H k(fM , U, Z)) = eπ∗(Tk). fM/F − 0, Z)) = pH k(N eX/F , N eX/F − 0, Z). follows the exact sequence: fM /F , N It (23) 0 /eπ∗(Tk) / H k(fM , Z) / H k(U, Z). In order to calculate the discriminant of Nk,r which is by Lemma 6.9 (ii), the primitive over lattice of p (U, Z) be the eπ∗(Tk) ⊕eπ∗(T2n−k), we need to know the p-torsion of H k(U, Z) ⊕ H 2n−k(U, Z). Let H k To lighten a bit the notation, in the following calculations, we denote ℓk Fp-vector spaces given by the p-torsion part of H k(U, Z). ∗(X) and Nk for Nk,r. ∗ for ℓk Lemma 6.10. Let k ≤ 2c − 1, we have if k = 2a: If k = 2a + 1: dimFp H k p (U, Z) ≤ dimFp H k p (U, Z) ≤ a−1Xj=0 a−1Xj=0 ℓ2j + + ℓ2j+1 − . ℓ2j+1 + + ℓ2j − . a−1Xj=0 aXj=0 Proof. Since G acts freely on V , we recall from Section 4 that H ∗(U, Z) ≃ H ∗ G(V, Z). Using the second page of the spectral sequence of equivariant cohomology, it follows that dimFp H k p (U, Z) ≤ kXi=1 dimFp H i(G, H k−i(V, Z)). (24) The group H 0(G, H k(V, Z) does not provide any contribution to H k H 0(G, H k(V, Z) = H k(V, Z)G and by Lemma 6.7, H k(V, Z) is p-torsion-free. p (U, Z) because by Corollary 4.2, Moreover, by Lemma 6.7 (i), (24) becomes: dimFp H k p (U, Z) ≤ kXi=1 dimFp H i(G, H k−i(X, Z)). Hence, by Corollary 4.2, if k = 2a: If k = 2a + 1: dimFp H k p (U, Z) ≤ dimFp H k p (U, Z) ≤ a−1Xj=0 a−1Xj=0 From (23) and Lemma 6.9 (ii): ℓ2j + + ℓ2j+1 − . ℓ2j+1 + + ℓ2j − . a−1Xj=0 aXj=0 dimFp(cid:18) Nk eπ∗(Tk ⊕ T2n−k)(cid:19) ≤ dimFp H k p (U, Z) ⊕ H 2n−k p (U, Z). 22 / / / When 2n − 2c + 1 ≤ k ≤ 2c − 1, we can apply Lemma 6.10 to H k k = 2a, p (U, Z) and to H 2n−k p (U, Z). When Hence, by Lemma 6.9 (i) and (3): Nk dimFp(cid:18) eπ∗(Tk ⊕ T2n−k)(cid:19) ≤ a−1Xj=0 logp discr Nk ≥ 2h2∗(Fix G, Z) − 2 a−1Xj=0 logp discr Nk ≥ 2h2∗+1(Fix G, Z) − 2 a−1Xj=0 Similarly, when k = 2a + 1, we obtain: ℓ2j + + ℓ2j+1 − + a−1Xj=0 n−a−1Xj=0 ℓ2j + + n−a−1Xj=0 ℓ2j+1 − . ℓ2j + + ℓ2j+1 − + a−1Xj=0 n−a−1Xj=0 ℓ2j + + n−a−1Xj=0 ℓ2j+1 + + ℓ2j − + aXj=0 n−a−2Xj=0 ℓ2j+1 + + ℓ2j+1 −  . − . n−a−1Xj=0 ℓ2j (25) (26) Now the strategy to finish the proof is the following. Equations (25) and (26) alone are not practical but, as we will see, their sum can be simplified. Hence, the objective is to use (25)+(26) to prove our theorem. Notice first that since c ≥ n 2 + 1, the interval {2n − 2c + 1, ..., 2c − 1} always has cardinal at least 3. Hence, we can always find k such that k, k+1 ∈ {2n − 2c + 1, ..., 2c − 1}. Using (25)+(26) we will provide an upper bound for logp discr Nk + logp discr Nk+1 which will provide the H k ⊕ H k+1-normality. Then, the claim follows covering {2n − 2c + 1, ..., 2c − 1} by pairs of consecutive numbers {k, k + 1}. So, let k, k +1 ∈ {2n − 2c + 1, ..., 2c − 1}, we can assume that k = 2a, if k is odd the proof is identical. Hence, summing (25)+(26) and applying Proposition 3.9 (ii), we have: ℓj + + ℓj + + logp discr Nk + logp discr Nk+1 − 2h∗(Fix G, Z) ≥ −2 k−1Xj=0 ≥ −2 k−1Xj=0 ≥ −2 k−1Xj=0 ≥ −2 X1≤q<p logp discr Nk + logp discr Nk+1 ≥ 2(cid:2)ℓk ℓj + + kXj=0 kXj=0 kXj=0 and, by Proposition 5.2, Hence, by hypothesis, 2n−k−2Xj=0 2n−k−2Xj=0 2nXj=k+2 +  . + (cid:3) , ℓ∗ q − ℓk + − ℓk+1 ℓj − + ℓj + + ℓj − + ℓ2n−j + + ℓj − + ℓj + + ℓj − 2n−k−1Xj=0 −  2n−k−1Xj=0 −  2nXj=k+1 ℓ2n−j ℓ2n−j + + ℓk+1 (27) αk = α2n−k = αk+1 = α2n−k−1 = 0. Remark 6.11. We have proved a bit more than what is claimed in Theorem 6.1. Under its hypotheses, we have seen thateπ∗(s∗(H k(X, Z)⊕H 2n−k(X, Z))) is primitive, in H ∗(fM , Z), for all 2n−2c+1 ≤ k ≤ 2c−1. Indeed, from (27) and Proposition 5.2, discr Nk = pℓk + . Moreover, from (16) and Corollary 3.8, + . So, by (18), we have: (X, Z))) = pℓk discr r∗(π∗(H k f (X, Z) ⊕ H 2n−k f H 2n−k(X, Z))) ⊂ N ⊥ logp discr Nk = discreπ∗(s∗(H k Since fM is smooth, by (4), we have discr Nk = discr N ⊥ k . So by (28), eπ∗(s∗(H k(X, Z) ⊕ H 2n−k(X, Z))) = N ⊥ H ∗(fM , Z). f (X, Z) ⊕ H 2n−k (X, Z))). f k . However, by Lemma 6.6 (ii),eπ∗(s∗(H k(X, Z) ⊕ k , which is primitive in (28) 23 6.2 Refinement on the codimension of the fixed locus We provide an extension of Theorem 1.1 when Fix G is a bit larger. As mentioned in Remark 6.4, we will need more technical conditions. However, when Codim Fix G =(cid:6) n result which is interesting in practice (see for instance the application in [13, Section 15]). We will see another example in Section 7.3. 2(cid:7), it is still possible to provide a Theorem 6.12. Let X be a compact Kähler manifold of complex dimension n and G = hgi be an automorphism group of prime order p ≤ 19. Assume that: (i) H ∗(X, Z) and H ∗(Fix G, Z) are p-torsion-free. (ii) h2∗(Fix G, Q) ≥ ℓ2∗ + (X) + ℓ2∗+1 − (X) if n is even and h2∗+1(Fix G, Q) ≥ ℓ2∗+1 + (X) + ℓ2∗ − (X) if n is odd. (iii) All fixed points of G are simple. 2(cid:7). (iv) Codim Fix G =(cid:6) n (v) When n is even, let Σ be the component of Fix G of dimension n 2 . We assume that Σ is connected. Let [Σ] be the cohomology class associated to Σ. We assume that [Σ] is not of torsion (in particular non 0) and is primitive in H n(X, Z). Then, (X, G) is H n-normal. This section is dedicated to the proof of this theorem. The proof is different when n is even or odd. When n is odd, it is a straightforward consequence of the results in the previous section. When n is even we have to do more work. Let us first prove the theorem when n is odd. Case: n is odd We have n = 2m + 1 and Codim Fix G = m + 1, so n ≤ 2m + 1. It follows that Lemmas 6.6, 6.7, 6.8, 6.9 remain true. The problem is that the interval {2n − 2c + 1, ..., 2c − 1} has only cardinality one; hence it is no longer possible to sum (25) and (26). However, (26) remains true and becomes Applying Proposition 3.9, it follows: logp discr Nn ≥ 2h2∗+1(Fix G, Z) − 2 m−1Xj=0 logp discr Nn ≥ 2h2∗+1(Fix G, Z) − 2 m−1Xj=0 ℓ2j+1 + + mXj=0 logp discr Nn ≥ 2h2∗+1(Fix G, Z) − 2(cid:2)ℓ2∗+1 Then, we get the result using hypothesis (ii) and Proposition 5.2. So: ℓ2j+1 + + ℓ2j − + mXj=0 m−1Xj=0 ℓ2j+1 + + mXj=0 ℓ2j − . ℓ2j − + ℓ2n−2j−1 + + m−1Xj=0 ℓ2n−2j − mXj=0  . + + ℓ2∗ − − ℓ2m+1 + (cid:3) . Case: n is even In this case n = 2m and 2m > 2c − 1. We have to provide a variation of each lemma of Section 6.1 which are not true anymore. Let us start with Lemma 6.7. Lemma 6.13. (i) H 2m−1(X, Z) = H 2m−1(V, Z), (ii) H 2m(V, Z) is p-torsion-free. 24 Proof. We have the following exact sequence: H 2m−1(X, V, Z) / H 2m−1(X, Z) / H 2m−1(V, Z) / H 2m(X, V, Z) ρ / H 2m(X, Z). By Thom's isomorphism H 2m−1(X, V, Z) = 0 and T : H 2m(X, V, Z) ≃ H 0(Σ, Z). In [29, Section 11.1.2], it is explained that the cohomology class associated to Σ can be expressed as [Σ] = ρ(T −1(1)). By assumption [Σ] is not of torsion, it follows that ρ is injective. So, Moreover, we obtain the following exact sequence: H 2m−1(X, Z) = H 2m−1(V, Z). 0 / H 0(Σ, Z) ρ◦T −1 / H 2m(X, Z) / H 2m(V, Z) / H 2m+1(X, V, Z). From the hypotheses, we know that H 2m(X, Z) and H ∗(Fix G, Z) are p-torsion-free; so by Thom's isomorphism H 2m+1(X, V, Z) is p-torsion-free. Moreover, we assumed that [Σ] = ρ(T −1(1)) is primitive. Hence, it follows (ii). We consider the following commutative diagram: g eπ∗ H 2m(N fM /F , N fM/F − 0, Z) = H 2m(fM , U, Z) H 2m(N eX/F , N eX/F − 0, Z) = H 2m(eX, V, Z) h / fM/F −0 are vector bundles minus the zero section. We denote R := h(H 2m(eX, V, Z)). / H 2m(fM , Z) / H 2m(eX, Z), (29) deπ∗ where N eX/F −0 and N We prove the following variation of Lemma 6.8: Lemma 6.14. We can write: with R = T ⊕ Z[Σ]. H 2m(eX, Z) = s∗(H 2m(X, Z)) ⊕⊥ T, Proof. As in proof of Lemma 6.8, the map h can be identified with the morphism j∗ : H 2m−2(F, Z) → H 2m(eX, Z); moreover, the map is surjective and its kernel is the image of the map (s∗, j∗) : H 2m(X, Z) ⊕ H 2m−2(F, Z) → H 2m(eX, Z) MSd⊂Fix G H 2m−2rd (Sd, Z) → H 2m(X, Z) ⊕ H 2m−2(F, Z), where rd is the codimension of the component Sd of Fix G. However, in our case MSd⊂Fix G H 2m−2rd (Sd, Z) = H 0(Σ, Z). So, we obtain H 2m(eX, Z) = s∗(H 2m(X, Z)) ⊕ T, and the orthogonality of the sum can also be deduced using the projection formula. 25 / / / / / / / /   /   (ii): Let eR be the minimal primitive over-lattice of eπ∗(R) in H 2m(fM , Z) and eT the minimal primitive over-lattice of eπ∗(T ) in H 2m(fM , Z). From Lemmas 6.14 and 6.6 (ii), by the same proof as Lemma 6.9 As in the previous proof, we need to calculate the discriminant of eT . By the property of the Thom Then, by commutativity of diagram (20) and Proposition 3.1, we have g(H 2m(fM , U, Z)) = eπ∗(R). It fM/F − 0, Z)) = pH 2m(N eX/F , N eX/F − 0, Z). deπ∗(H 2m(N N2m = eT 2⊕. follows the exact sequence: isomorphism, fM /F , N (30) / H 2m(U, Z). With the same proof as Lemma 6.10, we obtain from Lemma 6.13: 0 /eπ∗(R) / H 2m(fM , Z) m−1Xj=0 dimFp H 2m p (U, Z) ≤ ℓ2j + + ℓ2j+1 − . m−1Xj=0 However, (31) will provide only an upper bound for dimFp eR eπ∗(R) with eT eπ∗(T ) . eR eπ∗(R) . We need some lemmas to compare Lemma 6.15. There exists x ∈eπ∗(T ) such that x+(−1)n−1eπ∗(s∗(Σ)) Proof. Let s1 : Y → X be the blow-up of X in Σ and Σ1 the exceptional divisor, and s2 : eX → Y the 2(Σ1) andeΣ =eπ∗(Σ2). blow-up in the other components of Fix G such that s = s2 ◦ s1. We denote Σ2 = s∗ Consider the following diagram: ∈ H 2m(fM , Z). l2 p Σ2 g2 Σ1 g1 Σ  / eX s2 l1 l0 / Y s1 / X, where l0, l1 and l2 are the inclusions and gi := siΣi , i ∈ {1, 2}. We have eπ∗(O eX ) = OfM ⊕ L, with Lp = OfM(cid:16)−(cid:16)PSk⊂Fix G\ΣfSk(cid:17) −eΣ(cid:17), where each fSk is the exceptional divisor associated to the connected component Sk 6= Σ of Fix G. Thus, (31) (32) (33) It follows that By Lemma 3.6 (i), we get: p PSk⊂Fix G\ΣfSk +eΣ ∈ H 2(fM , Z). PSk⊂Fix G\ΣfSk +eΣ !m ∈ H 2m(fM , Z). x +eπ∗(Σm ∈ H 2m(fM , Z), 2 ) p p s∗ 1l0∗(Σ) = l1∗(cm−1(E)), 26 with x ∈eπ∗(T ). Now, it remains to calculate Σm 2 . By [11, Proposition 6.7], we have / / /     /       /    / where E := g∗ 1(NΣ/X )/NΣ1/Y . So s∗ (−1)m−1−ici(g∗ 1l0∗(Σ) =l1∗ m−1Xi=0 =l1∗ m−1Xi=1 + (−1)m−1l1∗(cid:0)c1(NΣ1/Y )m−1(cid:1) =l1∗ m−1Xi=1 1(NΣ/X )) · c1(NΣ1/Y )m−1−i! 1(NΣ/X )) · c1(NΣ1/Y )m−1−i! 1(NΣ/X )) · c1(NΣ1/Y )m−1−i! (−1)m−1−ici(g∗ (−1)m−1−ici(g∗ + (−1)m−1Σm 1 . By applying s∗ 2, we get: Σm 2 = (−1)m−1 (s∗(Σ) − s∗ 2l1∗(a)) , where a =Pm−1 i=1 (−1)m−1−ici(g∗ The result follows from (33). 1(NΣ/X )) · c1(NΣ1/Y )m−1−i ∈ T . And pushing forward viaeπ∗, we get: eπ∗(Σm 2 ) = (−1)m−1 (eπ∗(s∗(Σ)) −eπ∗ (s∗ 2l1∗(a))) . Lemma 6.16. We have: dimFp ≤ dimFp H 2m p (U, Z) − 1, where H 2m p (U, Z) is the p-torsion part of H 2m(U, Z). Proof. From (31): dimFp ≤ dimFp H 2m p (U, Z). eT eπ∗(T ) eR eπ∗(R) p ∈ p In both cases: ∈ (eR/eπ∗(R)) \ (eT /eπ∗(T )), if not x+(−1)m−1eπ∗(s∗(Σ)) But by Lemma 6.15, we already know that there exists x ∈ eπ∗(T ) such that x+(−1)m−1eπ∗(s∗(Σ)) H 2m(fM , Z). We are going to deduce dimFp eT /eπ∗(T ) from dimFp eR/eπ∗(R). If eπ∗(s∗(Σ)) is divisible by p in H 2m(fM , Z), then eπ∗(s∗(Σ)) ∈ (eR/eπ∗(R)) \ (eT /eπ∗(T )). eR eπ∗(R) m−1Xj=0 So, from Lemma 6.16 and (32): eT eπ∗(T ) m−1Xj=0 ℓ2j+1 − − 1. = dimFp ℓ2j + + dimFp dimFp (34) p − 1. ≤ From Theorem 5.4 and Lemma 6.14: eT eπ∗(T ) rk T = rk MSd⊂Fix G rd−2Mi=0 H 2m−2i−2(Sd, Z), where rd is the codimension of the component Sd of Fix G. Hence, we obtain: rk T = h2∗(Fix G, Z) − 2, With the same proof as Lemma 6.9, it follows: discreπ∗(T ) = ph2∗(Fix G,Z)−2. 27 (35) (36) Combining (36), (34) and (3), then applying Proposition 3.9, we have ℓ2j+1 −  ℓ2j + + ℓ2j+1 − + ℓ2j + + logp discreT − h2∗(Fix G, Z) ≥ −2 m−1Xj=0 ≥ − m−1Xj=0 logp discreT ≥ ℓ2m m−1Xj=0 m−1Xj=0 ≥ −ℓ2∗ + − ℓ2∗+1 − + ℓ2m + . + . Finally, we conclude with (30) and Remark 5.3. ℓ2n−2j + + m−1Xj=0 ℓ2n−2j−1 − m−1Xj=0  Hence, using hypothesis (ii): Remark 6.17. Same remark as Remark 6.11, we have also proved thateπ∗(s∗(H n(X, Z))) is primitive in H ∗(fM , Z). 7 Examples of applications 7.1 Simply connected surfaces Boissière, Sarti and Nieper-Wisskirchen have proved the following ([5, Proposition 4.5]): Proposition 7.1. Let X be a simply connected compact surface and G be an automorphism group with at least a fixed point. Then, the spectral sequence of equivariant cohomology with coefficients in Fp is degenerate a the second page. Then, from this proposition and Theorem 1.1, we obtain the following corollary. Corollary 7.2. Let X be a simply connected compact Kähler surface and G be an automorphism group of prime order p ≤ 19. Assume that Fix G is finite but not empty and contains simple points. Then, (X, G) is H 2-normal. This result can be applied for example to a K3 surface endowed with a symplectic involution. 7.2 The case of the K3[2]-type manifolds Boissière, Sarti and Nieper-Wisskirchen also proved in [5, Theorem 1.1] that for X a K3[2]-type man- ifold and G an automorphism group of prime order p /∈ {2, 5, 23}, the spectral sequence of equivariant cohomology with coefficients in Fp is degenerate a the second page. Together with Theorem 1.1, this implies: Corollary 7.3. Let X be a K3[2]-type manifold and G be an automorphism group of prime order p /∈ {2, 5, 23}. Let c := Codim Fix G. Assume that: (i) all fixed points of G are simple; (ii) c ≥ 3. Then, (X, G) is H 2 and H 4-normal. We remark that H ∗(X, Z) is torsion-free by [14] and H ∗(Fix G, Z) is torsion-free because Fix G con- tains only isolated points or smooth complex curves. So, H 4-normality follows from Theorem 1.1. It remains to show the H 2-normality. To do so, we will use Proposition 3.14. Lemma 7.4. Let X be a topological space. Let t and k be integers and p a prime number. Assume that H ∗(X, Z) is p-torsion-free. We denote by x ∈ H ∗(X, Fp) the reduction modulo p of an element x ∈ H ∗(X, Z). 28 If the cup product map Symt H k(X, Q) → H kt(X, Q) is an isomorphism and Hkt(X,Z) Symt Hk(X,Z) is p-torsion- free, then: S : Symt H k(X, Fp) → H kt(X, Fp) x1 ⊗ ... ⊗ xt 7→ x1 · ... · xt is an isomorphism. Proof. We prove the injectivity. Let x1 ⊗ ... ⊗ xt ∈ Symt H k(X, Fp) such that x1 · ... · xt = 0. Then, there exists y ∈ H kt(X, Z) y ∈ H kt(X, Z)/ Symt H k(X, Z) is a p-torsion element (here such that x1 · ... · xt = py. Hence, y is the class of y modulo Symt H k(X, Z)). Hence, by the hypothesis y = 0. It follows that y ∈ Symt H k(X, Z), so y = y1 · ... · yt with yi ∈ H k(X, Z). Since Symt H k(X, Q) → H kt(X, Q) is injective, x1 ⊗ ... ⊗ xt = py1 ⊗ ... ⊗ yt. So, x1 ⊗ ... ⊗ xt = 0. We prove the surjectivity. Let y ∈ H kt(X, Fp), with y ∈ H kt(X, Z). Since Symt H k(X, Q) → H kt(X, Q) is an isomor- phism, there is q ∈ N and x1 ⊗ ... ⊗ xt ∈ Symt H k(X, Z) such that 1 q x1 · ... · xt = y. Hence, y ∈ H kt(X, Z)/ Symt H k(X, Z) is a q-torsion element. But since H kt(X, Z)/ Symt H k(X, Z) is p-torsion-free, p does not divide q. And S ( 1 q x1 ⊗ ... ⊗ xt) = y. When X is a K3[2]-type manifold, we know from [28] that Sym2 H 2(X, Q) → H 4(X, Q) is an isomor- phism. Moreover, from [5, Proposition 6.6], we know that: H 4(X, Z)/ Sym2 H 2(X, Z) = (Z /2 Z)23 ⊕ (Z /5 Z). It follows from the previous lemma that for p /∈ {2, 5}, Sym2 H 2(X, Fp) ≃ H 4(X, Fp). (37) So, by Proposition 3.14, we obtain the following corollary which finishes the proof of Corollary 7.3. Corollary 7.5. Let X be a K3[2]-type manifold and G be an automorphism group of prime order p /∈ {2, 5}. If (X, G) is H 4-normal, (X, G) is H 2-normal. We describe the following example as an application of Corollary 7.3. First, we recall that an auto- morphism on a Hilbert scheme of points S[n] on a K3 surface S is called natural if it is the automorphism induced by an automorphism on the K3 surface S. Example 7.6. Let X be a Hilbert scheme of 2 points on a K3 surface and G be a natural non-symplectic automorphism group of order 3 on X. Then, (X, G) is H 4 and H 2-normal. Proof. Let G be such an isomorphism group on S[2]. Since the action on the K3 surface is non- symplectic, the fixed points will be simple. Moreover, from [4, Section 6.1], Fix G consists in 3 isolated points and 3 rational curves. Remark 7.7. An analogous of Corollary 7.3 could also be stated when p = 2 or 5 and X is a Hilbert scheme of 2 points on a K3 surface and G a natural automorphism group using statement (2) of [5, Theorem 1.1]. 7.3 Integral basis of the Hilbert scheme of two points on a surface In this section, we complete the work started in [13, Section 7]. Let A be a smooth compact Kähler surface with H ∗(A, Z) 2-torsion-free and A[2] the Hilbert scheme of two points. It can be constructed as follows: Consider the direct product A × A. Denote b : Bl∆(A × A) → A × A 29 the blow-up along the diagonal h : ∆ ≃ A with exceptional divisor E (for simplicity we denote by x the pull-back h∗(x) for all x ∈ H ∗(A, Z)). Let f be a fiber of the map E → ∆. Let j : E ֒→ Bl∆(A × A), g : ∆ ֒→ A × A be the embeddings and p1 : A × A → A, p2 : A × A → A the projections. We also denote by A the class which generates H 0(A, Z) and by pt the class which generates H 4(A, Z). We have the following commutative diagram: / Bl∆(A × A) b (38) E bE j g / A × A h ∆ ❆❆❆❆❆❆❆❆ ✵✵✵ ✵✵✵✵ ✵✵✵ ✵✵✵✵ h A A. p1 ysssssssssss ✆✆✆✆✆✆✆✆✆✆✆✆✆✆✆✆ p2 The action of S2 extends to an action on Bl∆(A × A). Then, A[2] = Bl∆(A × A)/S2 and we denote by π : Bl∆(A × A) → A[2] the quotient map. Moreover, from [26, Theorem 2.2], we know that H ∗(A[2], Z) is 2-torsion-free. We want to prove the following proposition: Proposition 7.8. Let A be a smooth compact Kähler surface with H ∗(A, Z) 2-torsion-free. The integral cohomology of A[2] can be expressed as follows: (i) H 1(A[2], Z) = π∗(b∗(H 1(A × A, Z))), (ii) H 2(A[2], Z) = π∗(b∗(H 2(A × A, Z))) ⊕ Z π∗(E) 2 , (iii) H 3(A[2], Z) = π∗(b∗(H 3(A × A, Z))) ⊕ 1 2 π∗j∗b∗ E(H 1(∆, Z)), (iv) The lattices π∗(b∗(H 4(A × A, Z))) and π∗j∗b∗ E(H 2(∆, Z)) are primitive in H 4(A[2], Z). Moreover: Q := H 4(A[2], Z) π∗(b∗(H 4(A × A, Z))) ⊕ π∗j∗b∗ E(H 2(∆, Z)) = (Z /2 Z)b2(A) and there exists an isometry θ of H 2(A, Z) such that the classes π∗j∗b∗ E (θ(x))+π∗b∗(x⊗x) 2 generate Q. (v) H 5(A[2], Z) = π∗(b∗(H 5(A × A, Z))) ⊕ π∗j∗b∗ E(H 3(∆, Z)), (vi) H 6(A[2], Z) = π∗(b∗(H 2(A × A, Z))) ⊕ π∗(f ), (vii) H 7(A[2], Z) = π∗(b∗(H 7(A × A, Z))). This section is dedicated to the proof of this proposition. Statements (iii) and (v) were proven in [13, Proposition 7.1]. Before proving the remaining statements, we will see that they are also interesting because they provide the ring structure of H ∗(A[2], Z). Lemma 7.9. (i) For all x1, x2, y1, y2 in H ∗(A, Z): (π∗b∗(x1 ⊗ y1)) · (π∗b∗(x2 ⊗ y2)) = π∗b∗ ((x1 · x2) ⊗ (y1 · y2) + (x1 · y2) ⊗ (y1 · x2)) . (ii) For all x, y, z in H ∗(A, Z): (iii) (cid:0) 1 2 π∗j∗(E)(cid:1)2 = 1 2(cid:16)π∗j∗b∗ E(z)(cid:17) = 2π∗j∗b∗ E(x · y · z). (π∗b∗(x ⊗ y)) ·(cid:16)π∗j∗b∗ E(c1(A)) − π∗b∗(∆)(cid:17). 30     /      / y  (iv) For all x, y in H ∗(A, Z): (cid:16)π∗j∗b∗ E(x)(cid:17) ·(cid:16)π∗j∗b∗ E(y)(cid:17) = 2(cid:16)π∗j∗b∗ E(x · y · c1(A)) − π∗b∗g∗(x · y)(cid:17) . (i) Using lemma 3.6 (ii), we have: Proof. (π∗b∗(x1 ⊗ y1)) · (π∗b∗(x2 ⊗ y2)) = = = 1 4 1 2 1 2 (π∗b∗(x1 ⊗ y1 + y1 ⊗ x1)) · (π∗b∗(x2 ⊗ y2 + y2 ⊗ x2)) π∗ ((b∗(x1 ⊗ y1 + y1 ⊗ x1)) · (b∗(x2 ⊗ y2 + y2 ⊗ x2))) π∗b∗ ((x1 ⊗ y1 + y1 ⊗ x1) · (x2 ⊗ y2 + y2 ⊗ x2)) = π∗b∗ ((x1 · x2) ⊗ (y1 · y2) + (x1 · y2) ⊗ (y1 · x2)) . (ii) Using Lemma 3.6 (ii), then the projection formula and finally the commutativity of diagram (38), we have: (π∗b∗(x ⊗ y)) ·(cid:16)π∗j∗b∗ E(z)(cid:17) = 1 2 (π∗b∗(x ⊗ y + y ⊗ x)) ·(cid:16)π∗j∗b∗ E(z)(cid:17) E(z)(cid:17)(cid:17) = π∗(cid:16)(b∗(x ⊗ y + y ⊗ x)) ·(cid:16)j∗b∗ E(z)(cid:17)(cid:17) = π∗j∗(cid:16)(j∗b∗(x ⊗ y + y ⊗ x)) ·(cid:16)b∗ E (g∗ (x ⊗ y + y ⊗ x) · z) . = π∗j∗b∗ By definition: x ⊗ y = p∗ write the h∗. So: 1(x) · p∗ 2(y). It follows: g∗(x ⊗ y) = h∗(x · y) and by convention, we do not (π∗b∗(x ⊗ y)) ·(cid:16)π∗j∗b∗ E(z)(cid:17) = 2π∗j∗b∗ E (x · y · z) . (iii) By [11, Proposition 6.7], we have: where F := b∗ E(N∆/A×A)/NE/ Bl∆(A×A). Moreover, j∗(NE/ Bl∆(A×A)) = j∗(E)2; it follows: b∗(∆) = j∗(c1(F )), b∗(∆) = j∗(b∗ E(c1(N∆/A×A))) − j∗(E)2. Furthermore, we remark that so: c1(N∆/A×A) = c1(cid:18) TA×A∆ T∆ (cid:19) = c1(T∆) = h∗c1(TA), j∗(E)2 = j∗(b∗ E(c1(A))) − b∗(∆). Then, the result follows from Lemma 3.6 (ii). (iv) By (ii) and then by (iii), (i): (cid:16)π∗j∗b∗ E(x)(cid:17) ·(cid:16)π∗j∗b∗ E(y)(cid:17) = 1 4 1 2 = Then, again by (ii): (cid:16)π∗j∗b∗ E(x)(cid:17) ·(cid:16)π∗j∗b∗ E(y)(cid:17) = (π∗b∗(x ⊗ 1)) · (π∗j∗(E))2 · (π∗b∗(y ⊗ 1)) (π∗b∗((x · y) ⊗ 1 + x ⊗ y)) ·(cid:16)π∗j∗(b∗ 2(cid:16)4π∗j∗b∗ E(c1(A))) − π∗b∗(∆)(cid:17) . E(x · y · c1(A)) − π∗b∗((x · y) ⊗ 1 + x ⊗ y) · π∗b∗(∆)(cid:17) . 1 (39) Now, we calculate π∗b∗((x · y) ⊗ 1 + x ⊗ y) · π∗b∗(∆) separately: π∗b∗((x · y) ⊗ 1 + x ⊗ y) · π∗b∗(∆) = 1 2 π∗b∗((x · y) ⊗ 1 + 1 ⊗ (x · y) + x ⊗ y + y ⊗ x) · π∗b∗(∆). 31 By Lemma 3.6 (ii), then by projection formula and finally by the definition x ⊗ y = p∗ we have: 1(x) · p∗ 2(y), π∗b∗((x · y) ⊗ 1 + x ⊗ y) · π∗b∗(∆) = π∗b∗ ([(x · y) ⊗ 1 + 1 ⊗ (x · y) + x ⊗ y + y ⊗ x] · ∆) = π∗b∗g∗g∗ ((x · y) ⊗ 1 + 1 ⊗ (x · y) + x ⊗ y + y ⊗ x) = 4π∗b∗g∗(x · y). So, by (39): (cid:16)π∗j∗b∗ E(x)(cid:17) ·(cid:16)π∗j∗b∗ E(y)(cid:17) = 1 2(cid:16)4π∗j∗b∗ E(x · y · c1(A)) − 4π∗b∗g∗(x · y)(cid:17) . Now we start the proof of Proposition 7.8, we first prove (i) and (vii). By Theorem 5.4, we have: H k(Bl∆(A × A), Z) = b∗(H k(A × A, Z)), for k = 1 and k = 7. It follows from Künneth formula that ℓ1 3.10, the coefficient of surjectivity α1(Bl∆(A × A)) = α7(Bl∆(A × A)) = 0. It follows (i) and (vii). +(Bl∆(A× A)) = 0. Hence, from Proposition Now let us prove (ii) and (vi). By Theorem 5.4, we have: H 2(Bl∆(A × A), Z) = b∗(H 2(A × A, Z)) ⊕ j∗b∗ E(H 0(∆, Z)). (40) We have j∗b∗ from Proposition 3.10: E(H 0(∆, Z)) = Z E. From Künneth formula, we remark that ℓ2 +(Bl∆(A × A)) = 1. Hence, α2(Bl∆(A × A)) + α6(Bl∆(A × A)) = 1. Moreover, we have π∗(OBl∆(A×A)) = OA[2] ⊕ L, with L2 = OA[2] (π∗(E)). So, as it is well known π∗(E) is divisible by 2. Hence, α2(Bl∆(A × A)) = 1 and α6(Bl∆(A × A)) = 0. So, (ii) follows from (40). Similarly, by Theorem 5.4, we have: H 6(Bl∆(A × A), Z) = b∗(H 6(A × A, Z)) ⊕ j∗b∗ E(H 4(∆, Z)), with j∗b∗ E(H 4(∆, Z)) = Z f . So, we obtain (vi) because α6(Bl∆(A × A)) = 0. It remains to study the most interesting part which is H 4(A[2], Z). Again by Theorem 5.4, we have: H 4(Bl∆(A × A), Z) = b∗(H 4(A × A, Z)) ⊕ j∗b∗ E(H 2(∆, Z)). (41) + (A × A) = 0 for all k ∈ {0, 1, 2, 3}, ℓ4 First, we are going to use Remark 6.17 to show that π∗(b∗(H 4(A × A, Z))) is primitive in H 4(A[2], Z). To do so, we verify the hypothesis of Theorem 6.12 for the couple (A × A, S2). By Künneth formula, we have ℓ2k+1 +(A × A) = 0. It follows condition (ii) of Theorem 6.12. Moreover, ∆ ∈ H 4(A × A, Z) is primitive because ∆ · A ⊗ pt = 1. So, by Theorem 6.12, (A × A, S2) is H 4-normal and by Remark 6.17 the lattice π∗(b∗(H 4(A × A, Z))) is primitive in H 4(A[2], Z). Moreover, by Corollary 3.8, log2 discr π∗(b∗(H 4(A × A, Z))) = b2(A) and by Lemma 3.6 (ii), log2 discr(π∗j∗b∗ +(A × A) = b2(A) and ℓ2 E(H 2(∆, Z))) = b2(A). So: discr π∗(b∗(H 4(A × A, Z))) = discr π∗j∗b∗ E(H 2(∆, Z)). (42) Since A[2] is smooth, we know by (4) that discr π∗(b∗(H 4(A × A, Z))) = discr π∗(b∗(H 4(A × A, Z)))⊥. E(H 2(∆, Z)) = π∗(b∗(H 4(A×A, Z)))⊥. Since π∗j∗b∗ So, π∗j∗b∗ E(H 2(∆, Z)) ⊂ π∗(b∗(H 4(A×A, Z)))⊥, by (42), π∗j∗b∗ E(H 2(∆, Z)) is also primitive in H 4(A[2], Z). Now, we are going to calculate H 4(A[2], Z) π∗(b∗(H 4(A × A, Z))) ⊕ π∗j∗b∗ E(H 2(∆, Z)) . 32 For a lattice L, we denote by AL := L∨/L its discriminant group and for x ∈ L∨, we denote by x its reduction in AL. Let denote N := π∗(b∗(H 4(A × A, Z))) and N ⊥ = π∗j∗b∗ E(H 2(∆, Z)). By Lemma 3.6 (ii), 1 2 N ⊥ = (N ⊥)∨. It follows that: AN ⊥ ≃ 1 2 N ⊥ N ⊥ = (Z /2 Z)b2(A). (43) Since H 4(A[2], Z) is unimodular, by (6), we have an isometry between the discriminant groups: ν : AN ≃ AN ⊥ . Again by Lemma 3.6 (ii), 1 classes 1 2 π∗b∗(x ⊗ x) ∈ N ∨, for all x ∈ H 2(A, Z). So, AN is generated by the 2 π∗b∗(x ⊗ x), where x ∈ H 2(A, Z). Let(cid:8) z(x) ∈ N ⊥(cid:12)(cid:12) x ∈ H 2(A, Z)(cid:9) be a basis of N ⊥ such that (cid:17) = z(x) 2 . Since ν is an isometry, we have: ν(cid:16) π∗b∗(x⊗x) 2 π∗b∗(x ⊗ x) π∗b∗(y ⊗ y) z(x) z(y) mod Z . (44) · 2 = 2 · 2 However, by Lemma 7.9 (i): 2 And by Lemma 7.9 (iv): π∗b∗(x ⊗ x) · π∗b∗(y ⊗ y) = 2(x · y)2. E(x) · π∗j∗b∗ Since (x · y) = (x · y)2 mod 2, it follows from (44) that: π∗j∗b∗ E(y) = 2x · y. z(x) 2 · z(y) 2 = π∗j∗b∗ E(x) 2 · π∗j∗b∗ E(y) 2 mod Z . So, we have constructed an isometry of AN ⊥ sending z(x) . However, [23, Theorem 3.6.3] 2 says that the natural map O(N ⊥) → O(AN ⊥ ) is surjective when N ⊥ is an indefinite lattice which is always the case apart when A is the projective space; but in this cases , rk N ⊥ = 1 and the natural map O(N ⊥) → O(AN ⊥ ) is obviously surjective. to π∗j∗b∗ E (x) 2 It follows that there is an isometry eθ of N ⊥ such that: Then, eθ provides an isometry θ on H 2(A, Z) such that: eθ(π∗j∗b∗ π∗j∗b∗ E(θ(x)) = z(x) mod 2N ⊥. E(x)) = z(x) mod 2N ⊥. Then, by definition of ν, the elements π∗j∗b∗ E(θ(x)) + π∗b∗(x ⊗ x) are divisible by 2 for all x ∈ H 2(A, Z). 7.4 Symplectic groups and Beauville -- Bogomolov forms We will study the Beauville -- Bogomolov form of some quotients of an irreducible symplectic 2n-fold X by a symplectic group G of prime order. So, M := X/G will be a singular primitively symplectic orbifold. One of our most important ingredients will be the Fujiki formula generalized by Matsushita in [16] (see [18, Section 1.2.2] for an overview). Assume that Codim Sing M ≥ 4, then if Y = M or X, we have for all α ∈ H 2(Y, Z): α2n = cY BY (α, α)n, (45) where cY is the Fujiki constant. Moreover, if ω 6= 0 is the holomorphic 2-form of Y , the convention is: BY (ω + ω, ω + ω) > 0. (46) Furthermore, the convention is to choose the Beauville -- Bogomolov form to be integral and indivisible. We know the Beauville -- Bogomolov form BX of X. Then, we will deduce the Beauville -- Bogomolov form BM of M from BX and H 2(M, Z). The link between BX and BM is the matter of the following proposition. 33 Proposition 7.10. Let X be an irreducible symplectic manifold of dimension 2n and G be a symplectic group of prime order p with Codim Fix G ≥ 4. Then, BM (π∗(α), π∗(β)) = ns cX p2n−1 cM BX (α, β), where cM is the Fujiki constant of M and α, β are in H 2(X, Z)G. Proof. By (45), we have: and: Moreover, by Lemma 3.6 (ii): So, by (47), (48), (49), we have: (π∗(α))2n = cM BM (π∗(α), π∗(α))n, α2n = cX BX (α, α)n. (π∗(α))2n = p2n−1α2n. BM (π∗(α), π∗(α))n = cX p2n−1BX (α, α)n cM . (47) (48) (49) By (46), we get the result. When we assume that X is a manifold of K3[2]-type, from [3, Section 9], we know that the Fujiki constant of X is 3. So for α, β in H 2(X, Z)G: BM (π∗(α), π∗(β)) =s 3p3 cM BX (α, β). (50) We need a variation of Lemma 2.11 in the case of the lattice of H 2(X, Z) endowed with the Beauville -- Bogomolov form. Lemma 7.11. Let X be an irreducible symplectic manifold of K3[2]-type and G = hgi be a symplectic automorphism group of prime order p ≥ 3. We have: (i) AH2(X,Z)G = (Z /2 Z) ⊕ (Z /p Z)ℓ2 p(X), (ii) discr H 2(X, Z)G = 2pℓ2 p(X). (iii) We denote AH2(X,Z)G,p := (Z /p Z)ℓ2 p(X). Then, the projection H 2(X, Z) H 2(X, Z)G ⊕ ker σ → AH2(X,Z)G,p is an isomorphism, where σ = id +g + ... + gp−1. (iv) Moreover, let x ∈ H 2(X, Z)G. We have x p ∈ (H 2(X, Z)G)∨ if and only if there is z ∈ H 2(X, Z) such that x = z + g(z) + ... + gp−1(z). (v) Also: π∗(H 2(X, Z)) π∗(H 2(X, Z)G) = (Z /p Z)ℓ2 p(X). Proof. We recall that p ≤ 11 by [21]. Thus, the first three assumptions follow from [5, Lemma 6.5] and its proof (the number aG(X) in [5, Lemma 6.5] equal ℓ2 p(X) from [5, Corollary 5.8]). Then, the proof of (iv) is identical to the proof of Lemma 2.11. It remains to prove (v). Let A :=(cid:26) z + ... + gp−1(z) p ∈ (H 2(X, Z)G)∨(cid:12)(cid:12)(cid:12)(cid:12) z ∈ H 2(X, Z)(cid:27) . 34 By (iv): So: A =(cid:26) x p The following morphism ∈ (H 2(X, Z)G)∨(cid:12)(cid:12)(cid:12)(cid:12) x ∈ H 2(X, Z)G(cid:27) . A H 2(X, Z)G = AH2(X,Z)G,p. f : A z+...+gp−1(z) p / π∗(H2(X,Z)) π∗(H2(X,Z)G) / π∗(z) is surjective by definition. Quotiented A by H 2(X, Z)G, f provides an isomorphism between AH2(X,Z)G,p and π∗(H2(X,Z)) π∗(H2(X,Z)G) . Then, the result follows from (iii). 7.5 Quotient of a K3[2]-type manifold by an automorphism of order 11 As mentioned in the introduction, there are two different examples of symplectic automorphisms of order 11 on a manifold of K3[2]-type (Example 4.5.1 and Example 4.5.2 in [19]). In both examples, the fixed locus is a set of 5 isolated points. But as it is explained in [19, Section 7.4.4], the lattice H 2(X, Z)G will Moreover, it is proved in [19, Section 7.4.4] that these are the only possible invariant lattices for p = 11. So, let X be a manifold of K3[2]-type and G a symplectic automorphism group of X of order 11. If be different in these cases. In the first case the lattice is H 2(X, Z)G is isomorphic to is isomorphic to 0 22, we will denote the quotient X/G by M 2 8 , we will denote the quotient X/G by M 1 8 , and in the second 11. Hence, we obtain the following 0 0 22. 2 2 8 −3 6 2 2 −3 2 2 8 −3 6 2 2 −3 11, and if H 2(X, Z)G 2 1 1 6 0 0 1 6 0 0 2 1 0 corollary. Corollary 7.12. Let X be a manifold of K3[2]-type. Let G be a symplectic automorphism group of X of order 11. Then, (X, G) is H 2-normal and H 4-normal. Proof. In the both cases, we have discr H 2(X, Z)G = 2 × 112. Hence, by Lemma 7.11 (ii): ℓ2 11(X) = 2. (51) It follows from Corollary 2.7 that ℓ2 +(X) = 1 and ℓ2 We recall from Section 2.2, that when p ≥ 3, we have ℓk Hence, by (37) and [5, Lemma 6.14 ]: −(X) = 0. +(X) = ℓk 1(X) and ℓk −(X) = ℓk p−1(X) for all k. So, it follows from Remark 3.11 that: ℓ4 +(X) = 1. α4(X) ≤ 1 2 . Since α4(X) is an integer, we have α4(X) = 0. So, (X, G) is H 4-normal. Therefore, the H 2-normality follows from Corollary 7.5. The end of the section is dedicated to the proof of Theorem 1.2. Assume H 2(X, Z)G ≃ 2 2 8 −3 6 2 2 −3 8  . 35 / / We denote by {a, b, c} an integral basis of H 2(X, Z)G with BX (a, a) = 6, BX (b, b) = 8, BX (c, c) = 8, BX (a, b) = BX (a, c) = 2 and BX (b, c) = −3. Let B =(cid:26) π∗(b) − π∗(c) 11 , 4π∗(a) − π∗(b) 11 , π∗(a) − 3π∗(c) 11 (cid:27) . We show that B is a basis of H 2(M 1 11, Z). By Lemma 7.11 (iv), B ⊂ π∗(H 2(X, Z)). Moreover, we have hBi π∗(H 2(X, Z)G) = (Z /11 Z)2. Hence, by Lemma 7.11 (v) and (51): It follows from Corollary 7.12 that B is a basis of H 2(M 1 The matrix of the sublattice of H 2(X, Z)G generated by b − c, 4a − b and a − 3c is hBi = π∗(H 2(X, Z)). 11, Z). Then, by (50), the Beauville -- Bogomolov form on H 2(M 1 11, Z) gives the lattice Since the Beauville -- Bogomolov form is integral and indivisible, it follows that cM 1 lattice 11 = 33, and we get the 3 × 11 8 × 11 −11 6 × 11 . 6  . 3 8 −1 2 −1 −1 3 −1 11 1 cM 1  2 × 11 −11 −11 3 × 11 −11  11s 3 · 113  H 2(X, Z)G ≃ B =(cid:26) π∗(a) − 2π∗(b) 2 −1 −1 3 −1 11 Now assume that The proof is identical taking the basis 3 8 −1 6  . 0 22 . 0 0 1 6 2 1 0 6π∗(a) − π∗(b) , 11 , π∗(c) 11 (cid:27) , where {a, b, c} is an integral basis of H 2(X, Z)G with BX (a, a) = 2, BX (b, b) = 6, BX (c, c) = 22, BX (b, c) = BX (a, c) = 0 and BX (a, b) = 1. 7.6 Quotient of a K3[2]-type manifold by a symplectic automorphisms of or- der 3 As mentioned in the introduction, in [19, Theorem 7.2.7], Mongardi distinguishes two kinds of symplectic automorphism of order 3 on a K3[2]-type manifold one with 27 isolated fixed points and another with a fixed abelian surface. We recall from [19, Example 4.2.1]: Remark 7.13. Let S be a K3 surface. A natural symplectic automorphism of order 3 on S[2] has 27 isolated fixed points. Moreover, Mongardi in [20, Theorem 2.5] shows the following result: Proposition 7.14. Let X be a K3[2]-type manifold and ϕ be a symplectic automorphism of order 3 with 27 isolated fixed points. Then, (X, ϕ) can be deformed to a couple (S[2], φ[2]), where S is a K3 surface and φ[2] is the automorphism induced on S[2] by a symplectic automorphism φ of order 3 on S. 36 In this section, we are considering these symplectic automorphisms ϕ of order 3, with 27 isolated points, on a K3[2]-type manifold X; and we denote the quotient X/ϕ by M3. Corollary 7.15. Let X be a manifold of K3[2]-type. Let G be a symplectic automorphisms group of order 3 of X with 27 isolated fixed points. Then, (X, G) is H 2-normal and H 4-normal. Proof. By Corollary 7.5, if (X, G) is H 4-normal then (X, G) is H 2-normal. Hence, we have only to show the H 4-normality. The action of G is symplectic on X, in this case, the only possible local action around the fixed points is given by: diag(ξ3, ξ3, ξ2 3, ξ2 3 ), where ξ3 is a 3-root of the unity. So, unfortunately, the fixed points of G are not simple. However, since G is of order 3, we can get around this difficulty using Proposition 5.10. Let X1 be the blow-up of X in Fix G and G1 the automorphisms group induced from G. From Lemma 5.6 the fixed points of G1 are simple. Hence, it will be possible to apply Theorem 6.1 to (X1, G1). First, using the same method as in the proof of Lemma 5.6, we remark that for each isolated fixed point of G, we get 2 rational fixed lines of G1. So, Fix G1 consists in 54 rational lines. Moreover, from [14], H ∗(X, Z) is torsion-free, then from Theorem 5.4, H ∗(X1, Z) is also torsion-free. Hence, it only remains to verify the condition (ii) of Theorem 6.1. From Theorem 5.4: H i(X1, Z) ≃ H i(X, Z) ⊕ H 0(pt, Z)27⊕ and H j(X1, Z) ≃ H j(X, Z), for i ∈ {2, 4, 6} and j ∈ {0, 1, 3, 5, 7, 8}. Since the exceptional divisors are fixed by the action of G1, it follows that: ℓi +(X1) = ℓi +(X) + 27, (52) for all i ∈ {2, 4, 6}. The others ℓj Proposition 6.2: ∗ are the same for X and X1. Moreover, from [5, Theorem 1.1] and h∗(Fix G, Q) = X1≤q<p ℓ∗ q(X). Since Fix G consists in 27 isolated points, we get: By (52): X1≤q<p 27 = X1≤q<p q(X1) = X1≤q<p ℓ∗ ℓ∗ q(X). ℓ∗ q(X) + 3 × 27. Since Fix G1 consists in 54 rational curves, we have: h∗(Fix G1, Q) = 2 × 54 = 4 × 27. Finally, from (53), (54) and (55), we get: h∗(Fix G1, Q) = X1≤q<p ℓ∗ q(X1). (53) (54) (55) It follows from Theorem 6.1 that (X1, G1) is H 4-normal. We conclude with Proposition 3.17. The end of the section is dedicated to the proof of Theorem 1.3. From [12, Theorem 4.1] and Proposition 7.14, there is an isometry of lattices H 2(X, Z)G ≃ U ⊕U (3)⊕ U (3) ⊕ A2 ⊕ A2 ⊕ (−2). In the rest of the proof, we identify H 2(X, Z)G with the lattice U ⊕ U (3) ⊕ U (3) ⊕ A2 ⊕ A2 ⊕ (−2) By Lemma 7.11 (iv), we have: 1 3 π∗(U (3)) ⊂ H 2(M3, Z). 37 (56) Let (a, b) an integral basis of A2, with BX (a, a) = BX (b, b) = −2 and BX (a, b) = 1. Idem, by Lemma 7.11 (iv), we have: π∗(a) − π∗(b) 3 ∈ H 2(M3, Z). The lattice generated by a − b and a + 2b is isomorphic to A2(3) =(cid:18)−6 3 −6(cid:19). So, we denote A2(3) := 3 ha − b, a + 2bi. We have: 1 3 π∗(A2(3)) ∈ H 2(M3, Z). (57) Then, by (56) and (57): Therefore, by (ii) and (v) of Lemma 7.11, we obtain: So by Corollary 7.15, 1 3 1 3 π∗(H 2(X, Z)) ⊃ π∗(cid:18)U ⊕ π∗(H 2(X, Z)) = π∗(cid:18)U ⊕ H 2(M3, Z) = π∗(cid:18)U ⊕ U s 81 U 2 3s 81 cM3! ⊕ cM3! ⊕ cM3! ⊕ U 2 s 9 = U 3s 9 cM3! ⊕ A2 1 3 1 3 1 3 A2 U (3)2 ⊕ U (3)2 ⊕ 1 3 U (3)2 ⊕ 1 3 1 3 A2(3)2 ⊕ (−2)(cid:19) . A2(3)2 ⊕ (−2)(cid:19) . A2(3)2 ⊕ (−2)(cid:19) . cM3! ⊕ −2s 81 cM3! cM3! ⊕ −6s 9 cM3! . 2 3s 81 2 s 9 Then, by (50), the Beauville -- Bogomolov form of H 2(M3, Z) gives the lattice Then knowing that the Beauville -- Bogomolov form is integral and indivisible, we have cM3 = 9 and we get the lattice U (3) ⊕ U 2 ⊕ A2 2 ⊕ (−6). References [1] M.A. Aguilar, C. Prieto, Transfers for ramified covering maps in homology and cohomology. Int. J. Math. Sci., (2006), Art. ID 94651, 28pp. [2] M.F. Atiyah, R. Bott, The moment map and equivariant cohomology. Topology, Vol. 23, (1984), no. 1, 1-28 [3] A. Beauville, Variétés kähleriennes dont la première classe de Chern est nulle, J. Differential geometry, 18 (1983) 755-782. [4] S. Boissière, C. Camere and A. Sarti, Classification of automorphisms on a deformation family of hyperkähler fourfolds by p-elementary lattices. Kyoto J. Math. 56 (2016), no.3, 465-499. [5] S. Boissière, M. Nieper-Wisskirchen, A. Sarti Smith theory and irreducible holomorphic symplectic manifolds, J. Topo. 6 (2013), no. 2, 361-390. [6] H. Cartan, Quotient d'une variété analytique par un groupe discret d'automorphismes, Séminaire Henri Cartan, tome 6, (1953-1954), exp. no 12, p.1-13. [7] C.W. Curtis, I. Reiner Representation theory of finite groups and associative algebras, Wiley Classics Library, John Wiley & Sons Inc., New York, 1988 Reprint of the 1962 original, A Wiley-Interscience Publication. MR 1013113 (90g:16001) 38 [8] P. Deligne, Théorème de Lefschetz et critères de dégénérescence de suites spectrales, Inst. Hautes Etudes Si. Publ. Math. (1968), no. 35, 259-278. [9] I.V. Dolgachev, Classical Algebraic Geometry, Cambridge University Press (2012). [10] A. Fujiki On Primitively Symplectic Compact Kähler V-manifolds of Dimension Four. Classification of algebraic and analytic manifolds (Katata, 1982), 71-250. [11] W. Fulton, Intersection theory, Second edition, Springer. [12] A. Garbagnati, A. Sarti, Symplectic automorphisms of prime order on K3 surfaces. J. Algebra 318 2007, no. 1, 323-350. [13] S. Kapfer, G. Menet, Integral cohomology of the generalized Kummer fourfold, arXiv:1607.03431. [14] E. Markman, Integral generatorsfor the cohomology ring of moduli spaces of sheaves over Poisson surfaces, Adv. Math. 208 (2007), no. 2, 622-646. [15] J.M. Masley, H.L. Montgomery, Cyclotomic fields with unique factorization, J. Reine Angew. Math. 286/287 (1976), 248-256. [16] D. Matsushita, On base manifolds of Lagrangian fibrations, Science China Mathematics (2014), 531-542. [17] G. Menet, Beauville -- Bogomolov lattice for a singular symplectic variety of dimension 4, Journal of pure and apply algebra (2014), 1455-1495. [18] G. Menet Cohomologie entière et fibration lagrangiennes sur certaines variétés holomorphiquement symplectiques singulières, PhD thesis of Lille 1 University, (2014). [19] G. Mongardi, Automorphisms of hyperkähler manifolds, PhD Thesis, University of Rome 3 (2013). [20] G. Mongardi, On natural deformations of symplectic automorphisms of manifolds of K3[n] type, C. R. Math. Acad. Sci. Paris315 (2013), no. 13-14, 561-564. [21] G. Mongardi, On symplectic automorphisms of hyperkähler fourfolds of K3[2] type, Michigan Math. J. 62 (2013), no. 3, 537-550. [22] Y. Namikawa, Extension of 2-forms and symplectic varieties, J. Reine Angew. Math. 539 (2001), 123-147. [23] V.V. Nikulin Integral symmetric bilinear forms and some of their applications, Math. USSR Izv. 14 (1980), 103-167. [24] D. Prill, Local classification of quotients of complex manifolds by discontinuous groups, Duck Math. J. 34 1967 375-386. [25] Z. Qin and W. Wang, Integral operators and integral cohomology classes of Hilbert schemes, Math. Ann. 331 (2005), no. 3, 669 -- 692. [26] B. Totaro, The integral cohomology of the Hilbert scheme of two points, Forum Math. Sigma 4 (2016). [27] L. Smith, Transfer and ramified coverings, Math. Proc. Camb. Phil. Soc. (1983), 93, 485-493. [28] M. Verbitsky, Cohomology of compact hyper-Kähler manifolds and its applications, Geom. Funct. Anal. 6 (1996), no. 4, 601-611. [29] C. Voisin, Hodge Theory and Complex Algebraic Geometry. I,II, Cambridge Stud. Adv. Math., 76, 77, Cambridge Univ. Press, 2003. Grégoire Menet IMB, Université de Dijon 9 avenue Alain Savary, 21000 DIJON (France) [email protected] 39
1812.09574
2
1812
2019-05-09T02:36:42
The homotopy Leray spectral sequence
[ "math.AG", "math.AT", "math.KT" ]
In this work, we build a spectral sequence in motivic homotopy that is analogous to both the Serre spectral sequence in algebraic topology and the Leray spectral sequence in algebraic geometry. Here, we focus on laying the foundations necessary to build the spectral sequence and give a convenient description of its $E_2$-page. Our description of the $E_2$-page is in terms of homology of the local system of fibers, which is given using a theory similar to Rost's cycle modules. We close by providing some sample applications of the spectral sequence and some hints at future work.
math.AG
math
THE HOMOTOPY LERAY SPECTRAL SEQUENCE ARAVIND ASOK, FR´ED´ERIC D´EGLISE, JAN NAGEL Abstract. In this work, we build a spectral sequence in motivic homotopy that is analogous to both the Serre spectral sequence in algebraic topology and the Leray spectral sequence in algebraic geometry. Here, we focus on laying the foundations necessary to build the spectral sequence and give a convenient description of its E2-page. Our description of the E2-page is in terms of homology of the local system of fibers, which is given using a theory similar to Rost's cycle modules. We close by providing some sample applications of the spectral sequence and some hints at future work. Contents Introduction 1. Notations and conventions 2. Homotopy t-structure and duality 2.1. Recollections on the homotopy t-structure 2.2. Recollection on purity and duality 3. Fiber homology and Gersten complexes 3.1. Fiber δ-homology 3.2. Gersten complexes 3.3. Products 4. The homotopy Leray spectral sequences 4.1. The homological version 4.2. The cohomological version 4.3. Remarkable properties of homotopy modules 5. Applications 5.1. Morphisms with A1-contractible fibers 5.2. Gysin and Wang sequences 5.3. Relative cellular spaces References 1 4 6 6 9 10 10 13 16 16 17 20 24 26 26 28 30 33 1. Introduction The goal of this paper is to study an algebro-geometric version of the Leray -- Serre spectral sequence for generalized cohomology theories. To explain the setup, suppose one is given a topological space X, a sheaf F of abelian groups on X and a filtration X• = X0 ⊂ . . . ⊂ Xp ⊂ Xp+1 ⊂ . . . ⊂ Xn = X by closed subsets. One naturally associates an exact couple with the preceding data and obtains a spectral sequence of the form: Ep,q 1 (X•, F ) = H p+q(Xp \ Xp−1, F ) ⇒ H p+q(X, F ). If X• is cellular with respect to F , i.e., if H i(Xp \Xp−1, F ) = 0 for all i 6= p, then there are isomorphisms of the form H k(X, F ) ∼= Ek,0 2 (X•, F ). If f : X → B is a continuous map of topological spaces and B• is a filtration of B that is cellular with respect to the direct image sheaves Rqf∗F for all q, then the E2 -- term of Leray spectral sequence Ep,q 2 = H p(B, Rqf∗F ) ⇒ H p+q(X, F ) is isomorphic to Ep,q 2 (X•, F ), where X• is the inverse image of B•. Date: December 2018. 1 2 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL On the other hand, suppose F −→ X f −→ B is a Serre fibration of topological spaces, where B has the homotopy type of a (connected) finite CW complex, and E is a (generalized) cohomology theory in the sense of classical stable homotopy theory. One may consider an associated Atiyah -- Hirzebruch spectral sequence (see, e.g., [DK01, §9.2-9.5]): the E2-page of this spectral sequence is given in terms of the ordinary (co)homology of B with coefficients in local systems attached to the E-(co)homology of F and converges to the E-(co)homology of X. When E is ordinary cohomology and F is the constant sheaf ZX , the two spectral sequences coincide. Indeed, Serre showed that the direct image sheaves Rqf∗ZX are local systems on B depending only on the cohomology of the fiber F and the action of the fundamental group of B on these cohomology groups. To explain our algebro -- geometric analog, recall that Arapura showed the Leray spectral sequence is "motivic" [Ara05]. In more detail, suppose f : X → B is a projective morphism of complex quasi -- projective algebraic varieties. After reducing to a situation where B is affine (via the Jouanolou trick), work of Beilinson and Nori [Nor02] shows that any constructible sheaf on B can be made cellular with respect to a filtration by closed algebraic subsets. In this situation, Arapura compares the Leray spectral sequence Ea,b 2 (f ) = H a(B, Rqf∗ZX ) ⇒ H a+b(X, ZX ) and the spectral sequence associated with the skeletal filtration and uses this to show that the Leray spectral sequence essentially reflects suitably functorial algebro-geometric structure present on cohomol- ogy groups (e.g., it lifts to Nori's category of mixed motives). A similar result holds for the perverse Leray spectral sequence Ea,b 2 = H a(B, pRbf∗ZX ) ⇒ H a+b(X, ZX ) that is obtained by replacing the classical truncation functor τ by the perverse truncation pτ [dCM10]. In this paper, we consider a variant of this setup. Let us now consider a simple case to highlight our considerations. Suppose k is a field, B is a finite dimensional irreducible smooth k-variety, and f : X → B is a smooth morphism of k-varieties. As usual, f may be thought of as an ´etale locally trivial fibration in algebraic geometry. We write B(n) for the set of codimension n points of B. If B has dimension d, we view the collection B(i) as i ranges from 0 through d as an algebro-geometric analog of the skeletal filtration of B. In this situation, the scheme-theoretic fiber Xb for b ∈ B(i) is then a smooth variety over b. There is a comparison theorem between the Leray spectral sequence and the spectral sequence as- If H ∗(X, A) is a suitable cohomology theory defined using a sociated to the Bloch-Ogus complex. Grothendieck topology that is finer than the Zariski topology (e.g., ´etale cohomology, or Betti coho- mology over C), one can consider the Leray spectral sequence associated with the morphism of sites π : Xfine → XZar. The higher direct image sheaves Rqπ∗A are the Zariski sheaves Hq associated with the presheaves U 7→ H q(π−1(U ), A). Bloch and Ogus show that the Leray spectral sequence Ep,q 2 = H p(X, Hq) ⇒ H p+q(X, A) can be identified from the E2 -- term onward with the spectral se- quence Ep,q 1 = ⊕x∈X (p)H q−p(k(x)) ⇒ N •H p+q(X) that gives the coniveau filtration on cohomology. (The argument uses Deligne's technique of "d´ecalage", see [Par96].) Furthermore, there are variants of the previous constructions where one replaces coho- mology by Borel -- Moore homology and works with the niveau (rather than coniveau) filtration. More generally, one can work with the bivariant theory H i(X f → Y ) = Hom(Rf!QX , QY [i]) that simultaneously generalizes both Borel -- Moore homology and cohomology. Building on this analysis, by analyzing formal properties of Gersten complexes, Rost developed a notion of "local coefficient system" on a k-scheme B, which he called a cycle module. Given a morphism f : X → B and a cycle module M on X, Rost defined Chow groups with coefficients in M , A∗(X, M ). For example, there is a cycle module built out of Milnor K-theory for which the associated Chow groups with coefficients coincide with usual Chow groups. Finally, Rost constructed a spectral sequence of the form: here Aq(X, M ) is a cycle module on B obtained by taking homology of the fibers. E2 p,q = Ap(B, Aq(X, M )) ⇒ Ap+q(X, M ); THE HOMOTOPY LERAY SPECTRAL SEQUENCE 3 Our approach in motivic homotopy theory essentially mixes all the ideas above and combines the classical Leray spectral sequence of a fibration with Rost's spectral sequence. The jumping off point is the work of the second author exploring the close relationship between the notion of cycle module in the sense of Rost and the heart of a t-structure on the motivic stable homotopy category. We use a relative version of Morel's homotopy t-structure, the so-called perverse homotopy t-structure, defined initially by Ayoub [Ayo07] and developed further in [BD17]. The approach to constructing this t-structure in [BD17] is rather flexible and works in great generality: the basic definition requires only the existence of a dimension function δ on the base scheme; for this reason, the t-structure was called the δ-homotopy t-structure in op. cit.. The heart of the δ-homotopy t-structure consists of δ-homotopy modules, or simply homotopy mod- ules. These objects are closely related to cycle modules and their homology (with respect to the trun- cation functors in the δ-homotopy t-structure) can be computed by a Gersten-type complex that is formally extremely similar to that written down by Rost (see Definition 3.2.4 for more details). The approach we take via t-structures has technical advantages: functoriality properties and multiplicative structure are easy to obtain (see Proposition 3.2.7 and Proposition 3.3.2 for more details). With these tools in hand, we state a version of our main result here (the notation and terminology in the statement that we have not yet mentioned may be found in the notations and conventions section below). Theorem 1 (see Theorem 4.1.2). Fix a base scheme S with a dimension function δ, and suppose f : X → B is a separated morphism of S-schemes having finite type. Let E be a motivic spectrum (or a mixed motive) over S, and put: E! X := f !E. Then there exists a convergent spectral sequence of the form: abutting to the E-homology of X relative to S, or rather, the bivariant theory of X/S with coefficients in E (see below our Notations and conventions). In fact, E2 p,q E2 p,q(f, E) = Aδ q (f∗E! p(cid:0)B, H δ X ))(cid:1) ⇒ Ep+q(X/S), i) vanishes outside a range of columns bounded by the maximum and minimum of the dimension function δ on B; and ii) is described as the homology of a Gersten type complex with coefficients in the homotopy module H δ q (f∗E! X )). The spectral sequence above is obtained by filtering f∗(E! X ) by truncations with respect to the δ- homotopy t-structure. It is closely related to the spectral sequence one would obtain by filtering E∗(−/S) via the pullback of the (δ-)niveau filtration on B along f (for the dimension function δ). The following result gives a precise formulation. Proposition 2 (See Proposition 4.1.9). Let δF f En(X/S), p ∈ Z, be the abuting filtration of the homo- p topy Leray spectral sequence. Define the δ-niveau filtration on E∗(X/S) relative to f by the following formula: δN f p E∗(X/S) = [i:Z→X,δ(Z)≤p Im(cid:0)i∗ : E∗(X ×B Z/S) → E∗(X/S)(cid:1). For any pair of integer (p, n) ∈ Z2, the following relation holds: δF f p En(X/S) = δN f n−p En(X/S). This proposition may be viewed as an analogue of Washnitzer's conjecture (that the coniveau filtration on de Rham cohomology coincides with the filtration arising from the second hypercohomology spectral sequence), which was established by Bloch and Ogus [BO74]. There is also a cohomological form of the homotopy Leray spectral sequence, which is better suited to analysis of product structures. We Theorem 3 (See Theorem 4.2.5). Fix a base scheme S with a dimension function δ, and suppose f : X → B is a separated morphism of S-schemes having finite type. If E is a motivic ring spectrum over X, then there is a convergent spectral sequence of the form Ep,q 2 (f, E) = Ap δ (B, H q δ (f∗EX )) ⇒ H p+q(X, E), where we write H ∗(X, E) for the groups usually denoted E∗(X) to emphasize the link with Leray spectral sequence. As before the E2-page is a) concentrated in a range of columns bounded by the maximum and minimum of the dimension function δ on B, and b) described in terms of (co)homology of the Gersten-style complex associated with homotopy module H q δ (f∗EX ). Moreover, i) each page Er has the structure of a differential graded algebra, and 4 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL ii) the induced product structure on the abutment coincides with the E-cohomology product structure. While the reader may consult the beginning of each section for further discussion of its contents, we close the introduction with a brief overview of the paper. Section 2 is devoted to reviewing the basic properties of this homotopy t -- structure. Section 3 then studies homotopy modules, Gersten complexes and the associated niveau spectral sequences. In particular, we describe multiplicative structure on the niveau spectral sequences when considering cohomology theories equipped with suitable multiplicative structure. Section 4 is the theoretical heart of the paper. We build the homotopy Leray spectral sequence (see, e.g., Theorem 4.1.2) and study an analog of "locally constant sheaves" in our context. Finally, Section 5 contains some sample applications of these spectral sequences. In particular, we analyze fibrations with A1-contractible fibers, construct Gysin and Wang sequences and, in the motivic case, prove a degeneration result for the homotopy Leray spectral sequence for relative cellular spaces. Acknowledgements. The authors want to thank Adrien Dubouloz for fruitful discussions about fibrations, and, especially, Fabien Morel for providing the initial impetus to pursue this work. Aravind Asok was partially supported by National Science Foundation Awards DMS-1254892 and DMS-1802060. F. D´eglise and J. Nagel received support from the French "Investissements dAvenir" program, project ISITE-BFC (contract ANR-lS-IDEX-OOOB). Notations and conventions Geometry. All schemes in this paper will be noetherian, finite dimensional and assumed to come equipped with a dimension function, usually denoted δ. While we fix such a dimension function through- out, and while it may even appear explicitly in various notions we use, we emphasize that the choice is inessential in the sense that the most of the relevant notions do not depend on the fixed choice of dimension function up to a suitable notion of canonical equivalence; see Remark 2.1.6 for a more precise statement.1 Anyway the dimension function is used to fix conventions regarding degrees of (co)homology as in 2.1.10. Fix a base scheme S. An S-scheme X will be said to have essentially finite type or, equivalently, X → S has essentially finite type, if X can be written as a (co)filtered limit a S-schemes of finite type with affine ´etale transition morphisms. By a point of a scheme X, we will mean a map x : Spec(K) → X where i) K is a field and ii) if k is the residue field of the image of x in X, then the extension K/k is finitely generated; equivalently, the morphism x : Spec(K) → X has essentially finite type. We will simply write such a point by x ∈ X(K). The dimension function δ on S, is extended to a dimension function on X as follows: ∀x/s ∈ X/S, δ(x) := δ(s) + trd(κ(x)/κ(s)). δ(X) = δ+(X) := max x∈X δ−(X) := min x∈X (δ(x)). (δ(x)). Two examples to keep in mind include: (S1) if S is the spectrum of a field k, δ = 0, then δ(x) = trd(κ(x)/k); (S2) if S is an excellent regular scheme of dimension less or equal than 3 and δ is the Krull dimension. In both cases, if X/S has finite type, then the integer δ(X) coincides with the Krull dimension of X. Finally, the following formulas will be used in the paper: (D1) if S is regular connected, δ = d − codimS where d = δ(S); (D2) if the morphism f : X → S has essentially finite type, and is lci with cotangent complex Lf , for any point x ∈ X, δ(x) = rk(cid:0)Lf,x(cid:1) + δ(s). Moreover, if X and S are irreducible, and d = dim(f ) is the rank of Lf , one has: δ(X) = d + δ(S). Motivic stable homotopy. We fix a motivic triangulated category T in the sense of [CD09, Def. 2.4.45], equipped with a motivic adjunction: SH ⇆ T . In brief, T consists of the following data: for any scheme S, a triangulated closed symmetric monoidal category T (S); for any morphism of schemes f and any separated morphism of finite type p, pairs of 1There is an exception to this rule. this is when one considers the underlying monoidal structure and its interaction with the δ-homotopy t-structure. It will appear once in our applications, when we consider products on the homotopy Leray spectral sequence. We refer the reader to Sections 3.3 and 4.2. THE HOMOTOPY LERAY SPECTRAL SEQUENCE 5 adjoint functors (f ∗, f∗), (p!, p!) satisfying the so-called Grothendieck six functor formalism (see [CD09, Th. 2.4.50] for a precise statement). Following the terminology from stable homotopy theory objects of T (S) will be called T -spectra over S. We write 1S for the monoidal unit in T (S), and 1S(1) for the Tate twist. In the sequel, various combinations of Tate twists and shifts naturally arise, and we introduce a separate notation for these twists: 1S{1} := 1S(1)[1], 1Sh1i := 1S(1)[2]. Given an object E in T (S), and a morphism f : X → S (resp. a separated morphism having finite type) we may define E-cohomology and bivariant E-theory by means of the formulas: • (Cohomology) En,i(X) = HomT (X)(1X , f ∗E(i)[n]). • (Bivariant theory En,i(X/S) = HomT (X)(cid:0)1X (i)[n], f !E(cid:1). This notation is a standard in motivic homotopy theory. When dealing with spectral sequences, and homotopy modules, it will both be useful and meaningful to use the following notations: • H n(X, E) = En,0(X). • Hn(X/S, E) = En,0(X/S). We will use the later notations exclusively in Sections 4 and 5. The definition of the bivariant theory attached to E will be extended to the situation where X has essentially finite type over S (see Paragraph 2.1.1). The indexing for cohomology used above may also be expressed in terms of the other conventions for twists mentioned above, with notation changed accordingly: one has equalities of the form En,i(X) = En−i,{i}(X) = En−2i,hii(X); similar notation will be used for the associated bivariant theory. When E = 1S, the various coho- mology groups will be referred to as T -cohomology and bivariant T -theory respectively, and we write H n,i(X, T ) (resp. H BM n,i (X/S, T )) for these groups. Write K(S) for the category of virtual vector bundles over S, associated with the category of vector bundles over S with morphisms the isomorphisms of vector bundles. The Thom space construction may be viewed as a functor ThS : K(S) → T (S) that sends sums to tensor products. Following [DJK18], we may twist cohomology and bivariant theories by pairs (n, v) ∈ Z × K(X); we use the following notation for these twists: En(X, v) = HomT (X)(1X , f ∗E ⊗ T hX(v)[n]) En(X/S, v) = HomT (X)(cid:0) ThX (v)[n], f !E(cid:1). Given an integer i ∈ Z, we denote by hii the unique free virtual vector bundle of rank i -- the underlying scheme is implicit -- so that this notation is compatible with our conventions on twists. If K is a perfect complex over X, we denote by hKi the associated virtual bundle (see [DJK18, 2.1.5]). In order to be able to apply the construction of the δ-homotopy t-structure of [BD17], we will require that T satisfies the following assumptions: (T1) T is generated by Tate twists of smooth schemes: more precisely, for any scheme S, T (S) is generated as a triangulated category by objects of the form MS(X)(i) with X/S smooth and i ∈ Z.2 (T2) T is continuous with respect to Tate twists: see [CD09, 4.3.2].3 (T3) T is homotopically compatible: see [BD17, 3.2.12].4 (T4) A suitable form of resolution of singularities holds: property (Resol) of [BD17, 2.4.1].5 2All objects in T (S) are obtained by taking extensions of arbitrary coproducts. Equivalently, an object K of T (S) is zero if and only if: ∀X/S smooth,(n, i) ∈ Z2, HomT (S) (cid:0)MS(X)(i)[n], K(cid:1) = 0. 3Recall that this expresses the compatibility of T with projective limits. 4This condition is automatically verified (see [BD17, 3.2.13]) if T satisfies absolute purity and the following vanishing statement holds: ∀fields E, ∀n > m, H n,m(Spec(E), T ) = 0. 5Under two geometric assumptions, this means that: • In case (S1), we will require that, for p the characteristic exponent of k, T is Z[1/p]linear. • In case (S2), we require that T is Q-linear. 6 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL Examples. The abstract setting being given, we now give our list of concrete frameworks, that will be used in all of our examples. We first fix some absolute base S0 and restrict our schemes to S0-schemes essentially of finite type. (1) Motivic case. Let R be a ring of coefficients, and assume one of the following situation: • S0 is the spectrum of a field k whose characteristic exponent is invertible in R and T = DM(−, R) is Voevodsky's cdh-local category of triangulated mixed motives, introduced in [CD15]. • S0 is a Dedekind ring (or more generally an excellent scheme of dimension greater or equal than 3), R is a Q-algebra and T = DM(−, R) is one the many models of the triangulated category of R-motives introduced in [CD09]. (2) Homotopical case. The absolute base S0 is the spectrum of a field with characteristic expo- nent p, and T = SH[p−1] is the Z[p−1]-linearization of the Morel-Voevodsky stable homotopy category of P1-spectra. Particular motivic ring spectra of interest to us will include: S = 1S the sphere spectrum. • S0 • HRS the motivic Eilenberg-MacLane ring spectrum with coefficients in R.6 • H RS the Milnor -- Witt motivic ring spectrum with coefficients in R. See [DF17]. To fix ideas, the reader may take the dimension function δ0 on S0 satisfying the conditions (S1) or (S2) fixed above, and for any S0-scheme essentially finite type, take the induced dimension function (as described above). 2. Homotopy t-structure and duality The goal of this section is to review a variant of the homotopy t-structure analyzed in [BD17] that makes sense over rather general base schemes. In more detail, Section 2.1 reviews the necessary theory from [BD17] and fixes additional notations and conventions to be used in the sequel. We highlight here Theorem 2.1.4 which summarizes the required existence result and Point 2.1.8 which describes homological conventions for t-structures that will be in force throughout the paper. On the other hand, Section 2.2 is somewhat orthogonal to the above considerations and is concerned with recalling some facts about purity and duality that will be used in the remainder of the paper. 2.1. Recollections on the homotopy t-structure. 2.1.1. For any T -spectrum E over a scheme S, bivariant E-theory extends canonically from the category of separated S-schemes of finite type to that of separated S-schemes essentially of finite type. Indeed, suppose X is a separated S-scheme essentially of finite type. By assumption, there exists a pro-scheme (Xλ)λ with affine and ´etale transition morphisms, such that each Xλ is separated and has finite type over S, and with limit X. We set: En,i(X/S) := lim−→ λ En,i(Xλ/S). This definition is independent of the pro-scheme presenting X as a limit [GD67, §8.2]. The resulting definition presents a canonical extension of the original functor by the continuity assumption on T (i.e., property (T2) from our conventions). Moreover, when X has finite type over S the new definition agrees with the old definition by using again the continuity assumption on T . Proposition 2.1.2. If E is a T -spectrum over S, then the following conditions are equivalent. (i) For any separated scheme X/S of finite type, one has: En,i(X/S) = 0 when n − i < δ−(X), respectively n − i > δ+(X). (ii) For any point x ∈ S(K) (see our Notations and conventions), one has: En,i(x) = 0 when n − i < δ(x), respectively n − i > δ(x). 6Recall this ring spectrum is obtained as the image of the constant motive under the canonical map: K ◦ γ∗ : DM(S, Z[p−1]) → SH(S)[p−1], by forgetting the transfer and then taking the Nisnevich Eilenberg-MacLane functor. See for example [CD09]. THE HOMOTOPY LERAY SPECTRAL SEQUENCE 7 Proof. The equivalence of (i) and (ii) follows by combining Theorem 3.3.1 and Corollary 3.3.5 of [BD17]. Alternatively, it is a straightforward consequence of the existence and convergence of the δ-niveau spectral sequence ([BD17, Def. 3.1.5] or Paragraph 3.2.2 in this paper). (cid:3) Remark 2.1.3. (1) In [BD17], the extension E∗∗ to separated schemes essentially of finite type was denoted by E∗∗. Since, according to paragraph 2.1.1, this extension is unique and well-defined -- using in particular the assumption (T2) for coherence, we will not follow this notational convention here (and we caution the reader that the decoration E is used with a different meaning in this paper. ) (2) If we use the δ-niveau spectral sequence, then the proof of the previous proposition does not use the assumptions (T1), (T3) and (T4). Therefore, the preceding proposition is true without assuming these conditions. We can now state the main theorem of [BD17] (see loc. cit. Th. 3.3.1 and Cor. 3.3.5). Theorem 2.1.4. Given any scheme S, there exists a t-structure on T (S) whose homologically non- negative (resp. non-positive) objects are the T -spectra E over S satisfying the equivalent conditions (i) and (ii) of the above proposition. This t-structure is, moreover, non-degenerate and satisfies gluing in the sense of [BBD82, 1.4.10] (see also Remark 2.1.12). Definition 2.1.5. Given any scheme S, the t-structure on T (S) of the above theorem will be called the δ-homotopy t-structure. Objects of the heart of this t-structure, denoted by T (S)♥, will be called δ-homotopy modules, or simply homotopy modules when this does not lead to confusion. Remark 2.1.6. The t-category T (S) is "independent" of the choice of δ in a sense we now explain. Given another choice δ′, we know that the function δ′ − δ is constant on connected components of S. As T (S ⊔ S′) = T (S) ⊕ T (S′), we may obviously assume S is connected so that δ′ = δ + n. It follows from the above definition that φδ,δ′ : (T (S), tδ) → (T (S), tδ′ ), E 7→ E[n] is an equivalence of t-categories. It may be useful to remember the formula: (2.1.6.a) τ δ+n ≥p = τ δ ≥p+n (see the conventions of Par. 2.1.8 for these truncation functors). Example 2.1.7. (1) Motivic case: Let k be a perfect field and assume δ is the obvious dimension function on k. In the motivic case, the δ-homotopy t-structure on DM(k, R) coincides with the stable version of Voevodsky's homotopy t-structure introduced in [D´eg11, Sec. 5.2]. See [BD17, Ex. 2.3.5] for details. In particular, a δ-homotopy module E over k is just a Z-graded homotopy invariant sheaf with transfers equipped with an isomorphism: (En+1)−1 ≃ En. We refer the reader to [D´eg11, 1.17] for more details. (2) Homotopical case: In the homotopical case, given a field k with the obvious dimension function, the δ-homotopy t-structure on SH(k) coincides with Morel's homotopy t-structure (see [Mor03, Sec. 5.2]). We refer the reader again to [BD17, Ex. 2.3.5] for more details. In this case, a δ-homotopy module E over k is a Z-graded strictly A1-invariant Nisnevich sheaf over smooth k-schemes with a given isomorphism: (En+1)−1 ≃ En. In other word, this is a homotopy module in the sense of Morel. Recall also that a spectrum E is called orientable if it admits the structure of a module over the ring spectrum MGL. Orientability for homotopy modules turns out to be equivalent to requiring that the Hopf map η (an element in the graded endomorphisms of the motivic sphere spectrum) acts trivially. In fact, orientability is also equivalent to requiring that E admits transfers, in which case these transfers are unique (see [D´eg13, 4.1.5, 4.1.7] for further details). (3) Over a general base S, in both the homotopical and motivic cases, the δ-homotopy t-structure can be compared with the perverse homotopy t-structure defined by Ayoub in [Ayo07, §2.2.4]. This comparison requires an appropriate choice of δ, and we refer the reader to [BD17, 2.3.11] for details. 8 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL 2.1.8. Homological conventions. -- For the most part, we adopt homological conventions, as they are better suited to issues that arise involving singularities. We will write E ≥δ n (resp. E ≤δ n) to say that E is concentrated in homological degree above n − 1 (resp. below n + 1) and denote by τ δ ≥n (resp. τ δ ≥n) the corresponding homological truncation functor. Using homological conventions, the truncation triangles read: τ δ ≥0(E) → E → τ δ and Hom(E, F) = 0 if E ≥δ 0 and F <δ 0. <0(E) +1−−→ We denote by H δ n the n-th homology functor with respect to the tδ-homotopy t-structure. Finally, we summarize conventions with respect to suspensions: H δ n(E[i]) = H δ n−i(E), τ δ ≥n(E[i]) = τ δ ≥n−i(E)[i], τ δ ≤n(E[i]) = τ δ ≤n−i(E)[i]. Remark 2.1.9. We may pass from homological to cohomological indexing by writing indices as super- scripts and reversing signs. In formulas: δ (E) = H δ −n(E), τ ≤n <−n. ≥−n, τ >n δ = τ δ δ = τ δ H n 2.1.10. For the record, we now give a number of equivalent characterizations of positivity or negativity with respect to the δ-homotopy t-structure -- this is simply a consequence of Proposition 2.1.2, using the above conventions and the construction in Paragraph 2.1.1. Given a T -spectrum E over S and an integer m ∈ Z, the following conditions are equivalent: (i) E ≥δ m (resp. E ≤δ m) (ii) For any separated scheme X/S having finite type, one has: En,i(X/S) = 0 when n − i < m + δ−(X), respectively n − i > m + δ+(X). (ii') For any separated scheme X/S being essentially of finite type, one has: En,i(X/S) = 0 when n − i < m + δ−(X), respectively n − i > m + δ+(X). (iii) For any point x ∈ S(K) (see again our Notations and conventions), one has: En,i(x) = 0 when n − i < m + δ(x), respectively n − i > m + δ(x). 2.1.11. t-exactness of the six operations. -- Given a functor F between triangulated categories equipped with t-structures, one says that F is left (resp. right) t-exact if it respects homologically negative (resp. positive) objects; such a functor F is t-exact if it is both left and right t-exact. One says that F has homological amplitude [a, b] if for any object E: • E ≥ 0 ⇒ F (E) ≥ a. • E ≤ 0 ⇒ F (E) ≤ b. Let f : X → S be a morphism essentially of finite type, and d the maximum dimension of its fibers. We consider the δf -homotopy t-structure on T (X), where δf is the dimension function on X induced by that of S with respect to the morphism f (see Notations and conventions page 4). Then one has the following results: • f ∗[d] is right tδ-exact. • If f is smooth, f ∗[d] is tδ-exact. • The functor f∗ has tδ-amplitude [0, d]. The first, second and third points are respectively proved in [BD17], 2.1.6(3), 2.1.12, 3.3.7. If in addition f is separated of finite type, we get: • f! is right tδ-exact. • f ! is tδ-exact. • If δ ≥ 0 then ⊗ is right tδ-exact. These points are respectively proved in [BD17], 2.1.6(1), 3.3.7(4), 2.1.6(2). Remark 2.1.12. As mentioned in the Theorem 2.1.4, the δ-homotopy t-structure satisfies gluing. We recall here precisely what this means. Consider a closed immersion i : Z → S with complementary open immersion j : U → S and a T -spectrum E over S. The following conditions are equivalent: • E is homologically non-tδ-negative (resp. non-tδ-positive). THE HOMOTOPY LERAY SPECTRAL SEQUENCE 9 • j∗E and i∗E are homologically non-tδ-negative (resp. j∗E and i!E are homologically non-tδ- positive). The equivalence of these conditions can be deduced from the localization property of T and the t- exactness stated above (see [BD17, Cor. 2.1.9]). Let us now explicitly state the following consequence of the result on the tensor product. Proposition 2.1.13. Assume the dimension function δ on S is non-negative. Let E be a T -spectrum over S. Then for any pair (p, q) ∈ Z2, one has a canonical pairing: ¯φ : τ δ ≥p(E) ⊗ τ δ ≥q(F) → τ δ ≥p+q(E ⊗ F) which is bi-functorial in E. Proof. According to the preceding paragraph, the assumption implies that the tensor product ⊗ in T (S) preserves non-negative objects. Consider the canonical map τ δ ≥p(E) ⊗ τ δ ≥q(F) φ −→ E ⊗ F. According to the preceding assertion, the left hand-side is in homological degrees ≥ p + q. Consider the distinguished triangle: ≥p+q(E ⊗ F) a−→ E ⊗ F τ δ b−→ τ δ <p+q(E ⊗ F) +1−−→ We deduce that the composition b ◦ φ is zero. So φ uniquely factors through a giving us the desired map ¯φ. The bifunctoriality of ¯φ follows from the uniqueness. (cid:3) Example 2.1.14. Consider the assumptions of the preceding proposition. The previous pairing is associative in an obvious sense - this follows from the uniqueness of the map ¯φ. Hence the symmetric monoidal structure on T (X) induces a canonical symmetric monoidal structure on T (X)♥, using the formula, for δ-homotopy modules E and F: Note in particular that the canonical functor: E ⊗H F := τ δ ≤0(E ⊗ F). T (X)≥δ 0 → T (X)♥, E 7→ τ δ ≤0(E) = H δ 0 (E) is monoidal. Remark 2.1.15. Beware that the monoidal structure defined above a priori depends on δ ≥ 0. To be more precise, though the heart for two different choices of dimension functions δ ≥ 0, δ′ ≥ 0 are equivalent as additive categories, according to Remark 2.1.6, this equivalence is not compatible in general with the tensor structures induced by δ and δ′. Indeed, if k is a field and we are in the motivic (or the homotopical) case, if we take a dimension function δ on k such that δ(k) = n > 0, then the induced tensor product on the heart is just the zero bi-functor ! The situation of a positive dimensional base is more complicated though, and it seems there are as many non-trivial tensor structures induced as in the above example as the dimension of the base. 2.2. Recollection on purity and duality. 2.2.1. Consider a T -spectrum E over a scheme S. Let us recall a construction from [DJK18]. Let f : Y → X be a quasi-projective lci morphism of S-schemes with cotangent complex Lf . Let EX be the pullback of E along X/S. Then we associate to f a purity transformation (see [DJK18, 4.3.1]); evaluated at the object EX , it gives a canonical map: pf : f ∗(EX ) ⊗ Th(Lf ) → f !(EX ). where Th(Lf ) is the Thom space associated with the perfect complex Lf . The following definition extends classical considerations; in the motivic case, see [DJK18, 4.3.9]. Definition 2.2.2. Consider the above notations. We will say that E is absolutely pure if for any quasi-projective morphism f : Y → X between regular schemes, the map pf is an isomorphism. If S is regular, we will say that E is S-pure if for any quasi-projective morphism f : X → S with X regular, pf is an isomorphism. Example 2.2.3. (1) Motivic case: the constant motive 1S0 is absolutely pure. Equivalently, 1S is S-pure for any regular scheme S. 10 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL (2) Homotopical case: Then any spectrum E over the base field k is absolutely pure (see [DJK18, 4.3.10(ii)]). Equivalently, any spectrum E over a regular base S is S-pure. An almost immediate corollary of the S-purity assumption is the following duality statement. Proposition 2.2.4. Let f : X → S be a morphism and E be a T -spectrum over S. We assume one of the following hypothesis is fulfilled: • f is essentially smooth. • f is essentially quasi-projective, X and S are regular and E is S-pure. Then for any pair (n, v) ∈ Z × K(X), the map pf gives an isomorphism: (2.2.4.a) En(X, v) ∼−→ E−n(X/S, hLf i − v) which is contravariantly natural in X with respect to ´etale maps. Proof. The case where f is of finite type is tautological, while the contravariance with respect to ´etale map follows from the compatibility of pf with ´etale pullbacks (apply [DJK18, 3.3.2(iii)] in the case where p is ´etale). The general case is obtained using the previous one, together with the naturality with respect to (cid:3) ´etale maps, and the extension of bivariant theory described in 2.1.1. Remark 2.2.5. (1) This isomorphism can be described, at least when f is of finite type, as the cap-product by the fundamental class ηf ∈ E0(X/S, hLf i) of f : see [DJK18, 4.3.9, 2.3.14]. In fact, the description of the above isomorphism in the general case immediately follows when one considers the obvious extension of the notion of fundamental classes to essentially quasi- projective lci morphisms. (2) Recall that if either E is an object of DM(S, R) or if E is an MGL-module7 over S then the Thom isomorphism implies we can canonically identify the twist by a virtual vector bundle with the Tate twist by its rank. In particular, the above isomorphism takes the following classical form: (2.2.5.a) En,i(X) ∼−→ E2d−n,d−i(X/S) where d is the relative dimension of f . 3. Fiber homology and Gersten complexes In this section, we investigate the heart of the homotopy t-structure discussed in the preceding section in greater detail. Section 3.1 is concerned with studying an approximation to the notation of homology associated with the truncation functors for the homotopy t-structure. Definition 3.1.2 introduces the notion of fiber δ-homology, which is esentially the restriction of δ-homology to points. This section concludes with Theorem 3.1.7, which introduces an "effective" variant of the t-stucture of Section 2.1, ameliorating some unboundedness issues that naturally arise because we work in a stable context. Section 3.2 recalls conventions for exact couples, and introduces Gersten complexes (see Definition 3.2.4), built out of fiber δ-homology, which essentially form the E1-page of the niveau spectral sequence. We also give a detailed description of the differentials of the Gersten complexes and important formal properties of these complexes are summarized in Proposition 3.2.7. Finally, Section ?? is devoted to analyzing multiplicative structure in these complexes and the associated spectral sequences. 3.1. Fiber δ-homology. 3.1.1. Recall from the introduction that we have also considered Gm-twists on bivariant theory. Indeed these twists are more natural with respect to the δ-homotopy t-structure, because the functor −(1)[1] is tδ-exact. Using this grading, one can reformulate Proposition 2.1.2 for a given T -spectrum E over S as the equivalence of the following conditions: (i) E ≥ 0 (resp. E ≤ 0). (ii) For any separated scheme X/S essentially of finite type, one has: (iii) For any point x ∈ S(K), one has: En,{∗}(X/S) = 0 when n < δ−(X) (cid:0)resp. n > δ+(X)(cid:1). 7in other words, an oriented spectrum; here module is to be understood in the sense of the monoidal category SH(S) En,{∗}(x) = 0 when n < δ(x) (cid:0)resp. n > δ(x)(cid:1). i.e. in the weak homotopical sense. THE HOMOTOPY LERAY SPECTRAL SEQUENCE 11 In view of this characterization of the δ-homotopy t-structure, we have adopted the following definition in [BD17]. Definition 3.1.2. Let E be a T -spectrum over S. We define the fiber δ-homology of E in degree n ∈ Z as the functor H δ n(E) : Pts(S) → A bZ, x 7→(cid:0)Eδ(x)+n,{δ(x)−r}(x)(cid:1)r∈Z where Pts(S) is the discrete category of points of S. Dually, we define the fiber δ-cohomology of E as H n δ (E) = H δ Given an object F of the δ-homotopy heart, we set Fδ −n(E) (the Gm-grading does not change). ∗ := H δ 0 (F). Therefore E ≥ 0 (resp. E ≤ 0) if and only if H δ n(E) = 0 for n < 0 (resp. n > 0). Remark 3.1.3. Fiber δ-homology is a good approximation of δ-homology. In fact, given an object F of the heart of the δ-homotopy t-structure on T (S), the functor Fδ ∗ is an approximation of a cycle module in the sense of Rost. In the motivic case, this idea can be turned into an equivalence of categories between the δ-homotopy heart of DM(S, R) and the category of Rost R-linear cycle modules over S in the sense of [Ros96]. The details of such a theorem have not yet been written up, but see [D´eg14a]. In the homotopical case, a generalization of Rost's theory is in the work (see [Fel18]). In any case, we have already obtained in [BD17, 4.2.2] that, in the general case of an abstract triangulated motivic category T satisfying our general assumptions, the functor: is conservative, exact and commmutes with colimits. Similarly, the family of functors ( H δ conservative on the whole category T (S). n)n∈Z is T (S)♥ → PSh(cid:0) Pts(S), R − modZ(cid:1), F 7→ Fδ ∗ Example 3.1.4. Let S be a regular connected scheme, and put d = δ(S). We fix a point x : Spec(K) → S. (1) Abstractly, using any S-pure spectrum E, one obtains, because of the duality isomorphism (2.2.4.a) and relation (D2) of dimension functions, a canonical isomorphism: (2) Assume we are in the motivic case. One then obtains (using the previous computation) a H δ n(E)r(x) ≃ E−n−r(cid:0) Spec(K), hLx/Si − hδ(x)i + hri(cid:1). canonical isomorphism: H n δ (1S)r(x) = H r−n−2d,r−d M (Spec(K), R). This follows because under our assumptions, motivic cohomology satisfies absolute purity, and is oriented. Observe, for example, that: H δ d (1S)∗ = H −d δ (1S)∗ = K M ∗ S, the restriction of the Milnor K-theory functor to the discrete category Pts(S). Additionally, the only vanishing that we have is: H δ n(1S) = 0 if n > d. In other words, 1S is concentrated in tδ-homological degrees ] − ∞, d]. On the other hand, we know for a fact, at least when S is a complex scheme, that for all n ≤ d, H δ n(S) 6= 0 (this is due to the existence and non-triviality of polylogarithm elements; see [BD17, Ex. 3.3.2]). So 1S is tδ-unbounded below. (3) Assume we are in the homotopical case. Then one obtains, using the computation of point (1), a non-canonical isomorphism: H δ n(1S)r(x) ≃ πA1 n+d(S0 K)r−d[1/p]. This is because absolute purity is automatic in the homotopical case: see point (2) of Example 2.2.3. In particular, Morel's computation of the zeroth stable A1-homotopy sheaf of the sphere gives an isomorphism, again non-canonical: H δ d (1S)r(x) ≃ K MW r (K) where the right hand side is the r-th Milnor -- Witt cohomology group of the field K. Moreover, the zero sphere spectrum 1S is concentrated in homological degrees ] − ∞, d] for the δ-homotopy t-structure. Not much else can be said in this case since, when S is a complex scheme, the rational homology of the zero sphere spectrum agrees with the rational homology 12 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL of the motivic spectrum, which is non-trivial. Thus, 1S is unbounded below for the δ-homotopy t-structure. (4) Again in the homotopical case, we can consider the Milnor -- Witt motivic ring spectrum H RS. Known computations of Milnor -- Witt cohomology imply, as in the case of motivic cohomology, that H RS is concentrated in homological degrees ] − ∞, d]. Moreover, the unit of this ring spectrum induces a canonical isomorphism: H δ d (1S)∗ ∼−→ H δ d (H RS)∗. Remark 3.1.5. In the non-oriented context, it appears more natural to introduce another twist for a virtual vector bundle v over a scheme X, namely: 1X {v} := ThX (v)[− rk(v)]. This notation is consistent with our notation 1X {r} for the Gm-twist (where r corresponds to the trivial bundle of rank r). Assuming δ(S) = 0 to simplify, the isomorphism of point (1) can be rewritten as: H δ n(E)r(x) ≃ E−n(cid:0) Spec(K), {Lx/S} + {r}(cid:1). The unboundedness of the constant object, observed in Points (2), (3), (4) of the previous example, appears to be a consequence of working in the stable context -- recall that stability with respect to Tate twists is among the axioms of triangulated motivic categories used in [CD09]. This potentially unpleasant feature can be corrected as follows. Definition 3.1.6. (see [BD17, Def. 2.2.1]) Let S be an arbitrary scheme. We define the triangulated category of δ-effective T -spectra over S, denoted by T δ−eff (S), as the full localizing subcategory of T (S) generated by objects of the form: where f : X → S is separated of finite type and δ(X) ≥ n. f!(1X )(n) Given a T -spectrum E over S, we define its effective fiber δ-homology H δeff (E) as the negatively n graded functor obtained from H δ n(E) by restricting the Gm-grading to negative integers. It follows from [BD17, 3.1.1] that one can define an effective version of the δ-homotopy t-structure. Theorem 3.1.7. Consider the notations of the previous definition. For any scheme S, there exists a t-structure on T δeff (S) whose homologically non-negative (resp. non-positive) objects are those T - spectra E over S such that H δeff (E) = 0 for n < 0 (resp. n > 0). This t-structure is non-degenerate and satisfies glueing (recall: [BBD82, 1.4.10], Remark 2.1.12). n Example 3.1.8. Let S be a regular connected scheme, and put d = δ(S). (1) Assume we are in the motivic case. Then the computation done in Example 3.1.4(1) shows that n(1S)≤0 = 0 if n 6= d. In other words, 1S[−d] is in the heart of the effective δ-homotopy H δ t-structure. This is exactly what happens for the perverse t-structure. (2) Assume we are in the homotopical case. The preceding example shows that the Eilenberg- MacLane motivic ring spectrum HRS[−d] is in the heart of the effective δ-homotopy t-structure. The computations of higher stable homotopy groups of spheres do not allow us at the moment to conclude anything about the sphere spectrum. However, the computations of Milnor-Witt motivic cohomology imply, as in the motivic case, that HMW RS[−d] is in the heart of the effective δ-homotopy t-structure. 3.1.9. The δ-effective categories T δeff (S), with the t-structure just defined, satisfy good properties. We refer the reader to [BD17] for the following facts. First, by construction, one has a pair of adjoint functors: such that s is fully faithful and w is tδ-exact. s : T δeff (S) ⇆ T (S) : w • If δ ≥ 0, the subcategory T δeff (S) of T (S) is stable under tensor products. So T δeff (S) becomes a closed symmetric monoidal category with internal hom given by: HomT δeff (S) = w ◦ HomT (S)(M, N ). Moreover, the tensor product on T δeff (S) is right tδ-exact. THE HOMOTOPY LERAY SPECTRAL SEQUENCE 13 • For f : X → S essentially of finite type with dim(f ) ≤ d, we get an adjunction of t-categories: When f is smooth of pure dimension d, f ∗(d)[2d] is tδ-exact. f ∗(d)[2d] : T δeff (S) ⇆ T δeff (X) : w ◦(cid:0)f∗(−d)[−2d](cid:1). • For f : X → S separated and essentially of finite type, we get an adjunction of t-categories: Moreover, w ◦ f ! is tδ-exact. f! : T δeff (X) ⇆ T δeff (S) : w ◦ f !. In the last two points, the dimension function on X is that induced by the one fixed on S (see our Notations and conventions). 3.2. Gersten complexes. We now build a spectral sequence using the theory of exact couples. Since conventions for exact couples vary in the literature, we now fix our conventions, which agree with those used in [BD17, 3.1.2]. Definition 3.2.1. Let A be an abelian category. A homological exact couple of degree d > 0 in A is the data of a pair of bigraded objects (D, E) of A together with a triangle of homogeneous maps (a, b, c) (−1,+1) D / D b ~⑦⑦⑦⑦⑦⑦⑦⑦⑦ (0,0) `❅❅❅❅❅❅❅❅❅ c a E (−d,d−1) such that consecutive maps fit together to form an exact sequence. As usual, one derives a (homologically indexed) spectral sequence from such an exact couple [McC01]: at page r, the r-th term is equal to the differential bigraded abelian group (E, b ◦ c). 3.2.2. We first recall δ-niveau spectral sequences, in the abstract setting and using dimension function following [BD17, 3.1.5].8 Let X be a separated S-scheme essentially of finite type. Recall from [BD17, 3.1.1] that a δ-flag of X is an increasing sequence of reduced closed subschemes Z∗ = (Zp)p∈Z of X such that δ(Zp) ≤ p for all p. The set F (X) of δ-flags, ordered by term-wise inclusion, is cofiltered. Given such a δ-flag, the classical properties of the bivariant theory E∗(−/−) (see our Notations and conventions) imply that we have a long exact sequence: Ep+q(Zp−1/S) a−→ Ep+q(Zp/S) b−→ Ep+q(Zp − Zp−1/S) c−→ Ep+q−1(Zp−1/S) where a (resp. b) is pushfoward (resp. pullback) along the obvious closed (resp. open) immersion, and c the boundary map. These long exact sequences are covariantly functorial with respect to inclusion of flags. Thus we can combine them into the following homological exact couple of degree 1: δDp,q = lim−→ δE1 p,q = lim−→ Z∗∈F (X)(cid:0)Ep+q(Zp/S)(cid:1), Z∗∈F (X)(cid:0)Ep+q(Zp − Zp−1/S)(cid:1). Finally, observe that one can express the E1-term as follows: where δE1 p,q = ⊕x∈X(p) Ep+q(x/S), X(p) = {x ∈ X δ(x) = p}. Definition 3.2.3. Under the assumptions and notations above, the spectral sequence: will be called the δ-niveau spectral sequence of X/S with coefficients in E. δE1 p,q = Mx∈X(p) Ep+q(x/S) ⇒ Ep+q(X/S) 8The main idea is of course classical, introduced by Grothendieck, and thoroughly developed in [BO74]. The main novelty of [BD17] is to work over an arbitrary base scheme S (rather than a field) and appeal to abstract dimension functions. This use of dimension functions can also be found in the homological indexing for Chow groups in [wc15]. See [BD17, 3.1.7] for more on this point. / ~ ` 14 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL The spectral sequence converges to the following filtration, called the δ-niveau filtration: NpE∗(X/S) = [i:Z→X,δ(Z)=p Im(cid:0)i∗ : E∗(Z/S) → E∗(X/S)(cid:1), where i runs over the closed immersions whose source has δ-dimension p. It is standard to consider this spectral sequence not just for E itself, but also for all Tate twists. In fact, if we apply this definition to the graded spectrum E(n) for an integer n ∈ Z, we get the following form: δE1 p,q = Mx∈X(p) Ep+q,n(x/S) ⇒ Ep+q,n(X/S). In view of Rost's theory of cycle modules, it will be useful to introduce the following definition. Definition 3.2.4. Under the assumptions of the preceding definition, we define the Gersten (δ-homologi- cal) complex of X/S with coefficients in E, denoted by Cδ ∗ (X, E), as the complex of abelian groups located at the line q = 0 of the E1-term of the δ-niveau spectral sequence. The p-th homology of this complex will be called the Gersten δ-homology of E and be denoted by: Note that the Gersten complex is concentrated in degrees [δ−(X), δ+(X)]. Moreover, using Definition Aδ p(X, E) = Hp(cid:0)Cδ ∗ (X, E)(cid:1). 3.1.2 for fiber δ-homology, the p-th term of this complex takes the following form: Cδ p (X, E) = Mx∈X(p) H δ 0 Ep(x/S). Remark 3.2.5. The complex defined above is closely linked with Rost's theory of cycle modules. We have used a simplification here. Indeed recall that cycle modules, as well as their associated complexes, are Z-graded. We can recover this Z-grading by applying the above definition to the Z-graded spectrum: It is also standard in stable A1-homotopy theory to consider this grading, called the Gm-grading. The complex Cδ ∗(X, E{∗}) is Z-graded, and has the following form: E{∗} =(cid:0)E(n)[n](cid:1)n∈Z. Cδ ∗ (X, E{∗}) = Mx∈X(p) H δ 0 Ep−∗(x/S). The main reason to use this grading is that the differentials of the Gersten complex are then homogeneous of degree −1. 3.2.6. Consider the notations of the above definition. The differentials of the Gersten complex associated with E and X/S take the following form: Cδ p (X, E) = Mx∈X(p) Ep(x/S) dp−→ Ms∈X(p−1) Ep(s/S) = Cδ p−1(X, E) The differential dp is obtained as the inductive limit of differentials dZ is described as follows: p associated with a δ-flag Z∗ and Ep(Zp/S) / Ep(Zp − Zp−1/S) c Ep−1(Zp−1/S) ,❨❨❨❨❨❨❨❨❨❨❨❨❨❨❨❨❨❨❨❨❨❨❨ dp Ep−1(Zp−2/S) / Ep−1(Zp−1/S) / Ep−1(Zp−1 − Zp−3/S) b where the rows arise as pieces of localization long exact sequences. Using the functoriality of the localization long exact sequences with respect to pullbacks along open immersions, it is therefore possible to explicitly describe the differentials of this complex as follows. Consider a pair (x, s) ∈ X(p) × X(p−1) and denote by (dp)x y : Ep(x/S)r−p → Ep−1(y/S)r−p the corre- sponding component of the above differential. Let us write Z(x) the reduced closure of x in X. From the construction of the Gersten complex given above, and the functoriality of localization long exact sequence in bivariant theory with respect to pullbacks along open immersions and pushforward along closed immersions, one then deduces that: / / / , / / THE HOMOTOPY LERAY SPECTRAL SEQUENCE 15 • if s ∈ Z(x): we let Z(x)(s) be the localization of Z(x) at s; this is a 1-dimensional scheme so that Z(x)(s) = {x, s}. Then (dp)x s is the boundary map of the localization long exact sequence associated with the closed immersion i : {s} → Z(x)(s). Explicitly, it is the middle map in the following exact sequence: Ep(cid:0)Z(x)(s)/S(cid:1) j∗ s = 0. • Otherwise, (dp)x −→ Ep(x/S) (dp)x s−−−→ Ep−1(s/S) i∗−→ Ep−1(cid:0)Z(x)(s)/S(cid:1) The next result summarizes the formal properties of Gersten complexes. Proposition 3.2.7. Consider the notations of the above definition. (1) The complex Cδ ∗ (X, E) is covariantly functorial in E. Given any integer p, the induced maps Cδ ∗ (X, E) → Cδ ∗ (X, τ≤pE), Cδ ∗(X, τ≥−pE) → Cδ ∗ (X, E) are isomorphisms provided p ≥ 0. In particular, one has a canonical isomorphism: Cδ ∗ (X, E) ≃ Cδ ∗ (X, H δ 0 E). (2) The complex Cδ ∗(X, E) is functorial in X/S, covariantly with respect to proper maps and con- travariantly with respect to ´etale maps. (3) If the T -spectrum E is homologically non-tδ-negative (resp. non-tδ-positive), there exists a canonical epimorphism (resp. monomorphism): Ep(X/S) → Aδ resp. Aδ p(X, E) p(X, E) → Ep(X/S). These two maps are functorial in the S-scheme X, covariantly with respect to proper maps and contravariantly with respect to ´etale maps. If E is in the δ-homotopy heart, or more generally is concentrated in one degree for the δ- homotopy t-structure, these two maps are inverse isomorphisms giving a functorial identification: Ep(X/S) ≃ Aδ p(X, E). Proof. Point (1) follows from the δ-niveau spectral sequence and its obvious functoriality in E (Paragraph 3.2.2). Point (2) follows from the classical functoriality of the (δ-)niveau spectral sequences. We refer the reader to [Jin16], proof of Proposition 3.11 for proper functoriality, and proof of Proposition 3.12 for ´etale contravariance. Note an important technical point here: we deal with the case of a general triangulated category T , in contrast with the special case T = DM which is oriented. In the case of proper functoriality, this does not come into play. The case of ´etale contravariance works as well as the tangent bundle of an ´etale map is trivial. Point (3) is a consequence of the convergence of the δ-niveau spectral sequence and of the computation (cid:3) of its E1-term. Remark 3.2.8. (1) According to the isomorphism of Point (3) above, one obtains using the main result of [DJK18] that, given a δ-homotopy module E, the Gersten homology Aδ ∗(X, E) is con- travariant in X with respect to any smoothable lci morphism f : Y → X. More precisely, one gets a canonical morphism: Aδ ∗(X, E) ≃ Ep(X/S) → Ep(Y /S, hLY /X i) ≃ Aδ ∗(Y, EhLY /X i) where the last identification uses the fact EhLY /X i is concentrated in one degree over any connected component of Y -- which is the virtual rank of LY /X,η where η is any generic point of the chosen connected component. (2) As another illustration of the connexion of our Gersten complexes with Rost's theory of cycle complexes, when S is the spectrum of a perfect field k, δ is the obvious dimension function, and we are in the motivic case T = DM(k, R). Then any object E in the heart of DM(k, R) is a homotopy module with transfers in the sense of [D´eg11] which canonically corresponds to a cycle modules E∗ = Eδ ∗ -- this notation corresponds to the one of Definition 3.1.2. Moreover, according to [D´eg12, 2.7(ii)], there exists a canonical isomorphism of complexes where the right hand-side is Rost cycle complex associated with the cycle module E∗. ∗(X, E) ≃ C∗(X, E∗) Cδ 16 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL 3.3. Products. 3.3.1. Consider a T -spectrum E over S. In order to describe products, we will use cohomological notations. We put: Cp δ (S, E) = Cδ −p(S, E), resp. Ap δ (S, E) := Aδ −p(S, E) and call it the Gersten δ-cohomological complex (resp. δ-cohomology) of S with coefficients in E. Note in particular that, when E is in the heart, we get from Proposition 3.2.7 a canonical isomorphism: Ap δ (S, E) ≃ H p(S, E). Consider now a morphism of T -spectra over S: µ : E ⊗ F → G. We deduce as usual a morphism at the level of cohomologies: Ep(S) ⊗Z Fq(S) → Gp+q(S) by sending a pair of maps (a : 1S → E[p], b : 1S → F[q]) to the following map: Using the preceding isomorphism, we deduce a canonical pairing: a⊗b−−→ E ⊗ F[p + q] µ −→ G[p + q]. 1S Ap δ (S, E) ⊗Z Aq δ(S, F) → Ap+q δ (S, G). We deduce the following results. Proposition 3.3.2. If E is a ring T -spectrum over S, then A∗ δ(S, E) has a ring structure. If F is a T -spectrum with equipped with an E-module-structure, then A∗(S, F) is equipped with the structure of a module over A∗ δ(S, E). Example 3.3.3. If E be a ring T -spectrum over S, then according to Example 2.1.14, the graded δ- homotopy module H δ ∗ (E) has the structure of a ring spectrum. According to the preceding proposition, A∗ ∗ (E)) has the structure of a bigraded ring. δ (S, H δ Remark 3.3.4. (1) Products are thus easy to obtain on our Gersten δ-homology. It is possible to follow Rost's approach in [Ros96] and to get a definition of the product on the level of complexes. Using classical techniques due to Levine, it is even possible to find a dg-algebra underlying our Gersten δ-homology (see [Lev06, Lev08, BY18]). An advantage of our approach is that it circumvents these technicalities. (2) One can also extend the preceding considerations to the case of smooth S-schemes. Indeed, in that case, one has: Ap δ (X, E) = Aδ −p(X, E) ≃ E−p(X/S) = H p(X, EhLX/Si) according to Proposition 2.2.4. The preceding proposition obviously extends to that case. Note that the product obtained for smooth S-schemes is now compatible with ´etale pullbacks as defined in Proposition 3.2.7, and even with respect to smooth pullbacks. 4. The homotopy Leray spectral sequences This section contains the main theoretical results of the paper and constructs the spectral sequences we mentioned in the introduction. We introduce two versions of this spectral sequence. Section 4.1 studies a homological version of the Leray spectral sequence; the main result is Theorem 4.1.2, which gives a description of the E2-page of the relevant spectral sequence. Proposition 4.1.4 summarizes formal properties of the resulting spectral sequences. On the other hand, Section 4.2 contains a cohomological version of the spectral sequence (summarized in Theorem 4.2.5) and discusses compatibility with product structures. Finally, Section 4.3 contains a discussion of an analog of locally constant sheaves in our situation. THE HOMOTOPY LERAY SPECTRAL SEQUENCE 17 4.1. The homological version. 4.1.1. Let us fix a T -spectrum E over S and consider the following geometric situation: B X ❅❅❅❅❅ π f S where f is any morphism of schemes, which plays the role of the fibration. We assume π is a separated X = π!E. Given a T -spectrum E over B, we put according morphism essentially of finite type and put E! to what was announced in the paragraph on Notations and conventions: Hp(B, E) = HomT (B)(1B[p], E). Then one can look at the tower of homological truncations of f∗(E! X ) for the δ-homotopy t-structure, which is the analogue of the Postnikov tower in our situation: X ) → τ δ ≥q+1(f∗E! X ) → τ δ ≥q(f∗E! . . . → τ δ ≥q−1(f∗E! X ) → . . . It is standard to deduce from this filtration a spectral sequence. Let us be more precise. We first consider the canonical distinguished triangle: ≥q(f∗E! τ δ ≥q+1(f∗E! ≥q+1(f∗E! X ) → τ δ X ) → τ δ X ) → τ δ =q(f∗E! X )[1] where: τ δ =q(f∗E! X ) = τ δ ≤qτ δ ≥q(f∗E! X ) = H δ q (f∗E! X )[q]. Applying the functor Hp+q(B, −) to the preceding distinguished triangle, we get a long exact sequence: Using the conventions of definition 3.2.1, we then get a homological exact couple of degree 2 such that: X )(cid:1) c−→ Hp+q−1(cid:0)B, τ δ ≥q+1(f∗E! X )(cid:1) . ≥q+1(f∗E! Hp+q(cid:0)B, τ δ X )(cid:1) a−→ Hp+q(cid:0)B, τ δ ≥q(f∗E! ≥q(f∗E! q (f∗E! We deduce from that exact couple our main construction. Dp,q = Hp+q(cid:0)B, τ δ Ep,q = Hp(cid:0)B, H δ q (f∗E! X )(cid:1) b−→ Hp(cid:0)B, H δ X ))(cid:1) , X )(cid:1) . Theorem 4.1.2. Fix assumptions and notations as in 4.1.1. The exact couple defined above gives a convergent spectral sequence of the form: The E2-term is concentrated in the range p ∈ [δ−(X), δ+(X)] and is the homology in degree p of the δ-homological Gersten complex Cδ E2 p,q(f, E) = Hp(cid:0)B, H δ ∗(cid:0)B, H δ . . . → Mx∈B(p) q (f∗E! X )(cid:1) ≃ Aδ p(cid:0)B, H δ q (f∗E! X ))(cid:1) ⇒ Ep+q(X/S). q (f∗E))(cid:1) (see Definition 3.2.4) which takes the form: dG Ep+q−1(Xs/S) → . . . Ep+q(Xx/S) p−−−→ Ms∈B(p−1) where Xx is the fiber of f above the point x ∈ B. Note, in particular, that the differentials dr p,q are trivial for r >(cid:0)δ+(X) − δ−(X)(cid:1). Proof. The spectral sequence follows from the theory of (derived) exact couples. One can compute the E2-term as in the above statement by using Proposition 3.2.7: Going back to the definition, this is the p-th homology of the following Gersten δ-homological complex: q (f∗E! Hp(cid:0)B, H δ q (f∗E! X )(cid:1) . p(cid:0)B, H δ X )(cid:1) ≃ Aδ p−−−→ Ms∈B(p−1) dG Ep+q(x/B, f∗E! X ) Ep+q−1(s/B, f∗E! X ) → . . . . . . → Mx∈B(p) Given any point x ∈ B(p) one considers the cartesian square: ix Xx fx  Spec(κx) x / X f / B The computation: E∗(x/B, f∗E! X ) = [x!1B[∗], f∗E! X ] = [1x[∗], x!f∗E! X ] = [1x[∗], fx∗i! x E! X ] = [1x[∗], i! xπ!E] = E∗(Xx/S) / / / /    18 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL yields the E2-term in the form of the statement and implies that the spectral sequence converges. (cid:3) Definition 4.1.3. The above spectral sequence will be called the (homological) homotopy Leray spectral sequence associated with f and with coefficients in E. The spectral sequence is obviously functorial in the T -spectrum E over S. If we want to analyze the behavior with respect to Tate twists, we can look at the spectral sequence with coefficients in E(n), which takes the following form: Similarly, twists by more general Thom spaces man be considered. The following result summarizes the formal properties of the spectral sequence. E2 p,q = Aδ q (f∗E! p(cid:0)B, H δ X )(n))(cid:1) ⇒ Ep+q,n(X/S). Proposition 4.1.4. We consider the following diagram of schemes: S π′ t❥❥❥❥❥❥❥ j❚❚❚❚❚❚❚ π Y X p f ′ *❚❚❚❚❚❚❚ 4❥❥❥❥❥❥❥ f B and a T -spectrum E over S. Set E! be the induced morphism on the fibers over a point x ∈ B. X = π!E and E! Y = π′!(E). Given any point x ∈ B, let px : Yx → Xx • If p is proper, then there is an adjunction map: φp : f ′ ∗ E! Y = f ′ ∗π′!(E) = f∗p!p!π!(E) ad(p!,p!) −−−−−→ f∗π!(E) = f∗(E! X ) which induces a morphism of spectral sequences converging to the indicated map on the abutment: q (f ′ ∗ E! q (f∗E! Aδ Aδ φp∗ p(cid:0)B, H δ p(cid:0)B, H δ Y ))(cid:1) X ))(cid:1) 3 Ep+q(Y /S) p∗ 3 Ep+q(X/S). Moreover, the map φp∗ is induced by the following morphism of Gersten complexes: Ep+q(Yx/S) Ep+q(Xx/S) Px px∗ / Lx∈B(p) / Lx∈B(p) dG p,Y dG p,X Ls∈B(p−1) / Ls∈B(p−1) Ep+q−1(Ys/S) . . . Ps ps∗ Ep+q−1(Xs/S) / . . . • If p is ´etale, then there is an adjunction map: ψp : f∗(E! X ) = f∗π!(E) ad(p∗,p∗) −−−−−−→ f∗p∗p∗π!(E) = f ′ ∗π′!(E) = f ′ ∗ E! Y which induces a morphism of spectral sequences, converging to the indicated map on the abut- ment: . . . . . . . . . . . . q (f∗E! q (f ′ ∗ E! Aδ Aδ ψp∗ p(cid:0)B, H δ p(cid:0)B, H δ X ))(cid:1) Y ))(cid:1) 3 Ep+q(X/S) p∗ 3 Ep+q(Y /S). Moreover, the map φp∗ is induced by the following morphism of Gersten complexes: Ep+q(Xx/S) Ep+q(Yx/S) Px p∗ x / Lx∈B(p) / Lx∈B(p) dG p,X dG p,Y Ls∈B(p−1) / Ls∈B(p−1) Ep+q−1(Xs/S) . . . Ps p∗ s Ep+q−1(Ys/S) / . . . Proof. Each point is obtained by using the functoriality of the homotopy Leray spectral sequence with respect to the T -spectrum E. The computation of the map on the abutment follows from the definition of the functoriality of the bivariant theory. The map on the E2-terms follows from point (2) of Proposition 3.2.7. (cid:3) t *   j 4   +   + / / /   / /   / / /   +   + / / /   / /   / / / THE HOMOTOPY LERAY SPECTRAL SEQUENCE 19 Remark 4.1.5. Using fundamental classes as defined in [DJK18], and the induced functoriality on bivari- ant theory, one can extend the contravariant ´etale functoriality to smoothable lci morphisms f : Y → X, up to considering twists by the Thom space of the cotangent complex of f . We leave the formulation to the reader. It is possible to describe the filtration on the abutment of the homotopy Leray spectral sequence in geometric terms. The following definition is the obvious generalization of the classical definition of Grothendieck (see [BO74]). Definition 4.1.6. Consider the setting of Paragraph 4.1.1. We define the δ-niveau filtration on E∗(X/S) relative to f as: δN f p E∗(X/S) = [i:Z→X,δ(Z)≤p Im(cid:0)i∗ : E∗(X ×B Z/S) → E∗(X/S)(cid:1). where i runs over the closed immersions with target X. Remark 4.1.7. One can also describe this filtration by the following formula: δN f p E∗(X/S) = [π:Y →X,δ(Y )≤p Im(cid:0)π∗ : E∗(X ×B Y /S) → E∗(X/S)(cid:1). where π : Y → X runs over the proper morphism with target X. Indeed, such a morphism always factors as Y ¯π−→ π(Y ) i−→ X where π(Y ) denotes the image of Y , with its canonical structure of a closed subscheme of X. As the map ¯π is surjective, one has δ(π(Y )) ≤ p. This concludes. 4.1.8. Before stating the computation of the filtration on the δ-homotopy spectral sequence, we will introduce notations in order to simplify the proof. We consider again the assumptions and notations of Paragraph 4.1.1. First, note that by definition, the filtration induced by the δ-homotopy spectral sequence on E∗(X/S) is defined as: δF f q following ind-schemes: Second, we can define the δ-niveau filtration relative to f at the level of schemes, by considering the E∗(X/S) = Im(cid:0)H∗(B, τ≥qf∗E! X ) → E∗(X/S)(cid:1). Therefore we get a closed ind-immersion B≤p by considering the following pro-objects: ip−→ B. We can define a kind of complementary immersion B≤p = " lim−→" Zp. Z∗∈F (B) B>p = "lim←−" Z∗∈F (B) (B − Zp) together with the pro-open immersion B>p jp−→ B. Using these notations, we can consider the localization long exact sequence E∗(X ×B B≤p/S) ip∗−−→ E∗(X/S) j∗ p−→ E∗(X ×B B>p/S) ∂p−→ E∗(X ×B B≤p/S) where the third (resp. first, fourth) member(s) is the obvious colimit, using the contravariance of E∗(−/S) with respect to open immersion (resp. contravariance with respect to closed immersions). This long exact sequence is nothing else than the filtered colimit of the localisation sequences with respect to the closed immersions X ×B Zp → X for δ-flags Z∗ of B. With these notations, the δ-niveau filtration relative to f simply equals the image of ip∗. Proposition 4.1.9. Consider the above assumptions and notation. Then for any pair of integer (p, n) ∈ Z2, one has the following relation: δF f p En(X/S) = δN f n−p En(X/S), where the left hand-side is the filtration on the abutment of the homotopy Leray spectral sequence asso- ciated with f and the right-hand side is the δ-niveau filtration relative to f . 20 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL Proof. Let us put E′ = f∗E! X = f∗π!E. We want to compare the following filtrations: δF f q δN f p E∗(X/S) = Im(cid:0)H∗(B, τ≥q E′) → H∗(B, E′)(cid:1), E∗(X/S) = Im(cid:0)H∗(B≤p, E′) → H∗(B, E′)(cid:1). So, reasoning with E′ instead of E, we reduce to the case X = B = S, f = IdX , π = IdX . We will start with the following lemma: Lemma 4.1.10. Let E be a T -spectrum over a scheme X. Then one has the following vanishing: Hn(cid:0)X≤p, τ δ Hn(cid:0)X>p, τ δ <q ≥q E(cid:1) = 0 if n ≥ p + q, E(cid:1) = 0 if n ≤ p + q. This follows from the case X = S, m = q, i = 0 of the equivalent conditions of Paragraph 2.1.10. We can now consider the following commutative diagram, whose rows and columns are exact se- quences: Hn+1(cid:0)X>p, τ δ ≥q E(cid:1) a ≥q <q Hn+1(cid:0)X≤p, τ δ E(cid:1) / Hn(cid:0)X≤p, τ δ E(cid:1) Hn(cid:0)X≤p, E(cid:1) Hn(cid:0)X≤pτ δ <q, E(cid:1) / Hn(cid:0)X>p, τ δ ≥q E(cid:1) b d c ≥q / Hn(cid:0)X, τ δ E(cid:1) / Hn(cid:0)X, E(cid:1) According to the preceding lemma, one gets that: • a is an isomorphism if n ≥ p + q. • b is an isomorphism if n ≤ p + q − 1, and an epimorphism if n = p + q. Therefore, we obtain that Im(c) = Im(d) if n = p + q. This concludes. (cid:3) Example 4.1.11. The preceding proposition recovers the conjecture of Washnitzer proved by Bloch and Ogus in [BO74, 6.9]. This is obtained as follows: • S is the spectrum of a perfect field k, X is smooth over k; • E = EdR is the (ring) spectrum representing De Rham cohomology as in [CD12, §3.1] while T = DM(−, Q). The fact the truncation of EdR for the homotopy t-structure agrees with the truncation of the De Rham complex follows from the construction of EdR ([CD12, 3.1.5]). Remark 4.1.12. The proof of Bloch and Ogus is less precise and more theoretical. It consists in proving that the niveau spectral sequence for De Rham cohomology agrees with the hypercohomology spectral sequence associated with the De Rham complex. We do not need such a comparison to prove our result, and our proof is more direct. But however, let us indicate that there is also an underlying comparison of spectral sequences. In fact, it is possible to prove that the homotopy Leray spectral sequence defined above agrees from E2-on with the δ-niveau spectral sequence 3.2.3 of X/S with coefficients in the spectrum f∗(E). The case where X = B, f = IdB, S is the spectrum of a perfect field was proved in [Bon10] and [D´eg14b]. The general case will be treated in future work. 4.2. The cohomological version. 4.2.1. Let us consider again the assumptions of Paragraph 4.1.1, but with X = S. So we fix a morphism of schemes: f : X → B and a T -spectrum E over X. To get products on the δ-homotopy spectral sequence of (f, E), we will use a classical construction of Douady (see [Dou59, Section II]) which uses the theory of spectral diagrams from [CE99, XV.7] rather than that of exact couples.   /   /   /   / THE HOMOTOPY LERAY SPECTRAL SEQUENCE 21 Here is how one gets such a spectral diagram underlying the spectral sequence (4.2.8.a). We first define the following tautological functors in a T -spectrum F over B (in the end F plays the role of f∗E): τ δ ≥−∞(F) = F, τ δ τ δ <−∞(F) = 0, τ δ ≥+∞(F) = 0, <+∞(F) = F. Consider the following poset: such that (p, q) ≤ (p′, q′) when p ≤ p′ and q ≤ q′. Then for any (p, q) ∈ P, we put: P = {(p, q) p, q ∈ Z ∪ {±∞}, p ≤ q}. τ δ [p,q[(F) := τ δ ≥pτ δ <q(F). This defines a contravariant functor τ δ(F) : P → T (X), (p, q) 7→ τ δ [p,q[(F). Using the truncation triangles associated with the δ-homotopy t-structure, we obtain distinguished triangles for p ≤ q ≤ r: [q,r[(F) → τ δ τ δ [p,r[(F) → τ δ [p,q[(F) ∂−→ τ δ [q,r[(F)[1] where the first two maps are given by the functoriality of τ δ. These formulas imply that the contravariant functor: defines a Z-graded spectral diagram in the sense of loc. cit..9 Again, this construction is obviously functorial in E. P 7→ A b, (p, q) 7→ H∗(cid:0)B, τ δ [p,q[(f∗E)(cid:1) Note that for F = f∗(E), the exact couple of Paragraph 4.1.1 is contained in the preceding spectral diagram -- it corresponds to pairs (p, p + 1). Therefore, the spectral sequence associated with the preceding spectral diagram in [CE99] coincides with the δ-homotopy Leray spectral sequence of (f, E). In particular, we get the following formula: Er p,q(f, E) = Im(cid:16)Hp+q(cid:0)B, τ δ [p−r,p[(f∗E)(cid:1) → Hp+q(cid:0)B, τ δ [p−1,p+r−1[(f∗E)(cid:1)(cid:17). Remark 4.2.2. After the seminal work of Cartan and Eilenberg, spectral diagrams have appeared in other forms in the literature. In the triangulated context, one can refer, mainly for historical purposes, to [Ver96]. The ∞-categorical context is much more recent, but also much more powerful and satisfactory. The notion of spectral diagram in the ∞-categorical context is introduced by Lurie in [Lur18, Section 1.2.2], under the name of Z-complex. The advantage of using ∞-category with pushouts (eg: stable ∞-categories) is that a Z-complex is essentially equivalent to a tower of objects: see [Lur18, Lemma 1.2.2.4]. The fact that we can stay in the old-fashioned world of triangulated categories stems from the good behavior of t-structures. To obtain products on the homotopy Leray spectral sequence following Douady, we need to refine Proposition 2.1.13 as follows. Proposition 4.2.3. We consider the preceding notations and assume δ ≥ 0. Suppose we are given a morphism of T -spectra over a scheme B: µ : F ⊗ F′ → F′′. Then for any triple of integers (p, q, r) such that r ≥ 0, there exists a canonical map: τ δ [p,p+r[(F) ⊗ τ δ [q,q+r[(F′) ϕp,q,r−−−−→ τ δ [p+q,p+q+r[(F′′). Moreover, this pairing satisfies the formulas (SPP 1) and (SPP 2) of [Dou59, II.A, D´efinition, p. 19- 06].10 9Property (SP.5) of loc. cit. is not immediate. It follows from the convergence of the homotopy Leray spectral sequence -- see Theorem 4.1.2. 10(SPP1) is the functoriality of this pairing with respect to (p, p + r) (resp. (q, q + r)) considered as an object of P. (SPP2) is the Leibniz rule for the boundary map of type ∂. 22 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL Proof. It is sufficient to treat the case where F′′ = F ⊗ F′ and µ is the identity. As by assumption δ ≥ 0, the tensor product respects homologically positive objects (see the end of Paragraph 2.1.11). According to Proposition 2.1.13, we get a canonical pairing: Let us consider the following diagram: τ δ ≥p(E) ⊗ τ δ ≥q(F) → τ δ ≥p+q(E ⊗ F). ≥p(E) ⊗ τ δ τ δ ≥q+r(F) ❑ ❑ ❑ ≥p(E) ⊗ τ δ τ δ ≥q(F) ❑ ❑ ❑ ❑ ❑ ❑ τ δ ≥p+q+r(E ⊗ F) τ δ ≥p+q(E ⊗ F) ❑ %❑ τ δ ≥p(F) ⊗ τ δ [q,q+r[(F′) /❴❴❴❴ τ δ [p+q,p+q+r[(F ⊗ F′) where the solid arrows form a commutative diagram. The two columns of this diagram come from distinguished triangles. First this implies that the slanted dotted arrow is zero. Second, it implies that there exists a unique dotted horizontal arrow making the bottom square commutative. Then we consider the following diagram: τ δ ≥p+r(E) ⊗ τ δ [q,q+r[(F) (1) *❯❯❯❯❯❯❯ 4✐✐✐✐✐✐✐ (2) [p,p+r[(F) ⊗ τ δ τ δ [q,q+r[(F′) τ δ ≥p(E) ⊗ τ δ [q,q+r[(F) τ δ [p+q,p+q+r[(E ⊗ F) As the tensor product is right tδ-exact, the object τ δ [q,q+r[(F) is in homological degree ≥ p + q + r. By definition, the object τ δ [p+q,p+q+r[(E ⊗ F) is in homological degree < p + q + r. So the map labeled (1) must be 0. Therefore, the map (2) must exist and gives the existence of the pairing ϕp,q,r. The uniqueness of the construction then guarantees Douady's coherence properties (SPP1) and (SPP2). (cid:3) ≥p+r(E) ⊗ τ δ 4.2.4. Granted Proposition 4.2.3, one may now apply the construction of [Dou59, Th. II]. Going back to the setting of Paragraph 4.2.1, and to our morphism f : X → B, we consider a pairing of T -spectra over X: µ : E ⊗ E′ → E′′. As f∗ is weakly monoidal (left adjoint of a monoidal functor), we get a pairing: µ : f∗(E) ⊗ f∗(E′) → f∗(E′′). Applying the preceding proposition and Douady's construction, we get a pairing of spectral sequences: Ep,q r (f, E) ⊗Z Es,t r (f, E′) → Ep+s,q+t r (f, E′′) such that the differentials dr satisfy the usual Leibniz rule. Following standard usage, when considering products, we renumber the δ-homological spectral se- quence cohomologically, and use our cohomological conventions (see in particular Paragraph 3.3.1). We then obtain the following result. Theorem 4.2.5. Suppose f : X → B is a morphism of schemes, and let E be a T -spectrum over X. The constructions of Paragraph 4.2.1 and 4.2.4 yield a convergent spectral sequence of the form: Ep,q 2 (f, E) = Ap δ (B, H q δ (f∗E)) ⇒ H p+q(X, E). If E admits a ring structure, then the spectral sequence is equipped with a multiplicative structure; the product on the E2-term is induced by the construction of Example 3.3.3. Before proceeding to a computation of the above E2-term, let us introduce the following general construction within the six functors formalism satisfied by T : / /   %   / /     /   * / /   4 THE HOMOTOPY LERAY SPECTRAL SEQUENCE 23 Proposition 4.2.6. Let f : X → S be a separated morphism essentially of finite type. Then there exists a pair of adjoint functors: f! : T (X) → T (S) : f ! such that for any factorisation X u−→ X0 f ! = u∗ ¯f !. ¯f −→ S where u is pro-´etale and ¯f is of finite type, one has: Proof. We write X as a projective limit of a pro-S-scheme (Xi)i∈I whose transition morphisms are ´etale affine of finite type and Xi/S is separated of finite type. If we choose an index i ∈ I, considering the induced factorisation X ui−→ Xi fi−→ S we put f! = fi!ui♯, and f ! = u∗ we have two indexes i, j ∈ I with a map j → i. Then we get a commutative diagram: i. This definition does not depend on the chosen index i. First assume i f ! X X uj ui / Xj ϕij / Xi fj fi X / X and we get: f ! = u∗ i f ! i = u∗ j ϕ∗ ij f ! i = u∗ j ϕ! ij f ! i = u∗ j f ! j. In general, given two indices i, j ∈ I, there exists an third index k ∈ I and maps k → i, k → j which gives the canonical identification. The definition does not depend on the pro-object chosen to present X/S. Indeed, any two such pro- objects are isomorphic. Finally, any factorisation as in the statement of the proposition can be taken as an element of a pro-object presenting X/S. (cid:3) Remark 4.2.7. (1) A more rigorous approach for the preceding proof will use a limit argument as in Deligne's construction of the functor f! when f is separated of finite type. Details are left to the reader. (2) The procedure described in the previous proposition extends a classical trick used in [BBD82, 2.2.12]. (3) The extension of the pair of adjoint functors (f!, f !) to the case where f : X → S is essentially of finite type immediately gives an extension of the definition of bivariant theories to schemes X/S essentially of finite type. It is straightforward to check this extension, under the continuity assumption (T2) agrees with that defined in Paragraph 2.1.1. 4.2.8. We endeavor to describe the E2-term of the cohomological form of the homotopy Leray spectral sequence associated with a morphism f : X → B and a T -spectrum E over X. Recall from Theorem 4.1.2 (with X = S and with cohomological conventions) that Ep,q 2 (f, E) is the p-cohomology of the Gersten δ-cohomological complex C∗ δ f∗E). Moreover, for p ∈ Z, one has: δ (B, H q (4.2.8.a) Cp δ (B, H q δ f∗E) = Mx∈B(p) H p+q(Xx/X, E), where we have used the extension of the bivariant theory defined by E as explained in the preceding remark. Proposition 4.2.9. Notations and assumptions as in 4.2.8, the following statements hold. (1) If f is smooth and E = f ∗E0 for a T -spectrum E0 over B, then, for any p ∈ Z, one has: Cp δ (B, H q δ f∗E) = Mx∈B(p) H p+q(cid:0)Xx, Ex(cid:1) where Ex = x!E. More generally, given any point x : Spec(K) → B such that δ(x) = −p, and any integer r ∈ Z, one has (recall Definition 3.1.2 and Paragraph 3.3.1): again with Xx = x−1X and Ex = x!E. H q δ f∗Er(x) = H p+q(cid:0)Xx, Ex{p + r}(cid:1), / / /   / / 24 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL (2) If X is regular, the fibers of f are regular and E is X-pure (Definition 2.2.2), then for any p ∈ Z, one has: Cp δ (B, H q δ f∗E) = Mx∈B(p) H p+q(cid:0)Xx, Eh−N (Xx/X)i(cid:1) where N (Xx/X) is the normal bundle of the regular closed immersion ix : Xx → X and we set Eh−N (Xx/X)i = i! x E ⊗ Th(−N (Xx/X)). Proof. Consider the first point. We directly prove the assertion concerning a general point x : Spec K → B, as it implies the remaining assertion.11 As in the proof of Theorem 4.1.2, consider the following cartesian diagram: ix Xx fx X f Spec(K) x / / B Since f is smooth, there is a canonical isomorphism of functors: i! x x! (using the notation intro- duced just before the statement of the proposition for the shriek functors). The following computation then concludes the proof: xf ∗ = f ∗ H p+q(Xx/X, E) =(cid:2)1Xx , i! xf ∗(E)[p + q](cid:3) =(cid:2)1Xx , f ∗ =(cid:2)fx♯(1x), Ex[p + q](cid:3) = H p+q(cid:0)Xx, Ex(cid:1). x x!(E)[p + q](cid:3) The second point is a direct consequence of the form (4.2.8.a) of the Gersten complex and of Propo- (cid:3) sition 2.2.4. Again, this spectral sequence has good functoriality properties (e.g., in the (ring) T -spectrum E). The next result summarizes other functoriality properties. Proposition 4.2.10. Consider a commutative diagram: Y ❆❆❆❆ g q B X ~⑥⑥⑥⑥ f and a ring T -spectrum E. Let us put EY = q∗E. There exists a morphism of converging spectral sequences: Ep,q 2 (f, E) = Ap δ(B, H q δ (f∗E)) p∗ 2 (f ′, E′) = Ap where q∗ is the usual pullback on cohomology. Ep,q δ (B, H q δ (g∗E′)) H p+q(X, E) q∗ 3 H p+q(Y, EY ) The morphism of spectral sequences is simply obtained using the functoriality in E with respect to the following map: E → q∗q∗(E) = q∗EY . 4.3. Remarkable properties of homotopy modules. 4.3.1. δ-effectivity. -- Consider the situation of Paragraph 4.1.1: E is a T -spectrum over S and one looks at morphisms: B X ❅❅❅❅❅ π f S where π is separated essentially of finite type, and put E! X = π!E. We assume in addition: X and 1B are δ-effective (Definition 3.1.6). (1) E! (2) f is proper. 11To be clear, recall: H p+q(Xx/X, E) = ( H q δ f∗E)−p(x). / /     / / ~ + 3     + / / THE HOMOTOPY LERAY SPECTRAL SEQUENCE 25 Then the first and second assumptions imply f∗(E! Moreover, one can compute the E2-term of the homotopy Leray spectral sequence as follows: X ) is δ-effective (see Paragraph 3.1.9). X ) = f!(E! HomT (B)(1B[p], H δ q f∗E) = HomT (B)(s1B[p], H δ q sf∗E) = HomT δeff (B)(1B[p], wH δ = HomT δeff (B)(1B[p], H δeff = HomT δeff (B)(1B[p], H δeff q f∗E). q sf∗E) q wsf∗E) The first identification uses the assumption (1), the second the adjunction (s, w), the third the fact w is tδ-exact and the last one the fact s is fully faithful. We have obtained the following remarkable result. Proposition 4.3.2. With the assumptions of 4.3.1 in place, the homotopy Leray spectral sequence takes the following form: E2 p,q = Aδ p(B, H δeff q f∗E! X ) ⇒ Ep+q(X/S). If X = S, then we can also consider the cohomological form of the δ-homotopy Leray spectral sequence: Ep,q 2 = Ap δ (B, H q δeff f∗E) ⇒ Ep+q(X). Example 4.3.3. The interest of the preceding proposition is that it is easier to get bounded objects with respect to the δ-effective category: see Example 3.1.8. Let us consider either the homotopical or motivic case. We consider a proper morphism f : X → B of schemes essentially of finite type over k such that X is k-smooth. We let our dimension functions be induced by that computed relative to k. Then 1X (n) for n ≥ 0 and 1B are both δ-positive. If we assume X is of pure dimension d, so that δ(X) = d according to the preceding choice, then the constant object 1X is concentrated in homological degree d. Therefore the preceding spectral sequence, in its homological form, is concentrated in degrees q ∈ [d, d + dim(f )]. The following definition is an analog of the notion of local system in classical topology. Definition 4.3.4. Suppose p : B → S is a separated morphism of finite type and E is a T -spectrum over B. We will say that E is S-simple if there exists a T -spectrum E0 over S and an isomorphism E ≃ p!E0. We will say that T is locally simple over S (or locally S-simple) if there exists a Nisnevich cover π : W → B such that π∗E is S-simple (with respect to the projection p ◦ π). Note that the terminology S-simple is analogous to the classical terminology of "simple local system". This corresponds to the case of trivial monodromy. Remark 4.3.5. This notion will come into play mainly when S is the spectrum of a base field k, or, in the motivic case, of a base Dedekind ring A. For us, the interest comes, as in topology, in the study of the homotopy Leray spectral sequence (Definition 4.1.3). Usually, we will start with a k-simple T -spectrum E over S -- thus EX is k-simple. If we know that the homotopy module H δ q f∗E over B is k-simple, then the E2-term depends only on the homotopy type (resp. motive) of B over k. We will give examples in Section 5. Remark 4.3.6. (1) Obviously, locally S-simple T -spectra over B are stable under suspensions, twists and even tensor products by Thom spaces of virtual bundles over B. The same is true for S-simple T -spectra except that one can only twist them by virtual bundles over B that come from S. Neither of these notions is stable under extensions or even direct factors in general. (2) Let f : B → S be smooth and consider the motivic case. Then f ! = f ∗(d)[2d] where d is the relative dimension of f . One deduces that S-simple motives over B are stable under tensor products. The same remark applies to oriented spectra over B, but not to arbitrary spectra. (3) Note that if k is a field, k-simple over a scheme B implies B-pure (Definition 2.2.2). (4) One could say that a homotopy module over B is S-constructible if it is obtained by a finite number of extensions and direct factors of locally S-simple homotopy modules, within the abelian category of homotopy modules over B. This notion is not so well-behaved, compared to its model for torsion ´etale sheaves, as it lacks some notion of finiteness. It would be desirable to have some good finiteness condition on S-simple homotopy modules. But even when S is a base field, it is not obvious to find such a finiteness condition; see [D´eg11, Rem. 6.7] for further discussion. Note the following fact. 26 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL Lemma 4.3.7. Consider a separated morphism f : B → S of finite type, and a T -spectrum E over B. Let us fix unrelated dimension functions δ on B and δ0 on S. If E is S-simple, then for any p ∈ Z, the homotopy module H δ q (E) over B is S-simple. Proof. We just need to be precise about dimension functions. By additivity, we can assume that B is connected. Then the dimension function δf 0 = δ + n for a fixed integer n ∈ Z (see Remark 2.1.6). In particular, H δ0 0 on B, induced by δ0, satisfies the relation δf q+n(E) according to loc. cit. q (E) = H δ By assumption, E = f !E0. it remains to apply the fact f ! is tδ0 -exact to conclude: H δ q (E) = H δ0 q−n(E) ≃ H δ0 q−n(f !E0) = f !H δ0 q−n(E0). Example 4.3.8. In the motivic case, if f : X → S is smooth, then the constant object 1X is S-simple. In contrast, if f has sufficiently complicated singularities, then 1X may fail to be S-simple. (cid:3) Similarly, the classical oriented ring spectra HRX , KGLX (K-theory), MGLX (algebraic cobordism), HQℓ,X (representing continuous ℓ-adic cohomology) are all S-simple. On the contrary, spectra representing non-orientable theories such as H RX , KQX (hermitian K- X , are not S-simple except when the tangent bundle of f is trivial (or theory), or the sphere spectrum S0 is the pullback of a vector bundle over S). In any case, they all are locally S-simple. Note finally that when f : X → S is arbitrary separated of finite type, the main result of [Jin18], Theorem 1.3, tells us that the spectrum GGLX representing algebraic G-theory in SH(X) is S-simple. As expected, here is the generic case where the homotopy modules appearing in the E2-term of the homotopy Leray spectral sequence are simple. Proposition 4.3.9. Let B and F be S-schemes separated essentially of finite type. We consider the trivial fibration f : X = F ×S B → B. Then for any S-simple T -spectrum E over X, the T -spectrum f∗E is S-simple. Proof. This is a trivial exercise on the six functors formalism. We consider the cartesian square: X q F f ∆ B p f0 / / S. Then we get an associated exchange isomorphism: Ex(∆! a T -spectrum E0 over S such that E = h!E0, h = pf = f0q. Thus we can do the computation: ∗) : p!f0∗ ∼−→ f∗q!. By assumption, there exists f∗E = f∗h!E0 = f∗q!f ! 0 E0 ∗)−1 Ex(∆! −−−−−−−→ p!f0∗f ! 0 E0. This concludes the proof. (cid:3) Corollary 4.3.10. Let f : X → B be Nisnevich locally principle fibration of S-schemes. Then for any locally S-simple T -spectrum E over B, the T -spectrum f∗E is locally S-simple. 5. Applications In this section, we study the homotopy Leray spectral sequence in various simple cases and include some applications. Section 5.1 studies morphism with (stably) A1-contractible fibers, and then mor- phisms with fibers that are "A1-homology spheres". Section 5.2 studies "fibrations" with either base or fiber that are motivic spheres. Finally, Section 5.3 is concerned with some applications of the spectral sequence to relative cellular spaces. 5.1. Morphisms with A1-contractible fibers. We first analyze the case where the fibers are A1- homotopically "as simple as possible", i.e., A1-contractible. Proposition 5.1.1. Suppose f : X → B be a smooth morphism with A1-contractible fibers. Let M be a spectrum (resp. a motive) that is a homotopy module over B and set MX := f ∗M . Then, H q δ f∗MX =(M q = 0, q 6= 0. 0 / /     THE HOMOTOPY LERAY SPECTRAL SEQUENCE 27 In particular, the homotopy Leray spectral sequence is concentrated on the line q = 0, and thus degen- erates; pullback along f yields identifications (in cohomological notation) of the form: H p(X, MX) = H p(B, H 0 δ f∗(MX )) = H p(B, M ). Note in particular that f∗ respects S-simple and locally S-constant objects. Proof. Consider the map α : H q δ (M ) → H q δ (f∗MX ) induced by the adjunction map M → f∗f ∗(M ). We need only to prove it is an isomorphism on fiber homology (Remark 3.1.3). We compute the above map, evatuated at a point x : Spec K → B and in Gm-degree r ∈ Z: H q δ (M )r(x) → H q δ (f∗MX )r(x). We use the computation of Proposition 4.2.9(1), which we can apply as f is smooth. So if we put δ(x) = −p and Mx = x!M (using the notation of Proposition 4.2.6), one obtains: δ (MX)r(x) = H p+q(x, Mx{p + r}), H q δ (f∗MX)r(x) = H p+q(Xx, Mx{p + r}). H q Moreover, the canonical map α is isomorphic to the pullback map: f ∗ x : H q+r−2p,r−p(x, Mx) → H q+r−2p,r−p(Xx, Mx), which is itself an isomorphism as fx is assumed to be an A1-weak equivalence. Note also that, because the spectral sequence is functorial with respect to pullbacks (Proposition 4.2.10), we know that the identification of the statement arises from the pullback map along f . The remaining assertions are straightforward. (cid:3) Example 5.1.2. There exist many examples of morphisms satisfying the hypotheses of Proposition 5.1.1. For concreteness, assume k is a field having characteristic 0. In [AD07, Theorem 1.3], it is shown that there exist connected k-schemes B of arbitrary dimension and smooth morphisms f : X → B of relative dimension ≥ 6 whose fibers are A1-contractible and such that fibers over distinct k-points of B are pairwise non-isomorphic. Because of the last point, such morphisms f are not Zariski locally trivial. These results were improved in [DF18, DPØ18], where it was shown that one could build f as above that are smooth of relative dimension ≥ 3. On the other hand, it is expected that the only A1-contractible smooth k-scheme of dimension 2 is A2, and a long-standing conjecture of Dolgachev -- Weisfeiler [VD74, 3.8.5] states that every flat morphism of (say) smooth schemes with all fibers isomorphic to affine space is Zariski locally trivial. Suppose f is a smooth morphism f with A1-contractible fibers. It is not clear to the authors whether such an f is unstably an A1-weak equivalence without imposing further hypotheses (e.g., that f is Nisnevich locally trivial). Nevertheless, the following remark demonstrates that such f are stable A1-weak equivalences in a strong sense, which makes Proposition 5.1.1 somewhat unsurprising. Remark 5.1.3. Given a map f : X → B as in Proposition 5.1.1, one can directly show that the adjunction map: Id → f∗f ∗ is an isomorphism of functors. Indeed, we can use the same argument as above and the fact the family of functors x! : SH(B) → SH(Spec(κ(x))), indexed by schematic points x ∈ B, is conservative.12 In particular, one does not need to work over a base field; nor does one need to invert any integers. As f is smooth, we also get a natural transformation f♯f ∗ → Id, which is an isomorphism according to the result of the preceding paragraph. In particular, the map Σ∞X+ = f♯f ∗(1B) → 1B = Σ∞B+ is an isomorphism in SH(B). In fact, f is a universal stable A1-weak equivalence since f is a stable A1-weak equivalence and the property mentioned above remains true after base change. 12One uses for example Proposition 4.2.6 to define x!. The conservativity property is obtained using the continuity property of SH and the localization property. 28 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL 5.2. Gysin and Wang sequences. The relative Atiyah -- Hirzebruch spectral sequence takes a partic- ularly simple form when the Serre fibration F → X → B under consideration has either the property that B is a sphere, or F is a homology sphere and the associated local system on B is trivial. In those cases, the spectral sequences yields the so-called Wang or Gysin long exact sequences. The fact that the differentials in the cohomological form of the spectral sequence are derivations yields additional structure in these long exact sequences that is frequently useful in computations. A1-homology spheres. In motivic homotopy theory, there are many smooth schemes over a base S that have the stable A1-homotopy type of a motivic sphere ΣiG∧j m . For example, Morel -- Voevodsky showed that An \ 0S has the A1-homotopy type of Σn−1G∧n m . Likewise, the split smooth affine quadric Q2n−1 S is A1-weakly equivalent to An \ 0 and [ADF17, defined by the hypersurface Pi xixn+i = 1 in A2n Theorem 2] demonstrates that the smooth affine quadric Q2n defined by the equation Pi xixn+i = x2n+1(1 − x2n+1) in A2n+1 m has no smooth model if i > j and conjecturally has no smooth model if i < j −1. We now formulate a definition of "homology sphere" in A1-homotopy theory. m . On the other hand, it is known that ΣiG∧j is a model of ΣnG∧n S Definition 5.2.1. We say that a (pointed) smooth S-scheme X is an A1-homology sphere if there exist integers p, q, r ≥ 0 and an A1-weak equivalence ΣrX ∼ ΣpG∧q m . The next proposition gives a construction of many A1-homology spheres, at least over a field. Proposition 5.2.2. Fix a smooth base scheme S, and an A1-contractible smooth S-scheme X. Assume there exists a closed immersion of S-schemes x : S → X with trivial normal bundle νx/X . (1) A choice of trivialization of νx/X determines an A1-weak equivalence ΣX \ x ∼= ΣAd \ 0, i.e., X \ x is an A1-homology sphere. (2) If S = Spec(k) for a perfect field k, X has dimension d ≥ 3 and is A1-connected, then X \ x is A1-simply connected as well. Proof. For the first point, note that there is a cofiber sequence of the form X \ x −→ X −→ X/(X \ x) −→ ΣX \ x −→ · · · Because the A1-local model structure is left proper, the fact that X is A1-contractible implies that the map T h(νx/X) → ΣX \ x is an A1-weak equivalence. Under the assumptions on S, there is a homotopy purity isomorphism X/(X \ x) ∼= T h(νx/X) and a choice of trivialization of νx/X determines an A1-weak equivalence T h(νx/X) ∼→ ΣdG∧d m , which can be written ΣAd \ 0. For the second point, since X \ x is A1-connected, X \ x has a non-empty set of k-points by the unstable 0-connectivity theorem [MV99, §2 Corollary 3.22]. Fix a base k-point in X \ x, and point X by its composite with the open immersion X \ x → X. Finally, since d ≥ 3, we may appeal to [AD09, Theorem 4.1] to conclude that the morphism πA1 1 (X) is an isomorphism. Since X is A1-contractible, the latter sheaf is trivial and X \ x is thus A1-simply connected. (cid:3) 1 (X \ x) → πA1 Remark 5.2.3. If k is a field, and X is furthermore affine, then X \ x is isomorphic to Ad \ 0 if and only if X is isomorphic to Ad. Indeed, if there exists an isomorphism from X \ x to Ad \ 0, then normality of X allows one to extend this isomorphism to an isomorphism of X with Ad; the other implication is immediate. Example 5.2.4. If X is the smooth affine threefold defined by x + x2y + z2 + t3 = 0, then the main result of [DF18] implies that X is A1-contractible, at least if k is an infinite field. In that case, for any extension L/k, B. Antieau observed that X is connected by chains of affine lines (see [DPØ18, Example 2.28] for a proof). If L is an infinite field, we may always assume our chains avoid a codimension ≥ 2 subset, and in particular it follows that X \ x is connected by chains of affine lines; it follows that, X \ x is A1-connected. Thus, if k is infinite and perfect, X \ x satisfies the hypotheses of the proposition and yields an "exotic motivic sphere". In dimension d ≥ 4, the examples in [AD07] or [ADF17] also satisfy the hypotheses of the theorem, at least over an infinite base field. Gysin and Wang sequences for homotopy modules. Our goal now is to analyze the homotopy Leray spectral sequence for f : X → B a smooth morphism where B is an A1-homology sphere in the sense above; the outcome will be a version of the Gysin sequence. We begin by observing that from the appropriate cohomological standpoint, A1-homology spheres behave in a fashion analogous to spheres in classical homotopy theory, i.e., cohomology with "locally constant coefficients" is concentrated in precisely 2 degrees. To make this precise, assume we work over a field, and let δ be the usual dimension THE HOMOTOPY LERAY SPECTRAL SEQUENCE 29 function relative to k. We consider homotopy modules M over k such that f∗M is k-simple in the sense of Definition 4.3.4. Lemma 5.2.5. Assume k is a field, and X is a (pointed) smooth k-scheme that is an A1-homology sphere as in Definition 5.2.1, with p, q ≥ 1. Let f : X → Spec(k) be the canonical projection. For any homotopy module M over k, MX := f ∗M , H i δ(f∗MX ) =  M M−q 0 if i = 0 if i = p − r otherwise. Proof. This result follows essentially immediately from [AF14, Lemma 4.5]. However, to keep the presentation self-contained, we sketch a proof in the spirit of this paper: one proceeds along the same lines as the proof of Proposition 5.1.1 and appeals to Proposition 4.2.9(1). Again we need only to compute fiber homology of f∗(MX ). To this end, take a pair (x, n) where x : Spec(K) → Spec(k) is a induced by a field extension of finite type, and n ∈ Z. To simplify the notation, let us assume that δ(x) = 0; set Mx = x!M . We may then appeal to the computation of Proposition 4.2.9(1): H i δ(f∗MX )n(x) ∼= H i(Xx, Mx{r}) = H i(K, Mx{n}) ⊕ H i(Xx, Mx{n}). As X/k is pointed, Xx/K is also pointed and we obtain an identification: H i(Xx, Mx{n}) = H i(K, Mx{n}) ⊕ H i(Xx, Mx{n}) where H ∗ stands for the reduced cohomology of a pointed scheme. Furthermore there are isomorphisms of the form: H i(Xx, Mx{n}) ∼= H i+r(ΣrXx, Mx) ∼= H i+r(ΣpG∧q m , Mx) ∼= H i+r−p(G∧q m , Mx). Since Mx is a homotopy module over K, the last group vanishes if i + r − p is not equal to 0 and is precisely M−q(Spec(k)) if i = p − r. The result then follows by unwinding the definitions. (cid:3) Proposition 5.2.6. Assume k is a field and f : X → B is a Zariski locally trivial smooth morphism of k-varieties with B connected and where the fibers F of f are A1-homology spheres (where ΣrF ∼ ΣpG∧q m with p, q ≥ 1). Assume furthermore that f trivializes on a Zariski open cover U = {Ui}i∈I of B, and that we may fix x ∈ ∩i∈I Ui(k) (i.e., the intersection is non-empty). If M is a homotopy module and H j δ (f∗MX ) is k-simple for each i ≥ 0, then and there is a long exact "Gysin" sequence of the form: H j δ (f∗MX ) =  MB (M−q)B 0 if j = 0 if j = p − r otherwise. , · · · −→ H i(B, M ) −→ H i(X, M ) −→ H i+r−p(B, M−q) ∂−→ H i+1(B, M ) −→ · · · . Proof. By Theorem 4.2.5, there is a spectral sequence with Ei,j 2 = Ai(B, H j δ (f∗MX)) =⇒ H i+j(X, M ). Granted the first statement, the spectral sequence is concentrated in two rows, and the existence of the resulting long exact sequence is immediate. Thus, it remains to prove the first statement. If B is a field, the computation of H j δ (f∗MX) is simply Lemma 5.2.5. The general case reduces to that one using the simplicity assumption and a Zariski patching argument. In more detail, by assumption, we may choose a k-point x : Spec(k) → B. By Lemma 5.2.5 there are induced morphisms M → x∗H 0 (f∗MX ). Since H j δ (f∗MX ) is k-simple by assumption, pulling back these morphisms along the structure map B → Spec(k) yields morphisms MB → H 0 (f∗MX ); moreover, these maps are compatible with restrictions to open sets U ⊂ B. The vanishing statement for H j δ (f∗MX) will be established in an identical manner. Since f is Zariski locally trivial, upon fixing a trivialization over the open cover U, appeal to Proposition 4.3.9 yields isomorphisms MUi δ (f∗MX)Ui for any i ∈ I, (f∗MX)Ui . To check the maps of sheaves on and similarly isomorphisms of the form (M−q)Ui B are isomorphisms, it suffices to check that the isomorphisms upon restriction to Ui just described are compatible on 2-fold intersections. However, since the all the sheaves in question are k-simple, this compatibility may be checked after pullback along x, where it is immediate. (cid:3) δ (f∗MX ) and (M−q)B → H p−r δ (f∗MX) and M−q → x∗H p−r ∼→ H 0 ∼→ H p−r δ δ δ 30 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL Remark 5.2.7. Proposition 5.2.6 avoids the "degenerate" case where f : X → B is a Gm-torsor. Such morphisms are, of course, always Zariski locally trivial, but behave like covering spaces in topology. Indeed, under the additional hypotheses in the statement, the sheaf H j δ (f∗MX) is only non-vanishing when j = 0, in which case it is isomorphic to MB ⊕ (M−1)B. ∗ or K MW Proposition 5.2.6 admits a refinement when M is a homotopy module that admits a ring structure ). In such cases, H 0(B, M ) is a ring, and we may fix a generator ξ of H 0(B, f∗MX,q) ∼= (e.g., K M H 0(B, M ) as a module over this ring. The class ∂(ξ) then determines an element of H p−r+1(B, Mq) that we will refer to as the Euler class of f . ∗ Theorem 5.2.8. Assume k is a field and f : X → B is a Zariski locally trivial smooth morphism of k-varieties with B connected and where the fibers F of f are A1-homology spheres (where ΣrF ∼ ΣpG∧q m with p, q ≥ 1). Assume furthermore that f trivializes on a Zariski open cover U = {Ui}i∈I of B, and that we may fix x ∈ ∩i∈I Ui(k) (i.e., the intersection is non-empty). Suppose M is a homotopy module that admits a ring structure. If H j δ (f∗MX) is k-simple for each i ≥ 0, then a choice of generator ξ of H 0(B, f∗MX,q) ∼= H 0(B, M ) determines an Euler class e(f ) ∈ H p−r+1(B, Mq), and the exact sequence of Proposition 5.2.6 takes the form: · · · −→ H i(B, M ) −→ H i(X, M ) −→ H i+r−p(B, M−q) ·e(f ) −→ H i+1(B, M ) −→ · · · , where the connecting homomorphism is given by product with e(f ) arising from the ring structure. Proof. Granted the identifications mentioned before the statement, the result follows from the existence of multiplicative structure in the homotopy Leray spectral sequence as discussed in Paragraph 4.2.4. (cid:3) Proposition 5.2.9. Assume k is a field, and f : X → B is a smooth morphism where B is an A1- homology sphere (where ΣrX ∼ ΣpG∧q δ (f∗M ) is k-simple for each j ≥ 0, then there is a long exact "Wang" sequence of the form: m with p, q ≥ 1). If M is a homotopy module such that H j · · · −→ H i(X, M ) −→ H i δ(f∗M )(k) θ−→ H i+1+r−p δ (f∗M )−p+r(k) −→ H i+1(X, M ) −→ · · · , where the map θ is a graded derivation. Proof. This result follows immediately by combining Theorem 4.2.5 and Lemma 5.2.5, which shows that the resulting spectral sequence is concentrated in two columns. (cid:3) 5.3. Relative cellular spaces. The notion of "algebraic cell decomposition" and the related notion of "relative algebraic cell decomposition" has a long history. E.g., if a split torus acts on a smooth projective scheme X over a field, then Bialynicki -- Birula [ByB73] showed that X may be decomposed as a disjoint union of smooth varieties that are total spaces of vector bundles over connected components of the fixed point loci (which are necessarily smooth). The cohomological consequences of the existence of such filtrations were observed almost immediately (e.g., one immediately computes Chow groups for smooth projective varieties equipped with a torus action with isolated fixed points). Karpenko was one of the first to exploit the existence of such algebro-geometric cell decompositions to produce motivic decompositions of varieties ([Kar00]), in his case absolute Chow motives (over a base field). Such results have been developed in numerous directions, but we will mainly be concerned with the relative version studied in [MNP13, 8.4.2]. We begin by introducing our own version of "cellularity". Definition 5.3.1. Suppose f : X → B be a morphism of schemes. Say that f (or X/B) admits a flat (resp. lci) cellular structure if there exists a filtration: ∅ = X−1 ⊂ X0 ⊂ . . . ⊂ Xn ⊂ Xn+1 = X of closed subschemes of X such that for all α, the following conditions are satisfied: (a) fα = f Xα : Xα → B is flat (resp. lci); (b) setting Uα := Xα − Xα−1, the restriction pα = f Uα : Uα → S is a smooth morphism and stable A1-weak equivalence (eg: see Remark 5.1.3). Remark 5.3.2. In Karpenko's definition, the morphisms pα are assumed to be vector bundles, which is sufficient for his purposes since he essentially uses the method of Bialynicki-Birula, where the geometric decomposition arises from the action of a split torus on a smooth projective variety. In contrast, our definition allows, e.g., cellular spaces where the "cells" are merely A1-contractible and we allow ourselves to consider varieties that are not necessarily projective. Examples show that this additional generality is natural. For example, the smooth affine quadrics Q2n of dimension 2n (the hypersurface in affine space defined by the equation Pi xiyi = z(1 − z)) admit a decomposition as an affine space (x1 = · · · = xn = THE HOMOTOPY LERAY SPECTRAL SEQUENCE 31 z = 0) of dimension n, and an open complement that is a strictly quasi-affine A1-contractible scheme [ADF17, Theorems 2.2.5 and 3.1.1] (in particular the complement is an A1-contractible variety that is not isomorphic to affine space). Likewise, the Panin -- Walter model of quaternionic projective space HPn admits a "cell decomposition" where the cells are strictly quasi-affine A1-contractible schemes [PW10, Theorem 3.1]. Since the morphism pα is smooth, its relative dimension is a Zariski locally constant function on Uα that we denote by dα. In other words, dα simply consists of integers for each connected component of Uα. To state the next results, we will use the following notation: (5.3.2.a) 1B(dα)[2dα] = Mx∈U (0) α 1B(dα(x))[2dα(x)]. 5.3.3. Let us first recall the computation obtained in [MNP13, 8.4.3], for Chow motives. We work over a quasi-projective base B over a perfect field k. Taking into account the comparison result of Jin [Jin16, Th. 3.17], the category of relative Chow motives defined by Corti-Hanamura corresponds to the weight 0 part of the triangulated category of rational mixed motives DM(B, Q). Given a projective morphism f : X → B such that X is smooth over k, one defines the Chow motive (i.e. cohomological motive) of X/B as: With this notation, [MNP13, Prop. 8.4.3] can be stated as follows. hB(X) = f∗(1X ). Proposition 5.3.4. Assume f : X → B is a projective morphism of k-varieties, with X smooth and admitting a flat relative cellular structure. There is a canonical isomorphism (using notation (5.3.2.a)) of the form: hB(X) =Mα 1B(−dα)[−2dα]. Note [MNP13, 8.4.3] is stated for k = C but this assumption is not used in the proof, which works over an arbitrary base field k. We appeal to the assumption k perfect implicitly via the comparison result (i.e., [Jin16]). 5.3.5. We can actually extend the previous computation to A1-homotopy, if one restricts to oriented spectra. Let us adopt the following definition. We will say that the abstract triangulated motivic category T is oriented if: • There exists a premotivic adjunction ϕ∗ : SH ⇆ T : ϕ∗; • The ring spectrum ϕ∗(1S) is oriented. These assumptions in place, we will use the theory of Borel-Moore objects as described in [BD17, section 1.3], together with the theory of fundamental classes as developed in [D´eg18] (see also [DJK18] for a more general account). For the convenience of the reader, we recall the definitions and properties we use. Given any separated B-scheme of finite type π : Y → B, we define its Borel-Moore T -motive 13 by the formula: We will need the following properties: MBM(Y /B) := π!(1Y ). (P1) According to the localization property of triangulated motivic categories, given any closed im- mersion i : Z → X of B-schemes with complementary open immersion j : U → X, one obtains a distinguished triangle: MBM(U/B) j∗−→ MBM(X/B) i∗ −→ MBM(Z/B) → MBM(U/B)[1]. (P2) Using the main construction of [D´eg18, Th. 2.5.3], one associates to any quasi-projective mor- phism f : Y → X of relative dimension d, a morphism (also called a fundamental class in loc. cit.): f! : MBM(Y /X) = f!(1Y ) → 1X (−d)[−2d]. 13Another appropriate terminology would be cohomological T -motive with compact support. 32 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL (P3) If in addition to the assumptions of the previous point, one suppose that f is smooth, then the following diagram is commutative: f!(1Y ) pf ∼ f! ,❨❨❨❨❨❨❨❨❨❨❨❨❨❨❨ 2❡❡❡❡❡❡❡❡❡❡❡ f♯(1Y )(−d)[−2d] f∗(−d)[−2d] 1X (−d)[−2d] where pf is the purity isomorphism associated with the smooth morphism f in the six functors formalism, f♯(1Y ) is the T -motive represented by Y /B, and f∗ refers to the classical functori- ality. Note in particular that f∗ is the image of the canonical map Σ∞Y+ → Σ∞B+ of spectra over B under the map ϕ∗ : SH(B) → T (B). (P4) If under the assumptions of point (2), X is separated of finite type over a scheme B, with structural morphism p, we obtain by applying p! a Gysin morphism: f! : MBM(Y /B) → MBM(X/B)(−d)[−2d]. It follows from the compatibility of fundamental classes with composition [D´eg18, 2.5.3] that these Gysin morphism are compatible with composition. Proposition 5.3.6. We consider a morphism f : X → B with an lci cellular structure, and B an arbitrary scheme. Then, using notation (5.3.2.a), there exists a canonical isomorphism in MGL−mod(B): Proof. We look at the localization triangle of property (P1) associated with the closed immersion iα : Xα : Xα−1 → Xα with complementary open immersion jα : Uα → Xα: f!(1X ) =Mα 1B(−dα)[−2dα] MBM(Uα/B) jα∗−−→ MBM(Xα/B) i∗ α−→ MBM(Xα−1/B) → MBM(Uα/B)[1]. Next we apply property (P4) to get a commutative diagram: MBM(Uα/B) jα∗ / MBM(Xα/B) 'PPPPPPPPPPP pα∗ v♥♥♥♥♥♥♥♥♥♥♥ fα∗ MBM(B/B)(−dα)[−2dα] Point (P3) applied to the smooth morphism pα, which by assumption is a stable weak A1-equivalence, implies the map pα∗ is an isomorphism. In particular, jα∗ is a split monomorphism (with spliting p−1 α∗ fα∗) and one gets an isomorphism MBM(Uα/B) = 1B(−dα)[−2dα]. This completes the proof. (cid:3) Corollary 5.3.7. Assume that f : X → B is a projective morphism with lci cellular structure. (1) Then one gets an isomorphism of the following cohomological T -motives: In particular, one gets an isomorphism of T -cohomology: 1B(−dα)[−2dα]. H ∗−2dα,∗−dα(B, 1B) f∗(1X ) =Mα H ∗∗(X, 1X ) ≃Mα f∗f ∗(E) ≃Mα (2) For any T -spectrum E over B, there exists a canonical isomorphism: 1B(−dα)[−2dα]. Only the second point needs a proof. One simply uses the preceding proposition and the projection formula from the six functors formalism: f∗f ∗(E) ≃ f!(1X ⊗ f ∗(E)) ≃ f!(1X ) ⊗ E. Example 5.3.8. (1) In the motivic case, one gets back the result of Proposition 5.3.4. In fact, with rational coefficients, we have shown that one can get rid of a base field, if one replaces the flatness by lci. (See also next Remark.)   , 2 ' / v THE HOMOTOPY LERAY SPECTRAL SEQUENCE 33 (2) We can apply the preceding result to classical oriented ring spectra, such as MGL and KGL, by using the theory of modules over ring spectra (see [CD09, §7]). The corollary extends a previously known result in [NZ06], from characteristic 0 to the absolute case (without the need of a base field). Remark 5.3.9. (1) The above corollary applies to strict MGL-modules M over B. That is the MGL-module structure on M must be defined at the model category level. This assumption is too strong in many situations. One can avoid it, assuming only that M is a spectrum in SH(B) with a module structure over MGLB. Indeed, it is possible to use the proof of Proposition 5.3.6 after replacing f!(1X ) = f!f ∗(1X ) with the spectrum f!f ∗(M ). One uses the theory of fundamental classes as developed in [DJK18] and the fact that one has Thom isomorphisms for M , using the MGLB-module structure. We leave the details to the reader. (2) The orientation assumptions in the preceding proposition and corollary is essential. Indeed, recall that the Chow-Witt ring does not satisfy the projective bundle formula though it is representable in SH by the Milnor-Witt homotopy module. Nevertheless, one can extend the validity of the above result by considering weaker orientabil- ity conditions. For example, we can consider a symplectically oriented (ring) spectrum M (see [PW18]), provided that we give for each index α a symplectic structure on the tangent bundle of Uα/B. (3) At least in the rational motivic case (which is equivalent to the rational orientable case), it is possible in principle to generalize both the flat and lci case. Indeed, it is visible in the proof that we only need a good theory of fundamental classes for the projections Xα → B. The recent work [Jin18] opens the way to define these fundamental classes for arbitrary finite tor-dimension morphisms, which contains both flat and lci cases. The main tool to do that is to extend Jin's work to a suitable representability theorem for relative K-theory (as defined in [BGI71, IV, 3.3]). The relative cellular space provides us with a situation where the homotopy Leray spectral sequence is particularly simple. Proposition 5.3.10. Let f : X → B be a proper morphism with an lci relative cellular structure. We consider a homotopy module M over B satisfying one of the following conditions: • Homotopical case. -- M is an oriented homotopy module in SH(B). • Motivic case. -- M is a homotopy module in DM(B, R). We consider the homotopy module MX = f !M . Then there exists an isomorphism: d f∗(MX ) ≃ H i δ(f∗MX)[−i] Xi=0 where d is the maximum of the dimension of the fibers of f . Moreover, the homotopy Leray spectral sequence: Ep,q 2 = Ap(B, H q δ (f∗MX)) ⇒ Ap+q(X, MX ) degenerates at E2, and the abutting filtration splits giving an isomorphism: An(X, MX) ≃ n Mp=0 Ap(B, H n−p δ (f∗MX )). References [AD07] A. Asok and B. Doran, On unipotent quotients and some A1-contractible smooth schemes, Int. Math. Res. Pap. IMRP (2007), no. 2, Art. ID rpm005, 51. MR 2335246 27, 28 [AD09] , A1-homotopy groups, excision, and solvable quotients, Adv. Math. 221 (2009), no. 4, 1144 -- 1190. MR 2518635 28 [ADF17] A. Asok, B. Doran, and J. Fasel, Smooth models of motivic spheres and the clutching construction, Int. Math. Res. Not. IMRN (2017), no. 6, 1890 -- 1925. MR 3658186 28, 31 [AF14] A. Asok and J. Fasel, Algebraic vector bundles on spheres, J. Topol. 7 (2014), no. 3, 894 -- 926. MR 3252968 29 [Ara05] D. Arapura, The Leray spectral sequence is motivic, Invent. Math. 160 (2005), no. 3, 567 -- 589. MR 2178703 2 J. Ayoub, Les six op´erations de Grothendieck et le formalisme des cycles ´evanescents dans le monde motivique [Ayo07] i, Ast´erisque, vol. 314, Soc. Math. France, 2007. 3, 7 [BBD82] A.A. Beilinson, J. Bernstein, and P. Deligne, Faisceaux pervers, Ast´erisque 100 (1982), 5 -- 171. 7, 12, 23 [BD17] M. Bondarko and F. D´eglise, Dimensional homotopy t-structures in motivic homotopy theory, Adv. Math. 311 (2017), 91 -- 189. MR 3628213 3, 5, 6, 7, 8, 9, 11, 12, 13, 31 34 ARAVIND ASOK, FR ´ED ´ERIC D ´EGLISE, JAN NAGEL [BGI71] P. Berthelot, A. Grothendieck, and L. Illusie, Th´eorie des intersections et th´eor`eme de Riemann-Roch, Lecture Notes in Mathematics, vol. 225, Springer-Verlag, 1971, S´eminaire de G´eom´etrie Alg´ebrique du Bois -- Marie 1966 -- 67 (SGA 6). 33 S. Bloch and A. Ogus, Gersten's conjecture and the homology of schemes, Ann. Sci. ´Ecole Norm. Sup. (4) 7 (1974), 181 -- 201 (1975). 3, 13, 19, 20 [BO74] [Bon10] M. V. Bondarko, Motivically functorial coniveau spectral sequences; direct summands of cohomology of function fields, Doc. Math. (2010), no. Extra vol.: Andrei A. Suslin sixtieth birthday, 33 -- 117. MR 2804250 20 [BY18] T. Bachmann and M. Yakerson, Towards conservativity of Gm-stabilization, arXiv:1811.01541, November 2018. 16 [ByB73] A. Bia l ynicki Birula, Some theorems on actions of algebraic groups, Ann. of Math. (2) 98 (1973), 480 -- 497. MR 0366940 30 [CD09] D.-C. Cisinski and F. D´eglise, Triangulated categories of mixed motives, Preprint, available at https://arxiv.org/abs/0912.2110 , 2009. 4, 5, 6, 12, 33 [CD12] [CD15] , Mixed weil cohomologies, Adv. in Math. 230 (2012), no. 1, 55 -- 130. 20 , Integral mixed motives in equal characteristics, Doc. Math. (2015), no. Extra volume: Alexander S. Merkurjev's sixtieth birthday, 145 -- 194. 6 [CE99] H. Cartan and S. Eilenberg, Homological algebra, Princeton Landmarks in Mathematics, Princeton Univer- sity Press, Princeton, NJ, 1999, With an appendix by David A. Buchsbaum, Reprint of the 1956 original. MR 1731415 20, 21 [dCM10] M. A. A. de Cataldo and L. Migliorini, The perverse filtration and the Lefschetz hyperplane theorem, Ann. of Math. (2) 171 (2010), no. 3, 2089 -- 2113. MR 2680404 2 [D´eg11] F. D´eglise, Modules homotopiques, Doc. Math. 16 (2011), 411 -- 455. 7, 15, 25 [D´eg12] [D´eg13] [D´eg14a] , Coniveau filtration and motives, Regulators, Contemporary Mathematics, vol. 571, 2012, pp. 51 -- 76. 15 , Orientable homotopy modules, Am. J. of Math. 135 (2013), no. 2, 519 -- 560. 7 , On the homotopy heart of mixed motives, http://deglise.perso.math.cnrs.fr/docs/2014/modhtpb.pdf, 2014. 11 [D´eg14b] , Suite spectrale du coniveau et t-structure homotopique, Ann. Fac. Sci. Toulouse Math. (6) 23 (2014), no. 3, 591 -- 609. 20 [D´eg18] [DF17] , Bivariant theories in motivic stable homotopy, Doc. Math. 23 (2018), 997 -- 1076. 31, 32 F. D´eglise and J. Fasel, The milnor-witt motivic ring spectrum and its associated theories, arXiv:1708.06102, 2017. 6 [DF18] A. Dubouloz and J. Fasel, Families of A1-contractible affine threefolds, Algebr. Geom. 5 (2018), no. 1, 1 -- 14. MR 3734108 27, 28 [DJK18] F. D´eglise, F. Jin, and A. Khan, Fundamental classes in motivic homotopy theory, arXiv:1805.05920v2, 2018. [DK01] 5, 9, 10, 15, 19, 31, 33 J. F. Davis and P. Kirk, Lecture notes in algebraic topology, Graduate Studies in Mathematics, vol. 35, American Mathematical Society, Providence, RI, 2001. MR 1841974 2 [Dou59] A. Douady, La suite spectrale d'Adams: structure multiplicative, exp. 19, S´eminaire Henri Cartan, Ecole Normale Sup´erieure, vol. tome 11, n2, 1958-1959, pp. 1 -- 13. 20, 21, 22 [DPØ18] A. Dubouloz, S. Pauli, and P.-A. Østvaer, A1-contractibility of affine modifications, Preprint, available at https://arxiv.org/abs/1805.08959, 2018. 27, 28 N. Feld, Milnor-witt cycle modules, arXiv:1811.12163, 2018. 11 [Fel18] [GD67] A. Grothendieck and J. Dieudonn´e, ´El´ements de g´eom´etrie alg´ebrique. IV. ´Etude locale des sch´emas et des [Jin16] morphismes de sch´emas IV, Publ. Math. IHES 20, 24, 28, 32 (1964-1967). 6 F. Jin, Borelmoore motivic homology and weight structure on mixed motives, Math. Zeit. 283, Issue 3-4 (2016), 1149 -- 1183. 15, 31 [Jin18] [Kar00] N. A. Karpenko, Cohomology of relative cellular spaces and of isotropic flag varieties, Algebra i Analiz 12 , Algebraic g-theory in motivic homotopy categories, arXiv:1806.03927, 2018. 26, 33 (2000), no. 1, 3 -- 69. MR 1758562 30 [Lev06] M. Levine, Chow's moving lemma and the homotopy coniveau tower, K-Theory 37 (2006), no. 1-2, 129 -- 209. 16 [Lev08] [Lur18] [McC01] J. McCleary, A user's guide to spectral sequences, second ed., Cambridge Studies in Advanced Mathematics, J. Lurie, Higher algebra, http://www.math.harvard.edu/~ lurie/papers/HA.pdf, Sep. 2018. 21 , The homotopy coniveau tower, J. Topol. 1 (2008), no. 1, 217 -- 267. 16 vol. 58, Cambridge University Press, Cambridge, 2001. 13 [MNP13] J. P. Murre, J. Nagel, and C. A. M. Peters, Lectures on the theory of pure motives, University Lecture Series, vol. 61, American Mathematical Society, Providence, RI, 2013. MR 3052734 30, 31 [Mor03] F. Morel, An introduction to A1-homotopy theory, Contemporary Developments in Algebraic K-theory, ICTP Lecture notes, vol. 15, 2003, pp. 357 -- 441. 7 [MV99] F. Morel and V. Voevodsky, A1-homotopy theory of schemes, Inst. Hautes ´Etudes Sci. Publ. Math. (1999), no. 90, 45 -- 143 (2001). MR 1813224 (2002f:14029) 28 [Nor02] M. V. Nori, Constructible sheaves, Algebra, arithmetic and geometry, Part I, II (Mumbai, 2000), Tata Inst. Fund. Res. Stud. Math., vol. 16, Tata Inst. Fund. Res., Bombay, 2002, pp. 471 -- 491. MR 1940678 2 [NZ06] A. Nenashev and K. Zainoulline, Oriented cohomology and motivic decompositions of relative cellular spaces, J. Pure Appl. Algebra 205 (2006), no. 2, 323 -- 340. MR 2203620 33 [Par96] K. H. Paranjape, Some spectral sequences for filtered complexes and applications, J. Algebra 186 (1996), no. 3, [PW10] 793 -- 806. MR 1424593 2 I. Panin and C. Walter, Quaternionic Grassmannians and Pontryagin classes in algebraic geometry, Preprint available at http://arxiv.org/abs/1011.0649 , 2010. 31 THE HOMOTOPY LERAY SPECTRAL SEQUENCE 35 [PW18] I. Panin and C. Walter, Quaternionic Grassmannians and Borel classes in algebraic geometry, arXiv: 1011.0650v2, March 2018. 33 [Ros96] M. Rost, Chow groups with coefficients, Doc. Math. J. (1996), 319 -- 393. 11, 16 [VD74] B. Ju. Veısfeıler and I. V. Dolgacev, Unipotent group schemes over integral rings, Izv. Akad. Nauk SSSR Ser. [Ver96] [wc15] Mat. 38 (1974), 757 -- 799. MR 0376697 27 J.-L. Verdier, Des cat´egories d´eriv´ees des cat´egories ab´eliennes, Ast´erisque (1996), no. 239, xii+253 pp. (1997), With a preface by Luc Illusie, Edited and with a note by Georges Maltsiniotis. MR 1453167 21 Collective work (198 collaborators), The stacks project, version 10/11/2018, 2015. 13 A. Asok, Department of Mathematics, University of Southern California, 3620 S. Vermont Ave., Los Angeles, CA 90089-2532, United States; E-mail address: [email protected] F. D´eglise, IMB - UMR5584, 9 avenue Alain Savary 21000 Dijon Cedex; France E-mail address: [email protected] J. Nagel, IMB - UMR5584, 9 avenue Alain Savary 21000 Dijon Cedex; France E-mail address: [email protected]
1904.04903
2
1904
2019-08-14T18:53:42
Mirror curve of orbifold Hurwitz numbers
[ "math.AG", "math.CO" ]
Edge-contraction operations form an effective tool in various graph enumeration problems, such as counting Grothendieck's dessins d'enfants and simple and double Hurwitz numbers. These counting problems can be solved by a mechanism known as topological recursion, which is a mirror B-model corresponding to these counting problems. We show that for the case of orbifold Hurwitz numbers, the mirror objects, i.e., the spectral curve and the differential forms on it, are constructed solely from the edge-contraction operations of the counting problem in genus $0$ and one marked point. This forms a parallelism with Gromov-Witten theory, where genus 0 Gromov-Witten invariants correspond to mirror B-model holomorphic geometry.
math.AG
math
MIRROR CURVE OF ORBIFOLD HURWITZ NUMBERS OLIVIA DUMITRESCU AND MOTOHICO MULASE Abstract. Edge-contraction operations form an effective tool in various graph enumeration prob- lems, such as counting Grothendieck's dessins d'enfants and simple and double Hurwitz numbers. These counting problems can be solved by a mechanism known as topological recursion, which is a mirror B-model corresponding to these counting problems. We show that for the case of orb- ifold Hurwitz numbers, the mirror objects, i.e., the spectral curve and the differential forms on it, are constructed solely from the edge-contraction operations of the counting problem in genus 0 and one marked point. This forms a parallelism with Gromov-Witten theory, where genus 0 Gromov-Witten invariants correspond to mirror B-model holomorphic geometry. Contents Introduction r-Hurwitz graphs 1. 2. Orbifold Hurwitz numbers as graph enumeration 2.1. Cell graphs 2.2. 2.3. Construction of r-Hurwitz graphs 2.4. The edge-contraction formulas 3. Construction of the mirror spectral curves for orbifold Hurwitz numbers References 1 3 3 6 7 10 12 16 1. Introduction The purpose of the present paper is to identify the mirror B-model objects that enable us to solve certain graph enumeration problems. We consider simple and orbifold Hurwitz numbers, by giving a graph enumeration formulation for these numbers. We then show that the mirror of these counting problems are constructed from the edge-contraction operations of [8] applied to orbifold Hurwitz numbers for the case of genus 0 and one-marked point. Edge-contraction operations provide an effective method for graph enumeration problems. It has been noted in [10] that the Laplace transform of edge-contraction operations on many counting problems corresponds to the topological recursion of [13]. In this paper, we examine the construction of mirror B-models corresponding to the simple and orbifold Hurwitz numbers. In general, enumerative geometry problems, such as computation of Gromov-Witten type invariants, are often solved by studying a corresponding problem on the mirror dual side. The effectiveness of the mirror method relies on complex analysis and holomorphic geometry technique that is available on the mirror B-model side. The question we consider in this paper is the following: Question 1.1. How do we find the mirror of a given enumerative problem? 2010 Mathematics Subject Classification. Primary: 14N35, 81T45, 14N10; Secondary: 53D37, 05A15. Key words and phrases. Topological recursion; ribbon graphs; Hurwitz numbers; mirror curves. The first author is supported by NSF grant DMS1802082. 1 2 O. DUMITRESCU AND M. MULASE We give an answer to this question for a class of graph enumeration problems that are equivalent to counting orbifold Hurwitz numbers. The key is the edge-contraction operations. The base case, or the case for the "moduli space" M0,1, of the edge contraction in the counting problem identifies the mirror dual object, and a universal mechanism of complex analysis, known as the topological recursion of [13], solves the B-model side of the counting problem. The solution is a collection of generating functions of the original counting problem for all genera. Bouchard and Marino [3] conjectured that generating functions for simple Hurwitz num- bers could be calculated by the topological recursion of [13], based on the spectral curve identified as the Lambert curve (1.1) x = ye−y. Here, the notion of spectral curve is the mirror dual object for the counting problem. They arrived at the mirror dual by a consideration of mirror symmetry of open Gromov-Witten invariants of toric Calabi-Yau threefolds [2]. The mirror geometry of a toric Calabi-Yau threefold is completely determined by a plane algebraic curve known as the mirror curve. The Lambert curve (1.1) appears as the infinite framing number limit of the mirror curve of C3. The Hurwitz number conjecture of [3] was then solved in a series of papers by one of the authors [12, 20], using the Lambert curve as a given input. Since conjecture is true, the Lambert curve (1.1) should be the mirror B-model for Hurwitz numbers. But why? In [12, 20], we did not attempt to give any explanation. The emphasis of our current paper is to prove that the mirror dual object is simply a consequence of the M0,1 case of the edge-contraction operation on the original counting problem. The situation is similar to several cases of Gromov-Witten theory, where the mirror is constructed by the genus 0 Gromov-Witten invariants themselves. To illustrate the idea, let us consider the number Td of connected trees consisting of labeled d nodes (or vertices). The initial condition is T1 = 1. The numbers satisfy a recursion relation (1.2) (d − 1)Td = 1 2 ab TaTb. (cid:18)d (cid:19) a (cid:88) a+b=d a,b≥1 ∞(cid:88) d=1 Td (d − 1)! xd. A tree of d nodes has d−1 edges. The left-hand side counts how many ways we can eliminate an edge. When an edge is eliminated, the tree breaks down into two disjoint pieces, one consisting of a labeled nodes, and the other b = d − a labeled nodes. The original tree is restored by connecting one of the a nodes on one side to one of the b nodes on the other side. The equivalence of counting in this elimination process gives (1.2). From the initial value, the recursion formula generates the tree sequence 1, 1, 3, 16, 125, 1296, . . . . We note, however, that (1.2) does not directly give a closed formula for Td. To find one, we introduce a generating function, or a spectral curve (1.3) y = y(x) := (cid:18) (cid:19) In terms of the generating function, (1.2) becomes equivalent to x2 ◦ d dx ◦ 1 x 1 2 d dx y2 ⇐⇒ dx dy (1.4) The initial condition is y(0) = 0 and y(cid:48)(0) = 1, which allows us to solve the differential equation uniquely. Lo and behold, the solution is exactly (1.1). y = = x y . x(1 − y) MIRROR OF ORBIFOLD HURWITZ NUMBERS 3 To find the formula for Td, we need the Lagrange Inversion Formula. Suppose that f (y) is a holomorphic function defined near y = 0, and that f (0) (cid:54)= 0. Then the inverse function of x = y f (y) near x = 0 is given by (1.5) y = (cid:18) d ∞(cid:88) dy k=1 (cid:19)k−1(cid:0)f (y)k(cid:1)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)y=0 xk k! . The proof is elementary and requires only Cauchy's integration formula. Since f (y) = ey in our case, we immediately obtain Cayley's formula Td = dd−2. The point we wish to make here is that the real problem behind the scene is not tree- counting, but simple Hurwitz numbers. This relation is understood by the correspondence between trees and ramified coverings of P1 by P1 of degree d that are simply ramified except for one total ramification point. When we look at the dual graph of a tree, elimination of an edge becomes contracting an edge, and this operation precisely gives a degeneration formula for counting problems on Mg,n. The base case for the counting problem is (g, n) = (0, 1), and the recursion (1.2) is the result of the edge-contraction operation for simple Hurwitz numbers associated with M0,1. In this sense, the Lambert curve (1.1) is the mirror dual of simple Hurwitz numbers. The paper is organized as follows. In Section 2, we present combinatorial graph enu- meration problems, and show that they are equivalent to counting of simple and orbifold Hurwitz numbers. In Section 3, the spectral curves of the topological recursion for simple and orbifold Hurwitz numbers (the mirror objects to the counting problems) are constructed from the edge-contraction formulas for (g, n) = (0, 1) invariants. 2. Orbifold Hurwitz numbers as graph enumeration Mirror symmetry provides an effective tool for counting problems of Gromov-Witten type invariants. The question is how we construct the mirror, given a counting problem. Al- though there is so far no general formalism, we present a systematic procedure for computing orbifold Hurwitz numbers in this paper. The key observation is that the edge-contraction operations for (g, n) = (0, 1) identify the mirror object. The topological recursion for simple and orbifold Hurwitz numbers are derived as the Laplace transform of the cut-and-join equation [1, 12, 20], where the spectral curves are identified by the consideration of mirror symmetry of toric Calabi-Yau orbifolds [1, 3, 14, 15]. In this section we give a purely combinatorial graph enumeration problem that is equivalent to counting orbifold Hurwitz numbers. We then show in the next section that the edge- contraction formula restricted to the (g, n) = (0, 1) case determines the spectral curve and the differential forms W0,1 and W0,2 of [1]. These quantities form the mirror objects for the orbifold Hurwitz numbers. 2.1. Cell graphs. To avoid unnecessary confusion, we use the terminology cell graphs in this article, instead of more common ribbon graphs. Ribbon graphs naturally appear for encoding complex structures of a topological surface (see for example, [17, 18]). Our purpose of using ribbon graphs are for degeneration of stable curves, and we label vertices, instead of faces, of a ribbon graph. Definition 2.1 (Cell graphs). A connected cell graph of topological type (g, n) is the 1-skeleton of a cell-decomposition of a connected closed oriented surface of genus g with n labeled 0-cells. We call a 0-cell a vertex, a 1-cell an edge, and a 2-cell a face, of the cell graph. We denote by Γg,n the set of connected cell graphs of type (g, n). Each edge consists of two half-edges connected at the midpoint of the edge. 4 O. DUMITRESCU AND M. MULASE Remark 2.2. • The dual of a cell graph is a ribbon graph, or Grothendieck's dessin d'enfant. We note that we label vertices of a cell graph, which corresponds to face labeling of a ribbon graph. Ribbon graphs are also called by different names, such as embedded graphs and maps. • We identify two cell graphs if there is a homeomorphism of the surfaces that brings one cell-decomposition to the other, keeping the labeling of 0-cells. The only possible automorphisms of a cell graph come from cyclic rotations of half-edges at each vertex. Definition 2.3 (Directed cell graph). A directed cell graph is a cell graph for which an arrow is assigned to each edge. An arrow is the same as an ordering of the two half-edges forming an edge. The set of directed cell graphs of type (g, n) is denoted by (cid:126)Γg,n. Remark 2.4. A directed cell graph is a quiver. Since our graph is drawn on an oriented surface, a directed cell graph carries more information than its underlying quiver structure. The tail vertex of an arrowed edge is called the source, and the head of the arrow the target, in the quiver language. An effective tool in graph enumeration is edge-contraction operations. Often edge con- traction leads to an inductive formula for counting problems of graphs. Definition 2.5 (Edge-contraction operations). There are two types of edge-contraction operations applied to cell graphs. • ECO 1: Suppose there is a directed edge (cid:126)E = −→ pipi in a cell graph γ ∈ (cid:126)Γg,n, connecting the tail vertex pi and the head vertex pj. We contract (cid:126)E in γ, and put the two vertices pi and pj together. We use i for the label of this new vertex, and call it again pi. Then we have a new cell graph γ(cid:48) ∈ (cid:126)Γg,n−1 with one less vertices. In this process, the topology of the surface on which γ is drawn does not change. Thus genus g of the graph stays the same. Figure 2.1. Edge-contraction operation ECO 1. The edge bounded by two vertices pi and pj is contracted to a single vertex pi. • We use the notation (cid:126)E for the edge-contraction operation (2.1) (cid:126)E : (cid:126)Γg,n (cid:51) γ (cid:55)−→ γ(cid:48) ∈ (cid:126)Γg,n−1. • ECO 2: Suppose there is a directed loop (cid:126)L in γ ∈ (cid:126)Γg,n at the i-th vertex pi. Since a loop in the 1-skeleton of a cell decomposition is a topological cycle on the surface, its contraction inevitably changes the topology of the surface. First we look at the half-edges incident to vertex pi. Locally around pi on the surface, the directed loop (cid:126)L separates the neighborhood of pi into two pieces. Accordingly, we put the incident half-edges into two groups. We then break the vertex pi into two vertices, pi1 and pi2, so that one group of half-edges are incident to pi1, and the other group to pi2. The order of two vertices is determined by placing the loop (cid:126)L upward near at vertex pi. Then we name the new vertex on its left by pi1, and on its right by pi2. pipjpiE MIRROR OF ORBIFOLD HURWITZ NUMBERS 5 Let γ(cid:48) denote the possibly disconnected graph obtained by contracting (cid:126)L and separating the vertex to two distinct vertices labeled by i1 and i2. Figure 2.2. Edge-contraction operation ECO 2. The contracted edge is a loop (cid:126)L of a cell graph. Place the loop so that it is upward near at pi to which (cid:126)L is attached. The vertex pi is then broken into two vertices, pi1 on the left, and pi2 on the right. Half-edges incident to pi are separated into two groups, belonging to two sides of the loop near pi. • If γ(cid:48) is connected, then it is in (cid:126)Γg−1,n+1. The loop (cid:126)L is a loop of handle. We use the same notation (cid:126)L to indicate the edge-contraction operation (cid:126)L : (cid:126)Γg,n (cid:51) γ (cid:55)−→ γ(cid:48) ∈ (cid:126)Γg−1,n+1. • If γ(cid:48) is disconnected, then write γ(cid:48) = (γ1, γ2) ∈ (cid:126)Γg1,I+1 × (cid:126)Γg2,J+1, where (cid:40) I (cid:116) J = {1, . . . ,(cid:98)i, . . . , n} g = g1 + g2 . The edge-contraction operation is again denoted by (cid:126)L : (cid:126)Γg,n (cid:51) γ (cid:55)−→ (γ1, γ2) ∈ (cid:126)Γg1,I+1 × (cid:126)Γg2,J+1. In this case we call (cid:126)L a separating loop. Here, vertices labeled by I belong to the connected component of genus g1, and those labeled by J are on the other component of genus g2. Let (I−, i, I+) (reps. (J−, i, J+)) be the reordering of I (cid:116){i} (resp. J (cid:116) {i}) in the increasing order. Although we give labeling i1, i2 to the two vertices created by breaking pi, since they belong to distinct graphs, we can simply use i for the label of pi1 ∈ γ1 and the same i for pi2 ∈ γ2. The arrow of (cid:126)L translates into the information of ordering among the two vertices pi1 and pi2. (2.2) (2.3) (2.4) Remark 2.6. The use of directed cell graphs enables us to define edge-contraction oper- ations, keeping track with vertex labeling. We refer to [?] for the actual motivation for quiver cell graphs. Since our main concern is enumeration of graphs, the extra data of directed edges does not plan any role. In what follows, we deal with cell graphs without directed edges. The edge-contraction operations are defined with a choice of direction, but the counting formula we derive does not depend of this choice. Remark 2.7. Let us define m(γ) = 2g − 2 + n for a graph γ ∈ Γg,n. Then every edge- contraction operation reduces m(γ) exactly by 1. Indeed, for ECO 1, we have m(γ(cid:48)) = 2g − 2 + (n − 1) = m(γ) − 1. The ECO 2 applied to a loop of handle produces m(γ(cid:48)) = 2(g − 1) − 2 + (n + 1) = m(γ) − 1. pi1pi2piL 6 O. DUMITRESCU AND M. MULASE For a separating loop, we have 2g1 − 2 + I + 1 2g2 − 2 + J + 1 +) 2g1 + 2g2 − 4 + I + J + 2 = 2g − 2 + n − 1. 2.2. r-Hurwitz graphs. We choose and fix a positive integer r. The decorated graphs we wish to enumerate are the following. Definition 2.8 (r-Hurwitz graph). An r-Hurwitz graph (γ, D) of type (g, n, d) consists of the following data. • γ is a connected cell graph of type (g, n), with n labeled vertices. • D = d is divisible by r, and γ has m = d/r unlabeled faces and s unlabeled edges, where (2.5) s = 2g − 2 + d r + n. • D is a configuration of d = rm unlabeled dots on the graph subject to the following conditions: (1) The set of d dots are grouped into m subsets of r dots, each of which is equipped with a cyclic order. (2) Every face of γ has cyclically ordered r dots. (3) These dots are clustered near vertices of the face. At each corner of the face, say at Vertex i, the dots are ordered according to the cyclic order that is consistent of the orientation of the face, which is chosen to be counter-clock wise. (4) Let µi denote the total number of dots clustered at Vertex i. Then µi > 0 for every i = 1, . . . , n. Thus we have an ordered partition In particular, the number of vertices ranges 0 < n ≤ d. d = µ1 + ··· + µn. (5) Suppose an edge E connecting two distinct vertices, say Vertex i and j, bounds the same face twice. Let p be the midpoint of E. The polygon representing the face has E twice on its perimeter, hence the point p appears also twice. We name them as p and p(cid:48). Which one we call p or p(cid:48) does not matter. Consider a path on the perimeter of this polygon starting from p and ending up with p(cid:48) according to the counter-clock wise orientation. Let r(cid:48) be the total number of dots clustered around vertices of the face, counted along the path. Then it satisfies 0 < r(cid:48) < r. (2.6) (2.7) For example, not all r dots of a face can be clustered at a vertex of degree 1. In particular, for the case of r = 1, the graph γ has no edges bounding the same face twice. An arrowed r-Hurwitz graph (γ, (cid:126)D) has, in addition to to the above data (γ, D), an arrow assigned to one of the µi dots from Vertex i for each index 1 ≤ i ≤ n. The counting problem we wish to study is the number Hr g,n(µ1 . . . , µn) of arrowed r- Hurwitz graphs for a prescribed ordered partition (2.6), counted with the automorphism weight. The combinatorial data corresponds to an object in algebraic geometry. Let us first identify what the r-Hurwitz graphs represent. We denote by P1[r] the 1-dimensional orbifold modeled on P1 that has one stacky point(cid:2)0(cid:14)(cid:0)Z/(r)(cid:1)(cid:3) at 0 ∈ P1. MIRROR OF ORBIFOLD HURWITZ NUMBERS 7 Example 2.9. The base case is Hr morphism P1[r] ∼−→ P1[r]. 0,1(r) = 1 (see Figure 2.3). This counts the identity Figure 2.3. The graph has only one vertex and no edges. All r dots are clustered around this unique vertex, with an arrow attached to one of them. Because of the arrow, there is no automorphism of this graph. Definition 2.10 (Orbifold Hurwitz cover and Orbifold Hurwitz numbers). An orbifold Hurwitz cover f : C −→ P1[r] is a morphism from an orbifold C that is modeled on a smooth algebraic curve of genus g that has to(cid:2)0(cid:14)(cid:0)Z/(r)(cid:1)(cid:3) ∈ P1[r], (1) m stacky points of the same type as the one on the base curve that are all mapped (2) arbitrary profile (µ1, . . . , µn) with n labeled points over ∞ ∈ P1[r], (3) and all other ramification points are simple. If we replace the target orbifold by P1, then the morphism is a regular map from a smooth r, . . . , r) over 0 ∈ P1, labeled profile (µ1, . . . , µn) over ∞ ∈ curve of genus g with profile ( P1, and a simple ramification at any other ramification point. The Euler characteristic condition (2.5) of the graph γ gives the number of simple ramification points of f through the Riemann-Hurwitz formula. The automorphism weighted count of the number of the topological types of such covers is denoted by H r g,n(µ1, . . . , µn). These numbers are referred to as orbifold Hurwitz numbers. When r = 1, they count the usual simple Hurwitz numbers. m(cid:122) (cid:125)(cid:124) (cid:123) The counting of the topological types is the same as counting actual orbifold Hurwitz covers such that all simple ramification points are mapped to one of the s-th roots of unity ξ1, . . . , ξs, where ξ = exp(2πi/s), if all simple ramification points of f are labeled. Indeed, such a labeling is given by elements of the cyclic group {ξ1, . . . , ξs} of order s. Let us construct an edge-labeled Hurwitz graph from an orbifold Hurwitz cover with fixed branch points on the target as above. We first review the case of r = 1, i.e., the simple Hurwitz covers. Our graph is essentially the same as the dual of the branching graph of [21]. 2.3. Construction of r-Hurwitz graphs. First we consider the case r = 1. Let f : C −→ P1 be a simple Hurwitz cover of genus g and degree d with labeled profile (µi, . . . , µn) over ∞, unramified over 0 ∈ P1, and simply ramified over B = {ξ1, . . . , ξs} ⊂ P1, where ξ = exp(2πi/s) and s = 2g − 2 + d + n. We denote by R = {p1, . . . , ps} ⊂ C the labeled simple ramification points of f , that is bijectively mapped to B by f : R −→ B. We choose a labeling of R so that f (pα) = ξα for every α = 1, . . . , s. On P1, plot B and connect each element ξα ∈ B with 0 by a straight line segment. We also connect 0 and ∞ by a straight line z = t exp(πi/s), 0 ≤ t ≤ ∞. Let ∗ denote the configuration of the s line segments. The inverse image f−1(∗) is a cell graph on C, for which f−1(0) forms the set of vertices. We remove all inverse images f−1(0ξα) of the line segment 0ξα from this graph, except for the ones that end at one of the points pα ∈ R. Since pα is a simple ramification point of f , the line segment ending at pα extends to another vertex, i.e., another point in f−1(0). We denote by γ∨ the graph after this removal of line 8 O. DUMITRESCU AND M. MULASE segments. We define the edges of the graph to be the connected line segments at pα for some α. We use pα as the label of the edge. The graph γ∨ has d vertices, s edges, and n faces. An inverse image of the line 0∞ is a ray starting at a vertex of the graph γ∨ and ending up with one of the points in f−1(∞), which is the center of a face. We place a dot on this line near at each vertex. The edges of γ∨ incident to a vertex are cyclically ordered counter-clockwise, following the natural cyclic order of B. Let pα be an edge incident to a vertex, and pβ the next one at the same vertex according to the cyclic order. We denote by dαβ the number of dots in the span of two edges pα and pβ, which is 0 if α < β, and 1 if β < α. Now we consider the dual graph γ of γ∨. It has n vertices, d faces, and s edges still labeled by {p1, . . . , ps}. At the angled corner between the two adjacent edges labeled by pα and pβ in this order according to the cyclic order, we place dαβ dots. The data (γ, D) consisting of the cell graph γ and the dot configuration D is the Hurwitz graph corresponding to the simple Hurwitz cover f : C −→ P1 for r = 1. It is obvious that what we obtain is an r = 1 Hurwitz graph, except for the condition (5) of the configuration D, which requires an explanation. The dual graph γ∨ for r = 1 is the branching graph of [21]. Since B = s is the number of simple ramification points, which is also the number of edges of γ∨, the branching graph cannot have any loops. This is because two distinct powers of ξ in the range of 1, . . . , s cannot be the same. This fact reflects in the condition that γ has no edge that bounds the same face twice. This explains the condition (5) for r = 1. Remark 2.11. If we consider the case r = 1, g = 0 and n = 1, then s = d − 1. Hence the graph γ∨ is a connected tree consisting of d nodes (vertices) and d − 1 labeled edges. Except for d = 1, 2, every vertex is uniquely labeled by incident edges. The tree counting of Introduction is relevant to Hurwitz numbers in this way. rm with labeled profile (µi, . . . , µn) over ∞, m isomorphic stacky points over(cid:2)0(cid:14)(cid:0)Z/(r)(cid:1)(cid:3) ∈ Now let us consider an orbifold Hurwitz cover f : C −→ P1[r] of genus g and degree d = P1[r], and simply ramified over B = {ξ1, . . . , ξs} ⊂ P1[r], where s = 2g − 2 + m + n. By R = {p1, . . . , ps} ⊂ C we indicate the labeled simple ramification points of f , that is again bijectively mapped to B by f : R −→ B. We choose the same labeling of R so that f (pα) = ξα for every α = 1, . . . , s. On P1[r], plot B and connect each element ξα ∈ B with the stacky point at 0 by a straight line segment. We also connect 0 and ∞ by a straight line z = t exp(πi/s), 0 ≤ t ≤ ∞, as before. Let ∗ denote the configuration of the s line segments. The inverse image f−1(∗) is a cell graph on C, for which f−1(0) forms the set of vertices. We remove all inverse images f−1(0ξα) of the line segment 0ξα from this graph, except for the ones that end at one of the points pα ∈ R. We denote by γ∨ the graph after this removal of line segments. We define the edges of the graph to be the connected line segments at pα for some α. We use pα as the label of the edge. The graph γ∨ has m vertices, s edges. The inverse image of the line 0∞ form a set of r rays at each vertex of the graph γ∨, connecting m vertices and n centers f−1(∞) of faces. We place a dot on each line near at each vertex. These dots are cyclically ordered according to the orientation of C, which we choose to be counter-clock wise. The edges of γ∨ incident to a vertex are also cyclically ordered in the same way. Let pα be an edge incident to this vertex, and pβ the next one according to the cyclic order. We denote by dαβ the number of dots in the span of two edges pα and pβ. Let γ denote the dual graph of γ∨. It now has n vertices, m faces, and s edges still labeled by {p1, . . . , ps}. At the angled corner between the two adjacent edges labeled MIRROR OF ORBIFOLD HURWITZ NUMBERS 9 by pα and pβ in this order according to the cyclic order, we place dαβ dots, again cyclically ordered as on γ∨. The data (γ, D) consisting of the cell graph γ and the dot configuration D is the r-Hurwitz graph corresponding to the orbifold Hurwitz cover f : C −→ P1[r]. We note that γ∨ can have loops, unlike the case of r = 1. Let us place γ∨ locally on an oriented plane around a vertex. The plane is locally separated into r sectors by the r rays f−1(0∞) at this vertex. There are s half-edges coming out of the vertex at each of these r sectors. A half-edge corresponding to ξα cannot be connected to another half-edge corresponding to ξβ in the same sector, by the same reason for the case of r = 1. But it can be connected to another half-edge of a different sector corresponding again to the same ξα. In this case, within the loop there are some dots, representing the rays of f−1(0∞) in between these half-edges. The total number of dots in the loop cannot be r, because then the half-edges being connected are in the same sector. Thus the condition (5) is satisfied. Example 2.12. Theorem 2.15 below shows that This is the weighted count of the number of 2-Hurwitz graphs of type (g, n, d) = (0, 2, 4) with an ordered partition 4 = 3 + 1. H2 0,2(3, 1) = 9 2 . Figure 2.4. Hurwitz covers counted in H2 simple ramification points, and one ramification point of degree 3. 0,2(3, 1) have two orbifolds points, two Figure 2.5. There are two 2-Hurwitz graphs. The number of graphs is 3/2 for the graph on the left counting the automorphism, and 3 for the one on the right. The total is thus 9/2. In terms of formulas, the 2-Hurwitz cover corresponding to the graph on the left of Figure 2.5 is given by √ To make the simple ramification points sit on ±1, we need to divide f (x) by f (i/ x = ±1/ graph on the right of Figure 2.5 is given by 3), where 3 are the simple ramification points. The 2-Hurwitz cover corresponding to the √ x (x − 1)2(x + 1)2 f (x) = f (x) = (x − 1)2(x + 1)2 x − a . , 10 O. DUMITRESCU AND M. MULASE 3/2. The real parameter a changes the topological √ 3 2 , the graph is the same as on the left, and 3 2 < a < √ where a is a real number satisfying a > type of the 2-Hurwitz cover. For −√ for a > 2.4. The edge-contraction formulas. √ 3 2 , the graph becomes the one on the right. Definition 2.13 (Edge-contraction operations). The edge-contraction operations (ECOs) on an arrowed r-Hurwitz graph (γ, (cid:126)D) are the following procedures. Choose an edge E of the cell graph γ. • ECO 1: We consider the case that E is an edge connecting two distinct vertices Vertex i and Vertex j. We can assume i < j, which induces a direction i E−→ j on E. Let us denote by F+ and F− the faces bounded by E, where F+ is on the left side of E with respect to the direction. We now contract E, with the following additional operations: (1) Remove the original arrows at Vertices i and j. (2) Put the dots on F± clustered at Vertices i and j together, keeping the cyclic order of the dots on each of F±. (3) Place a new arrow to the largest dot on the corner at Vertex i of Face F+ with respect to the cyclic order. (4) If there are no dots on this particular corner, then place an arrow to the first dot we encounter according to the counter-clock wise rotation from E and centered at Vertex i. The new arrow at the joined vertex allows us to recover the original graph from the new one. Figure 2.6. After contracting the edge, a new arrow is placed on the dot that is the largest (according to the cyclic order) around Vertex i in the original graph, and on the face incident to E which is on the left of E with respect to the direction i → j. The new arrow tells us where the break is made in the original graph. If there are no dots on this particular face, then we go around Vertex i counter-clock wise and find the first dot in the original graph. We place an arrow to this dot in the new graph after contracting E. Here again the purpose is to identify which of the µi dots come from the original Vertex i • ECO 2: This time E is a loop incident to Vertex i twice. We contract E and separate the vertex into two new ones, as in ECA 3 of Definition ??. The additional operations are: (1) The contraction of a loop does not change the number of faces. Separate the (2) Look at the new vertex to which the original arrow is placed. We keep the same dots clustered at Vertex i according to the original configuration. name i to this vertex. The other vertex is named i(cid:48). (3) Place a new arrow to the dot on the corner at the new Vertex i that was the largest in the original corner with respect to the cyclic order. MIRROR OF ORBIFOLD HURWITZ NUMBERS 11 (4) If there are no dots on this particular corner, then place an arrow to the first dot we encounter according to the counter-clock wise rotation from E and centered at Vertex i on the side of the old arrow. (5) We do the same operation for the new Vertex i(cid:48), and put a new arrow to a dot. (6) Now remove the original arrow. Figure 2.7. New arrows are placed so that the original graph can be recovered from the new one Although cumbersome, it is easy to show that Lemma 2.14. The edge-contraction operations preserve the set of r-Hurwitz graphs. An application of the edge-contraction operations is the following counting recursion formula. Theorem 2.15 (Edge-Contraction Formula). The number of arrowed Hurwitz graphs sat- isfy the following edge-contraction formula. (cid:18) 2g − 2 + Hr g,n(µ1 . . . , µn) (cid:19) g,n−1(µ1, . . . , µi−1, µi + µj, µi+1, . . . ,(cid:99)µj, . . . , µn) Hr (cid:88) g−1,n+1(α, β, µ1, . . . ,(cid:98)µi, . . . , µn)  . Hr g1,I+1(α, µI )Hr g2,J+1(β, µJ ) d + n r µiµjHr (cid:88) n(cid:88) i<j 1 2 µi (cid:88) i=1 α+β=µi α,β≥1 (2.8) = + + Here,(cid:98) indicates the omission of the index, and µI = (µi)i∈I for any subset I ⊂ {1, 2, . . . , n}. I(cid:116)J={1,...,i,...,n} g1+g2=g Remark 2.16. The edge-contraction formula (ECF) is a recursion with respect to the number of edges s = 2g − 2 + µ1 + ··· + µn + n. r g,n(µ1 . . . , µn) from the base case Hr Therefore, it calculates all values of Hr 0,1(r). However, it does not determine the initial value itself, since s = 0. We also note that the recursion is not for Hr Proof. The counting is done by applying the edge-contraction operations. The left-hand side of (2.8) shows the choice of an edge, say E, out of s = 2g − 2 + d r + n edges. The first line of the right-hand side corresponds to the case that the chosen edge E connects Vertex g,n as a function in n integer variables. 12 O. DUMITRESCU AND M. MULASE i and Vertex j. We assume i < j, and apply ECO 1. The factor µiµj indicates the removal of two arrows at these vertices (Figure 2.6). When the edge E we have chosen is a loop incident to Vertex i twice, then we apply ECO 2. The factor µi is the removal of the original arrow (Figure 2.7). The second and third lines on the right-hand side correspond whether E is a handle-cutting loop, or a separation loop. The factor 1 2 is there because of the symmetry between α and β of the (cid:3) partition of µi. This complete the proof. Theorem 2.17 (Graph enumeration and orbifold Hurwitz numbers). The graph enumera- tion and counting orbifold Hurwitz number are related by the following formula: (2.9) Hr g,n(µi, . . . , µn) = µ1µ2 ··· µnH r g,n(µi, . . . , µn). Proof. The simplest orbifold Hurwitz number is H r 0,1(r), which counts double Hurwitz num- bers with the same profile (r) at both 0 ∈ P1 and ∞ ∈ P1. There is only one such map f : P1 −→ P1, which is given by f (x) = xr. Since the map has automorphism Z/(r), we have H r We notice that (2.8) is exactly the same as the cut-and-join equation of [1, Theorem 2.2], after modifying the orbifold Hurwitz numbers by multiplying µ1 ··· µn. Since the initial value is the same, and the formulas are recursion based on s = 2g − 2 + d r + n, (2.9) holds (cid:3) by induction. This completes the proof. 0,1(r) = 1/r. Thus (2.9) holds for the base case. 3. Construction of the mirror spectral curves for orbifold Hurwitz numbers In the earlier work on simple and orbifold Hurwitz numbers in connection to the topo- logical recursion [1, 3, 5, 12, 20], the spectral curves are determined by the infinite framing limit of the mirror curves to toric Calabi-Yau (orbi-)threefolds. The other ingredients of the topological recursion, the differential forms W0,1 and W0,2, are calculated by the Laplace transform of the (g, n) = (0, 1) and (0, 2) cases of the ELSV [11] and JPT [16] formulas. Certainly the logic is clear, but why these choices are the right ones is not well explained. In this section, we show that the edge-contraction operations themselves determine all the mirror ingredients, i.e., the spectral curve, W0,1, and W0,2. The structure of the story is the following. The edge-contraction formula (2.8) is an equation among different values of (g, n). When restricted to (g, n) = (0, 1), it produces an equation on Hr 0,1(d) as a function in one integer variable. The generating function of Hr g,n(µ1, . . . , µn) is reasonably complicated, but it can be expressed rather nicely in terms of the generating function of the (0, 1)-values Hr 0,1(d), which is essentially the spectral curve of the theory. The edge- contraction formula (2.8) itself has the Laplace transform that can be calculated in the spectral curve coordinate. Since (2.8) contains (g, n) on each side of the equation, to make it a genuine recursion formula for functions with respect to 2g − 2 + n in the stable range, we need to calculate the generating functions of Hr 0,2(µ1, µ2), and make the rest of (2.8) free of unstable terms. The result is the topological recursion of [1, 12]. Let us now start with the restricted (2.8) on (0, 1) invariants: (cid:19) (cid:18) d r − 1 Hr 0,1(d) = 1 2 d 0,1(d) and Hr (cid:88) Hr 0,1(α)Hr α+β=d α,β≥1 0,1(β). (3.1) MIRROR OF ORBIFOLD HURWITZ NUMBERS 13 At this stage, we introduce a generating function Hr 0,1(d)xd. y = y(x) = (3.2) ∞(cid:88) (cid:19) d=1 (cid:18) xr+1 ◦ d dx ◦ 1 xr y = 1 2 rx d dx y2, In terms of this generating function, (3.1) is a differential equation (3.3) or simply Its unique solution is − ry(cid:48) = y(cid:48) r y x Cxr = ye−ry . (cid:40) 0 1 1 ≤ d < r; d = r, with a constant of integration C. As we noted in the previous section, the recursion (2.8) does not determine the initial value Hr 0,1(d). For our graph enumeration problem, the values are (3.4) Hr 0,1(d) = which determine C = 1. Thus we find (3.5) xr = ye−ry, which is the r-Lambert curve of [1]. This is indeed the spectral curve for the orbifold Hurwitz numbers. Remark 3.1. We note that rHr 0,1(rm) satisfies the same recursion equation (3.1) for r = 1, with a different initial value. Thus essentially orbifold Hurwitz numbers are determined by the usual simple Hurwitz numbers. Remark 3.2. If we define Td = (d − 1)!Hr=1 0,1 (d), then (3.1) for r = 1 is equivalent to (1.2). This is the reason we consider the tree recursion as the spectral curve for simple and orbifold Hurwitz numbers. For the purpose of performing analysis on the spectral curve (3.5), let us introduce a global coordinate z on the r-Lambert curve, which is an analytic curve of genus 0: (3.6) We denote by Σ ⊂ C2 this parametric curve. Let us introduce the generating functions of general Hr g,n, which are called free energies: (3.7) Fg,n(x1, . . . , xn) := 1 µ1 ··· µn Hr g,n(µ1, . . . , µn) xµi i . n(cid:89) i=1 (cid:40) x = x(z) := ze−zr y = y(z) := zr. (cid:88) µ1,...,µn≥1 We also define the exterior derivative (3.8) Wg,n(x1, . . . , xn) := d1 ··· dnFg,n(x1, . . . , xn), which is a symmetric n-linear differential form. By definition, we have (3.9) y = y(x) = x d dx F0,1(x). 14 O. DUMITRESCU AND M. MULASE The topological recursion requires the spectral curve, W0,1, and W0,2. From (3.8) and (3.9), we have (3.10) W0,1(x) = y dx x = yd log(x). Remark 3.3. For many examples of topological recursion such as ones considered in [10], we often define W0,1 = ydx, which is a holomorphic 1-form on the spectral curve. For Hurwitz theory, due to (3.9), it is more natural to use (3.10). As a differential equation, we can solve (3.9) in a closed formula on the spectral curve Σ of (3.6). Indeed, the role of the spectral curve is that the free energies, i.e., Fg,n's, are actually analytic functions defined on Σn. Although we define Fg,n's as a formal power series in (x1, . . . , xn) as generating functions, they are analytic, and the domain of analyticity, or the classical sense of Riemann surface, is the spectral curve Σ. The coordinate change (3.6) gives us (3.11) hence (3.9) is equivalent to x d dx = z 1 − rzr d dz , zr−1(1 − rzr) = Since z = 0 =⇒ x = 0 =⇒ F0,1(x) = 0, we find 1 r (cid:0)x(z)(cid:1) = (3.12) F0,1 (cid:0)x(z)(cid:1). d dz F0,1 zr − 1 2 z2r. The calculation of F0,2 is done similarly, by restricting (2.8) to the (g, n) = (0, 1) and (0, 2) terms. Assuming that µ1 + µn = mr, we have (cid:18) d (cid:19) (3.13) − 1 = µ1µ2Hr r Hr 0,2(µ1, µ2) 0,1(µ1 + µ2) + µ1 (cid:88) α+β=µ1 α,β>0 Hr 0,1(α)Hr 0,2(β, µ2) + µ2 (cid:88) α+β=µ2 α,β>0 Hr 0,1(α)Hr 0,2(µ1, β). As a special case of [1, Lemma 4.1], this equation translates into a differential equation for F0,2: (cid:18) (3.14) 1 r = + 1 (cid:18) x1 − x2 ∂ ∂x1 x1 (cid:18) x1 + x2 F0,2(x1, x2) ∂ ∂x1 (cid:18) x2 1 ∂ ∂x2 F0,1(x1) − x2 2 ∂ ∂x1 (cid:19)(cid:18) (cid:19) x1 ∂ ∂x1 (cid:19) ∂ ∂x2 (cid:18) − (cid:19) F0,1(x2) (cid:19) (cid:18) (cid:0)x(z1), x(z2)(cid:1) = x2 + F0,1(x1) F0,2(x1, x2) F0,1(x2) x2 F0,2(x1, x2) . Denoting by xi = x(zi) and using (3.11), (3.14) becomes simply 1 − x2zr x1 − x2 ∂ ∂z2 ∂ ∂z1 (3.15) x1zr + z2 F0,2 1 r z1 2 − (zr 1 + zr 2) on the spectral curve Σ. This is a linear partial differential equation of the first order with analytic coefficients in the neighborhood of (0, 0) ∈ C2, hence by the Cauchy-Kovalevskaya (cid:19) (cid:19) F0,1(x2) ∂ ∂x2 x1 ∂ ∂x1 ∂ ∂x2 F0,1(x1) + x2 (cid:19)(cid:18) ∂ ∂x2 Proof. We first note that log solution to (3.15) is a straightforward calculation that can be verified as follows: x(z1)−x(z2) is holomorphic around (0, 0) ∈ C2. (3.16) being a z1−z2 (cid:0)x(z1), x(z2)(cid:1) = log (cid:19) − z1e−zr ∂ ∂z2 (cid:18) + z2 ∂ z1 ∂z1 z1 − z2 z1 − z2 = 1 − x1 − x2 (cid:0)x(0), x(z2)(cid:1) = log ezr x1 − x2 2 − zr = z1 − z2 x(z1) − x(z2) log 1 (1 − rzr 1) − z2e−zr x1 − x2 = r 2 x1zr 2 (1 − rzr 2) 1 − x2zr x1 − x2 2 . x1zr 1 − x2zr x1 − x2 + r MIRROR OF ORBIFOLD HURWITZ NUMBERS 15 theorem, it has the unique analytic solution around the origin of C2 for any Cauchy problem. Since the only analytic solution to the homogeneous equation (cid:18) (cid:19) z1 ∂ ∂z1 + z2 ∂ ∂z2 f (z1, z2) = 0 is a constant, the initial condition F0,2(0, x2) = F0,2(x1, 0) = 0 determines the unique solution of (3.15). Proposition 3.4. We have a closed formula for F0,2 in the z-coordinates: (3.16) F0,2 z1 − z2 x(z1) − x(z2) − (zr 1 + zr 2). Since F0,2 2 = 0, (3.16) is the desired unique solution. (cid:3) In [1], the functions (3.12) and (3.16) are derived by directly computing the Laplace transform of the JPT formulas [16] H r 0,1(d) = (3.17) H r 0,2(µ1, µ2) = d(cid:98) d r (cid:99)−2 (cid:98) d r(cid:99)! , r(cid:104) µ1 0 r (cid:105)+(cid:104) µ1 r (cid:105) 1 µ1+µ2 (cid:98) µ2 r (cid:99) r (cid:99) (cid:98) µ1 µ µ 1 2 (cid:98) µ1 r (cid:99)!(cid:98) µ2 r (cid:99)! µ1 + µ2 ≡ 0 mod r otherwise. Here, q = (cid:98)q(cid:99) + (cid:104)q(cid:105) gives the decomposition of a rational number q ∈ Q into its floor and the fractional part. We have thus recovered (3.17) from the edge-contraction formula alone, which are the (0, 1) and (0, 2) cases of the ELSV formula for the orbifold Hurwitz numbers. Acknowledgement. The paper is based a series of lectures by M.M. at Mathematische Ar- beitstagung 2015, Max-Planck-Institut fur Mathematik in Bonn. The authors are grateful to the American Institute of Mathematics in California, the Banff International Research Station, the Institute for Mathematical Sciences at the National University of Singapore, Kobe University, Leibniz Universitat Hannover, the Lorentz Center for Mathematical Sci- ences, Leiden, Max-Planck-Institut fur Mathematik in Bonn, and Institut Henri Poincar´e, Paris, for their hospitality and financial support during the authors' stay for collaboration. The research of O.D. has been supported by GRK 1463 Analysis, Geometry, and String Theory at Leibniz Universitat Hannover and MPIM. The research of M.M. has been sup- ported by NSF grants DMS-1309298, DMS-1619760, DMS-1642515, and NSF-RNMS: Geo- metric Structures And Representation Varieties (GEAR Network, DMS-1107452, 1107263, 1107367). 16 O. DUMITRESCU AND M. MULASE References [1] V. Bouchard, D. Hern´andez Serrano, X. Liu, and M. Mulase, Mirror symmetry for orbifold Hurwitz numbers, arXiv:1301.4871 [math.AG] (2013). [2] V. Bouchard, A. Klemm, M. Marino, and S. Pasquetti, Remodeling the B-model, Commun. Math. Phys. 287, 117 -- 178 (2009). [3] V. Bouchard and M. Marino, Hurwitz numbers, matrix models and enumerative geometry, Proc. Symposia Pure Math. 78, 263 -- 283 (2008). [4] R. Dijkgraaf, E. Verlinde, and H. Verlinde, Loop equations and Virasoro constraints in non-perturbative two- dimensional quantum gravity, Nucl. Phys. B348, 435 -- 456 (1991). [5] N. Do, O. Leigh, and P. Norbury, Orbifold Hurwitz numbers and Eynard-Orantin invariants, arXiv:1212.6850 (2012). [6] O. Dumitrescu and M. Mulase, Quantum curves for Hitchin fibrations and the Eynard-Orantin theory, Lett. Math. Phys. 104, 635 -- 671 (2014). [7] O. Dumitrescu and M. Mulase, Quantization of spectral curves for meromorphic Higgs bundles through topological recursion, arXiv:1411.1023 (2014). [8] O. Dumitrescu and M. Mulase, Edge contraction on dual ribbon graphs and 2D TQFT, Journal of Algebra, vol. 494, January 2018. [9] O. Dumitrescu and M. Mulase, Lectures on the topological recursion for Higgs bundles and quantum curves, to appear in the lecture notes series of the National University of Singapore. [10] O. Dumitrescu, M. Mulase, A. Sorkin and B. Safnuk, The spectral curve of the Eynard-Orantin recursion via the Laplace transform, in "Algebraic and Geometric Aspects of Integrable Systems and Random Matrices," Dzhamay, Maruno and Pierce, Eds. Contemporary Mathematics 593, 263 -- 315 (2013). [11] T. Ekedahl, S. Lando, M. Shapiro, A. Vainshtein, Hurwitz numbers and intersections on moduli spaces of curves, Invent. Math. 146, 297 -- 327 (2001) [arXiv:math/0004096]. [12] B. Eynard, M. Mulase and B. Safnuk, The Laplace transform of the cut-and-join equation and the Bouchard- Marino conjecture on Hurwitz numbers, Publications of the Research Institute for Mathematical Sciences 47, 629 -- 670 (2011). [13] B. Eynard and N. Orantin, Invariants of algebraic curves and topological expansion, Communications in Number Theory and Physics 1, 347 -- 452 (2007). [14] B. Fang, C.-C. M. Liu, and Z. Zong, All genus open-closed mirror symmetry for affine toric Calabi-Yau 3- orbifolds, arXiv:1310.4818 [math.AG] (2013). [15] B. Fang, C.-C. M. Liu, and Z. Zong, On the Remodeling Conjecture for Toric Calabi-Yau 3-Orbifolds, arXiv:1604.07123 (2016). [16] P. Johnson, R. Pandharipande, and H.H. Tseng, Abelian Hurwitz-Hodge integrals, Michigan Math. J. 60, 171 -- 198 (2011) [arXiv:0803.0499]. [17] M. Kontsevich, Intersection theory on the moduli space of curves and the matrix Airy function, Communications in Mathematical Physics 147, 1 -- 23 (1992). [18] M. Mulase and M. Penkava, Ribbon graphs, quadratic differentials on Riemann surfaces, and algebraic curves defined over Q, The Asian Journal of Mathematics 2 (4), 875 -- 920 (1998). [19] M. Mulase and P. Su(cid:32)lkowski, Spectral curves and the Schrodinger equations for the Eynard-Orantin recursion, arXiv:1210.3006 (2012). [20] M. Mulase and N. Zhang, Polynomial recursion formula for linear Hodge integrals, Communications in Number Theory and Physics 4, 267 -- 294 (2010). [21] A. Okounkov and R. Pandharipande, Gromov-Witten theory, Hurwitz numbers, and matrix models, I, Proc. Symposia Pure Math. 80, 325 -- 414 (2009). [22] C. Teleman, The structure of 2D semi-simple field theories, Inventiones Mathematicae 188, 525 -- 588 (2012). [23] G. 't Hooft, A planer diagram theory for strong interactions, Nuclear Physics B 72, 461 -- 473 (1974). [24] E. Witten, Two dimensional gravity and intersection theory on moduli space, Surveys in Differential Geometry 1, 243 -- 310 (1991). O. Dumitrescu: Department of Mathematics, Central Michigan University, Mount Pleasant, MI 48859 E-mail address: [email protected] and Simion Stoilow Institute of Mathematics, Romanian Academy, 21 Calea Grivitei Street, 010702 Bucharest, Romania M. Mulase: Department of Mathematics, University of California, Davis, CA 95616 -- 8633 E-mail address: [email protected] and Kavli Institute for Physics and Mathematics of the Universe, The University of Tokyo, Kashiwa, Japan
1703.00758
4
1703
2018-02-01T08:46:31
Effective Adjunction Theory
[ "math.AG" ]
Here we investigate the property of effectivity for adjoint divisors. Among others, we prove the following results: (i) A normal projective variety $X$ with at most canonical singularities is uniruled if and only if for each very ample Cartier divisor $H$ on $X$ we have $H^0(X, m_0K_X+H)=0$ for some $m_0=m_0(H)>0$. (ii) Let $(X,L)$ be a polarized manifold of dimension $4$ and let $t$ be an integer with $t \ge 3$. If $K_X+tL$ is pseudo-effective, then $H^0(X, K_X+tL) \ne 0$.
math.AG
math
EFFECTIVE ADJUNCTION THEORY MARCO ANDREATTA AND CLAUDIO FONTANARI Abstract. Here we investigate the property of effectivity for adjoint divisors. Among others, we prove the following results: A projective variety X with at most canonical singularities is uniruled if and only if for each very ample Cartier divisor H on X we have H 0(X, m0KX + H) = 0 for some m0 = m0(H) > 0. Let X be a projective 4-fold, L an ample divisor and t an integer with t ≥ 3. If KX + tL is pseudo-effective, then H 0(X, KX + tL) 6= 0. 1. Introduction Let X be a normal projective variety over the complex field C; let KX be its canonical divisor. We assume that X has at most canonical singularities. In the paper we fix a suitable Cartier divisor H on X and we discuss when the effectivity or non-effectivity of some adjoint divisors aKX + bH determines the geometry of X. In the first part we consider the notion of Termination of Adjunction. This turns out to be rather delicate, since in the literature there are different meanings for such a property. The following are four possibilities, where m0 and m are natural numbers. (A) For every (for some) big Cartier divisor H there exists m0 = m0(H) > 0 such that mKX + H /∈ Ef f (X) (i.e. it is not pseudo-effective) for m ≥ m0. (B) For every big Cartier divisor H we have H 0(X, m0KX + H) = 0 for some m0 = m0(H) > 0. (C) For every very ample Cartier divisor H we have H 0(X, m0KX +H) = 0 for some m0 = m0(H) > 0. (D) For some (for every) big Cartier divisor H0 we have H 0(X, m0KX + kH0) = 0 for every k > 0 and some m0 = m0(k) > 0. It is clear that (A) =⇒ (B) =⇒ (C) =⇒ (D). We prove that these four definitions are equivalent and moreover that Ad- junction Terminates in the above sense if and only if X is uniruled (see Theorem 3, Corollary 1 and Corollary 2). The results follow by some characterizations of pseudo-effective Cartier divi- sor (see Theorem 2), which are direct consequences of a fundamental result We would like to thank Paolo Cascini, Roberto Pignatelli and Luis Sola-Conde for fruitful conversations. We are grateful to J´anos Koll´ar for pointing out his examples and for suggesting projective varieties with canonical singularities as a good category to settle our results. We also thank the referees for useful comments. The research project was partially supported by GNSAGA of INdAM, by PRIN 2015 "Geometria delle variet`a algebriche", and by FIRB 2012 "Moduli spaces and Applications". 2010 Mathematics Subject Classification: 14E30, 14J40, 14J35, 14N30. 1 2 MARCO ANDREATTA AND CLAUDIO FONTANARI of Siu ([22]). The connection with uniruledness follows in turn from the fact that a projective variety X with canonical singularities is uniruled if and only if KX is not pseudo-effective (see [3], Corollary 0.3, or [5], Corollary 1.3.3). A characterization of rationally connected manifolds along the same lines has been given in [6]. The examples described in [15], Theorem 39, show that, for varieties with singularities worst then canonical, uniruledness is not connected to Termi- nation of Adjunction. We consider also the following more general definition. (C') Let H be an effective Cartier divisor on X. We say that Adjunction Terminates in the classical sense for H if there exists an integer m0 ≥ 1 such that H 0(X, H + mKX) = 0 for every integer m ≥ m0. We conjecture that such a definition is actually equivalent to the previous ones; a partial result in this direction is provided by Proposition 2. In dimension two, Castelnuovo and Enriques indeed proved that Condition (C') implies that X is uniruled (see [7] and also [20]). In the second part of the paper we assume that X is a projective variety of dimension n with at most terminal Q-factorial singularities. We take a nef and and big Cartier divisor L on X and we call (X, L) a quasi polarized pair. The following is a straightforward consequence of Theorem D in [5], see Remark 5 at the beginning of Section 5. Proposition 1. Let (X, L) be a quasi polarized pair and t > 0. If KX +tL ∈ Ef f (X), then there exists N ∈ N such that H 0(X, N (KX + tL)) 6= 0. Note that for t = 0 the statement of the Proposition would amount to Abundance Conjecture, together with MMP. The next Conjecture is an effective version of the above Proposition. Conjecture 1. Let (X, L) be a quasi polarized pair and t > 0. If KX + tL ∈ Ef f (X), then H 0(X, KX + tL) 6= 0. The case t = 1 is a version of the so-called Ambro-Ionescu-Kawamata conjec- ture, which is true for n ≤ 3 (see Theorem 1.5 in [14]), while for t = n − 1 we recover a conjecture by Beltrametti and Sommese (see [4], Conjecture 7.2.7). Note that if Conjecture 1 holds for t = 1 then it holds also for every t > 0. In the paper we consider the following conjecture. Conjecture 2. Let (X, L) be a quasi polarized pair and s > 0. Then H 0(X, KX + tL) = 0 for every integer t with 1 ≤ t ≤ s if and only if KX + sL is not pseudo-effective. Since L is big, in particular pseudo-effective, then the if part is obvious. Note that Conjecture 2 for s = 1 implies Conjecture 1. EFFECTIVE ADJUNCTION 3 We prove that Conjecture 2 is true for s = n (see Proposition 4); we actually show that this case happens if and only if the pair (X, L) is bira- tionally equivalent (via a 0-reduction, see the definition in the next section) to the pair (Pn, O(1)). For s = n − 1 the conjecture was essentially proved by Horing, see [14], Theorem 1.2. We prove a slightly more explicit version of his result (see Proposition 7), namely, we show that this case happens if and only if the pair (X, L) is birationally equivalent to a finite list of pairs. Finally, we focus on the case n = 4 (see Theorem 8 and Proposition 6) and we generalize previous work by Fukuma ([12], Theorem 3.1). 2. Notation and preliminaries Let X be a normal complex projective variety of dimension n. We adopt [16] and [17] as the standard references for our set-up. In particular, we denote by Div(X) the group of all Cartier divisors on X and by N um(X) the subgroup of numerically trivial divisors. The quotient group N 1(X) = Div(X)/N um(X) is the Neron-Severi group of X. In the vector space N 1(X)R := N 1(X) ⊗ R, whose dimension is ρ(X) := rkN 1(X), we consider some convex cones. (a) Amp(X) ⊂ N 1(X)R the convex cone of all ample R-divisor classes; it is an open convex cone. (b) Big(X) ⊂ N 1(X)R the convex cone of all big R-divisor classes; it is an open convex cone. (e) Ef f (X) ⊂ N 1(X)R the convex cone spanned by the classes of all effective R-divisors. (n) N ef (X) = Amp(X) ⊂ N 1(X)R the closed convex cone of all nef R-divisor classes. (p) Ef f (X) = Big(X) ⊂ N 1(X)R the closed convex cone of all pseudo- effective R-divisor classes. The above definitions actually lean on some fundamental results like the openess of the ample and big cones, the facts that int{Ef f (X)} = Big(X) and N ef (X) = Amp(X); for more details see [17]. Note that Amp(X) ⊂ N ef (X) ∩ Big(X) and that there are no inclusions between N ef (X) and Big(X). Note also that if π : X ′ → X is a birational morphism and D is a Cartier divisor on X then D is big (resp. pseudo-effective) if and only if π∗D is big (resp. pseudo-effective). We consider projective varieties with singularities of special type, as in the Minimal Model Program. For reader convenience we recall their definition (see [16], Definition 2.11 and Definition 2.12). Definition 1. Let X be a normal projective variety. We say that X has canonical (respectively terminal) singularities if i) KX is Q-Cartier, and ii) ν∗O X (mK X ) = OX (mKX) for one (or for any) resolution of the singularities ν : X → X (respectively 4 MARCO ANDREATTA AND CLAUDIO FONTANARI ii) ν∗O X (mK X − E) = OX (mKX ) for one (or for any) resolution of the singularities ν : X → X, where E ⊂ X is the reduced exceptional divisor). In the category of projective varieties with canonical singularities the pseudo- effectivity of the canonical bundle is a birational invariant, as noticed by Mori in [19], (11.4.1). He actually conjectured the following beautiful result ([19], (11.4.2) and (11.5)), which was proved in [3], Corollary 0.3 and in [5], Corollary 1.3.3. Theorem 1. Let X be a projective variety with at most canonical singular- ities. Then X is uniruled if and only if KX is not pseudoeffective. As for the invariance of the global sections of adjoint bundles (or of pluri- canonical bundles if L is trivial) we have the following. Lemma 1. Let π : Y → X be a birational morphism between projective varieties with at most canonical singularities, let L be a Cartier divisor on X and let a, b ∈ N. Then H 0(X, aKX + bL) = H 0(Y, aKY + bπ∗(L)). Proof. Since Y and X have canonical singularities we have π∗aKY = aKX. This is straightforward from the definition of canonical singularities and by taking a resolution of Y , ν : Y ′ → Y , and π ◦ ν : Y ′ → X as a resolution of X. Since L is Cartier, by projection formula it follows π∗(aKY + bπ∗(L)) = π∗(aKY + π∗(bL)) = π∗(aKY ) + bL = aKX + bL; by taking global sections we obtain our statement. (cid:3) 3. Termination of Adjunction Much of this section is based on the following Lemma, which was proved in the analytic setting by Siu (see [22], Proposition 1). For reader convenience we provide an algebraic proof relying on [18] (see also [21], Chapter V, Corollary 1.4). Lemma 2. Let X be a smooth projective variety of dimension n and let H be a very ample divisor on X. If G := (n + 1)H + KX, then for every pseudo-effective divisor F on X we have H 0(X, F + G) 6= 0. Proof. Since F is pseudo-effective we have that F + H is big, hence there exists a positive integer m > 0 such that m(F + H) ∼ A + E with A ample and E effective (see for instance [17], Corollary 2.2.7). Let D := m E and L := F + H, so that L − D = 1 1 m A is big and nef; apply [18], Proposition 9.4.23, to get H 0(X, KX + L + kH + I(D)) 6= 0. Since the multiplier ideal I(D) is an ideal of OX , it follows that H 0(X, KX + L + kH) 6= 0 for every k ≥ n, i.e. H 0(X, KX + F + (k + 1)H) 6= 0 as soon as k + 1 ≥ n + 1. (cid:3) The following characterization of pseudo-effective divisors is probably well- known to the specialists; however, we did not find it explicitly in the litera- ture. EFFECTIVE ADJUNCTION 5 Theorem 2. Let X be a smooth projective variety and let F be a divisor on X. The following statements, where m and N denote natural numbers, are equivalent: i) F ∈ Ef f (X) (i.e it is pseudo-effective). ii) There is a big divisor G such that H 0(X, N (mF + G)) 6= 0 for every m > 0 and for some N > 0. iii) There is a big divisor G such that H 0(X, mF +G) 6= 0 for all m > 0. iv) There is a very ample divisor G such that H 0(X, mF + G) 6= 0 for all m > 0. v) For every big divisor H we have H 0(X, mF + kH) 6= 0 for all m > 0 and all k ≥ k0(H). Proof. First of all note that the implications v) =⇒ iv), iv) =⇒ iii) and iii) mF +G m . =⇒ ii) are obvious. Moreover ii) =⇒ i) follows from F ≡ limm→+∞ The difficult part is to prove i) =⇒ v); for this we use Lemma 2 together with Kodaira's Lemma (see for instance [17], Proposition 2.2.6). Namely, let G be the divisor of Lemma 2; then H 0(X, G) 6= 0 (just take F = OX ). If H is a big divisor on X, then by Kodaira's Lemma H 0(X, kH − G) 6= 0 for every k ≥ k0(H). Hence dim H 0(X, mF + kH) = dim H 0(X, mF + k0H − G + G + (k − k0)H) ≥ ≥ dim H 0(X, mF + (k − k0)H + G) > 0, where the last inequality follows from Lemma 2 by taking as a pseudo- effective divisor mF + (k − k0)H. (cid:3) Remark 1. Note that i) =⇒ iii) is just Lemma 2, while i) =⇒ ii) follows easily from int{Ef f (X)} = Big(X); this last fact was first noticed by Mori in [19], (11.3) on p. 318. Indeed, let G ∈ Big(X) and F ∈ Ef f (X); then the set [G, F ) := {G + mF : m ∈ R+} is contained in int{Ef f (X)} = Big(X). The next Theorem proves the equivalence of the different definitions of Ter- mination of Adjunction stated in the Introduction. Theorem 3. Let X be a projective variety with at most canonical singular- ities. The following statements, where m and m0 denote natural numbers, are equivalent: (i) X is uniruled (i.e. KX is not pseudo-effective). (ii) For every big Cartier divisor H there exists m0 = m0(H) > 0 such that mKX + H /∈ Ef f (X) for m ≥ m0. (iii) For every big Cartier divisor H we have H 0(X, m0KX + H) = 0 for some m0 = m0(H) > 0. (iv) For every very ample Cartier divisor H we have H 0(X, m0KX + H) = 0 for some m0 = m0(H) > 0. (v) For some big Cartier divisor H0 we have H 0(X, m0KX + kH0) = 0 for every k > 0 and some m0 = m0(k) > 0. Proof. (i) =⇒ (ii) is implied by the properties of the cone described in Sec- tion 2; indeed, it follows by contradiction from KX ≡ limm→+∞ (ii) =⇒ (iii), (iii) =⇒ (iv) and (iv) =⇒ (v) are straightforward. mKX +H . m 6 MARCO ANDREATTA AND CLAUDIO FONTANARI (v) =⇒ (i) requires a resolution of the singularities ν : X → X. Assume by contradiction that X is not uniruled. Therefore also X is not uniruled and K X is pseudo-effective. If H is any big Cartier divisor on X, then H = ν∗(H) is big and by [17], Corollary 2.2.7, we have l H = A + N with A ample and N effective for some l > 0. It follows that hl H = hA + hN with hA very ample for some h > 0. Hence, by Lemma 1, for every m0 > 0 we have dim H 0(X, m0KX + (n + 1)hlH) = dim H 0( X, m0K X + (n + 1)hl H) = dim H 0( X, (m0 −1)K X +(K X +(n+1)hA)+(n+1)hN ) ≥ dim H 0( X, (m0 − 1)K X + (K X + (n + 1)hA)). Lemma 2 says that this last term is positive, thus contradicting our assumption. (cid:3) Remark 2. Note that Mori in [19], (11.4) on p. 318, suggests that in prin- ciple (i) could have been stronger then (iv): We say that X is κ-uniruled if KX is not pseudo-effective. We note that κ-uniruledness is slightly stronger than saying that adjunction terminates, i.e. H 0(X, mKX + H) = 0 for each very ample divisor H and some m = m(H) > 0. The following two corollaries show that the two formulations, respectively for some and for every, of (A) and (D) in the Introduction are equivalent. Corollary 1. Let X be a projective variety with at most canonical singu- larities. The following statements, where m and m0 denote natural numbers, are equivalent: (i) For every big Cartier divisor H there exists m0 = m0(H) > 0 such that mKX + H /∈ Ef f (X) for m ≥ m0. (ii) For some big Cartier divisor H0 there exists m0 = m0(H0) > 0 such that mKX + H0 /∈ Ef f (X) for m ≥ m0. Proof. It is obvious that (i) implies (ii). Conversely, if (ii) holds then KX is not pseudoeffective, hence X is uniruled. It follows from Theorem 3 that (i) holds. (cid:3) Corollary 2. Let X be a projective variety with at most canonical singu- larities. The following statements, where m and m0 denote natural numbers, are equivalent: (i) For some big Cartier divisor H0 we have H 0(X, m0KX + kH0) = 0 for every k > 0 and some m0 = m0(k) > 0. (ii) For every big Cartier divisor H we have H 0(X, m0KX + kH) = 0 for every k > 0 and some m0 = m0(k, H) > 0. Proof. It is obvious that (ii) implies (i). Conversely, if (i) holds then by Theorem 3 X is uniruled, i.e. KX is not pseudoeffective. Assume by con- tradiction that there exist a big divisor H and some k0 > 0 such that mKX +k0H H 0(X, mKX + k0H) 6= 0 for every m > 0. Then KX = limm→+∞ is pseudo-effective, a contradiction. (cid:3) m As pointed out by the referee, since every divisor is a difference of very ample ones, (C) is actually equivalent to the following stronger condition. (C*) For every Cartier divisor D we have H 0(X, m0KX + D) = 0 for some m0 = m0(D) > 0. EFFECTIVE ADJUNCTION 7 The following is a more general definition of Termination of Adjunction. Definition 2. (Condition (C')) Let X be a normal projective variety; let H be an effective Cartier divisor on X. We say that Adjunction Terminates in the classical sense for H if there exists an integer m0 ≥ 1 such that for every integer m ≥ m0. H 0(X, H + mKX) = 0 We conjecture that such a definition is actually equivalent to the previous ones. The following partial result in this direction is straightforward. Proposition 2. Let X be a projective variety with canonical singularities. Let H be any effective divisor and assume that Adjunction Terminates in the classical sense for H. Then X has negative Kodaira dimension. Proof. Recall that the Kodaira dimension of a singular variety is defined to be the Kodaira dimension of any smooth model (see for instance [17], Example 2.1.5). Assume by contradiction that X has non-negative Ko- daira dimension, i.e. H 0( X, n0K X) 6= 0 for some integer n0 ≥ 1, where ν : X → X is any resolution of the singularities. Since X has canonical sin- gularities, from Lemma 1 it follows that H 0(X, n0KX ) = H 0( X, n0K X) 6= 0. Hence H 0(X, H + nn0KX) 6= 0 for every integer n ≥ 1, contradicting the assumption that H 0(X, H + mKX ) = 0 for m >> 0. (cid:3) Together with the standard conjecture that negative Kodaira dimension im- plies uniruledness (see for instance [19], (11.5) on p. 319, and [3], Conjecture 0.1), from Proposition 2 it would follow that Termination of Adjunction in the classical sense implies uniruledness. In dimension two such an implica- tion holds unconditionally, as it was proved by Castelnuovo and Enriques in [7] (for a modern proof we refer to [20]). We conclude this section with a characterization of uniruled varieties which may suggest a different way to consider (effective) termination of adjunction. It follows as a straightforward consequence of Lemma 2 and the main result in [3]. Proposition 3. Let X be a smooth projective variety of dimension n and let H be a very ample divisor on X. If H 0(X, mKX + (n + 1)H) = 0 for some natural number m ≥ 1, then X is uniruled. Proof. Assume by contradiction that X is not uniruled, so that KX is pseudo-effective by [3]. Lemma 2 with F = (m − 1)KX gives the sought-for contradiction. (cid:3) Theorem 3.1 in [9] gives a statement similar to the last proposition; there the variety is singular and H is just nef and big. However m > 1 and H has to be multiplied by a higher number, for instance n2. 4. Quasi polarized pairs A quasi polarized pair is a pair (X, L) where X is a projective variety with at most Q-factorial terminal singularities and L is a nef and big Cartier divisor on X. If L is ample we call the pair (X, L) a polarized pair. 8 MARCO ANDREATTA AND CLAUDIO FONTANARI In [1], Section 4, following T. Fujita's ideas as revisited by A. Horing in [14] and using the MMP developed in [5], we described a MMP with scaling related to divisors of type KX + rL, for r a positive rational number. In particular we introduced the 0-reduction of a quasi polarized pair (X, L) (see [1], Definition 4.4) as quasi polarized pair (X ′, L′) birational to (X, L) obtained from (X, L) via a Minimal Model Program with scaling: (X, L) ∼ (X, ∆) := (X0, ∆0) → − − −− → (Xs, ∆s) ∼ (X ′, L′), which contracts or flips all extremal rays R+[C] on X such that L · C = 0. At every step of the MMP given above, we have a quasi polarized variety (Xi, Li) with at most terminal Q-factorial singularities. If πi : (Xi, ∆i) → (Xi+1, ∆i+1) is birational then Li = π∗ πi : (Xi, ∆i) → (Xi+1, ∆i+1) is a flip then Li and π∗ in codimension one. i (Li+1), while if i (Li+1) are isomorphic Remark 3. By using Lemma 1 and Hartogs theorem we deduce H 0(X, aKX + bL) = H 0(X ′, aKX ′ + bL′) for a, b ∈ N. The following has been proved in [1], Theorem 5.1 and in [13], Proposition 1.3. Theorem 4. Let (X, L) be a quasi polarized pair. Then KX + tL is pseudo- effective for all t ≥ n unless the 0-reduction (X ′, L′) is (Pn, O(1)). Actually, KX + (n − 1)L is pseudo-effective unless (X ′, L′) is one of the following pairs: • (Pn, O(1)), • (Q, O(1)Q), where Q ⊂ Pn+1 is a quadric, • Cn(P2, O(2)), a generalized cone over (P2, O(2)), • X has the structure of a Pn−1-bundle over a smooth curve C and L restricted to any fiber is O(1). Moreover, except in the above cases, KX ′ + (n − 1)L′ is nef. The first-reduction of a quasi polarized pair (X, L) (see [1], Definition 5.5) is a quasi polarized pair (X ′′, L′′) birational to (X, L) obtained from a 0-reduction (X ′, L′) via a morphism ρ : X ′ → X ′′ consisting of a series of divisorial contractions to smooth points, which are weighted blow-ups of weights (1, 1, b, . . . , b) with b ≥ 1 (see [2], Theorem 1.1). Remark 4. According to [1], Proposition 5.4, we have H 0(KX + tL) = H 0(KX ′′ + tL′′) for any 0 ≤ t ≤ n − 2. The following has been proved in [1], Theorem 5.7. Theorem 5. Let (X, L) be a quasi polarized pair. KX + (n − 2)L is not pseudo-effective if and only if any first-reduction (X ′′, L′′) is either one of the pairs listed in the statement of Theorem 4 or one of the following pairs: • a del Pezzo variety, that is −KX ′′ ∼Q (n − 1)L with L ample, • (P4, O(2)), EFFECTIVE ADJUNCTION 9 • (P3, O(3)), • (Q, O(2)Q), where Q ⊂ P4 is a quadric, • X has the structure of a quadric fibration over a smooth curve C and L restricted to any fiber is O(1)Q, • X has the structure of a Pn−2-bundle over a normal surface S and L restricted to any fiber is O(1), • n = 3, X is fibered over a smooth curve Z with general fiber P2 and L restricted to it is O(2). If KX + (n − 2)L is pseudo-effective then on any first-reduction (X ′′, L′′) the divisor KX ′′ + (n − 2)L′′ is nef. The following definition was given by Horing (see ([14], Definition 1.2). Definition 3. A quasi polarized pair (X, L) is a (generalized) scroll if X is smooth and there is a fibration X → Y onto a projective manifold Y such that the general fiber F admits a birational morphism τ : F → Pm and that OF (L) = τ ∗OPm(1). A quasi polarized pair (X, L) is birationally a scroll if there is a birational morphism ν : X ′ → X such that (X ′, ν∗L) is a (generalized) scroll. The next is Theorem 1.4 in [14]. Theorem 6. Let (X, L) be a quasi polarized pair. tionally a scroll then ΩX ⊗ L is generically nef. If (X, L) is not bira- A key step in the proofs of Theorem 7 and of Theorem 8 is the following lemma due to Horing (see [14], p. 741, Step 2 in the proof of Theorem 1.2). Lemma 3. Let (X, L) be a quasi polarized pair. Assume that KX + (n − 2)L is pseudo-effective and that KX + (n − 1)L is nef and big. Then Ln−2[(2(K 2 X + c2(X)) + 6nLKX + (n + 1)(3n − 2)L2] > 0. We consider now a quasi polarized pair (X, L) and we assume moreover that X is smooth. We borrow from Y. Fukuma the following set-up for the computation of the Hilbert polynomial of KX + tL. Let F0(t) Fi(t) := dim H 0(X, KX + tL), := Fi−1(t + 1) − Fi−1(t) for every integer i with 1 ≤ i ≤ n. The following statement can be easily checked by reverse induction on b ≤ a. Lemma 4. Fix an integer a ≥ 1. 1 ≤ t ≤ a, then Fa−b(c) = 0 for all integers b, c with 1 ≤ c ≤ b ≤ a. If F0(t) = 0 for every integer t with If one defines then it follows easily that Ai(X, L) := Fn−i(1) (1) dim H 0(X, KX + tL) = n Xj=0 (cid:18) t − 1 n − j(cid:19)Aj(X, L). 10 MARCO ANDREATTA AND CLAUDIO FONTANARI Moreover, by taking a = n − i + 1 and b = c = 1 in Lemma 4, we obtain the following implication. Corollary 3. If H 0(X, KX + tL) = 0 for every integer t with 1 ≤ t ≤ n − i + 1, then Ai(X, L) = 0. On the other hand, by Kawamata-Viehweg vanishing theorem and Serre duality, we have dim H 0(X, KX + tL) = χ(X, −tL); therefore from the Riemann-Roch theorem we obtain the following explicit computations (for further details, see [10], (2.2), and [11], Proposition 3.2). Lemma 5. Let (X, L) be a polarized manifold of dimension n and let g(X, L) denote the sectional genus of (X, L). Then we have A0(X, L) = Ln A1(X, L) = g(X, L) + Ln − 1 24 · A2(X, L) = Ln−2[(2(K 2 48 · A3(X, L) = (n − 2)(n2 − 1)Ln + n(3n − 5)KX Ln−1 + X + c2(X)) + 6nLKX + (n + 1)(3n − 2)L2] +2(n − 1)K 2 X Ln−2 + 2c2(X)(KX + (n − 1)L)Ln−3. 5. Polarized Abundance The aim of this section is to argue around the Conjectures stated in the introduction. We start showing that Proposition 1 is a direct consequence of (the more general) Theorem D in [5]. Remark 5. Let (X, L) be a quasi-polarized variety and let t be a positive rational number. Then there exists an effective Q-divisor ∆t on X such that ∆t ∼Q tL and (X, ∆t) is Kawamata log terminal. This is well-known If KX + tL ∈ Ef f (X), to the specialists, a proof can be found in [1]. then KX + ∆t ∈ Ef f (X) and by [5], Theorem D, there exists an R-divisor D ≥ 0 such that KX + ∆t ∼R D. That is, there exists N ∈ N such that H 0(X, N (KX + tL)) > 0. We consider Conjecture 2; for s = n we recover the following easy fact. Proposition 4. Let (X, L) be a quasi polarized pair of dimension n. We have H 0(X, KX + tL) = 0 for every integer t with 1 ≤ t ≤ n if and only if KX + nL is not pseudo-effective. Moreover this is the case if and only if the 0-reduction (X ′, L′) of the pair (X, L) is (Pn, O(1)). Proof. By Remark 3 we have H 0(X, KX + tL) = H 0(X ′, KX ′ + tL′) for any t ≥ 0. Hence if H 0(X, KX + tL) = 0 for every integer t with 1 ≤ t ≤ n then from Corollary 3 it follows that A1(X ′, L′) = g(X ′, L′) + L′n − 1 = 0. Since we have g(X ′, L′) = 0 and L′n = 1 if and only if (X ′, L′) = (Pn, O(1)), the claim follows from [1], Theorem 5.1 (2). (cid:3) Next, for s = n − 1, the following is a slightly more explicit version of [14], Theorem 1.2; the proof is essentially the one of [14]. EFFECTIVE ADJUNCTION 11 Theorem 7. Let (X, L) be a quasi polarized pair of dimension n. We have H 0(X, KX + tL) = 0 for every integer t with 1 ≤ t ≤ n − 1 if and only if KX + (n − 1)L is not pseudo-effective. That is, by Theorem 4, if and only if the 0-reduction (X ′, L′) of the pair (X, L) is one of the following: (i) (Pn, O(1)), (ii) (Q, O(1)Q), where Q ⊂ Pn+1 is a quadric, (iii) Cn(P2, O(2)), a generalized cone over (P2, O(2)), (iv) X has the structure of a Pn−1-bundle over a smooth curve C and L restricted to any fiber F is O(1). Proof. Let (X ′, L′) be the 0-reduction of the pair (X, L) and let ( X ′, L′) be its desingularization (namely, ν : X ′ → X ′ and L′ = ν∗(L′)). By Remark 3 and Lemma 1 we have H 0(X, KX + tL) = H 0(X ′, KX ′ + tL′) = H 0( X ′, K X ′ + t L′) for any t ≥ 0. The if part is obvious. In order to prove the only if part, assume that H 0(X, KX + tL) = H 0(X ′, KX ′ +tL′) = H 0( X ′, K X ′ + t L′) = 0 for every integer t with 1 ≤ t ≤ n − 1. Corollary 3 implies that A2( X ′, L′) = 0. (2) Assume by contradiction that (X ′, L′) is not one of the pairs in (i), (ii), (iii), (iv); then, by Theorem 4, KX ′ + (n − 1)L′ is nef. The required contradiction is provided by [14], Theorem 1.2. (cid:3) The next step s = n − 2 should work as follows. Conjecture 3. Let (X, L) be a quasi polarized manifold of dimension n. We have H 0(X, KX + tL) = 0 for every integer t with 1 ≤ t ≤ n − 2 if and only if KX + (n − 2)L is not pseudo-effective, that is if and only if the first-reduction (X ′′, L′′) is one of the pairs (X, L) listed in Theorems 4 and 5. Once again, the if part is obvious. Conversely, from Corollary 3 it follows that A3(X, L) = 0, but the proof of the only if part seems to be elusive. From now on, we focus on the case n = 4; here formula (1) reads simply as: (3) H 0(X, KX + tL) = (cid:18)t − 1 +(cid:18)t − 1 2 (cid:19)A2(X, L) +(cid:18)t − 1 4 (cid:19)A0(X, L) +(cid:18)t − 1 1 (cid:19)A3(X, L) +(cid:18)t − 1 3 (cid:19)A1(X, L) + 0 (cid:19)A4(X, L) where A1(X, L) = g(X, L) + L4 − 1, A2(X, L) = dim H 0(X, KX + 3L) − 2 dim H 0(X, KX + 2L) + + dim H 0(X, KX + L), A3(X, L) = dim H 0(X, KX + 2L) − dim H 0(X, KX + L), A4(X, L) = dim H 0(X, KX + L). 12 MARCO ANDREATTA AND CLAUDIO FONTANARI We prove the following generalization of [12], Theorem 3.1. Theorem 8. Let (X, L) be a polarized manifold of dimension 4 and let t be an integer with t ≥ 3. If KX +tL is pseudo-effective, then H 0(X, KX +tL) 6= 0. In particular, • H 0(X, KX + tL) 6= 0 for t ≥ 5 • H 0(X, KX + 4L) = 0 if and only if (X, L) is (P4, O(1)) • H 0(X, KX + 3L) = 0 if and only if (X, L) is either (Q, O(1)Q), where Q ⊂ P5 is a quadric, or X has the structure of a P3-bundle over a smooth curve C and L restricted to any fiber is O(1). Proof. Since L is ample (X, L) is a 0-reduction, in particular by Theorem 4 we can assume that KX + tL is nef for t ≥ 4. We can also assume that KX + 3L is nef. Indeed, if not then (X, L) is one of the exceptions listed in the statement of Theorem 4. If (X, L) is (P4, O(1)) or (Q, O(1)), where Q ⊂ P5 is a quadric hypersurface, then Theorem 8 is obvious. The case of a generalized cone over (P2, O(2)) does not occur since X is smooth, while the case of a P3-bundle over a smooth curve will be considered in Proposition 5. Now, assume that ΩX ⊗ L is generically nef. By using the formulas in Lemma 5 and Miyaoka inequality as stated in [14], Corollary 2.11, with D := 4L, we compute: A2(X, L) ≥ 24 (cid:0)2(KX + 3L)2L2 + 6(KX + 3L)L3 + 2L4(cid:1) A3(X, L) ≥ − (KX + 3L)L3. 1 1 24 Hence from (3) and the nefness of KX + 3L it follows that dim H 0(X, KX + tL) ≥ (t − 1)A3(X, L) + (t − 1)(t − 2) 2 A2(X, L) > 0 for every t ≥ 3. Finally, assume that ΩX ⊗ L is not generically nef. By Theorem 6 and Lemma 1 we may assume that X is a (generalized) scroll and the claim is a consequence of the following proposition. (cid:3) Proposition 5. Let (X, L) be a generalized scroll of dimension 4 and let t be an integer such that t ≥ 3. If KX + tL is nef, then H 0(X, KX + tL) 6= 0. Proof. Let X → Y be the scroll fibration and let F be the generic fiber with a birational morphism τ : F → Pm as in Definition 3. If X = P4 the claim is obvious; therefore we can assume that m ≤ 3 and that A1(X, L) = g(X, L)+L4 −1 > 0 (since we have g(X, L) = 0 and L4 = 1 if and only if (X, L) = (P4, O(1))). We also have that A0(X, L) = L4 ≥ 1 and A4(X, L) = dim H 0(X, KX + L) ≥ 0. If m = 3, then KX + sLF = τ ∗OP3(−4 + s), hence H 0(X, KX + sL) = 0 for s ≤ 3. Thus we have A2(X, L) = A3(X, L) = 0 and from (3) it follows that for t ≥ 4 we have dim H 0(X, KX + tL) ≥ A1(X, L) > 0 EFFECTIVE ADJUNCTION 13 If m = 2, then KX + sLF = τ ∗OP2(−3 + s), hence H 0(X, KX + sL) = 0 for s ≤ 2. In particular, we have A3(X, L) = 0 and A2(X, L) = dim H 0(X, KX + 3L). For t = 3, i.e. if we assume KX + 3L is nef, by Theorem 1.2 in [14] we must have H 0(X, KX + 3L) 6= 0 since H 0(X, KX + sL) = 0 for s ≤ 2. For t ≥ 4 we deduce from (3) that dim H 0(X, KX + tL) ≥ A1(X, L) > 0. If m = 1, then KX + sLF = τ ∗OP1(−2 + s), hence H 0(X, KX + L) = 0. In particular, we have A3(X, L) ≥ 0. If H 0(X, KX + 2L) = 0, then A2(X, L) = dim H 0(X, KX + 3L) and we conclude exactly as in the previous case m = 2. If H 0(X, KX + 2L) 6= 0, then KX + 2L is pseudo-effective and KX + 3L is pseudo-effective and big. Passing to the 0-reduction we may assume that KX + 3L is nef and big. Therefore Lemma 3 applies and by Lemma 5 we get A2(X, L) > 0. Hence from (3) it follows that for t ≥ 3 we have dim H 0(X, KX + tL) ≥ A2(X, L) > 0. (cid:3) The statement of Theorem 8 should hold also for t = 2, but we have only the following partial result. Proposition 6. Let (X, L) be a polarized manifold of dimension 4. If KX + 2L is pseudo-effective, then H 0(X, KX + 2L) 6= 0 unless ΩX < 1 2 L > is not generically nef. Proof. By Theorem 5 and Remark 4 we may assume that KX + 2L is nef. Assume that ΩX < 1 2 L > is generically nef. By using the formula for A3(X, L) in Lemma 5 and Miyaoka inequality, as stated in [14], Corol- lary 2.11, with D := 2L, we compute: A3(X, L) ≥ 1 16 (KX + 2L)2L2 + 1 12 (KX + 2L)L3 + 1 48 L4. Hence from (3) it follows that dim H 0(X, KX + tL) ≥ A3(X, L) > 0. (cid:3) References [1] M. Andreatta: Minimal model program with scaling and adjunction theory. Internat. J. Math. 24 (2013), no. 2, 1350007, 13 pp. [2] M. Andreatta, L. Tasin: Fano-Mori contractions of high length on projective varieties with terminal singularities. Bull. Lond. Math. Soc. 46 (2014), no. 1, 185 -- 196. [3] S. Boucksom, J.-P. Demailly, M. Paun, Th. Peternell: The pseudo-effective cone of a compact Kahler manifold and varieties of negative Kodaira dimension. J. Algebraic Geom. 22 (2013), no. 2, 201 -- 248. [4] M. C. Beltrametti, A. J. Sommese: The adjunction theory of complex projective varieties. De Gruyter Expositions in Mathematics, 16. Walter de Gruyter & Co., Berlin, 1995. [5] C. Birkar, P. Cascini, C. Hacon, J. McKernan: Existence of minimal models for varieties of log general type. J. Amer. Math. Soc. 23 (2010), no. 2, 405 -- 468. 14 MARCO ANDREATTA AND CLAUDIO FONTANARI [6] F. Campana, J.-P. Demailly, Th. Peternell: Rationally connected manifolds and semi- positivity of the Ricci curvature. Recent advances in algebraic geometry, 71-91, Lon- don Math. Soc. Lecture Note Ser., 417. Cambridge Univ. Press, Cambridge, 2015. [7] G. Castelnuovo, F. Enriques: Sopra alcune questioni fondamentali nella teoria delle superficie algebriche, Ann. di Mat. Pura e Applicata (3), 6 (1901), 165 -- 225. [8] J-P. Demailly: Analytic Methods in Algebraic Geometry. Universit´e de Grenoble I, Institut Fourier, 2011. [9] G. Di Cerbo: Uniform bounds for the Iitaka fibration. Ann. Sc. Norm. Super. Pisa (5) 13 (2014), no. 4, 1133 -- 1143. [10] Y. Fukuma: A formula for the sectional geometric genus of quasi polarized manifolds by using intersection numbers. J. Pure Appl. Algebra 194 (2004), no. 1-2, 113 -- 126. [11] Y. Fukuma: A study on the dimension of global sections of adjoint bundles for po- larized manifolds. J. Algebra 320 (2008), no. 9, 3543 -- 3558. [12] Y. Fukuma: On a conjecture of Beltrametti-Sommese for polarized 4-folds. Kodai Math. J. 38 (2015), no. 2, 343 -- 351. [13] A. Horing: The sectional genus of quasi-polarised varieties. Arch. Math. (Basel) 95 (2010), no. 2, 125 -- 133. [14] A. Horing: On a conjecture of Beltrametti and Sommese. J. Algebraic Geom. 21 (2012), no. 4, 721 -- 751. [15] J. Koll´ar: Is there a Topological Bogomolov-Miyaoka-Yau Inequality?. Pure and Ap- plied Math. Quarterly 4 (2008), no. 2 , 203 -- 236. [16] J. Koll´ar, S. Mori: Birational geometry of algebraic varieties. Cambridge Tracts in Mathematics, 134. Cambridge University Press, Cambridge, 1998. [17] R. Lazarsfeld: Positivity in algebraic geometry. I. Classical setting: line bundles and linear series. Springer-Verlag, Berlin, 2004. [18] R. Lazarsfeld: Positivity in algebraic geometry. II. Positivity for vector bundles, and multiplier ideals. Springer-Verlag, Berlin, 2004. [19] S. Mori: Classification of higher-dimensional varieties. Algebraic geometry, Bowdoin, 1985 (Brunswick, Maine, 1985), 269 -- 331, Proc. Sympos. Pure Math., 46, Part 1, Amer. Math. Soc., Providence, RI, 1987. [20] D. Mumford: Enriques' classification of surfaces in char p. I. Global Analysis (Papers in Honor of K. Kodaira) Univ. Tokyo Press, Tokyo, 1969, 325 -- 339. [21] N. Nakayama: Zariski-decomposition and abundance. MSJ Memoirs, 14. Mathemat- ical Society of Japan, Tokyo, 2004. [22] Y.-T. Siu: Invariance of plurigenera. Invent. Math. 134 (1998), no. 3, 661 -- 673. E-mail address: [email protected], [email protected] Current address: Dipartimento di Matematica, Universit`a degli Studi di Trento, Via Sommarive 14, 38123 Trento, Italy.
1402.0460
2
1402
2015-06-24T17:01:39
Triangulations of monotone families I: Two-dimensional families
[ "math.AG", "math.AT", "math.LO" ]
Let $K \subset {\mathbb R}^n$ be a compact definable set in an o-minimal structure over $\mathbb R$, e.g., a semi-algebraic or a subanalytic set. A definable family $\{ S_\delta|\> 0< \delta \in {\mathbb R} \}$ of compact subsets of $K$, is called a monotone family if $S_\delta \subset S_\eta$ for all sufficiently small $\delta > \eta >0$. The main result of the paper is that when $\dim K \le 2$ there exists a definable triangulation of $K$ such that for each (open) simplex $\Lambda$ of the triangulation and each small enough $\delta>0$, the intersection $S_\delta \cap \Lambda$ is equivalent to one of the five standard families in the standard simplex (the equivalence relation and a standard family will be formally defined). The set of standard families is in a natural bijective correspondence with the set of all five lex-monotone Boolean functions in two variables. As a consequence, we prove the two-dimensional case of the topological conjecture in [6] on approximation of definable sets by compact families. We introduce most technical tools and prove statements for compact sets $K$ of arbitrary dimensions, with the view towards extending the main result and proving the topological conjecture in the general case.
math.AG
math
TRIANGULATIONS OF MONOTONE FAMILIES I: TWO-DIMENSIONAL FAMILIES SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Abstract. Let K ⊂ Rn be a compact definable set in an o-minimal structure over R, e.g., a semi-algebraic or a subanalytic set. A definable family {Sδ 0 < δ ∈ R} of compact subsets of K, is called a monotone family if Sδ ⊂ Sη for all sufficiently small δ > η > 0. The main result of the paper is that when dim K ≤ 2 there exists a definable triangulation of K such that for each (open) simplex Λ of the triangulation and each small enough δ > 0, the intersection Sδ ∩ Λ is equivalent to one of the five standard families in the standard simplex (the equivalence relation and a standard family will be formally defined). The set of standard families is in a natural bijective correspondence with the set of all five lex-monotone Boolean functions in two variables. As a consequence, we prove the two-dimensional case of the topological conjecture in [7] on approximation of definable sets by compact families. We introduce most technical tools and prove statements for compact sets K of arbitrary dimensions, with the view towards extending the main result and proving the topological conjecture in the general case. 1. Introduction Let K ⊂ Rn be a compact definable set in an o-minimal structure over R, for example, it may be a semi-algebraic or a subanalytic set. Consider a one-parametric definable family {Sδ}δ>0 of compact subsets of K, defined for all sufficiently small positive δ ∈ R. Definition 1.1. The family {Sδ}δ>0 is called monotone family if the sets Sδ are monotone increasing as δ (cid:38) 0, i.e., Sδ ⊂ Sη for all sufficiently small δ > η > 0. It is well known that there exists a definable triangulation of K (see [6, 12]). In this paper we suggest a more general notion of a definable triangulation of K compatible with the given monotone family {Sδ}δ>0. The intersection of each set Sδ with each open simplex of such a triangulation is a topologically regular cell and is topologically equivalent, in a precise sense, to one of the families in the finite list of model families. Model families are in a natural bijective correspondence with all lex-monotone Boolean functions in dim K Boolean variables (see Figure 1 for the lists of model families and corresponding lex-monotone functions in dimensions 1 and 2). We conjecture that such a triangulation always exist, and we prove the conjecture in the case when dim K ≤ 2 (Theorem 9.13). In the course of achieving this goal, we study a problem that is important on its own, of the existence of a definable cylindrical decomposition of Rn compatible with K such that each cylindrical cell of the decomposition is topologically regular. 2010 Mathematics Subject Classification 14P10, 14P15, 14P25. S. Basu was partially supported by NSF grants CCF-0915954 and CCF-1319080. A. Gabrielov and S. Basu were partially supported by NSF grant DMS-1161629. 1 2 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Cylindrical decomposition is a fundamental construction in o-minimal geometry [6, 12], as well as in semi-algebraic geometry [5]. The elements of a decomposition are called cylindrical cells and are definably homeomorphic to open balls of the corresponding dimensions. By definition, a cylindrical decomposition depends on a chosen linear order of coordinates in Rn. It is implicitly proved in [6, 12] that for a given finite collection of definable sets in Rn there is a linear change of coordi- nates in Rn and a cylindrical decomposition compatible with these sets, such that each cylindrical cell is a topologically regular cell. Without a suitable change of coordinates, the cylindrical cells defined in various proofs of existence of cylindrical decomposition (e.g., in [6, 12]) can fail to be topologically regular (see Example 4.3 in [2]). It remains an open problem, even in the category of semi-algebraic sets, whether there always exists a cylindrical decomposition, with respect to a given order of coordinates, compatible with a given definable bounded set K, such that the cells in the decomposition, contained in K, are topologically regular. We conjecture that such regular cylindrical decompositions always exist, and prove this conjecture in the case when dim K ≤ 2 (in this case a weaker result was obtained in [9]). Topological regularity is a difficult property to verify in general. An important tool that we use to prove it for cylindrical cells is the concept of a monotone cell introduced in [1] (see Definition 2.5 below). It is proved in [1] that every non-empty monotone cell is a topologically regular cell. In fact, everywhere in this paper when we prove that a certain cylindrical cell is topologically regular, we actually prove the stronger property that it is a monotone cell. History and Motivation. An important recurring problem in semi-algebraic ge- ometry is to find tight uniform bounds on the topological complexity of various classes of semi-algebraic sets. Naturally, in o-minimal geometry, definable sets that are locally closed are easier to handle than arbitrary ones. A typical example of this phenomenon can be seen in the well-studied problem of obtaining tight up- per bounds on Betti numbers of semi-algebraic or sub-Pfaffian sets in terms of the complexity of formulae defining them. Certain standard techniques from alge- braic topology (for example, inequalities stemming from the Mayer-Vietoris exact sequence) are directly applicable only in the case of locally closed definable sets. Definable sets which are not locally closed are comparatively more difficult to ana- lyze. In order to overcome this difficulty, Gabrielov and Vorobjov proved in [7] the following result. Suppose that for a bounded definable set S ⊂ K ⊂ Rn in an o-minimal structure over R there is a definable monotone family {Sδ}δ>0 of compact subsets of S such δ Sδ. Suppose also that for each sufficiently small δ > 0 there is a definable family {Sδ,ε}ε>0 of compact subsets of K such that for all ε, ε(cid:48) ∈ (0, 1), ε Sδ,ε. Finally, assume that for all δ(cid:48) > 0 sufficiently smaller than δ, and all ε(cid:48) > 0 there exists an open in K set U ⊂ K such that Sδ ⊂ U ⊂ Sδ(cid:48),ε(cid:48). The main theorem in [7] states that under a certain technical condition on the family {Sδ}δ>0 (called "separability" which will be made precise later), for all ε0 (cid:28) δ0 (cid:28) ε1 (cid:28) δ1 (cid:28) ··· (cid:28) εn (cid:28) δn (where "(cid:28)" stands for "sufficiently smaller than") the compact definable set Sδ0,ε0 ∪ ··· ∪ Sδn,εn is homotopy equivalent to S. that S = (cid:83) if ε(cid:48) > ε then Sδ,ε ⊂ Sδ,ε(cid:48), and Sδ = (cid:84) TRIANGULATIONS OF 2D FAMILIES 3 The separability condition is automatically satisfied in many cases of interest, such as when S is described by equalities and inequalities involving continuous de- finable functions, and the family Sδ is defined by replacing each inequality of the kind P > 0 or P < 0 in the definition of S, by P ≥ δ or P ≤ −δ respectively. How- ever, the property of separability is not preserved under taking images of definable maps (in particular, under blow-down maps) which restricts the applicability of this construction. The following conjecture was made in [7]. Conjecture 1.2. The property that the approximating set Sδ0,ε0 ∪ ··· ∪ Sδn,εn is homotopy equivalent to S remains true even without the separability hypothesis. Conjecture 1.2 would be resolved if one could replace Sδ0,ε0 ∪ ··· ∪ Sδn,εn by a homotopy equivalent union Vδ0,ε0 ∪ ··· ∪ Vδn,εn for another, separable, family {Vδ,ε}δ,ε>0, satisfying the same properties as the family Sδ,ε with respect to S. This motivates the problem of trying to find a finite list of model families inside the standard simplex ∆ such that for each simplex Λ of the triangulation of K, the family {Sδ ∩ Λ}δ>0 is topologically equivalent to one of the (separable or non- separable) model families. Such families {Sδ ∩ Λ}δ>0 are called standard. The main result of this paper is a proof of the existence of a triangulation yielding standard families in the two-dimensional case. As a consequence we obtain a proof of Conjecture 1.2 in the case when dim K ≤ 2. This triangulation presents an independent interest. We will show in Section 4 that there is a bijective correspondence between monotone families {Sδ}δ>0 and non-negative upper semi-continuous definable functions f : K → R, with Sδ = {x ∈ K f (x) ≥ δ}. Then, for a given f , a triangulation into simplices Λ yielding standard families {Λ ∩ Sδ}δ>0 can be interpreted as a topological resolution of singularities of the continuous map graph(f ) → K induced by f , in the sense that we obtain a partition of the domain into a finite number of simplices on each of which the function f behaves in a canonical way up to a certain topological equivalence relation. A somewhat loose analogy in the analytic setting is provided by the "Local Flattening Theorem" [8, Theorem 4.4.]. When f : K → R is the distance function to a singular point x ∈ K, the set Sδ for small δ > 0 becomes the complement to a neighbourhood of x in K, and the boundary ∂Sδ becomes the link of x. Then the triangulation of K, compatible with {Sδ}, can provide a new technique for the study of bi-Lipschitz classification of germs of two-dimensional definable sets [3]. Relation to triangulations of functions and maps. It is well known [6, 12] that con- tinuous definable functions f : K → R, where K is a compact definable subset of Rn, can be triangulated. A simple example (that of the blow-down map corre- sponding to the plane R2 blown up at a point) shows that definable maps which are not functions (i.e., maps of the form f : K → Rm, m ≥ 2) need not be triangulable, and this leads to various difficulties in studying topological properties of definable maps. For example, the question whether a definable map admitting a continuous section, also admits a definable one would have an immediate positive answer if the map was definably triangulable. However, at present this remains a difficult open problem in o-minimal geometry. The version of the topological resolution of singularities described above can be viewed as an alternative to the traditional notion of triangulations compatible 4 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV with a map. Towards this end, we have identified a special class of definable sets and maps, which we call semi-monotone sets and monotone maps respectively (see below for definitions), such that general definable maps could be obtained from these simple ones via appropriate gluing. Relation to preparation theorems. An important line of research in o-minimal ge- ometry has been concentrated around preparation theorems. Given a definable function f : Rn+1 → R, the goal of a preparation theorem (along the lines of clas- sical preparation theorems in algebra and analysis, due to Weierstrass, Malgrange, etc.) is to separate the dependence on the last variable, as a power function with real exponent, from the dependence on the remaining variables. For example, van den Dries and Speissegger [14], following earlier work by Macintyre, Marker and Van den Dries [13], Lion and Rolin [10], proved that in a polynomially bounded o-minimal structure there exists a definable decomposition of Rn into definable cells such that over each cell C the function f can be written as f (x, y) = (y − θC(x))λC gC(x)uC(x, y). where λC ∈ R, while θC, gC, uC are definable functions with uC being a unit. From this viewpoint, the triangulation yielding standard families, could be seen as a topological analogue of a preparation theorem such as the one mentioned above. Allowing the unit uC in the preparation theorem gives additional flexibility which is not available in the situation considered in this paper. Organization of the paper. Although the main results of the paper are proved in the case when dim K ≤ 2, most of the definitions and many technical statements are formulated and proved in the general case. We consider this paper as the first in the series, and will be using these general definitions and statements in future work. The rest of the paper is organized as follows. In Section 2, we recall the definition of monotone cells and some of their key properties needed in this paper. In Section 3, we recall the definition of definable cylindrical decomposition compatible with a finite family of definable subsets of Rn. The notions of "top", "bottom" and "side wall" of a cylindrical cell that are going to play an important role later are also defined in this section. We prove the existence of a cylindrical cell decomposition with monotone cylindrical cells in the case when dim K ≤ 2 (Theorem 3.20). In Section 4, we establish a connection between monotone definable families of compact sets, and super-level sets of definable upper semi-continuous functions. This allows us to include monotone families in the context of cylindrical decompo- sitions. In Section 5, we recall the notion of "separability" introduced in [7], and discuss certain topological properties of monotone families inside regular cells which will serve as a preparation for later results on triangulation. In Section 6 we define the combinatorially standard families and model families. A combinatorially standard family is a combinatorial equivalence class of monotone families inside the standard simplex ∆n. There is a bijective correspondence be- tween the set of all combinatorially standard families and all lex-monotone Boolean functions on {0, 1}n (Definition 6.12). The model families are particular piece-wise linear representatives of the combinatorially standard families (Definition 6.14). After applying a barycentric subdivision to any model family, the monotone family TRIANGULATIONS OF 2D FAMILIES 5 inside each of the sub-simplices of the barycentric sub-division is guaranteed to be separable (Lemma 6.24). In Section 7, we define the notion of topological equivalence and prove the exis- tence of certain "interlacing" homeomorphisms in the two dimensional case. This allows us to prove that in two dimensional case combinatorial equivalence is the same as topological equivalence. Section 8 is devoted to a technical problem of proving the existence of monotone curves (and, more generally, families of monotone curves) connecting any two points inside a monotone cell. Construction of such curves is an essential tool in obtaining a stellar sub-division of a monotone cell into simplices with an additional requirement that the simplices are monotone cells. In Section 9, we prove the existence of a triangulation of two-dimensional com- pact K such that the restriction of the monotone family to each simplex is standard. In Section 10, we prove for any given monotone family {Sδ}δ>0 in two-dimensional compact K the existence of a homotopy equivalent monotone family {Rδ}δ>0 in K and a definable triangulation of K such that the restriction Λ ∩ Rδ to each its simplex Λ is separable. In Section 11, we prove the motivating conjecture of this paper, Conjecture 1.2, in the case when the dimension of the set S is at most two (Theorem 11.4). Acknowledgements. The authors thank the anonymous referee for many helpful remarks. 2. Monotone cells In [2, 1] the authors introduced the concepts of a semi-monotone set and a monotone map. Graphs of monotone maps are generalizations of semi-monotone sets, and will be called monotone cells in this paper (see Definition 2.5 below). Definition 2.1. Let Lj,c := {x = (x1, . . . , xn) ∈ Rn xj = c} for j = 1, . . . , n, and c ∈ R. Each intersection of the kind S := Lj1,c1 ∩ ··· ∩ Ljm,cm ⊂ Rn, where m = 0, . . . , n, 1 ≤ j1 < ··· < jm ≤ n, and c1, . . . , cm ∈ R, is called an affine coordinate subspace in Rn. In particular, the space Rn itself is an affine coordinate subspace in Rn. We now define monotone maps. The definition below is not the one given in [1], We first need a preliminary definition. For a coordinate subspace L of Rn we but equivalent to it as shown in [1, Theorem 9]. denote by ρL : Rn → L the projection map. Definition 2.2. Let a bounded continuous map f = (f1, . . . , fk) defined on an open bounded non-empty set X ⊂ Rn have the graph Y := {(x, f1(x), . . . , fk(x)) ∈ Rn+k x = (x1, . . . , xn) ∈ X}. We say that f is quasi-affine if for any coordinate subspace L of Rn+k, the restriction ρLY of the projection is injective if and only if the image ρL(Y ) is n-dimensional. Definition 2.3. Let a bounded continuous quasi-affine map f = (f1, . . . , fk) de- fined on an open bounded non-empty set X ⊂ Rn have the graph Y ⊂ Rn+k. We say that the map f is monotone if for each affine coordinate subspace S in Rn+k the intersection Y ∩ S is connected. 6 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Notation 2.4. Let the space Rn have coordinate functions x1, . . . , xn. Given a subset I = {xj1, . . . , xjm} ⊂ {x1, . . . , xn}, let W be the linear subspace of Rn where all coordinates in I are equal to zero. By a slight abuse of notation we will denote by span{xj1, . . . , xjm} the quotient space Rn/W . Similarly, for any affine coordinate subspace S ⊂ Rn on which all the functions xj ∈ I are constant, we will identify S with its image under the canonical surjection to Rn/W . Again, by a slight abuse of notation, span{x1, . . . xi}, where i ≤ n, will be denoted by Ri. Definition 2.5 ([2, 1]). A set Y ⊂ Rn = span {x1, . . . , xn} is called a monotone cell if it is the graph of a monotone map f : X → span H, where H ⊂ {x1, . . . , xn} and X ⊂ span({x1, . . . , xn} \ H) . In a particular case, when H = ∅ (i.e., span H coincides with the origin) such a graph is called a semi-monotone set. We refer the reader to [2], Figure 1, for some examples of monotone cells in R2 (actually, semi-monotone sets), as well as some counter-examples. In particular, it is clear from the examples that the intersection of two monotone cells in the plane is not necessarily connected and hence not a monotone cell. Notice that any bounded convex open subset X of Rn is a semi-monotone set, while the graph of any linear function on X is a monotone cell in Rn+1. The following statements were proved in [1]. Proposition 2.6 ([1], Theorem 1). Every monotone cell is a topologically regular cell. Proposition 2.7 ([1], Corollary 7, Theorem 11). Let X ⊂ Rn be a monotone cell. Then (i) for every coordinate xi in Rn and every c ∈ R, each of the intersections X ∩{xi = c}, X ∩{xi < c}, X ∩{xi > c} is either empty or a monotone cell; (ii) Let Y ⊂ X be a monotone cell such that dim Y = dim X − 1 and ∂Y ⊂ ∂X. Then X \ Y is a disjoint union of two monotone cells. Proposition 2.8 ([1], Theorem 10). Let X ⊂ Rn be a monotone cell. Then for any coordinate subspace L in Rn the image ρL(X) is a monotone cell. Let Rn >0 := {x = (x1, . . . , xn) ∈ Rn xi > 0 for all i = 1, . . . , n}, and X ⊂ Rn >0. Lemma 2.9. Consider the following two properties, which are obviously equivalent. (i) For each x ∈ X, the box Bx := {(y1, . . . , yn) ∈ Rn 0 < y1 < x1, . . . , 0 < yn < xn} is a subset of X. (ii) For each x ∈ X and each j = 1, . . . , n, the interval Ix,j := {(y1, . . . , yn) ∈ Rn 0 < yj < xj, yi = xi for i (cid:54)= j} is a subset of X. If X is open and bounded, then either of the properties (i) or (ii) implies that X is semi-monotone. If an open and bounded subset Y ⊂ Rn >0 also satisfies the conditions (i) or (ii), then both X ∪ Y and X ∩ Y satisfy these conditions, and hence are semi-monotone. Proof. The proof of semi-monotonicity of X is by induction on n, the base for n = 0 being obvious. According to Corollary 1 in [1], it is sufficient to prove that X is connected, and that for every k, 1 ≤ k ≤ n and every c ∈ R the intersection TRIANGULATIONS OF 2D FAMILIES 7 If an open and bounded Y ⊂ Rn X ∩ {xk = c} is semi-monotone. The set X is connected because for every two points x, z ∈ X the boxes Bx and Bz are connected and Bx ∩ Bz (cid:54)= ∅. Since the property (ii) is true for the intersection X ∩ {xk = c}, by the inductive hypothesis this intersection is semi-monotone, and we proved semi-monotonicity of X. >0 satisfies the conditions (i) or (ii), then both sets X ∪ Y and X ∩ Y obviously also satisfy these conditions, hence are semi- (cid:3) monotone. Definition 2.10. Let Y ⊂ span {x1, . . . , xn} be a monotone cell and f : Y → span {y1, . . . , yk} a continuous map. The map f is called monotone on Y if its graph Z ⊂ span {x1, . . . , xn, y1, . . . , yk} is a monotone cell. In the case k = 1, the map f is called a monotone function on Y . Remark 2.11. Let Y be a monotone cell and L a coordinate subspace such that ρLY is injective. Then, according to Theorem 7 and Corollary 5 in [1], Y is the graph of a monotone map defined on ρL(Y ). 3. Cylindrical decomposition We now define, closely following [12], a cylindrical cell and a cylindrical cell decomposition. Definition 3.1. When n = 0, there is a unique cylindrical cell, 0, in Rn. Let n ≥ 1 and (i1, . . . , in) ∈ {0, 1}n. A cylindrical (i1, . . . , in)-cell is a definable set in Rn obtained by induction on n as follows. A (0)-cell is a single point x ∈ R, a (1)-cell is one of the intervals (x, y) or (−∞, y) or (x,∞) or (−∞,∞) in R. Suppose that (i1, . . . , in−1)-cells, where n > 1, are defined. An (i1, . . . , in−1, 0)- cell (or a section cell) is the graph in Rn of a continuous definable function f : C → R, where C is a (i1, . . . , in−1)-cell. Further, an (i1, . . . , in−1, 1)-cell (or a sector cell) is either a set C × R, or a set {(x, t) ∈ C × R f (x) < t < g(x)}, or a set {(x, t) ∈ C × R f (x) < t}, or a set {(x, t) ∈ C × R t < g(x)}, where C is a (i1, . . . , in−1)-cell and f, g : C → R are continuous definable functions such that f (x) < g(x) for all x ∈ C. In the case of a sector cell C, the graph of f is called the bottom of C, and the graph of g is called the top of C. In the case of a section (i1, . . . , in−1, 0)-cell C, let k be the largest number in {1, . . . , n − 1} with ik = 1. Then C is the graph of a map C(cid:48) → Rn−k, where C(cid:48) is a sector (i1, . . . , ik)-cell. The pre-image of the bottom of C(cid:48) by ρRkC is called the bottom of C, and the pre-image of the top of C(cid:48) by ρRkC is called the top of C. Let CT be the top and CB be the bottom of a cell C. The difference C \ (C ∪ CT ∪ CB) is called the side wall of C. In some literature (e.g., in [6]) section cells are called graphs, while sector cells -- bands. Note that in the case of a sector cell, the top and the bottom are cylindrical section cells. On the other hand, the top or the bottom of a section cell C is not necessarily a graph of a continuous function since it may contain blow-ups of the function ϕ of which C is the graph. Consider the following example. Example 3.2. Let n = 3, C(cid:48) = {(x, y) x ∈ (−1, 1),x < y < 1}, and ϕ(x, y) = x/y. In this example, the bottom of the cell C, defined as the graph of ϕC(cid:48), is not the graph of a continuous function. 8 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Lemma 3.14 below provides a condition under which the top and the bottom of a cylindrical section cell are cylindrical section cells. When it does not lead to a confusion, we will sometimes drop the multi-index (i1, . . . , in) when referring to a cylindrical cell. Lemma 3.3. Let C ⊂ Rn be a cylindrical (i1, . . . , ik−1, 0, ik+1, . . . , in)-cell. Then C(cid:48) := ρspan{x1,...,xk−1,xk+1,...,xn}(C) is a cylindrical (i1, . . . , ik−1, ik+1, . . . , in)-cell, and C is the graph of a continuous definable function on C(cid:48). Proof. Proof is by induction on n with the base case n = 1 being trivial. ρRk (C) is the graph of a continuous function ϕ : ρRk−1(C) → span{xk}. span{xn}. Thus C(cid:48) is the graph of the continuous function By the definition of a cylindrical (i1, . . . , ik−1, 0, ik+1, . . . , in)-cell, the image If C is a section cell, then it is the graph of a continuous function f : ρRn−1 (C) → f ◦ (x1, . . . , xk−1, ϕ(x1, . . . , xk−1), xk+1, . . . , xn−1) on ρspan{x1,...,xk−1,xk+1,...,xn−1}(C). The latter is a cylindrical cell by the induc- tive hypothesis, since ρRn−1 (C) is a cylindrical (i1, . . . , ik−1, 0, ik+1, . . . , in−1)-cell. Hence C(cid:48) is a cylindrical cell, being the graph of a continuous function on a cylin- drical cell. By the inductive hypothesis, ρRn−1 (C) is the graph of a continuous function h on ρspan{x1,...,xk−1,xk+1,...,xn−1}(C). The cell C is the graph of the con- tinuous function f ◦ h ◦ ρRn−1C(cid:48) on C(cid:48). If C is a sector cell, then let f, g : ρRn−1(C) → span{xn} be its bottom and its top functions. Thus, C(cid:48) is a sector between graphs of functions f ◦ (x1, . . . , xk−1, ϕ, xk+1, . . . , xn−1) and g ◦ (x1, . . . , xk−1, ϕ, xk+1, . . . , xn−1) B B on C(cid:48) on C(cid:48) B be its bottom and C(cid:48) on the cylindrical cell ρspan{x1,...,xk−1,xk+1,...,xn−1}(C). Hence C(cid:48) is a cylindrical cell. Let C(cid:48) T its top. The cell C is the graph of the continuous function on C(cid:48) since the bottom of C is the graph of the continuous function f ◦ h ◦ ρRn−1C(cid:48) B, while the top of C is the graph of the continuous function g ◦ h◦ ρRn−1C(cid:48) T . Hence each intersection of C with the straight line parallel to xn-axis projects bijectively by ρspan{x1,...,xk−1,xk+1,...,xn} onto an intersection of (cid:3) C(cid:48) with the straight line parallel to xn-axis. Lemma 3.4. Let C be a two-dimensional cylindrical cell in Rn such that C is the graph of a quasi-affine map (see Definition 2.2). Then the side wall W of C has exactly two connected components each of which is either a single point or a closed curve interval. Proof. Let C be a cylindrical (i1, . . . , in)-cell and ij the first 1 in the list i1, . . . , in. The image of the projection, ρspan{xj}(C) is an interval (a, b). Consider the disjoint sets A := (ρspan{xj}C)−1(a) and B := (ρspan{xj}C)−1(b). Then W = A ∪ B, and A (respectively, B) is the Hausdorff limit of the intersections C ∩{xj = c} as c (cid:38) a (respectively, c (cid:37) b). Since C is the graph of a quasi-affine map, for every c ∈ (a, b) the intersection C ∩ {xj = c} is a curve interval which is also the graph of a quasi- affine map. Hence each of the Hausdorff limits A, B is either a single point or a (cid:3) closed curve interval. TRIANGULATIONS OF 2D FAMILIES 9 Definition 3.5. A cylindrical cell decomposition of Rn is a finite partition of Rn into cylindrical cells defined by induction on n as follows. When n = 0 the cylindrical cell decomposition of Rn consists of a single point. Let n > 0. For a partition D of Rn into cylindrical cells, let D(cid:48) be the set of all cells C(cid:48) ⊂ Rn−1 such that C(cid:48) = ρRn−1(C) for some cell C of D. Then D is a cylindrical cell decomposition of Rn if D(cid:48) is a cylindrical cell decomposition of Rn−1. In this case we call D(cid:48) the cylindrical cell decomposition of Rn−1 induced by D. (i) A cylindrical cell decomposition D of Rn is compatible with Definition 3.6. a definable set X ⊂ Rn if for every cell C of D either C ⊂ X or C ∩ X = ∅. (ii) A cylindrical cell decomposition D(cid:48) of Rn is a refinement of a decomposition D of Rn if D(cid:48) is compatible with every cell of D. Remark 3.7. It is easy to prove that for a cylindrical cell decomposition D of Rn compatible with X ⊂ Rn, the cylindrical cell decomposition D(cid:48) of Rn−1 induced by D is compatible with ρRn−1(X). Remark 3.8. Let D be a cylindrical cell decomposition of Rn and C be a cylindrical cell in D such that the dimension of C(cid:48) := ρR1(C) equals 0, i.e., C(cid:48) = {c} ⊂ R1 for some c ∈ R. It follows immediately from the definitions that D is compatible with the hyperplane {x1 = c} in Rn, and the set of all cells of D, contained in {x1 = c}, forms a cylindrical cell decomposition D(cid:48) of the hyperplane {x1 = c} when the latter is identified with Rn−1. Moreover, any refinement of D(cid:48) leads to a refinement of D. Proposition 3.9 ([12], Theorem 2.11). Let A1, . . . , Am ⊂ Rn be definable sets. There is a cylindrical cell decomposition of Rn compatible with each of the sets Ai. Definition 3.10. Let A1, . . . , Am ⊂ Rn be definable bounded sets. We say that a cylindrical cell decomposition D of Rn is monotone with respect to A1, . . . , Am if D i Ai is a monotone cell. Lemma 3.11. Let D be a cylindrical cell decomposition of Rn, and c ∈ R. Then the collection of sets is compatible with A1, . . . , Am, and each cell contained in(cid:83) C ∩ {x1 = c}, C ∩ {x1 < c} and C ∩ {x1 > c} for all cylindrical cells C of D forms a refinement D(cid:48) of D. Moreover, for any cylindrical cell C of D which is a monotone cell, all cells of D(cid:48) contained in C are monotone cells. Proof. A straightforward induction on n, taking into account that intersections of a monotone cell with a hyperplane or a half-space are monotone cells (Proposi- (cid:3) tion 2.7). Definition 3.12. A cylindrical cell decomposition D of Rn satisfies the frontier condition if for each cylindrical cell C its frontier C \ C is a union of cells of D. It is clear that if a cylindrical cell decomposition D of Rn satisfies the frontier condition, then the induced decomposition (see Definition 3.5) also satisfies the frontier condition. It is also clear that the side wall of each cell is a union of some cells in D of smaller dimensions. We next prove that the tops and the bottoms of cells in a cylindrical decomposition satisfying the frontier condition are each cells of the same decomposition. Before proving this claim, we first consider an example. 10 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Example 3.13. One can easily check that there is a cylindrical cell decomposition of R3 containing the cell C from Example 3.2 and the cells {−1 < x < 1, y = x}×R and {−1 < x < 1, y = 1} × R. This decomposition does not satisfy frontier condition. The following lemma implies that in fact C cannot be a cell in any cylindrical decomposition that satisfies the frontier condition. Lemma 3.14. Let D be a cylindrical decomposition of Rn satisfying the frontier condition. Then, the top and the bottom of each cell of D of Rn are cells of D. Proof. Let CT be the top of a cylindrical cell C of D. Suppose C is a sector cell of D. By the definition of a cylindrical cell decomposition, ρRn−1(C) is a cylindrical cell. Since CT is contained in a union of some cells of D, and ρRn−1(C) = ρRn−1 (CT ), it is a cylindrical cell of D. Suppose now C is a section cell. Then C is the graph of a map C(cid:48) → Rn−k, where C(cid:48) is a sector cell in the induced cylindrical cell decomposition D(cid:48) of Rk. Applying T is a cylindrical cell of D(cid:48). By the above argument to C(cid:48) we conclude that its top C(cid:48) the frontier condition, CT is a union of some (k − 1)-dimensional cells of D. This is because the pre-image of a cell in a cylindrical cell decomposition is always a union of cells, and CT consists of the cylindrical cells in the pre-image ρ−1Rk (C(cid:48) T ) which are contained in ¯C, due to the frontier condition. As C \ C is (k − 1)-dimensional, all cells in CT are (k − 1)-dimensional and project surjectively onto C(cid:48) T , and thus they are disjoint graphs of continuous functions over C(cid:48) T . The proof for the bottom of C is similar. Finally, the closure of a graph of a definable function is a graph of a definable function everywhere except, possibly over a subset of codimension at least 2. Thus C cannot contain two disjoint graphs over the (k − 1)-dimensional cell C(cid:48) T . (cid:3) Definition 3.15. Let D be a cylindrical decomposition of Rn satisfying the frontier condition. By Lemma 3.14, the top and the bottom of each cell of D are cells of D. For a cell C of D define vertices of C by induction as follows. If dim C = 0 then C itself is its only vertex. Otherwise, the set of vertices of C is the union of the sets of vertices of its top and of its bottom. Lemma 3.16. Let X be an open subset in R2, and f = (f1, . . . , fk) : X → Rk a quasi-affine map. If each component f(cid:96) is monotone, then the map f itself is monotone. Proof. Without loss of generality, assume that none of the functions f1, . . . , fk is constant. Let X ⊂ span{x1, x2} and let F ⊂ span{x1, x2, y1, . . . , yk} be the graph of f . Note that for each i = 1, . . . , k the graph Fi ⊂ span{x1, x2, yi} of the function fi coincides with the image of the projection of F to span{x1, x2, yi}, and this projection is a homeomorphism. By Theorem 9 in [1], it is sufficient to prove that the intersection of F with any affine coordinate subspace of codimension 1 or 2 is connected. First consider the case of codimension 1. For every i = 1, . . . , k and every b ∈ R the image of the projection of F ∩ {yi = b} to span{x1, x2, yi} coincides with Fi ∩ {yi = b}, and this projection is a homeomorphism. Since fi is monotone, the intersection Fi ∩ {yi = b} is connected, hence the intersection F ∩ {yi = b} is also connected. For every i = 1, . . . , k, every j = 1, 2, and every c ∈ R the image of the projection of F ∩ {xj = c} to span{x1, x2, yi} coincides with Fi ∩ {xj = c}, thus F ∩ {xj = c} is connected since Fi ∩ {xj = c} is connected. TRIANGULATIONS OF 2D FAMILIES 11 Now consider the case of codimension 2. The intersection F∩{x1 = c1, x2 = c2}, for any c1, c2 ∈ R is obviously a single point. The intersection F ∩ {xj = c}, for any j = 1, 2 and c ∈ R is the graph of a continuous map on an interval X ∩ {xj = c}, taking values in span{y1, . . . , yk} and this map is quasi-affine. Hence each It follows that the intersection component of this map is a monotone function. F∩{xj = c, yi = b} for every i = 1, . . . k and every b ∈ R is either empty, or a single point, or an interval. Finally, the intersection F∩{yi = bi}, for any i = 1, . . . , k and bi ∈ R is the graph of a continuous map on the curve X ∩ {fi = bi} (this curve is the graph of a monotone function), taking values in span{y1, . . . , yi−1, yi+1, . . . , yk}, and this map is quasi-affine. Hence each component of this map is a monotone It follows that the intersection F ∩ {yi = bi, y(cid:96) = b(cid:96)} for every (cid:96) = function. 1, . . . , i − 1, i + 1, . . . , k and every b(cid:96) ∈ R is either empty, or a single point, or an (cid:3) interval. Remark 3.17. Let V1, . . . , Vk be bounded definable subsets in Rn. According to Section (2.19) of [12] (see also Section 4 of [2]), there is a cylindrical cell decompo- sition of Rn compatible with each of V1, . . . , Vk, with cylindrical cells being van den Dries regular. One can prove that one- and two-dimensional van den Dries regular cells are topologically regular cells. Hence, in case dim(V1 ∪ ··· ∪ Vk) ≤ 2, there exists a cylindrical cell decomposition of Rn, compatible with each of Vi, such that cylindrical cells contained in V1 ∪···∪ Vk are topologically regular. This covers the greater part of the later work [9]. Our first goal will be to generalize these results by proving the existence of a cylindrical cell decomposition of Rn, monotone with respect to V1, . . . , Vk. Lemma 3.18. Let f : X → R be a quasi-affine function on an open bounded domain X ⊂ R2. Then there is a cylindrical cell decomposition of R2 compatible with X, obtained by intersecting X with straight lines of the kind {x1 = c} ⊂ R2, and half-planes of the kind {x1 ≶ c} ⊂ R2, where c ∈ R, such that the restriction fB to each cell B ⊂ X is a monotone function. Proof. Every non-empty intersection of the kind X ∩ {x1 = c}, where c ∈ R, is a finite union of pair-wise disjoint intervals. Let I(c) be family of such intervals. Let γ := {(x1, x2) ∈ X x2 is an endpoint of an interval in I(x1)}. Let the real numbers c1, . . . , ct be such that the intersection γ ∩ {ci < x1 < ci+1}, for each 1 ≤ i < t, is a disjoint union of monotone 1-dimensional cells with the images under ρR1 coinciding with (ci, ci+1). By Theorem 1.7 in [2], the intersection X ∩ {ci < x1 < ci+1} for every 1 ≤ i < t is a disjoint union of one- and two- dimensional semi-monotone sets. By the definition of γ, the intersection X ∩{x1 = ci} for every 1 ≤ i < t is a disjoint union of intervals. We have constructed a cylindrical decomposition D of R2 compatible with X and having semi-monotone cylindrical cells. Take any two-dimensional cylindrical cell C in D. Then ρR1 (C) = (ci, ci+1) for some 1 ≤ i < t. Since f is quasi-affine, its restriction fC is also quasi-affine, hence (cf. the second part of Remark 7 in [1]) fC is either strictly increasing in or strictly decreasing in or independent of each of the variables x1, x2. This also implies that the restriction of fC to any non-empty C ∩ {x2 = c}, where c ∈ R, is a monotone function. Let real numbers b1, . . . , br ∈ (ci, ci+1) be such that the restrictions of both functions inf x2 f and supx2 f to the interval (bj, bj+1) ⊂ R1 for each 1 ≤ j < r 12 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV are monotone functions. Note that the intersection of {bj < x1 < bj+1} with any two-dimensional cylindrical cell in D is also a cylindrical (and semi-monotone) cell, in particular the intersection B := C∩{bj < x1 < bj+1} is such a cell. By Theorem 3 in [1], the restriction fB is a monotone function. We have proved that there exists a cylindrical decomposition of R2 monotone with respect to X (in particular, the two-dimensional cells of of the decomposition, contained in X, are semi-monotone), (cid:3) and such that the restrictions of f to each cell is a monotone function. Lemma 3.19. Let V1, . . . , Vk be bounded definable subsets in Rn with dim Vi ≤ 2 for each i = 1, . . . , k. Then there is a cylindrical cell decomposition of Rn, compatible 1≤i≤k Vi is the with every Vi, such that every cylindrical cell contained in V :=(cid:83) graph of a quasi-affine map. Proof. Let W be the smooth two-dimensional locus of V . Stratify W with respect to critical points of its projections to 2- and 1-dimensional coordinate subspaces. More precisely, let Wi,j ⊂ W , for 1 ≤ i < j ≤ n, be the set of all locally two- dimensional critical points of the projection map ρi,j : W → span{xi, xj}, and i,j, for 1 ≤ i < j ≤ n, be the set of all critical points of the projection map W (cid:48) ρi,j : W → span{xi, xj}, having local dimension at most 1. Consider a cylindrical decomposition D of Rn compatible with each of V1, . . . , Vk, Wi,j, W (cid:48) i,j, Note that C ∩ W (cid:48) i,j = ∅ for all pairs 1 ≤ i < j ≤ n since dim(W (cid:48) where 1 ≤ i < j ≤ n. Let C be a two-dimensional cylindrical cell in this decompo- sition. Then C is the graph of a smooth map defined on a cylindrical cell in some 2-dimensional coordinate subspace. We now prove that this map is quasi-affine. i,j) < 2 while D is compatible with W (cid:48) i,j. Since D is compatible with every Wi,j, if C ∩ Wi,j (cid:54)= ∅ then C ⊂ Wi,j. Therefore, if C ∩ Wi,j (cid:54)= ∅ for a pair i, j, 1 ≤ i < j ≤ n, then dim(ρi,j(C)) ≤ 1. Suppose now that C∩Wi,j = ∅ for a pair i, j, 1 ≤ i < j ≤ n. Then dim(ρi,j(C)) = 2. Assume that the projection ρi,jC is not injective, i.e., there are distinct points a = (a1, . . . , an), b = (b1, . . . , bn) ∈ C such that ρi,j(a) = ρi,j(b). There exists (cid:96) ∈ {1, . . . , n} \ {i, j} such that a(cid:96) (cid:54)= b(cid:96). The set ρi,j,(cid:96)(C) ⊂ span{xi, xj, x(cid:96)} is two-dimensional, smooth, connected, and contains points (ai, aj, a(cid:96)), (ai, aj, b(cid:96)). Hence there is a critical point of its projection to the subspace span{xi, xj}. This contradicts the condition C ∩ (Wi,j ∪ W (cid:48) i,j) = ∅. It follows that C is the graph of a quasi-affine map. Finally, D can be refined so that any 1-dimensional cylindrical cell B of the of the refinement is a monotone (hence quasi-affine) 1-dimensional cell. Indeed, if dim ρR1 (B) = 1, then a refinement exists by Lemma 3.11. Otherwise, B is contained in an affine subspace {x1 = c} for some c ∈ R and a refinement exists by (cid:3) Remark 3.8. Theorem 3.20. Let V1, . . . , Vk be bounded definable subsets in Rn with dim Vi ≤ 2 for each i = 1, . . . , k. Then there is a cylindrical cell decomposition of Rn satisfying the frontier condition, and monotone with respect to V1, . . . , Vk. Proof. First, using Lemma 3.19, construct a cylindrical cell decomposition D of Rn, compatible with every Vi, such that each cylindrical cell contained in V is the graph of a quasi-affine map. TRIANGULATIONS OF 2D FAMILIES 13 cell decomposition with every cell contained in V :=(cid:83) We now construct, inductively on n, a refinement of D, which is a cylindrical 1≤i≤k Vi being a monotone cell. The base case n = 1 is straightforward. Suppose the construction exists for all dimensions less than n. Each cylindrical cell X in D such that dim ρR1 (X) = 0 belongs to a cylindrical cell decomposition in the (n−1)-dimensional affine subspace {x1 = c} for some c ∈ R, and in this subspace the refinement can be carried out by the inductive hypothesis. According to Remark 3.8, this refinement is also a refinement of D. Now let X be a cylindrical cell in D, contained in V , with dim ρR1(X) = 1. If dim X = 1, then X, being quasi-affine, is already a monotone cell. Suppose dim X = 2. Let α be the smallest number among {1, . . . , n} such that X(cid:48) := ρspan{x1,xα}(X) is two-dimensional. Then X is a graph of a quasi-affine map f = (f1, . . . , fn−2) defined on X(cid:48). Since f is quasi-affine, each fj is quasi-affine too. By Lemma 3.18, for each fj there exists a cylindrical decomposition of span{x1, xα}, compatible with X(cid:48), obtained by intersecting X(cid:48) with straight lines of the kind {x1 = c} ⊂ span{x1, xα}, and half-planes of the kind {x1 ≶ c} ⊂ span{x1, xα}, where c ∈ R, such that the restriction fjY (cid:48) for each cylindrical semi-monotone cell Y (cid:48) ⊂ X(cid:48) is a monotone function. According to Lemma 3.11, the intersections of all cylindrical cells in D with {x1 = c} or {x1 ≶ c} form a cylindrical cell decomposition. Performing such a refinement for each fj we obtain a cylindrical cell decomposition of X(cid:48) into cylindrical cells Y (cid:48)(cid:48), such that the restriction fY (cid:48)(cid:48) is a monotone map by Lemma 3.16. Therefore all elements of the resulting cylindrical cell decomposition, contained in X, are monotone cells. Decomposing in this way each two-dimensional set X of D we obtain a refinement D(cid:48) of D which is a cylindrical cell decomposition with monotone cylindrical cells. It remains to construct a refinement of the cylindrical cell decomposition D(cid:48) satisfying the frontier condition. Let X be a two-dimensional cylindrical cell in D(cid:48). Since X is a monotone cell, its boundary ∂X is homeomorphic to a circle. Let U be a partition of ∂X into points and curve intervals so that U is compatible with all 1-dimensional cylindrical cells of D(cid:48), and each curve interval in U is a monotone cell. Let c = (c1, . . . , cn) be point (0-dimensional element) in U such that c1 is not a 0-dimensional cell in the cylindrical decomposition induced by D(cid:48) on R1. By Lemma 3.11, intersections of the cylindrical cells of D(cid:48) with {x1 = c1} or {x1 ≶ c1} form the refinement of D(cid:48) with cylindrical cells remaining to be monotone cells. Let T ∈ U be one of monotone curve intervals having c as an endpoint. If T is a subset of a two-dimensional cylindrical cell Z of D(cid:48), then T divides Z into two two-dimensional cylindrical cells, hence by Theorem 11 in [1], these two cells are monotone cells. Obviously, adding T to the decomposition, and replacing one two-dimensional cell Z (if it exists) by two cells, we obtain a refinement of D(cid:48). Let c = (c1, . . . , cn) be point in U such that (c1, . . . , ci−1), where i < n, is a 0-dimensional cell in the cylindrical decomposition induced by D(cid:48) on Ri−1, while (c1, . . . , ci) is not a 0-dimensional cell in the cylindrical decomposition induced by D(cid:48) on Ri. In this case we apply the same construction as in the previous case, replacing Rn by {x1 = c1, . . . , xi−1 = ci−1}. By Remark 3.8, the refinement in {x1 = c1, . . . , xi−1 = ci−1} is also a refinement of D(cid:48). 14 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Application of this procedure to all two-dimensional cells X of D(cid:48), all c and all T completes the construction of a refinement of D(cid:48) which satisfies the frontier (cid:3) condition. Corollary 3.21. Let U1, . . . , Um be bounded definable subsets in Rn with dim Ui ≤ 2 for each i = 1, . . . , m and let A be a cylindrical decomposition of Rn. Then there is a refinement of A, satisfying the frontier condition, and monotone with respect to U1, . . . , Um. Proof. Apply Theorem 3.20 to the family V1, . . . , Vk consisting of sets U1, . . . , Um and all cylindrical cells of the decomposition A. (cid:3) The following example shows that there may not exist a cylindrical cell decom- position of R3 compatible with a two-dimensional definable subset, such that each component of the side wall of each two-dimensional cell is a one-dimensional cell of this decomposition. We will call the latter requirement the strong frontier condition. Example 3.22. Let V = {x > y > 0, z > 0, y = xz} and V (cid:48) = {x > y > 0, z > 0, y = 2xz} be two cylindrical cells in R3. Then any cylindrical decomposition of R3 compatible with V and V (cid:48) and satisfying the strong frontier condition, must be compatible with two intervals I1 := {x = y = 0, 0 ≤ z ≤ 1/2} and I2 := {x = y = 0, 1/2 ≤ z ≤ 1} on the z-axis, and the point v = (0, 0, 1/2). Observe that the interval I := {x = y = 0, 0 ≤ z ≤ 1} is the (only) 1-dimensional component of the side wall of V , while I1 is the (only) 1-dimensional component of the side wall of V (cid:48). In order to satisfy the strong frontier condition, we have to partition V into cylindrical cells so that there is a 1-dimensional cell γ ⊂ V such that v = γ ∩ I. Then the tangent at the origin of the projection β := ρR2 (γ) would have slope 1/2. The lifting γ(cid:48) of β to V (cid:48) (i.e., γ(cid:48) := (ρR2V (cid:48))−1(β)) would satisfy the condition v(cid:48) = γ(cid:48) ∩ I, where v(cid:48) = (0, 0, 1/4). (Note that the tangent to γ at v or the tangent to γ(cid:48) at v(cid:48) may coincide with the z-axis.) The point v(cid:48) must be a 0-dimensional cell of the required cell decomposition. Iterating this process, we obtain an infinite sequence of points (0, 0, 2−k), for all k > 0, on I, all being 0-dimensional cells of a cylindrical cell decomposition. This is a contradiction. Lemma 3.23. Let X be a cylindrical sector cell in Rn with respect to the ordering x1, . . . , xn of coordinates. Suppose that the top and the bottom of X are monotone cells. Then X itself is a monotone cell. Proof. Let X(cid:48) := ρRn−1(X). Then X = {(x1, . . . , xn) ∈ Rn x := (x1, . . . , xn−1) ∈ X(cid:48), f (x) < xn < g(x)}, where f, g : X(cid:48) → R are monotone functions, having graphs F and G respectively. Note that ρRn−1(F ) = ρRn−1(G) = X(cid:48). According to Theorem 10 in [1], X(cid:48) is monotone cell. It easily follows from Theorem 9 in [1] that for any a ∈ R the cylinder C := (X(cid:48) × R) ∩ {−a < xn < a} is a monotone cell. Choose a so that −a < inf xn f and a > supxn g. Then we have the following inclusions: F ⊂ C, G ⊂ C, ∂F ⊂ ∂C and ∂G ⊂ ∂C. By Theorem 11 in [1], the set X is a monotone cell, being a connected component of C \ (F ∪ G). (cid:3) The following statement is a generalization of the main result of [9]. TRIANGULATIONS OF 2D FAMILIES 15 Corollary 3.24. Let U1, . . . , Uk be bounded definable subsets in Rn, with dim Ui ≤ 3. Then there is a cylindrical decomposition of Rn, compatible with each Ui, such that (i) for p ≤ 2 each p-dimensional cell X ⊂ Rn of the decomposition, contained in (ii) each 3-dimensional sector cell X ⊂ Rn of the decomposition, contained in U , U :=(cid:83) i Ui, is a monotone cell; is a monotone cell; (iii) if n = 3, then each 3-dimensional cell, contained in U , is a semi-monotone set and all cells of smaller dimensions, contained in U , are monotone cells. Proof. Construct a cylindrical decomposition D of Rn compatible with each Ui, and let V1, . . . , Vr be all 0-, 1- and 2-dimensional cells contained in U . Using Theorem 3.20 obtain a cylindrical decomposition D(cid:48) of Rn monotone with respect to V1, . . . , Vr. Observe that the decomposition D(cid:48) is a refinement of D and hence is compatible with each Ui. Therefore (i) is satisfied. Since for each 3-dimensional sector cell X, contained in U , its top and its bottom are monotone cells, Lemma 3.23 implies that X is itself a monotone cell, and thus (ii) is satisfied. If n = 3, then every 3-dimensional cell in D is a sector cell, which implies (iii). (cid:3) 4. Monotone families as superlevel sets of definable functions Convention 4.1. In what follows we will assume that for each monotone family {Sδ} in a compact definable set K ⊂ Rn (see Definition 1.1) there is δ1 > 0 such that Sδ = ∅ for all δ > δ1. Lemma 4.2. Let K ⊂ Rn be a compact definable set, and {Sδ}δ>0 a monotone de- finable family of compact subsets of K. There exists δ0 > 0, such that the monotone definable family {S(cid:48) (cid:40) δ}δ>0 defined by Sδ ∅ S(cid:48) δ = for for 0 < δ ≤ δ0 δ > δ0 δ}. Then, has the following property. For each x ∈ K let Mx := {δ ∈ R>0 x ∈ S(cid:48) either Mx = ∅, or Mx = (0, bx] for some bx > 0. Proof. By Hardt's theorem for definable families [6, Theorem 5.22], there exists δ0 > 0, and a fiber-preserving homeomorphism H : (0, δ0] × Sδ0 → S(0,δ0], where for any subset I ⊂ R, SI := {(δ, x) δ ∈ I, x ∈ Sδ} ⊂ Rn+1. Let x ∈ K be such that (cid:17)(cid:111) Mx is not empty. Then, there exists c, 0 < c ≤ δ0, such that x ∈ S(cid:48) Mx = (0, c] ∪ {δ ∈ [c, δ0] x ∈ S(cid:48) (cid:17)(cid:111) Now S(cid:48) δ ∈ [c, δ0] (δ, x) ∈ H−1(cid:16) δ} = (0, c] ∪(cid:110) [c,δ0] is compact, and hence H−1(cid:16) (cid:17) δ ∈ [c, δ0] (δ, x) ∈ H−1(cid:16) (cid:110) is projection of a compact set is compact, (cid:3) compact as well. Convention 4.3. In what follows we identify two families {Sδ} and {Vδ} if Sδ = Vδ for small δ > 0. In particular, the families {Sδ} and {S(cid:48) δ} from Lemma 4.2 will be identified. is also compact, and since the c. Then, S(cid:48) S(cid:48) [c,δ0] S(cid:48) [c,δ0] [c,δ0] . 16 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV δ}, if there is δ > 0 with x ∈ S(cid:48) With any monotone definable family {Sδ}δ>0 of compact sets contained in a compact definable set K (see Definition 1.1), we associate a definable non-negative upper semi-continuous function f : K → R as follows. Definition 4.4. Associate with the given family {Sδ}δ>0 the family {S(cid:48) δ}δ>0 sat- isfying the conditions of Lemma 4.2. Define for each x ∈ K the value f (x) as max{δ x ∈ S(cid:48) δ, or as 0 otherwise (by Lemma 4.2, the function f is well-defined). Note that, by Convention 4.1, the function f is bounded. Convention 4.5. We identify any two non-negative functions f, g : K → R if they have the same level sets {x ∈ K f (x) = δ} = {x ∈ K g(x) = δ} for small δ > 0. Lemma 4.6. For a compact definable set K ⊂ Rn, there is a bijective corre- spondence between monotone definable families {Sδ}δ>0 of compact subsets of K and non-negative upper semi-continuous definable functions f : K → R, with Sδ = {x ∈ K f (x) ≥ δ}. Proof. The bijection is defined in Definition 4.4. The lemma follows from the fact that a function ψ : X → R on a topological space X is upper semi-continuous if and only if the set {x ∈ X ψ(x) < b} is open for every b ∈ R, and the identifications (cid:3) made in Conventions 4.3 and 4.5. S :=(cid:83) Remark 4.7. Observe that due to the correspondence in Lemma 4.6, the union δ Sδ coincides with {x ∈ K f (x) > 0}, the complement K \ S coincides with the 0-level set {x ∈ K f (x) = 0} of the function f . Lemma 4.8. For a compact definable set K ⊂ Rn, there is a bijective correspon- dence between arbitrary non-negative definable functions h : K → R and monotone definable families {Sδ}δ>0 of subsets of K, with Sδ = {x ∈ K h(x) ≥ δ}, satisfying the following property. There exists a cylindrical decomposition D of Rn, compatible with K, such that for small δ > 0 and every cylindrical cell X in D, the intersection Sδ ∩ X is closed in X. Proof. Let h : K → R be a non-negative definable function. Consider a cylindrical decomposition D(cid:48) of Rn+1 compatible with K and the graph of h in Rn+1. Note that D(cid:48) induces a cylindrical decomposition D of Rn compatible with K. Then the family {{x ∈ K h(x) ≥ δ}}δ>0 and the decomposition D satisfy the requirements, since by Definition 3.5 (of a cylindrical decomposition), for every cell X the restriction hX is a continuous function. Conversely, given a family {Sδ}δ>0 and a cylindrical decomposition D such that for small δ > 0 and every cylindrical cell X in D, the intersection Sδ ∩ X is closed in X, consider the family of compact sets {Sδ ∩ X}δ>0 in X (note that Sδ ∩ X∩X = Sδ∩X, since Sδ is closed in X). Applying Lemma 4.6 to {Sδ ∩ X}δ>0, we obtain an upper semi-continuous function hX : X → R such that Sδ ∩ X = {x ∈ X hX (x) ≥ δ}, and therefore, Sδ ∩ X = {x ∈ X hX (x) ≥ δ}. The function h : K → R is now defined by the restrictions of hX on all cylindrical cells X of D. (cid:3) Definition 4.9. Let K ⊂ Rn be a compact definable set, {Sδ}δ>0 a monotone definable family of compact subsets of K, and f the corresponding non-negative upper semi-continuous definable function. Let F be the graph of the function f . TRIANGULATIONS OF 2D FAMILIES 17 We say that a cylindrical cell decomposition C of Rn+1 is monotone with respect to the function f if C is monotone with respect to sets F and {xn+1 = 0}. Remark 4.10. By Theorem 3.20, for each K and {Sδ}δ>0 there exists a cylindrical cell decomposition of Rn+1 satisfying the frontier condition and monotone with respect to f . Let D be the cylindrical decomposition induced by C on Rn. Then (i) all cylindrical cells Z of C contained in the graph F and all cylindrical cells Y (ii) for every Y of D contained in K, the restriction fY is a monotone function on Y (see Definition 2.10), either positive or identically zero, and such that of D contained in K are monotone cells; Sδ ∩ Y = {x ∈ Y fY (x) ≥ δ} for small δ > 0. Indeed, each cylindrical cell Z of C is a monotone cell hence, by Proposition 2.8, each cylindrical cell Y in D is a monotone cell. Since the graph of the restriction fY is a cylindrical cell in C, it is a monotone cell. The function fY is either positive or identically zero due to Remark 4.7. 5. Separability and Basic Conditions Definition 5.1. By the standard m-simplex in Rm we mean the set ∆m := {(t1, . . . , tm) ∈ Rm >0 t1 + ··· + tm < 1}. We will assume that the vertices of ∆m are labeled by numbers 0, . . . , m so that the vertex at the origin has the number 0, while the vertex with xi = 1 has the number i. We will be dropping the upper index m for brevity, in cases when this does not lead to a confusion. Definition 5.2. An ordered m-simplex in Rn is an open simplex with some total order on its vertices. An ordered simplicial complex is a finite simplicial complex, such that all its simplices are ordered and the orders are compatible on the faces of simplices. Remark 5.3. For each ordered m-simplex Σ there is a canonical affine map from ∆m to a standard simplex Σ preserving the order of vertices. Definition 5.4. An ordered definable triangulation of a compact definable set K ⊂ Rn compatible with subsets A1, . . . , Ar ⊂ K is a definable homeomorphism Φ : C → K, where C is an ordered simplicial complex, such that each Ai is the union of images by Φ of some simplices in C. Proposition 5.5 ([6]). Let K be a compact definable subset in Rn and A1, . . . , Ar be definable subsets in K. Then there exists an ordered definable triangulation of K compatible with A1, . . . , Ar. Definition 5.6. For m ≤ n, by an ordered definable m-simplex we mean a pair (Λ, Φ) where Λ ⊂ Rn is a bounded definable set, and Φ : C → Λ is an ordered definable triangulation of its closure Λ, where C is the complex consisting of a standard m-simplex ∆ and all of its faces, such that Φ(∆) = Λ. The images by Φ of the faces of ∆ are called faces of Λ. Zero-dimensional faces are called vertices of Λ. If (Σ, Ψ) is another definable m-simplex in Rn, then a homeomorphism h : Λ → Σ is called face-preserving if Ψ−1 ◦ h ◦ Φ is a face-preserving homeomorphism of ∆ (i.e., sends each face to itself). 18 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Convention 5.7. In what follows, whenever it does not lead to a confusion, we will assume that for a given bounded definable set Λ the map Φ is fixed, and will refer to an ordered definable simplex by just Λ. Obviously, in the definable triangulation of any compact definable set the image by Φ of any simplex of the simplicial complex is a definable simplex. Let {Sδ}δ>0 be a monotone definable family of subsets of a compact definable set K ⊂ Rn, and f : K → R the corresponding upper semi-continuous function, so that Sδ = {x ∈ K f (x) ≥ δ}. Convention 5.8. In what follows, for a given definable simplex Λ we will write, slightly abusing the notation, Sδ instead of Sδ ∩ Λ, and will say that Sδ is a family in Λ. A family Sδ is proper if neither Sδ = Λ, nor Sδ = ∅. Notation 5.9. For a subset A ⊂ Λ of a definable simplex Λ we will use the following terms and notations. • The interior int(A) := {x ∈ A a neighbourhood of x in Λ lies in A}. • The boundary ∂A := A \ int(A). • The boundary in Λ, ∂ΛA := ∂A ∩ Λ. Definition 5.10. The set {f = δ} ∩ ∂ΛSδ is called the moving part of ∂ΛSδ, the set {f > δ} ∩ ∂ΛSδ is called the stationary part of ∂ΛSδ. Remark 5.11. Obviously the moving part and the stationary part form a partition of ∂ΛSδ. It is easy to see that the stationary part of ∂ΛSδ coincides with {x ∈ Λ lim inf y∈Λ, y→x f (y) ≤ δ < f (x)}. At each point of stationary part of the boundary, the function fΛ is discontinuous. In particular, if fΛ is continuous then the stationary part is empty. Definition 5.12 ([7], Definition 5.7). A family Sδ in a definable n-simplex Λ is called separable if for small δ > 0 and every face Λ(cid:48) of Λ, the inclusion Λ(cid:48) ⊂ Λ \ Sδ is equivalent to Sδ ∩ Λ(cid:48) = ∅. Remark 5.13. Observe that the implication (Sδ ∩ Λ(cid:48) = ∅) ⇒ (Λ(cid:48) ⊂ Λ \ Sδ) is true regardless of separability, since Sδ ∩ Λ(cid:48) = ∅ is equivalent to Λ(cid:48) ⊂ Λ \ Sδ, while Λ \ Sδ ⊂ Λ \ Sδ. Notation 5.14. It will be convenient to label the face of an ordered definable simplex Λ opposite to vertex j by Λj, and the face of Λ opposite to vertices j1, . . . , jk by Λj1,...,jk . Clearly, Λj1,...,jk does not depend on the order of j1, . . . , jk. More generally, for a subset A ⊂ Λ, let Aj := A∩ Λj and Aj1,...,jk = Aj1,...,jk−1 ∩ Λj1,...,jk . Note that the set Aj1,...,jk generally depends on the order of j1, . . . , jk. We call Aj1,...,jk a restriction of A to Λj1,...,jk . Definition 5.15. Given the ordered standard simplex ∆m, the canonical simplicial map (identification) between a face ∆m and the ordered standard simplex ∆m−k is the simplicial map preserving the order of vertex labels. j1,...,jk We still adhere to Convention 5.8, i.e., for a given (ordered) definable simplex Λ we write, slightly abusing the notation, Sδ instead of Sδ ∩ Λ, and say that Sδ is a TRIANGULATIONS OF 2D FAMILIES 19 family in Λ. Similarly, for an upper semi-continuous function f : K → R, defined on a compact set K ⊂ Rn, and a definable simplex Λ ⊂ K we write f instead of fΛ. Definition 5.16. Let f : Λ → R be an upper semi-continuous definable function on a definable ordered m-simplex Λ. Define fj : (Λ ∪ Λj) → R, where 0 ≤ j ≤ m, as the unique extension, by semicontinuity, of f to the facet Λj. Define the function fj1,...,jk : (Λ ∪ Λj1,...,jk ) → R, where (j1, . . . , jk) is a sequence of pair-wise distinct numbers in {0, . . . , m}, by induction on k, as the unique extension, by semicontinuity, of fj1,...,jk−1 to the face Λj1,...,jk . Consider a monotone family Sδ in a definable ordered n-simplex Λ. According to Lemma 4.6, there is a non-negative upper semi-continuous definable function f : Λ → R, with Sδ = {x ∈ Λ f (x) ≥ δ}. Similarly, for any pair-wise distinct numbers j1, . . . , jk in {0, . . . , n}, we have (Sδ)j1,...,jk = {x ∈ Λj1,...,jk fj1,...,jk (x) ≥ δ}. Convention 5.17. The vertices of a face Λj1,...,jk of Λ inherit the labels of vertices from Λ. When considering a family (Sδ)j1,...,jk in Λj1,...,jk we rename the vertices so that they have labels 0, 1, . . . , n − k but the order of labels is the same as it was in Λ. Definition 5.18. We say that a property of a family Sδ in Λ is hereditary if it holds true for any face of Λ and any restriction of Sδ to this face (assuming the Convention 5.17). Consider the following Basic Conditions (A) -- (D), satisfied for small δ > 0. (A) If Sδ (cid:54)= ∅ then (Sδ)j (cid:54)= ∅ for any j > 0. (B) If (Sδ)0 = Λ0 then Sδ = Λ. (C) For every pair (cid:96), m such that 0 ≤ (cid:96) < m ≤ n and ((cid:96), m) (cid:54)= (0, 1) we have (D) Either Sδ = ∅ or(cid:83) (Sδ)(cid:96),m = (Sδ)m,(cid:96). δ int(Sδ) = Λ. Lemma 5.19. Let a family Sδ in a definable n-dimensional simplex Λ be separable, satisfy the basic condition (D), and this condition is hereditary. Then for each n- dimensional simplex Σ of any triangulation of Λ (in particular, of a barycentric subdivision) the restriction Sδ ∩ Σ is separable in Σ. Proof. By Remark 5.13, it is sufficient to prove that for small δ > 0 and every face Σ(cid:48) of Σ, if Σ(cid:48) ⊂ Σ \ Sδ then Sδ ∩ Σ ∩ Σ(cid:48) = ∅. Take Σ(cid:48) such that Σ(cid:48) ⊂ Σ \ Sδ. By the basic condition (D), if Σ(cid:48) ⊂ Λ then Σ(cid:48) (cid:54)⊂ Σ \ Sδ for small δ > 0, hence Σ(cid:48) ⊂ Λ(cid:48) for some face Λ(cid:48) of Λ. Again by (D), Λ(cid:48) ⊂ Λ \ Sδ, and therefore Sδ ∩ Λ(cid:48) = ∅. It follows that the intersection Sδ ∩ Σ ∩ Σ(cid:48) = ∅ of smaller sets is also empty. (cid:3) 6. Standard and model families Convention 6.1. In this section we assume that all monotone definable families satisfy the basic conditions (A) -- (D), and these conditions are hereditary. Definition 6.2. Let Sδ be a monotone family in a definable ordered n-simplex Λ. We assign to Sδ a Boolean function ψ : {0, 1}n → {0, 1} using the following inductive rule. 20 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV • If n = 0 (hence Λ is the single vertex 0), then there are two possible types of Sδ. If Sδ = Λ, then ψ ≡ 1, otherwise Sδ = ∅ and ψ ≡ 0. • If n > 0, then ψxj =0 is assigned to (Sδ)j for every j (cid:54)= 0, and ψx1=1 is assigned to (Sδ)0 (here vertices of Λj are renamed i → i − 1 for all i > j, cf. Definition 5.15). Remark 6.3. (i) It is obvious that the function ψ is completely defined by its restrictions ψx1=1 and ψx1=0, hence by the restrictions (Sδ)0 and (Sδ)1. (ii) The basic condition (A) implies that if ψ(0, . . . , 0) = 0 then ψ ≡ 0 and Sδ = ∅. The condition (B) implies that if ψ(1, . . . , 1) = 1 then ψ ≡ 1 and Sδ = Λ. (iii) Because of the basic condition (C), for every pair 0 ≤ (cid:96) < m ≤ n such that (cid:96) < m and ((cid:96), m) (cid:54)= (0, 1) the restrictions (Sδ)(cid:96),m and (Sδ)m,(cid:96) have the same Boolean function assigned to them. It follows that under (C) Definition 6.2 is consistent. It is easy to give an example of a family Sδ where (C) is not satisfied and Definition 6.2 becomes contradictory. Definition 6.4. A Boolean function ψ is monotone (decreasing) if replacing 0 by 1 at any position of its argument (while keeping other positions unchanged) either preserves the value of ψ or changes it from 1 to 0. Lemma 6.5. The Boolean function ψ assigned to Sδ is monotone. Proof. If Sδ = Λ, then ψ ≡ 1, hence ψ is trivially monotone. Now assume that Sδ (cid:54)= Λ, and continue the proof by induction on n. The base of the induction, for n = 0, is obvious. Restriction of ψ to {xj = 0}, for any j, or to {x1 = 1}, is the Boolean function assigned to a facet of Λ, which is monotone by the inductive hypothesis. Hence, if Boolean values are assigned to n − 1 variables among x1, . . . , xn, except on x2 = . . . = xn = 1, the function ψ is monotone in the remaining variable. Suppose ψ is not monotone in x1 with x2 = ··· = xn = 1, i.e., ψ(1, 1, . . . , 1) = 1 and ψ(0, 1, . . . , 1) = 0. Then ψ is identically 1 on {x1 = 1}, i.e., (Sδ)0 = Λ0, while Sδ (cid:54)= Λ. This contradicts the basic condition (B). (cid:3) Definition 6.6. Two monotone families Sδ and Vδ, in definable simplices Σ and Λ respectively, are combinatorially equivalent if for every sequence (j1, . . . , jk), where the numbers j1, . . . , jk ∈ {0, . . . , n} are pair-wise distinct, the restrictions (Sδ)j1,...,jk ⊂ Σ and (Vδ)j1,...,jk ⊂ Λ are simultaneously either empty or non-empty. Remark 6.7. Obviously, if two families Sδ and Vδ are combinatorially equiva- lent, then for every sequence (j1, . . . , jk) of pair-wise distinct numbers the families (Sδ)j1,...,jk and (Vδ)j1,...,jk are combinatorially equivalent. Remark 6.8. Let Sδ and Vδ be two proper monotone families in a 2-simplex Λ, such that ∂ΛSδ and ∂ΛVδ are curve intervals. Then Sδ and Vδ are combinatorially equivalent if and only if the endpoints of ∂ΛSδ can be mapped onto endpoints of ∂ΛVδ so that the corresponding endpoints belong to the same faces of Λ for small δ > 0. Lemma 6.9. Two families Sδ and Vδ are assigned the same Boolean function if and only if these families are combinatorially equivalent. Proof. Suppose Sδ and Vδ are assigned the same Boolean function. According to Definition 6.2, for any (j1, . . . , jk) the restrictions (Sδ)j1,...,jk and (Vδ)j1,...,jk are also TRIANGULATIONS OF 2D FAMILIES 21 assigned the same Boolean function (after renaming the vertices whenever appro- priate). In particular, this function is identical 0 or not identical 0 simultaneously for both faces. We prove the converse statement by induction on n, the base case n = 0, being obvious. According to Remark 6.7, for every j the restrictions (Sδ)j and (Vδ)j are combinatorially equivalent. By the inductive hypothesis these restrictions are assigned the same (n − 1)-variate Boolean function. According to Definition 6.2, for every j, the restrictions to xj = 0 of Boolean functions ψ and ϕ, assigned to Sδ and Vδ respectively coincide, and also restrictions of ψ and ϕ to x1 = 1 coincide. (cid:3) Hence, ψ = ϕ. Observe that when n = 1 (respectively, n = 2) there are exactly three (respec- tively, six) distinct monotone Boolean functions. Therefore, Lemma 6.9 implies that in this case there are exactly three (respectively, six) distinct combinatorial equivalence classes of monotone families. Definition 6.10. A Boolean function ψ : {0, 1}n → {0, 1} is lex-monotone if it is monotone with respect to the lexicographic order of its arguments, assuming x1 < ··· < xn. Note that when n = 1 all monotone functions are lex-monotone, whereas for n = 2 all functions except one are lex-monotone. In general, for the number of all monotone Boolean functions (Dedekind number) no closed-form expression is known at the moment of writing. On the other hand, the number of lex-monotone functions is easy to obtain. (i) There are 2n + 1 lex-monotone functions ψ : {0, 1}n → {0, 1}. Lemma 6.11. (ii) If ψ is lex-monotone then its restriction ψxi=α, for any i = 1, . . . , n and any α ∈ {0, 1}, is a lex-monotone function in n − 1 variables. Proof. (i) Every monotone function ψ : A → {0, 1}, where A is a totally ordered finite set of cardinality k, can be represented as a k-sequence of the kind ψ = (1, . . . , 1, 0, . . . , 0). There are k + 1 such sequences. In our case k = 2n. (ii) Straightforward. (cid:3) Definition 6.12. A combinatorial equivalence class of a family Sδ is called combi- natorially standard, and the family itself is called combinatorially standard, if the corresponding Boolean function is lex-monotone. The following picture shows representatives of all combinatorially standard fam- ilies in cases n = 1 and n = 2. Lemma 6.13. If a family Sδ is combinatorially standard, then the family (Sδ)j1,...,jk for each (j1, . . . , jk) is combinatorially standard. Proof. Follows immediately from the definition of the combinatorial equivalence. (cid:3) Definition 6.14. Let ψ(x1, . . . , xn) be a lex-monotone Boolean function. A family Vδ in the ordered standard simplex ∆n is called the model family assigned to ψ, if it is constructed inductively as follows. (1) The non-proper family Vδ = ∅ is assigned to ψ ≡ 1 when δ > 1, and to ψ ≡ 0. The non-proper family Vδ = ∆n is assigned to ψ ≡ 1, when δ ≤ 1. In case n = 0, there are no other families. 22 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Figure 1. Standard families in dimensions one and two, with cor- responding lex-monotone functions. (2) If ψ (cid:54)≡ 0 while ψx1=1 ≡ 0, then let Vδ = ∆n ∩ {t1 + ··· + tn ≤ 1 − δ}. (3) If ψx1=0 = ψx1=1, then Vδ is the pre-image of the model family in ∆n−1 assigned to ψx1=0 under the projection map ∆n → ∆n−1 gluing together vertices 0 and 1 of ∆n. (4) Let ψx1=0 (cid:54)= ψx1=1 (cid:54)≡ 0. Define Uδ in ∆n as the family assigned as in (3), to the Boolean function ϕ(x1, . . . , xn) such that ϕx1=0 = ϕx1=1 = ψx1=0. ξ(x1, . . . , xn) such that ξx1=0 = ξx1=1 = ψx1=1. Let (cid:98)∆ denote the closure Define Wδ in ∆n as the family assigned as in (3), to the Boolean function of ∆n∩{2t1 + t2 ··· + tn < 1}. Define the family Vδ in ∆n as (Uδ ∩(cid:98)∆)∪ Wδ. Here are the lists of all model families in ordered ∆n in cases n = 1 and n = 2. Case n = 1. (0) ∅ (1) {t ≤ 1 − δ} ∩ ∆1 (2) (0, 1) = ∆1 Case n = 2. (0) ∅ (1) {t1 + t2 ≤ 1 − δ} ∩ ∆2 (2) {t2 ≤ 1 − δ} ∩ ∆2 (3) ({t2 ≤ 1 − δ} ∪ {2t1 + t2 ≤ 1}) ∩ ∆2 (4) ∆2 In cases n ≤ 2, the Figure 1 actually shows all model families. In the case n = 1, all monotone Boolean functions are lex-monotone, hence families (0) -- (2) represent all combinatorial classes of families satisfying the basic conditions, and these classes are combinatorially standard. In the case n = 2, there is only one monotone function which is not lex-monotone, ψx1=0 ≡ 1, ψx1=1 ≡ 0, and we can define the non-standard model family for ψ as 010011012012012012012e1e1e1e1e1e1e1e1e2e2e2e2e2 TRIANGULATIONS OF 2D FAMILIES 23 (5) ({t1 + t2 ≤ 1 − δ} ∪ {2t1 + t2 ≤ 1}) ∩ ∆2. Remark 6.15. The non-standard model family (5) can be obtained as a result of the procedure in item (4) of Definition 6.14. Also (5) is combinatorially equivalent to {t1/(1 − δ) + t2 ≤ 1} ∩ ∆2 which corresponds to the standard blow-up at the vertex labeled by 2 of ∆2. Note that for large n most of the monotone Boolean functions are not lex- monotone (Dedekind number grows superexponentially), hence most families are not standard. Lemma 6.16. There is a bijection between all combinatorially standard equivalence classes and all lex-monotone Boolean functions. Proof. According to Lemma 6.9, to any two combinatorially equivalent families the same Boolean function is assigned. It is straightforward to show that the model family, corresponding by Definition 6.14 to a given lex-monotone function, is assigned this same function by Definition 6.2. By Lemma 6.9, any two families (cid:3) having the same Boolean function belong to the same equivalence class. Definition 6.17. A lex-monotone Boolean function ψ(x1, . . . , xn) is called separa- ble if the set {(x1, . . . , xn) ψ(x1, . . . , xn) = 1} consists of either 0 or 2k points, for some k ∈ {0, . . . , n}. Remark 6.18. The number of separable functions in n variables is n + 2, since for each k ∈ {0, . . . , 2n} there is the unique lex-monotone function ψ with cardinality of {ψ = 1} equal to k (cf. the proof of Lemma 6.11, (i)). It follows that all lex- monotone functions are separable for n ≤ 1 and there is a single non-separable lex-monotone function for n = 2. Lemma 6.19. A lex-monotone Boolean function ψ(x1, . . . , xn), where n ≥ 0, is separable if and only if either ψx1=1 ≡ 0, or ψx1=0 and ψx1=1 are separable and equal. Proof. Let ψx1=1 ≡ 0, or ψx1=0 and ψx1=1 be separable and equal. By Re- mark 6.18, ψ is separable when n = 0. If ψx1=1 ≡ 0 then, by lex-monotonicity, {ψx1=0 = 1} consists of a single point. If ψx1=1 (cid:54)≡ 0 then the cardinality of {ψ = 1} is twice the cardinality of {ψx1=0 = 1}. The latter, by the assumption, is a power of 2. Conversely, there are exactly n + 2 such distinct functions ψ, while the number (cid:3) of different separable functions is also n + 2, by Remark 6.18. Lemma 6.20. Let Vδ = {f ≥ δ} ∩ ∆ be a model family, and ψ its corresponding lex-monotone Boolean function. The following properties are equivalent. (i) Vδ is separable; (ii) ψ is separable; (iii) f is continuous in ∆. Proof. We prove the lemma by induction on n, the basis for n = 0 being obvious. Let ψ be separable, and n > 0. By Lemma 6.19, either ψx1=1 ≡ 0, or ψx1=0 and ψx1=1 are separable and equal. In the first case, by Definition 6.14, (2), the family Vδ coincides with ∆n ∩ {t1 + ··· + tn ≤ 1 − δ}, and hence is separable. It is clear that the function f for this family is continuous. 24 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV 1 In the second case, by Definition 6.14, (3), Vδ is the pre-image of the separable (by the inductive hypothesis) family in ∆n−1, assigned to the separable function ψx1=0 and having a continuous defining function, under the projection map ∆n → ∆n−1 gluing together vertices 0 and 1 of ∆n, and hence is separable. It is clear that the the function f for Vδ is also continuous. Let ψ be not separable. Then, by Lemma 6.19, there are two possibilities. First, ψx1=1 (cid:54)≡ 0, and at least one of the restrictions ψx1=0 or ψx1=1 is not separable. Let, for definiteness, it be ψx1=0, and let ∆(cid:48) be a face of ∆n−1 such \ (Vδ)1. Note that ∆(cid:48) is also a face of ∆n, and ∆(cid:48) ⊂ ∆n−1 \ Vδ. that ∆(cid:48) ⊂ ∆n−1 By the inductive hypothesis, (V δ)1 ∩ ∆(cid:48) (cid:54)= ∅, hence V δ ∩ ∆(cid:48) (cid:54)= ∅, and we conclude that Vδ is not separable. Also by the inductive hypothesis, the defining function f1 of the family (Vδ)1 (see Definition 5.16) is not continuous, hence the function f is not continuous. The second possibility is that both restrictions, ψx1=0 and ψx1=1, are separable but different functions. Then the faces ∆(cid:48) and ∆(cid:48)(cid:48) of ∆n of the largest dimensions 1 \ (Vδ)1 are also different, one of them is a such that ∆(cid:48) ⊂ ∆n face of another (say, ∆(cid:48)(cid:48) ⊂ ∆(cid:48)), and both lie in ∆n \ Vδ. It follows that V δ ∩ ∆(cid:48) (cid:54)= ∅, hence Vδ is not separable. The values of the function f at points of ∆n 0 sufficiently close to ∆(cid:48) and sufficiently far from ∆(cid:48)(cid:48), are close to 0, while the values of f at points of ∆n 1 with the same property are separated from 0. It follows that f is not (cid:3) continuous. 0 \ (Vδ)0 and ∆(cid:48)(cid:48) ⊂ ∆n 1 Lemma 6.21. For each model family Vδ in the ordered standard simplex ∆n its interior int(Vδ) is a semi-monotone set. Proof. Consider cases (1) -- (4) of Definition 6.14. If Vδ is defined according to either (1) or (2), then int(Vδ) is a convex set, hence semi-monotone. For other cases we prove by induction on n, that the family int(Vδ) satisfies condition (ii) in Lemma 2.9, this implies its semi-monotonicity. The base for n = 0 is obvious. δ be the family in ∆n−1 assigned to ψx1=0 under the projection ∆n → ∆n−1. Take any x = (x(cid:48), x1) ∈ int(Vδ) and j ∈ {2, . . . , n}, then, by the inductive hypothesis, the interval Suppose Vδ is defined according to case (3), and let V (cid:48) x(cid:48),j := {(y2, . . . , yn) ∈ Rn−1 0 < yj < xj, yi = xi for i (cid:54)= j} I(cid:48) x(cid:48),j × span{x1}) ∩ ∆n lies lies in int(V (cid:48) δ ). Since ∆n is convex, it follows that (I(cid:48) in int(Vδ), therefore Ix,j ⊂ int(Vδ) and Ix,1 ⊂ int(Vδ), i.e., int(Vδ) satisfies the condition (ii) in Lemma 2.9. Now suppose that Vδ is defined according to case (4). By the same argument as was applied to Vδ in case (3), we show that families int(Uδ) and int(Wδ) satisfy the condition (ii) in Lemma 2.9. Simplex (cid:98)∆ satisfies this condition trivially. Then Lemma 2.9 implies that the set int(Vδ) = (int(Uδ) ∩ int((cid:98)∆)) ∪ int(Wδ) also satisfies the condition (ii) in Lemma 2.9. Using Lemma 2.9, we conclude that in all cases, int(Vδ) is semi-monotone. (cid:3) Remark 6.22. For a proper model family Vδ its boundary ∂∆nVδ in ∆n is not necessarily a monotone cell. However, ∂∆n Vδ is always a regular (n−1)-cell, because TRIANGULATIONS OF 2D FAMILIES 25 its complement in the whole boundary ∂Vδ is a union of monotone (hence, regular) (n − 1)-cells glued together with the same nerve as that of the complement of a vertex in the boundary of the simplex ∆n (considered as a simplicial complex of all of its faces). An exception is the family with ψx1=1 ≡ 0, for which the nerve is the same as for the complement of an (n − 1)-face instead of a vertex. Lemma 6.23. Every model family Vδ in the standard ordered simplex ∆n satisfies the following properties. (1) Vδ satisfies the basic conditions (A) -- (D), and these conditions are heredi- tary. (2) For any (j1, . . . , jk) the restriction (Vδ)j1,...,jk is a model family. (3) int(Vδ) is a regular n-cell while ∂∆Vδ is a regular (n − 1)-cell. Proof. Properties (1) and (2) follow immediately from Definition 6.14. The property (3) follows from Lemma 6.21, the fact that a semi-monotone set is (cid:3) a regular cell, and Remark 6.22. Lemma 6.24. Let Vδ be a model family in the standard ordered simplex ∆n. Then for each simplex Σ of the barycentric subdivision of ∆n the restriction Vδ ∩ Σ is a separable family. Proof. We first describe, by induction on n, a triangulation of ∆ (which is coarser than the barycentric subdivision) such that the restriction of Vδ to each its simplices is separable. If n = 0 then the family is already separable. If Vδ is defined according to either (1) or (2) of Definition 6.14, then it is already separable. δ be the family in ∆n−1 Suppose Vδ is defined according to case (3), and let V (cid:48) assigned to ψx1=0 under the projection ∆n → ∆n−1. By the inductive hypothesis, V (cid:48) δ can be partitioned into separable families. The pre-images of these families form a partition of Vδ into separable families. < := ∆n∩{2t1 + t2 ··· + tn < 1} and ∆n Now suppose that Vδ is defined according to case (4). Partition ∆n into two n- > := ∆n∩{2t1 + t2 ··· + tn > 1}, := ∆n ∩ {2t1 + t2 ··· + tn = 1}, where the vertex δ := Wδ ∩ ∆n simplices, ∆n and one (n − 1)-simplex ∆n−1 = (1/2, 0, . . . , 0) on the edge ∆n 2,...,n has label 1 in ∆n < and W (cid:48) δ are partitioned into separable families as in the case (3), while the (n−1)-dimensional model family (it corresponds to the Boolean function ψx1=0) is partitioned (U into separable families, according to the inductive hypothesis. Define U(cid:48) (cid:48) δ ∪ W δ := Uδ ∩ ∆n < and label 0 in ∆n >. <. Families U(cid:48) δ and W (cid:48) (cid:48) δ) ∩ ∆n−1 = There is a refinement of the described triangulation of ∆n which is the barycen- tric subdivision of ∆n. By Lemma 5.19, the restriction of Vδ to each simplex of this (cid:3) barycentric subdivision is separable. 7. Topological equivalence Definition 7.1. Consider two monotone families Sδ and Vδ in definable ordered m-simplices (Σ, Φ) and (Λ, Ψ) in Rn respectively, where Φ : ∆ → Σ, Ψ : ∆ → Λ are homeomorphisms, and ∆ is the closure of the standard ordered m-simplex in Rm (see Definition 5.6). 26 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Families Sδ and Vδ are topologically equivalent if there exist two face-preserving homeomorphisms h1 : Λ → Σ and h2 : Σ → Λ, not depending on δ, such that for small δ > 0 the inclusions Sδ ⊂ h1(Vδ) and Vδ ⊂ h2(Sδ) are satisfied. Families Sδ and Vδ are strongly topologically equivalent if they are topologically equivalent, and for small δ > 0 there is a face-preserving homeomorphism hδ : Σ → Λ such that hδ(Sδ) = Vδ. Remark 7.2. It is clear that if two families Sδ and Vδ are topologically equiva- lent with homeomorphisms h1, h2, then there exist two face-preserving homeomor- phisms, namely, h(cid:48) := Φ−1 ◦ h1 ◦ Ψ, h(cid:48)(cid:48) := Ψ−1 ◦ h2 ◦ Φ : ∆ → ∆, with the property that for small δ > 0 the inclusions Φ−1(Sδ) ⊂ h(cid:48)(Ψ−1(Vδ)) and Ψ−1(Vδ) ⊂ h(cid:48)(cid:48)(Φ−1(Sδ)) are satisfied. Conversely, given two homeomorphisms h(cid:48), h(cid:48)(cid:48) : ∆ → ∆, satisfying these inclusion properties, there are homeomorphisms h1 := Φ ◦ h(cid:48) ◦ Ψ−1 : Λ → Σ and h2 := Ψ ◦ h(cid:48)(cid:48) ◦ Φ−1 : Σ → Λ, realizing the topological equivalence of Sδ and Vδ. Introduce the following new basic condition on a monotone family Sδ in a defin- able ordered m-dimensional simplex Λ. (E) The interior int(Sδ) is either empty or a regular m-cell and the boundary ∂ΛSδ in Λ is either empty or a regular (m − 1)-cell. Convention 7.3. In this section we assume that all monotone definable families satisfy the basic conditions (A) -- (E), and these conditions are hereditary. Remark 7.4. If Sδ contains a facet Λj, then Sδ contains a neighborhood of Λj in Λ for all small positive δ. If, in addition, (Sδ)i contains a face Λi,j for every i (cid:54)= j, then Sδ contains the neighborhood in Λ of the closed facet Λj. Lemma 7.5. Given a proper family Sδ in a definable m-dimensional simplex Λ, the Hausdorff limit L of ∂ΛSδ, as δ (cid:38) 0, is the closure of a face Λ0,1,...,r of Λ for some r ∈ {0, 1, . . . , m − 1}. Proof. (1) We first prove that L ⊂ ∂Λ. Let, contrary to claim, L ∩ Λ (cid:54)= ∅. Then, by the basic condition (D), there exists x ∈ Λ such that x ∈ L ∩ int(Sδ) for all sufficiently small δ, which is a contradiction. (2) Next we prove that L ⊂ Λ0. Suppose that x ∈ L but x (cid:54)∈ Λ0. Since, by (1), L ⊂ ∂Λ, the point x lies in a face Λ(cid:48) of Λ such that Λ(cid:48) ∩ Λ0 = ∅. By the hereditary basic condition (D) applied to Λ(cid:48), we have that x ∈ Sδ ∩ Λ(cid:48) for small δ. Then in the neighbourhood of x in Λ there is a point y such that y ∈ Λ \ Sδ for any small δ. This contradicts to the basic condition (D). (3) We now prove that ∂Λ0 (Sδ)0 = ∂(∂ΛSδ) ∩ Λ0. By the basic condition (E), ∂(∂ΛSδ) is an (m − 2)-sphere in ∂Λ. By Newman's theorem (3.13 in [11]), it is the boundary of the closed (m − 1)-ball Sδ ∩ ∂Λ. Hence ∂(∂ΛSδ) ∩ Λ0 is the boundary in Λ0 of Sδ ∩ Λ0 = (Sδ)0 which we conventionally denote by ∂Λ0(Sδ)0. (4) Now we are able to finalise the proof the lemma by induction on m, the base being trivial. If L = Λ0 we are done. Otherwise, let L0 be the Hausdorff limit of ∂Λ0 (Sδ)0, as δ (cid:38) 0, and we prove that L = L0. TRIANGULATIONS OF 2D FAMILIES 27 By the inductive hypothesis, L0 is a closure of a face Λ0,1,...,r of Λ for some (cid:3) We have L0 ⊂ L, because, by (3), ∂Λ0 (Sδ)0 ⊂ ∂(∂ΛSδ) ⊂ ∂ΛSδ for small δ, and we can pass to Hausdorff limit in these inclusions. Conversely, there does not exist x ∈ L such that x (cid:54)∈ L0, because, by the hereditary basic condition (D), the latter implies that x ∈ (Sδ)0 for all small δ, hence x (cid:54)∈ L. r ∈ {1, . . . , m − 1}, hence the same is true for L. Lemma 7.6. Let Sδ be a separable family in the standard simplex ∆, and Vδ the combinatorially equivalent model family in ∆. Then for any δ > 0 there exists η > 0 such that Sδ ⊂ Vη and Vδ ⊂ Sη. Proof. For non-proper families the statement is trivial. Let Sδ be proper. Let ∆(cid:48) be the face of ∆ such that its closure is the Hausdorff limit of ∂∆Sδ as δ (cid:38) 0 (such ∆(cid:48) exists by Lemma 7.5). Then ∆(cid:48) is the unique face of ∆ of the maximal dimension such that ∆(cid:48) ⊂ ∆ \ Sδ. Let min(Sδ) = min x∈Sδ max(Sδ) = max x∈Sδ min(Vδ) = min x∈Sδ max(Vδ) = max x∈Sδ dist(x, ∆(cid:48)), dist(x, ∆(cid:48)), dist(x, ∆(cid:48)), dist(x, ∆(cid:48)). All functions max(·), min(·) are monotone decreasing with δ (cid:38) 0. The separability of both families implies that the minima are positive. For a given δ, choosing η so that max(Vη) < min(Sδ), we get Sδ ⊂ Vη, while choosing η so that max(Sη) < min(Vδ), we get Vδ ⊂ Sη. (cid:3) Definition 7.7. A combinatorial equivalence class of a family Sδ (satisfying the basic conditions (A) -- (E)) is called standard, and the family itself is called standard, if the corresponding Boolean function is lex-monotone. δ , . . . T r Remark 7.8. A combinatorially standard family, from Definition 6.12, does not need to satisfy the condition (E). Consider, for example, Sδ = Λ2 \ Bδ, where Bδ ⊂ Λ2 is homeomorphic to an open disk, with Bδ intersecting the boundary of Λ2 at a point x (cid:54)= Λ2 1,2, such that Bδ contracts to x as δ (cid:38) 0. The family Sδ satisfies (A) -- (D), is combinatorially standard (combinatorially equivalent to Vδ = Λ2) but not standard. Lemma 7.9. Let {T 1 δ } be a finite set of standard monotone families in the standard ordered 1-simplex ∆1. There is a face-preserving homeomorphism h1 : ∆1 → ∆1 not depending on δ, satisfying the following property. For every i = 1, . . . , r there exists one of the model families, W i δ , (among types (1), (2), (3)) of the same combinatorial type as T i δ , such that for small δ > 0 the inclusions δ ⊂ h1(W i T i Proof. Observe that for non-proper families any homeomorphism h1 is suitable. The basic conditions (A) -- (E) imply that every proper family T i δ coincides with (0, ui(δ)] for some functions ui(δ) : (0, 1) → (0, 1), such that limδ→0 ui(δ) = 1 (in particular, functions ui are monotone decreasing for small positive δ). In this case homeomorphism h1 can be defined by any monotone function h1(t) satisfying the conditions: δ ⊂ h1(T i δ ) and W i δ ) hold. 28 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV δ , . . . , T r and for small δ > 0. • h1(0) = 0, • h1(1) = 1, • h1(ui(δ)) > 1 − δ and h1(1 − δ) > ui(δ) for each i such that T i δ is proper, To achieve the last property, the graph of h1(0,1) should be situated above para- metric curves (ui(δ), 1 − δ), (1 − δ, ui(δ)) ⊂ (0, 1)2 for small δ > 0. (cid:3) Theorem 7.10. Let {T 1 δ } be finite sets of standard mono- tone families in the standard ordered 1-simplex ∆1 and 2-simplex ∆2, respectively. There exist face-preserving homeomorphisms h1 : ∆1 → ∆1 and h2 : ∆2 → ∆2 not depending on δ, satisfying the following properties. The homeomorphism h1 satis- fies Lemma 7.9. The restriction of h2 to closures of all facets of ∆2 coincide with h1, after canonical identification of each facet with the standard ordered 1-simplex (see Definition 5.15). For every i = 1, . . . , k there exists a model family, V i δ , such that for every small δ > 0 the inclusions Si δ)) hold. Proof. We may assume all families Si δ to be proper (of type (1), (2) or (3)). δ } and {S1 δ ⊂ h2(V i δ ⊂ h2(Si δ ) and V i We first define the homeomorphism h1. It will satisfy Lemma 7.9 for families δ and three other special families in ∆1. These special families are needed T 1 δ , . . . , T r for h1 to become the restriction to edges of ∆2 of the homeomorphism h2, required in the theorem. δ , . . . , Sk To construct the special families, we introduce four auxiliary functions s(δ), a(δ), b(δ), c(δ). Each of these functions is defined and continuous for small δ > 0, has values in (0, 1) and tends to 1 as δ → 0. In particular, each function is monotone decreasing for small positive δ. Let s(δ) := maxi{si(δ)}, where si(δ) is the boundary of the monotone family Si δ 0 (under the canonical affine map identifying δ with combinatorial types δ holds for all small δ > 0. Such δ contains the neighborhood Let a(δ) be a function such that for all families Si of the type (2) or (3), restricted to ∆2 this edge with ∆1). either (2) or (3), the inclusion {t2 ≤ a(δ)}∩ ∆2 ⊂ Si a function exists, since by Remark 7.4, the closure Si in ∆2 of the closed edge ∆2 2. ({t2 = b(δ)}∩ ∆2)\ Si its end (1 − b(δ), b(δ)) on the edge ∆2 0 of ∆2. Let b(δ) be a function, such that s(δ) < b(δ) < 1. Then for any Si δ the difference δ contains a neighborhood in the segment {t2 = b(δ)}∩ ∆2 of Let Si δ be of the type either (1) or (2), and V i δ the combinatorially equivalent model family. According to Lemma 7.6, for any δ > 0 there exists η(δ) > 0 such that Sδ ⊂ Vη(δ) and Vδ ⊂ Sη(δ). Let ci(δ) be a monotone continuous function defined for all small positive δ so that ci(δ) > 1 − η(δ). Define the function c(δ) as c(δ) := maxi{ci(δ)}. We construct the homeomorphism h1, as in Lemma 7.9, for T 1 δ and the following three families: T r+1 = (0, c(δ)], all defined on ∆1. It follows, in particular, that h1(1−δ) > b(δ), h1(a(δ)) > 1 − δ, and h1(1 − δ) > c(δ) for all small δ > 0. We will also assume, without loss of generality, that h1(t) > t for all t ∈ ∆1. = (0, 1 − a−1(1 − δ)], and T r+3 = (0, b(δ)], T r+2 δ , . . . , T r δ δ δ To describe the homeomorphism h2, we need two more auxiliary monotone de- creasing functions, p(t) and q(t). Let p(t) be a function on [0, 1] satisfying the following properties. (i) 0 < p(t)/(1 − t) < 1/2 for all t ∈ (0, 1), and p(0) = 1/2; TRIANGULATIONS OF 2D FAMILIES 29 (ii) p(t)/(1 − t) is decreasing as a function in t; (iii) there exists δ0 > 0 such that, for each Si holds. inclusion ({t1 = p(t2)} ∩ ∆2) ⊂ Si δ0 δ with combinatorial type (3), the δ is monotone, this implies the inclusion ({t1 = p(t2)} ∩ ∆2) ⊂ Si Such a function exists. The property (iii) can be satisfied because, by Remark 7.4, a family Si δ of the combinatorial type (3) contains a neighborhood of the open edge 1 in ∆2 for all small positive δ. By the curve selection lemma (Lemma 3.2 in ∆2 for all t2 ∈ (0, 1). [6]) there exists δ0 and a function p(t) such that (p(t2), t2) ∈ Si Since Si δ for all positive δ < δ0. Finally, there exists a function = q(t) on [0, 1] such that (i) 1 > q(t)/(1 − t) > 1/2 for all t ∈ (0, 1) and q(0) = 1/2; (ii) the point (t1, b(δ)) /∈ Si Since Si that (t1, t2) /∈ Si δ for all small positive δ and all t1 ∈ [q(b(δ)), 1 − b(δ)). δ is a monotone family and b(δ) is monotone decreasing for small δ, it follows δ for all small positive δ and all t2 ∈ [b(δ), 1), t1 ∈ [q(t2), 1 − t2). Now we describe the homeomorphism h2 : ∆2 → ∆2. We set h2 := ψ ◦ ϕ, where ψ and ϕ are homeomorphisms ∆2 → ∆2 defined as δ0 follows. Let ϕ(t1, t2) := ((1 − h1(t2))h1(t1/(1 − t2)), h1(t2)) when t2 < 1, and ϕ(t1, 1) = (0, 1). Then ϕ restricted to all edges of ∆2 equals to h1 (after identification of these edges with ∆1). Next, we define the homeomorphism ψ so that: (1) ψ preserves the segments {t2 = const} ∩ ∆2, (2) ψ(p(t2), t2) = ((1 − t2)/2, t2) for all t2 close to 1, (3) ψ((1 − t2)/2, t2) = (q(t2), t2) for all t2 close to 1. Observe that one can choose ψ to be piecewise linear on each segment {t2 = const}∩ ∆2. Then ψ acts as identity on the boundary of ∆2. We now prove that the homeomorphism h2 satisfies the property, required in the theorem, for the case of a family Sδ combinatorially equivalent to the model family Vδ = {t2 ≤ 1− δ}∪{2t1 + t2 ≤ 1} in the equivalence class (3). The proofs for other combinatorial types are similar (and simpler). Define Aδ := ({t2 ≤ a(δ)} ∪ {t1 ≤ p(t2)}) ∩ ∆2 Bδ := ({t2 ≤ b(δ)} ∪ {t1 ≤ q(t2)}) ∩ ∆2. and Then Aδ and Bδ have the same combinatorial type as Sδ and Vδ, and Aδ ⊂ Sδ ⊂ Bδ for all small δ > 0 (the inclusion Aδ ⊂ Sδ uses the fact that Sδ is a 2-ball, according to the basic condition (E)). We now prove that Vδ ⊂ h2(Aδ) and Bδ ⊂ h2(Vδ) for all small positive δ, which immediately implies the theorem. δ := ({t2 ≤ 1 − δ} ∪ {t1 ≤ p(t2)}) ∩ ∆2 A(cid:48) δ := ({t2 ≤ b(δ)} ∪ {2t1 + t2 ≤ 1}) ∩ ∆2. V (cid:48) and Let Then it is easy to check, using the properties of functions p(t2), q(t2), a(δ), b(δ), that, for all small δ > 0, the following inclusions hold. (1) A(cid:48) δ ⊂ ϕ(Aδ), 30 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV δ ⊂ ϕ(Vδ), (2) V (cid:48) (3) Vδ ⊂ ψ(A(cid:48) δ), (4) Bδ ⊂ ψ(V (cid:48) δ ). For example, let us prove (1). According to the construction of the homeomor- phism h1, we have h1(a(δ)) > 1 − δ, hence ({t2 ≤ 1 − δ} ∩ ∆2) ⊂ ϕ({t2 ≤ a(δ)} ∩ ∆2). Furthermore, ϕ maps each interval {t2 = const}∩∆2 parallel to itself along the rays through the vertex 2. Condition that p(t2)/(1 − t2) is decreasing as a function of t2, guarantees that the image of the curve (p(t2), t2) under ϕ is to the right (along the coordinate t1) of that curve, hence ({t1 ≤ p(t2)} ∩ ∆2) ⊂ ϕ({t1 ≤ p(t2)} ∩ ∆2), and (1) is proved. Proofs of the inclusions (2) -- (4) are analogous. Inclusions (1) and (3) imply Vδ ⊂ h2(Aδ), while (2) and (4) imply Bδ ⊂ h2(Vδ). (cid:3) Remark 7.11. (i) The statement analogous to Lemma 7.9 or Theorem 7.10 for families in ∆0 is trivial (the homeomorphism is the identity). (ii) The proof of Theorem 7.10 remains valid for the case when the families Si δ satisfying basic conditions, are not necessarily standard, but we do not need this fact in the sequel. For n = 2 there is only one non-standard model family, of type (5). The only addition one has to make to accommodate type (5), is in the description of the function p(t). More precisely, in item (iii) of the description of this function and in the proof of its existence, the phrase "combinatorial type (3)" should be replaced by "combinatorial type either (3) or (5)". Accordingly, Corollary 7.12 and Lemma 7.13 below remain true for not necessarily standard families satisfying basic conditions. (iii) Theorem 7.10 can be generalized to standard monotone families in the stan- dard ordered m-simplex ∆m for any m. The proof will appear elsewhere. Corollary 7.12. Let m ∈ {0, 1, 2}. Given a finite set of standard monotone fam- ilies {S1 δ } in definable m-simplices (Σi, Φi) with Φi : ∆ → Σi, there is a homeomorphism h : ∆ → ∆ such that the homeomorphisms δ , . . . , Sk hi,j : Φj ◦ h ◦ Φ−1 : Σi → Σj i and hj,i : Φi ◦ h ◦ Φ−1 : Σj → Σi j provide topological equivalence in pairs of combinatorially equivalent families in the set. In particular, if two families Si δ, Sj δ are combinatorially equivalent, then they are topologically equivalent. Proof. Apply Theorem 7.10 to the set of standard monotone families Φ−1 i = 1, . . . , k in ∆. δ), (cid:3) The following lemma shows, in particular, that for m ≤ 2, the property to be topologically equivalent becomes unnecessary in the definition of the strong topolog- ical equivalence (this is not true in the case m = 3). (Si i Lemma 7.13. For two standard monotone families Sδ and Vδ in definable or- dered simplices Σ and Λ respectively with dim Σ = dim Λ ≤ 2, the following three statements are equivalent. TRIANGULATIONS OF 2D FAMILIES 31 (1) The families are combinatorially equivalent. (2) The families are topologically equivalent. (3) The families are strongly topologically equivalent. (4) For small δ > 0 there is a face-preserving homeomorphism hδ : Σ → Λ mapping Sδ onto Vδ. Proof. The statement (1) implies (2) by Corollary 7.12. The statement (2) implies (1) since the homeomorphisms h, Φ and Ψ are face-preserving. Let v0, v1, v2 be the vertices of Σ having labels 0, 1, 2 respectively, let w0, w1, w2 be the vertices of Λ having labels 0, 1, 2 respectively. Due to the basic condition (E), the boundaries ∂ΣSδ of Sδ and ∂ΛVδ of Vδ in Σ and Λ respectively, are homeomorphic to intervals, hence (a very special case of the Schonflies theorem, Proposition 6.10 in [2]) the pairs (Σ, ∂ΣSδ) and (Λ, ∂ΛVδ) δ : Σ → Λ, mapping are homeomorphic, in particular there is a homeomorphism h(cid:48) δ : Λ → Λ, which Sδ onto Vδ. We prove that there is another homeomorphism, h(cid:48)(cid:48) preserves Vδ and maps the images h(cid:48) δ(v2) in Λ of vertices of Σ into vertices w1, w2, w3 respectively of Λ. δ(v0), h(cid:48) δ(v1), h(cid:48) δ(vi)) = wi for each h(cid:48) Indeed, consider the closed 2-ball Vδ, and the boundary ∂ΛVδ of Vδ in Λ. The set ∂ΛVδ lies in the boundary of Vδ which is a 1-sphere. It is easy to construct a homeomorphism g, of the boundary of Vδ onto itself such that the restriction g∂ΛVδ δ(vi) ∈ Vδ. The homeomorphism g is the identity, and g(h(cid:48) can be extended from the boundary to the homeomorphism g1 : Vδ → Vδ of the whole 2-ball ([11], Lemma 1.10). Repeating the same argument for the closed 2-ball Λ \ Vδ we obtain a homeomorphism g2 of this ball onto itself, such that g2∂ΛVδ is the identity and g2(h(cid:48) δ to coincide with g1 on Vδ and with g2 on Λ \ Vδ. δ ◦ h(cid:48) Taking hδ := h(cid:48)(cid:48) By Remark 6.8, the statement (4) implies (1). Since the statement (3), according to the definition, is the conjunction of (2) and (4), all the statements (1) -- (4) are (cid:3) pair-wise equivalent. δ, we conclude that the statement (1) implies (4). δ(vi)) = wi for each h(cid:48) δ(vi) ∈ Λ \ Vδ. Define h(cid:48)(cid:48) Corollary 7.14. A monotone family Sδ in a definable ordered simplex Σ with dim Σ ≤ 2, satisfying the basic conditions (A) -- (E) is standard if and only if it is topologically equivalent to a standard family. Proof. Follows directly from Definition 7.7 and Lemma 7.13. (cid:3) 8. Constructing families of monotone curves Let a1 < b1, . . . , an < bn, where n ≥ 2, be real numbers and consider an open box Let X ⊂ B be a monotone cell, with dim X = 2, such that points B := {(x1, . . . , xn) ∈ Rn ai < xi < bi}. a := (a1, . . . , an), b := (b1, . . . , bn) belong to X. Theorem 8.1. There is a definable family {γt 0 < t < 1} of disjoint open curve intervals ("curves") in X such that the following properties are satisfied. (i) The curve γt for each t is a monotone cell, that is, each coordinate function among x1, . . . , xn is monotone on γt. SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV 32 (iii) The union(cid:83) (ii) The endpoints of each γt are a and b. 0<t<1 γt coincides with X. Proof. Since at a the minima of all coordinate functions x1, . . . , xn on X are at- tained, on any germ of a smooth curve in X at a, all coordinate functions are monotone increasing from a. Choose generically such a germ α and a point c ∈ α. Consider all level curves through c of all coordinate functions (the same level curve can correspond to different functions). Observe that the cyclic order (of coordinate functions) in which the level curves pass through c is the same as the cyclic order in which level curves pass through any other point in X, because X is a monotone cell. The level curves through c divide X into open sectors, and there is a unique pair of neighbours in the ordered set of level curves such that α belongs to the intersection of two of sectors bounded by these curves. Let this pair of neighbours correspond to coordinate functions x(cid:96), xm (if a level curve corresponds to several coordinate functions, choose any corresponding function). Let X(cid:48) := ρspan{x(cid:96),xm}(X), and B(cid:48) := {(x(cid:96), xm) a(cid:96) < x(cid:96) < b(cid:96), am < xm < bm} = ρspan{x(cid:96),xm}(B). Note that X(cid:48) is a semi-monotone set and X(cid:48) ⊂ B(cid:48). It is clear, from the choice of coordinate functions x(cid:96) and xm, that if on a curve in X(cid:48) both x(cid:96) and xm are monotone, then all other coordinate functions are also t 0 < t < 1} of monotone. It follows that it is sufficient to construct a family {γ(cid:48) curves in X(cid:48) with properties analogous to (i) -- (iii), i.e., t is the graph of a monotone function in either x(cid:96) or xm; • each γ(cid:48) • the endpoints of each γ(cid:48) • the union(cid:83) t are (a(cid:96), am) and (b(cid:96), bm); 0<t<1 γ(cid:48) t coincides with X(cid:48). Then, for each t, define γt as (ρspan{x(cid:96),xm}X )−1(γ(cid:48) t). Note first that the top and the bottom of the semi-monotone set X(cid:48) are graphs of non-decreasing functions on the interval X(cid:48)(cid:48) ⊂ span{x(cid:96)}. It follows that for each c ∈ R such that a(cid:96)+am < c < b(cid:96)+bm, the line {x(cid:96)+xm = c} intersects X(cid:48) by an open interval (p(c) = (p(cid:96)(c), pm(c)), q(c) = (q(cid:96)(c), qm(c))) with p(cid:96)(c), pm(c), q(cid:96)(c), qm(c) being continuous non-decreasing functions of c. The limits of both p(c) and q(c) as c → a(cid:96) + am (respectively, as c → b(cid:96) + bm) are equal to (a(cid:96), am) (respectively, to (b(cid:96), bm)). For each t ∈ (0, 1) define the curve βt := {(1 − t)p(c) + tq(c) c ∈ (a(cid:96) + am, b(cid:96) + bm)} ⊂ X(cid:48). For each t the curve βt is continuous, and connects points a and b. As the parameter c varies from a(cid:96) + am to b(cid:96) + bm both coordinates of points on βt are non-decreasing as the functions of c, and strictly increasing everywhere except the parallelograms in X(cid:48) where either both p(cid:96)(c) and q(cid:96)(c) are constants or both pm(c) and qm(c) are constants. It remains to modify the family of curves inside each such parallelogram P . Suppose that pm(c) and qm(c) are constant in P , so the curves βt∩P are horizon- tal. The left and the right sides of P can be parameterized by t ∈ (0, 1). Consider a monotone increasing continuous function φ(t) : [0, 1] → [0, 1] such that φ(0) = 0, φ(1) = 1, and t < φ(t) for every t ∈ (0, 1) (e.g., strictly concave). For each t, connect a point corresponding to t on the left side of P with the point corresponding to φ(t) on the right side. This gives a family of monotone TRIANGULATIONS OF 2D FAMILIES 33 (with respect to c) curves αt,P inside P . The family γt can now be defined as βs outside all parallelograms and αs,P inside each parallelogram P , with the natural (cid:3) re-parameterizations. 9. Combinatorial equivalence to standard families Let K ⊂ Rn be a compact definable set with dim K ≤ 2, and {Sδ}δ>0 a monotone definable family of compact subsets of K. According to Lemma 4.6, there is a non- negative upper semi-continuous definable function f : K → R such that Sδ = {x ∈ K f (x) ≥ δ}. Moreover, by Remark 4.10, there is a cylindrical decomposition D(cid:48) of Rn+1, satisfying the frontier condition, and monotone with respect to the function f (Definition 4.9). Let D be the cylindrical cell decomposition induced by D(cid:48) on Rn. (see Definition 5.16) is monotone. Theorem 9.1. There is a definable ordered triangulation of K, which is a refine- ment of the decomposition D, with the following properties. (i) Every definable simplex Λ of the triangulation is a monotone cell. (ii) For every simplex Λ, the restriction fΛ is a monotone function. (iii) For every two-dimensional simplex Λ and every i = 0, 1, 2 the function (fΛ)i (iv) For every simplex Λ the family Sδ ∩ Λ satisfies the basic conditions (A) -- (E), (v) For every simplex Λ the family Sδ∩Λ is a standard family (see Definition 7.7). The theorem will follow from a series of lemmas. First we introduce some nec- essary notations. Let Z be a cylindrical two-dimensional section cell of D(cid:48), and Y corresponding two-dimensional cell of D, i.e., Z is the graph of fY . Let (cid:96) be minimal positive integer for which dim ρR(cid:96)Y = 1, and m, with (cid:96) < m ≤ n, be minimal for which dim ρRmY = 2. Then, by Lemma 3.3, and these conditions are hereditary. X := ρspan{x(cid:96),xm}(Y ) = ρspan{x(cid:96),xm}(Z) is a two-dimensional cylindrical cell, Y and Z are graphs of continuous definable maps on X. Note that X is a semi-monotone set, while Y and Z are graphs of monotone maps on X. Also observe that ρspan{x(cid:96),xm} maps Y and Z onto X, but Y and Z are not necessarily graphs of some maps over X. Let (a, b) ⊂ span {x(cid:96)} be the common image ρspan {x(cid:96)}(Z) = ρspan {x(cid:96)}(Y ) = ρspan {x(cid:96)}(X). intervals in Rn+1, and that Notice that the bottom U and the top V of Z (see Definition 3.1) are open curve ρspan {x(cid:96)}(Z) = ρspan {x(cid:96)}(Y ) = ρspan {x(cid:96)}(X) = = ρspan {x(cid:96)}(U ) = ρspan {x(cid:96)}(V ) = (a, b). The side wall W of Z is the pre-image (ρspan {x(cid:96)}Z)−1({a, b}), and by Lemma 3.4 has exactly two connected components each of which is either a single point or a closed curve interval. For a 1-dimensional cylindrical cell C of D, contained in the closure of Y , denote by fY,C the unique extension, by semicontinuity, of fY to C. Definition 9.2. Let ϕ : W → R be a continuous function on a definable set W ⊂ Rn, having the graph Φ ⊂ Rn+1. A point x ∈ W \ W is called a blow-up point of ϕ if ρ−1Rn (x) ∩ Φ contains an open interval. 34 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Infimum of a blow-up point x is inf{y {(x, y) ∈ ρ−1Rn (x) ∩ Φ}}. Lemma 9.3. (i) The set ∂Sδ ∩ Y , for small δ > 0 is either empty or a 1- dimensional regular cell, i.e., an open curve interval with distinct endpoints on ∂Y . (ii) The Hausdorff limit of a non-empty ∂Sδ ∩ Y is either a vertex of one of the 1-dimensional cylindrical cells on ∂Y or a curve interval in ∂Y with the endpoints at two distinct vertices. (iii) For small δ > 0, an endpoint of a non-empty ∂Sδ ∩ Y is a vertex of one of the 1-dimensional cylindrical cells on ∂Y if and only if this endpoint is a blow-up point of fY with infimum 0. Proof. Due to Proposition 2.7, the intersection {xn+1 = δ} ∩ Z, for each δ, is either empty or a monotone cell. By Proposition 2.8, the projection ρRn ({xn+1 = δ} ∩ Z) = {fY = δ} ∩ Y is also either empty or a monotone cell. According to Remark 5.11, {fY = δ} ∩ Y coincides with ∂Sδ ∩ Y . Hence (i) is proved. Let the Hausdorff limit of a non-empty ∂Sδ ∩ Y be either a single point x which is not a vertex of any 1-dimensional cylindrical cell on ∂Y and belongs to a 1- dimensional cell C on ∂Y , or an interval with at least one endpoint x belonging to a 1-dimensional cell C on ∂Y . Then fY,C(x) = 0 while at some other point in the neighbourhood of x in C the function fY,C is positive. This contradicts the supposition that the cylindrical decomposition D(cid:48) satisfies the frontier condition, and is monotone with respect to the function f (see Definition 4.9). This proves (ii). The item (iii) follows directly from item (ii) and Definition 9.2. (cid:3) Definition 9.4. Let Z be a two-dimensional cell of the decomposition D(cid:48), i ∈ {1, . . . , n + 1}, and c ∈ R. The intersection Z ∩ {xi = c} is called a separatrix in Z if • the sets {xi > c} ∩ Z and {xi < c} ∩ Z are not empty; • the set {xi = c} ∩ Z is not contained in {xi > c} ∩ Z ∩ {xi < c} ∩ Z. If Z ∩ {xi = c} is a separatrix in Z then its extension is Z ∩ {xi = c}. Observe that for each fixed i = 1, . . . n + 1 any two different separatrices are disjoint and their extensions are disjoint. Lemma 9.5. If a cylindrical cell Z has no separatrices, then in any cylindrical decomposition C of Rn+1 compatible with Z, each cylindrical cell contained in Z has no separatrices. Proof. Suppose, contrary to the claim of the lemma, that there is a cylindrical 2-cell C of C which contains a separatrix Z ∩ {xi = c}. By Definition 9.4, there is a point u = (u1, . . . , un+1) ∈ C with ui = c such that the closure of one of the sets, {xi > c} ∩ C or {xi < c} ∩ C, contains u, while the closure of the other one does not contain u. The point u can belong neither to the top V nor to the bottom U of C. Indeed, V is a monotone cell, hence if u ∈ V , then the function xi is constant on V . It follows that either xi is constant on whole C, or xi > c, or xi < c on C. Any of these alternatives contradicts to C ∩ {xi = c} being a separatrix. The same argument shows that u (cid:54)∈ U . Now let the point u belong to one of the two connected components W of the side wall of C. Observe that W = Z ∩ {xj = a} for some j (cid:54)= i and a ∈ R. The TRIANGULATIONS OF 2D FAMILIES 35 function xj is a monotone function on the monotone cell Z ∩ {xi = c}. On the other hand, it is constant on some interval in Z ∩{xi = c} and non-constant on the whole Z ∩ {xi = c}, which is a contradiction. (cid:3) Lemma 9.6. There is a refinement E(cid:48) of the cell decomposition D(cid:48) monotone with respect to f , and such that no 2-cell Z in E(cid:48) contains a separatrix. Proof. First notice that for given Z and i there is a finite number of separatrices. Indeed, the 1-dimensional set A of points u = (u1, . . . , un+1) ∈ ∂Z such that Z ∩ {xi = ui} is a separatrix, is definable, hence has a finite number of connected components. Since A and each separatrix extension are 1-dimensional and compact, each connected component of A is contained in one of these extensions. It follows that the number of separatrix extensions, and hence separatrices, does not exceed the number of connected components of A. Apply Corollary 3.21 to the set of all separatrices of all 2-cells of D(cid:48) (as the sets U1, . . . , Um in the corollary) and the decomposition D(cid:48) (as A). By the corol- lary, there is a refinement E(cid:48) of D(cid:48), monotone with respect to the function f and U1, . . . , Um, such that no 2-cell Z of E(cid:48) intersects with a separatrix of any 2-cell of D(cid:48). (cid:3) Lemma 9.7. Suppose that a 2-cell Z in a cylindrical decomposition E(cid:48) does not contain a separatrix. Then any point v = (v1, . . . , vn+1) ∈ Z can be connected to any point u = (u1, . . . , un+1) ∈ Z by a curve γ which is 1-dimensional monotone cell. Proof. The point v belongs to a certain sign condition set of the functions xi − ui for all i = 1, . . . , n + 1. Let, for definiteness, v ∈ {x1 > u1, . . . , xn+1 > un+1}. Let According to Lemma 9.5, no 2-cell in E(cid:48) contains a separatrix. B := {(x1, . . . , xn+1) ∈ Rn+1 ui < xi < vi}. Observe that Z ∩ B is a monotone cell. The closure Z ∩ B contains u, otherwise there would be a separatrix Z ∩ {xi = ui} for some i, which contradicts the main property of the cylindrical decomposition E(cid:48) (Lemma 9.6). Applying Theorem 8.1 to Z ∩ B as F, and (v, u) as (a, b), connect v to u by a curve γ, which is a (cid:3) monotone cell. Lemma 9.8. Let Z be a cylindrical cell of E(cid:48) having side wall W (cid:48), bottom U(cid:48), and top V (cid:48). Let P (cid:48) be a finite subset of W (cid:48), containing all vertices of Z, and v(cid:48) = (cid:96)} and b(cid:48) := V (cid:48)∩{x(cid:96) = v(cid:48) (cid:96)}, 1, . . . , v(cid:48) (v(cid:48) where (cid:96) is the minimal positive integer for which dim ρR(cid:96)Z = 1. There is a definable triangulation B(cid:48) of Z such that (i) the triangulation B(cid:48) is a cylindrical decomposition of Z, in particular each n+1) a point in Z. Introduce a(cid:48) := U(cid:48)∩{x(cid:96) = v(cid:48) simplex is a cylindrical cell; (ii) each simplex is a monotone cell; (iii) each 2-simplex does not contain a separatrix; (iv) the set of all vertices of the triangulation is P (cid:48) ∪ {v(cid:48), a(cid:48), b(cid:48)}; (v) the edges connecting v(cid:48) with a(cid:48) and b(cid:48) are contained in {x(cid:96) = v(cid:48) (cid:96)}. Proof. If u is any point in P (cid:48), then by Lemma 9.7, there exists a curve γ which is 1- dimensional monotone cell, connecting points v(cid:48) and u. If u be either a(cid:48) or b(cid:48), then (cid:96)} ∩ Z. u(cid:96) = v(cid:48) Thus we have constructed a triangulation of Z, denote it by B(cid:48). (cid:96) and we connect v(cid:48) to u by a monotone cell (interval) γ ⊂ {x(cid:96) = v(cid:48) 36 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Each Λ(cid:48) in B(cid:48) is a cylindrical cell. Indeed, since all 1-dimensional simplices are (cid:96)} ∩ Z, is monotone cells, each such simplex, except the ones contained in {x(cid:96) = v(cid:48) the graph of a monotone function on an interval in span{x(cid:96)}, hence is a cylindrical 1-cell. Each two-dimensional simplex has the top and the bottom among these graphs, moreover the corresponding two functions are defined on the same interval. Hence each two-dimensional simplex is a cylindrical two-cell. According to the Lemma 9.5, each two-dimensional simplex contains no separatrix. Now we show that each two-dimensional Λ(cid:48) in B(cid:48) is a monotone cell. Let Λ(cid:48) be (cid:96)}, bounded in the definable simplex contained in the monotone cell Z ∩ {x(cid:96) < v(cid:48) (cid:96)} and the curve γ, connecting v(cid:48) with a vertex of Z. Then, by Z by Z ∩ {x(cid:96) = v(cid:48) (cid:96)} into two monotone cells, one of Theorem 11 in [1], the curve γ divides Z ∩{x(cid:96) < v(cid:48) which is Λ(cid:48), and the other is the union of the remaining definable two-dimensional simplices and their boundaries in Z ∩ {x(cid:96) < v(cid:48) (cid:96)}. Applying inductively the same argument, we prove that each of these remaining simplices is a monotone cell, and (cid:96)} are also monotone cells. Hence each simplex Λ(cid:48) is a all simplices in Z ∩ {x(cid:96) > v(cid:48) (cid:3) monotone cell. Corollary 9.9. Let Y be a cylindrical cell of E having side wall W , bottom U , and top V . Let P be a finite subset of W , containing all vertices of Y , and v = (v1, . . . , vn) a point in Y . Let Q ⊂ P be the subset of all points w ∈ P at which the function fY has a blow-up, and for each w ∈ Q fix one of the limit values α of fY at w. Introduce a := U ∩{x(cid:96) = v(cid:96)} and b := V ∩{x(cid:96) = v(cid:96)}, where (cid:96) is the minimal positive integer for which dim ρR(cid:96)Y = 1. There is a definable triangulation B of Y such that (i) the triangulation B is a cylindrical decomposition of Y , in particular each simplex is a cylindrical cell; (ii) each simplex is a monotone cell; (iii) each 2-simplex does not contain a separatrix; (iv) the set of all vertices of the triangulation is P ∪ {v, a, b}; (v) the edges connecting v with a and b are contained in {x(cid:96) = v(cid:96)}; (vi) for every simplex Λ, the restriction fΛ is a monotone function; (vii) for each w ∈ Q, the limit of fY at w along the edge connecting v and w equals α. Proof. Let Z be a cylindrical cell of E(cid:48) such that ρRn (Z) = Y , i.e., Z is the graph of fY . Since Q is the set of all blow-up points of fY , for each point u ∈ P \ Q there is a unique pre-image ρ−1(u) in the side wall of Z. Applying Lemma 9.8 to Z and the union of the set (ρRnZ)−1(P \ w) ∪ {(w, α) w ∈ Q} and the set of all vertices of Z as P (cid:48), and the point (ρRnZ)−1(v) as v(cid:48), we obtain a triangulation B(cid:48) of Z. In particular, all 1-dimensional simplices of B(cid:48) are monotone cells. Projections by ρRn of these 1-dimensional simplices are monotone cells, connecting v with the points in P , a and b. The properties (i) -- (v) of the triangulation B can be proved exactly as the anal- ogous properties of the triangulation B(cid:48) in Lemma 9.8. For each simplex Λ in B, the function fΛ is monotone since its graph is a simplex Λ(cid:48) of the triangulation B(cid:48), and hence is a monotone cell. Finally, the property (vii) of B is valid by the choice of edges connecting v and points w ∈ Q. (cid:3) TRIANGULATIONS OF 2D FAMILIES 37 The triangulation constructed in Corollary 9.9 is not ordered. To label the vertices of simplices so that conditions (iv) and (v) of Theorem 9.1 are satisfied we will need to perform a further refinement of the triangulation B, as follows. Let Y be a cylindrical cell of the cell decomposition E. For the side wall W of Y define the finite set P ⊂ W as the set of all vertices of cylindrical cells of E, contained in W (including vertices of Y ). Apply Corollary 9.9 to Y , with this P , and arbitrary v. If there is a vertex w of Y where fY has a blow-up point with infimum 0, then choose the value α = 0. For all other points w ∈ Q choose α arbitrarily. According to the corollary, there is a definable triangulation B(Y ) of Y satisfying properties (i) -- (vii). Let Λ be a simplex of the triangulation B(Y ). By Corollary 9.9, (i), the simplex Λ is a cylindrical cell. One of the connected components of its side wall is a curve interval, while another is a single point. Apply Corollary 9.9 to Λ (as Y ), with the set P consisting of all three vertices of Λ and the point in the middle of the 1-dimensional component of its side wall. If any of the vertices of Λ is a blow-up point w of fΛ with infimum 0, then choose a value α > 0. The resulting definable triangulation B(Λ) is then a barycentric subdivision of Λ. Label the vertices of B(Λ) as follows: the center of the subdivision is assigned 0, the vertices of Λ are assigned 2, while the remaining three vertices are assigned 1. Observe that any simplex Σ of the triangulation B(Λ) may have at most one vertex at which fY has a blow-up, and this vertex is Σ0,1 (having the label 2). In this picture the triangulation B(Λ) is shown for just one simplex Λ, the only simplex of the triangulation B(Y ) to which the restriction of the family Sδ is not separable. This construction is illustrated on Figire 2. Figure 2. Iterated stellar subdivision of Y . According to Convention 5.8, for a simplex Σ of B(Λ) we will write Sδ meaning the monotone family {Sδ ∩ Σ}. Lemma 9.10. For all cylindrical cells Y in E, for all simplices Λ in B(Y ), for all 1-dimensional simplices Σ in B(Λ), and for all i = 0, 1, 2 the restriction (Sδ)i has one of the three combinatorial types of 1-dimensional model families. Proof. Since the cylindrical decomposition D(cid:48) is monotone with respect to the func- tion f (see Definition 4.9) the vertex Σ1,2 (labelled by 0) belongs to Sδ ∩ Y for small δ > 0 whenever Sδ ∩ Y (cid:54)= ∅. It follows that the restrictions (Sδ)1 and (Sδ)2 have the ab0111222a)b)c)wwv 38 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV combinatorial type either (1) or (2) of 1-dimensional model family. If in the edge Σ0 the vertex Σ0,2 (labelled by 1) lies in Y , then (Sδ)0 has the type either (1) or (2) by the same argument. If, on the other hand, in Σ0 the vertex Σ0,2 lies in ∂Y , then it, and the whole edge Σ0 lies in a 1-dimensional cell C of the decomposition E, and vertices of C are labelled by 2, in particular the vertex Σ0,2 lies in C. By Lemma 9.3, (iii), C is compatible with the Hausdorff limit of ∂ΣSδ, hence (Sδ)0 may have any of the three combinatorial types of 1-dimensional model family. (cid:3) Lemma 9.11. For all cylindrical cells Y in E, for all simplices Λ in B(Y ), and for all simplices Σ in B(Λ), the family Sδ satisfies the basic conditions (A) -- (E), and these conditions are hereditary. Proof. Recall that by Definition 7.7, the standard family satisfies the basic condi- tions (A) -- (E). Since the cylindrical decomposition D(cid:48) is monotone with respect to the function f (see Definition 4.9) the vertex Σ1,2 (labelled by 0) belongs to Sδ ∩ Y for small δ > 0 whenever Sδ ∩ Y (cid:54)= ∅. It follows that the basic condition (A) is satisfied. Since, by the construction, Σ may have at most one vertex at which fY has a blow-up, and this vertex is Σ0,1 (having the label 2), the vertex Σ0,2 (having the label 1) is not a blow-up vertex. Thus, it cannot happen that one of the two sets (Sδ)0,2, (Sδ)2,0 is empty while another is not. It follows that the basic condition (C) is satisfied. Because of the basic condition (C), the only possibility for the basic condition (B) to fail for a simplex Σ would be to have (Sδ)0 = Σ0 while (Sδ)1,0 = ∅ for small δ > 0. In this case the vertex Σ0,1 of Σ, labeled by 2, is a blow-up point of fΣ. By the construction, the function fΣ1(x) → α > 0 as x → Σ0,1. Since fY is monotone, hence continuous, we have fΣ,Σ1 = fΣ1 . It follows that (Sδ)1,0 (cid:54)= ∅, for small δ > 0, which is a contradiction. The basic condition (D) holds true since the cylindrical decomposition D(cid:48) is monotone with respect to the function f . Let Σ(cid:48) be the graph of the function fΣ. Due to Proposition 2.7, for every Σ, the intersections Σ(cid:48) ∩ {xn+1 = c}, Σ(cid:48) ∩ {xn+1 ≶ c} for each c ∈ R are either empty or monotone cells. By Proposition 2.8, the projections of these sets to Rn, in particular the δ-level sets ∂ΣSδ of the function fΣ, are also either empty or monotone cells. Hence, the basic condition (E) is satisfied. All basic conditions are hereditary for the restrictions of Sδ to edges of Σ by (cid:3) Lemma 9.10. Lemma 9.12. For all cylindrical cells Y in E, for all simplices Λ in B(Y ), and for all simplices Σ in B(Λ), the family Sδ is standard. Proof. By Lemma 6.5, since the basic conditions (A) -- (E) are satisfied for the family Sδ for every Σ, it belongs to either one of the standard combinatorial types, or to the combinatorial type (5) of the non-standard model family ({t1 + t2 ≤ 1 − δ} ∪ {2t1 + t2 ≤ 1}) ∩ ∆2. However the latter alternative is not possible. Suppose this is the case. Then the vertex Σ0,1 in Σ (labeled by 2), is a blow-up point w of fY . Since Σ is an element of a barycentric subdivision of some Λ, it is one simplex of the two, in the subdivision, having the vertex w. TRIANGULATIONS OF 2D FAMILIES 39 First let Σ be the simplex whose edge Σ0 lies in the edge of Λ connecting the internal point of Y with w. Then the vertex Σ0,2 (labeled by 1) of Σ belongs to Sδ for small δ > 0, which contradicts to the family Sδ being of the type (5). Now let Σ be the other simplex in the barycentric subdivision of Λ with the vertex w. Recall that, by the construction, fY (x) → 0 as x → w along the edge γ of Λ connecting the internal point of Y with w. Under the supposition that Sδ is of the type (5), at each point x ∈ γ in the neighbourhood of w we have fY (x) > δ (cid:3) for small δ > 0 which is a contradiction. Proof of Theorem 9.1. The theorem follows immediately from Lemma 9.6, Corol- (cid:3) lary 9.9, Lemmas 9.11 and 9.12. Corollary 7.12 and Theorem 9.1 immediately imply the following theorem. Theorem 9.13. When dim K ≤ 2 there exists a definable triangulation of K such that for each simplex Λ and small δ > 0, the intersection Sδ ∩ Λ is topologically equivalent to one of the model families Vδ in the standard simplex ∆. 10. Triangulations with separable families j := Φ(∆j). In ∆1 consider definable families T i Consider the triangulation of K from Theorem 9.13, i.e., the definable home- C → K, where C is the finite ordered simplicial complex. Let omorphism Φ : Σ1, . . . , Σr be all 1-simplices in C and ∆1, . . . , ∆k be all 2-simplices in C. Then there are affine face-preserving homeomorphisms Ψi : ∆1 → Σi and Φi : ∆2 → ∆i, where ∆1 and ∆2 are a standard ordered 1-simplex and 2-simplex respectively. Let (Φ−1(Sδ ∩ i := Φ(Σi), Λ2 Λ1 i δ := Φ−1 Λ1 i )), and in ∆2 consider definable families Si i )). For each i = 1, . . . , r, in ∆1 consider the model family W i δ combinatorially equivalent to the family T i δ combinato- rially equivalent to the family Sj Lemma 10.1. For all small positive δ, the sets Φ−1(Sδ) and Vδ are homotopy equivalent. Proof. Without loss of generality, assume that Sδ (and hence, Φ−1(Sδ) and Vδ) is connected. Consider the covering of the set Φ−1(Sδ) by all sets Aδ in δ , and for each j = 1, . . . k in ∆2 consider the model family V j δ ). δ . Let Vδ :=(cid:83) δ ) ∪(cid:83) (Φ−1(Sδ ∩ Λ2 δ := Ψ−1 j Φj(V j i Ψi(W i i {Ψi(T i δ ), Φi(Sj δ ) i = 1, . . . , r, j = 1, . . . , k}. and let NΦ−1(Sδ) be the nerve of this covering. Observe that every finite non- empty intersection Ai1 is either a single point or an interval in one of 1-simplices Σi, hence contractible. By the Nerve Theorem ([4], Theorem 6), Φ−1(Sδ) is homotopy equivalent to the geometric realization of NΦ−1(Sδ). δ ∩ ··· ∩ Ait δ Analogously, the covering of the set Vδ by all sets Bδ in {Ψi(W i δ ), Φi(V j δ ) i = 1, . . . , r, j = 1, . . . , k} has the nerve NVδ whose geometric realization is homotopy equivalent to Vδ. Since each model family W i δ , and each model δ , the nerves NΦ−1(Sδ) and NVδ are δ is combinatorially equivalent to Sj family V j isomorphic, hence their geometric realizations are homotopy equivalent. It follows (cid:3) that Φ−1(Sδ) and Vδ are homotopy equivalent. δ is combinatorially equivalent to T i 40 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Theorem 10.2. There exists a definable monotone family Rδ in K and a defin- able ordered triangulation of K such that Rδ is homotopy equivalent to Sδ for all small positive δ, and for each ordered simplex Λ and each small enough δ > 0, the intersection Rδ ∩ Λ is topologically equivalent to one of the separable families in the standard simplex. Proof. By Lemma 10.1, for all small positive δ, the sets Φ−1(Sδ) and Vδ are homo- topy equivalent. According to Lemma 6.24, the restrictions of W i δ to each simplex of the barycentric subdivision of ∆1, and of V j δ to each simplex of the barycentric subdivision of ∆2 are separable model families. As Rδ take Φ(Vδ) and as trian- C(cid:48) → K, where C(cid:48) is the barycentric subdivision of the gulation of K take Φ : (cid:3) simplicial complex C. Let, as before, S :=(cid:83) 11. Approximation by compact families δ>0 Sδ ⊂ K ⊂ Rn. In [7] the following construction was For each δ > 0, let {Sδ,ε}ε>0 be a definable family of compact subsets of K such introduced. (2) Sδ =(cid:84) that the following conditions hold: (1) for all ε, ε(cid:48) ∈ (0, 1), if ε(cid:48) > ε, then Sδ,ε ⊂ Sδ,ε(cid:48); (3) for all δ(cid:48) > 0 sufficiently smaller than δ, and all ε(cid:48) > 0, there exists an open ε>0 Sδ,ε; in K set U ⊂ K such that Sδ ⊂ U ⊂ Sδ(cid:48),ε(cid:48). For a sequence ε0 (cid:28) δ0 (cid:28) ε1 (cid:28) δ1 (cid:28) ··· (cid:28) εm (cid:28) δm, introduce the compact set Tm(Sδ,ε) := Sδ0,ε0 ∪ ··· ∪ Sδm,εm. Here m ≥ 0, and (cid:28) stands for "sufficiently smaller than" (for the precise meaning of (cid:28) see Definition 1.7 in [7]). Let Hi(X) be the singular homology group of a topological space X with coef- ficients in some fixed Abelian group. Without loss of generality, assume that S is connected in order to make the homotopy groups πk(S) and πk(Tm(Sδ,ε)) indepen- dent of a base point. Proposition 11.1 ([7], Theorem 1.10). (i) For every 1 ≤ k ≤ m, there are epi- morphisms ψk : πk(Tm(Sδ,ε)) → πk(S), ϕk : Hk(Tm(Sδ,ε)) → Hk(S). (ii) If there is a triangulation of S such that the restriction of Sδ to each simplex is separable, then ψk and ϕk are isomorphisms for all k ≤ m − 1. Herewith if m ≥ dim S, then Tm(Sδ,ε) is homotopy equivalent to S. k ≤ m − 1 even without the separability condition. It was conjectured in [7], Remark 1.11, that ψk and ϕk are isomorphisms for all We now show that this conjecture is true in case when dim K ≤ 2. Let a family {Sδ,ε} have the corresponding monotone family {Sδ}. Consider the triangulation from Theorem 9.1, corresponding to {Sδ}. Construct new families {Vδ} and {Vδ,ε} in K as follows. We start with Vδ. For each definable simplex Λ of the triangulation Φ, if the restriction Λ ∩ Sδ is separable, let Λ ∩ Vδ coincide with Λ ∩ Sδ. If Λ ∩ Sδ is not separable, consider two cases. TRIANGULATIONS OF 2D FAMILIES 41 In the first case Λ0 ⊂ Sδ,ε. Then, for Λ ∩ Vδ, replace the restriction Λ ∩ Sδ by the (separable) family of combinatorial type of the model family (4) (i.e., by the full simplex Λ). In the second case Λ0 (cid:54)⊂ Sδ,ε. Let Φ : ∆ → Λ, where ∆ is the standard 2-simplex, be a homeomorphism preserving the order of vertices. Let Wδ be the monotone family in ∆ having the combinatorial type of the model family (2), such that ∂∆Wδ is the straight line interval parallel to ∆2 with the endpoint Φ−1(∂Λ0(Λ ∩ Sδ ∩ Λ0)) (hence, the restrictions to ∆0 of Φ−1(Λ∩ Sδ) and Wδ coincide). For Λ∩ Vδ, replace the restriction Λ ∩ Sδ by a (separable) family Φ(Wδ). Now we construct the family Vδ,ε. Observe that the set Φ−1(Λ ∩ Sδ,ε ∩ Λ0) consists of two semi-open intervals (∆0,2, a] and [b, ∆0,1) in ∆0 (in particular, the open endpoint of the first interval is the vertex 1 of ∆, while the open endpoint of the second interval is the vertex 2). Let α be the intersection of ∆ with the straight line passing through a parallel to ∆2, and β be the intersection of ∆ with the straight line passing through b parallel to ∆1. Let Λ1, . . . , Λr be all two-dimensional simplices of the triangulation such that Λi ∩ Sδ is non-separable and Λi ∩ Vδ (cid:54)= Λi for each i = 1, . . . , r. For each δ let For each i = 1, . . . , r let Λi ∩ Vδ,ε be the union of two sectors, closed in Λi. One sector lies between Λ2 and the curve Φ(α). Another sector lies between Λ1 and the curve Φ(β). i Λi coincide with Sδ,ε \(cid:83) Vδ,ε \(cid:83) i Λi. ginning of this section. Remark 11.2. (i) The family Vδ,ε satisfies the conditions (1), (2), (3) at the be- (ii) The family Λ∩Vδ is separable for each definable simplex Λ of the triangulation. δ>0(Λ∩ Vδ) coincide, being δ>0 Vδ = S, since the unions(cid:83) δ>0(Λ∩ Sδ) and(cid:83) (iii) (cid:83) simultaneously either empty or equal to Λ for each Λ. Lemma 11.3. For m ≥ 1 and the families Sδ,ε, Vδ,ε, we have Tm(Sδ,ε) = Tm(Vδ,ε). Proof. By the definition, for all δ and ε, the sets Sδ,ε and Vδ,ε coincide everywhere outside the union of all simplices Λ such that Λ ∩ Sδ is non-separable. Then so do also the sets Tm(Sδ,ε) and Tm(Vδ,ε). On the other hand, for each such Λ, we have Λ ∩ T1(Sδ,ε) = Λ ∩ T1(Vδ,ε) = Λ, and therefore Λ ∩ Tm(Sδ,ε) = Λ ∩ Tm(Vδ,ε) = Λ. (cid:3) Hence, Tm(Sδ,ε) = Tm(Vδ,ε). Theorem 11.4. If dim K ≤ 2 then for every 1 ≤ k ≤ m, there are epimorphisms ψk : πk(Tm(Sδ,ε)) → πk(S), ϕk : Hk(Tm(Sδ,ε)) → Hk(S) which are isomorphisms for all k ≤ m − 1. Tm(Sδ,ε) is homotopy equivalent to S. Proof. Construct the family {Vδ,ε} for {Sδ,ε} as described above. Since(cid:83) δ>0 Vδ = S, and each family Λ ∩ Vδ is separable (see Remark 11.2, (ii)), there exist, by Proposition 11.1, (ii), epimorphisms In particular, if m ≥ dim S, then ψk : πk(Tm(Vδ,ε)) → πk(S), ϕk : Hk(Tm(Vδ,ε)) → Hk(S) which are isomorphisms for all k ≤ m − 1. By Lemma 11.3, we have Tm(Sδ,ε) = (cid:3) Tm(Vδ,ε), which implies the theorem. 42 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV References [1] S. Basu, A. Gabrielov, and N. Vorobjov. Monotone functions and maps. Rev. R. Acad. Cienc. Exactas F´ıs. Nat. Ser. A Math. RACSAM, 107(1):5 -- 33, 2013. 2, 5, 6, 7, 10, 11, 12, 13, 14, 36 [2] S. Basu, A. Gabrielov, and N. Vorobjov. Semi-monotone sets. J. Eur. Math. Soc. (JEMS), 15(2):635 -- 657, 2013. 2, 5, 6, 11, 31 [3] L. Birbrair. Lipschitz geometry of curves and surfaces definable in O-minimal structures. Illinois J. Math., 52(4):1325 -- 1353, 2008. 3 [4] A. Bjorner. Nerves, fibers and homotopy groups. J. Combin. Theory Ser. A, 102(1):88 -- 93, 2003. 39 [5] J. Bochnak, M. Coste, and M.-F. Roy. G´eom´etrie alg´ebrique r´eelle (Second edition in english: Real Algebraic Geometry), volume 12 (36) of Ergebnisse der Mathematik und ihrer Grenzge- biete [Results in Mathematics and Related Areas ]. Springer-Verlag, Berlin, 1987 (1998). 2 [6] M. Coste. An introduction to o-minimal geometry. Istituti Editoriali e Poligrafici Internazion- ali, Pisa, 2000. Dip. Mat. Univ. Pisa, Dottorato di Ricerca in Matematica. 1, 2, 3, 7, 15, 17, 29 [7] A. Gabrielov and N. Vorobjov. Approximation of definable sets by compact families, and upper bounds on homotopy and homology. J. Lond. Math. Soc. (2), 80(1):35 -- 54, 2009. 1, 2, 3, 4, 18, 40 [8] H. Hironaka. Introduction to real-analytic sets and real-analytic maps. Istituto Matematico "L. Tonelli" dell'Universit`a di Pisa, Pisa, 1973. Quaderni dei Gruppi di Ricerca Matematica del Consiglio Nazionale delle Ricerche. 3 [9] D. Lazard. CAD and topology of semi-algebraic sets. Math. Comput. Sci., 4(1):93 -- 112, 2010. 2, 11, 14 [10] J.-M. Lion and J.-P. Rolin. Th´eor`eme de pr´eparation pour les fonctions logarithmico- exponentielles. Ann. Inst. Fourier (Grenoble), 47(3):859 -- 884, 1997. 4 [11] C. P. Rourke and B. J. Sanderson. Introduction to piecewise-linear topology. Springer Study Edition. Springer-Verlag, Berlin, 1982. Reprint. 26, 31 [12] L. van den Dries. Tame topology and o-minimal structures, volume 248 of London Mathe- matical Society Lecture Note Series. Cambridge University Press, Cambridge, 1998. 1, 2, 3, 7, 9, 11 [13] L. van den Dries, A. Macintyre, and D. Marker. The elementary theory of restricted analytic fields with exponentiation. Ann. of Math. (2), 140(1):183 -- 205, 1994. 4 [14] L. van den Dries and P. Speissegger. O-minimal preparation theorems. In Model theory and applications, volume 11 of Quad. Mat., pages 87 -- 116. Aracne, Rome, 2002. 4 Department of Mathematics, Purdue University, West Lafayette, IN 47907, USA E-mail address: [email protected] Department of Mathematics, Purdue University, West Lafayette, IN 47907, USA E-mail address: [email protected] Department of Computer Science, University of Bath, Bath BA2 7AY, England, UK E-mail address: [email protected]
1906.04100
1
1906
2019-06-10T16:21:50
Equivariant Landau--Ginzburg mirror symmetry
[ "math.AG" ]
We give a new proof of the computation of Hodge integrals we have previously obtained for the quantum singularity (FJRW) theory of chain polynomials. It uses the classical localization formula of Atiyah--Bott and we phrase our proof in a general framework that is suitable for future studies of gauged linear sigma models (GLSM). As a by-product, we obtain the first equivariant version of mirror symmetry without concavity, generalizing the work of Chiodo--Iritani--Ruan on the Landau--Ginzburg side.
math.AG
math
EQUIVARIANT LANDAU -- GINZBURG MIRROR SYMMETRY J´ER´EMY GU´ER´E Abstract. We give a new proof of the computation of Hodge integrals we have previously obtained for the quantum singularity (FJRW) theory of chain polynomials. It uses the classical localization formula of Atiyah -- Bott and we phrase our proof in a general framework that is suitable for future studies of gauged linear sigma models (GLSM). As a by-product, we obtain the first equivariant version of mirror symmetry without concavity, generalizing the work of Chiodo -- Iritani -- Ruan on the Landau -- Ginzburg side. Contents Introduction 0. 1. Localization formula for localized Chern characters 1.1. Localized Chern character 1.2. Localization formula 2. Application to FJRW theory 2.1. Hodge virtual cycle map 2.2. Comparaison with previous work 2.3. Equivariant mirror symmetry References 1 2 2 4 6 6 9 11 12 0. Introduction In our previous work [13], we gave the first genus-zero computation of the vir- tual cycle in the quantum singularity theory, also called Fan -- Jarvis -- Ruan -- Witten (FJRW) theory [9, 10, 18], in a range of cases where the state-of-the-art techniques relying on the concavity condition did not apply. As an application, we proved a mirror symmetry theorem for these theories. Later in [14], we generalized our results and obtained the first all-genus com- putation on the moduli space of Landau -- Ginzburg spin curves, providing we first cap the virtual cycle with the Euler class of the Hodge vector bundle. It lead to Hodge integral calculations in the quantum singularity (FJRW) theory in a range of cases where the techniques relying on Teleman's reconstruction theorem for gener- ically semi-simple Cohomological Field Theories (CohFTs) did not apply. As an application, we proved in [3] the DR/DZ conjecture for 3-spin, 4-spin, and 5-spin theories, that is the equivalence of the Double Ramification (DR) hierarchy with the Dubrovin -- Zhang (DZ) hierarchy for these theories. Interestingly, there are up to now no counterparts of [13, 14] for Gromov -- Witten theory of hypersurfaces in weighted projective spaces, although such a parallel story should appear under the light of Landau -- Ginzburg/Calabi -- Yau correspondence 1 2 GU ´ER ´E [5]. Genus-zero Gromov -- Witten invariants of hypersurfaces in weighted projective spaces are still unknown, as soon as the convexity condition fails. Even for pro- jective hypersurfaces, there is no general description of Hodge integrals in positive genus. Both papers [13, 14] relied on a new technique based on the notion of recursive complexes and it has been our main focus for the last five years to understand how to carry this technique into the Gromov -- Witten side. To achieve this ambitious goal, we shed new light on our previous results by changing our strategy to a more Gromov -- Witten-like approach: we make use of the localization formula of Atiyah -- Bott [1], developped in the algebraic category by [7, 8], to carry the computation of Hodge integrals in FJRW theory. We then give a new and shorter proof of the results in [13, 14]. We also upgrade our mirror theorem [13, Theorem 4.4] to an equivariant version of it, in the spirit of [5, Section 4.3], see Theorem 2.10. We highlight the fact it was out of reach with the previous technique. Importantly, we phrase our new method in a very general framework that is relevant when working with Landau -- Ginzburg models. We thus believe it is suitable for the study of any Gauged Linear Sigma Model (GLSM) [11]. Indeed, following our work [6, Section 6], we see that the definition of virtual cycles in hybrid GLSMs can be phrased as a localized Chern character of a two-periodic complex on a big moduli space denoted (cid:3) in [6], and that the picture in [6, Section 1.5] is a special case of the one we describe in Section 1. Furthermore, it is worth noticing that Gromov -- Witten theory of a complete intersection in a toric Deligne -- Mumford (DM) stack is a special instance of a GLSM via the comparaison [16]. In particular, it is absolutely clear that the strategy developed in Section 1 applies, with little changes compared to Section 2, to the case of hypersurfaces in weighted projective spaces which are defined by chain polynomials. However, writing this paper, we discovered a more direct way to pursue this goal and we decided to leave this result to another paper [15]. Acknowledgement. The author is grateful to Alexander Polishchuk who sug- gested first to look for a proof of the results in [13,14] using the localization formula. 1. Localization formula for localized Chern characters Here, we describe the localization method that we apply to FJRW theory in the next section. We explain it in a more general framework, so that it can serve as a reference for future works regarding GLSM models. 1.1. Localized Chern character. We work over an arbitrary field K and we consider the following set-up: T j Y V E X p S (cid:9) K∗ where S is a proper DM stack, X is a DM stack over S, the substack Y is a local complete intersection in X and is proper over S, and V, E, and T are locally free sheaves (vector bundles) over the DM stacks X, S, and Y . Moreover, we have an action of the multiplicative group K∗ on the fibers of p; precisely an action on X EQUIVARIANT LG MIRROR SYMMETRY 3 and on S such that the action on S is trivial and the projection morphism p : X → S is equivariant. We assume the closed substack Y to be K∗-invariant and the vector bundles V, E, and T to be K∗-equivariant. Furthermore, we assume the K∗-fixed locus of X to be a closed substack of Y that we denote YF . We also consider four global sections α, α′ ∈ H 0(X, V ∨) , α′′ ∈ H 0(X, p∗E) , and β ∈ H 0(X, V ) such that α, α′′, and β are K∗-equivariant but α′ is not, and that For every t1, t2 ∈ K, we define global sections β(α) = β(α′) = 0. α(t1) = α+t1α′ ∈ H 0(X, V ∨) and eα(t1, t2) = α+t1α′+t2α′′ ∈ H 0(X, V ∨⊕p∗E) and Koszul two-periodic complexes K(t1) = (Λ•(V ∨), δ(t1)) and eK(t1, t2) =(cid:16)Λ•(V ∨ ⊕ p∗E),eδ(t1, t2)(cid:17) and eδ(t1, t2) =eα(t1, t2) ∧ · + β(·). over the DM stack X, where the maps are δ(t1) = α(t1) ∧ · + β(·) We observe the following K(t1) is K∗-equivariant ⇐⇒ t1 = 0, eK(t1, t2) is K∗-equivariant ⇐⇒ t1 = 0, and we have the equality of two-periodic complexes (1) Furthermore, we assume eK(t1, 0) = K(t1) ⊗ Λ• (p∗E) . K(t1) is strictly exact off Y ⇐⇒ t1 6= 0, eK(t1, t2) is strictly exact off Y ⇐⇒ (t1, t2) 6= 0. We recall from [17] that a two-periodic complex is strictly exact off Y if it is exact off Y and the images of the maps are subbundles. As a consequence, we get localized Chern characters ∀ t1 6= 0 , ChX Y (K(t1)) ∈ A(Y → X)Q, ∀ (t1, t2) 6= 0 , ChX Y (eK(t1, t2)) ∈ A(Y → X)Q in the bivariant Chow rings. For their constructions, we refer to [12] for complexes and to [17] for two-periodic complexes. Definition 1.1. We call virtual class for the above input data the Chow class1 (2) cvir := (p ◦ j)∗(cid:16)Td(T ) ∪ ChX Y (K(1))[X](cid:17) ∈ A∗(S). We call Hodge virtual class the following Chow class (3) ctop(E) · cvir ∈ A∗(S), where ctop stands for the top Chern class of a vector bundle. Hodge integrals refer to integrals of the Hodge virtual class against other Chow classes on the space S. 1In this formula, we recall that the projection p ◦ j : Y → S is proper. 4 GU ´ER ´E Remark 1.2. The reason for the name "virtual class" is explained in Section 2. Indeed, once we choose appropriate input data, we prove it corresponds to the virtual class from the quantum singularity theory or even from Gromov -- Witten theory of hypersurfaces or more general GLSM, see [6, Section 6]. Moreover, the vector bundle E plays the role of the Hodge bundle over the moduli space of stable curves, hence the second definition. Remark 1.3. It is a huge open challenge to express the virtual class cvir in a simple way. However, the Hodge virtual class can be expressed in terms of the fixed loci of the K∗-action, see Theorem 1.7, and thus it is often computable. We also recall that Hodge integrals play an important role in the definition of the Double Ramification hierarchy, see [2]. Proposition 1.4. We have an equality ctop(E) · cvir = (−1)rk(E)(p ◦ j)∗(cid:16)Td(T ⊕ j∗p∗E∨) ∪ ChX in the Chow ring of S, where rk stands for the rank of a vector bundle. Y (eK(0, 1))[X](cid:17) Proof. Using invariance of the localized Chern character by homotopy and equation (1), we see that ∀t1, t2 6= 0 , ChX Y (eK(0, 1)) = ChX Y (eK(t1, t2)) Y (eK(t1, 0)) = ChX = Ch(λ−1(p∗E)) · ChX = Ch(λ−1(p∗E)) · ChX Y (K(t1)) Y (K(1)), where the lambda-class is defined in K-theory on a vector bundle W by λt(W ) =Xq≥0 ΛqW tq ∈ K 0[t]. Therefore, we obtain the desired equality after using the classical formulae Ch(λ−1(W ∨))Td(W ) = ctop(W ) and ctop(W ∨) = (−1)rk(W )ctop(W ). (cid:3) 1.2. Localization formula. We recall all input data, except the global section α′ ∈ H 0(X, V ∨), are K∗-equivariant. We denote by q the equivariant parameter, by AK∗ ∗ (·)q the ring obtained from it by inverting the equivariant parameter q, see [7] for a detailed construction of the equivariant Chow ring. ∗ (·) the equivariant Chow ring, and by AK∗ Proposition 1.5 ([8, Thm 1]). We have an isomorphism of groups ∗ (Y )q ≃ AK∗ given by the equivariant pushforward of embeddings. ∗ (YF )q ≃ AK∗ AK∗ ∗ (X)q, Since the substack Y is a complete intersection inside the stack X, we have an explicit description of the second isomorphism above, yielding a localization formula as the one proved by Atiyah -- Bott [1] in equivariant cohomology. Theorem 1.6 (Localization formula). Denote by Nj = NY ⊂X the normal bundle of the local complete intersection, it is a K∗-equivariant vector bundle over Y . We have the following formula [X] = j∗ [Y ] top(Nj)! ∈ AK∗ cK∗ ∗ (X)q, EQUIVARIANT LG MIRROR SYMMETRY 5 top is the K∗-equivariant2 top Chern class. Furthermore, if the closed im- where cK∗ mersion ιF : YF ֒→ Y of the fixed locus inside Y is also a local complete intersection, we have the same result replacing Y by YF , j by j ◦ ιF , and the normal bundle Nj by Nj◦ιF . Proof. By the surjectivity of the second map in Proposition 1.5, there exists an equivariant class a ∈ AK∗ ∗ (Y )q such that Therefore, we have [X] = j∗(a) ∈ AK∗ ∗ (X)q. [Y ] = j∗[X] = j∗j∗(a) = cK∗ top(Nj) ∪ a, where in the last equality we use that j is a local complete intersection morphism. Therefore, dividing3 both sides by cK∗ (cid:3) top(Nj), we get the result. Applying the localization formula to the right-hand side of the equality in Propo- sition 1.4, we obtain a simple formula for the Hodge virtual class. Theorem 1.7. The Hodge virtual class equals ctop(E∨) · cvir = lim q→0 top(E∨) · (p ◦ j)∗ cK∗ top(j∗V ) cK∗ top(Nj) cK∗ Td(j∗T ) Td(j∗V )! , · where we recall the map p ◦ j is proper. On the right-hand side, the Todd, Chern, and top Chern classes are all taken K∗-equivariantly. Remark 1.8. The Chow class on the right-hand side lives in the equivariant Chow ring AK∗ ∗ (S)q ≃ A∗(S)((q)), as the action on S is trivial. Precisely, Theorem 1.7 means that it contains no negative powers of q and that the constant term in q, that we interpret as a limit q → 0, equals the left-hand side. Proof. By Proposition 1.4, we rewrite the Hodge virtual class in terms of the local- ized Chern character of the two-periodic complex eK(0, 1). Since it is K∗-equivariant, we use the localization formula to get ChX Y (eK(0, 1))[X] = ChX = ChY cK∗ Y (eK(0, 1))"j∗ [Y ] top(Nj)!# Y (j∗eK(0, 1))" top(Nj)# Ch(j∗eK(0, 1)) cK∗ top(Nj) ∗ (Y )q. ∈ AK∗ cK∗ [Y ] = Furthermore, we have a K-theoretic equality j∗eK(0, 1) = λ−1(j∗V ∨ ⊕ j∗p∗E) = λ−1(j∗V ∨) ⊗ λ−1(j∗p∗E). 2We put the upper-script K∗ to emphasize that it is the equivariant top Chern class, even though it should be clear from the context. In particular, the localized Chern characters in the proof below are also K∗-equivariant, although we do not write the upper-script K∗. 3Since the substack Y ⊂ X is K∗-invariant, its normal bundle has no fixed part under K∗ and its equivariant top Chern class is then invertible. 6 GU ´ER ´E Thus, we obtain Td(T ⊕ j∗p∗E∨) · ChX in the ring AK∗ (4) (p◦j)∗(cid:16)Td(T ⊕ E∨) ∪ ChX Y (eK(0, 1))[X] = j∗p∗(cK∗ Y (eK(0, 1))[X](cid:17) = cK∗ top(E∨)) · Td(j∗T ) Td(j∗V ) · cK∗ top(j∗V ) cK∗ top(Nj) top(E∨)·(p◦j)∗ Td(j∗T ) Td(j∗V ) top(Nj)! . cK∗ top(j∗V ) cK∗ · ∗ (Y )q. Eventually, we push-forward to the space S and get Notice that at this point, it is a K∗-equivariant equality, taking place in the ring AK∗ ∗ (S)q, which equals the ring A∗(S)((q)) since the K∗-action on S is trivial. It is not clear that the right-hand side of equality (4) contains no negative powers of the equivariant parameter q, but it follows from the fact the left-hand side has a non-equivariant limit q → 0. At last, once we take the non-equivariant limit, or equivalently the constant term in q, in equality (4), the left-hand side becomes the Hodge virtual class by Proposition 1.4 and we obtain the desired equality. (cid:3) 2. Application to FJRW theory In this section, we work over the base field K = C and we give an application of Theorem 1.7 to the quantum singularity (FJRW) theory, shedding new light on the results of [13, 14]. We then use most of the same notations and refer to [13, 14] for more details. Once for all, we fix a quasi-homogeneous polynomial W of chain type W (x1, . . . , xN ) = xa1 1 x2 + · · · + xaN −1 N −1 xN + xaN N , where integers a1, . . . , aN are positive. We denote by d its degree and by w1, . . . , wN the weights of the variables x1, . . . , xN . We also fix two non-negative integers g and n such that 2g − 2 + n > 0, i.e. the space of stable curves Mg,n is non-empty. We further consider an admissible group G, which is a subgroup of the maximal group Aut(W ) of (diagonal) symmetries of W , containing the grading element j, see [13, Equation (3)]. 2.1. Hodge virtual cycle map. Let us consider the Landau -- Ginzburg orbifold (W, G) (see [13, Definition 1.2]) and denote by H the state space of its FJRW theory. It decomposes as H =Mγ∈G Hγ where we recall that γ is a diagonal matrix γ = diag(γ1, . . . , γN ) and that γ is called broad if at least one of its entries equals 1 and narrow otherwise. In particular, for γ narrow, we have Hγ ≃ C. For any γ = (γ(1), . . . , γ(n)) ∈ Gn satisfying the selection rule γ(1) · · · · · γ(n) = j2g−2+n, where j ∈ G is the grading element, we have a moduli space of (W, G)-spin curves, that we denote by Sg,n(γ), see [9, Section 2] or [14, Section 1.2]. We denote by Sg,n the disjoint union over all possible γ and we have a finite map o : Sg,n → Mg,n. EQUIVARIANT LG MIRROR SYMMETRY 7 The virtual cycle map4 is a linear map cvir : H⊗n → A∗(Sg,n) such that for entries u1 ∈ Hγ(1), . . . , un ∈ Hγ(n) we have cvir(u1, . . . , un) ∈ H 2degvir(Sg,n(γ(1), . . . , γ(n)), where the integer degvir is explicitly determined by the genus g and the matrices γ(1), . . . , γ(n), see [13, Equation (16)]. The Hodge bundle E on the moduli space Mg,n is the rank-g vector bundle given by π∗ω, where π is the map from the universal curve and ω is the relative dualizing sheaf. Its fiber on the point representing a curve C is then H 0(C, ωC ). Definition 2.1. We call Hodge virtual cycle map the product λg · cvir : H⊗n → A∗(Sg,n), where λg := ctop(E) is the top Chern class of the (pull-back of the) Hodge bundle. From now on, we fix entries u1 ∈ Hγ(1), . . . , un ∈ Hγ(n) which are invariant under the maximal group Aut(W ), see [14, Section 1.3] for an explicit description. In particular, we are given a subset Cγ(i) ⊂ {x1, . . . , xN } of variables for each marking, see [13, Definition 1.5]. Before stating Theorem 2.4, we introduce a few more notations. Let us denote by π : C → Sg,n the universal curve over the moduli space of (W, G)-spin curves and by L1, . . . , LN the universal line bundles. In [13, Equation (72)] or [14, Equation (11)], we define the modified line bundles LC N on C, which are obtained from the universal line bundles by twisting down some of the markings. We also need the following definition. 1 , . . . , LC Definition 2.2. The equivariant Euler class of a vector bundle V on a space S is defined by eq(V ) = qrk(V ) · 1 + c1(V ) q + c2(V ) q2 + · · · + crk(V )(V ) qrk(V ) ! , (5) and extended multiplicatively to K-theory as a map eq : K 0(S) → A∗(S)[q, q−1]]. Remark 2.3. Let C∗ act trivially on a space S and fiberwise on a vector bundle V . The C∗-equivariant top Chern class of V equals the equivariant Euler class, i.e. cC∗ top(V ) = eq(V ), where q is the equivariant parameter. Theorem 2.4. Under the previous assumptions and notations, we have (6) λg · cvir(u1, . . . , un) = lim q→0 e−qN +1(E) · eqj (−R•π∗(LC j )), NYj=1 where q1 := q and qj+1 := (−a1) · · · (−aj) q for 1 ≤ j ≤ N . Proof. First of all, we take resolutions of the higher push-forwards R•π∗(LC all j by vector bundles over Sg,n j ) for 4We use the construction of the virtual cycle map by Polishchuk and Vaintrob [18]. Rπ∗LC j = [Aj → eBj], 8 GU ´ER ´E such that there exist the appropriate morphisms from [14, Equation (12)] (7) eαj eβj : O → Symaj−1 A∨ : j → A∨ j , j−1 ⊗ eB∨ eB∨ j ⊕ (Symaj −1A∨ j ⊗ A∨ j , j+1) ⊗ eB∨ N +eα′′ N where with the convention (A0, AN +1) = (0, AN ). Moreover, we decompose the morphism eαN into the sum eα′ N : O → SymaN −1 A∨ N and we also consider the morphism from [14, Equation (19)] eα′ eαN +1 : O → SymaN A∨ N −1 ⊗ eB∨ N ⊗ E. We apply Theorem 1.7 to the following data: • S := Sg,n(γ(1), . . . , γ(n)) is the moduli space of (W, G)-spin curves, • E := E is the Hodge bundle, • X := Tot (A1 ⊕ · · · ⊕ AN ) is the total space of the vector bundle, with p : X → S the projection, • Y := Sg,n(γ(1), . . . , γ(n)) embedded in X via the zero section Y ֒→ X, • V := p∗B1 ⊕ · · · ⊕ p∗BN , • T := B1 ⊕ · · · ⊕ BN , N viewed as a global section of V ∨ on X, N viewed as a global section of V ∨ on X, • α :=eα1 + · · · +eαN −1 +eα′ • α′ :=eα′′ • α′′ :=eαN +1 viewed as a global section of p∗E on X, • β := eβ1 + · · · +eβN viewed as a global section of V on X. It follows from [13, Section 3.5] that β(α) = β(α′) = 0. Furthermore, we take the following C∗-action: • trivial on S, • scaling fibers with weight 1 on A1 and on B1, i.e. λ · v = λ v on a vector v, • scaling fibers with weight (−a1) · · · (−aj) on Aj+1 and on Bj+1, i.e. λ · v = λ(−a1)···(−aj ) v on a vector v, • scaling fibers with weight (−a1) · · · (−aN ) on the Hodge bundle E. It is straightforward to see that the global sections α, α′′, and β are C∗-equivariant, and that α′ is not. Moreover, the C∗-fixed locus in X is given by the constraint ∀λ ∈ C∗ , λ · (x1, . . . , xN ) = (λ x1, . . . , λ(−a1)···(−aN −1) xN ) = (x1, . . . , xN ), yielding (x1, . . . , xN ) = 0, i.e. the C∗-fixed locus is YF = Y = S. Following Section 1, we form the two-periodic complexes K(t1) and K(t1, t2) for (t1, t2) ∈ C2, and they are C∗-equivariant when t1 = 0. Looking at the common vanishing locus of the global sections α, α′′, and β, it follows that K(t1) is strictly exact when t1 6= 0 and K(t1, t2) is strictly exact when (t1, t2) 6= 0. As a consequence, all assumptions from Section 1 are fulfilled. By definition of the virtual cycle map, the equality (8) cvir(u1, . . . , un) = p∗(cid:16)Td(T ) ∪ ChX Y (K(1))[X](cid:17) ∈ A∗(S) EQUIVARIANT LG MIRROR SYMMETRY 9 is exactly [4, Lemma 5.3.8]. Therefore, we obtain by Theorem 1.7 ctop(E∨) · cvir(u1, . . . , un) = lim q→0 top(E∨) · (p ◦ j)∗ cK∗ top(j∗V ) cK∗ top(Nj) cK∗ Td(j∗T ) Td(j∗V )! · = lim q→0 cK∗ top(E∨) · = lim q→0 eqN +1(E∨) · cK∗ top(j∗V ) cK∗ top(Nj) eq1(B1) · · · eqN (BN ) eq1(A1) · · · eqN (AN ) , where we use that the normal bundle Nj equals the vector bundle A1 ⊕ · · · ⊕ AN . At last, the equality eq(E∨) = (−1)rkEe−q(E) concludes the proof. (cid:3) 2.2. Comparaison with previous work. Theorem 2.4 computes exactly the same class as [14, Theorem 2.2]. We then have to compare the two formulae. Remark 2.5. It is interesting to see that Equation (6) from Theorem 1.7 was al- ready written in [14, Equation (24)], where it was deduced from the computation of the sum over dual graphs, see [14, Section 3]. In particular, the expression of Equation (6) as a sum over dual graphs is presented in [14, Corollary 3.5]. Nev- ertheless, we give below a more comprehensive way to understand the relationship between Theorem 2.4 and [14, Theorem 2.2]. In [13, 14], we use the notion of recursive complexes to obtain the formula for the Hodge virtual cycle map and it uses a multiplicative characteristic class defined as follows. On a vector bundle V over a space S, it is ct(V ) = Ch(λ−tV ∨)Td(V ) ∈ A∗(S)[t] and in terms of its roots α1, . . . , αv, it is ct(V ) = eαk − t eαk − 1 · αk. vYk=1 It is multiplicative on vector bundles and then extended multiplicatively to K-theory into a function ct : K 0(S) → A∗(S)[[t]]. We have the fundamental property (9) ∀R, R′ ∈ K 0(S) , ct(R + R′) = ct(R) · ct(R′). Note also that the characteristic class ct is actually defined for t 6= 1, and not only for a formal parameter t, using Chern characters. Explicitly, for R ∈ K 0(S), we have (10) (11) with the functions ct(R) = (1 − t)Ch0(R) · exp−Xl≥1 (k − 1)!(cid:18) t sl(t)Chl(R) , 1 − t(cid:19)k + (−1)l Bl(0) sl(t) = γ(l, k). l lXk=1 Here, the number γ(l, k) is defined by the generating function γ(l, k) zl l! := (ez − 1)k k! . Xl≥0 10 GU ´ER ´E We notice that γ(l, k) vanishes for k > l and that the sum over l in (10) is finite because Chl vanishes for l > dim(S). Similarly, the equivariant Euler class (see Definition 2.2) is actually defined for q 6= 0 using Chern characters. Explicitly, for R ∈ K 0(S), we have (12) eq(R) = qCh0(R) · exp−Xl≥1 (l − 1)! (−q)l Chl(R) . It also has the multiplicativity property (13) ∀R, R′ ∈ K 0(S) , eq(R + R′) = eq(R) · eq(R′), and in terms of roots α1, . . . , αv of a vector bundle V , it takes the simple form eq(V ) = q + αk. vYk=1 Proposition 2.6. Let R ∈ K 0(S) and q be a formal parameter or be in K∗. We have the relation (14) eq(R) = ce−q (R) · Td(R ⊗ O(q)) Td(R) , where O(q) is a formal line bundle with first Chern class q. Precisely, we have (15) Td(R ⊗ O(q)) Td(R) Proof. The equality = exp−Xk≥0 Td(R) = exp −Xl>1 l>k ql−kChk(R) . Chl(R)! Bl(0) l Bl(0) l is easy to prove using the multiplicativity of the Todd class and the formula for a line bundle L. Thus, we get Equation (15) using Td(L) = c1(L) 1 − e−c1(L) Chl(R ⊗ O(q)) = Chk(R) ql−k (l − k)! . lXk=0 Similarly, since the classes eq, ce−q , and Td are multiplicative, it is enough to check equation (14) on a line bundle L. Denoting α := c1(L), we obtain ce−q (L) · Td(L ⊗ O(q)) Td(L) = eα − e−q eα − 1 · α · α + q 1 − e−α−q · 1 − e−α α = α + q = eq(L). Definition 2.7. We say that a formal power series in q, q−1 is convergent when q → 0 if all coefficients of negative powers in q are zero. These coefficients yield relations in the coefficient ring of the formal power series. Moreover, we call limit of a convergent formal power series its constant term in q. (cid:3) EQUIVARIANT LG MIRROR SYMMETRY 11 q → 0. Furthermore, under convergence, we have Corollary 2.8. Let R1, . . . , RN ∈ K 0(S) and k1, . . . , kN ∈ Z. We set qj := kj · q, j=1 ctj (Rj) is convergent j=1 eqj (Rj ) is convergent when tj := tkj , and t := e−q. Then, the formal power series QN when q → 0 if and only if the formal power seriesQN NYj=1 eqj (Rj) = lim q→0 ctj (Rj) = lim t→1 NYj=1 lim q→0 NYj=1 ctj (Rj) and the two sets of relations in A∗(S) are equivalent. Proof. Since we have q 2 + q2 · f (q) + q · g(q), with f (q) ∈ C[[q]] and g(q) ∈ Adeg≥1(S)[[q]], then we get exp−Xk≥0 l>k Bl(0) l ql−kChk(R) = 1 + NYj=1 eqj (Rj ) ctj (Rj ) = 1 + O(q) and the claims follow easily. (cid:3) Corollary 2.9. Theorem 2.4 gives the same result as [14, Theorem 2.2]. In partic- ular, it recovers [13, Theorem 3.21] as a special case for genus zero. Furthermore, the set of tautological relations presented in [14, Section 2.3] is equivalent to the set of tautological relations obtained by looking at the coefficients of negative powers of q in the right-hand side of Equation (6). 2.3. Equivariant mirror symmetry. [13, Section 4] can be entirely rewritten with the specialization of the twisted theory given by sj 0 := 1 qj and sj l := (l − 1)! (−qj)l for l ≥ 1, with q1 := q and qj+1 := (−a1) · · · (−aj) q for 1 ≤ j < N , instead of the specializa- tion given by [13, Equation (67)]. According to Corollary 2.9, it recovers the same big I-function of [13, Theorem 4.2] and the same small I-function and Picard -- Fuchs equation of [13, Theorem 4.4] once we take the limit q → 0. However, it is interest- ing to consider the equivariant version of these results, i.e. without taking the limit q → 0. In [13, Theorem 4.2], the only change is that, with the notations from there, the contribution Mj(γ) becomes j (γ)−1 Mj(γ) =  Y0≤m≤DC Y1≤m≤−DC j (γ) (ωC j (Γ) + m)z + qj when DC j (γ) ≥ 1, 1 when DC j (γ) = 0, 1 (ωC j (Γ) − m)z + qj , when DC j (γ) ≤ −1. 12 GU ´ER ´E As a consequence, we deduce the equivariant small I-funtion for chain polynomials of Calabi -- Yau type5 Theorem 2.10 (Equivariant Mirror Symmetry). Let W be a chain polynomial of Calabi -- Yau type. The equivariant I-function6 defined for t ∈ C∗ by (16) I(t, −z) = −z (bz + qj) ejk , δj := −δ{N −j is odd} tkQN ∞Xk=1 b≥0,hbi=hqj ki j=1Q δj <b<qj k Q0<b<k bz lies on the Lagrangian cone L of the equivariant FJRW theory of the Landau -- Ginzburg orbifold (W, G). This function satisfies the Picard -- Fuchs equation (17) (cid:20)td NYj=1 wj −1Yc=0 (qjzt ∂ ∂t + cz + qj) − (zt dYc=1 ∂ ∂t − cz)(cid:21) · I(t, −z) = 0. References [1] Michael Atiyah and Raoul Bott, The moment map and equivariant cohomology, Topology 23 (1984), 1-28. [2] Alexandr Buryak, Double ramification cycles and integrable hierarchies, Communications in Mathematical Physics 336 (2015), no. 3, 1085-1107. [3] Alexandr Buryak and J´er´emy Gu´er´e, Towards a description of the double ramification hier- archy for Witten's r-spin class, Journal de Math´ematiques Pures et Appliqu´ees 106 (2016), no. 5, 837-865. [4] Alessandro Chiodo, The Witten top Chern class via K-theory, Journal of Algebraic Geometry 15 (2006), 681-707. [5] Alessandro Chiodo, Hiroshi Iritani, and Yongbin Ruan, Landau -- Ginzburg/Calabi -- Yau cor- respondence, global mirror symmetry and Orlov equivalence, Publications math´ematiques de l'IH´ES 119 (2014), no. 1, 127-216. [6] Ionut Ciocan-Fontanine, David Favero, J´er´emy Gu´er´e, Bumsig Kim, and Mark Shoemaker, Fundamental Factorization of a GLSM, Part I: Construction, available at arXiv:1802.05247. [7] Dan Edidin and William Graham, Equivariant intersection theory, Inventiones Math. 131 (1998), 595-634. [8] , Localization in equivariant intersection theory and the Bott residue formula, Amer- ican Journal of Mathematics 120 (1998), no. 3, 619-636. [9] Huijun Fan, Tyler Jarvis, and Yongbin Ruan, The Witten equation, mirror symmetry and quantum singularity theory, Ann. of Math. 178 (2013), no. 1, 1-106. [10] [11] , The Witten equation and its virtual fundamental cycle, available at arXiv:0712.4025. , A mathematical theory of the gauged linear sigma model, Geometry and Topology (2017). [12] William Fulton, Intersection theory, Springer -- Verlag. [13] J´er´emy Gu´er´e, A Landau -- Ginzburg Mirror Theorem without Concavity, Duke Mathematical Journal 165 (2016), no. 13, 2461-2527. [14] , Hodge integrals in FJRW theory, Michigan Mathematical Journal 66 (2017), no. 4, 831-854. [15] [16] Bumsig Kim and Jeongseok Oh, Localized Chern characters for 2-periodic complexes, , Hodge -- Gromov -- Witten theory (2019). avalaible at arXiv:1804.03774. [17] Alexander Polishchuk and Arkady Vaintrob, Algebraic construction of Witten's top Chern class, Contemp. Math. 276 (2001). [18] , Matrix factorizations and cohomological field theories, J. Reine Angew. Math. 714 (2016), 1-122. 5The Calabi -- Yau condition means that the degree equals the sum of the weights, i.e. d = w1 + · · · + wN . 6There is a typo in [13, Equation (98)] as the condition b ≥ 0 is not specified. EQUIVARIANT LG MIRROR SYMMETRY 13 Univ. Grenoble Alpes, CNRS, IF, 38000 Grenoble, France E-mail address: [email protected]
1201.3558
3
1201
2013-01-05T05:25:59
P1-bundles over projective manifolds of Picard number one each of which admit another smooth morphism of relative dimension one
[ "math.AG" ]
We give a complete classification of P1-bundles over a projective manifold of Picard number one which admit another smooth morphism of relative dimension one.
math.AG
math
P1-BUNDLES OVER PROJECTIVE MANIFOLDS OF PICARD NUMBER ONE EACH OF WHICH ADMIT ANOTHER SMOOTH MORPHISM OF RELATIVE DIMENSION ONE KIWAMU WATANABE Abstract. We give the complete classification of P1-bundles over projective manifolds of Picard number one each of which admit another smooth morphism of relative dimension one. Contents Introduction 1. 2. A Part of Theorem 1.1 (I) and Structures of A2(X)Q and N 2(X)Q 3. Computation of the discriminant ∆(E ) 4. Proof of Theorem 1.1 5. Proof of Theorem 1.3 References 1 3 4 9 12 12 1. Introduction R. Munoz, G. Occhetta and L. Sol´a Conde studied rank 2 vector bundles on Fano manifolds in [11]. In their paper [11, Theorem 6.5], they obtained a complete list of P1-bundles over Fano manifolds with b2 = b4 = 1 that have another second P1- bundle structure. The purpose of this paper is to generalize their result. Actually, we give the complete classification of P1-bundles over projective manifolds of Picard number 1 each of which admit another smooth morphism of relative dimension 1. Our main result is the following: Theorem 1.1. Let X be a complex projective manifold of Picard number ρ = 1 and E a rank 2 vector bundle on X. Assume that Z := P(E ) → X admits another smooth morphism Z → Y of relative dimension 1 and n := dim X ≥ 2. Then, (I) X and Y are Fano manifolds of ρ = 1 and there exists a rank 2 vector bundle E ′ on Y such that Z → Y is given by PY (E ′). Furthermore, (II) if E and E ′ are normalized by twisting with line bundles (i.e., c1 = 0 or −1), then ((X, E ), (Y, E ′)) is one of the following, up to changing the pairs (X, E ) and (Y, E ′): (a) ((P2, TP2), (P2, TP2)), where TP2 is the tangent bundle of the projective plane P2, Date: February 7, 2012. 2000 Mathematics Subject Classification. Primary 14J45, 14J60, Secondary 14M17. Key words and phrases. P1-bundle, rank 2 vector bundle, Fano manifold, homogeneous manifold. 1 2 KIWAMU WATANABE (b) ((P3, N ), (Q3, S )), where N is the null-correlation bundle on P3 (see [13]) and S is the restriction to the 3-dimensional quadric Q3 of the universal quotient bundle of the Grassmannian G(1, P3), (c) ((Q5, C ), (K(G2), Q)), where C is a Cayley bundle on Q5 (see [14]), K(G2) is the 5-dimensional Fano homogeneous contact manifold of type G2 which is a linear section of the Grassmannian G(1, P6) and Q the restriction of the universal quotient bundle on G(1, P6). Consequently, Z is the full-flag manifold of type A2, B2 or G2. In particular, X, Y and Z are rational homogeneous manifolds. Another motivation of our main result is the following conjecture proposed by Campana and Peternell: Conjecture 1.2 ([3]). A Fano manifold M with nef tangent bundle is homogeneous. This conjecture is true in dimension ≤ 4. The most difficulty lies in the case where M is a Fano 4-fold of ρ = 1 which carries a rational curve C with −KM .C = 3. In this case, N. Mok [10] proved the conjecture under the additional assumption that b4(M ) = 1. Later on, J. M. Hwang pointed out that the assumption b4(M ) = 1 can be removed in [7]. More generally, they obtained the following result. We will also prove this as a corollary of Theorem 1.1. Theorem 1.3 ([10, Main Theorem], [7]). Let M be a Fano manifold of ρ = 1 with nef tangent bundle. Assume that M carries a rational curve C such that −KM .C = 3. Then M is isomorphic to P2, Q3 or K(G2). The contents of this paper are organized as follows. Section 2 is devoted to study the structures of the Chow group A2(X)Q of 2-dimensional cycles with Q- coefficients and its quotient N 2(X)Q by numerical equivalence, according to a similar argument as in [7]. In Section 3, we give two computations of the discriminant ∆(E ) := c2 1(E )−4c2(E ) which are based on ideas in [10] and [11] (see Proposition 3.8 and 3.14). In Section 4, by comparing the computational results, we narrow down the possible values of some invariants of X and Y . Then we can show the existence of a rank 2 vector bundle E ′ on Y such that Z = PY (E ′) (Proposition 4.5). The main novelty of this paper is to show the existence of E ′ by using the above two computational results of the discriminant. After this paper was submitted, the draft of [11] was revised and they obtained the same result (see [11, Lemma 6.2]). Since Z admits double P1-bundle structures π : Z → X and φ : Z → Y , the same argument as in [11] implies Theorem 1.1 (Theorem 4.6). In the final section, we will show Theorem 1.3 as a corollary of Theorem 1.1. In this paper, we use notation as in [5] and every point on a variety we deal with is a closed point. We work over the field of complex numbers. Acknowledgements. This work was done while the author was visiting the Uni- versity of Freiburg. The author would like to express his gratitude to his host Prof. Stefan Kebekus for all his assistance and his kindness. He would also like to thank the staff in the department of mathematics at the University of Freiburg, especially Patrick Graf for his support and encouragement. My deepest appreciation goes to Luis E. Sol´a Conde and Roberto Munoz for inviting me to visit Universidad Rey Juan Carlos. The author would like to express his gratitude to Gianluca Occhetta for sending the final version of their paper [11]. The author is partially supported by P1-BUNDLES OVER PROJECTIVE MANIFOLD OF PICARD NUMBER ONE 3 Research Fellowships of the Japan Society for the Promotion of Science for Young Scientists. 2. A Part of Theorem 1.1 (I) and Structures of A2(X)Q and N 2(X)Q First, we partially prove Theorem 1.1 (I). We will complete the proof of Theo- rem 1.1 (I) in Proposition 4.5. Proposition 2.1. Let X be a projective manifold of ρX = 1 and E a rank 2 vector bundle on X. Assume that π : Z = P(E ) → X admits another smooth morphism φ : Z → Y of relative dimension 1 and n := dim X ≥ 2. Then (i) Y is a Fano manifold of ρY = 1, (ii) φ−1(y) is isomorphic to P1 for every y ∈ Y , and (iii) X is also a Fano manifold of ρX = 1. It is easy to see (i). In fact, since ρZ = 2, it turns out that ρY = 1. Additionally, Y is covered by rational curves which are the images of fibers of π. Thus Y is a Fano manifold of ρY = 1. We use the following lemma to prove Proposition 2.1 (ii): Lemma 2.2. Let X be a projective manifold and Y a Fano manifold of ρ = 1. Assume that a projective manifold Z admits two different smooth morphisms π : Z → X and φ : Z → Y of relative dimension 1 and π−1(x) is isomorphic to P1 for every x ∈ X. Given y ∈ Y , define inductively (i) V 0 (ii) V m+1 y := {y}, and y := φ(π−1(π(φ−1(V m y )))). Then there exists a natural number l such that V l y = Y . y = Y provided dim V k y = dim V k+1 y y = dim V k+1 Proof. The idea of this proof is in [9]. Since V k Y , it is sufficient to show that V k dim V k Assume that dim V k that V k codimension of V k y is an irreducible closed subset of . Remark that y is independent of the choice of y ∈ Y . It follows from flatness of π and φ. . Supposing y does not coincide with Y , we shall derive a contradiction. Let q be the y in Y and T ⊂ Y a (q − 1)-dimensional projective subvariety. From our assumption, we have q ≥ 1. Denote Sy∈T V k y by A. Since ρY = 1, A is an ample divisor on Y . Hence, for any point x ∈ X, φ(π−1(x)) ∩ A 6= ∅, then there yx 6= ∅. This implies that φ(π−1(x)) is exists a point yx ∈ T such that φ(π−1(x)) ∩ V k contained in V k+1 for any y ∈ Y . Then V k yx ⊂ A. However this contradicts the surjectivity of φ. y = V k+1 y yx = V k (cid:3) y Proof of Proposition 2.1(ii) and (iii). Assume that there exists an irrational fiber of φ. Then every fiber of φ is not rational. Let f be a fiber of π and ν the restriction of φ to f ∼= P1. Consider a smooth family of curves ν ∗Z → f ∼= P1. Since a fiber of ν ∗Z → f ∼= P1 is not rational, the family is isotrivial. Furthermore, the family is trivial by virtue of the simply-connectedness of P1. It turns out that π(φ−1(y1)) = π(φ−1(y2)) for any y1, y2 ∈ φ(π−1(x)) provided we fix a point x ∈ X. From Lemma 2.2, it follows that any two point can be connected by a chain of rational curves φ(π−1(x)) of finite length. Hence we see that π(φ−1(y1)) = π(φ−1(y2)) for any y1, y2 ∈ Y . However, this is a contradiction to the surjectivity of π and dim ≥ 2. As a consequence, every fiber of φ is rational. Now (iii) follows in a similar way to (i). (cid:3) 4 KIWAMU WATANABE According to a similar argument as in [7, Sect. 4], we prove the following: Proposition 2.3. Let X be an n-dimensional Fano manifold of ρ = 1 and E a rank 2 vector bundle on X. Assume that Z := P(E ) → X admits another smooth morphism φ : Z → Y whose fiber is isomorphic to P1. If n ≥ 2, then A2(X)Q and N 2(X)Q are isomorphic to a 1-dimensional vector space Q over the field of rational numbers. Proof. For a point y ∈ Y , we define inductively the varieties W k y as follows: y andgW k (i) W 0 (ii) W k y := W 0 y := φ−1(y), gW 0 ^ W k−1 y := y ×Y Z, gW k y ×X Z and y := W k y ×X Z. y Remark that y as above. has a natural morphism to Y defined by the composition of a y admits a natural y → Z and π : Z → X. Hence projection morphism to X by the composition of a projection W k we can define W k ^ W k−1 y ^ W k−1 y → Z and φ : Z → Y . On the other hand, W k y and gW k ^ W l−1 y → Y is surjective. Hence W l y)Q → A2(X)Q is at most 1. Since W k y ) and A0(gW k y ) and A1(gW k For a point y ∈ Y , the image of the composition of a projection gW k φ : Z → Y coincides with V k+1 such that is A2(W l the rank of A2(W l connected, A0(W k follows from Lemma 2.4 below that A1(W k whose images in Z are either a fiber of π or a fiber of φ. Furthermore, A2(W k y → Z and as in Lemma 2.2. Therefore, there exists l ∈ N y → X is also surjective. Then, so y)Q → A2(X)Q. Thus, to prove A2(X)Q ∼= Q, we only have to show that y are rationally y ) are isomorphic to the ring of integers Z. Then it y ) are generated by curves y ) and y ) are generated by surfaces whose images in Z are either a curve or surfaces of the form φ−1(φ(C)) for some fiber C of π or π−1(π(C ′)) for some fiber C ′ of φ. Hence the rank of A2(W l y)Q → A2(X)Q is at most 1. Since A2(X)Q → N2(X)Q is surjective, N2(X)Q is also isomorphic to Q. Thus we obtain N 2(X)Q ∼= Q. Lemma 2.4. Let p : W ′ → W be a P1-bundle with a section σ : W → W ′. Then any γ ∈ Ak(W ′) is of the form γ = σ∗α + p∗β for some α ∈ Ak(W ) and β ∈ Ak−1(W ). Proof. See [4, Theorem 3.3]. y and gW k A2(gW k (cid:3) (cid:3) 3. Computation of the discriminant ∆(E ) Throughout this section, we work under the following assumptions: Assumptions 3.1. Let X and Y be n-dimensional Fano manifolds of ρ = 1 and E a normalized rank 2 vector bundle over X, i.e., c1 := c1(E ) = 0 or −1 (when the Picard group of X is identified with Z). Assume that π : Z = P(E ) → X admits another smooth morphism φ : Z → Y whose fibers are isomorphic to P1 and n := dim X ≥ 2. Notation 3.2. the ample generator of Pic(X) (resp. Pic(Y )). Note that X and Y are Fano manifolds of ρ = 1 and hence Pic(X) ∼= Z and Pic(Y ) ∼= Z. • HX (resp. HY ): P1-BUNDLES OVER PROJECTIVE MANIFOLD OF PICARD NUMBER ONE 5 X, dY := H n Y . 1(E ) − 4c2(E ) • iX (resp. iY ): the Fano index of X (resp. Y ). • H := π∗HX, H ′ := φ∗HY . • dX := H n • f (resp. f ′): a fiber of π (resp. φ). • µ := H.f ′, µ′ := H ′.f . • ∆(E ) := c2 • Σ: an effective cycle on X of codimension 2 such that N 2(X)Q = QΣ (cf. Proposition 2.3) • c2(E ) =: c2Σ, H 2 • Kπ := KZ − π∗KX. • L: a divisor associated with the tautological line bundle of P(E ). • τ := τ (E ): the unique real number such that −Kπ + τ H is nef but not • υ := υ(E ): the unique real number such that −Kπ + υH is pseudoeffective X =: dΣ, ∆(E ) =: (d∆)Σ. ample. but not big. Remark 3.3. The following holds: (i) τ ≥ υ. (ii) K 2 (iii) E is not trivial. π = π∗∆(E ) = ∆H 2. Proof. (i) If τ < υ, then −Kπ + υH is ample. This contradicts the definition of υ. (ii) By using the Chern-Wu relation L2 − π∗c1(E ).L + π∗c2(E ) = 0, π = π∗∆(E ) = ∆H 2. (1) a direct computation implies that K 2 (iii) Assume that E is trivial. Then Z = X × P1, in particular, Z is a Fano manifold. So φ is a KZ -negative extremal contraction, hence Y = P1. However it contradicts dim Y = n ≥ 2. (cid:3) We review the definition of (semi)stability of vector bundles and some results in [11]. Definition 3.4. Under the same setting as in Assumptions 3.1, let A be an ample divisor on X. Then E is said to be stable (resp. semistable) if, for any line bundle L ⊂ E , c1( L).An−1 < c1(E ).An−1 (resp. c1(L).An−1 ≤ c1(E ).An−1). 1 2 1 2 Theorem 3.5. Let (X, E ) be as in Assumptions 3.1. If E is semistable, then, for an ample divisor A ∈ Pic(X), we have ∆(E ).An−2 ≤ 0. Proof. This follows from the Bogomolov inequality and the Mehta-Ramanathan the- orem. (cid:3) Theorem 3.6 ([11, Theorem 2.3, Proposition 3.5, Remark 3.6]). Let (X, E ) be as in Assumptions 3.1. Then the following holds: (i) τ ≥ 0, and the equality holds if and only if E ∼= O ⊕2 X . (ii) If E is not semistable, then υ ≤ 0, and the equality holds if and only if E is strictly semistable. 6 KIWAMU WATANABE Thanks to Proposition 2.3, the same argument as in [11, Proposition 4.12] can be applied to our case. In particular, we obtain the following Proposition 3.7 and 3.8. For the readers convenience, we recall their argument. Proposition 3.7 (cf. tions 3.1, the following holds: [11, Proposition 4.12]). Under the setting as in Assump- (i) τ = υ = iX − 2 (ii) E is stable. µ ∈ Q>0, and Proof. (i) Since ρZ = 2, the Kleiman-Mori cone of Z is spanned by [f ] and [f ′]. Furthermore, we have −KZ .f = −KZ.f ′ = 2. By Kleiman's criterion for ampleness, this implies that −KZ is ample, that is, Z is a Fano manifold. So the nef cone of Z is a rational polyhedral cone. This implies that τ is a rational number. It follows from Kawamata-Shokurov base point free theorem that −Kπ + CH is semiample. Then it turns out that φ is defined by the linear system m(−Kπ + τ H) if m is sufficiently large and divisible. This implies that (−Kπ + τ H).f ′ = 0. Thus we see that τ = iX − 2 µ . Furthermore, since φ is a morphism of relative dimension 1, −Kπ + τ H is nef but not big. This means that τ ≤ υ. By combining Remark 3.3 (i), we get τ = υ. If τ = 0, then E is trivial by Theorem 3.6 (i). However it contradicts Remark 3.3 (iii). (ii) Since we have τ > 0, Theorem 3.6 (iii) concludes that E is stable. (cid:3) Proposition 3.8 (cf. tions 3.1, the following holds: [11, Proposition 4.4]). Under the setting as in Assump- (i) ∆ < 0, (ii) √−∆ = τ tan( π (iii) n = 2, 3 or 5. n+1 ), and Proof. (i) By Proposition 3.7, E is stable. Then ∆ ≤ 0 by Theorem 3.5. Again by Proposition 3.7, −Kπ +τ H is nef but not big. So we have (−Kπ +τ H)n+1 = 0. Since K 2 π = ∆H 2 (see Remark 3.3), H n+1 = 0 and −Kπ.H n > 0, (−Kπ + τ H)n+1 = 0 is equivalent to (2) (cid:18)n + 1 i (cid:19)τ n+1−i∆ n+1Xi=0 i≡1(2) i−1 2 = 0. If ∆ = 0, then τ n = 0 by (2). Proposition 3.7 (i). As a consequence, we have ∆ < 0. (ii) From the above equality (2), we obtain It means that τ = 0. However this contradicts (τ + √∆)n+1 − (τ − √∆)n+1 = 0. We denote the argument of the complex number (τ +√∆)n+1 by arg(cid:16)τ + √∆(cid:17)n+1 [0, 2π). Then (3) is equivalent to ∈ (3) (4) Since we have ∆ < 0 by (i), (4) implies π n + 1 . arg(cid:16)τ + √∆(cid:17) = 0 or √−∆ = τ tan(cid:18) π n + 1(cid:19) . P1-BUNDLES OVER PROJECTIVE MANIFOLD OF PICARD NUMBER ONE 7 (iii) From (ii), we obtain n + 1(cid:19) = −∆ tan2(cid:18) π The algebraic degree of tan(cid:16) π n+1(cid:17) over Q is known (see [12, pp. 33-41] and [2, Proposition 2]). Then we see that n = 2, 3 or 5. τ 2 ∈ Q. On the other hand, we give another description of ∆(E ) via a computation of the total Chern class c(π∗E ). First, we prepare the following lemma. (cid:3) Lemma 3.9. Under the setting as in Assumptions 3.1, let σ denote the restriction of π to f ′ ∼= P1. If σ∗E ∼= OP1(a) ⊕ ØP1(b) (a ≥ b), then we have (cid:19) . (a, b) =(cid:18)−1 + (c1 − iX )µ (c1 + iX )µ , 1 + 2 2 Proof. Let us consider a P1-bundle σ∗Z ∼= P(OP1(a) ⊕ ØP1(b)) over f ′ ∼= P1. Then It implies that σ∗f ′ ∼= P(OP1(b)). Hence σ∗f ′ is an exceptional curve on σ∗Z. b = L.f ′ = 1 + (c1−iX )µ . On the other hand, we have a + b = c1µ. Thus a = −1 + (c1+iX )µ (cid:3) Lemma 3.10. Under the setting as in Assumptions 3.1, the total Chern class c(π∗E ) is given by 2 2 . c(π∗E ) = 1 + 1 µ (a + b)H +(cid:18) ab µ2 H 2 + a − b µµ′ HH ′ − 1 µ′2 H ′2(cid:19) . Proof. Let P be the kernel of π∗E → L. Then we have an exact sequence (5) 0 → P → π∗E → L → 0. In general, any saturated subsheaf of a locally-free sheaf is again locally-free. So P is a line bundle. Since Lf ∼= OP1(1) and Lf ′ ∼= OP1(b), we see that Pf = ker(π∗E f → OP1(1)) ∼= OP1(−1) and Pf ′ = ker(π∗E f ′ → OP1(b)) ∼= OP1(a). Remark that A1(Z)Q = hH, H ′iQ. Hence we obtain 1 µ′ H ′, 1 µ′ H ′, c(P ) = 1 + c(L) = 1 + b µ a µ H − H + and c(π∗E ) = c(L) · c(P ) =(cid:18)1 + (a + b)H +(cid:18) ab = 1 + 1 µ H + b µ µ2 H 2 + 1 µ′ H ′(cid:19) ·(cid:18)1 + a − b µµ′ HH ′ − a µ 1 H − 1 µ′ H ′(cid:19) µ′2 H ′2(cid:19) . By using Lemmas 3.10 and 3.12 below, the equality (5) will be rewritten in more simple form in Proposition 3.13. (cid:3) 8 KIWAMU WATANABE Lemma 3.11. Under the setting as in Assumptions 3.1, let ν be the restriction of φ to f ∼= P1 and ζ a projection ν ∗Z → Z. If Nν ∗f /ν ∗Z ∼= OP1(−e), then e > 0 and we have (ζ ∗H)2 = µe µ′ (ζ ∗Hζ ∗H ′). Proof. We consider a P1-bundle ψ : ν ∗Z → f ∼= P1. Then ν ∗f is an exceptional curve on ν ∗Z. From Nν ∗f /ν ∗Z ∼= OP1(−e), we obtain e > 0. Furthermore, we see that ν ∗Z ∼= P(OP1⊕OP1(−e)) and ν ∗f ∼= P(OP1(−e)). Let M be the tautological line bundle of ν ∗Z ∼= P(OP1 ⊕ OP1(−e)) and Q the kernel of ψ∗ (OP1 ⊕ OP1(−e)) → M . Then we have an exact sequence 0 → Q → ψ∗ (OP1 ⊕ OP1(−e)) → M → 0. Then we see that Mν ∗f ∼= OP1(−e) and Mν ∗f ′ ∼= OP1(1). These imply that Qν ∗f = ker (ψ∗ (OP1 ⊕ OP1(−e))ν ∗f → OP1(−e)) ∼= OP1, and Qν ∗f ′ = ker(cid:0)ψ∗ (OP1 ⊕ OP1(−e))ν ∗f ′ → OP1(1)(cid:1) ∼= OP1(−1). Remark that A1(ν ∗Z)Q = hν ∗H, ν ∗H ′iQ. Hence we obtain e µ′ ζ ∗H ′, c(M ) = 1 + 1 µ 1 µ ζ ∗H − ζ ∗H. c(Q) = 1 − and c(ψ∗ (OP1 ⊕ OP1(−e))) = c(M ) · c(Q) =(cid:18)1 + µ′ ζ ∗H ′ +(cid:18)− = 1 − e 1 µ ζ ∗H − 1 µ2 (ζ ∗H)2 + e µ′ ζ ∗H ′(cid:19) ·(cid:18)1 − µµ′ ζ ∗Hζ ∗H ′(cid:19) . e 1 µ ζ ∗H(cid:19) Furthermore, we obtain 1 µ2 (ζ ∗H)2 + − As a consequence, we get e µµ′ ζ ∗Hζ ∗H ′ = c2 (ψ∗ (OP1 ⊕ OP1(−e))) = ψ∗ (c2 (OP1 ⊕ OP1(−e))) = 0. (ζ ∗H)2 = µe µ′ (ζ ∗Hζ ∗H ′) as desired. (cid:3) Lemma 3.12. Under the setting as in Assumptions 3.1, we have a − b eµ2 H 2 = a − b µµ′ HH ′ − 1 µ′2 H ′2 ∈ N 2(Z)Q. Proof. In Lemma 3.10, we have seen that π∗ (c2 (E )) = c2(π∗E ) = ab µ′ 2 H ′2. Since we have N 2(X)Q ∼= Q, there exists g ∈ Q such that 1 µ2 H 2 + a−b µµ′ HH ′− (6) a − b µµ′ HH ′ − Pulling back to ν ∗Z by ζ, we obtain gH 2 + ζ ∗(gH 2 + a − b µµ′ HH ′) = ζ ∗(cid:18)gH 2 + 1 µ′2 H ′2 = 0 ∈ N 2(Z)Q. a − b µµ′ HH ′ − 1 µ′2 H ′2(cid:19) = 0 ∈ N 2(ν ∗Z)Q. P1-BUNDLES OVER PROJECTIVE MANIFOLD OF PICARD NUMBER ONE 9 (cid:18) gµe By Lemma 3.11, (ζ ∗H)2 = µe µ′ (ζ ∗Hζ ∗H ′). It turns out that a − b µµ′ (cid:19) ζ ∗Hζ ∗H ′ = 0 ∈ N 2(ν ∗Z)Q. µ′ + Hence we have g = b−a µ2e . Substituting this in the equation (6), we obtain a − b eµ2 H 2 = µ′2 H ′2 ∈ N 2(Z)Q. a − b µµ′ HH ′ − 1 By combining Lemmas 3.10 and 3.12, we get the following: (cid:3) Proposition 3.13. Under the setting as in Assumptions 3.1, the total Chern class c(π∗E ) is given by c(π∗E ) = 1 + 1 µ (a + b)H +(cid:18) ab µ2 + a − b eµ2 (cid:19) H 2 ∈ 1 ⊕ N 1(Z) ⊕ N 2(Z)Q. Proposition 3.14. Under the setting as in Assumptions 3.1, e is defined by Nν ∗f /ν ∗Z ∼= OP1(−e) as in Lemma 3.11. Then we have ∆ = τ 2 − 4τ eµ . Proof. From the definition of ∆ and Proposition 3.13, ∆H 2 = c1(π∗E )2 − 4c2(π∗E ) − 4(cid:18) ab (a + b)H(cid:19)2 = (cid:18) 1 µ µ2 + a − b eµ2 (cid:19) H 2 = a − b µ2e (e (a − b) − 4) H 2. By Lemma 3.9, a − b = iXµ − 2 = τ µ. Thus, we obtain ∆ = τ 2 − 4τ eµ as desired. (cid:3) 4. Proof of Theorem 1.1 In this section, we prove Theorem 1.1. It is sufficient to work under the same setting as in Assumptions 3.1. Then it follows from Proposition 3.8 that n = 2, 3 or 5. Theorem 4.1. With the same setting as in Assumptions 3.1, if n = 2, then (X, E ) is isomorphic to (P2, TP2). Proof. Since a Fano surface of ρ = 1 is isomorphic to P2, we have X ∼= P1 and iX = 3. By virtue of Proposition 3.8 (ii) and Proposition 3.14, we have ∆ = −3τ 2 and ∆ = τ 2− 4τ eµ . Recall that τ > 0 by Proposition 3.7 (i). Thus (3µ−2)e = τ µe = 1. Hence (iX , µ, e) = (3, 1, 1), τ = 1 and ∆ = −3. By Lemma 3.9, (c1 +iX)µ is divisible by 2. This implies that c1 = −1. Here we take a point on X ∼= P2 as a base Σ of N 2(X)Q. Then d = 1. Since ∆ = −3 and c1 = −1, we see that c2 = 1. Hence E is a rank 2 stable vector bundle over P2 with (c1, c2) = (−1, 1). Then E is isomorphic to TP2 by [6]. (cid:3) Lemma 4.2. With the same setting as in Assumptions 3.1, if n = 3, then (iX , µ, e) = (4, 1, 1), (3, 1, 2), (2, 2, 1), (1, 3, 2) or (1, 4, 1). Proof. By virtue of Proposition 3.8 (ii) and Proposition 3.14, we have ∆ = −τ 2 and ∆ = τ 2 − 4τ eµ . Recall that τ > 0 by Proposition 3.7 (i). Thus (iX µ − 2)e = τ µe = 2. This implies that (iX , µ, e) = (4, 1, 1), (3, 1, 2), (2, 2, 1), (1, 3, 2) or (1, 4, 1). (cid:3) 10 KIWAMU WATANABE Lemma 4.3. Under the same setting as in Lemma 4.2, if (iX , µ, e) = (2, 2, 1), then c1 = 0. Proof. In this case, X is a del Pezzo 3-fold of ρ = 1. So H 4(X, Z) is generated by a line l on X. Here a line means a rational curve with HX.l = 1. We take l as a base Σ of N 2(X)Q. Then c2 is an integer. Now assume the contrary of our claim, that is, c1 = −1. Then Riemann-Roch theorem tells us that c2 is even. On the other hand, we see that ∆ = −1. Thus, we obtain dX = d = 2c2. From the classification of del Pezzo 3-fold of ρ = 1 [8], it follows that dX ≤ 5. It turns out that (dX , c2) = (4, 2). Again, according to [8], X is a complete intersection of two quadric 4-folds in P5. Consider a morphism π ◦ ζ : ν ∗Z → X. Since ν ∗Z ∼= P(OP1 ⊕ OP1(−1)), π ◦ ζ factors through g : P2 → X. This can be obtained by taking the Stein factorization of π◦ζ = g◦h, where h : ν ∗Z → P2 is a blow-up at a point o ∈ P2. Remark that h sends every fiber of ν ∗Z → f ∼= P1 to a line through o ∈ P2. Hence, for a line lo through o ∈ P2, we have g∗HX.lo = µ = 2. This implies that g∗OX (HX) ∼= OP2(2). Let S denote the image of g. Then the degree of S ⊂ P5 satisfies that deg(g) deg(S) = 4. On the other hand, S is a member of the linear system OX (sHX) for some s > 0. So we have deg(S) = 4s. Hence we see that (s, deg(g)) = (1, 1). Then, it is easy to see that S is smooth, that is, S ∼= P2. However this contradicts the adjunction formula. (cid:3) By the same way as in Lemma 4.2, we can prove the following: Lemma 4.4. With the same setting as in Assumptions 3.1, if n = 5, then (iX , µ, e) = (5, 1, 1), (3, 1, 3), (1, 5, 1), or (1, 3, 3). Now we prove the remaining part of Theorem 1.1 (I). Proposition 4.5. Under the same setting as in Assumptions 3.1, there exists a rank 2 vector bundle E ′ on Y such that Z = PY (E ′). Proof. If n = 2, this follows from Theorem 4.1. Thus, we deal with the cases where n = 3 and 5. From the above lemmas, it follows that (n, iX , µ, e) = (3, 4, 1, 1), (3, 3, 1, 2), (3, 2, 2, 1), (3, 1, 3, 2), (3, 1, 4, 1), (5, 5, 1, 1), (5, 3, 1, 3), (5, 1, 5, 1) or (5, 1, 3, 3). Recall that L.f ′ = 1 + (c1−iX )µ It is enough to find a line bundle V on Z which satisfies V.f ′ = 1. Indeed, E ′ := φ∗OZ(V ) satisfies the property desired. Hence it is sufficient to deal with the case where µ := H.f ′ 6= 1, that is, (n, iX , µ, e) = (3, 2, 2, 1), (3, 1, 3, 2), (3, 1, 4, 1), (5, 1, 5, 1), and (5, 1, 3, 3). due to Lemma 3.9. If (n, iX , µ, e) = (3, 2, 2, 1), then, by Lemma 4.3, we have c1 = 0. This means L.f ′ = −1. Hence V := H ⊗ L satisfies V.f ′ = 1. If (n, iX , µ, e) = (3, 1, 3, 2), (5, 1, 5, 1) or (5, 1, 3, 3), then we see that L.f ′ = −2,−4, −2, respectively. It turns out that V := H⊗L satisfies V.f ′ = 1. If (n, iX , µ, e) = (3, 1, 4, 1), then L.f ′ = −1 or −3. Hence we can take H ⊗ L⊗3 or H ⊗ L as V . 2 According to this proposition, we see that Z admits double P1-bundle structures π : Z → X and φ : Z → Y . By symmetry of X and Y , all the results on X as above also hold for Y . By the same way as in Notation 3.2, we define rational (cid:3) P1-BUNDLES OVER PROJECTIVE MANIFOLD OF PICARD NUMBER ONE 11 1, c′ 2, d′ and ∆′ for Y and E ′. Here E ′ may be normalized. Moreover, numbers c′ Kφ and L′ stand for the relative canonical divisor and a divisor associated with the tautological line bundle of P(E ′), respectively. Then we define τ ′ := τ (E ′) and υ′ := υ(E ′) as in Notation 3.2. Applying the argument as in [11], we complete the proof of Theorem 1.1 as follows: Theorem 4.6. Under the same setting as in Assumptions 3.1 and the above, if n ≥ 3, then ((X, E ), (Y, E ′)) is isomorphic to ((P3, N ), (Q3, S )) or ((Q5, C ), (K(G2), Q)) up to changing the pairs (X, E ) and (Y, E ′). Proof. As we have seen in Proposition 3.7, τ = iX − 2 obtain the following table: µ and τ ′ = iY − 2 µ′ . Then we H H ′ 0 µ µ′ 0 f f ′ L 1 1 + (c1−iX )µ 2 L′ 1 + (c′ 1−iY )µ′ 1 2 This table represents intersection numbers of divisors in the first row and f or f ′. For example, H.f = 0 and H.f ′ = µ etc. Applying this table, we obtain ( H ′ = − µ′ L′ = {− µ′ 2 (c1 − τ )H + µ′L 4 (c1 − τ )(c′ 1 − τ ′) + 1 µ}H + µ′ 2 (c′ 1 − τ ′)L. Since {H, L} and {H ′, L′} are Z-bases of Pic(P(E )), the determinant of the matrix of base change is equal to 1 or −1. This implies that µ = µ′. Hence we can write H ′ = µ 2 (−Kπ + τ H). Furthermore, we get )n (−Kπ + τ H)nH/µ = ( dY dX µ 2 −KπH n/2 = ( µ 2 )n−1 im((τ + √∆)n) . √−∆ Since we have √−∆ = τ tan( π n+1 ), (8) is equivalent to dY dX = ( τ µ 2cos(π/n + 1) )n−1. (7) (8) (9) (10) By symmetry of X and Y , we get a similar equation dX dY = ( τ ′µ′ 2cos(π/n + 1) )n−1. These equations imply 4cos2(π/n + 1) Since τ, τ ′ > 0 and µ = µ′ > 0, (9) provides τ µτ ′µ′ ( )n−1 = 1. (iX µ − 2)(iY µ − 2) = τ τ ′µ2 =(cid:26) 2 (n = 3) 3 (n = 5) We may assume that iX ≥ iY . From (10), we have (iX , iY , µ) = (4, 3, 1) provided n = 3. Hence we see that X ∼= P3. Here we take a line on X ∼= P3 as a base Σ of N 2(X)Q. Then d = 1. On the other hand, if n = 5, then we have (iX , iY , µ) = (5, 3, 1). Hence we obtain that X ∼= Q5. Here H 4(X, Q5) ∼= Z and we take its positive generator as a base Σ of N 2(X)Q. Then d = 1. In both cases, easy calculations imply the following table: 12 KIWAMU WATANABE n iX d µ τ ∆ c1 1 1 2 −4 0 3 1 1 3 −3 −1 5 4 5 c2 1 1 Since vector bundles N and C are determined by their Chern classes among sta- ble bundles (see [13, Lemma 4.3.2] and [14]), hence (X, E ) is isomorphic to (P3, N ) or (Q5, C ). Then the structure of (Y, E ′) is well-known (for instance, see [15, Propo- sition 2.6], [11, Example 6.4] and [14, 1.3]). Consequently, Theorem 1.1 holds. (cid:3) 5. Proof of Theorem 1.3 Let M be an n-dimensional Fano manifold of ρ = 1 with nef tangent bundle. Assume that M carries a rational curve C such that −KM .C = 3. This assumption implies that n ≥ 2. Let K be a minimal rational component of M (see [10, 1.2]) and π : U → K its universal family. Denote the evaluation map by ι : U → M . Lemma 5.1 ([10, Lemma 1.2.1, Lemma 1.2.2, Corollary 1.3.1]). Under the above setting, the following holds: (i) K is a projective manifold of dimension n, (ii) every fiber of ι is isomorphic to P1, and (iii) π : U → K is a P1-bundle. Since ι is a smooth morphism whose fibers are P1, one can check that the second Betti numbers satisfy b2(U ) = b2(M ) + 1. This implies b2(K ) = 1. Furthermore, K is covered by rational curves which are images of fibers of ι, that is, a uniruled manifold. Consequently, K is a Fano manifold of ρK = 1. Applying Theorem 1.1, we obtain Theorem 1.3. References [1] E. Arrondo, I. Sols, Classification of smooth congruence of low degree, J. Reine Angew. Math. 393 (1989), 199-219. [2] J. S. Calcut, Rationality and the tangent function, preprint, available at http://www.oberlin.edu/faculty/jcalcut/tanpap.pdf. [3] F. Campana, T. Peternell, Projective manifolds whose tangent bundles are numerically effective, Math. Ann. 289 (1991), 169-187. [4] W. Fulton, Intersection Theory, Ergeb. Math. Grenzgeb. (3), Springer-Verlag, Berlin, Heidel- berg, New York, 1984. [5] R. Hartshorne, Algebraic geometry. Graduate Texts in Mathematics, No. 52. Springer-Verlag, New York-Heidelberg, 1977. [6] K. Hulek, Stable rank-2 vector bundles on P2 with c1 odd, Math. Ann. 242, 241-266, (1979). [7] J.M. Hwang, Rigidity of rational homogeneous spaces, Proceedings of ICM. 2006 Madrid, volume II, European Mathematical Society, 2006, 613-626. [8] V. A. Iskovskih, Fano 3-folds. I, II, Math. USSR Izv. 11 (1977) 485-529; 12 (1978). 496-506. [9] J. Koll´ar, Y. Miyaoka and S. Mori, Rational curves on Fano varieties, Proc. Alg. Geom. Conf. Trento, Springer Lecture Notes 1515 (1992) 100-105. [10] N. Mok, On Fano manifolds with nef tangent bundles admitting 1-dimensional varieties of minimal rational tangents, Trans. Amer. Math. Soc. 354 (2002), 2639 -2658. [11] R. Munoz, G. Occhetta, L. Sol´a Conde, On rank 2 vector bundles on Fano manifolds, arXiv:1104.1490. [12] I. Niven, Irrational Numbers, The Carus Mathematical Monographs, no. 1 1, MAA, 1956 [13] C. Okonek, M. Schneider and H. Spindler, Vector bundles over complex projective space, Progress in Math., vol. 3, Birkhauser, Boston, Basel, Stuttgart, 1980. [14] G. Ottaviani, On Cayley bundles on the five-dimensional quadric, Boll. Un. Mat. Ital. A (7) 4 (1990). P1-BUNDLES OVER PROJECTIVE MANIFOLD OF PICARD NUMBER ONE 13 [15] M. Szurek, J. A. Wi´sniewski, Fano bundles over P3 and Q3, Pacific J. Math. 141 (1990), no. 1, 197-208. Course of Mathematics, Programs in Mathematics, Electronics and Informatics, Graduate School of Science and Engineering, Saitama University. Shimo-Okubo 255, Sakura-ku Saitama-shi, 338-8570 JAPAN E-mail address: [email protected]
1608.06053
2
1608
2017-02-10T12:23:26
K-stability of smooth del Pezzo surfaces
[ "math.AG" ]
In an algebro-geometric way, we completely determine whether smooth del Pezzo surfaces are K-(semi)stable or not.
math.AG
math
K-STABILITY OF SMOOTH DEL PEZZO SURFACES JIHUN PARK AND JOONYEONG WON Abstract. In a new algebro-geometric way we completely determine whether smooth del Pezzo surfaces are K-(semi)stable or not. In the present article, all varieties are defined over an algebraically closed field k of charac- teristic 0. 1. Introduction Since entering the 21st century we have witnessed dramatic developments in the study of the Yau-Tian-Donaldson conjecture concerning the existence of Kahler-Einstein metrics on Fano manifolds and stability. The challenge to the conjecture has been highlighted by Chen, Don- aldson, Sun and Tian who have completed the proof of the following celebrated statement ( [6], [7], [8], [25]). Theorem 1.1. Let X be a smooth Fano variety defined over C. It admits a Kahler-Einstein metric if and only if the pair (X, −KX ) is K-polystable. The motivation of this article cannot be expressed in a better way than quoting the following phrase in one of the three articles by Chen, Donaldson and Sun ( [8]): "On the other hand, we should point out that as things stand at present the result is of very limited use in concrete cases, so that there is no manifold X known to us, not covered by other existence results and where we can deduce that X has a Kahler-Einstein metric. This is because it seems a very difficult matter to test K-stability by a direct study of all possible degenerations. However, we are optimistic that this situation will change in the future, with a deeper analysis of the stability condition." There are not so many results concerning K-stability of specified smooth Fano varieties, not deduced by Kahler-Einstein metrics. It seems almost infeasible to consider all possible degenerations of a given Fano manifold. Even for del Pezzo surfaces, we do not have complete description of their degenerations. There are a classification only for the projective plane and some partial classifications for the others ( [18], [19], [14], [22]). We also have a classification of del Pezzo surfaces with quotient singularities and Picard rank one that admit Q-Gorenstein smoothings ( [14]). These are however not enough to directly test K-stability of del Pezzo surfaces. Meanwhile, since asymptotic Chow-stability implies K-semistability (for instance, see [23]), an algebro-geometric proof for the K-semistability of the projective spaces can be yielded by the celebrated result of Kempf in [16] that a homogeneous rational variety embedded with a complete linear system is Chow-stable. There are a few of algebro-geometric methods known to us that can be utilized to prove K- stability in concrete cases. One of the ways is based on the α-invariant originally introduced by Tian ( [24]). The original definition of the α-invariant was given in an analytic way. However, This work has been supported by IBS-R003-D1, Institute for Basic Science in Korea. The authors are grateful to Kento Fujita and Yuji Odaka who brought the article [13] to their attention. They also thank Giulio Codogni who informed them of the article [16]. 1 2 JIHUN PARK AND JOONYEONG WON there is an algebro-geometric way to define the α-invariant over an arbitrary field of characteristic zero. Definition 1.2. Let X be a Fano orbifold defined over k. The global log canonical threshold of X is defined by the number α(X) = sup(cid:26)λ ∈ Q (cid:12)(cid:12)(cid:12)(cid:12) the pair (X, λD) is log canonical for every effective Q-divisor D numerically equivalent to −KX . (cid:27) . It has been verified that for a Fano orbifold defined over C its α-invariant coincides with the global log canonical threshold ( [5]). For this reason, the same name α-invariant and the same notation α(X) will be used for the global log canonical threshold of the Fano orbifold defined over k in the present article. The original purpose of the α-invariant is to show the existence of Kahler-Einstein metrics on given Fano manifolds ( [24]). Fujita, Odaka and Sano however reinterpret the α-invariant as a sufficient condition for a Q-Fano variety to be K-stable. Theorem 1.3 ( [20], [11]). Let X be a Q-Fano variety. Either if (1.4) α(X) > dim(X) dim(X) + 1 or if α(X) = dim(X) dim(X)+1 and X is smooth, then the pair (X, −KX ) is K-stable. This enables us to have a detour studying K-stability for some specific Q-Fano varieties. Even though the method is only a one-side implication and the α-invariants are not easy to compute at all, this is the only pragmatic way, known to us so far, to verify K-stability in concrete cases. It is an indisputable expectation that computing the α-invariant of a given Fano variety should be much more doable than directly investigating its degenerations. For instance, Cheltsov has computed the exact values of the α-invariants of all the smooth del Pezzo surfaces. For the purpose of K-stability, we summarize his computation in the following way. Theorem 1.5 ( [1]). Let S be a smooth del Pezzo surface of degree d. • If d ≥ 5, then α(S) < 2 3 . • If d ≤ 4, then α(S) ≥ 2 3 . With this estimation, the method of Fujita-Odaka-Sano immediately implies that a smooth del Pezzo surface of degree at most 4 is K-stable with respect to its anticanonical polarisation. Even though we are not able to completely determine the K-stability of all the smooth del Pezzo surfaces with their α-invariants, this result demonstrates that the α-invariant is a very practical tool to test K-stability for given Fano varieties. The following higher dimensional smooth Fano varieties are also instructive examples to which we can apply the method of Fujita- Odaka-Sano: • a double cover of Pn ramified along a smooth hypersurface of degree 2n, n ≥ 3; • a smooth hypersurface of degree n in Pn, n ≥ 4. All the smooth Fano varieties in these families satisfy the condition of Theorem 1.3 ( [3], [2]). Therefore, they are K-stable, and hence the Fano manifolds defined over C in these families admit Kahler-Einstein metrics. In particular, a smooth hypersurface of degree n in Pn with generalized Eckardt points ( [2]) is an example of a Kahler-Einstein Fano manifold whose Kahler-Einstein metric is verified to exist only by proving its K-stability ( [2], [4], [11]). Recently Fujita and Odaka provided a new algebro-geometric way to test K-(semi)stability of Fano varieties. To introduce their method, let X be a Q-factorial variety with at worst log K-STABILITY OF SMOOTH DEL PEZZO SURFACES 3 canonical singularities, Z ⊂ X a closed subvariety and D an effective Q-divisor on X. The log canonical threshold of D along Z is the number Because log canonicity is a local property, we see that cZ (X, D) = supnc(cid:12)(cid:12)(cid:12) the pair (X, cD) is log canonical along Z.o . cZ (X, D) = inf p∈Z {cp(X, D)} . If X = An and D = (f = 0), where f is a polynomial defined over An, then we also use the notation c0(f ) for the log canonical threshold of D at the origin. Definition 1.6 ( [13]). Let X be a Q-Fano variety and let m be a positive integer such that the plurianticanonical linear system − mKX is non-empty. Set ℓm = h0(X, OX (−mKX)). For a section s in H0(X, OX (−mKX )), we denote the effective divisor of the section s by D(s). If ℓm sections s1, . . . , sℓm form a basis of the space H0(X, OX (−mKX)), then the anticanonical Q-divisor is said to be of m-basis type. For a positive integer m, we set D := 1 ℓm 1 m D(si) ℓmXi=1 δm(X) = inf D: m-basis type cX(X, D). We set δm(X) = 0 if − mKX is empty. The δ-invariant of X is defined by the number δ(X) = lim sup m δm(X). Using the δ-invariant, Fujita and Odaka set up a conjectural criterion for K-(semi)stability. They then proved that the δ-invariant gives a sufficient condition for K-(semi)stability. Conjecture 1.7 ( [13]). A Q-Fano variety X is K-stable (resp. K-semistable) with respect to −KX if and only if δ(X) > 1 (resp. ≥ 1). Note that Conjecture 1.7 is true if Berman-Gibbs stability ( [12]) is equivalent to K-stability. Theorem 1.8 ( [13]). Let X be a Q-Fano variety. If δ(X) > 1 (resp. ≥ 1), then (X, −KX ) is K-stable (resp. K-semistable). Proof. See [13]. (cid:3) In the present article we utilize this theorem to verify the K-(semi)stability of smooth del Pezzo surfaces, i.e., we prove the following. Main Theorem. Let S be a smooth del Pezzo surface of degree d. • If d ≤ 5, then δ(S) ≥ 15 • If S ∼= P2, P1 × P1 or the del Pezzo surface of degree 6, then δ(S) = 1. • If S ∼= F1 or the del Pezzo surface of degree 7, then δ(S) < 1. 14 > 1. Through Theorem 1.8, these estimations of the δ-invariants immediately yield the following. Corollary 1.9. Let S be a smooth del Pezzo surface of degree d. Then the pair (S, −KS ) is (cid:26) K-semistable if S ∼= P2, P1 × P1 or the del Pezzo surface of degree 6. K-stable if d ≤ 5. Since the sufficient condition for K-stability given by the α-invariant in (1.4) involves the dimensions of Fano varieties, the following theorem shows that the α-invariant itself is paralyzed in testing K-stability of products of Fano varieties. 4 JIHUN PARK AND JOONYEONG WON Theorem 1.10. Let X and Y be smooth Fano varieties. Then α(X × Y ) = min {α(X), α(Y )} . Proof. See [5, Lemma 2.29]. (cid:3) We strongly believe that the same formula holds for the δ-invariant. Conjecture 1.11. Let X and Y be smooth Fano varieties. Then δ(X × Y ) = min {δ(X), δ(Y )} . Remark 1.12. One can easily verify the conjecture in case when either X or Y is 1-dimensional, i.e., P1. From this formula we see that the δ-invariant, unlike the α-invariant, keeps its ability to test K-stability in products of Fano varieties. For instance, Main Theorem implies the following. Corollary 1.13. Assume that Conjecture 1.11 holds. Then a product of smooth del Pezzo surfaces of degrees ≤ 5 is K-stable. A product of K-semistable smooth del Pezzo surfaces is K-semistable. It is certain that the δ-invariant produces a new method to verify existence of Kahler-Einstein metrics on Fano varieties through Theorem 1.1. It is no exaggeration to say that this new method is simpler and more coherent than the method by the α-invariant. Corollary 1.14. A smooth del Pezzo surface of degree ≤ 5 defined over C admits a Kahler- Einstein metric. Since Conjecture 1.7 has not been verified completely, at this moment we cannot say that the last statement of Main Theorem implies that the Hirzebruch surface F1 and the del Pezzo surface of degree 7 are not K-semistable. Yet it should be remarked here that algebro-geometric ways to prove their non-K-semistability are already known. Proposition 1.15. The del Pezzo surfaces of degree 7 and the Hirzebruch surface F1 are not K-semistable. Proof. For instance, see [10, Examples 6.5 and 6.5], [15, Proposition 3.1], [21] (cid:3) Since the del Pezzo surfaces in the second statement of Main Theorem are all toric varieties, they cannot be K-stable. Therefore, Corollary 1.9 and Proposition 1.15 verify that Conjecture 1.7 holds good for 2-dimensional Fano manifolds. 2. Preliminaries Let f be a polynomial over the field k in variables z1, . . . , zn. Assign positive integral weights w(zi) to the variables zi. Let w(f ) be the weighted multiplicity of f at the origin (= the lowest weight of the monomials occurring in f ) and let fw denote the weighted homogeneous leading term of f (= the term of the monomials in f with the weighted multiplicity of f ). Let g be a polynomial over the field k in z2, . . . , zn and set h(z1, . . . , zn) = f (z1 + g(z2, . . . , zn), z2, . . . , zn). If z1 + g(z2, . . . , zn) is weighted homogeneous with respect to the given weights w(z1), . . . , w(zn), it is clear that hw(z1, . . . , zn) = fw(z1 + g(z2, . . . , zn), z2, . . . , zn). Let f1, . . . , fℓ be polynomials over the field k in z1, . . . , zn. With respect to the given weights w(z1), . . . , w(zn), we easily see that ℓYi=1 fi!w = (fi)w, w ℓYi=1 ℓYi=1 fi! = w(fi). ℓXi=1 K-STABILITY OF SMOOTH DEL PEZZO SURFACES 5 Let m ⊂ k[z1, . . . , zn] be the maximal ideal of the origin in An. Let f1, . . . , fℓ be poly- nomials over the field k in z1, . . . , zn that induce a basis for the d-jet space at the origin, i.e., k[z1, . . . , zn]/md+1, where d is a positive integer and ℓ = dimk k[z1, . . . , zn]/md+1. The k-linear map of k[z1, . . . , zn]/md+1 induced by the coordinate change z1 − g(z2, . . . , zn) 7→ z1 and zi for i ≥ 2 is an automorphism. Therefore, the new polynomials f1(z1 + g(z2, . . . , zn), z2, . . . , zn), . . . , fℓ(z1 + g(z2, . . . , zn), z2, . . . , zn) also induce a basis for the d-jet space. 7→ zi Proposition 2.1. Let f be a polynomial over An. Assign integral weights w(zi) to the variables and let w(f ) be the weighted multiplicity of f . Then • c0(fw) ≤ c0(f ) ≤ P w(zi) w(f ) • If . (cid:18)An,P w(zi) w(f ) · (fw = 0)(cid:19) is log canonical outside the origin, then c0(f ) = P w(zi) w(f ) . Proof. See [17, Propositions 8.13 and 8.14]. (cid:3) 3. K-semistable del Pezzo surfaces I Fix a positive integer m and put ℓm = h0(P2, OP2(−mKP2)). Let [x : y : z] be a homogeneous coordinate for P2. Since the ℓm monomials of degree 3m in x, y, z form a basis of the space H0(P2, OP2(−mKP2)), the effective divisor defined by the equation xyz = 0 is an anticanonical divisor of m-basis type. This shows that for each m. Therefore, δ(P2) ≤ 1. Theorem 3.1. The δ-invariant of P2 is 1. δm(P2) ≤ 1 Proof. Let {s1, . . . , sℓm} be a basis of H0(P2, OP2(−mKP2)). We denote the effective divisor of the section si by Di and set D := Di. ℓmXi=1 It is enough to show that for an arbitrary point p in P2 cp(P2, D) ≥ 1 mℓm . By a linear coordinate change, we may assume that p = [0 : 0 : 1]. Consider the ℓm monomials of degree 3m in the variables x, y, z. Putting z = 1, we obtain the ℓm monomials of degrees at most 3m 1, x, y, x2, xy, xy2, . . . , xk, xk−1y, . . . , xyk−1, yk, . . . , x3m, x3m−1y, . . . , xy3m−1, y3m, where they are written in lexicographic order. In this order, denote the i-th monomial by xi for each i = 1, . . . , ℓm. We may consider x and y as local coordinates around the point p. Then each Di is defined around the point p by a polynomial fi of degree at most 3m in the variables x, y. The divisor D is defined by f :=Q fi in an affine neighborhood of p. Since the sections s1, . . . , sℓm form a basis for H0(P2, OP2(−mKP2)) and they induce the polynomials f1, . . . , fℓm, respectively, we may assume that fi contains the monomial xi for each i = 1, . . . , ℓm. 6 JIHUN PARK AND JOONYEONG WON Now we consider the Newton polygon of the polynomial f in R2, where we use coordinate functions (s, t) for R2. Claim 1. The Newton polygon of the polynomial f contains the point (mℓm, mℓm) corre- sponding to the monomial xmℓmymℓm. Since each fi contains the monomial xi, we have w(fi) ≤ w(xi) with respect to given weights w(x), w(y), and hence w(f ) ≤ w(Q xi) = w(xmℓm ymℓm). This proves the claim. If the line s = t intersects the Newton polygon of f at one of its vertices, mℓm . Claim 2. then c0(f ) ≥ 1 In such a case, we can take weights w(x), w(y) such that fw is of the form ǫxaya, where ǫ is a non-zero constant. Claim 1 implies that a ≤ mℓm. We then obtain from Proposition 2.1. c0(f ) ≥ c0(fw) = 1 a ≥ 1 mℓm We may therefore assume that any vertex of the Newton polygon of f does not lie on the line s = t. Step A. Let Λ be the edge of the Newton polygon of f that intersects the line s = t. Let w(x), w(y) be the weights such that the monomials of fw are plotted on the edge Λ. If Λ is either vertical ( fw = xag(y)) or horizontal (fw = yah(x)), where g and h are polyno- mials of multiplicity at most a − 1 at the origin, then Claim 1 implies a ≤ mℓm. It then follows from Proposition 2.1 that c0(f ) ≥ 1 mℓm . Therefore, we may assume that Λ is neither vertical nor horizontal. The slope of the edge Λ is equal to − w(x) w(y) . Step B. We write an irreducible decomposition of fw as fw = ǫxayb(xα1 + g1(x, y))c1 · · · (xαr + gr(x, y))cr , where ǫ is a non-zero constant and gi(x, y) is an irreducible weighted homogeneous polynomial of degree w(xαi) such that it does not contain the monomial xαi. Note that a, b ≤ mℓm. Let c = max{ci}. We may assume that c1 = c. For the convenience, we set α = α1. Since g1(x, y) is irreducible, it must possess the monomial yβ for some positive integer β. Claim 3. If c ≤ mℓm, then c0(f ) ≥ 1 mℓm . It immediately follows from Proposition 2.1 that c0(fw) = min(cid:26) 1 1 b mℓ , so that c0(fw) ≥ 1 mℓ . Therefore, c0(f ) ≥ c0(fw) ≥ 1 mℓ . w(fw) (cid:27) . w(x) + w(y) a 1 c , , , By Claim 1, w(x)+w(y) w(fw) ≥ 1 Step C. Suppose that c > mℓm. Since w((xα + yβ)c) ≤ w(xmℓm ymℓm) by Claim 1, if α, β ≥ 2, then we immediately obtain c ≤ mℓm. Therefore, either α = 1 or β = 1. By exchanging coordinates if necessary, we may assume that α = 1. Note that α = 1 implies that w(x) ≥ w(y) and w(x) w(y) is the integral number β. Therefore, fw = ǫxayb(x + A1yβ)c(xα2 + g2(x, y))c2 · · · (xαr + gr(x, y))cr , K-STABILITY OF SMOOTH DEL PEZZO SURFACES 7 where A1 is a non-zero constant. The weighted leading term fw contains the monomial x(a+c+Pr i=2 αici ≥ c > mℓm. We now apply a change of coordinate x + A1yβ 7→ x to the polynomials fi and f . Set i=2 αici)yb and a + c +Pr for each i and f (1) i (x, y) := fi(x − A1yβ, y) f (1)(x, y) := f (x − A1yβ, y). Then f (1) Since f (1) w (x, y) = fw(x − A1yβ, y) and f (1) =Q f (1) i . 1 , . . . , f (1) ℓm form a basis for the (3m)-jet space k[x, y]/(x, y)3m+1, we may again assume contains the monomial xi for each i. The Newton polygon of the polynomial f (1) again that f (1) contains the point corresponding to xmℓmymℓm. i Now we go back to Step A with f (1) instead of f , i.e., let Λ(1) be the edge of the Newton polygon of f (1) that intersects the line s = t. We also find weights w(1)(x), w(1)(y) with respect to w(1) are plotted on the edge Λ(1). The slope of the edge Λ(1) is − w(1)(x) which the monomials of f (1) . w(1)(y) Let L be the line in R2 determined by the edge Λ. We observe that there is no monomial in f (1) plotted under the line L and that there is no mono- mial in f (1) plotted on the line L with s < c. (*) But f (1) possesses the monomial xa+c+Pr strictly steeper than Λ, i.e., − w(1)(x) i=2 αiciyb that is plotted on L. This shows that Λ(1) is w(1)(y) < − w(x) w(y) . As before, we write w(1) = ǫ(1)xa(1) f (1) yb(1) (xα (1) 1 + g(1) 1 (x, y))c (1) 1 · · · (x (1) α r(1) + g(1) r(1)(x, y)) c (1) r(1) , where ǫ(1) is a non-zero constant and g(1) mial of degree w(1)(xα (1) i i ) such that it does not contain the monomial xα (1) i . (x, y) is an irreducible weighted homogeneous polyno- Let c(1) = maxnc(1) i o . Again we assume that c(1) = c(1) for some positive integer β(1). must contain the monomial yβ(1) 1 . The irreducible polynomial g(1) 1 If c(1) ≤ mℓm, then the proof is done by Claim 3. If c(1) > mℓm, then we follow Step C. is the integral w(y) = β(1) − β ≥ 1. Now we go back to Step A with the new We here remark that α(1) must be 1 because w(1)(x) w(1)(y) number β(1) and w(1)(x) w(1)(y) coordinate-changed polynomials f (2) w(y) ≥ 1. Note w(1)(x) and f (2). > w(x) − w(x) w(1)(y) For the proof, it is enough to show that this procedure terminates in a finite number of loops. The slope of Λ(i) is bounded below by −mℓm since the Newton polygon of f (i) must contain the point corresponding to xmℓmymℓm and it has the property (*) above. The termination is therefore guaranteed because the slope of Λ(i) drops by at least 1 for each loop. (cid:3) i 4. K-stable del Pezzo surfaces In this section, we prove the first statement of Main Theorem. Before we start, we should remark here that the estimations of the δ-invariants in this section are not sharp at all. Since we have only to check that the δ-invariants are strictly bigger than 1, our estimations have been made in such a way that we can reduce the amount of computation as much as possible. Let Sd be a smooth del Pezzo surface of degree d ≤ 5. Let m be a positive integer and let s1, . . . , sℓm be sections in H0(Sd, OSd(−mKSd)) that form a basis for H0(Sd, OSd(−mKSd)), 8 JIHUN PARK AND JOONYEONG WON where ℓm = dm(m+1) 2 + 1. Denote the divisor of the section si by Dm i . Set Dm := Dm i . ℓmXi=1 Since d ≤ 5, for a given point p ∈ Sd, we can choose mutually disjoint (9 − d) (−1)-curves M1, . . . , M9−d which do not pass through the point p. By contracting these (−1)-curves, we obtain a birational morphism π : Sd → P2 such that it is an isomorphism in a neighborhood of p. Set q = π(p). By a suitable coordinate change, we may assume that q = [0 : 0 : 1]. Note that i )) for i = 1, . . . , ℓm and cp(Sd, Dm) = cq(P2, π(Dm)). Denote π(Dm cp(Sd, Dm i ) by ¯Dm i i ) = cq(P2, π(Dm for each i and π(Dm) by ¯Dm. For an effective divisor C in − mKSd, π(C) is an effective divisor of degree 3m on P2 which passes through the points π(M1), . . . , π(M9−d) with multiplicities at least m. Such divisors produce an ℓm-dimensional subspace of H0(P2, OP2(−mKP2)). We denote this subspace by Lm. The effective divisors ¯Dm ℓm induce a basis for Lm. 1 , . . . , ¯Dm As before, let [x : y : z] be a homogeneous coordinate for P2. We may consider x and y as local coordinates in the affine chart U by z 6= 0. For a polynomial g(x, y) = gi(x, y) + gi+1(x, y) + . . ., where gj(x, y) is a homogeneous polynomial of degree j, we will call the lowest degree term gi(x, y) the Zariski tangent term of g(x, y). In U , the divisor ¯Dm is defined by a i polynomial fm,i. Note that fm,i! = cq(cid:18)P2, 1 mℓm ¯Dm(cid:19) = cp(cid:18)Sd, 1 mℓm Dm(cid:19) . For each 0 ≤ n ≤ 3m, we define a subspace of the k-vector space of homogeneous polynomials of degree n in the variables x and y as follows: mℓm · c0 ℓmYi=1 g(x, y) ∈ k[x, y]n (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Tn,m := there is an effective divisor C in − mKSd such that the divisor π(C) is defined in U by a polynomial whose Zariski tangent term is g(x, y). ,  where k[x, y]n denotes the space of homogeneous polynomials of degree n in x, y. Note that the space k[x, y]n is spanned by the standard basis Set Sn =(cid:8)xn, xn−1y, . . . , xyn−1, yn(cid:9) . Tm := Tn,m. 3mMn=0 A basis for Tm consisting only of homogeneous elements will be called a basis for Tm. Put ℓ = dimk(Tm). Let {t1, . . . , tℓ} be a basis of Tm. For each 1 ≤ j ≤ ℓ we take a polynomial hm,j on U originated from Lm such that its Zariski tangent term is tj. It is obvious that hm,1, . . . , hm,ℓ are linearly independent, so that ℓ ≤ ℓm. Let G be a member in Lm. It can be expressed on U by a polynomial g(x, y) = gi(x, y) + gi+1(x, y) + . . ., where gj(x, y) is a homogeneous polynomial of degree j. If i = 3m, then g(x, y) is a linear combination of hm,1, . . . , hm,ℓ. Assume that if a + 1 ≤ i ≤ 3m, then g(x, y) is a linear combination of hm,1, . . . , hm,ℓ. For i = a, we have constants α1, . . . , αℓ such that the Zariski tangent term of P αjhj(x, y) is gi(x, y). Since g(x, y) −P αjhm,j(x, y) is induced by some element in Lm, the polynomial g(x, y) is a linear combination of hm,1, . . . , hm,ℓ. Thus inductively we can conclude that every polynomial on U from Lm is a linear combination of hm,1, . . . , hm,ℓ. Consequently, the dimension of Tm is equal to the dimension of H0(Sd, OSd(−mKSd)), i.e., ℓ = ℓm. K-STABILITY OF SMOOTH DEL PEZZO SURFACES 9 Now let us explain how to estimate the log canonical threshold c0 ℓmYi=1 fm,i! . Step 0. Consider the space L1. Using a suitable coordinate change, we choose a ba- sis {t1, . . . , td+1} for the space T1 with deg(t1) ≤ . . . ≤ deg(td+1) such that t1, . . . , td are mono- mials. Indeed, in each proof, we will immediately see that this is possible whenever we need this step. Denote the monomial ti by x1,i for i = 1, . . . , d. Choose an appropriate monomial from those in td+1 and denote it by x1,d+1. Step 1. Consider the finite sets Cm :=(d+1Yi=1 m :=(d+1Yi=1 The monomialQd+1 Bs xni tni i 1,i (cid:12)(cid:12)(cid:12) ni are non-negative integers with n1 + · · · + nd+1 = m) , (cid:12)(cid:12)(cid:12) ni are non-negative integers with n1 + · · · + nd+1 = m) . 1,i is contained in the homogeneous polynomialQd+1 d (cid:1) ≤ ℓm. Set a = ℓm − b. m by b. Note that b =(cid:0)m+d i=1 tni i . Since the sets {t1, . . . , td+1} and {x1,1, . . . , x1,d+1} are linearly independent respectively, the sets Cm and Bs m are linearly independent respectively. i=1 xni Step 2. We choose linearly independent a homogeneous elements u1, . . . , ua in Tm such that Denote the number of elements in Bs the set {u1, . . . , ua} ∪ Bs m forms a basis for Tm. Lemma 4.1. There is an injective map (4.2) ιm : {u1, . . . , ua} ∪ Bs m → Sn 3m[n=0 such that (1) ιm(Qd+1 i=1 tni i ) =Qd+1 i=1 xni 1,i; (2) ιm(ui) is contained in the homogeneous polynomial ui for each i. Proof. See Appendix at the end. (cid:3) Step 3. Consider all the possibilities of the image ιm({u1, . . . , ua}) with linearly indepen- m forms a basis dent a homogeneous elements u1, . . . , ua in Tm such that the set {u1, . . . , ua} ∪ Bs for Tm. Lemma 4.3. There is an injective map (4.4) such that Im : {fm,i 1 ≤ i ≤ ℓm} → Sn 3m[n=0 (1) the monomial Im(fm,i) is contained in fm,i; (2) its image coincides with the image of ιm. Proof. See Appendix at the end. (cid:3) 10 JIHUN PARK AND JOONYEONG WON Lemma 4.3 implies that the Newton polygon of the polynomialQℓm corresponding to the monomial (4.5) xαyβ := Im(fm,i) = ℓmYi=1 aYi=1 ιm(ui) · Yx∈Cm x. i=1 fm,i contains the point Varying the possible image ιm({u1, . . . , ua}), we find an attainable maximum vm of the value i=1 fm,i always contains the point corre- max{α, β}. Then we see that the Newton polygon ofQℓm sponding to the monomial xvmyvm. Step 4. With the monomial xvmyvm, we can follow Steps A, B, C in the proof of Theorem 3.1. Then we obtain This implies c0( ℓmYi=1 fm,i! = cq(cid:18)P2, mℓm · c0 ℓmYi=1 for an arbitrary point p on Sd. Therefore, fm,i) ≥ 1 vm . 1 mℓm ¯Dm(cid:19) = cp(cid:18)Sd, 1 mℓm Dm(cid:19) ≥ mℓm vm δ(Sd) ≥ lim sup m mℓm vm . i i In Step C we define the polynomials f (1) by applying a change of coordinate x + A1yβ 7→ x to the polynomial fi for each i. We furthermore define f (k) inductively for each i and each k. We here apply the same to fm,i and denote the coordinate-changed polynomials by f (k) m,i. In the change of coordinate x + A1yβ 7→ x, we may assume that β > 1. Indeed, if β = 1, then we replace fm,i(x, y) by fm,i(x − A1y, x). Then they also form a basis for Lm and the log canonical threshold of their product at the origin is equal to that of the product of the original fm,i. Note that the edge of the Newton polygon of the product of the replaced fm,i that intersects the line s = t is not parallel to the line s = −t. This implies that β for the replaced fm,i cannot be 1 in Step C. Therefore, it is strictly bigger than 1 and hence β(k) for f (k) m,i is also strictly bigger than 1 for all k. The lemma below guarantees that we are able to go through Step C in Step 4. m :=Qℓm i=1 f (k) Lemma 4.6. For each k there is an injective map Im,k : {f (k) m,i 1 ≤ i ≤ ℓm} → such that (1) the monomial Im,k(f (k) (2) its image coincides with the image of ιm. m,i) is contained in f (k) m,i; Proof. See Appendix at the end. Sn 3m[n=0 (cid:3) Note that in some cases, we may skip Steps 0 and 1, i.e., we may start from Step 2 with a = ℓm. As we see, the δ-invariant is defined in an asymptotic way. Therefore, it is enough to see the asymptotic behaviors of mℓm and vm with respect to m. Since mℓm = d 2 m2 + m, a lower bound of the δ-invariant of Sd can be determined by the coefficient of the term m3 in vm. For reader's convenience, we here provide the following diagrams. In each diagram, the 2 m3 + d K-STABILITY OF SMOOTH DEL PEZZO SURFACES 11 exponent of x in the product of all the monomials in x, y corresponding to the integral points of a triangle (or a quadrilateral) is given by a cubic polynomial with respect to m. The number in the triangle (or the quadrilateral) is the coefficient of the term m3 in this cubic polynomial. The diagrams in Figure 1 enable us to immediately figure out the coefficients of the term m3 in the exponents vm that appear in the present and the next sections. Figure 1. 20 24 5 24 16 24 9 24 9 24 11 24 28 24 2m 5 24 7 24 m x 3m 3m 2m m y 4 24 1 24 1 24 3 24 3 24 3 24 3 24 y 3m 2m m 3m 2m m y y y 3m 2m 3 2 m m 14 24 2 24 1 5 4 m 2m 5 2 m 3m x 25 24 3 2 m 2m x 3m 7 24 m m 2m x x 3m Now we are ready to estimate the values of the δ-invariants of smooth del Pezzo surfaces. We keep the same notations as before. Theorem 4.7. Let S be a del Pezzo surface of degree 1. Then δ(S) ≥ 3 2 . Proof. Let C be a cubic curve in P2 that passes through the points π(M1), . . . , π(M8) and q. Since π(M1), . . . , π(M8) are in general position, the curve C must be irreducible and reduced. Case 1. The curve C is smooth at q. The effective divisor G produced by a section in Lm can pass through the point q with multiplicity at most m. To see this, we write G = m′C + Ω, where m′ is a non-negative 12 JIHUN PARK AND JOONYEONG WON integer not bigger than m and Ω is an effective divisor whose support does not contain C. If multq(G) > m, then multq(Ω) > m − m′. This yields an absurd inequality 9(m − m′) = C · Ω ≥ multq(Ω) + multπ(Mi)(C) · multπ(Mi)(Ω) > 9(m − m′). Therefore, the image of the injective map in (4.2) is always contained in the set 8Xi=1 These monomials are plotted in the shade area of the following diagram: {xn1yn2 0 ≤ n1 + n2 ≤ m}. y 3m 2m m m 2m 3m x Since mℓm = 1 2 m(m2 + m + 2), Figure 1 shows lim sup m mℓm vm = 3. Case 2. The curve C is singular at q. Then C is a unique curve in P2 that passes through the points π(M1), . . . , π(M8) and q. Note that dimk T1 = 2. Using a suitable coordinate change, if C has a node at q, then we may assume that and if C has a cusp at q, then we may assume that t1 = 1, t2 = xy, t1 = 1, t2 = (x + y)2. Therefore, in both the cases we can always take We then obtain x1,1 = 1, x1,2 = xy. Cm = {1, xy, x2y2, . . . , xmym}. Let G be the effective divisor of a section in Lm. We write G = m′C + Ω as in Case 1. Since 9(m − m′) = C · Ω ≥ 8(m − m′) + 2 multq(Ω), we obtain multq(G) = m′ · multq(C) + multq(Ω) ≤ 2m′ + m − m′ 2 . This shows that we can define the injective map ιm in (4.2) in such a way that its image is contained in the set m[i=0(cid:26)xi+n1yi+n2 0 ≤ n1, n2 and 0 ≤ n1 + n2 ≤ m − i 2 (cid:27) . These monomials are plotted in the shade area of the following diagram: K-STABILITY OF SMOOTH DEL PEZZO SURFACES 13 y 3m 2m m m 2m 3m x The exponent of x in the product of all the monomials corresponding to the integral points of the shade area in the diagram above is clearly smaller than the exponent from the shade area in the below. y 3m 2m m Therefore, Figure 1 shows m 2m 3m x lim sup m mℓm vm ≥ 3 2 . Consequently, from Case 1 and Case 2 we obtain δ(S) ≥ 3 2 . (cid:3) Theorem 4.8. Let S be a del Pezzo surface of degree 2. Then δ(S) ≥ 6 5 . Proof. Note that dimk(T1) = 3. We have two possibilities: either deg(t1) = 0, deg(t2) = deg(t3) = 1 or deg(t1) = 0, deg(t2) = 1, deg(t3) = 2. Case 1. deg(t1) = 0, deg(t2) = deg(t3) = 1. In this case, we may assume that t1 = 1, t2 = x, t3 = y by a suitable coordinate change. Therefore, x1,1 = 1, x1,2 = x and x1,3 = y. We then obtain Cm = {1, x, y, x2, xy, y2, . . . , xm, xm−1y, . . . , ym}. We claim that the effective divisor G yielded by a section in Lm cannot pass through q with multiplicity more than 2m. For the claim, we consider the effective plurianticanonical divisor on S. Since 2m = H · G ≥ multp( G) for a general member H in − KS passing through the point p, we obtain multp( G) ≤ 2m. This implies the claim. 7Xi=1 G = π∗(G) − m Mi JIHUN PARK AND JOONYEONG WON Therefore, the image of {u1, . . . , ua} under the injective map ιm in (4.2) is always contained 14 in Sn \ Cm, 2m[n=0 where a = ℓm − (m+1)(m+2) 2 = m(m−1) 2 . In case when ιm({u1, . . . , ua}) = {xn1yn2 0 ≤ n1, n2, m + 2 ≤ n1 + n2 ≤ 2m and m + 2 ≤ n1 ≤ 2m} the value α in the monomial xαyβ of (4.5) attains the possible maximum vm. Therefore, the value vm can be asymptotically evaluated by the shade area in the following diagram: y 3m 2m m i.e., Figure 1 shows m 2m 3m x lim sup m mℓm vm = 6 5 . Case 2. deg(t1) = 0, deg(t2) = 1, deg(t3) = 2. In this case, there is a unique cubic C on P2 that passes through the points π(M1), . . . , π(M7) and that has multiplicity 2 at the point q. Let G be the effective divisor of a section in Lm. Subcase 1. The curve C is irreducible. In this subcase, we may assume that the Zariski tangent term of the defining polynomial of C on U contains the monomial xy. We write G = m′C + Ω, where m′ is a non-negative integer not bigger than m and Ω is an effective divisor whose support does not contain C. Since 9(m − m′) = C · Ω ≥ 7(m − m′) + 2 multq(Ω), the injective map ιm in (4.2) can be defined in a way that its image is contained in the set m[i=0(cid:8)xi+n1yi+n2 0 ≤ n1, n2 and 0 ≤ n1 + n2 ≤ m − i(cid:9) . This set can be depicted as follows: K-STABILITY OF SMOOTH DEL PEZZO SURFACES 15 y 3m 2m m Figure 1 shows m 2m 3m x lim sup m mℓm vm = 2. Subcase 2. The curve C is reducible. The cubic C consists of a line L and an irreducible conic Q. Since (−KS)2 = 2, it cannot consist of three lines. Note that L passes through exactly two points of π(M1), . . . , π(M7) and that Q passes through five of them. We may assume that the Zariski tangent term of the defining polynomial of L on U contains the monomial x and that of Q contains the monomial y. We write G = m1L + m2Q + Ω, where m1 and m2 are non-negative integers and Ω is an effective divisor whose support contains neither L nor Q. From the inequalities 3m = L · (m1L + m2Q + Ω) ≥ m1 + 2m2 + 2(m − m1) + multq Ω, 6m = Q · (m1L + m2Q + Ω) ≥ 2m1 + 4m2 + 5(m − m2) + multq Ω, we obtain −m1 + 2m2 ≤ m and 2m1 − m2 ≤ m. These imply m1, m2 ≤ m. Moreover, we obtain multq(Ω) ≤ m + min{−2m1 + m2, m1 − 2m2}. Therefore, we can define the injective map ιm of (4.2) in a way that its image is contained in the sets either or [0≤m1≤m2≤m(cid:8)xm1+n1ym2+n2 0 ≤ n1, n2 and 0 ≤ n1 + n2 ≤ m + m1 − 2m2(cid:9) [0≤m2≤m1≤m(cid:8)xm1+n1ym2+n2 0 ≤ n1, n2 and 0 ≤ n1 + n2 ≤ m − 2m1 + m2(cid:9) . Both the sets sit in the shade area of the following diagram: y 3m 2m m m 2m 3m x 16 Therefore, JIHUN PARK AND JOONYEONG WON lim sup m mℓm vm = 2. Consequently, from Cases 1 and 2 we obtain δ(S) ≥ 6 5 . (cid:3) Theorem 4.9. Let S be a del Pezzo surface of degree 3. Then δ(S) ≥ 36 31 . Proof. Let C be the unique cubic on P2 that passes through the points π(M1), . . . , π(M6) and that is singular at the point q. Let G be the effective divisor of a section in Lm. Case 1. The curve C is irreducible. In this case, the Zariski tangent term of the defining polynomial of C on U may be assumed to contain the monomial xy. We write G = m′C + Ω, where m′ is a non-negative integer not bigger than m and Ω is an effective divisor whose support does not contain C. Since 9(m − m′) = C · Ω ≥ 6(m − m′) + 2 multq(Ω), we have multq(Ω) ≤ 3 2 (m − m′). This implies that the injective map ιm in (4.2) can be chosen in such a way that its image is contained in the set m[i=0(cid:26)xi+n1yi+n2 0 ≤ n1, n2 and 0 ≤ n1 + n2 ≤ 3 2 (m − i)(cid:27) . These monomials are plotted in the shade area as below: y 3m 2m m m 2m 3m x lim sup m mℓm vm = 12 7 We then obtain from Figure 1. Case 2. The curve C is reducible. The cubic curve C then consists of either one line and one irreducible conic or three lines. Subcase 1. The curve C consists of a line L and an irreducible conic Q. In this subcase, we may assume that the Zariski tangent term of the defining polynomial of L on U contains the monomial x and that of Q contains the monomial y. We write G = m1L + m2Q + Ω, where m1 and m2 are non-negative integers and Ω is an effective divisor whose support contains neither L nor Q. K-STABILITY OF SMOOTH DEL PEZZO SURFACES 17 We first suppose that L passes through exactly one of the points π(M1), . . . , π(M6). From 6m = Q · (m1L + m2Q + Ω) ≥ 2m1 + 4m2 + 5(m − m2) + multq Ω, 3m = L · (m1L + m2Q + Ω) ≥ m1 + 2m2 + (m − m1) + multq Ω, we obtain m2 ≤ m, 2m1 − m2 ≤ m and multq Ω ≤ min {m − 2m1 + m2, 2m − 2m2} . Therefore, the injective map ιm in (4.2) can be defined in such a way that its image sits in the set [0≤m2≤m 2m1−m2≤m (cid:26)xm1+n1ym2+n2 0 ≤ n1, n2 and 0 ≤ n1 + n2 ≤ min(cid:26) m − 2m1 + m2 2m − 2m2 (cid:27)(cid:27) . It is easy to check that the monomials in this set are plotted in the shade area of the following diagram: y 3m 2m m m 2m 3m x We now suppose that L passes through two points of π(M1), . . . , π(M6). Then, from the inequalities 6m = Q · (m1L + m2Q + Ω) ≥ 2m1 + 4m2 + 4(m − m2) + multq Ω, 3m = L · (m1L + m2Q + Ω) ≥ m1 + 2m2 + 2(m − m1) + multq Ω, we obtain m1 ≤ m, 2m2 − m1 ≤ m and multq Ω ≤ min {m − 2m2 + m1, 2m − 2m1} . As the previous case, the injective map ιm in (4.2) can be chosen in a way that its image is contained in the set [0≤m1≤m 2m2−m1≤m (cid:26)xm1+n1ym2+n2 0 ≤ n1, n2 and 0 ≤ n1 + n2 ≤ min(cid:26) m − 2m2 + m1 2m − 2m1 (cid:27)(cid:27) . The monomials in this set are plotted in the shade area of the following diagram: 18 JIHUN PARK AND JOONYEONG WON y 3m 2m m Therefore, Figure 1 implies m 2m 3m x lim sup m mℓm vm = 36 31 . Subcase 2. The curve C consists of three lines L1, L2, L3 with multq(C) = 2. We may assume that q is the intersection point of L1 and L2. In addition, we may assume that L1 is defined by x = 0 and L2 by y = 0. We write G = m1L1 + m2L2 + m3L3 + Ω, where m1, m2, m3 are non-negative integers and Ω is an effective divisor whose support contains none of L1, L2, L3. From the inequalities 3m = L1 · (m1L1 + m2L2 + m3L3 + Ω) ≥ m1 + m2 + m3 + 2(m − m1) + multq(Ω), 3m = L2 · (m1L1 + m2L2 + m3L3 + Ω) ≥ m1 + m2 + m3 + 2(m − m2) + multq(Ω), 3m = L3 · (m1L1 + m2L2 + m3L3 + Ω) ≥ m1 + m2 + m3 + 2(m − m3), we see that the injective map ιm in (4.2) can be defined in such a way that its image is contained in the set [0≤m1,m2,m3≤m m1+m2≤m3+m (cid:26)xm1+n1ym2+n2 0 ≤ n1, n2 and 0 ≤ n1 + n2 ≤ min(cid:26) m − m1 + m2 − m3 m + m1 − m2 − m3(cid:27)(cid:27) . The monomials in this set are plotted in the shade area of the following diagram: y 3m 2m m Therefore, m 2m 3m x lim sup m mℓm vm = 18 11 . Subcase 3. The curve C consists of three lines L1, L2, L3 with multq(C) = 3. The line L1 can be assumed to be defined by x = 0, L2 by y = 0 and L3 by x + y = 0. K-STABILITY OF SMOOTH DEL PEZZO SURFACES 19 We write G = m1L1 + m2L2 + m3L3 + Ω, where m1, m2, m3 are non-negative integers and Ω is an effective divisor whose support contains none of L1, L2, L3. The three inequalities 3m = L1 · (m1L1 + m2L2 + m3L3 + Ω) ≥ m1 + m2 + m3 + 2(m − m1) + multq(Ω), 3m = L2 · (m1L1 + m2L2 + m3L3 + Ω) ≥ m1 + m2 + m3 + 2(m − m2) + multq(Ω), 3m = L3 · (m1L1 + m2L2 + m3L3 + Ω) ≥ m1 + m2 + m3 + 2(m − m3) + multq(Ω) show that the injective map ιm in (4.2) can be arranged in such a way that its image is contained in the set [0≤m1,m2,m3≤m  xm1+m3+n1ym2+n2 0 ≤ n1, n2 and 0 ≤ n1 + n2 ≤ min The monomials in this set sit in the shade area in the diagram below: m + m1 − m2 − m3 m − m1 + m2 − m3 m − m1 − m2 + m3 .   y 3m 2m m m 2m 3m x Figure 1 then implies mℓm vm Consequently, Cases 1 and 2 imply δ(S) ≥ 36 31 . lim sup m = 9 7 . (cid:3) Theorem 4.10. Let S be a del Pezzo surface of degree 4. Then δ(S) ≥ 12 11 . Proof. Note that at most two (−1)-curves can pass through a given point p on S. Let G be the effective divisor of a section in Lm. Case 1. There is no (−1)-curve that passes through the point p. In this case, there is an irreducible cubic curve C on P2 that passes through the points π(M1), . . . , π(M5) and that is singular at the point q. We may assume that the Zariski tangent term of the defining polynomial of C on U contains the monomial xy. We write G = m′C + Ω, where m′ is a non-negative integer not bigger than m and Ω is an effective divisor whose support does not contain C. From the inequality 9(m − m′) = C · Ω ≥ 5(m − m′) + 2 multq(Ω), we obtain multq(Ω) ≤ 2(m − m′). Therefore, the injective map ιm in (4.2) can be chosen in such a way that its image is contained in the set m[i=0(cid:8)xi+n1yi+n2 0 ≤ n1, n2 and 0 ≤ n1 + n2 ≤ 2(m − i)(cid:9) . The monomials in this set correspond to the integral points in the shade area below: 20 JIHUN PARK AND JOONYEONG WON y 3m 2m m Therefore, Figure 1 implies m 2m 3m x lim sup m mℓm vm = 3 2 . Case 2. There is only one (−1)-curve that passes through the point p. In this case, there is either a line passing through the point q and two of the points π(M1), . . . , π(M5) or an irreducible conic passing through q and π(M1), . . . , π(M5). We consider the former case only since the latter can be dealt in almost the same way. Let L be the line in the former case. We may assume that it passes through π(M1) and π(M2). Also, there is an irreducible conic Q that passes through the points q and π(M2), . . . , π(M5). We write G = m1L + m2Q + Ω, where m1, m2 are non-negative integers and Ω is an effective divisor whose support contains neither L nor Q. From the inequalities 3m = L · (m1L + m2Q + Ω) ≥ m1 + 2m2 + 2(m − m1) + multq(Ω), 6m = Q · (m1L + m2Q + Ω) ≥ 2m1 + 4m2 + 4(m − m2) + multq(Ω), we obtain This implies that multq(Ω) ≤ min{m + m1 − 2m2, 2m − 2m1}. multq(G) = m1 + m2 + multq(Ω) ≤ min{m + 2m1 − m2, 2m − m1 + m2} ≤ 2m. Therefore, the image of the injective map ιm in (4.2) is always contained in the same set as in the previous case, and hence lim sup m mℓm vm = 3 2 . Case 3. There are two (−1)-curves that pass through the point p. There are two distinct lines L1, L2 passing through the point q and two of the points π(M1), . . . , π(M5). We may assume that L1 passes through π(M1), π(M2) and that L2 passes through π(M3), π(M4). Let L3 be the line determined by q and π(M5). By suitable coordinate changes, we may assume that the line L1 is defined by x = 0, L2 by y = 0 and L3 by x + y = 0. We then see that t1 = 1, t2 = x, t3 = y, t4 = xy, t5 = xy(x + y) form a basis for T1. Furthermore, we may take x1,1 = 1, x1,2 = x, x1,3 = y, x1,4 = xy, x1,5 = x2y. K-STABILITY OF SMOOTH DEL PEZZO SURFACES 21 Then the set Cm =( 5Yi=1 xni 1,i (cid:12)(cid:12)(cid:12) ni are non-negative integers with n1 + · · · + n5 = m) consists of the monomials corresponding the integral points in the shade area of the following diagram: y 3m 2m m m 2m 3m x We now write G = m1L1 + m2L2 + m3L3 + Ω, where m1, m2, m3 are non-negative integers and Ω is an effective divisor whose support contains none of L1, L2, L3. The inequities 3m = L1 · (m1L1 + m2L2 + m3L3 + Ω) ≥ m1 + m2 + m3 + 2(m − m1) + multq(Ω), 3m = L2 · (m1L1 + m2L2 + m3L3 + Ω) ≥ m1 + m2 + m3 + 2(m − m2) + multq(Ω), 3m = L3 · (m1L1 + m2L2 + m3L3 + Ω) ≥ m1 + m2 + m3 + (m − m3) + multq(Ω) imply multq(Ω) ≤ min{m + m1 − m2 − m3, m − m1 + m2 − m3, 2m − m1 − m2}. This shows that the images ιm(u1), . . . , ιm(ua) can be plotted only above or on the line joining (m, 0) and (2m, m). Since deg(G) = 3m, they must sit only below or on the line joining (3m, 0) and (0, 3m). Consequently, the maximum value vm can be attained when the image of the injective map ιm in (4.2) are contained in the set of the monomials corresponding the integral points in the shade area of the following diagram: y 3m 2m m m 2m 3m x Therefore, mℓm vm Consequently, Cases 1, 2 and 3 imply δ(S) ≥ 12 11 . lim sup m = 12 11 . (cid:3) Theorem 4.11. Let S be a del Pezzo surface of degree 5. Then δ(S) ≥ 15 14 . 22 JIHUN PARK AND JOONYEONG WON Proof. For a point p on S, we may or may not have a member in − KS that has multiplicity 3 at p. Case 1. No divisor in − KS has multiplicity 3 at p. In this case, the monomials t1 = 1, t2 = x, t3 = y, t4 = x2, t5 = xy, t6 = y2 form a basis for T1. Therefore, x1,1 = 1, x1,2 = x, x1,3 = y, x1,4 = x2, x1,5 = xy, x1,6 = y2. Then we obtain the set Cm = {xn1yn2 0 ≤ n1, n2 and 0 ≤ n1 + n2 ≤ 2m} . Let G be the effective divisor of a section in Lm and let H be a general curve of degree 3 on P2 that has multiplicity 2 at q and passes through π(M1), π(M2), π(M3) and π(M4). Then the inequality 9m ≥ multq(H) multq(G) + multπ(Mi)(H) multπ(Mi)(G) ≥ 4m + 2 multq(G) 4Xi=1 shows that the maximum value vm can be attained when the image of the injective map ιm in (4.2) is contained in the set of the monomials corresponding to the integral points in the shade area of the following diagram: y 3m 2m m 5 4 m 2m 3m x To be precise, the monomials of the set Cm correspond to the integral points below and on the line joining (2m, 0) and (0, 2m). The images ιm(u1), . . . , ιm(ua) sit in the shade area above the line joining (2m, 0) and (0, 2m). Then, Figure 1 shows that lim sup m mℓm vm = 15 14 . Case 2. There is a divisor C in − KS that has multiplicity 3 at p. The curve π(C) consists of three lines L1, L2, L3 passing through the point q. We may assume that the points π(M1) and π(M2) belong to L1. Then the points π(M3) and π(M4) belong to the union of L2 and L3. Furthermore, we may assume that L1 is defined by x = 0, L2 by y = 0 and L3 by x + y = 0. Then t1 = 1, t2 = x, t3 = y, t4 = x2, t5 = xy, t6 = xy(x + y) form a basis for T1. We may take x1,1 = 1, x1,2 = x, x1,3 = y, x1,4 = x2, x1,5 = xy, x1,6 = xy2. K-STABILITY OF SMOOTH DEL PEZZO SURFACES 23 The set Cm consists of the monomials corresponding the integral points in the shade area of the following diagram: y 3m 2m m m 2m 3m x Note that the number of monomials in this set is exactly ℓm. Therefore, a = 0 in Step 2 and Figure 1 shows that lim sup m mℓm vm = 15 13 . Cases 1 and 2 complete the proof. (cid:3) 5. K-semistable del Pezzo surfaces II In this section, together with Section 3, we complete the proof of the second statement of Main Theorem. Let S be the del Pezzo surface of degree 6. This surface can be obtained by blowing up P2 at p1 = [0 : 0 : 1], p2 = [0 : 1 : 0] and p3 = [1 : 0 : 0]. Let φ3 : S → P2 be the blow-up with the exceptional curves E, F , G. For a fixed positive integer m, the space H0(S, OS (−mKS)) can be regarded as the subspace Lm of H0(P2, OP2(3m)) consisting of the sections vanishing at p1, p2, p3 with order at least m. The set B3 =nxaybzc a, b, c are integers with a + b + c = 3m and 0 ≤ a, b, c ≤ 2mo forms a basis for Lm. Let Da,b,c be the divisor defined by xaybzc on P2. Then the divisor D = 1 m(3m2 + 3m + 1) Xxaybzc∈B3 (φ∗ 3(Da,b,c) − mE − mF − mG) is an anticanonical Q-divisor of m-basis type on S. We immediately see that D = Lx + Ly + Ly + E + F + G, where Lx, Ly, Lz are the proper transforms of the lines on P2 defined by x = 0, y = 0, z = 0, respectively. Therefore, δ(S) ≤ 1. Theorem 5.1. The δ-invariant of the del Pezzo surface of degree 6 is 1. Proof. Fix a positive integer m and set ℓm = h0(S, OS (−mKS)). Let {s1, . . . , sℓm} be a basis of H0(S, OS (−mKS)). We denote the effective divisor of the section si by Di. Put D =P Di. Let p be an arbitrary point on S. Case 1. The point p is not an intersection point of two (−1)-curves. In this case, there is a birational morphism π : S → P2 that is an isomorphism around the point p. We can use the exactly same method as in the previous section. In order to apply the same method, we set q = π(p) and use the same notations as before. By a suitable coordinate change, we assume that q = [0 : 0 : 1]. 24 JIHUN PARK AND JOONYEONG WON Subcase 1. The point p does not lie on any (−1)-curve. Note that dimk T1 = 7. Since the three lines determined by the points q and π(Mi), i = 1, 2, 3, are distinct, by a suitable coordinate change we may assume that t1 = 1, t2 = x, t3 = y, t4 = x2, t5 = xy, t6 = y2, t7 = xy(x + y) form a basis for T1. We may therefore take x1,1 = 1, x1,2 = x, x1,3 = y, x1,4 = x2, x1,5 = xy, x1,6 = y2, x1,7 = x2y. Then the set Cm consists of the monomials corresponding to the integral points in the shade area of the following diagram: y 3m 2m m m 2m 3m x Note that the number of monomials in this set is exactly ℓm. Therefore, a = 0 in Step 2 and Figure 1 shows that lim sup m mℓm vm = 9 8 . Subcase 2. The point p lies on a single (−1)-curve. In this case, by a suitable coordinate change we may assume that t1 = 1, t2 = x, t3 = y, t4 = x2, t5 = xy, t6 = x2y, t7 = xy2 form a basis for T1. Then x1,1 = 1, x1,2 = x, x1,3 = y, x1,4 = x2, x1,5 = xy, x1,6 = x2y, x1,7 = xy2. The set Cm consists of the monomials corresponding to the integral points in the shade area of the following diagram: y 3m 2m m m 2m 3m x K-STABILITY OF SMOOTH DEL PEZZO SURFACES 25 The number of monomials in this set is exactly ℓm. Therefore, a = 0 in Step 2 and Figure 1 shows that lim sup m mℓm vm = 1. Case 2. The point p is an intersection point of two (−1)-curves. There is a birational morphism φ : S → P1 × P1 that is an isomorphism around the point p. The morphism φ is obtained by contracting two suitable disjoint (−1)-curves M1, M2 on S. For an effective divisor C ∈ − mKS, the divisor φ(C) is a curve of bidegree (2m, 2m) on P1 × P1 which passes through the points φ(M1) and φ(M2) with multiplicities at least m. Such divisors yield an ℓm-dimensional subspace of H0(P1 × P1, OP1×P1(−mKP1×P1)). The effective divisors φ(D1), . . . , φ(Dℓm ) induce a basis for this subspace. Let G be an effective divisor on P1 × P1 that comes from − mKS. Let L1 and L2 be the two 0-curves that pass through the point φ(p). We write G = m1L1 + m2L2 + Ω, where m1 and m2 are non-negative integers and Ω is an effective divisor whose support contains neither L1 nor L2. We may assume that φ(M1) belongs to L1 and φ(M2) belongs to L2. We use a bihomogeneous coordinate system ([x : u], [y : v]) for P1 × P1. For the proof, we may assume that φ(p) = ([0 : 1], [0 : 1]). Putting u = 1 and v = 1, we may regard x and y as local coordinates around the point φ(p). Even though we maps the surface S onto P1 × P1 instead of P2, the original method to estimate cp(S, mℓm D) works verbatim for this case. 1 The inequalities 2m = L1 · (m1L1 + m2L2 + Ω) ≥ m2 + (m − m1) + multφ(p)(Ω), 2m = L2 · (m1L1 + m2L2 + Ω) ≥ m1 + (m − m2) + multφ(p)(Ω) imply m1 − m2 ≤ m and m − m1 − m2 ≥ multφ(p)(Ω). Note that 0 ≤ m1, m2 ≤ 2m. This implies that the image of the injective map ιm in (4.2) can be arranged to be contained in the set of the monomials corresponding to the integral points in the shade area of the following diagram: y 3m 2m m m 2m 3m x Note that the number of monomials in this set is exactly ℓm. Figure 1 shows that lim sup m mℓm vm = 1. From Cases 1 and 2, we can draw the conclusion that δ(S) = 1. (cid:3) 26 JIHUN PARK AND JOONYEONG WON Now we consider the space P1×P1. We use a bihomogeneous coordinate system ([x : u], [y : v]) for P1 × P1. As before, fix a positive integer m and set ℓm = h0(P1 × P1, OP1×P1(−mKP1×P1)). The ℓm monomials of bidegree (2m, 2m) in variables x, u; y, v form a basis for H0(P1 × P1, OP1×P1(−mKP1×P1)). Therefore, the devisor defined by xuyv = 0 is an anticanonical di- visor of m-basis type. This shows Theorem 5.2. The δ-invariant of P1 × P1 is 1. δ(P1 × P1) ≤ 1. Proof. Let {s1, . . . , sℓm} be a basis of the space H0(P1 × P1, OP1×P1(−mKP1×P1)). Denote the effective divisor defined by the section si by Di. Set D =P Di. For a given point p ∈ P1 × P1, as before, the inequality will be verified. cp(P1 × P1, D) ≥ 1 mℓm We may assume that p = ([0 : 1], [0 : 1]). Putting u = 1 and v = 1, we may regard x and y as local coordinates around the point p. As Case 2 in Theorem 5.1, the original method to estimate cp(P1 × P1, mℓm D) works verbatim. 1 Let G be an effective divisor in − mKP1×P1. Let L1 and L2 be the two 0-curves that pass through the point p. We write G = m1L1 + m2L2 + Ω, where m1 and m2 are non-negative integers and Ω is an effective divisor whose support contains neither L1 nor L2. The inequalities 2m = L1 · (m1L1 + m2L2 + Ω) ≥ m2 + multp(Ω), 2m = L2 · (m1L1 + m2L2 + Ω) ≥ m1 + multp(Ω) imply 0 ≤ m1, m2 ≤ 2m and 2m − max{m1, m2} ≥ multp(Ω). This implies that the image of the injective map ιm in (4.2) is contained in the set of the monomials corresponding to the integral points in the shade area of the following diagram: y 3m 2m m Figure 1 then shows that m 2m 3m x lim sup m mℓm vm = 1. (cid:3) K-STABILITY OF SMOOTH DEL PEZZO SURFACES 27 6. Non-K-semistable del Pezzo surfaces In this section, we verify the last statement of Main Theorem. Since Conjecture 1.7 has not been verified completely, at this moment we cannot say that the last statement of Main Theorem implies that the Hirzebruch surface F1 and the del Pezzo surface of degree 7 are not K-semistable. However, it is able to serve as a good evidence for Conjecture 1.7. Let [x : y : z] be a homogeneous coordinate for P2. The Hirzebruch surface F1 can be obtained by blowing up P2 at p1 = [0 : 0 : 1]. Let φ1 : F1 → P2 be the blow-up with the exceptional divisor E. For a fixed positive integer m, the space H0(F1, OF1(−mKF1)) can be regarded as the subspace M1 of H0(P2, OP2(3m)) consisting of the sections vanishing at p1 with order at least m. The set B1 =nxaybzc a, b, c are non-negative integers with a + b + c = 3m, c ≤ 2mo forms a basis for M1. Let Da,b,c be the divisor defined by xaybzc on P2. Then the divisor D = 1 m(2m + 1)2 Xxaybzc∈B1 (φ∗ 1(Da,b,c) − mE) is an anticanonical Q-divisor of m-basis type on F1. The multiplicity of D along the curve E is 7m+4 6m+3 . Therefore, and hence δm(F1) ≤ 6m + 3 7m + 4 , δ(F1) ≤ 6 7 . We now let S be the del Pezzo surface of degree 7. It can be obtained by blowing up P2 at p1 = [0 : 0 : 1] and p2 = [0 : 1 : 0]. Let φ2 : S → P2 be the blow-up with the exceptional curves E and F . For a fixed positive integer m, the space H0(S, OS (−mKS)) can be regarded as the subspace M2 of H0(P2, OP2(3m)) consisting of the sections vanishing at p1 and p2 with order at least m. The set B2 =nxaybzc a, b, c are non-negative integers with a + b + c = 3m, b ≤ 2m, c ≤ 2mo forms a basis for M2. Let Ca,b,c be the divisor defined by xaybzc on P2. The anticanonical Q-divisor C = 2 7m2(m + 1) + 2m Xxaybzc∈B2 (φ∗ 2(Ca,b,c) − mE − mF ) is of m-basis type on S. The multiplicity of C along the proper transform of the curve defined by x = 0 is 25m2+27m+8 21m(m+1)+6 . Therefore, and hence δm(S) ≤ 21m(m + 1) + 6 25m2 + 27m + 8 , δ(S) ≤ 21 25 . 28 JIHUN PARK AND JOONYEONG WON Appendix In Appendix Lemmas 4.1, 4.3 and 4.6 are verified. All the notations are the same as those in the beginning of Section 4. We consider the vector space Vλ := k[x, y]n λMn=0 with an ordered basis {xαyβ α + β ≤ λ}, where k[x, y]n is the (n + 1)-dimensional vector space of homogeneous polynomials of degree n in variables x, y. The order of the basis is given in the following way: (1) the graded lexicographic order with x ≺ y except for Case 2 in Theorem 4.7 and Case 2 in Theorem 4.11; (2) the graded lexicographic order with y ≺ x for Case 2 in Theorem 4.11; (3) the order for Case 2 in Theorem 4.7 satisfies the properties: i=1 xni 1,i in Cm smaller than any other monomials (a) xα1yβ1 ≺ xα2yβ2 if α1 + β1 < α2 + β2; (b) xαyα ≺ xα2yβ2 if α2 6= β2 and α2 + β2 = 2α. Note that these orders make the monomialQd+1 that appear inQd+1 i=1 tni i . Since fm,i is a member of the vector space V3m, we may express the polynomial fm,i as a 1 × σ matrix with respect to the given ordered basis, where σ = (3m+1)(3m+2) . By writing these 1 × σ matrices as rows, we can express the ℓm polynomials fm,1, · · · , fm,ℓm altogether as a single ℓm × σ matrix MF . Since fm,1, · · · , fm,ℓm are linearly independent, the rank of the matrix MF is exactly ℓm. 2 Let E be a row echelon form of the matrix MF . Then there is an ℓm × ℓm invertible matrix T such that MF = T E. Since the rank of MF is ℓm, the matrix E does not have any zero row. The i-th row of E represents a polynomial hm,i that belongs to Lm. Its Zariski tangent term tm,i is represented by the pivot (the first non-zero entry from the left in a row) and the entries whose corresponding monomials have the same degree as the monomial corresponding to the pivot. In particular, the Zariski tangent term contains the monomial corresponding to the pivot. The polynomials hm,i form a basis for the space Lm and their Zariski tangent terms tm,i form a basis for the space Tm. contain the set Cm. By collecting the ℓm pivot columns of E in order, we obtain an ℓm×ℓm upper triangular matrix in Cm is smaller than any other monomials that appear i , the set of the monomials corresponding to the columns with the pivots of E must Since the monomial Qd+1 inQd+1 with the pivots on the diagonal. Denote this minor matrix of E by eE. We also denote the ℓm×ℓm matrix TeE by fMF . The entries of i-th row of fMF are the coefficients of the monomials in fm,i corresponding to the pivot columns of E. Since the matrix fMF is nonsingular, we can choose a single non-zero entry from each column of fMF in such a way that the non-zero entries are For Lemma 4.6, we consider the vector space Vλ with a sufficiently large positive integer λ so that we could write polynomials of bigger degrees as matrices. Note that the change of coordinate x + A1yβ 7→ x in Step C is given with β > 1. Since selected exactly one time from each row. This proves Lemmas 4.1 and 4.3. i=1 xni 1,i i=1 tni fm,i(x, y) = ℓmXj=1 Tijhm,j(x, y) K-STABILITY OF SMOOTH DEL PEZZO SURFACES 29 for each i, where Tij is the entry of T in the i-th row and the j-th column, we have f (1) m,i(x, y) = fm,i(x − A1yβ, y) = Tijhm,j(x − A1yβ, y). ℓmXj=1 Since β > 1, we immediately see that the change of coordinate does not give any effect on the Zariski tangent term of hm,i at all. It therefore leaves the positions of the pivot columns of E unchanged. Therefore, by the same argument as for Lemmas 4.1 and 4.3, we can obtain an injection in Lemma 4.6. Furthermore, the same argument works inductively for {f (k) m,i}, k > 1. This completes the proof of Lemma 4.6. References [1] Cheltsov, I.: Log canonical thresholds of del Pezzo surfaces. Geom. Funct. Anal. 11, 1118–1144 (2008) [2] Cheltsov, I., Park, J.: Global log-canonical thresholds and generalized Eckardt points. Sb. Math. 193 (5–6), 779–789 (2002) [3] Cheltsov, I., Park, J.: Sextic double solids. In: Cohomological and geometric approaches to rationality problems, Progr. Math., 282, Birkhauser, Boston, MA, 75–132 (2010) [4] Cheltsov, I., Park, J., Won. J: Log canonical thresholds of certain Fano hypersurfaces. Math. Z. 276 (1–2), 51–79 (2014) [5] Cheltsov, I., Shramov, K.: Log-canonical thresholds for nonsingular Fano threefolds. With an appendix by J.-P. Demailly. Russian Math. Surveys 63 (5), 859–958 (2008) [6] Chen, X., Donaldson, S., Sun, S.: Kahler-Einstein metrics on Fano manifolds I: approximation of metrics with cone singularities. J. Amer. Math. Soc. 28 (1), 183–197 (2015) [7] Chen, X., Donaldson, S., Sun, S.: Kahler-Einstein metrics on Fano manifolds, II: limits with cone angle less than 2π. J. Amer. Math. Soc. 28 (1), 199–234 (2015) [8] Chen, X., Donaldson, S., Sun, S.: Kahler-Einstein metrics on Fano manifolds III: limits as cone angle approaches 2π and completion of the main proof. J. Amer. Math. Soc. 28 (1), 235–278 (2015) [9] Demailly, P., Koll´ar, J.: Semicontinuity of complex singularity exponents and Kahler-Einstein metrics on Fano orbifolds. Ann. Ec. Norm. Sup 34, 525–556 (2001) [10] Fujita, K.: On K-stability and the volume functions of Q-Fano varieties. Proc. Lond. Math. Soc. 113 (5), 541–582 (2016) [11] Fujita, K.: K-stability of Fano manifolds with not small alpha invariants. preprint arXiv:1606.08261 [12] Fujita, K.: On Berman-Gibbs stability and K-stability of Q-Fano varieties. Compositio Math. 152, 288–298 (2016) [13] Fujita, K., Odaka, Yu.: On the K-stability of Fano varieties and anticanonical divisors. preprint arXiv:1602.01305 [14] Hacking, P., Prokhorov, Yu.: Smoothable del Pezzo surfaces with quotient singularities. Compositio Math. 146 (1), 169–192 (2010) [15] Hwang, J.-M., Kim, H., Lee, Y., Park, J.: Slopes of smooth curves on Fano manifolds. Bull. Lond. Math. Soc. 43 (5), 827–839 (2011) [16] Kempf, G.: Instability in invariant theory. Ann. of Math. (2) 108 (2), 299–316 (1978) [17] Koll´ar, J.: Singularities of pairs. In: Algebraic geometry (Santa Cruz, 1995) Part 1, Proc. Sympos. Pure Math., 62, Amer. Math. Soc., 221–287 (1997) [18] Manetti, M.: Normal degenerations of the complex projective plane. J. Reine Angew. Math. 419, 89–118 (1991) [19] Manetti, M.: Normal projective surfaces with ρ = 1, P−1 ≥ 5. Rend. Sem. Mat. Univ. Padova 89, 195–205 (1993) [20] Odaka, Yu., Sano, Yu.: Alpha invariant and K-stability of Q-Fano varieties. Adv. Math. 229 (5), 2818–2834 (2012) [21] Panov, D., Ross, J.: Slope stability and exceptional divisors of high genus. Math. Ann. 343 (1), 79–101 (2009) [22] Prokhorov, Yu.: A note on degenerations of del Pezzo surfaces. Ann. Inst. Fourier (Grenoble) 65 (1), 369–388 (2015) [23] Ross, J., Thomas, R.: A study of the Hilbert-Mumford criterion for the stability of projective varieties. J. Algebraic Geom. 16 (2), 201–255 (2007) [24] Tian, G.: On Kahler-Einstein metrics on certain Kahler manifolds with c1(M ) > 0. Invent. Math. 89, 225–246 (1987) 30 JIHUN PARK AND JOONYEONG WON [25] Tian, G.: K-stability and Kahler-Einstein metrics. Comm. Pure Appl. Math. 68 (7), 1085–1156 (2015). Corrigendum: K-Stability and Kahler-Einstein Metrics. 68 (11), 2082–2083 (2015) Jihun Park Center for Geometry and Physics, Institute for Basic Science (IBS) 77 Cheongam-ro, Nam-gu, Pohang, Gyeongbuk, 37673, Korea. Department of Mathematics, POSTECH 77 Cheongam-ro, Nam-gu, Pohang, Gyeongbuk, 37673, Korea. [email protected] Joonyeong Won Center for Geometry and Physics, Institute for Basic Science (IBS) 77 Cheongam-ro, Nam-gu, Pohang, Gyeongbuk, 37673, Korea. [email protected]
1903.02038
2
1903
2019-11-26T12:21:30
Minimal Newton strata in Iwahori double cosets
[ "math.AG" ]
The set of Newton strata in a given Iwahori double coset in the loop group of a reductive group G is indexed by a finite subset of the set B(G) of Frobenius-conjugacy classes. For unramified $G$ we show that it has a unique minimal element and determine this element. Under a regularity assumption we also compute the dimension of the corresponding Newton stratum. We derive corresponding results for affine Deligne-Lusztig varieties.
math.AG
math
MINIMAL NEWTON STRATA IN IWAHORI DOUBLE COSETS EVA VIEHMANN Abstract. The set of Newton strata in a given Iwahori double coset in the loop group of a reductive group G is indexed by a finite subset of the set B(G) of Frobenius-conjugacy classes. For unramified G, we show that it has a unique minimal element and determine this element. Under a regularity assumption we also compute the dimension of the corresponding Newton stratum. We derive corresponding results for affine Deligne-Lusztig varieties. 1. Introduction Let F be a local field with ring of integers OF , uniformizer t and residue field Fq of characteristic p. Let L denote the completion of the maximal unramified extension of F , and OL its ring of integers. Throughout the paper we assume that F is of equal characteristic p. However, most results can then be translated literally to the arithmetic case, i.e. to F = Qq, and to the reduction of Shimura varieties (see Section 3). In particular, our main theorem (Theorem 1.1 below) also holds in that context. Let G be an unramified reductive group over F , and let K be reductive over OF with KF = G. Denote by σ the Frobenius of L over F . For b ∈ G(L) let [b]G = [b] = {g−1bσ(g) g ∈ G(L)} be its σ-conjugacy class and let B(G) = {[b] b ∈ G(L)}. We fix a Borel subgroup B and a maximal torus T ⊂ B of G, both defined over OF . Then the elements [b] ∈ B(G) are classified by Kottwitz [K1] by two invariants, the Newton point νb ∈ X∗(T )dom,Q and the Kottwitz point κG(b) ∈ π1(G)Γ. Here π1(G) is the fundamental group of G, i.e. the quotient of X∗(T ) by the coroot lattice, and Γ denotes the absolute Galois group of F . There is a partial order (cid:22) on B(G) defined by [b] (cid:22) [b′] if and only if κG(b) = κG(b′) and νb′ − νb is a non-negative rational linear combination of positive coroots. A σ-conjugacy class [b] is called basic if νb is central in G. If [b] is basic, it is the unique minimal element of the set of [b′] with κG(b′) = κG(b). This induces a bijection between π1(G)Γ and the set of basic [b] ∈ B(G). Let I ⊂ K be the Iwahori sub-group scheme whose reduction modulo t is the opposite of the Borel subgroup B. Let fW be the Iwahori-Weyl group of G, where for all details on the notation we refer to Section 2. Then G(L) =`x∈fW IxI. Here we use the same letter x to denote an element of fW and a chosen representative in G(L). For fixed x ∈ fW we consider the set B(G)x = {[b] ∈ B(G) [b] ∩ IxI 6= ∅} and for [b] ∈ B(G) let Nx,[b] = N G x,[b] := [b] ∩ IxI. The author was partially supported by ERC Consolidator Grant 770936: NewtonStrat. 1 By [RR] this is the set of closed points of a locally closed (reduced) subscheme of IxI which we call the Newton stratum of [b] in IxI. In this context two very natural questions are: • What is B(G)x? • What is the codimension of Nx,[b] in IxI? In general, still very little is known. Let us describe some partial answers that have been obtained due to the joined effort of several people. An obvious necessary condition for [b] ∈ B(G)x is that κG(b) = κG(x). Recent results of Gortz, He and Nie [GHN] give a necessary and sufficient condition determining if the unique basic element of B(G) satisfying κG(b) = κG(x) is indeed in B(G)x, compare Section 6. If this is the case, it is the unique smallest element of B(G)x with respect to the order on B(G). Another element of B(G)x that is of particular interest is the unique maximal element [bx] of B(G)x, which coincides with the generic σ-conjugacy class in the irreducible double coset IxI. There are descriptions of [bx] that also give finite algorithms to compute it, in [V2] via the partial order on fW and in [M] via the quantum Bruhat graph. However, none of them provides a closed formula for [bx]. A complete description of B(G)x is only known in very particular cases such as for example if x is of minimal length in its σ-conjugacy class (in which case B(G)x = {[x]}, compare [He1]) or if G = SL3 (see [B]). In general (already for G = SL3), the partially ordered set B(G)x may be non-saturated, i.e. there may be gaps in form of elements [b1] (cid:22) [b2] (cid:22) [b3] ∈ B(G) with [b1], [b3] ∈ B(G)x and [b2] /∈ B(G)x. Although the dimensions of Nx,[b] ⊂ IxI are a priori infinite, such dimensions and the codimension can be defined in a meaningful (and finite) way. Then they can be related to the dimension of the corresponding affine Deligne-Lusztig variety defined as Xx(b) = {g ∈ G(L)/I g−1bσ(g) ∈ IxI}. Indeed, by [He3], Theorem 2.23 we have (1) dim Xx(b) = ℓ(x) − h2ρ, ν(b)i − codim(N[b] ⊂ IxI). By [MV], Corollary 2.11, a similar statement holds for the individual irreducible components. Again, very little is known about these dimensions. Notice also that in general there are Newton strata and affine Deligne-Lusztig varieties which are not equidimensional (see [GH], 5.2 for an example for G of type A3 and b = 1 basic). There is a good approximation to dim Xx(b) by the so-called virtual dimension introduced by He in [He1], 10.1 as (2) dG x (b) = dx(b) = 1 2(cid:0)ℓ(x) + ℓ(η(x)) − def Gb(cid:1) − hρ, νG b i, where for notation we refer again to Section 2. By [He3], Theorem 2.30 we have If Xx(b) 6= ∅ we set dG x (b) ≥ dim Xx(b). ∆x(b) = ∆G x (b) = dG x (b) − dim Xx(b). If the basic locus in IxI is non-empty, and one assumes in addition that x is in the shrunken Weyl chambers, then He shows that for this basic locus, ∆x(b) = 0 (compare Section 6). Results of E. Milicevic and the author ([MV], Theorem 2.19) study those x where ∆x(bx) = 0 for the generic [bx] in IxI, and call them cordial. For cordial x, none of the above-mentioned phenomena (∆x(b) 6= 0 for some [b] ∈ B(G)x, 2 non-equidimensionality of some Xx(b), or gaps in B(G)x) occurs in IxI. Corollary 1.2 below provides a partial converse to this result. Altogether it therefore seems promising to approach the above two questions using the following three steps: I. What are the minimal elements of B(G)x, and what is the dimension of the corresponding stratum? II. What is the maximal element [bx] of B(G)x (in terms of a closed formula), and what is ∆x(bx)? III. What are the gaps in B(G)x, which strata are not equidimensional, and how are these phenomena reflected in the values of ∆x? The main result of this work is concerned with the first of these questions. Theorem 1.1. element [mx], which is obtained as follows. Let J be minimal such that xa is a (J, w, δ)-alcove for some w ∈ W , where a is the base alcove corresponding (1) For every x ∈ fW the set B(G)x has a unique minimal to I. Then w−1xδ(w) ∈ fWJ . Let MJ be the Levi subgroup defined by J and [b0]MJ the unique basic element of B(MJ ) with κMJ (b0) = κMJ (w−1xδ(w)). Then [mx] = [b0]G. (2) Assume that x is in the regular shrunken Weyl chambers. Then dim Xx(mx) = dx(mx), or equivalently ∆x(mx) = 0. Here, δ is the automorphism of fW induced by σ. The slightly technical notions of (J, w, δ)-alcove and of being regular shrunken are explained in Section 4 and Section 2.2, respectively. We prove the first assertion in Section 6, and the second in Section 7. On the way, we prove several comparison results between G and a Levi subgroup. From [MV], Theorem 2.19 and Theorem 1.1 above we obtain the following corol- lary which gives another result towards Step III above, complementing the results of [MV]. Corollary 1.2. Suppose that x ∈fW satisfies ∆G x (bx) > 0 and x is in the regular shrunken Weyl chambers. Then there is a [b′] ∈ B(G) such that x (mx), e.g. if ∆G x (bx) > ∆G (a) [mx] < [b′] < [bx] but [b′] /∈ B(G)x (i.e. B(G)x is not saturated), or (b) [mx] < [b′] < [bx] and [b′] ∈ B(G)x, but the closure of [b′] ∩ IxI in IxI is (cid:3) not the union of all [b′′] ∩ IxI for [b′′] ∈ B(G)x with [b′′] < [b′]. Acknowledgment. We thank the two referees for helpful remarks, in particular for suggesting a simplification of the proof of Theorem 7.3. We thank U. Gortz for a helpful discussion on sign conventions in Bruhat-Tits buildings. 2. Notation 2.1. We fix an L-split maximal torus S defined over OF and let T be the centralizer of S in G. Let be the Iwahori-Weyl group where NS is the normalizer of S in G and where T (L)0 ⊂ K(OL) is the unique parahoric subgroup of T . Then the Frobenius automorphism fW = NS(L)/T (L)0 σ of L over F induces an automorphism of fW , which we denote by δ. For each x ∈ fW we choose a representative in NS(L) which we denote by the same letter x. We choose B ⊂ K a Borel subgroup defined over OF and let I be the Iwahori 3 subgroup whose reduction modulo t agrees with the reduction of the opposite of B. It corresponds to a σ-invariant base alcove a containing the vertex corresponding to K in the apartment for S of the Bruhat-Tits building of GL. We use the convention that a lies in the dominant Weyl chamber, and that a translation element λ ∈ fW acts by translation by λ on the apartment for S. We have G(L) = ax∈fW IxI. Let Wa be the affine Weyl group (for more details on the relevant notation and fur- ther references compare [GHN], 2). Then our choice of I induces a length function and a Bruhat order on Wa. There is a short exact sequence 1 → Wa → fW → Ω → 1 where Ω is the stabilizer of a in the Bruhat-Tits building. It identifies fW with Wa ⋊ Ω. We extend the length function and the partial order from Wa to fW by setting ℓ(x) = 0 for x ∈ Ω and x ≤ y if and only if x and y are of the form x′t and y′t for some t ∈ Ω and x′ ≤ y′ ∈ Wa. Let W = NS(L)/T (L) be the (finite) Weyl group of G. Then the natural pro- the (hyper)special vertex of a corresponding to K. jection fW → W has kernel X∗(T ). We have a splitting W → fW associated with fW ∼= X∗(T ) ⋊ W . Let Φ be the set of roots of G over L relative to S. For a ∈ Φ we denote by Ua the corresponding root subgroup of G. Our choice of Borel subgroup determines a basis S of Φ of simple roots, which we also identify with the set of simple reflections in W . Let Φ+ be the set of positive roots. For a subset J ⊆ S let WJ be the subgroup of W generated by the simple reflections in J and let ΦJ ⊆ Φ be the roots spanned by J. Let Φ+ J = Φ+ ∩ ΦJ . Let MJ be the subgroup of G generated by T and all It induces an isomorphism we also use σg = σ(g). For g ∈ G(L) (or for a subset of G(L)) and x ∈ G(L) we write xg = xgx−1, and Ua for a ∈ ΦJ . If δ(J) = J, then MJ is defined over F . Let fWJ = X∗(T ) ⋊ WJ . Then fWJ is the Iwahori-Weyl group of MJ. Definition 2.1 ([GHN], 3.3). Let J ⊆ S with δ(J) = J and w ∈ W . Let x ∈ fW . Then xa is called a (J, w, δ)-alcove if w−1xδ(w) ∈ fWJ and for every α ∈ w(Φ+ −Φ+ In this context, we write IM = wMJ ∩ I (following [GHN]), and IJ = MJ ∩ I we have Uα ∩ xI ⊆ Uα ∩ I. J ) (which we use more frequently). 2.2. Affine Deligne-Lusztig varieties in affine flag varieties. For x ∈ fW and b ∈ G(L) the associated affine Deligne-Lusztig set is defined as Xx(b) = {g ∈ G(L)/I g−1bσ(g) ∈ IxI}. It is the set of Fp-valued points of a locally closed, reduced subvariety of the affine flag variety, called the affine Deligne-Lusztig variety. For µ ∈ X∗(T ) let tµ be the image of the fixed uniformizer t under the map µ : Gm → T . We consider a map η : fW → W which is defined as follows. We write x ∈ fW as x = vtµw where v, w ∈ W and tµw maps the base alcove to the dominant chamber. Equivalently we can require that w ∈ µW , i.e. that it is the shortest representative of its Wµ-coset Wµw where Wµ is the centralizer of µ in W . Then η(x) = δ−1(w)v. 4 Using this notation one can explain the definition of the virtual dimension in (2). Namely, ℓ(x) and ℓ(η(x)) are the lengths of the two elements in fW resp. in W , and the defect is given by def Gb = rkF G − rkF Jb where rkF denotes the rank of a maximal F -split torus and where Jb is the reductive group over F with Jb(F ) = {g ∈ G(L) g−1bσ(g) = b}. Previous work on affine Deligne-Lusztig varieties in affine flag varieties indicates that the theory is more accessible and more is known if one assumes that the alcove xa is sufficiently far from the walls of the Weyl chambers in the Bruhat-Tits building. To make such an assumption precise, the shrunken Weyl chambers are defined as the set of x such that Uα ∩ xI 6= Uα ∩ I for all roots α. An element x is called regular shrunken if it is shrunken and in addition x can be written as v1tµv2 with v1, v2 ∈ W and hα, µi > 0 for all α ∈ Φ+. 3. The arithmetic case By [He1], Theorem 6.1, Corollary 6.2, and Section 6.2, the condition Xx(b) 6= ∅ and also dim Xx(b) are expressed in terms of certain class polynomials and thus only depend on the combinatorial datum (fW ∼= X∗(T )⋊W, δ, x, [b]), and are independent of the group G itself. Here we use that [b] can be given by a representative y in fW . Claim. dx(b), and thus ∆x(b) can be expressed in terms of this combinatorial datum. This is clear for ℓ(x) and ℓ(η(x)). If we assume [b] to be given by a representative n yδ(y) · · · δn−1(y) where n is chosen such that yδ(y) · · · δn−1(y) ∈ X∗(T ). Finally, by [K2] (1.9.1), def G(b) = dim a − dim awb where wb ∈ W is the image of y in the finite Weyl group. This proves the claim. y infW , then νb = 1 Let us now explain the analog of our main result in the arithmetic case. Let F be a local field of mixed characteristic, and let all other data be as before (w.r.t. our new choice of F ). Then the affine Deligne-Lusztig varieties are defined as locally closed subschemes of Zhu's Witt-vector affine Grassmannian, compare [Z]. In particular, there is a meaningful notion of dimension for them. We claim that then Theorem 1.1 also holds in this context. Indeed, in the same way as above one proves that also in the arithmetic case the statement of Theorem 1.1 is an assertion that only depends on the combinatorial datum (fW ∼= X∗(T ) ⋊ W, δ, x, y) where y is a representative of [b] in fW . But the combinatorial datum associated with the problem in the arithmetic case is also represented by a group G over a local function field, and [y] ∈ B(G) for the function field case. Thus Theorem 1.1 for the function field case implies the same assertion for the arithmetic case. Similar considerations apply to other assertions concerning non-emptiness and dimension of affine Deligne-Lusztig varieties, for example to Theorem 7.3. The arithmetic version of Theorem 1.1 in its turn can be applied to the reduction modulo p of Shimura varieties with Iwahori level structure. Here, it implies that for every KR stratum in the reduction there is a unique minimal Newton stratum intersecting it non-trivially. 4. Normalized (J, w, δ)-alcoves Construction 4.1. Assume that xa is a (J, w, δ)-alcove. Notice that in general, wMJ (and hence IM ) need not be σ-invariant if w 6= δ(w). However, σ-conjugation 5 with w−1 gives an isomorphism IM xσ(IM ) = (wMJ w−1 ∩ I)x(σ(w)MJ σ(w)−1 ∩ I) −→ (MJ ∩ w−1 I)(w−1xδ(w))(MJ ∩ δ(w−1)I) that also preserves the Newton stratifications. By definition of (J, w, δ)-alcoves, xa is a (J, w, δ)-alcove if and only if it is a (J, wm, δ)-alcove for any m ∈ WJ . Replacing w by the minimal length representa- tive of wWJ we obtain MJ ∩ w−1 I = IJ = MJ ∩ I. Then the above isomorphism reads IM xσ(IM ) ∼=−→ IJ (w−1xδ(w))IJ . Notice however that we still have that w−1 I 6= I in general. Definition 4.2. For x ∈ fW , we call xa a normalized (J, w, δ)-alcove if it is a (J, w, δ)-alcove such that w is of minimal length in wWJ , or equivalently such that MJ ∩ w−1 I = IJ . Then we also call x a (J, w, δ)-alcove element. In this case let x = w−1xδ(w). lemma. for MJ for any δ-stable proper subset J ′ ⊂ J and w′ ∈ WJ . S such that there is a w ∈ W such that xa is a (J, w, δ)-alcove. Let w be such that Lemma 4.3. Let x ∈ fW and let J be a minimal element in the set of subsets of xa is a normalized (J, w, δ)-alcove. Then x ∈ fWJ is not a (J ′, w′, δ)-alcove element Construction 4.1 shows that for every x ∈ fW , there exists a pair (J, w) as in the Proof. Assume that there is a proper subset J ′ ( J such that x ∈ fWJ is a (w′)−1 xδ(w′) ∈ fWJ ′ . We have to verify that for every α ∈ (ww′)(Φ+ −Φ+ (J ′, w′, δ)-alcove element for some w′ ∈ WJ . We want to show that then xa is a (J ′, ww′, δ)-alcove. By the first property of (J ′, w′, δ)-alcoves, (ww′)−1xδ(ww′) = J ′ ) we have J ) = w(Φ+−Φ+ Uα∩xI ⊆ Uα∩I. We first consider the case that α ∈ (ww′)(Φ+−Φ+ J ), where the equality follows from w′ ∈ WJ . In this case the property holds since xa is a (J, w, δ)-alcove. Let now α ∈ (ww′)(Φ+ J ′ ). Conjugating by w the desired inclusion Uα ∩ xI ⊆ Uα ∩ I is equivalent to J − Φ+ (3) Uw−1α ∩ w−1xI ⊆ Uw−1α ∩ w−1 I. J ) ⊆ ΦJ . Since x is normalized, we have MJ ∩ w−1 We have w−1α ∈ w′(Φ+ I = IJ . Applying σ to this equality and using that the right hand side and MJ are by definition σ-invariant, we also obtain MJ ∩ δ(w)−1 I = IJ . By the first equality the right hand side of (3) is equal to Uw−1α ∩ I. The left hand side is equal to Uw−1α ∩ xδ(w−1)I. From x ∈ fWJ we obtain that this is equal to Uw−1α ∩ xδ(w−1)IJ = Uw−1α ∩ x(δ(w−1)I ∩ MJ ) = Uw−1α ∩ xIJ . Since w−1α ∈ w′(Φ+ J ′ ), (3) is thus nothing but Uw−1α ∩ xI ⊆ Uw−1α ∩ I, the second property of x being a (J ′, w′, δ)- alcove element. (cid:3) J − Φ+ 5. Comparing sets of σ-conjugacy classes If G is as above and H a subgroup, then the inclusion ι induces a natural map B(ι) : B(H) → B(G). In general, and even if we assume that H is the Levi component of a parabolic subgroup of G, this map is neither injective nor surjective, as can be seen in the following example. 6 Example 5.1. Let G = GL2 and H the diagonal torus. Consider b1 =(cid:18) 1 0 0 t (cid:19) , b2 =(cid:18) 0 t 1 0 (cid:19) , b3 =(cid:18) t 0 1 (cid:19) 0 Then b1, b3 ∈ H(L) with [b1]H 6= [b3]H because their images in π1(H) ∼= Z2 are different, but b1 = sb3σ(s)−1 where s ∈ G(L) is the morphism exchanging the two basis elements. Thus [b1]G = [b3]G and B(ι) is not injective. Furthermore, [b2]G does not have a representative in H(L) because its Newton point is (1/2, 1/2) ∈ Q2, and elements of the split torus H have integral Newton points. Thus B(ι) is also not surjective. However, if [b]G ∈ B(G) is basic and H is a standard Levi subgroup of G defined over F , then the fiber B(ι)−1([b]G) has at most one element by [GHN], Proposition 3.5.1. Notice that only this weaker statement is in fact shown in loc. cit., compare also their erratum. More can also be said if we require the classes to be comparable for the partial order. Since it seems to be of independent interest, we formulate the following theorem in greater generality than what is needed for the present purpose. Theorem 5.2. Let F be a local field and let H ⊆ G be quasi-split reductive groups over F . Let [b]H (cid:22) [b′]H ∈ B(H), and let [b]G, [b′]G be the corresponding classes in B(G). (1) [b]G (cid:22) [b′]G. (2) If moreover [b]H 6= [b′]H , then also [b]G 6= [b′]G. Proof. We choose maximal F -tori TH ⊂ T of H and G and Borel subgroups BH ⊂ B containing them, all defined over F . From κH (b) = κH (b′) and functoriality of the Kottwitz map we obtain κG(b) = κG(b′). For (1) it is thus enough to show the following statement: Let ν, ν ′ ∈ X∗(T )Q be H-dominant and ν′ − ν a non-negative rational linear combination of positive coroots of H. Let νdom, ν ′ dom be the G-dominant representatives in the Weyl group orbits of ν, ν ′. Then ν ′ dom −νdom is a non-negative rational linear combination of positive coroots of G. For (2) we have to show an analogous statement replacing non-negative by non-trivial non-negative, by which we mean that all coefficients of the positive coroots are non-negative and at least one coefficient is non-zero. Mul- tiplying by a suitable positive integer we may assume that ν, ν ′ ∈ X∗(T )H−dom are integral and that the above difference is a non-negative integral linear combination of coroots of H. Since these assertions do not depend on any Galois action any more, we may pass to an algebraic closure of F and may thus assume that H and G are split. Let K be a hyperspecial maximal compact subgroup of G and KH = K ∩ H. The condition that ν ′ − ν a non-negative rational linear combination of positive coroots of H is equivalent to KH ν(t)KH ⊂ KHν ′(t)KH . In particular, Kνdom(t)K∩ Kν ′ dom(t)K is invariant under this action, this implies Kνdom(t)K ⊂ Kν ′ dom(t)K. This last condition is again equivalent to the condition that ν ′ dom − νdom is a non-negative integral linear combination of positive coroots of G, which implies (1). dom(t)K 6= ∅. Since Kνdom(t)K is an orbit for the K ×K-action and Kν ′ For (2) we have to show that νdom 6= ν ′ dom for the G-dominant representatives of the Weyl group orbits of ν and ν ′. Claim. There is a Borel subgroup B′ of G with B′ ∩ H = BH and such that ν is G-dominant with respect to B′. 7 To prove the claim let M ′ be the centralizer of ν, a Levi subgroup containing T . Let W M ′ be the subset of elements of W that are shortest representatives of their WM ′ -cosets. Let w ∈ W M ′ be such that w(ν) is dominant with respect to B. Let B′ = w−1Bw. Then ν is dominant with respect to B′. It remains to consider the intersection with H. Let α ∈ Φ+ H , the set of roots in BH , or equivalently in B ∩ H = B′ ∩ H. Assume first that hα, νi > 0. Then since ν is also dominant with respect to B′, the root α lies in B′. Let now α ∈ Φ+ H with hα, νi = 0. Then α is a positive root in M ′ ∩ H. Since w ∈ W M ′ , the conjugate w(α) is again a positive root, hence in B. In particular, α is a root in B′, which finishes the proof of the claim. We replace B by B′ and may thus assume that ν is dominant. As usual let ρ denote the half-sum of the positive roots of G, which agrees with the sum of all fundamental weights. Recall that ν′ − ν 6= 0 is a non-trivial non-negative linear combination of positive coroots of H and thus also of G. Using the second interpretation for ρ we see that hρ, ν ′i > hρ, νi. Using the first interpretation we obtain that h2ρ, ν ′ domi = Xα∈Φ+ = Xα∈Φ+ 1 hα, ν ′ domi = 2 Xα∈Φ hα, ν ′i ≥ Xα∈Φ+ hα, ν ′ domi = 1 2 Xα∈Φ hα, ν ′i = h2ρ, ν ′i. hα, ν ′i (cid:3) Altogether we find hρ, ν ′ domi > hρ, νi, thus in particular ν ′ dom 6= ν. From [GHN], Theorem 3.3.1 we obtain Corollary 5.3. Let x ∈ fW be a normalized (J, w, δ)-alcove element. Then every element of IxI is I-σ-conjugate to an element of IM xσ(IM ), which is in turn w−1- σ-conjugate to IJ xIJ . In particular, the restriction of B(ι) composed with w-σ- conjugation induces a surjection B(MJ )x → B(G)x. (cid:3) Example 5.4. The map of the preceeding corollary is still not injective in general. For example let G = GL4 and M ∼= GL2 × GL2 the subgroup of block diagonal matrices of size (2, 2). We choose B to be the Borel subgroup of lower triangular matrices. Let x ∈ fWJ be such that the image in the affine Weyl group of each factor GL2 is of the form x0 = t(1,0)s where s is the non-trivial element in WGL2 = S2. Let µ = (0, 1). Then x0 ∈ WGL2tµWGL2 implies that every [b] ∈ B(GL2)x0 satisfies [b] ≤ [µ]. We have x0δ(x0) = t(1,1), hence νx0 = ( 1 2 ). Since µ ≤ x0, the formula for the generic σ-conjugacy class in Ix0I in [V2] shows that it is equal to [µ]. Thus B(GL2)x0 = {[x0], [µ]}. Let w = (14) ∈ WG. Then one can check that the corresponding element x = wxδ(w−1) is a (J = (s1, s3), w, δ)-alcove element with B(M )x consisting of all four pairs of elements of B(GL2)x0 , and B(G)x consisting of the three elements that are obtained by identifying ([x0], [µ]) and ([µ], [x0]). 2 , 1 6. The basic locus and the proof of Theorem 1.1(1) Let x ∈fW . Then there is a unique basic element [b0,x] ∈ B(G) with κG(b0,x) = κG(x). It agrees with the unique minimal element of B(G) mapping to κG(x) under κG. If Nx,[b0,x] 6= ∅, it is therefore the unique minimal element of B(G)x. By Theorem A of [GHN], Nx,[b0,x] = ∅ if and only if there is a pair (J, w) such that xa is a (J, w, δ)-alcove and κMJ (w−1xδ(w)) /∈ κMJ ([b0,x] ∩ MJ (L)). In [He1], Corollary 12.2 (see also [He2], Theorem 5.3 and the corresponding footnote for the generalization to non-split G), He proves that if x is in the shrunken 8 Weyl chambers and Nx,[b0,x] 6= ∅, then (4) dim Xx(b0,x) = dx(b0,x). Thus if [b0,x] ∈ B(G)x, all assertions of Theorem 1.1 are shown. We now return to the general case. Proof of Theorem 1.1(1). We choose J, w as in the theorem. We may further choose w such that xa is a normalized (J, w, δ)-alcove, compare Construction 4.1. By Corollary 5.3, B(G)x is the image of B(MJ )x in B(G). By Lemma 4.3 and the results we just recalled, we obtain that B(MJ )x has a unique minimal element for the partial order on B(MJ ), which is the unique basic element [b0,x] of B(MJ ) with κMJ ([b0,x]) = κMJ (x). By Theorem 5.2 its image in B(G) is the unique minimal element of B(G)x. (cid:3) 7. Dimension its G-dominant Newton point. Recall that Let [b] ∈ B(G)x, and νG b := dG x (b) − dim Xx(b) 1 (ℓ(x) + ℓ(η(x)) − def Gb) − hρ, νG 2 x (b) ∆G = (5) b i − dim Xx(b). Proposition 7.1. Let xa be a normalized (J, w, δ)-alcove. (1) Let b ∈ IJ xIJ and denote its MJ -dominant Newton point by νb. Then ℓG(x) − ℓMJ (x) = h2(ρG − ρMJ ), νbi. (2) If b ∈ IJ xIJ is basic in MJ then νb is G-dominant. Notice that in general νb need not be G-dominant, compare Example 5.4. Proof. By the defining property of (J, w, δ)-alcoves we have that ℓG(x) = dim I/I ∩ xIx−1 = Xα∈w(ΦJ ) dim (Uα ∩ I)/(Uα ∩ I ∩ xI) + Xα∈w(Φ+−Φ+ J ) dim (Uα ∩ I/Uα ∩ xI). We denote the two sums by S1 and S2, and first consider S1. Since xa is a normalized (J, w, δ)-alcove, we have MJ ∩I = MJ ∩ w−1 first description, this group is σ-invariant, thus we also have MJ ∩I = MJ ∩ δ(w)−1 We conjugate this equality by x = w−1xδ(w) ∈ MJ , and obtain I. Using the I. MJ ∩ xI = x(MJ ∩ I) = x(MJ ∩ δ(w)−1 I) = MJ ∩ xδ(w)−1 I = MJ ∩ w−1xI. We have S1 = dim (wMJ ∩ I)/(wMJ ∩ I ∩ xI) = dim (MJ ∩ w−1 I)/(MJ ∩ w−1 I ∩ w−1xI) and using the above equalities this is = dim (MJ ∩ I)/(dim MJ ∩ I ∩ xI) = ℓMJ (x). For the first assertion it remains to show that S2 = h2(ρG − ρMJ ), νbi. Claim. Let µ ∈ X∗(T )MJ −dom with x ∈ WJ tµWJ . Then h2(ρG − ρMJ ), νbi = h2(ρG − ρMJ ), µi. From b ∈ MJ (OL)µ(t)MJ (OL) we obtain νb (cid:22)MJ µ, i.e. µ − νb is a linear combi- nation of coroots of T in MJ (in fact a non-negative linear combination of positive coroots). Let α∨ be a coroot of T in MJ . Then hρG, α∨i = 1 and analogously for ρMJ . The claim follows. 9 By definition of S2 we have S2 = Xα∈w(Φ+−Φ+ J ) dim (Uα ∩ I/Uα ∩ xI), and conjugating by w−1 this is = Xα∈Φ+−Φ+ J dim (Uα ∩ w−1 I/Uα ∩ w−1xI) = dim NJ ∩ w−1 I − dim NJ ∩ w−1xI. Since NJ , I, I ∩ NJ , and hence also the dimension of subsets of NJ are invariant under σ, this is = dim NJ ∩ δ(w−1)I − dim NJ ∩ w−1xδ(w)δ(w−1)I = dim NJ ∩ δ(w−1)I − dim NJ ∩ xδ(w−1)I. w2µ with w1, w2 ∈ WJ and µ ∈ X∗(T )MJ −dom. We use We now decompose x = w1 that NJ = w1NJ , and furthermore that NJ ∩ I (and hence the notion of dimensions for subsets of NJ ) is invariant under conjugation by w1. We obtain = dim NJ ∩ δ(w−1)I − dim NJ ∩ w2 µδ(w−1)I = Xα∈Φ+−Φ+ J hα, w2 µi. The element w2 ∈ WJ induces a permutation of Φ+ − Φ+ J , thus this is = Xα∈Φ+−Φ+ J hα, µi. = h2(ρG − ρMJ ), µi. Together with the claim this finishes the proof of the first assertion. For the second assertion we assume that νb is central in MJ . Let α0 ∈ S − J and let J ∪ {δi(α0) i ∈ Z} = J ′ ⊂ S and M ′ ⊃ MJ the corresponding Levi subgroup of G. We have to show that hα0, νbi ≥ 0. Since hα, νMJ i = 0 for all roots α in MJ and since νb is δ-invariant, this is equivalent to hρM ′ − ρMJ , νMJ i ≥ 0. By an analog of the calculation of S2 above this expression is equal to dim (Uα ∩ I) − dim (Uα ∩ xI). b b Xα∈w(Φ+ J ′ −Φ+ J ) This in its turn is non-negative by the assumption that xa is a (J, w, δ)-alcove. (cid:3) Lemma 7.2. Let xa be a normalized (J, w, δ)-alcove in the regular shrunken Weyl chambers. Then ℓG(ηG(x)) = ℓMJ (ηMJ (x)). Proof. We want to show the two equalities (6) ℓG(ηG(x)) = ℓG(ηG(x)) = ℓMJ (ηMJ (x)). We begin with the first. Our assumption on x to be regular implies that there is a unique decomposition x = v1tµv2 with v1, v2 ∈ W = µW and µ dominant, and hence that ηG(x) = σ−1(v2)v1. Further, the same consideration implies that ηG(w−1xσ(w)) = σ−1(v2σ(w))w−1v1 = ηG(x), implying the first equality. For the 10 second equality notice that the length function on WMJ is the restriction of the length function for WG. Thus it is enough to show that ηG(x) = ηMJ (x). Recall that x = w−1xσ(w) ∈ fWJ . The elements ηG(x) resp. ηMJ (x) are computed via the unique decompositions x = (w−1v1)tµ(v2σ(w)) and x = w1tµJ w2 where w−1v1, v2σ(w) ∈ W = µW , µ is G-dominant, wi ∈ WJ and µJ is MJ -dominant. We want to show that already these two decompositions coincide, i.e. that µJ is G-dominant. For this it is enough to prove that if we write x = tµ w with w ∈ WJ , then hβ, µi ≥ 0 for all β ∈ Φ+ − Φ+ J . Conjugation by x (or rather by its image w in the finite Weyl group) induces a J . Let α = w(β) ∈ Φ. Since x is a (J, w, δ)-alcove element permutation of Φ+ − Φ+ and shrunken, we obtain Uα ∩ xI ( Uα ∩ I. Conjugation by w−1 yields (7) Uβ ∩ xδ(w−1)I ( Uβ ∩ w−1 I. Normalizing the Uγ such that Uγ ∩ I ⊆ Uγ(OL) with equality if and only if γ is negative, we obtain that the right hand side is contained in Uβ(OL). By the strict inclusion, the left hand side is contained in Uβ(tOL). We write Uβ ∩ xδ(w−1)I = µ(t)(Uβ ∩ wδ(w−1)I). We have Uβ ∩ wδ(w−1)I ⊇ Uβ(tOL). Hence by (7), hβ, µi ≥ 0 which is what we wanted. (cid:3) Altogether we have Theorem 7.3. Let xa be a (J, w, δ)-alcove in the regular shrunken Weyl chambers. Let [b]G ∈ B(G)x. Then for every [b]MJ ∈ B(MJ )x mapping to [b]G and such that νMJ b = νG Proof. We have b we have ∆G x (b) = ∆MJ x (b). • Proposition 7.1 to compare the first and fourth summand in (5) to the analogous summands for MJ instead of G. The proposition gives the de- sired comparison under the additional assumption that the MJ-dominant Newton point νb of b is equal to its G-dominant Newton point νG b (which by (2) holds if νb is basic in MJ ). • ℓG(ηG(x)) = ℓMJ (ηMJ (w−1xσ(w))) for x regular shrunken by Lemma 7.2. • defG(b) = def MJ (b) for all b and all x, by definition of the defect. • dim (X G (b)) by [GHKR], Theorem 1.1.4(b). x (b)) = dim (X MJ x Together with (5), this implies the theorem. (cid:3) Proof of Theorem 1.1. We apply the above theorem to the case where J, w, and [b] are chosen such that [b]MJ is basic in MJ . By (4) we have ∆MJ x (b) = 0. The theorem follows. (cid:3) Appendix - Comparing closure relations of Newton strata The aim of this appendix is to complement the other results of the paper by proving the following more general geometric version of Corollary 5.3. It gives a direct means to compare closures of Newton strata for Iwahori double cosets for G and a Levi subgroup. We use the same notation as in the other sections. Recall from Construction 4.1 that if x is normalized, then IM xIM = wIJ xIJ σ(w−1). Theorem 8.4. Let x ∈ fW such that xa is a normalized (J, w, δ)-alcove. Let b, b′ ∈ IJ xIJ . Then b ∈ [b′]G ∩ IJ xIJ if and only if wbσ(w−1) ∈ [b′]G ∩ IxI where the closure is taken in the respective Schubert cell. More precisely, assume that wbσ(w−1) ∈ IxI is a specialization of a point bη ∈ ([b′]G ∩ IxI)(κ(bη)) defined over 11 some field κ(bη), and let C be the I-σ-conjugacy class of bη. Then the closure of C ∩ IM xIM contains wbσ(w−1). Our proof is a refinement of the proof of [GHN], Theorem 3.3.1, or [GHKR], Theorem 2.1.2. A technical ingredient is the following lemma. Lemma 8.5. Let (R, m) be a complete discrete valuation ring with algebraically closed residue field. Let M ∈ Mn×n(R) and v ∈ Rn. (1) There is a w ∈ Rn with w − M σ(w) = v. (2) There is a ring (R′, m′) satisfying the same assumptions as R, a surjective finite flat morphism Spec R′ → Spec R and a w ∈ (R′)n with σ(w) − M w = v. (3) If in addition v ≡ 0 (mod m) in (1) or (2), then we may choose w ≡ 0 modulo m resp. m ′. Proof. For the first statement, [GHKR], Lemma 5.1.1 implies that the equation has a solution modulo m. Then we use induction on i to show that the solution can be lifted to a solution modulo mi. Indeed, if wi ∈ An(R) is a solution modulo mi, then for d ≡ 0 (mod mi) we have (wi + d) − M σ(wi + d) ≡ (wi + d) − M σ(wi) (mod mqi), thus wqi = wi + (v − wi + M σ(wi)) is a solution modulo mqi. Since R is complete, this proves (1), and also (3) for this case. For (2) let R = R[w1, . . . , wn]/(wq − M w − v), a finite integral extension of R. Let p be any minimal prime ideal of R. Then p ∩ R = {0}, hence R ֒→ R/p is a finite integral extension, and R/p is again an integral domain. Let R′ be the integral closure of R/p in its fraction field (which is a finite field extension of QuotR). Then since R is a complete discrete valuation ring, R′ has all claimed properties of (2). For (3) we choose p to be contained in (m, w1, . . . , wn) which is a maximal ideal since v ∈ m. (cid:3) Proof of Theorem 8.4. By σ-conjugation with w, the condition b ∈ [b′]MJ ∩ IJ xIJ is equivalent to wbσ(w−1) ∈ w[b′]MJ σ(w−1) ∩ IM xσ(IM ). Since the latter set is contained in [b′]G ∩ IxI, one implication follows. The I-σ-conjugacy class C is admissible in G (i.e. invariant under right mul- tiplication by some open subgroup of I, see for example [VW], Prop. 2.3), and IM xσ(IM ) and the IM -σ-conjugacy class of wbσ(w−1) are admissible in wMJ = M . Thus for the other direction it is enough to show that for all elements I ′ of a fun- damental system of neighborhoods of 1 consisting of subgroups of I, the image of c := wbσ(w−1) in G(L)/I ′ is contained in the closure of the intersection of [b′]GI ′/I ′ and IM xσ(IM )I ′/I ′. Furthermore, the assertion holds for wbσ(w−1) if and only if it holds for some IM -σ-conjugate of wbσ(w−1). Thus we may assume that wbσ(w−1) ∈ IM x. By our assumption there is an element b ∈ (IxI)(R) ⊆ LG(R) for some complete discrete valuation ring (R, m) with unformizer ε and algebraically closed residue field k such that its generic point bη is in [b′]G ∩ IxI and such that its special point satisfies b0 = wbσ(w−1). By [HV], Remark 5.7 we can decompose b = i1xi2 with i1, i2 ∈ I(R). Here, i2 is unique up to left multiplication by elements of I ∩ x−1Ix. For the reductions modulo m we have ¯i1x¯i2 ∈ IM (k)x, thus ¯i2 ∈ (I ∩x−1Ix)(k). We lift ¯i2 to some element d ∈ (I ∩x−1Ix)(R) and replace i2 by d−1i2 and i1 by i1xdx−1. Thus we may assume that i2 ≡ 1 (mod m). Replacing R by σ−1(R) = R(ε1/q), we have σ−1(i2) ∈ I(R). We σ-conjugate b by σ−1(i2) (without modifying b0) and may thus assume that i2 = 1. Hence it is enough to prove the following claim. 12 Claim. There is a fundamental system of neighborhoods of 1 consisting of sub- groups I ′ of I with the following property. For every I ′ there is a complete discrete valuation ring R′ with algebraically closed residue field k, a surjective finite flat morphism Spec R′ → Spec R, and an element i ∈ I(R′) with i ≡ 1 (mod mR′) and ibσ(i−1) ∈ IM xI ′/I ′. For the proof of this claim we follow the proof of the surjectivity statement of [GHN], Theorem 3.3.1, which claims a similar assertion for R replaced by an algebraically closed field (so that in their case no extension is needed). For n ∈ N let T (L)n be the corresponding congruence subgroup as in [PR], 2.6. For r ≥ 0 let Ir be the subgroup of I generated by T (L)n for n ≥ r and all affine root subgroup schemes Hα+m with α ∈ Φ and m ≥ r such that α + m is a positive affine root. Let Ir+ =Ss>r Is. Then for all r > 0, we obtain normal subgroups Ir and Ir+ of I. Let M = wMJ , let N be the subgroup of G generated by the root subgroups for roots in w(Φ+ − Φ+ J ) and let N be the subgroup of G generated by the root subgroups for roots in w(Φ− −Φ− J ). By intersecting with Ir resp. with Ir+ we obtain normal subgroups Nr, N r resp. Nr+ , N r+ . The condition that xa is a (J, w, δ)-alcove then implies that xσNr ⊆ Nr and xσN r ⊇ N r. Lemma 8.6. Let (R, m) be a complete discrete valuation ring with algebraically closed residue field k. Let m ∈ IM (R) and r ≥ 0. (1) For every i+ ∈ Nr(R) there is a b+ ∈ Nr(R) with b+i+ + ∈ Nr+(R). (2) For every i− ∈ σ(N r(R)) there is a ring (R′, m′) satisfying the same as- sumptions as R, a surjective finite flat morphism Spec R′ → Spec R and an element b− ∈ N r(R′) such that (mx)−1 − ∈ σ(N r+ ). σb−1 mxσb−1 b−i− (3) If i− resp. i+ are congruent to 1 modulo m, then b− resp. b+ can be chosen to be congruent to 1 modulo m resp. m′. Proof. The proof of (1) and (2) is the same as that of [GHN], Lemma 3.4.1, replacing references to [GHKR], Lemma 5.1.1 by Lemma 8.5 above. Assertion (3) follows in the same way using (3) of the lemma. (cid:3) We return to the proof of the theorem. We consider a generic Moy-Prasad filtration as explained in [GHKR], Section 6. This yields a filtration I =Sr≥0 I[r] with I[r] ⊃ I[s] for r < s such that each I[r] is normal in I and a semidirect product of I[r+] =Ss>r I[s] and some Ihri which is either an affine root subgroup or contained in T (L)0. In particular, we also get a corresponding decomposition for every R-valued point of I[r]. By the same argument as in [GHKR], Section 6 one shows that for every i, there is an extension R′ of R as in the claim, and an h ∈ (δ−1(x−1)I ∩ I)(R′) with h ≡ 1 (mod m′) such that hbσ(h−1) ∈ I[i+]IM x. Since IM x is bounded, the subgroups x−1IM I[i+]IM x form a fundamental system of neighborhoods of 1 as in the claim. (cid:3) References [B] [BT] E. Beazley, Codimensions of Newton strata for SL(3) in the Iwahori case, Math. Z. 263 (2009), 499 -- 540. F. Bruhat, J. Tits, Groupes r´eductifs sur un corps local : I. Donn´ees radicielles valu´ees, Publ. Math. Inst. Hautes ´Etudes Sci., 41 (1972), 5 -- 251. [GHKR] U. Gortz, T. Haines, R. Kottwitz, D. Reuman, Affine Deligne-Lusztig varieties in affine flag varieties, Compos. Math. 146 (2010), 1339-1382. 13 [GH] U. Gortz, X. He, Dimensions of affine Deligne-Lusztig varieties in affine flag varieties, Doc. Math. 15 (2010), 1009 -- 1028. [GHN] U. Gortz, X. He, S. Nie, P -alcoves and nonemptiness of affine Deligne-Lusztig varieties, [Ha1] [Ha2] [HV] [He1] [He2] [He3] [K1] [K2] [M] [MV] [PR] [RR] [V1] [V2] [VW] [Z] Ann. Sci. ´Ec. Norm. Sup´er. (4) 48 (2015), 647 -- 665. P. Hamacher, The dimension of affine Deligne-Lusztig varieties in the affine Grassman- nian of unramified groups, Int. Math. Res. Not. IMRN 2015, 12804 -- 12839. P. Hamacher, The almost product structure of Newton strata in the Deformation space of a Barsotti-Tate group with crystalline Tate tensors, Math. Z. 287 (2017), 1255 -- 1277. P. Hamacher, E. Viehmann, Irreducible components of minuscule affine Deligne-Lusztig varieties, Algebra Number Theory 12 (2018), 1611 -- 1634. X. He, Geometric and homological properties of affine Deligne-Lusztig varieties, Ann. of Math. (2) 179 (2014), 367 -- 404. X. He, Note on affine Deligne-Lusztig varieties, preprint, arXiv:1309.0075. X. He, Hecke algebras and p-adic groups, Current developments in mathematics 2015, 73 -- 135, Int. Press, Somerville, MA, 2016. R. E. Kottwitz, Isocrystals with additional structure, Compos. Math. 56 (1985), 201 -- 220. R. E. Kottwitz, Dimensions of Newton strata in the adjoint quotient of reductive groups, Pure Appl. Math. Q. 2 (2006) (Special Issue: In honor of Robert MacPherson, Part 1 of 3), 817 -- 836. E. Milicevic, Maximal Newton points and the quantum Bruhat graph, preprint, 2016, arXiv:1606.07478. E. Milicevic, E. Viehmann, Generic Newton points and the Newton poset in Iwahori double cosets, preprint, 2019, arXiv:1902.02415. G. Prasad, M. Raghunathan, Topological central extensions of semisimple groups over local fields, Ann. of Math. (2) 119 (1984), 143 -- 201. M. Rapoport, M. Richartz, On the classification and specialization of F -isocrystals with additional structure, Compos. Math. 103 (1996), 153 -- 181. E. Viehmann, Newton strata in the loop group of a reductive group, Amer. J. Math. 135 (2013), 499 -- 518. E. Viehmann, Truncations of level 1 of elements in the loop group of a reductive group, Ann. of Math. (2) 179 (2014), 1009 -- 1040. E. Viehmann, H. Wu, Central leaves in loop groups, Math. Res. Lett. 25 (2018), 989 -- 1008. X. Zhu, Affine Grassmannians and the geometric Satake in mixed characteristic, Ann. of Math. (2) 185 (2017) 403 -- 492. Technische Universitat Munchen, Fakultat fur Mathematik - M11, Boltzmannstr. 3, 85748 Garching bei Munchen, Germany E-mail address: [email protected] 14
1301.7633
1
1301
2013-01-31T14:54:11
Seshadri constants and degrees of defining polynomials
[ "math.AG" ]
In this paper, we study a relation between Seshadri constants and degrees of defining polynomials. In particular, we compute the Seshadri constants on Fano varieties obtained as complete intersections in rational homogeneous spaces of Picard number one.
math.AG
math
SESHADRI CONSTANTS AND DEGREES OF DEFINING POLYNOMIALS ATSUSHI ITO AND MAKOTO MIURA Dedicated to Professor Yujiro Kawamata on the occasion of his sixtieth birthday. Abstract. In this paper, we study a relation between Seshadri con- stants and degrees of defining polynomials. In particular, we compute the Seshadri constants on Fano varieties obtained as complete intersec- tions in rational homogeneous spaces of Picard number one. 1. Introduction Seshadri constant was introduced by Demailly in [Dem], as an invariant which measures the local positivity of ample line bundles. Definition 1.1. Let L be an ample line bundle on a projective variety X, and take a (possibly singular) closed point p ∈ X. The Seshadri constant of L at p is defined to be ε(X, L; p) := max{ t ≥ 0 µ∗L − tE is nef }, where µ : eX → X is the blowing up at p and E = µ−1(p) is the exceptional Equivalently, the Seshadri constant can be also defined as divisor. ε(X, L; p) = inf C (cid:26) C.L multp(C)(cid:27) , where the infimum is taken over all reduced and irreducible curves C on X passing through p. We call a curve C a Seshadri curve of L at p if ε(X, L; p) = C.L/ multp(C). Seshadri constants have many interesting properties (see [La, Chapter 5] for instance), but it is very difficult to compute them in general. Many authors study surface cases, but computations in higher dimensions are rare. Let X be a projective variety embedded in PN and p ∈ X. The pur- pose of this paper is to study the Seshadri constant ε(X, OX (1); p) by in- vestigating homogeneous polynomials which define X. It is easy to see that ε(X, OX (1); p) ≥ 1 for such X and p. Furthermore it is known that 2010 Mathematics Subject Classification. 14C20. Key words and phrases. Seshadri constant, defining polynomial, rational homogeneous space. 1 SESHADRI CONSTANTS AND DEGREES OF DEFINING POLYNOMIALS 2 ε(X, OX (1); p) = 1 holds if and only if there exists a line on X passing through p (cf. [Ch, Lemma 2.2]). In [Ba] and [Ch], Bauer and Chan give a lower bound of ε(X, OX (1); p) by using the degree deg(X) := OX(1)dim X when X is a surface and a 3-fold respectively. Theorem 1.2 (cf. [Ba, Theorem 2.1], [Ch, Theorem 1.4]). Let X be a smooth projective surface or a 3-fold in PN , and p ∈ X a point. If there exists no line on X passing through p, it holds that ε(X, OX (1); p) ≥ deg(X) deg(X) − 1 . Remark 1.3. Bauer proved the sharpness of the lower bound as well, that is, for each d ≥ 3, there exist a smooth surface X and p ∈ X such that d = deg(X) and ε(X, OX (1); p) = d/(d − 1). Chan also constructed such X and p in 3-dimensional case for d ≥ 4, although X might have finitely many singular points. Instead of the degree, we introduce dp(X) for a projective variety X ⊂ PN and p ∈ X (cf. [La, Definition 1.8.37]). Definition 1.4. Let X be a projective variety in PN and p ∈ X a point. We define dp(X) to be the least integer d such that the natural map H 0(PN , IX ⊗ OPN (d)) ⊗ OPN → IX ⊗ OPN (d) is surjective at p, where IX ⊂ OPN is the ideal sheaf corresponding to X. In other words, X is cut out scheme theoretically by hypersurfaces of degree dp(X) at p. By using dp(X), we give a lower bound of the Seshadri constant on X ⊂ PN (which may be singular) in any dimensions. Theorem 1.5. Let X be a projective variety in PN and p ∈ X a point. If there exists no line on X passing through p, it holds that ε(X, OX (1); p) ≥ dp(X) dp(X) − 1 . Furthermore, this lower bound is sharp, i.e., for any n ≥ 1 and d ≥ 2, there exist a smooth projective variety X ⊂ PN and p ∈ X such that n = dim X, d = dp(X), and ε(X, OX (1); p) = d/(d − 1). Remark 1.6. It holds dp(X) ≤ deg(X) for any projective variety X ⊂ PN and p ∈ X (see the proof of [Mu, Theorem 1]). Thus Theorem 1.5 improves Theorem 1.2 even in the cases dim X = 2, 3. In Section 3, we compute the Seshadri constants on some varieties X ⊂ PN by finding a curve C such that deg(C)/ multp(C) coincides with the lower bound dp(X)/(dp(X) − 1). As a special case, we obtain the following theorem. SESHADRI CONSTANTS AND DEGREES OF DEFINING POLYNOMIALS 3 Theorem 1.7 (=Corollary 3.4). Let Y ⊂ PN be a rational homogeneous space of Picard number 1, which is embedded by the ample generator. Let X be a complete intersection variety in Y of hypersurfaces of degrees d1 ≤ . . . ≤ dr such that −KX = OX(1) and p ∈ X. If there exists no line on X passing through p, it holds that ε(X, OX (1); p) = dp(X) dp(X) − 1 =(cid:26) dr/(dr − 1) when dr ≥ 2, when dr = 1. 2 Remark 1.8. If X is a complete intersection variety in Y of hypersurfaces of degrees d1 ≤ . . . ≤ dr such that −KX = OX (i) for i ≥ 2, it is easy to check that X is covered by lines (see Lemma 2.2). Thus ε(X, OX (1); p) = 1 holds for any p ∈ X. Hence Theorem 1.7 states that we can compute ε(X, OX (1); p) for any Fano variety X obtained as a complete intersection in any rational homoge- neous space of Picard number 1. Throughout this paper, all schemes are defined over the complex number field C. In Section 2, we give the proof of Theorem 1.5. In Section 3, we compute Seshadri constants on some varieties. Acknowledgments. The authors would like to express their gratitude to Professor Yujiro Kawamata for his valuable advice, comments, and warm encouragement. They are also grateful to Professors Katsuhisa Furukawa and Kiwamu Watanabe for their useful comments and suggestions. 2. Lower bounds In this paper, a line means a projective curve of degree 1 in PN . The moduli of lines plays an important role in this paper. Definition 2.1. Let X be a projective scheme in PN and p ∈ X a point. We denote by Fp(X) the moduli space of lines on X passing through p. Note that Fp(X) is naturally embedded in Fp(PN ) ∼= PN −1. For a graded ring S, we denote by Si the set of all homogeneous elements of degree i. We will use the following lemma in Sections 2, 3. Lemma 2.2. Let X be a projective scheme in PN and p ∈ X a point. Fix homogeneous coordinates x0, . . . , xN on PN such that p = [1 : 0 : · · · : 0]. Assume that X is defined by homogeneous polynomials {fj}1≤j≤r around p, j ∈ C[x1, . . . , xN ]i. Then f i j for dj = deg fj and f i i=1 xdj−i Fp(X) ⊂ Fp(PN ) ∼= Proj C[x1, . . . , xN ] is defined by {f i Proof. The proof is similar to that of Proposition 2.13 (a) in [Deb]. We leave the details to the reader. (cid:3) j }1≤j≤r,1≤i≤dj . and write fj =Pdj 0 For a variety X, a point p ∈ X, and an effective Cartier divisor D, we define ordp(D) := max{ m ∈ N f ∈ mm p }, where f is a defining function of D at p. The following lemma is easy and well known, but we prove it for the convenience of the reader. SESHADRI CONSTANTS AND DEGREES OF DEFINING POLYNOMIALS 4 Lemma 2.3. Let C be a curve on a projective variety X and p ∈ C a point. For an effective Cartier divisor D on X not containing C, it holds C.D ≥ ordp(D) · multp(C). Proof. Let ν : eC → C be the normalization. Then there exists an effective divisor E on eC such that O eC (−E) = ν−1mC,p and deg(E) = multp(C). Since OC(−DC ) ⊂ m ordp(DC ) C,p , we have O eC(−ν∗DC) ⊂ O eC (− ordp(DC )E). This means D.C = deg(ν∗DC ) ≥ deg(ordp(DC )E) = ordp(DC ) · multp(C). Since ordp(DC ) ≥ ordp(D), this lemma is proved. (cid:3) Now we can prove Theorem 1.5. The idea is simple. For any curve C on X passing through p, we find a suitable divisor D ∈ OX (i) not containing C for some i and apply Lemma 2.3. Proof of Theorem 1.5. Let x0, . . . , xN be homogeneous coordinates on PN such that p = [1 : 0 : · · · : 0], and set d = dp(X). Choose and fix a basis f1, . . . , fr of H 0(PN , IX ⊗ OPN (d)) ⊂ C[x0, . . . , xN ]d. As in Lemma 2.2, we can write for some f i fj = xd−1 j + xd−2 f 1 j ∈ C[x1, . . . , xN ]i. 0 0 For 1 ≤ i ≤ d − 1, 1 ≤ j ≤ r, we set j + · · · + x0f d−1 f 2 j + f d j Di j = (xi−1 0 f 1 j + · · · + x0f i−1 j + f i j = 0) ⊂ PN . Note that Di j can be PN itself. By the definition of Di j, it holds X ∩ (∗) \1≤i≤d−1,1≤j≤r Di j ⊂ \1≤j≤r (fj = 0) ∩ \1≤i≤d−1,1≤j≤r Di j = \1≤i≤d,1≤j≤r (f i j = 0). By the definition of dp(X), we have X = (f1 = · · · = fr = 0) around p. Hence the last term of (∗) is nothing but Cone Fp(X) by Lemma 2.2, where Cone Fp(X) is the projective cone of Fp(X) in PN with the vertex at p. Since Fp(X) = ∅ by assumption, we have Fix a curve C ⊂ X passing through p. To show the inequality in this (†) \1≤i≤d−1,1≤j≤r Di jX = {p}. theorem, it suffices to show deg(C) multp(C) ≥ d d − 1 . SESHADRI CONSTANTS AND DEGREES OF DEFINING POLYNOMIALS 5 By (†), there exist 1 ≤ i ≤ d − 1, 1 ≤ j ≤ r such that Di C. In particular, Di xd−i 0 f 1 0 it holds that jX does not contain jX is an effective divisor on X not containing C. Since j ) on X, j + · · · + x0f i−1 j) = −(xd−i−1 · · · + x0f d−1 j + f d j + f i (xi−1 f i+1 j 0 ordp(Di jX ) = ordp xi−1 = ordp − Since Di jX ∼ OX(i), we have 0 f 1 j + · · · + x0f i−1 j + f i j xi 0 · · · + x0f d−1 j + f d j xd−i−1 0 f i+1 j (cid:12)(cid:12)(cid:12)(cid:12)X! xd 0 (cid:12)(cid:12)(cid:12)(cid:12)X! ≥ i + 1. i deg(C) = (Di jX ).C ≥ ordp(Di j X) · multp(C) ≥ (i + 1) multp(C) by Lemma 2.3. Hence it holds deg(C)/ multp(C) ≥ (i + 1)/i ≥ d/(d − 1) by 1 ≤ i ≤ d − 1, and the inequality of this theorem is shown. Now, we show the sharpness of the lower bound. First, assume d ≥ n + 1. Set f = xd−1 0 f 1 + · · · + xd−n+1 0 f n−1 + x0f d−1 + f d ∈ C[x0, . . . , xn+1] for a general f i ∈ C[x1, . . . , xn+1]i for each 1 ≤ i ≤ n − 1 and i = d − 1, d. Then f defines a smooth hypersurface X ⊂ Pn+1 containing p = [1 : 0 : · · · : 0]. By the generality of f i, Fp(X) = (f 1 = · · · = f n−1 = f d−1 = f d = 0) ⊂ Proj C[x1, . . . , xn+1] is empty. Set C = (f 1 = · · · = f n−1 = x0f d−1 + f d = 0) ⊂ Pn+1. By definition, C is contained in X and contains p. Since all f i are general, C is a complete intersection curve. Hence deg(C) = (n − 1)! · d, multp(C) = (n − 1)! · (d − 1) hold, and we have ε(X, OX (1); p) ≤ deg(C)/ multp(C) = d/(d − 1). Since dp(X) = d and Fp(X) = ∅, it holds ε(X, OX (1); p) = d/(d − 1) by the inequality of this theorem. When d ≤ n, we use Theorem 1.7, which is proved in the next section. (Of course, we do not use the sharpness assertion in Theorem 1.5 to show Theorem 1.7.) For 2 ≤ d ≤ n, choose positive integers r and d1 ≤ · · · ≤ dr such that d = dr andPr j=1 dj = n + r. Applying Theorem 1.7 to Y = Pn+r, we have ε(X, OX (1); p) = dr/(dr − 1) = d/(d − 1) for a smooth complete intersection X of hypersurfaces of degrees d1 ≤ . . . ≤ dr and general p ∈ X. Note that we can easily check Fp(X) = ∅ for general p ∈ X by using Lemma 2.2. Since dp(X) = dr = d, the sharpness is proved. (cid:3) SESHADRI CONSTANTS AND DEGREES OF DEFINING POLYNOMIALS 6 3. Finding Seshadri curves In [It], the first author computes Seshadri constants on some Fano mani- folds at a very general point. In the paper, toric degenerations are used to estimate Seshadri constants from below. Since the lower semicontinuity is used there for lower bounds, we have to assume some very generality in the method. Instead of toric degenerations, we use the lower bound in Theorem 1.5 here. For upper bounds, we find Seshadri curves similar to [It]. That is, for some variety X, we can find a curve C ⊂ X passing through p such that deg(C)/ multp(C) coincides with the lower bound in Theorem 1.5. To show Theorem 1.7, we treat the case dr ≥ 2 in Subsection 3.1. In that case, we construct a Seshadri curve by cutting a suitable cone in X by hypersurfaces. In Subsection 3.2, we treat the case dr = 1. In that case, we show the existence of a conic C ⊂ X passing through p by using the deformation theory. In Subsection 3.3, we prove Theorem 1.7. 3.1. Cutting cones by hypersurfaces. In this subsection, we prove the following theorem, from which the case dr ≥ 2 in Theorem 1.7 follows im- mediately. Theorem 3.1. Let Y be a projective variety in PN and p ∈ Y a point. For a subvariety X ⊂ Y containing p, we assume the following: i) Around p, X is a locally complete intersection in Y of hypersurfaces of degrees d1 ≤ . . . ≤ dr. That is, r = codim(X, Y ) and there exists fj ∈ H 0(PN , OPN (dj)) for each j such that X = Y ∩ \1≤j≤r (fj = 0) holds in a neighborhood of p. ii) Fp(Y ) 6= ∅ and Pr iii) dp(Y ) ≤ dr. Then it holds that j=1 dj ≤ dim Fp(Y ) + 1. ε(X, OX (1); p) =(cid:26) 1 when Fp(X) 6= ∅, dr/(dr − 1) when Fp(X) = ∅. Proof. When Fp(X) 6= ∅, this theorem is clear. Thus we may assume Fp(X) = ∅, and show ε(X, OX (1); p) = dr/(dr − 1). As in the proof of Theorem 1.5, we take homogeneous coordinates x0, . . . , xN on PN such that j ∈ C[x1, . . . , xN ]i. By i=1 xdj −i p = [1 : 0 : · · · : 0], and write fj =Pdj Lemma 2.2, we have f i j for f i 0 (f i j = 0) ⊂ Proj C[x1, · · · , xN ]. Fp(X) = Fp(Y ) ∩ \i,j Hence Fp(X) is an intersection of Fp(Y ) and P dj hypersurfaces. Thus Pr j=1 dj = dim Fp(Y ) + 1 must hold by the condition ii) and Fp(X) = ∅. We define a curve C ⊂ PN to be C = Cone Z ∩ \1≤j≤r−1 (f 1 j = · · · = f dj−1 j = f dj j = 0) SESHADRI CONSTANTS AND DEGREES OF DEFINING POLYNOMIALS 7 By Theorem 1.5, we have ε(X, OX (1); p) ≥ dr/(dr − 1) since dp(X) ≤ max{dp(Y ), dr} = dr. To show the opposite inequality, we may assume that each f i j is general because of the lower semicontinuity of Seshadri constants (cf. [La, Example 5.1.11]). For general f i j , we can find a curve C ⊂ X through p such that deg(C)/ multp(C) = dr/(dr − 1) as follows. Fix an irreducible component Z of Fp(Y ) such that dim Z = dim Fp(Y ). r = 0). Note dr ≥ 2 holds because X is not linear and dp(X) ≤ dr. Since r = · · · = f dr−2 = x0f dr−1 + f dr ∩ (f 1 r r Cone Z ⊂ Cone Fp(Y ) ⊂ Y, C is contained in X. By definition, C is cut out from Cone Z byPr−1 (dr − 1) = Pr i=1 dj + j=1 dj = dim Fp(Y ) + 1 = j are general, C is a reduced and irreducible curve. j=1 dj − 1 hypersurfaces, and Pr dim Cone Z. Since all f i By definition, we have deg(C) = deg(Cone Z) · d1! · · · dr−1! · (dr − 2)! · dr, multp(C) = multp(Cone Z) · d1! · · · dr−1! · (dr − 2)! · (dr − 1). Thus it holds that deg(C)/ multp(C) = dr/(dr − 1) because deg(Cone Z) = deg(Z) = multp(Cone Z). Hence ε(X, OX (1); p) ≤ deg(C)/ multp(C) = dr/(dr − 1) holds and this theorem follows. (cid:3) 3.2. Finding conics. When Y ⊂ PN is a rational homogeneous space of Picard number 1 other than a projective space, dp(Y ) = 2 for any p ∈ Y . Hence we cannot apply Theorem 3.1 to the case when dr = 1, i.e., X is a section of Y by hyperplanes. Since dp(X) = dp(Y ) = 2 holds for such X, the lower bound obtained by Theorem 1.5 is dp(X)/(dp(X) − 1) = 2. Thus if there exists a (smooth) conic C ⊂ X passing through p, we have 2 ≤ ε(X, OX (1); p) ≤ deg(C) multp(C) = 2. For the following proposition, we prepare some notations. For a subvariety X in a variety Y , we denote by IX/Y the ideal sheaf on Y corresponding to X. A conic in PN is a smooth projective curve of degree 2, and a plane in PN is a 2-dimensional linear projective subspace. For a projective variety Y ⊂ PN , we say that Y is covered by lines (resp. conics, planes) if for general p ∈ Y , there exists a line (resp. a conic, a plane) on Y containing p. Proposition 3.2. Let Y ⊂ PN be a smooth projective variety satisfying the following: i) IY /PN ⊗ OPN (2) is globally generated, SESHADRI CONSTANTS AND DEGREES OF DEFINING POLYNOMIALS 8 ii) for a general p ∈ Y , dim Rp(Y ) = 2 holds, where Rp(Y ) is the subscheme of the Hilbert scheme Hilb(Y ) which parametrizes conics on Y passing through p. Then for general p ∈ Y and general hyperplane H ⊂ PN containing p, there exists a conic C on Y ∩ H passing through p. Proof. Fix a general point p ∈ Y . Since dim Rp(Y ) = 2, we can choose an irreducible component R of Rp(Y ) of dimension 2. We define the incidence variety I as I = {(C, H) ∈ R × OPN (1) ⊗ mp C ⊂ H}, and consider the natural projections I π1 ✁✁✁✁✁✁✁✁ R &▲▲▲▲▲▲▲▲▲▲▲▲ π2 OPN (1) ⊗ mp. To show this proposition, we have to show that the projection π2 is gener- ically surjective. For a fixed C ∈ R, a hyperplane H ∈ OPN (1) ⊗ mp con- tains C if and only if H contains the plane spanned by C. Thus π1 is a PN −3-bundle and it holds that dim I = dim R + N − 3 = N − 1 = dim OPN (1) ⊗ mp. Hence it suffices to show that dim π−1 2 (H) = 0 for a general H ∈ π2(I). Step 1. In this step, we show the following claim. Claim 3.3. In the above setting, the composition of the natural maps H 0(PN , IC/PN ⊗ OPN (1)) ⊗ OPN → H 0(Y, IC/Y ⊗ OY (1)) ⊗ OY → IC/Y ⊗ OY (1) is surjective for C ∈ R. Proof of Claim 3.3. Fix C ∈ R, and let PC ⊂ PN be the plane spanned by C. Since dim Rp(Y ) = 2 and dim Rp(PC ) = 4, PC is not contained in Y . Choose a general section f ∈ H 0(PN , IY /PN ⊗ OPN (2)), and let D ⊂ PN be the corresponding hypersurface of degree 2. By the condition i), the generality of f , and PC 6⊂ Y , we have PC 6⊂ D . Thus as an effective divisor on PC , C is contained in DPC . Since C is a conic and DPC is an effective divisor of degree 2 on PC ∼= P2, C and DPC coincide as schemes. This means IC/PN = ID/PN + IPC /PN . Thus it holds that IC/Y = IC/PN /IY /PN = (ID/PN + IPC /PN )/IY /PN = (IY /PN + IPC /PN )/IY /PN . The last equality follows from ID/PN ⊂ IY /PN . Since H 0(PN , IPC /PN ⊗ OPN (1)) ⊗ OPN → IPC /PN ⊗ OPN (1) & SESHADRI CONSTANTS AND DEGREES OF DEFINING POLYNOMIALS 9 is surjective and H 0(PN , IPC /PN ⊗ OPN (1)) = H 0(PN , IC/PN ⊗ OPN (1)), this claim follows. (cid:3) Step 2. Let (C, H) ∈ I be a general element, and set X = Y ∩ H. In this step, we show that X is smooth. (Note that we do not know H is general in OPN (1) ⊗ mp a priori. Thus we have to check the smoothness of X.) It is clear that X \ C is smooth by Claim 3.3 and the generality of (C, H). Hence it is enough to show that X is smooth along C. Consider B := {(q, H) ∈ C × OPN (1) ⊗ IC/PN Y ∩ H is singular at q}. For each q ∈ C, the fiber of the projection B → C over q corresponds to the kernel of H 0(PN , IC/PN ⊗ OPN (1)) → H 0(Y, IC/Y ⊗ OY (1)) Y,q, → (IC/Y + m2 Y,q)/m2 where mY,q is the maximal ideal sheaf on Y at q. By Claim 3.3, this map is surjective. Hence the projection B → C is a Pk-bundle for k = dim OPN (1) ⊗ IC/PN − codim(C, Y ). Thus we have dim B = k + 1 = dim OPN (1) ⊗ IC/PN + 2 − dim Y. If dim Y ≥ 3, the natural projection B → OPN (1)⊗IC/PN is not generically surjective since dim B < dim OPN (1) ⊗ IC/PN . This means X = Y ∩ H is smooth along C for general H ∈ OPN (1) ⊗ IC/PN . When dim Y = 2, it holds KY .C = −4 because dim R = 2 and C is a free rational curve. Thus we have C 2 = 2. By the Hodge index theorem, Y ⊂ PN is a quadric surface and C ∼ OY (1). Hence X must coincide with C, which is smooth. Step 3. Let (C, H) ∈ I and X = Y ∩ H be as in Step 2. To prove dim π−1 for n = dim X because dim π−1 2 (H) = 0, it is enough to show NC/X 2 (H) ≤ h0(C, NC/X ⊗ mp). Write ∼= O⊕n−1 C f ∗NC/Y = OP1(ai) nMi=1 for integers a1 ≥ . . . ≥ an, where f is an isomorphism P1 → C. (We use f not to confuse OPN (1)C and the degree 1 invertible sheaf on C.) Since C is free on Y , it follows that ai ≥ 0 for any i. Furthermore,Pi ai = dim R = 2 holds since C is free. Hence we have f ∗NC/Y = OP1(2) ⊕ O⊕n−1 or OP1(1)⊕2 ⊕ O⊕n−2. Let α ∈ H 0(PN , IC/PN ⊗ OPN (1)) be a section corresponding to H. From the natural surjection IC/Y ⊗ OY (1) → IC/Y /I 2 C/Y ⊗ OY (1)C ∼= N ∨ C/Y ⊗ OY (1)C SESHADRI CONSTANTS AND DEGREES OF DEFINING POLYNOMIALS 10 and Claim 3.3, we obtain a surjective map δ : H 0(PN , IC/PN ⊗ OPN (1)) ⊗ OC → N ∨ C/Y ⊗ OY (1)C . Furthermore, there exist natural isomorphisms N ∨ C/Y ⊗ OY (1)C ∼= HomC(NC/Y , OY (1)C ) ∼= HomC(NC/Y , NX/Y C ). The image δ(α) ∈ H 0(C, N ∨ induces an exact sequence C/Y ⊗ OY (1)C ) ∼= HomC(NC/Y , NX/Y C) of α 0 → NC/X → NC/Y δ(α) → NX/Y C → 0. Since f ∗NC/Y = OP1(2) ⊕ O⊕n−1 or OP1(1)⊕2 ⊕ O⊕n−2 and f ∗NX/Y C = if the restriction of δ(α) on O(2) or O(1)⊕2 OP1(2), we have NC/X is surjective. Since α is general, this follows from the subjectivity of δ. (cid:3) ∼= O⊕n−1 C 3.3. Proof of Theorem 1.7. As a corollary of Theorem 3.1 and Proposi- tion 3.2, we obtain Theorem 1.7. As stated in Remark 1.8, Theorem 1.7 can be rephrased as follows. 1 2 when Fp(X) 6= ∅, Corollary 3.4 (=Theorem 1.7). Let Y ⊂ PN be a rational homogeneous space of Picard number 1, which is embedded by the ample generator. Let X be a complete intersection variety in Y of hypersurfaces of degrees d1 ≤ . . . ≤ dr such that −KX is ample. For p ∈ X, it holds that dr/(dr − 1) when Fp(X) = ∅ and dr ≥ 2, when Fp(X) = ∅ and dr = 1. ε(X, OX (1); p) = Proof. By the adjunction formula, −KX is ample if and only ifPr j=1 dj ≤ i(Y ) − 1, where i(Y ) is the positive integer satisfying −KY = OY (i(Y )), i.e., the Fano index of Y . It is well known that dim Fp(Y ) = i(Y ) − 2. For instance, the existence of lines is proved in [Ko, Theorem V.1.1.15], and dim Fp(Y ) is computed by the deformation theory. By Lemma 2.2, it is easy j=1 dj < i(Y ) − 1 as in the first to show that X is covered by lines if Pr paragraph of the proof of Theorem 3.1. Furthermore, Y ⊂ PN is cut out by quadrics, i.e., dp(Y ) ≤ 2 holds for any p ∈ Y (see [Li] for example). Thus we can apply Theorem 3.1 if dr ≥ 2, and this corollary follows in that case. j=1 dj + 1 = r + 1 and Y is not a projective space. Since dp(X) ≤ dp(Y ) = 2, ε(X, OX (1); p) ≥ 2 follows from Theorem 1.5. Assume Fp(X) = ∅ and dr = 1. In this case, i(Y ) =Pr semicontinuity of Seshadri constants, we may assume X = Y ∩Tr To show the opposite inequality, we use Proposition 3.2. By the lower j=1 Hj for general hyperplanes Hj ⊂ PN and p ∈ X is a general point. Set Y ′ = Y ∩ Hj, r−1\j=1 SESHADRI CONSTANTS AND DEGREES OF DEFINING POLYNOMIALS 11 and let us check that Y ′ satisfies the conditions i) and ii) in Proposition 3.2. Since Y is cut out by quadrics, so is Y ′. Hence i) holds for Y ′. By Lemma 2.2, Fp(Y ′) is an intersection of Fp(Y ) and general r − 1 hyperplanes in Fp(PN ). Thus Fp(Y ′) is a non-empty 0-dimensional set because dim Fp(Y ) = i(Y ) − 2 = r − 1. Furthermore Fp(Y ) is not an irreducible linear space in Fp(PN ) ∼= PN −1 by [Hw, Proposition 5]. Thus we have #Fp(Y ′) ≥ 2. Since any line in Fp(Y ′) is free by the generality of p, the union l ∪ l′ of two lines l 6= l′ ∈ Fp(Y ′) is smoothable into a free conic C on Y ′ (cf. [Deb, Proposition 4.24]). Thus Y ′ is covered by conics. Since −KY ′ = OY ′(2), we have dim Rp(Y ′) = −KY ′.C − 2 = 2 for C ∈ Rp(Y ′), which is nothing but ii). Thus we can apply Proposition 3.2 to Y ′, and we have a conic on X = Y ′ ∩ Hr passing through p. This means ε(X, OX (1); p) ≤ 2 and the proof is finished. (cid:3) References [Ba] T. Bauer, Seshadri constants on algebraic surfaces, Math. Ann. 313 (1999), no. 3, 547 -- 583. [Ch] K. Chan, A lower bound on Seshadri constants of hyperplane bundles on threefolds, Math. Z. 264 (2010), no. 3, 497 -- 505. [Deb] O. Debarre, Higher-dimensional algebraic geometry, Universitext. Springer-Verlag, New York, 2001. xiv+233 pp. [Dem] J.P. Demailly, Singular Hermitian metrics on positive line bundles, Complex alge- braic varieties (Bayreuth, 1990), 87-104, Lecture Notes in Math., 1507, Springer, Berlin, 1992. [Hw] J.-M. Hwang, On the degrees of Fano four-folds of Picard number 1, J. Reine Angew. Math. 556 (2003), 225-235. [It] A. Ito, Seshadri constants via toric degenerations, arXiv:1202.6664. [Ko] J. Koll´ar, Rational curves on algebraic varieties, Ergeb. Math. Grenzgeb. (3), vol. 32, Springer, 1996. [La] R. Lazarsfeld, Positivity in algebraic geometry I, Ergebnisse der Mathematik undihrer Grenzgebiete, vol. 48. Springer, Berlin (2004). [Li] W. Lichtenstein, A system of quadrics describing the orbit of the highest weight vector, Proc. Amer. Math. Soc. 84 (1982), no. 4, 605 -- 608. [Mu] D. Mumford, Varieties defined by quadratic equations, 1970 Questions on Algebraic Varieties (C.I.M.E., III Ciclo, Varenna, 1969) pp. 29 -- 100 Edizioni Cremonese, Rome. Graduate School of Mathematical Sciences, The University of Tokyo, 3- 8-1 Komaba, Meguro, Tokyo, 153-8914, Japan. E-mail address: [email protected] Graduate School of Mathematical Sciences, The University of Tokyo, 3- 8-1 Komaba, Meguro, Tokyo, 153-8914, Japan. E-mail address: [email protected]
1705.08239
1
1705
2017-05-10T02:34:06
On the splitting of Lazarsfeld-Mukai bundles on K3 surfaces II
[ "math.AG" ]
In this paper, we say that a rank 2 bundle splits if it is given by an extension of two line bundles. In the previous works, we gave a necessary condition for Lazarsfeld-Mukai bundles of rank 2 to split, under a numerical condition ([W2], Theorem 3.1). We gave the splitting types of them on a smooth quartic hypersurface in P3 ([W2], Proposition 3.1) as a corollary of it. However, the assertion of it contains a few mistakes. In this paper, we correct them, and give an application of the results in [W2].
math.AG
math
On the splitting of Lazarsfeld-Mukai bundles on K3 surfaces II Kenta Watanabe ∗ Keywords Lazarsfeld-Mukai bundle, ACM bundle Abstract In this paper, we say that a rank 2 bundle splits if it is given by an ex- tension of two line bundles. In the previous works, we gave a necessary condition for Lazarsfeld-Mukai bundles of rank 2 to split, under a numeri- cal condition ([W2], Theorem 3.1). We gave the splitting types of them on a smooth quartic hypersurface in P3 ([W2], Proposition 3.1) as a corollary of it. However, the assertion of it contains a few mistakes. In this paper, we correct them, and give an application of the results in [W2]. 1 Introduction Let X be a K3 surface, C be a smooth curve of genus g ≥ 3, and Z be a base point free divisor on C with h0(OC(Z)) = r (r ≥ 2). Then the Lazarsfeld-Mukai bundle EC,Z of rank r associated with C and Z is defined as the dual of the kernel FC,Z of the evaluation map ev : H 0(OC(Z)) ⊗ OX → OC(Z). By the construction of it, if KC ⊗ OC(−Z) is not empty, EC,Z is globally gen- erated off the base points of it. Moreover, we have h1(EC,Z) = h2(EC,Z) = 0, by easy computation. Conversely, it is known that if a vector bundle E of rank r with h1(E) = h2(E) = 0 is globally generated, then there exist a smooth curve C and a base point free divisor Z on C with dim Z = r − 1 such that E = EC,Z (for example [C-P], Lemma 1.2). In particular, if a rank 2 vector bundle E is given by the extension 0 −→ M −→ E −→ N −→ 0 ∗Nihon University, College of Science and Technology, 7-24-1 Narashinodai Funabashi city Chiba 274-8501 Japan , E-mail address:[email protected], Telephone numbers: 090- 9777-1974 1 2 of two non-trivial and base point free line bundles N and M satisfying h1(N) = h1(M) = 0, then it is of type EC,Z. Therefore, the problem of when a Lazarsfeld- Mukai bundle EC,Z of rank 2 is given by an extension of two line bundles is natural and interesting. For simplicity, in this paper, we say that a rank 2 bundle splits if it is given by an extension of two line bundles. Donagi and Morrison proved that if EC,Z is not simple, then the following assertion holds. Theorem 1.1 ([D-M], Lemma 4.4). Let X, C, and Z be as above. Assume that h0(OC(Z)) = 2. If EC,Z is not a simple bundle, then there exist two line bundles M and N on X and a subscheme Z ′ ⊂ X of finite length such that: (a) h0(M) ≥ 2, h0(N) ≥ 2; (b) N is base point free; (c) There exists an exact sequence 0 −→ M −→ EC,Z −→ N ⊗ JZ ′ −→ 0. Moreover, if h0(M ⊗ N ∨) = 0, then the length of Z ′ is zero. The exact sequence as in Theorem 1.1 is called Donagi-Morrison's extension. ∼= N ⊕ M. However, Theorem Moreover, if h0(M ⊗ N ∨) = 0, then we have EC,Z 1.1 does not give an answer of the above problem, in the case where EC,Z is simple. In the previous work, we remarked that if C is a very ample smooth curve on X, the Lazarsfeld-Mukai bundle EC,Z of rank 2 defined by C and a base point free pencil Z on C is ACM and initialized with respect to H = OX (C) ([W2], Proposition 2.3). Moreover, we have characterized a necessary condition for EC,Z to split, by ACM line bundles with respect to H, under a numerical condition ([W2], Theorem 3.1). Moreover, in the case where X is a quartic hypersurface in P3 and C is a hyperplane section of X, we have investigated the splitting type of EC,Z, by using a numerical characterization of ACM line bundles on X ([W2], Proposition 3.1). However, the statement of it contains some mistakes, and we indicated no concrete application of them. Therefore, in this paper, we correct the mistakes and give some applications of the previous results as in [W2]. Our plan of this paper is as follows. In section 2, we recall some basic facts about line bundles on K3 surfaces. In section 3, we recall several important theorems about the Clifford index of polarized K3 surfaces proved by Green and Lazarsfeld, and prepare some notations. In section 4, we review the previous results about the splitting of Lazarsfeld-Mukai bundles of rank 2 as in [W2], and correct the mistakes of the assertion. In section 5, we recall the existence theorem for polarized K3 surfaces which are dealt with in this paper, and prepare some lemmas to prove our main theorem. In section 6, we prove our main theorem. 3 Notations and conventions. We work over the complex number field C. A curve and a surface are smooth projective. For a curve or a surface Y , we denote by KY the canonical line bundle of Y and denote by D the linear system defined by a divisor D on Y . For a curve C, the Clifford index of a line bundle A on C is definded as Cliff(A) := deg(A) − 2(h0(A) − 1). The Clifford index of a curve C is defined to be the minimum value of the Clifford index of line bundles on C. We denote it by Cliff(C). It is well known that 0 ≤ Cliff(C) ≤ ⌊ g − 1 2 ⌋, and a curve which has the maximal Clifford index is general in the moduli space of curves of genus g. For a surface X, Pic(X) denotes the Picard lattice of X. A regular surface X (i.e., a surface X with h1(OX) = 0) is called a K3 surface if the canonical bundle KX of it is trivial. For a vector bundle E, we denote by E ∨ the dual of it, and if we fix a very ample line bundle H as a polarization on X, then we write E ⊗ H ⊗l = E(l). We will say that a vector bundle E is initialized with respect to a given polarization H, if H 0(E) 6= 0, and H 0(E(−1)) = 0. 2 Linear systems on K3 surfaces In this section, we recall some classical results about the properties of base point free line bundles on K3 surfaces. Proposition 2.1 ([SD], Proposition 2.7). Let L be a numerical effective line bundle on a K3 surface X. Then L is not base point free if and only if there exists an elliptic curve F , a smooth rational curve Γ and an integer k ≥ 2 such that F.Γ = 1 and L ∼= OX(kF + Γ). Proposition 2.2 ([SD], Proposition 2.6). Let L be a line bundle on a K3 surface X such that L 6= ∅. Assume that L is base point free. Then one of the following cases occurs. (i) L2 > 0 and the general member of L is a smooth irreducible curve of genus + 1. L2 2 (ii) L2 = 0 and L ∼= OX (kF ), where k ≥ 1 is an integer and F is a smooth curve of genus one. In this case, h1(L) = k − 1. 4 We note that, a linear system C given by an irreducible curve C with C 2 > 0 is base point free. Hence, by Proposition 2.2, any line bundle L on a K3 surface has no base point outside its fixed components. On the other hand, by the classification of hyperelliptic linear systems on K3 surfaces which was given by B. Saint-Donat ([SD]), we have the following assertion. [M-M], and [SD], Theorem 5.2). Let L be a numerical Proposition 2.3 (cf. effective line bundle with L2 ≥ 4 on a K3 surface X. Then L is very ample if and only if the following conditions are satisfied. (i) There is no irreducible curve F with F 2 = 0 and F.L ≤ 2. (ii) There is no irreducible curve B with B2 = 2 and L ∼= OX (2B). (iii) There is no (−2)-curve Γ with Γ.L = 0. 3 Clifford index of polarized K3 surfaces In this section, we recall several works about the Clifford index of polarized K3 surfaces, and prepare some notations. First of all, we remark the following theorem. Theorem 3.1 ([G-L]). Let H be a base point free and big line bundle on a K3 surface X of sectional genus g. Then, for all smooth irreducible curve C ∈ H, the Clifford index of C is constant. Moreover, if Cliff(C) < ⌊ ⌋, then there exists a line bundle L on X such that L ⊗ OC computes the Clifford index of C for any smooth irreducible curve C ∈ H. 2 g − 1 By Theorem 3.1, for any base point free and big line bundle H on a K3 surface X, we can define the Clifford index of it by the Clifford index of the curves belonging to the linear system H. Hence, we denote it by Cliff(H). If a given polarized K3 surface (X, H) is not general (i.e., Cliff(H) < ⌊ ⌋), then we can choose the line bundle L as in Theorem 3.1 so that it satisfies the properties as in the following theorem. 2 g − 1 Theorem 3.2 ([Kn], Theorem 8.3). Let X, H, and g be as in Theorem 3.1, and g − 1 ⌋, then there exists a smooth curve D on assume that Cliff(H) = c. If c < ⌊ X satisfying 0 ≤ D2 ≤ c + 2, 2D2 ≤ D.L, and 2 c = Cliff(OX(D) ⊗ OC) = D.H − D2 − 2, for any smooth curve C ∈ H. 5 Let X and H be as in Theorem 3.2. Then we set A(H) := {L ∈ Pic(X) h0(L) ≥ 2 and h0(H ⊗ L∨) ≥ 2}. If A(H) 6= ∅, then we set the integer µ(H) as follows. µ(H) := min{L.(H ⊗ L∨) − 2 L ∈ A(H)}. Then we set A0(H) := {L ∈ A(H) L.(H ⊗ L∨) = µ(H) + 2}. Since the smooth curve D as in Theorem 3.2 satisfies h0(OX (D)C) = h0(OX(D)) and h0(H ⊗ OX (−D)C) = h1(OX (D)C), for any smooth curve C ∈ H, we have the following assertion. Corollary 3.1 . Let X and H be as in Theorem 3.2. Then Cliff(H) = min{µ(H), ⌊ g − 1 2 ⌋}. Moreover, it is known that a line bundle L belonging to A0(H) has the following properties. Proposition 3.1 ([J-K], Proposition 2.6). Let (X, H) be a polarized K3 surface of sectional genus g ≥ 2 such that A(H) 6= ∅. Then µ(H) ≥ 0 and any line bundle L ∈ A0(H) satisfies the following properties. (i) The (possibly empty) base divisor ∆ of L satisfies H.∆ = 0. (ii) h1(L) = 0. 4 Splitting of Lazarsfeld-Mukai bundles In this section, we review previous works in [W2]. Let X be a K3 surface, and let H = OX(1) be a given very ample line bundle on X. Then we say that a vector bundle E is arithmetically Cohen-Macaulay (ACM for short) with respect to H if H 1(E(l)) = 0 for any integer l ∈ Z. We can easily see that if we let C ∈ H be a smooth curve, then the Lazarsfeld-Mukai bundle EC,Z of rank 2 defined by a base point free divisor Z on C with h0(OC(Z)) = 2 is ACM with respect to H ([W2], Proposition 2.3). By using this fact, we can obtain the following necessary condition for EC,Z to split, under a cohomological condition. 6 Proposition 4.1 ([W2], Theorem 3.1). Let X, C and Z be as above. Assume that L is a line bundle on X such that h1(H ⊗ L∨) = 0 and EC,Z fits the exact sequence 0 −→ H ⊗ L∨ −→ EC,Z −→ L −→ 0. Then: (i) If (H ⊗ L∨)2 < 0, then h0(L ⊗ OC(−Z)) 6= 0. (ii) L is initialized with respect to H, and h0(L) ≥ 2. (iii) H ⊗ L∨ and L are ACM with respect to H. Conversely, by the following assertions, we have a sufficient condition for EC,Z to split. Proposition 4.2 . Let the notations be as in Proposition 4.1. If a line bundle L on X satisfies L ⊗ OC ∼= OC(Z) and h1(H ⊗ L∨) = 0, then: (i) L is an elliptic pencil. (ii) There exists the following exact sequence. 0 −→ H ⊗ L∨ −→ EC,Z −→ L −→ 0 Proof. Let g be the genus of C. Note that since C is not hyperelliptic, g ≥ 3. (i) Since Z is a pencil, by the Riemann-Roch theorem, we have g + 1 − deg Z = h1(OC(Z)) ≥ 0. Since H.(L ⊗ H ∨) = deg Z − (2g − 2) ≤ 3 − g ≤ 0, by the ampleness of H, we have h0(L ⊗ H ∨) = 0. Since LC the following exact sequence. ∼= OC(Z), we have 0 −→ L ⊗ H ∨ −→ L −→ OC(Z) −→ 0 Moreover, since h1(H ⊗ L∨) = 0, we have h0(L) = h0(OC(Z)) = 2. Since OC(Z) is base point free and H is ample, L is base point free. Therefore, by Proposition 2.2, L is an elliptic pencil. (ii) Since h1(H ⊗ L∨) = 0 and the exact sequence 0 −→ H 0(OC(Z))∨ ⊗ L ⊗ H ∨ −→ L ⊗ H ∨ ⊗ EC,Z −→ OC −→ 0, we have h0(L ⊗ H ∨ ⊗ EC,Z) = h0(OC) = 1. Hence, we have H ⊗L∨ ⊂ EC,Z. Let M be a saturated subsheaf of EC,Z satisfying H ⊗ L∨ ⊂ M. 7 Assume that h0(M) = 0. Since H ⊗ L∨ ⊂ M, we have H ⊗ M ∨ ⊂ L. On the other hand, by the exact sequence 0 −→ M −→ EC,Z −→ EC,Z/M −→ 0, we have h0(H ⊗ M ∨) ≥ h0(EC,Z) = g + 3 − deg Z ≥ 2. By the assertion of (i), we have L = H ⊗ M ∨. Since M = H ⊗ L∨ and L is an elliptic pencil, we have M.(H ⊗ M ∨) = (H ⊗ L∨).L = H.L = deg Z. Hence, we have the assertion. Assume that h0(M) > 0. By the exact sequence 0 −→ H 0(OC(Z))∨ ⊗ M ∨ −→ EC,Z ⊗ M ∨ −→ M ∨ ⊗ KC(−Z) −→ 0, we have h0(M ∨ ⊗ KC(−Z)) ≥ h0(EC,Z ⊗ M ∨) > 0. Hence, we have deg(M ∨ ⊗ KC(−Z)) = 2g − 2 − deg Z − M.H ≥ 0. On the other hand, since M.H ≥ (H ⊗ L∨).H = 2g − 2 − deg Z, we have M.H = (H ⊗ L∨).H. Therefore, we have M = H ⊗ L∨. By the same reason as above, we have the assertion. (cid:3) Lemma 4.1 ([LC], Proof of Lemma 3.2). Let X be a K3 surface, C be a smooth curve on X, and Z be a base point free divisor on C with h0(OC(Z)) = 2 such that KC ⊗ OC(−Z) 6= ∅. Assume that Q is a torsion free sheaf of rank one on X. Then if there exists a surjective morphism ϕ : EC,Z −→ Q, then Q∨∨ is base point free and not trivial. By Lemma 4.1, we have the following proposition. Proposition 4.3 . Let the notations be as in Proposition 4.1, and let deg Z = d. If Cliff(H) = d − 2 and a saturated subsheaf L of EC,Z satisfies L.H ≥ d, then L is ACM and initialized with respect to H, and there exists the following exact sequence. 0 −→ L −→ EC,Z −→ H ⊗ L∨ −→ 0 Proof. Since L ⊂ EC,Z is saturated, we have (EC,Z/L)∨∨ = H ⊗ L∨. Since L.(H ⊗ L∨) ≤ d, by the assumption, we have L2 ≥ L.H − d ≥ 0. Hence, we have h0(L) ≥ 2. Since h0(KC(−Z)) = h1(OC(Z)) = g + 1 − d and d ≤ ⌊ g + 3 2 ⌋, 8 by Lemma 4.1, h0(H ⊗ L∨) ≥ 2. Hence, by Corollary 3.1, we have Cliff(H) = d − 2 ≤ µ(H) ≤ L.(H ⊗ L∨) − 2. Therefore, we have L.(H ⊗ L∨) = d. Then we have the following exact sequence. 0 −→ L −→ EC,Z −→ H ⊗ L∨ −→ 0 By Proposition 3.1, we have h1(L) = 0. Hence, by Proposition 4.1, L is ACM and initialized with respect to H. (cid:3) If X is a quartic hypersurface in P3 and C ⊂ X is a hyperplane section of X, then we have a complete numerical characterization of ACM line bundles with respect to H. Proposition 4.4 ([W1], Theorem 1.1). Let X be a smooth quartic hypersurface in P3. Let C be a smooth hyperplane section of X, let H = OX(C), and let D be a non-zero effective divisor on X. Then the following conditions are equivalent. (i) OX(D) is an ACM and initialized line bundle with respect to H. (ii) One of the following cases occurs. (a) D2 = −2 and 1 ≤ C.D ≤ 3. (b) D2 = 0 and 3 ≤ C.D ≤ 4. (c) D2 = 2 and C.D = 5. (d) D2 = 4, C.D = 6 and D − C = 2C − D = ∅. In the case where X and C are of as in Proposition 4.4, we get more concrete descriptions about the splitting type of EC,Z. Although we have stated it in [W2] already, the assertion is not correct. Therefore, we correct it as follows. Proposition 4.5 ([W2], Proposition 3.1). Let X, C and H be as in Proposition 4.4. Moreover, let Z be a base point free divisor on C such that h0(OC(Z)) = 2. Then, if there exists a line bundle L with h1(H ⊗ L∨) = 0 such that EC,Z fits the exact sequence 0 −→ H ⊗ L∨ −→ EC,Z −→ L −→ 0, then we get (L.H, L2) = (5, 2), (3, 0) or (4, 0). Proof. We note that, by Propsition 4.1 (ii), we have h0(L) ≥ 2. Hence, L is not trivial. Since Z is a pencil, we have h1(OC(Z)) = 4 − deg Z. Since C is a trigonal curve, we have deg Z = 3 or 4. Since h1(H ⊗L∨) = 0, by Proposition 4.1 (iii), L is ACM. By Proposition 4.1 (ii), L is initialized. Hence, by Proposition 4.4, we have (L.H, L2) = (1, −2), (2, −2), (3, 0), (4, 0), or (5, 2). 9 Since H is very ample, if L.H ≤ 2, the movable part of L is empty and hence, h0(L) = 1. However, this is a contradiction. (cid:3) By Proposition 4.4, the line bundle L satisfying (L.H, L2) = (3, 0) or (4,0) as in Proposition 4.5 is ACM with respect to H and hence, h1(H ⊗ L∨) = 0. Hence, if ∼= OC(Z), by Proposition 4.2, there exists such a line bundle L satisfies L ⊗ OC an exact sequence as in Proposition 4.5. 5 Existence theorem In this section, we recall the following existence theorem for a certain polarized K3 surface of Picard number 2. Theorem 5.1 (cf. [J-K], Proposition 4.2 and Lemma 4.3). Let d and g be g + 3 ⌋. Then there exists a K3 surface X with integers with g ≥ 3 and 3 ≤ d ≤ ⌊ Pic(X) = ZH ⊕ ZF such that H is a base point free line bundle with H 2 = 2g − 2 and the general member of F is an elliptic curve with H.F = d. Moreover, Cliff(H) = d − 2. 2 Note that, in Theorem 5.1, since H is base point free, the general member C of H is a smooth curve, by Proposition 2.2. Moreover, the gonality of C is d and it can be computed by the pencil on C which is given by the restriction of the elliptic pencil F on X ([J-K], proof of Lemma 4.3 and Theorem 4.4). We prepare the following proposition. Proposition 5.1 . Let the notations be as in Theorem 5.1. Then the following statements hold. (i) X contains a (−2)-curve if and only if dg. Moreover, in this case, if we set g = md, then H ⊗ F ∨⊗m is a unique (−2)-vector up to sign in Pic(X) and the member of H ⊗ F ∨⊗m is a (−2)-curve. (ii) Let L be a line bundle which is given by an elliptic curve on X. Then; (a) If dg, then L = F . (b) If d /g, then L = F or there exist integers m and n such that n > 0, n(g − 1) + md = 0, n(g − 1) = lcm(g − 1, d), and L = H ⊗n ⊗ F ⊗m. (iii) H is very ample. (iv) If d ≤ g − 1, then H ⊗ F ∨ is base point free. 10 Proof. (i) By easy computation, H ⊗F ∨⊗m with g = md is a unique (−2)-vector up to sign in Pic(X) and this case occurs precisely when dg. In this case, since H.(H ⊗ F ∨⊗m) = g − 2 > 0, we have H ⊗ F ∨⊗m 6= ∅. By the uniqueness of the (−2)-vector, the member of H ⊗ F ∨⊗m is a (−2)-curve. (ii) Let L = H ⊗n ⊗ F ⊗m, and assume that L is a line bundle which is given by an elliptic curve on X. Then L2 = 2n2(g − 1) + 2nmd = 0. If n = 0, then, by the assumption, we have m = 1. Assume that n 6= 0. Then we have n(g − 1) + md = 0. Since L.F = nd ≥ 0, we have n > 0, and since the general member of L is irreducible, we have n(g − 1) = lcm(g − 1, d). Moreover, this case occurs only if d /g. In fact, if dg, we set k = . Then, by the statement of (i), the member of H ⊗ F ∨⊗k is a (−2)-curve and L.(H ⊗ F ∨⊗k) = −n < 0. Hence, L is not base point free. This is a contradiction. g d g d (iii) First of all, we set m = . Since H.(H ⊗ F ∨⊗m) = g − 2 > 0, by the assertion of (i), H is ample. Moreover, if L is a line bundle given by an elliptic curve on X, by the statement of (ii), we have L.H ≥ d ≥ 3. Assume that there exists a base point free line bundle B of sectional genus 2 with H ∼= B⊗2. Since H 2 = 8, we have g = 5. In this case, by easy computation, we can find a unique 2-vector H ⊗ F ∨ up to sign in Pic(X) precisely when H.F = 3. Since H.(H ⊗ F ∨) = 5 > 0, we have B ∼= H ⊗ F ∨. However, we have the contradiction H ∼= F ⊗2. Therefore, by Proposition 2.3, H is very ample. (iv) In the case where d ≤ g − 1, we have (H ⊗ F ∨)2 = 2g − 2 − 2d ≥ 0. Since H.(H ⊗ F ∨) = 2g − 2 − d ≥ g − 1 > 0, we have H ⊗ F ∨ 6= ∅. If d /g, by the assertion of (i), there is no (−2)-curve on X. Hence, H ⊗ F ∨ is base point free. , the member of H ⊗ F ∨⊗m is a unique (−2)-curve on X, If dg, then for m = and since (H ⊗ F ∨).(H ⊗ F ∨⊗m) = g − 2 − d > 0 and F.(H ⊗ F ∨⊗m) = d ≥ 3, by Proposition 2.1, we have the assertion. (cid:3) g d 6 Main results In this section, we give some applications of previous works as in section 4. ⌋. Then Theorem 6.1 . Let d and g be integers with g ≥ 3 and 3 ≤ d ≤ ⌊ let X be a K3 surface with Pic(X) = ZH ⊕ ZF such that H is a base point free line bundle with H 2 = 2g − 2 and the general member of F is an elliptic curve with H.F = d. Moreover, let C ∈ H be a smooth curve and Z be a divisor on C which gives a gonality pencil Z on C, (that is, deg Z = d). Then EC,Z has the following properties. 2 g + 3 (i) Assume that d = g (i.e., d = g = 3), or g − 1, and let L be a line bundle on X. Then the following conditions are equivalent. 11 (a) There exists the following exact sequence. 0 −→ H ⊗ L∨ −→ EC,Z −→ L −→ 0 (b) L is an elliptic pencil on X, and OC(Z) = LC. (ii) If d = g (i.e., d = g = 3), EC,Z is H-slope-stable. (iii) If d = g − 1, the following conditions are equivalent. (a) EC,Z is not H-slope-stable. (b) EC,Z is strictly H-slope-semistable. (c) OC(Z) ∼= F C or H ⊗ F ∨C. (iv) If d < g − 1, then the following conditions are equivalent. (a) EC,Z is not H-slope-stable. (b) EC,Z is not H-slope-semistable. (c) OC(Z) ∼= F C. First of all, we prove the following lemma to prove Theorem 6.1. Lemma 6.1 . Let EC,Z be as in Theorem 6.1, and let L be a line bundle such that EC,Z fits the following exact sequence. 0 −→ H ⊗ L∨ −→ EC,Z −→ L −→ 0 Then L is a non-trivial and base point free line bundle satisfying h1(L) = h0(L ⊗ H ∨) = 0. Proof. By the Riemann-Roch theorem, we have h0(KC ⊗ OC(−Z)) = g − d + 1 ≥ 1, and hence, we have KC ⊗ OC(−Z) 6= ∅. Therefore, EC,Z is globally generated off the base points of KC ⊗ OC(−Z). Assume that EC,Z fits the following exact sequence. 0 −→ H ⊗ L∨ −→ EC,Z −→ L −→ 0 Since h1(EC,Z) = h2(EC,Z) = 0, if we assume that L is not initialized with respect to H, then we have h1(L) = h2(H ⊗ L∨) = h0(L ⊗ H ∨) 6= 0. 12 By Lemma 4.1, L is globally generated and not trivial. Hence, by Proposition 2.2, there exist an elliptic curve D on X and an integer r ≥ 2 such that L ∼= OX (rD). However, by Proposition 5.1 (ii), we have the contradiction d = L.(H ⊗ L∨) = L.H ≥ rd > d. Therefore, we have h1(L) = h0(L ⊗ H ∨) = 0. (cid:3) Proof of Theorem 6.1. (i) (a)⇒(b). Assume that there exists a line bundle L such that EC,Z fits the following exact sequence. 0 −→ H ⊗ L∨ −→ EC,Z −→ L −→ 0 We consider the case where d = g = 3. If we let Γ = H ⊗F ∨, then by Proposition 5.1 (i), the member of Γ is a (−2)-curve. Since Pic(X) is generated by H and Γ, we set L ∼= H ⊗n ⊗ Γ⊗m (m, n ∈ Z). Note that m < 0. In fact, since L is globally generated and not trivial, we have h0(L) ≥ 2. We have n ≥ 1. Therefore, if m ≥ 0, then L is not initialized. This contradicts Lemma 6.1. Since H ⊗ Γ∨ = F has no base point and L ∼= H ⊗(n+m) ⊗ F ∨m, we have L.F = 3(n + m) ≥ 0. Since h0(L ⊗ H ∨) = 0, we must have n + m = 0 and h1(L) = h2(H ⊗ L∨) = 0. We have m = −1. Hence, H ⊗ L∨ = Γ. Since (H ⊗ L∨)2 = −2 < 0, by Proposition 4.1 (i), we have h0(L ⊗ OC(−Z)) > 0 and deg(L ⊗ OC(−Z)) = L.H − deg Z = L.(H ⊗ L∨) − deg Z = 0. Hence, we have LC ∼= OC(Z). We consider the case where d = g − 1. Let L = H ⊗n ⊗ F ⊗m. Since L.F ≥ 0, we have n ≥ 0. Moreover, since L is globally generated, we have (2g − 2)n(m + n) = L2 ≥ 0. Since h1(L) = 0, if n = 0, then we have m = 1 and hence, L ∼= F . If n ≥ 1, we have m ≥ −n. Since h1(L) = 0, if m = −n, then we have L ∼= H ⊗ F ∨. If m > −n, then there exists an injection (H ⊗ F ∨)⊗(n−1) ֒→ L ⊗ H ∨. Since h0(H ⊗ F ∨) > 0, we have h0(L ⊗ H ∨) 6= 0. Hence, by Lemma 6.1, this case does not occur. Therefore, by Proposition 5.1, H ⊗ L∨ is an elliptic pencil. Hence, by the exact sequence 0 −→ H 0(OC(Z))∨ ⊗ L ⊗ H ∨ −→ EC,Z ⊗ L ⊗ H ∨ −→ L ⊗ OC(−Z) −→ 0, we have h0(EC,Z ⊗ L ⊗ H ∨) = h0(L ⊗ OC(−Z)). Since H ⊗ L∨ ⊂ EC,Z, we have h0(L ⊗ OC(−Z)) > 0. Since deg(L ⊗ OC(−Z)) = L.H − deg Z = L.(H ⊗ L∨) − deg Z = 0, we have LC ∼= OC(Z). 13 (b)⇒(a). By Proposition 5.1 (ii), if L is an elliptic pencil, then L = F or H ⊗F ∨. Since h1(H ⊗ L∨) = 0, by Proposition 4.2, we have the assertion. Before the proof of the latter part of Theorem 6.1, we prepare the following lemma. Lemma 6.2 . Let the notations be as in Theorem 6.1. Let L be a non-trivial line bundle on X. Then L is ACM and initialized with respect to H if and only if L = F or H ⊗ F ∨. Proof. We assume that L is ACM and initialized with respect to H. If we set L = H ⊗n ⊗ F ⊗m, then h1(L ⊗ H ∨⊗n) = 0. Hence, we have m = −1, 0 or 1. Since L is non-trivial, we have L = F or H ⊗ F ∨. Conversely, we assume that L = F or H ⊗ F ∨. By Proposition 5.1, F is initialized with respect to H. Moreover, the general member of H ⊗ F ∨ is a smooth curve. Hence, for any l ∈ Z such that l ≥ 0, we have h1(H l ⊗ F ∨) = 0. Obviously, since h1(H l ⊗ F ) = 0 for any l ≥ 0, F is ACM with respect to H. Hence, H ⊗ F ∨ is also ACM. Since h0(H ⊗ F ∨) > 0, it is initialized with respect to H. (cid:3) Now we return to the proof of Theorem 6.1. (ii) We consider the case where d = g = 3. We assume that there exists a saturated sub-line bundle L of EC,Z such that L.H ≥ = 2. By Lemma 4.1, H ⊗ L∨ = (EC,Z/L)∨∨ is base point free. Hence, by Proposition 2.3 and the Hodge index theorem, we have H.(H ⊗ L∨) ≥ 3. However, this means that L.H ≤ 1. This is a contradiction. H 2 2 (iii) We consider the case where d = g − 1. (b)⇒(a) is clear. (a)⇒(c). Assume that EC,Z is not H-slope-stable. Let M ⊂ EC,Z be a saturated = g − 1. Since d = g − 1, by Proposition 4.3, sub-line bundle with M.H ≥ M is ACM and initialized with respect to H, and there exists the following exact sequence. H 2 2 0 −→ M −→ EC,Z −→ H ⊗ M ∨ −→ 0 Since h1(M) = 0, by the exact sequence 0 −→ H 0(OC(Z))∨ ⊗ M ∨ −→ M ∨ ⊗ EC,Z −→ M ∨ ⊗ KC(−Z) −→ 0, we have h0(M ∨ ⊗ KC(−Z)) = h0(M ∨ ⊗ EC,Z) > 0. By Lemma 6.2, we have M = F or H ⊗ F ∨. Therefore, we have deg(M ∨ ⊗ KC(−Z)) = 2g − 2 − d − M.H = 0. 14 Since H ⊗ M ∨ = H ⊗ F ∨ or F , we have OC(Z) = H ⊗ F ∨C or F C. (c)⇒(b). Assume that OC(Z) = F C or H ⊗ F ∨C. Let L = F or H ⊗ F ∨. Since h1(H ⊗ L∨) = 0, by Proposition 4.2, we have the following exact sequence. 0 −→ H ⊗ L∨ −→ EC,Z −→ L −→ 0 Assume that EC,Z is not H-slope semi-stable. Since the maximal destabilizing sheaf M of EC,Z satisfies M.H > g − 1 = d, by Proposition 4.3, M is initialized and ACM with respect to H. Therefore, by Lemma 6.2, we have M = F or H ⊗ F ∨. However, since M.H = d, this is a contradiction. Hence, EC,Z is strictly H-slope semi-stable. (iv) We consider the case where d < g − 1. (b)⇒(a) is clear. (a)⇒(c). We assume that there exists a saturated sub-line bundle M ⊂ EC,Z H 2 2 such that M.H ≥ = g − 1. By the same way of the proof of (iii) (a)⇒(c), we have M = F or H ⊗ F ∨. Since d < g − 1, we have M = H ⊗ F ∨. Since h1(H ⊗ F ∨) = 0, by the exact sequence 0 −→ H 0(OC(Z))∨ ⊗ F ⊗ H ∨ −→ F ⊗ H ∨ ⊗ EC,Z −→ F ⊗ H ∨ ⊗ KC(−Z) −→ 0, we have h0(F ⊗ H ∨ ⊗ KC(−Z)) = h0(F ⊗ H ∨ ⊗ EC,Z) > 0. On the other hand, since deg(F ⊗ H ∨ ⊗ KC(−Z)) = 0, we have OC(Z) ∼= F C. (c)⇒(b). We assume that OC(Z) = F C. Since h1(H ⊗ F ∨) = 0, by Proposition 4.2, we have H ⊗F ∨ ⊂ EC,Z. Since d < g −1, we have H.(H ⊗F ∨) = 2g −2−d > g − 1. Hence, EC,Z is not H-slope semi-stable. (cid:3) In Theorem 6.1, the case where d = g − 1 occurs precisely when (g, d) = (4, 3) or (5,4). In particular, if (g, d) = (4, 3), then we have the following assertion. Corollary 6.1 . Let the notations be as in Theorem 6.1. If (g, d) = (4, 3), then W 1 3 (C) = {F C, H ⊗ F ∨C}. Proof. Let Z be a divisor on C with h0(OC(Z)) = 2 and deg(Z) = 3. Since h0(EC,Z) = 4, we have dim V2 H 0(EC,Z) = 6 and h0(H) = 5. Hence, the deter- minant map 2 det : ^ H 0(EC,Z) −→ H 0(H) is not injective. Therefore, there exist two sections s1, s2 ∈ H 0(EC,Z) which are linearly independent, and det(s1 ∧ s2) = 0. s1 and s2 form a sub-line bundle L of EC,Z with h0(L) ≥ 2. By taking the saturation of L, we can assume that L is saturated. Since (EC,Z/L)∨∨ ∼= H ⊗ L∨, we have L.(H ⊗ L∨) ≤ 3 and, by 15 Lemma 4.1, we have L ∈ A(H). Since Cliff(C) = 1, L.(H ⊗ L∨) ≥ 3 and hence, L ∈ A0(H). Therefore, we have the following exact sequence. 0 −→ L −→ EC,Z −→ H ⊗ L∨ −→ 0 By Theorem 6.1 (i), H ⊗ L∨ is an elliptic pencil and OC(Z) = H ⊗ L∨C. Since, by Proposition 5.1 (ii), L = F or H ⊗ F ∨, we have the assertion. (cid:3) Acknowledgements The author is partially supported by Grant-in-Aid for Scientific Research (16K05101), Japan Society for the Promotion of Science. References [C-P] C. Ciliberto, G. Pareschi, Pencils of minimal degree on curves on a K3 surface, J. reine angew. Math. 460 15-36 (1995). [D-M] R.Donagi, D.R. Morrison, Linear systems on K3 sections, J. Differen- tial Geom. 29 49-64 (1989). [G-L] M. Green, R. Lazarsfeld, Special divisors on curves on a K3 surface, Invent. Math. 89, 357-370 (1987). [J-K] T. Johnsen, A.L. Knutsen, K3 Projective Models in Scrolls, Lecture Notes in Mathematics, 1842. Springer, Berlin (2004). [Kn] A.L. Knutsen, On kth order embeddings of K3 surfaces and Enriques surfaces, Manuscripta Math. 104, 211-237 (2001). [LC] M. Lelli-Chiesa, Stability of rank 3 Lazarsfeld-Mukai bundles on K3 surfaces, Proc. Lond. Math. Soc. 107(2) 451-479 (2013). [M-M] S. Mori, S. Mukai, The uniruledness of the moduli space of curves of genus 11, In Algebraic geometry, Lecture Notes in Math. 1016, Springer-Verlag, Berlin, 334-353 (1983). [SD] B. Saint-Donat, Projective models of K3 surfaces, Amer. J. Math. 96 (4), 602-639 (1974). [W1] K. Watanabe, The classification of ACM line bundles on quartic hy- persurfaces on P3, Geom. Dedic. 175, 347-353 (2015). [W2] K. Watanabe, On the splitting of Lazarsfeld-Mukai bundles on K3 surfaces, J. Algebra 447, 445-454 (2016).
1402.5651
3
1402
2015-01-10T05:02:37
Tropicalization of Del Pezzo Surfaces
[ "math.AG" ]
We determine the tropicalizations of very affine surfaces over a valued field that are obtained from del Pezzo surfaces of degree 5, 4 and 3 by removing their (-1)-curves. On these tropical surfaces, the boundary divisors are represented by trees at infinity. These trees are glued together according to the Petersen, Clebsch and Schl\"afli graphs, respectively. There are 27 trees on each tropical cubic surface, attached to a bounded complex with up to 73 polygons. The maximal cones in the 4-dimensional moduli fan reveal two generic types of such surfaces.
math.AG
math
Tropicalization of Del Pezzo Surfaces Qingchun Ren, Kristin Shaw and Bernd Sturmfels 5 1 0 2 n a J 0 1 ] Abstract We determine the tropicalizations of very affine surfaces over a valued field that are obtained from del Pezzo surfaces of degree 5, 4 and 3 by removing their (−1)-curves. On these tropical surfaces, the boundary divisors are represented by trees at infinity. These trees are glued together according to the Petersen, Clebsch and Schlafli graphs, respectively. There are 27 trees on each tropical cubic surface, attached to a bounded complex with up to 73 polygons. The maximal cones in the 4-dimensional moduli fan reveal two generic types of such surfaces. . G A h t a m [ 3 v 1 5 6 5 . 2 0 4 1 : v i X r a Introduction 1 A smooth cubic surface X in projective 3-space P3 contains 27 lines. These lines are charac- terized intrinsically as the (−1)-curves on X, that is, rational curves of self-intersection −1. The tropicalization of an embedded surface X is obtained directly from the cubic polynomial that defines it in P3. The resulting tropical surfaces are dual to regular subdivisions of the size 3 tetrahedron. These come in many combinatorial types [15, §4.5]. If the subdivision is a unimodular triangulation then the tropical surface is called smooth (cf. [15, Prop. 4.5.1]). Alternatively, by removing the 27 lines from the cubic surface X, we obtain a very affine surface X 0. In this paper, we study the tropicalization of X 0, denoted trop(X 0), via the embedding in its intrinsic torus [11]. This is an invariant of the surface X. The (−1)-curves on X now become visible as 27 boundary trees on trop(X 0). This distinguishes our approach from Vigeland’s work [27] on the 27 lines on tropical cubics in TP3. It also highlights an important feature of tropical geometry [17]: there are different tropical models of a single classical variety, and the choice of model depends on what structure one wants revealed. Throughout this paper we work over a field K of characteristic zero that has a non- archimedean valuation. Examples include the Puiseux series K = C{{t}} and the p-adic numbers K = Qp. We use the term cubic surface to mean a marked smooth del Pezzo surface X of degree 3. A tropical cubic surface is the intrinsic tropicalization trop(X 0) described above. Likewise, tropical del Pezzo surface refers to the tropicalization trop(X 0) for degree ≥ 4. Here, the adjective “tropical” is used solely for brevity, instead of the more accurate “tropicalized” used in [15]. We do not consider non-realizable tropical del Pezzo surfaces, nor tropicalizations of surfaces defined over a field K with positive characteristic. The moduli space of cubic surfaces is four-dimensional, and its tropical version is the four-dimensional Naruki fan. This was constructed combinatorially by Hacking, Keel and Tevelev [11], and it was realized in [21, §6] as the tropicalization of a very affine variety Y 0, 1 obtained from the Yoshida variety Y in P39 by intersecting with (K∗)39. The Weyl group W (E6) acts on Y by permuting the 40 coordinates. The maximal cones in trop(Y 0) come in two W (E6)-orbits. We here compute the corresponding cubic surfaces: Theorem 1.1. There are two generic types of tropical cubic surfaces. They are contractible and characterized at infinity by 27 metric trees, each having 10 leaves. The first type has 73 bounded cells, 150 edges, 78 vertices, 135 cones, 189 flaps, 216 rays, and all 27 trees are trivalent. The second type has 72 bounded cells, 148 edges, 77 vertices, 135 cones, 186 flaps, (For more data see Table 1.) 213 rays, and three of the 27 trees have a 4-valent node. Here, by cones and flaps we mean unbounded 2-dimensional polyhedra that are affinely to that for tropical planes in [12]. Indeed, by [12, Theorem 4.4], every tropical plane L in isomorphic to R2≥0 and [0, 1]×R≥0 respectively. The characterization at infinity is analogous TPn−1 is given by an arrangement of n boundary trees, each having n − 1 leaves, and L is uniquely determined by this arrangement. Viewed intrinsically, L is the tropicalization of a very affine surface, namely the complement of n lines in P2. Theorem 1.1 offers the analogous characterization for the tropicalization of the complement of the 27 lines on a cubic surface. Tropical geometry has undergone an explosive development during the past decade. To the outside observer, the literature is full of conflicting definitions and diverging approaches. The text books [15, 17] offer some help, but they each stress one particular point of view. An important feature of the present paper is its focus on the unity of tropical geometry. We shall develop three different techniques for computing tropical del Pezzo surfaces: • Cox ideals, as explained in Section 2; • fan structures on moduli spaces, as explained in Section 3; • tropical modifications, as explained in Section 4. The first approach uses the Cox ring of X, starting from the presentation given in [26]. Propositions 2.1 and 2.2 extend this to the universal Cox ideal over the moduli space. For any particular surface X, defined over a field such as K = Q(t), computing the tropicalization is a task for the software gfan [13]. In the second approach, we construct del Pezzo surfaces from fibers in the natural maps of moduli fans. Our success along these lines completes the program started by Hacking et al. [11] and further developed in [21, §6]. The third approach is to build tropical del Pezzo surfaces combinatorially from the tropical projective plane TP2 by the process of tropical modifications in the sense of Mikhalkin [16]. It mirrors the classical construction by blowing up points in P2. All three approaches yield the same results. Section 5 presents an in-depth study of the combinatorics of tropical cubic surfaces and their trees, including an extension of Theorem 1.1 that includes all degenerate surfaces. We now illustrate the rich combinatorics in our story for a del Pezzo surface X of degree 4. Del Pezzo surfaces of degree d ≥ 6 are toric surfaces, so they naturally tropicalize as polygons with 12− d vertices [17, Ch. 3]. On route to Theorem 1.1, we prove the following for d = 4, 5: Proposition 1.2. Among tropical del Pezzo surfaces of degree 4 and 5, each has a unique generic combinatorial type. For degree 5, this is the cone over the Petersen graph. For degree 4, the surface is contractible and characterized at infinity by 16 trivalent metric trees, each with 5 leaves. It has 9 bounded cells, 20 edges, 12 vertices, 40 cones, 32 flaps, and 48 rays. 2 Figure 1: Tropical del Pezzo surfaces of degree 4 illustrated by coloring the Clebsch graph To understand degree 4, we consider the 5-regular Clebsch graph in Figure 1. Its 16 nodes are the (−1)-curves on X, labelled E1, . . . , E5, F12, . . . , F45, G. Edges represent intersecting pairs of (−1)-curves. In the constant coefficient case, when K has trivial valuation, the tropicalization of X is the fan over this graph. However, over fields K with non-trivial valuation, trop(X 0) is usually not a fan, but one sees the generic type from Proposition 1.2. Here, the Clebsch graph deforms into a trivalent graph with 48 = 16·3 nodes and 72 = 40+32 edges, determined by the color coding in Figure 1. Each of the 16 nodes is replaced by a trivalent tree with five leaves. Incoming edges of the same color (red or blue) form a cherry (= two adjacent leaves) in that tree, while the black edge connects to the non-cherry leaf. Corollary 1.3. For a del Pezzo surface X of degree 4, the 16 metric trees on its tropical- ization trop(X 0), obtained from the (−1)-curves on X, are identical up to relabeling. Proof. Moving from one (−1)-curve on X to another corresponds to a Cremona transforma- tion of the plane P2. Each (−1)-curve on X has exactly five marked points arising from its intersections with the other (−1)-curves. Moreover, the Cremona transformations preserve the cross ratios among the five marked points on these 16 P1’s. From the valuations of all the various cross ratios one can read off the combinatorial trees along with their edge lengths, as explained in e.g. [15, Proposition 6.5.1] or [21, Example 5.2]. We then obtain the following relabeling rules for the leaves on the 16 trees, which live in the circular nodes of Figure 1. We start with the tree G whose leaves are labeled E1, E2, E3, E4, E5. For the specific 3 GF12E2E4F24F25F13F35E3F45F23F34F15F14E1E5 Figure 2: The bounded complex of the tropical del Pezzo surface in degree 4 example in Figure 1, this is the caterpillar tree ({E1, E4}, E3,{E2, E5}). Now, given any trivalent tree for G, the tree Fij is obtained by relabeling the five leaves as follows: Ei (cid:55)→ Ej, Ej (cid:55)→ Ei, Ek (cid:55)→ Flm, El (cid:55)→ Fkm, Em (cid:55)→ Fkl. (1.1) Here {k, l, m} = {1, 2, 3, 4, 5}\{i, j}. The tree Ei is obtained from the tree G by relabelling (1.2) Ei (cid:55)→ G and Ej (cid:55)→ Fij where j (cid:54)= i. This explains the color coding of the graph in Figure 1. The bounded complex of trop(X 0) is shown in Figure 2. It consists of a central rectangle, with two triangles attached to each of its four edges. There are 12 vertices, four vertices of the rectangle, labeled S, and eight pendant vertices, labeled T. To these 12 vertices and 20 edges, we attach the flaps and cones, according to the deformed Clebsch graph structure. The link of each S vertex in the surface trop(X 0) is the Petersen graph (Figure 3), while the link of each T vertex is the bipartite graph K3,3. The bounded complex has 16 chains TST consisting of two edges with different colors. These are attached by flaps to the bounded parts of the 16 trees. The Clebsch graph (Figure 1) can be recovered from Figure 2 as follows: its nodes are the TST chains, and two chains connect if they share precisely one vertex. Out at infinity, T vertices attach along cherries, while S vertices attach along non-cherry leaves. Each such attachment between two of the 16 trees links to the bounded complex by a cone. 2 Cox Ideals We study del Pezzo surfaces over K of degrees 5, 4 and 3. Such surfaces X are obtained from P2 by blowing up 4, 5 or 6 points in general position, and we obtain moduli by varying these points. From an algebraic perspective, it is convenient to represent X by its Cox ring Cox(X) = H 0(X,L). (2.1) (cid:77) L∈Pic(X) 4 SSSSTTTTTTTT The Cox ring of a del Pezzo surface X was first studied by Batyrev and Popov [4]. We shall express this ring explicitly as a quotient of a polynomial ring over the ground field K: The number of variables xC in our three polynomial rings is 10, 16 and 27 respectively. The ideal IX is the Cox ideal of the surface X. It was conjectured already in [4] that the ideal IX is generated by quadrics. This conjecture was proved in several papers, including [25, 26]. Cox(X) = K(cid:2) xC : C is a (−1)-curve on X(cid:3) modulo an ideal IX generated by quadrics. the N-graded subring Cox(X)[L] =(cid:76) The Cox ring encodes all maps from X to a projective space. Such a map is given by m≥0 H 0(X, mL) for a fixed line bundle L ∈ Pic(X). The image of the map X → Proj(Cox(X)[L]) is contained in the projective space PN , where N = dim(H 0(X,L)) − 1, provided Cox(X)[L] is generated in degree 1. This applies to both the anticanonical map and to the blow-down map to P2. In what follows, we give explicit generators for all relevant Cox ideals IX. Some of this is new and of independent interest. The tropicalization of X 0 we seek is defined from the ideal IX. So, in principle, we can compute trop(X 0) from IX using the software gfan [13]. Recall that X 0 denotes the very affine surface obtained from X by removing all (−1)-curves. Figure 3: The tropical del Pezzo surface trop(M0,5) is the cone over the Petersen graph. Del Pezzo Surfaces of Degree 5 Consider four general points in P2. This configuration is projectively unique, so there are no moduli. The surface X is the moduli space M 0,5 of rational stable curves with five marked points, see for example [14]. The very affine variety X 0 is simply M0,5, the moduli space of rational curves with five distinct marked points. It is the complement of a hyperplane arrangement whose underlying matroid is the graphical matroid of the complete graph K4. The Cox ideal is the Plucker ideal of relations among 2 × 2-minors of a 2 × 5-matrix: IX = (cid:104) p12p34 − p13p24 + p14p23, p12p45 − p14p25 + p15p24, p12p35 − p13p25 + p15p23, p13p45 − p14p35 + p15p34, p23p45 − p24p35 + p25p34 (cid:105). The affine variety of IX in ¯K 10 is the universal torsor of X, now regarded over the algebraic closure ¯K of the given valued field K. From the perspective of blowing up P2 at 4 points, the ten variables (representing the ten (−1)-curves) fall in two groups: the four exceptional fibers, and the six lines spanned by pairs of points. For example, we may label the fibers by E1 = p15, E2 = p25, E3 = p35, E4 = p45, 5 12452315343513142425 and the six lines by F12 = p34, F13 = p24, F14 = p23, F23 = p14, F24 = p13, F34 = p12. The Cox ideal IX is homogeneous with respect to the natural grading by the Picard group Pic(X) = Z5. In Plucker coordinates, this grading is given by setting deg(pij) = ei + ej, where ei represents the ith standard basis vector in Z5 = Pic(X). This translates into an action of the torus ( ¯K∗)5 = Pic(X) ⊗Z ¯K∗ on the universal torsor in ¯K 10. We remove the ten coordinate hyperplanes in ¯K 10, and we take the quotient modulo ( ¯K∗)5. The result is precisely the very affine del Pezzo surface we seek to tropicalize: X 0 = M0,5 ⊂ ( ¯K ∗ )10/( ¯K ∗ )5. (2.2) The 2-dimensional balanced fan trop(X 0) is the Bergman fan of the graphical matroid of K4. It is known from [2] that this is the cone over the Petersen graph. This is also easy to check directly with gfan on IX. This fan is also the moduli space of 5-marked rational tropical curves, that is, 5-leaf trees with lengths on the two bounded edges (cf. [15, §4.3]). Del Pezzo Surfaces of Degree 4 Consider now five general points in P2. There are two degrees of freedom. The moduli space is our previous del Pezzo surface M0,5. Indeed, fixing five points in P2 corresponds to fixing a point (p12, . . . , p45) in M0,5, using Cox-Plucker coordinates as in (2.2). Explicitly, if we write the five points as a 3 × 5-matrix then the pij are the Plucker coordinates of its kernel. Replacing K with the previous Cox ring, we may consider the universal del Pezzo surface Y. The universal Cox ring is the quotient of a polynomial ring in 26 = 10 + 16 variables: K[Y] = Cox(M0,5)[E1, E2, E3, E4, E5, F12, F13, . . . , F45, G]/IY. (2.3) As before, Ei represents the exceptional divisor over point i, and Fij represents the line spanned by points i and j. The variable G represents the conic spanned by the five points. Proposition 2.1. Up to saturation with respect to the product of the 26 variables, the uni- versal Cox ideal IY for degree 4 del Pezzo surfaces is generated by the following 45 trinomials: Base Group p12p34−p13p24+p14p23 p13p45−p14p35 + p15p34 p23p45−p24p35+p25p34 p12p35−p13p25+p15p23 p12p45−p14p25 + p15p24, Group 1 Group 2 Group 3 Group 4 Group 5 F23F45−F24F35+F25F34 F13F45−F14F35+F15F34 p23p45F24F35−p24p35F23F45−GE1 p23p45F25F34−p25p34F23F45−GE1 p24p35F25F34−p25p34F24F35−GE1 p13p45F14F35−p14p35F13F45−GE2 p13p45F15F34−p15p34F13F45−GE2 p14p35F15F34−p15p34F14F35−GE2 p12p45F14F25−p14p25F12F45−GE3, p14p25F15F24−p15p24F14F25−GE3 p12p35F13F25−p13p25F12F35−GE4 p12p35F15F23−p15p23F12F35−GE4 p13p25F15F23−p15p23F13F25−GE4 p12p34F13F24−p13p24F12F34−GE5 p12p34F14F23−p14p23F12F34−GE5 p13p24F14F23−p14p23F13F24−GE5 p12p45F15F24−p15p24F12F45−GE3 F12F45−F14F25+F15F24 F12F35−F13F25+F15F23 F12F34−F13F24+F14F23 6 Group 1’ Group 2’ Group 3’ Group 4’ Group 5’ p25F12E2−p35F13E3+p45F14E4 p23F12E2−p34F14E4+p35F15E5 p15F12E1−p35F23E3+p45F24E4 p13F12E1−p34F24E4+p35F25E5 p15F13E1−p25F23E2+p45F34E4 p12F13E1−p24F34E4+p25F35E5 p15F14E1−p25F24E2+p35F34E3 p12F14E1−p23F34E3+p25F45E5 p14F15E1−p24F25E2+p34F35E3 p12F15E1−p23F35E3+p24F45E4 p24F12E2−p34F13E3+p45F15E5 p23F13E3−p24F14E4+p25F15E5 p14F12E1−p34F23E3+p45F25E5 p13F23E3−p14F24E4+p15F25E5 p14F13E1−p24F23E2+p45F35E5 p12F23E2−p14F34E4+p15F35E5 p13F14E1−p23F24E2+p35F45E5 p12F24E2−p13F34E3+p15F45E5 p13F15E1−p23F25E2+p34F45E4 p12F25E2−p13F35E3+p14F45E4 Proposition 2.1 will be derived later in this section. For now, let us discuss the structure and symmetry of the generators of IY. We consider the 5-dimensional demicube, here denoted D5. This is the convex hull of the following 16 points in the hyperplane {a0 = 0} ⊂ R6: (cid:8) (1, a1, a2, a3, a4, a5) ∈ {0, 1}6 : a1 + a2 + a3 + a4 + a5 is odd(cid:9). (2.4) The group of symmetries of D5 is the Weyl group W (D5). It acts transitively on (2.4). There is a natural bijection between the 16 variables in the Cox ring and the vertices of D5: E1 ↔ (1, 1, 0, 0, 0, 0), E2 ↔ (1, 0, 1, 0, 0, 0), . . . , E5 ↔ (1, 0, 0, 0, 0, 1), F12 ↔ (1, 0, 0, 1, 1, 1), F13 ↔ (1, 0, 1, 0, 1, 1), . . . , F45 ↔ (1, 1, 1, 1, 0, 0), (2.5) G ↔ (1, 1, 1, 1, 1, 1). This bijection defines the grading via the Picard group Z6. We regard the pij as scalars, so they have degree 0. Generators of IY that are listed in the same group have the same Z6 degrees. The action of W (D5) on the demicube D5 gives the action on the 16 variables. Consider now a particular smooth del Pezzo surface X of degree 4 over the field K, so the pij are scalars in K that satisfy the Plucker relations in the Base Group. The universal Cox ideal IY specializes to the Cox ideal IX for the particular surface X. That Cox ideal is minimally generated by 20 quadrics, two per group. The surface X 0 is the zero set of the ideal IX inside ( ¯K∗)16/( ¯K∗)6. The torus action is obtained from (2.5), in analogy to (2.2). Proof of Proposition 1.2. We computed trop(X 0) by applying gfan [13] to the ideal IX. If K = Q with the trivial valuation then the output is the cone over the Clebsch graph in Figure 1. This 5-regular graph records which pairs of (−1)-curves intersect on X. This also works over a field K with non-trivial valuation. The software gfan uses K = Q(t). If the vector (p12, . . . , p45) tropicalizes into the interior of an edge in the Petersen graph then trop(X 0) is the tropical surface described in Proposition 1.2. Each node in Figure 1 is replaced by a trivalent tree on 5 nodes according to the color coding explained in Section 1. The surface trop(X 0) can also be determined by tropical modifications, as in Section 4. The same tropicalization method works for the universal family Y 0. Its ideal IY is given by the 45 polynomials in 26 variables listed above, and Y 0 is the zero set of IY in the 15- dimensional torus ( ¯K∗)10/( ¯K∗)5 × ( ¯K∗)16/( ¯K∗)6. The tropical universal del Pezzo surface trop(Y 0) is a 4-dimensional fan in R26/R11. We compute it by applying gfan to the universal 7  . (cid:81) (2.6) (2.7) Cox ideal IY. The Grobner fan structure on trop(IY) has f-vector (76, 630, 1620, 1215). It is isomorphic to the Naruki fan described in [21, Table 5] and discussed further in Section 3. Del Pezzo Surfaces of Degree 3 (Cubic Surfaces) There exists a cuspidal cubic through any six points in P2. See e.g. [20, (4.4)] and [21, (6.1)]. Hence any configuration of six points in P2 can be represented by the columns of a matrix  1 D = d1 1 d3 d3 1 1 1 1 1 d2 d3 d4 d5 d6 5 d3 2 d3 6 3 d3 4 d3 The maximal minors of the matrix D factor into linear forms, [ijk] = (di − dj)(di − dk)(dj − dk)(di + dj + dk), and so does the condition for the six points to lie on a conic: [conic] = = [134][156][235][246] − [135][146][234][256] (d1 + d2 + d3 + d4 + d5 + d6) · 1≤i<j≤6(di − dj). The linear factors in these expressions form the root system of type E6. This corresponds to an arrangement of 36 hyperplanes in P5. Similarly, the arrangement of type E7 consists of 63 hyperplanes in P6, as in [20, (4.4)]. To be precise, for m = 6, 7, the roots for Em are di − dj di + dj + dk di1 + di2 + ··· + di6 1 ≤ i < j ≤ m, for for for 1 ≤ i1 < i2 < ··· < i6 ≤ m. 1 ≤ i < j < k ≤ m, (2.8) Linear dependencies among these linear forms specify a matroid of rank m, also denoted Em. The moduli space of marked cubic surfaces is the 4-dimensional Yoshida variety Y defined in [21, §6]. It coincides with the subvariety Y 0 of ( ¯K∗)26/( ¯K∗)11 cut out by the 45 trinomials in Proposition 2.1. This is the embedding of Y 0 in its intrinsic torus, as in [21, Theorem 6.1], and it differs from the embedding of Y 0 into ( ¯K∗)40/( ¯K∗) referred to in Section 3 below. The universal family for cubic surfaces is denoted by G0. This is the open part of the Gopel variety G ⊂ P134 constructed in [20, §5]. The base of this 6-dimensional family is the 4-dimensional Y 0. The map G0 → Y 0 was described in [11]. Thus the ring K[Y] in (2.3) is the natural base ring for the universal Cox ring for degree 3 surfaces. At this point it is essential to avoid confusing notation. To aim for a clear presentation, we use the uniformization of Y by the E6 hyperplane arrangement. Namely, we take R = Z[d1, d2, d3, d4, d5, d6] instead of K[Y] as the base ring. We write X for the universal cubic surface over R. The universal Cox ring is a quotient of the polynomial ring over R in 27 variables, one for each line on the cubic surface. Using variable names as in [26, §5], we write (2.9) This ring is graded by the Picard group Z7, similarly to (2.5). The role of the 5-dimensional demicube D5 is now played by the 6-dimensional Gosset polytope with 27 vertices, here also denoted by E6. The symmetry group of this polytope is the Weyl group W (E6). Cox(X ) = R[E1, E2, . . . , E6, F12, F13, . . . , F56, G1, G2, . . . , G6]/IX . 8 Proposition 2.2. Up to saturation by the product of all 27 variables and all 36 roots, the universal Cox ideal IX is generated by 270 trinomials. These are clustered by Z7-degrees into 27 groups of 10 generators, one for each line on the cubic surface. For instance, the 10 generators of IX that correspond to the line G1 involve the 10 lines that meet G1. They are (d3−d4)(d1+d3+d4)E2F12 − (d2−d4)(d1+d2+d4)E3F13 + (d2−d3)(d1+d2+d3)E4F14, (d3−d5)(d1+d3+d5)E2F12 − (d2−d5)(d1+d2+d5)E3F13 + (d2−d3)(d1+d2+d3)E5F15, (d3−d6)(d1+d3+d6)E2F12 − (d2−d6)(d1+d2+d6)E3F13 + (d2−d3)(d1+d2+d3)E6F16, (d4−d5)(d1+d4+d5)E2F12 − (d2−d5)(d1+d2+d5)E4F14 + (d2−d4)(d1+d2+d4)E5F15, (d4−d6)(d1+d4+d6)E2F12 − (d2−d6)(d1+d2+d6)E4F14 + (d2−d4)(d1+d2+d4)E6F16, (d5−d6)(d1+d5+d6)E2F12 − (d2−d6)(d1+d2+d6)E5F15 + (d2−d5)(d1+d2+d5)E6F16, (d4−d5)(d1+d4+d5)E3F13 − (d3−d5)(d1+d3+d5)E4F14 + (d3−d4)(d1+d3+d4)E5F15, (d4−d6)(d1+d4+d6)E3F13 − (d3−d6)(d1+d3+d6)E4F14 + (d3−d4)(d1+d3+d4)E6F16, (d5−d6)(d1+d5+d6)E3F13 − (d3−d6)(d1+d3+d6)E5F15 + (d3−d5)(d1+d3+d5)E6F16, (d5−d6)(d1+d5+d6)E4F14 − (d4−d6)(d1+d4+d6)E5F15 + (d4−d5)(d1+d4+d5)E6F16. defined by IX in P5 × ( ¯K∗)27/( ¯K∗)7 is 6-dimensional. It is the universal family X 0. Proof of Propositions 2.1 and 2.2. We consider the prime ideal in [20, §6] that defines the embedding of the Gopel variety G into P134. By [20, Theorem 6.2], G is the ideal-theoretic intersection of a 35-dimensional toric variety T and a 14-dimensional linear space L. The latter is cut out by a canonical set of 315 linear trinomials, indexed by the 315 isotropic planes in (F2)6. Pulling these linear forms back to the Cox ring of T , we obtain 315 quartic trinomials in 63 variables, one for each root of E7. Of these 63 roots, precisely 27 involve the unknown d7. We identify these with the (−1)-curves on the cubic surface via The remaining 260 trinomials are obtained by applying the action of W (E6). The variety di − d7 (cid:55)→ Ei, di + dj + d7 (cid:55)→ Fij, −dj + di (cid:55)→ Gj. (2.10) Moreover, of the 315 quartics, precisely 270 contain a root involving d7. Their images under the map (2.10) are the 270 Cox relations listed above. Our construction ensures that they generate the correct Laurent polynomial ideal on the torus of T . This proves Proposition 2.2. The derivation of Proposition 2.1 is similar, but now we use the substitution 7(cid:88) i=1 6(cid:88) i=1 di − d6 (cid:55)→ Ei, di + dj + d6 (cid:55)→ Fij, di (cid:55)→ G. We consider the 45 quartic trinomials that do not involve d7. Of these, precisely five do not involve d6 either. They translate into the five Plucker relations for M0,5. With this identifica- tion, the remaining 40 quartics translate into the ten groups listed after Proposition 2.1. Remark 2.3. The relations (2.10) is not unique. The non-uniqueness comes from both the symmetry of E7 and the choice of ± sign for each variable Ei, Fij, Gj. For the rest of this paper, we fix the relations (2.10). Other choices give the same result up to symmetry. 9 We now fix a K-valued point in the base Y 0, by replacing the unknowns di with scalars in K. In order for the resulting surface X to be smooth, we require (d1 : d2 : d3 : d4 : d5 : d6) to be in the complement of the 36 hyperplanes for E6. The corresponding specialization of IX is the Cox ideal IX of X. Seven of the ten trinomials in each degree are redundant over K. Only three are needed to generate IX. Hence, the Cox ideal IX is minimally generated by 81 quadrics in the Ei, Fij and Gi. Its variety is the surface X 0 = V (IX) ⊂ ( ¯K∗)27/( ¯K∗)7. Proposition 2.4. Each of the marked 27 trees on a tropical cubic surface has an involution. Proof. Every line L on a cubic surface X over K, with its ten marked points, admits a double cover to P1 with five markings. The preimage of one of these marked points is the pair of markings on L given by two other lines forming a tritangent with L. Tropically, this gives a double cover from the 10-leaf tree for L to a 5-leaf tree with leaf labelings given by these pairs. The desired involution on the 10-leaf tree exchanges elements in each pair. For instance, for the tree that corresponds to the line L = G1, the involution equals E2 ↔ F12, E3 ↔ F13, E4 ↔ F14, E5 ↔ F15, E6 ↔ F16. Indeed, this involution fixes the 10 Cox relations displayed in Proposition 2.2. The induced action on the trees corresponding to the tropicalization of the lines can be seen in Figures 4 and 5, where the involution reflects about a vertical axis. The corresponding 5-leaf tree is the tropicalization of the line in Proj(K[E2F12, E3F13, E4F14, E5F15, E6F16]) (cid:39) P4 that is the intersection of the 10 hyperplanes defined by the polynomials in Proposition 2.2. We aim to compute trop(X 0) by applying gfan to the ideal IX. This works well for K = Q with the trivial valuation. Here the output is the cone over the Schlafli graph which records which pairs of (−1)-curves intersect on X. This is a 10-regular graph with 27 nodes. However, for K = Q(t), our gfan calculations did not terminate. Future implementations of tropical algorithms will surely succeed; see also Conjecture 5.3. To get the tropical cubic surfaces, and to prove Theorem 1.1, we used the alternative method explained in Section 3. 3 Sekiguchi Fan to Naruki Fan In the previous section we discussed the computation of tropical del Pezzo surfaces directly from their Cox ideals. This worked well for degree 4. However, using the current imple- mentation in gfan, this computation did not terminate for degree 3. We here discuss an alternative method that did succeed. In particular, we present the proof of Theorem 1.1. The successful computation uses the following commutative diagram of balanced fans: (3.1) (cid:121) (cid:121) Berg(E7) −−−→ trop(G0) Berg(E6) −−−→ trop(Y 0) 10 This projection induces the vertical map from Berg(E7) to Berg(E6) on the left in (3.1). This diagram was first derived by Hacking et al. [11], in their study of moduli spaces of marked del Pezzo surfaces. Combinatorial details were worked out by Ren et al. in [21, §6]. The material that follows completes the program that was suggested at the very end of [21]. The notation Berg(Em) denotes the Bergman fan of the rank m matroid defined by the (36 resp. 63) linear forms listed in (2.8). Thus, Berg(E6) is a tropical linear space in TP35, and Berg(E7) is a tropical linear space in TP62. Coordinates are labeled by roots. The list (2.8) fixes a choice of injection of root systems E6 (cid:44)→ E7. This defines coordinate projections R63 → R36 and TP62 (cid:57)(cid:57)(cid:75) TP35, namely by deleting coordinates with index 7. On the right in (3.1), we see the 4-dimensional Yoshida variety Y ⊂ P39 and the 6- dimensional Gopel variety G ⊂ P134. Explicit parametrizations and equations for these vari- eties were presented in [20, 21]. The corresponding very affine varieties G0 ⊂ ( ¯K∗)135/ ¯K∗ and Y 0 ⊂ ( ¯K∗)40/ ¯K∗ are moduli spaces of marked del Pezzo surfaces [11]. Their tropicalizations trop(G0) and trop(Y 0) are known as the Sekiguchi fan and Naruki fan, respectively. The modular interpretation in [11] ensures the existence of the vertical map trop(G0) → trop(Y 0). This map is described in [11, Lemma 5.4], in [21, (6.5)], and in the proof of Lemma 3.1 below. The two horizontal maps in (3.1) are surjective and (classically) linear. The linear map Berg(E7) → trop(G0) is given by the 135 × 63 matrix A in [20, §6]. The corresponding toric variety is the object of [20, Theorem 6.1]. The map Berg(E6) → trop(Y 0) is given by the 40× 36-matrix in [21, Theorem 6.1]. We record the following computational result. It refers to the natural simplicial fan structure on Berg(Em) described by Ardila et al. in [3]. Lemma 3.1. The Bergman fans of E6 and E7 have dimensions 5 and 6. Their f-vectors are fBerg(E6) = fBerg(E7) = (1, 6091, 315399, 3639804, 14982660, 24607800, 13721400). (1, 750, 17679, 105930, 219240, 142560), The moduli fans trop(Y 0) and trop(G0) have dimensions 4 and 6. Their f-vectors are ftrop(Y 0) = ftrop(G0) = (1, 76, 630, 1620, 1215), (1, 1065, 27867, 229243, 767025, 1093365, 547155). Proof. The f-vector for the Naruki fan trop(Y 0) appears in [21, Table 5]. For the other three fans, only the numbers of rays (namely 750, 6091 and 1065) were known from [21, §6]. The main new result in Lemma 3.1 is the computation of all 57273155 cones in Berg(E7). The fans Berg(E6) and trop(G0) are subsequently derived from Berg(E7) using the maps in (3.1). We now describe how fBerg(E7) was found. We did not use the theory of tubings in [3]. Instead, we carried out a brute force computation based on [10] and [19]. Recall that a point lies in the Bergman fan of a matroid if and only if the minimum is obtained twice on each circuit. We computed all circuits of the rank 7 matroid on the 63 vectors in the root system E7. That matroid has precisely 100662348 circuits. Their cardinalities range from 3 to 8. This furnishes a subroutine for deciding whether a given point lies in the Bergman fan. Our computations were mostly done in sage [24] and java. We achieved speed by exploiting the action of the Weyl group W (E7) given by the two generators in [20, (4.2)]. The two matrices derived from these two generators using [20, (4.3)] act on the space R7 with coordinates d1, d2, . . . , d7. This gives subroutines for the action of W (E7) on R63, e.g. for deciding whether two given sequences of points are conjugate with respect to this action. 11 Let r1, . . . , r6091 denote the rays of Berg(E7), as in [11, Table 2] and [21, §6]. They form 11 orbits under the action of W (E7). For each orbit, we take the representative ri with smallest label. For each pair i < j such that ri is a representative, our program checks if ri + rj lies in Berg(E7), using the precomputed list of circuits. If yes, then ri and rj span a 2-dimensional cone in Berg(E7). This process gives representatives for the W (E7)-orbits of 2-dimensional cones. The list of all cones is produced by applying the action of W (E7) on the result. For each orbit, we keep only the lexicographically smallest representative (ri, rj). Next, for each triple i < j < k such that (ri, rj) is a representative, we check if ri + rj + rk lies in Berg(E7). If so, then {ri, rj, rk} spans a 3-dimensional cone in Berg(E7). The list of all 3-dimensional cones can be found by applying the action of W (E7) on the result. As before, we fix the lexicographically smallest representatives. Repeating this process for dimensions 4, 5 and 6, we obtain the list of all cones in Berg(E7), and hence the f-vector of this fan. We now describe the procedure to derive trop(G0) by applying the top horizontal map φ : Berg(E7) → trop(G0). Each ray r in Berg(E7) maps to either (a) 0, (b) a ray of trop(G0), or (c) a positive linear combination of 2 or 3 rays, as listed in [21, §6]. For each ray in case (c), our program iterates through all pairs and triples of rays in trop(G0) and writes the image explicitly as a positive linear combination of rays. With this data, we give a first guess of trop(G0) as follows: for each maximal cone σ = span(ri1, . . . , ri6) of Berg(E7), we write φ(ri1), . . . , φ(ri6) as linear combinations of the rays of trop(G0) and take σ(cid:48) ⊂ TP134 to be the cone spanned by all rays of trop(G0) that appear in the linear combinations. From this we get a list of 6-dimensional cones σ(cid:48). Let Φ ⊂ TP134 be the union of these cones. To certify that Φ = trop(G0) we need to show (1) for each σ ∈ Berg(E7), we have φ(σ) ⊂ σ(cid:48) for some cone σ(cid:48) ⊂ Φ is the union of some φ(σ) for σ ∈ Berg(E7); and (3) the intersection of any two cones σ(cid:48) 1 and σ(cid:48) 2. The claim (1) follows from the procedure of constructing Φ. For (2), one only needs to verify the cases where σ(cid:48) is one of the 9 representatives by the action of W (E7). For each of these, our program produces a list of φ(σ), and we check manually that σ(cid:48) is indeed the union. For (3), one only needs to iterate through the cases where σ(cid:48) 1 is a representative, and the procedure is straightforward. Therefore, our procedure shows that Φ is exactly trop(G0). Then the f -vector is obtained from the list of all cones in the fan Φ. Finally, we recover the list of all cones in Berg(E6) and trop(Y 0) by following the same ⊂ Φ; (2) each cone σ(cid:48) procedure with the left vertical map and the bottom horizontal map. 1, σ(cid:48) 2 in Φ is a face of both σ(cid:48) Remark 3.2. The (reduced) Euler characteristic of the link of Berg(E7) is the alternating sum of the entries of the f-vector of this Bergman fan. We see from Lemma 3.1 that this is 1 − 6091 + 315399 − 3639804+14982660 − 24607800 + 13721400 = 765765 = 1 · 5 · 7 · 9 · 11 · 13 · 17. This is the product of all exponents of W (E7), thus confirming the prediction in [20, (9.2)]. The Naruki fan trop(Y 0) is studied in [21, §6]. Under the action of W (E6) through Berg(E6), it has two classes of rays, labelled type (a) and type (b). It also has two W (E6)- orbits of maximal cones: there are 135 type (aaaa) cones, each spanned by four type (a) rays, and 1080 type (aaab) cones, each spanned by three type (a) rays and one type (b) ray. 12 The map trop(G0) → trop(Y 0) tropicalizes the morphism G0 → Y 0 between very affine K-varieties of dimension 6 and 4. That morphism is the universal family of cubic surfaces. In order to tropicalize these surfaces, we examine the fibers of the map trop(G0) → trop(Y 0). The next lemma concerns the subdivision of trop(Y 0) induced by this map. By definition, this is the coarsest subdivision such that each cone in trop(G0) is sent to a union of cones. Lemma 3.3. The subdivision induced by the map trop(G0) → trop(Y 0) is the barycentric subdivision on type (aaaa) cones. For type (aaab) cones, each cone in the subdivision is a cone spanned by the type (b) ray and a cone in the barycentric subdivision of the (aaa) face. Thus each (aaaa) cone is divided into 24 cones, and each (aaab) cone is divided into 6 cones. Proof. The map π : trop(G0) → trop(Y 0) can be defined via the commutative diagram (3.1): for x ∈ trop(G0), take any point in its preimage in Berg(E7), then follow the left vertical map and the bottom horizontal map to get π(x) in trop(Y 0). It is well-defined because the kernel of the map Berg(E7) → trop(G0) is contained in the kernel of the composition Berg(E7) → Berg(E6) → trop(Y 0). With this, we can compute the image in trop(Y 0) of any cone in trop(G0). For each orbit of cones in trop(Y 0), pick a representative σ, and examine all cones in trop(G0) that map into σ. Their images reveal the subdivision of σ. Figure 4: The 27 trees on tropical cubic surfaces of type (aaaa) Lemma 3.3 shows that each (aaaa) cone of the Naruki fan trop(Y 0) is divided into 24 subcones, and each (aaab) cone is divided into 6 subcones. Thus, the total number of cones in the subdivision is 24× 135 + 6× 1080 = 9720. For the base points in the interior of a cone, the fibers are contained in the same set of cones in trop(G0). The fiber changes continuously as the base point changes. Therefore, moving the base point around the interior of a cone simply changes the metric but not the combinatorial type of marked tropical cubic surface. Corollary 3.4. The map trop(G0) → trop(Y 0) has at most two combinatorial types of generic fibers up to relabeling. Proof. We fixed an inclusion E6 (cid:44)→ E7 in (2.8). The action of StabE6(W (E7)) on the fans is compatible with the entire commutative diagram (3.1). Hence, the fibers over two points 13 ×24×3 that are conjugate under this action have the same combinatorial type. We verify that the 9720 cones form exactly two orbits under this action. One orbit consists of the cones in the type (aaaa) cones, and the other consists of the cones in the type (aaab) cones. Therefore, there are at most two combinatorial types, one for each orbit. Figure 5: The 27 trees on tropical cubic surfaces of type (aaab) We can now derive our classification theorem for tropical cubic surfaces. Proof of Theorem 1.1. We compute the two types of fibers of π : trop(G0) → trop(Y 0). In what follows we explain this for a cone σ of type (aaaa). The computation for type (aaab) is similar. Let r1, r2, r3, r4 denote the rays that generate σ. We fix the vector x = r1 + 2r2 + 3r3 + 4r4 that lies in the interior of a cone in the barycentric subdivision. The fiber π−1(x) is found by an explicit computation. First we determine the directions of the rays. They arise from rays of trop(G0) that are mapped to zero by π. There are 27 such ray directions in π−1(x). These are exactly the image of the 27 type A1 rays in Berg(E7) that correspond to the roots in E7\E6. We label them by Ei, Fij, Gj as in (2.10). Next, we compute the vertices of π−1(x). They are contained in 4-dimensional cones σ(cid:48) = pos{R1, R2, R3, R4} with x ∈ π(σ(cid:48)). The coordinates of each vertex in TP134 is The part of the fiber contained in each cone in trop(G0) is spanned by the vertices and the Ei, Fij, Gj rays it contains. Iterating through the list of cones and looking at this data, we get a list that characterizes the polyhedral complex π−1(x). In particular, that list verifies that π−1(x) is 2-dimensional and has the promised f-vector. For each of the 27 ray directions Ei, Fij, Gj, there is a tree at infinity. It is the link of the corresponding point at infinity π−1(x) ⊂ TP134. The combinatorial types of these 27 trees are shown in Figure 4. The computed by solving y1π(R1) + y2π(R2) + y3π(R3) + y4π(R4) = x for y1, y2, y3, y4. metric on each tree can be computed as follows: the length of a bounded edge equals the lattice distance between the two vertices in the corresponding flap. The surface π−1(x) is homotopy equivalent to its bounded complex. We check directly that the bounded complex is contractible. This can also be inferred from Theorem 4.4. 14 ×12×12×3 Remark 3.5. We may replace x = r1 +2r2 +3r3 +4r4 with a generic point x = x1r1 +x2r2 + x3r3 + x4r4, where x1<x2<x3<x4. This lies in the same cone in the barycentric subdivision, so the combinatorics of π−1(x) remains the same. Repeating the last step over the field Q(x1, x2, x3, x4) instead of Q, we write the length of each bounded edge in the 27 trees in terms of the parameters. Each length either equals x1, x2, x3, x4 or is xi − xj for some i, j. 4 Tropical modifications In Section 2 we computed tropical varieties from polynomial ideals, along the lines of the book by Maclagan and Sturmfels [15]. We now turn to tropical geometry as a self-contained subject in its own right. This is the approach presented in the book by Mikhalkin and Rau [17]. Central to that approach is the notion of tropical modification. In this section we explain how to construct our tropical del Pezzo surfaces from the plane R2 by modifications. This leads to proofs of Proposition 1.2 and Theorem 1.1 purely within tropical geometry. Tropical modification is an operation that relates topologically different tropical models of the same variety. This operation was first defined by Mikhalkin in [16]; see also [17, Chapter 5]. Here we work with a variant known as open tropical modifications. These were introduced in the context of Bergman fans of matroids in [23]. Brugall´e and Lopez de Medrano [7] used them to study intersections and inflection points of tropical plane curves. → Y ⊂ Rn+1 is a new tropical variety to be described below. One should think of Y (cid:48) as where Y (cid:48) being an embedding of the complement of a divisor in Y into a higher-dimensional torus. We fix a tropical cycle Y in Rn, as in [17]. An open modification is a map p : Y (cid:48) Consider a piecewise integer affine function g : Y → R. The graph Γg(Y ) = (cid:8)(y, g(y)) y ∈ Y(cid:9) ⊂ Rn+1 is a polyhedral complex which inherits weights from Y . However, it usually not balanced. There is a canonical way to turn Γg(Y ) into a balanced complex. If Γg(Y ) is unbalanced around a codimension one face E, then we attach to it a new unbounded facet FE in direction −en+1. (We here use the max convention, as in [17]). The facet FE can be equipped with a unique weight wFE ∈ Z such that the complex obtained by adding FE is balanced at E. The resulting tropical cycle is Y (cid:48) ⊂ Rn+1. By definition, the open modification of Y given by g is the map p : Y (cid:48) The tropical divisor divY (g) consists of all points y ∈ Y such that p−1(y) is infinite. This is a polyhedral complex. It inherits weights on its top-dimensional faces from those of Y (cid:48). A tropical cycle is effective if the weights of its top-dimensional faces are positive. Therefore, the cycle Y (cid:48) is effective if and only if Y and the divisor divY (g) are effective. Given a tropical variety Y and an effective divisor divY (g), we say the tropical modification p : Y (cid:48) → Y is along divY (g). See [16, 17, 23] for basics concerning cycles, divisors and modifications. Open tropical modifications are related to re-embeddings of classical varieties as follows. → Y , where p comes from the projection Rn+1 → Rn with kernel Ren+1. Fix a very affine K-variety X ⊂ ( ¯K∗)n and Y = trop(X) ⊂ Rn. Given a polynomial function f ∈ K[X], let D be its divisor in X. Then X\D is isomorphic to the graph of the restriction of f to X\D. In this manner, the function f gives a closed embedding of X\D into ( ¯K∗)n+1. For the next proposition we require the tropicalization of a variety to be locally irreducible. 15 Let y be a point in a tropical variety Y , then Stary(Y ) = {y (cid:48) ∃  > 0 s.t. ∀ 0 < δ <  : y + y (cid:48) ∈ Y }, is a balanced tropical fan with weights inherited from Y . A tropical variety Y is locally irreducible if at every point y ∈ Y , we have that Stary(Y ) is not a proper union of two tropical varieties, taking weights into consideration. Proposition 4.1. Let X ⊂ ( ¯K∗)n be a very affine variety. For a function f ∈ K[X], let D be the divisor divX(f ), and let X(cid:48) = X\D ⊂ ( ¯K∗)n+1 denote the graph of X along f as described above. Let Y = trop(X) ⊂ Rn and Y (cid:48) = trop(X(cid:48)) ⊂ Rn+1. Suppose that Y is locally irreducible, then there exists a piecewise integer affine function g : Y → R such that divY (g) = trop(D) and the coordinate projection Y (cid:48) → Y is the open modification of Y along that divisor. Proof. The coordinate projection p : Rn+1 → Rn takes Y (cid:48) onto Y , since tropicalization acts coordinate-wise. We claim that the fiber over a point y ∈ Y is either a single point or a half-line in the −en+1 direction. The fiber, p−1(y) is 1-dimensional and closed in Y (cid:48). Let y(cid:48) be an endpoint of a connected component of p−1(y). Then p(Stary(cid:48)(Y (cid:48))) has the same dimension as Y . Since otherwise, Stary(cid:48)(Y (cid:48)) contains a space of linearity in the direction en+1 and y(cid:48) cannot be an endpoint of the fiber. If the one dimensional fiber p−1(y) contains two endpoints y1 and y2 then Y must be reducible at y; it can be split into more than one component coming from the projection of p(Stary1(Y (cid:48))) and p(Stary2(Y (cid:48))). Therefore, p−1(y) consists of either a single point, a line, or a half line. However, since f ∈ K[X] is a regular function, the fiber of a point y ∈ Y cannot be unbounded in the +en+1 direction. Thus the only possibilities are that p−1(y) is a single point or a half line in the −en+1 direction. Finally, we obtain the piecewise integer affine function g by taking g(y) = p−1(y) for y ∈ Y \trop(D) and then extending by continuity to the rest of Y . Then Y (cid:48) is the modification Any two tropical rational functions g and g(cid:48) that define the same tropical divisor on Y must differ by a map which is integer affine on Y , see [1, Remark 3.6]. This leads to the following corollary. Corollary 4.2. Under the assumptions of Proposition 4.1, the tropicalization of X(cid:48) = X\D ⊂ Rn+1 is determined uniquely by those of D and X, up to an integer affine map. In general, trop(X(cid:48)) is not determined by the tropical hypersurface of f ∈ K[X], as the tropicalization of the divisor D = divX(f ) may differ from the tropical stable intersection of trop(X(cid:48)) and that tropical hypersurface. Examples 4.2 and 4.3 of [7] demonstrate both this and that Proposition 4.1 can fail without the locally irreducibility hypothesis. ⊂ ( ¯K∗)n+k is obtained from X ⊂ ( ¯K∗)n by taking the graph of a along the function g described above. Suppose now that X(cid:48) list of k ≥ 2 polynomials f1, f2, . . . , fk. This gives us a sequence of projections (cid:48) X = Xk → Xk−1 → ··· → X2 → X1 → X0 = X, (4.1) where Xi ⊂ ( ¯K∗)n+i is obtained from X by taking the graph of (f1, . . . , fi). We further get a corresponding sequence of projections of the tropicalizations: (cid:48) trop(X ) = trop(Xk) → trop(Xk−1) → ··· → trop(X0) = trop(X). (4.2) 16 Figure 6: The tropical divisors in Example 4.3. The positions of trop(G1 ∩ H1) in trop(X1) for three choices of a are marked on the downward purple edge. For a = 1 we get M0,5. We may ask if it is possible to recover trop(X(cid:48)) ⊂ Rn+k just from trop(X) and the k tropical divisors trop(Di) considered in trop(X). The answer is “yes” in the special case when trop(X) = Rn and the arrangement of divisors trop(Di) are each locally irreducible and intersect properly, meaning the intersection of any l of these divisors has codimension l. However, in general, iterating modifications to recover trop(X(cid:48)) is a delicate procedure. In most cases, the outcome is not solely determined by the configuration of tropical divisors in trop(X), even if the divisors intersect pairwise properly. We illustrate this by deriving the degree 5 del Pezzo surface trop(M0,5). Example 4.3. This is a variation on [23, Example 2.29]. Let X = ( ¯K∗)2 and consider the functions f (x) = x1 − 1, g(x) = x2 − 1 and h(x) = ax1 − x2, for some constant a ∈ K∗ with val(a) = 0. Denote divX(f ) by F , and analogously for G and H. The tropicalization of each divisor is a line through the origin in R2. The directions of trop(F ), trop(G), and trop(H) are (1, 0), (0, 1), and (1, 1) respectively. Let X(cid:48) ⊂ ( ¯K∗)5 denote the graph of X along the three functions f, g, and h, in that order. This defines a sequence of projections, (cid:48) X −→ X2 −→ X1 −→ X = ( ¯K ∗ )2. Here, X2 = {(x1, x2, x1 − 1, x2 − 1)} ⊂ ( ¯K∗)4. The tropical plane trop(X2) contains the face σ = {0}×{0}× (−∞, 0]× (−∞, 0], corresponding to points with val(x1) = val(x2) = 0. Let H2 denote the graph of f and g restricted to H. This is a line in 4-space, namely, ∗ )4. The tropical line trop(H2) depends on the valuation of a − 1. It can be determined from H2 = {(x1, ax1, x1 − 1, ax1 − 1)} ⊂ X2 ⊂ ( ¯K trop(G1 ∩ H1) = (cid:8)(cid:0)0, 0,−val( a − 1)(cid:1)(cid:9). 1 Here, H1, G1 denote the graph of f restricted to H and G, respectively. Figure 6 shows the possibilities for trop(G1 ∩ H1) in trop(X1), and Figure 7 shows trop(H2) ∩ σ in trop(X2). a − 1, for instance by taking a = 1+tv when K = C{{t}}. In these cases, the tropical plane trop(X(cid:48)) is not a fan. However, We can prescribe any value v ∈ (0,∞) for the valuation of 1 1 17 v = 0 v ∈ (0,∞) v = ∞ Figure 7: The different possibilities for trop(H2) ∩ σ in Example 4.3 it becomes a fan when v moves to either endpoint of the interval [0,−∞]. For instance, v = 0 a is not equal to 1 and trop(X(cid:48)) is the fan obtained from happens when the constant term of 1 R2 by carrying out the modifications along the pull-backs of the tropical divisors on the left in Figure 6. See Definition 2.16 of [23] for pull-backs of tropical divisors. The other extreme is when a = 1. Here, F, G, H are concurrent lines in ( ¯K∗)2, and trop(H2) contains a ray in the direction e3 +e4. Upon modification, we obtain the fan over the Petersen graph in Figure 3. This is the tropicalization of the degree 5 del Pezzo surface in (2.2). Thus beginning from the tropical divisors trop(F ), trop(G), and trop(H) in R2, we recover trop(M0,5) if we know that they represent tropicalizations of concurrent lines in ( ¯K∗)2. The open tropical modification described above represents the tropicalization of the very affine variety M0,5. In this case the compactification of M0,5 which produces the del Pezzo surface of degree 5 is indeed the tropical compactification given by the fan trop(M0,5), (see [15, §6.4] for an introduction to tropical compactifications). There is no direct connection between open tropical modifications and birational transformations, the link depends a choice of compactification of the very affine variety. Upon removing divisors one can find more interesting compactifications of the complement. For example, (K∗)2 cannot be compactified to a del Pezzo surface of degree less than 6, but upon deleting the three divisors above one ♦ can compactify the complement to a del Pezzo surface of degree 5. We now explain how this extends to a del Pezzo surface X of degree d ≤ 4. As before, we write X 0 for the complement of the (−1)-curves in X. Then X(cid:48) = X 0 is obtained from ( ¯K∗)2 by taking the graphs of the polynomials f1, . . . , fk of the curves in ( ¯K∗)2 that give rise to (−1)-curves on X. More precisely, fix p1 = (1 : 0 : 0), p2 = (0 : 1 : 0) , p3 = (0 : 0 : 1), p4 = (1 : 1 : 1), and take p5, . . . , p9−d to be general points in P2. If d = 4 then there is only one extra point p5, we have k = 8 in (4.1), and f1, . . . , f8 are the polynomials defining F14, F15, F24, F25, F34, F35, F45, G. (4.3) For d = 3, there are two extra points p5, p6 in X, we have k = 18, and f1, . . . , f18 represent F14, F15, F16, F24, F25, F26, F34, F35, F36, F45, F46, F56, G1, G2, G3, G4, G5, G6. (4.4) We write Pi = trop(pi) ∈ TP2 for the image of the point pi under tropicalization. The tropical points P1, P2, . . . are in general position if any two lie in a unique tropical line, these lines are distinct, any five lie in a unique tropical conic, and these conics are distinct in TP2. 18 Figure 8: The tropical conic and the tropical lines determined by the 5 points for a marked del Pezzo surface of degree 4. The diagram is drawn in R2 on the left and in TP2 on the right. The 16 trivalent trees corresponding to the (−1)-curves of the del Pezzo surface, seen at the nodes in in Figure 1, arise from the plane curves shown here by tropical modifications. A configuration in general position for d = 4 is shown in Figure 8. Our next result implies that the colored Clebsch graph in Figure 1 can be read off from Figure 8 alone. For d = 3, in order to recover the tropical cubic surface from the planar configuration, the points Pi must satisfy further genericity assumptions, to be revealed in the proof of the next theorem. Theorem 4.4. Fix d ∈ {3, 4, 5} and points p1, . . . , p9−d in P2 whose tropicalizations Pi are sufficiently generic in TP2. The tropical del Pezzo surface trop(X 0) can be constructed from TP2 by a sequence of open modifications that is determined by the points P1, . . . , P9−d. Proof. The sequence of tropical modifications we use to go from R2 to trop(X 0) is determined if we know, for each i, the correct divisor on each (−1)-curve C in the tropical model trop(Xi). Then, the preimage of C in the next surface trop(Xi+1) is the modification C(cid:48) of the curve C along that divisor. By induction, each intermediate surface trop(Xi) is locally irreducible, since it is obtained by modifying a locally irreducible surface along a locally irreducible divisor. With this, Theorem 4.4 follows from Proposition 4.1, applied to both the i-th surface and its (−1)-curves. The case d = 5 was covered in Example 4.3. From the metric tree that represents the boundary divisor C of X 0 we can derive the corresponding trees on each intermediate surface trop(Xi) by deleting leaves. Thus, to establish Theorem 4.4, it suffices to prove the following claim: the final arrangement of the (16 or 27) metric trees on the tropical del Pezzo surface trop(X 0) is determined by the locations of the points Pi in TP2. Consider first the case d = 4. The points P4 and P5 determine an arrangement of plane tropical curves (4.3) as shown in Figure 8. The conic G through all five points looks like an “inverted tropical line”, with three rays in directions P1, P2, P3. By the genericity assumption, the points P4 and P5 are located on distinct rays of G. These data determine a trivalent metric tree with five leaves, which we now label by E1, E2, E3, E4, E5. Namely, P4 forms a cherry together with the label of its ray, and ditto for P5. For instance, in Figure 8, the cherries on the tree G are {E1, E4} and {E2, E5}, while E3 is the non-cherry leaf. This is precisely the tree sitting on the node labeled G in Figure 1. The lengths of the two bounded 19 P4P5GF34F35F14F45F15F24F25P1P2P3 edges of the tree G are the distances from P4 resp. P5 to the unique vertex of the conic G in R2. Thus the metric tree G is easily determined from P4 and P5. The other 15 metric trees can also be determined in a similar way from the configuration of points and curves in R2 and by performing a subset of the necessary modifications. Alternatively, we may use the transition rules (1.1) and (1.2) to obtain the other 15 trees from G. This proves the above claim, and hence Theorem 4.4, for del Pezzo surfaces of degree d = 4. Consider now the case d = 3. Here the arrangement of tropical plane curves in R2 ⊂ TP2 consists of three lines at infinity, F12, F13, F23, nine straight lines, F14, F15, . . . , F36, three honest tropical lines, F45, F46, F56, three conics that are “inverted tropical lines” G4, G5, G6, and three conics with one bounded edge, G1, G2, G3. Each of these looks like a tree already in the plane, and it gets modified to a 10-leaf tree, like to ones in Figures 4 and 5. We claim that these labeled metric trees are uniquely determined by the positions of P4, P5, P6 in R2. Consider one of the 9 straight lines in our arrangement, say, F14. If the points P4, P5, P6 are generically chosen, 7 of the 10 leaves on the tree Fij can be determined from the diagram in R2. These come from the 7 markings on the line F14 given by E1, E4, F23, F25, F26, F35, F36. The markings E1 and F23 are the points at infinity, the marking E4 is the location of point P4, and the markings F25, F26, F35, F36 are the points of intersection with those lines. Under our hypothesis, these 7 marked points on the line F14 will be distinct. With this, F14 is already a metric caterpillar tree with 7 leaves. The three markings which are missing are G1, G4 and F56. Depending on the positions of P4, P5, P6, the intersection points of these three curves with the line F14 may coincide with previously marked points. Whenever this happens, the position of the additional marking on the tree F14 can be anywhere on the already attached leaf ray. Again, the actual position of the point on that ray may be determined by performing modifications along those curves. Alternatively, we use the involution given in Corollary 2.4. The involution on the ten leaves of the desired tree F14 is E1 ↔ G4, E4 ↔ G1, F23 ↔ F56, F25 ↔ F36, F26 ↔ F35. Since the involution exchanges each of the three unknown leaves with one of the seven known leaves, we can easily construct the final 10-leaf tree from the 7-leaf caterpillar. A similar argument works the other six lines Fij, and the conics G4, G5, G6. In these cases, 8 of the 10 marked points on a tree are determined from the arrangement in the plane, provided the choice of points is generic. Finally, the conics G1, G2, G3 are dual to subdivisions of lattice parallelograms of area 1. They may contain a bounded edge. Suppose no point Pj lies on the bounded edge of the conic Gi, then the positions of all 10 marked points of the tree are visible from the arrangement in the plane. If Gi does contain a marked point Pj on its bounded edge, then the tropical line Fij intersects Gi in either a bounded edge or a single point with intersection multiplicity 2, depending on the dual subdivision of Gi. In the first case the position of the marked point Fij is easily determined from the involution; the distance from a vertex of the bounded edge of Gi to the marked point Fij must be equal to the distance from Pj to the opposite vertex of the bounded edge of Gi. If Gi ∩ Fij is a single point of intersection multiplicity two, then Pj and Fij form a cherry on the tree Gi which is invariant under the involution. We claim that this cherry attaches to the rest of the tree at a 4-valent vertex. The involution on the 10-leaf tree can also be seen as a tropical double cover from our 10-leaf tree to a 5-leaf tree, h : T → t, where the 5-leaf 20 tree t is labeled with the pair of markings interchanged by the involution. As mentioned in Corollary 2.4, this double cover comes from the classical curve in the del Pezzo surface X. In particular, the double cover locally satisfies the tropical translation of the Riemann-Hurwitz condition [5, Definition 2.2]. In our simple case of a degree 2 map between two trees, this local condition for a vertex v of T is deg(v)− dh,v(deg(h(v))− 2)− 2 ≥ 0, where deg denotes the valency of a vertex, and dh,v denotes the local degree of the map h at v. Suppose the two leaves did not attach at a four valent vertex, then they form a cherry, this cherry attaches to the rest of the tree by an edge e which is adjacent to another vertex v of the tree. The Riemann-Hurwitz condition is violated at v, since deg(v) = deg(h(v)) = 3 and dh,v = 2. We conclude that the tree arrangement can be recovered from the position of the points in R2. Therefore it is also possible to recover the tropical del Pezzo surface P1, P2, . . . trop(X 0) by open modifications. In each case, we recover the corresponding final 10 leaf tree from the arrangement in TP2 plus our knowledge of the involution in Corollary 2.4. Remark 4.5. Like in the case d = 4, knowledge of transition rules among the 27 metric trees on trop(X 0) can greatly simplify their reconstruction. We give such a rule in Proposition 5.2. In this section we gave a geometric construction of tropical del Pezzo surfaces of degree d ≥ 3, starting from the points P1, . . . , P9−d in the tropical plane TP2. The lines and conics in TP2 that correspond to the (−1)-curves are transformed, by a sequence of open tropical modifications, into the trees that make up the boundary of the del Pezzo surface. Knowing these well-specified modifications of curves ahead of time allows us to carry out a unique sequence of open tropical modifications of surfaces, starting with R2. In each step, going from right to left in (4.1), we modify the surface along a divisor given by one of the trees. This gives a geometric construction for the bounded complex in a tropical del Pezzo surface: it is the preimage under (4.1) of the bounded complex in the arrangement in R2. For instance, Figure 2 is the preimage of the parallelogram and the four triangles in Figure 8. The same modification approach can be used to construct (the bounded complexes of) any tropical plane in TPn from its tree arrangement. This provides a direct link between the papers [12] and [23]. That link should be useful for readers of the text books [15] and [17]. 5 Tropical Cubic Surfaces and their 27 Trees This section is devoted to the combinatorial structure of tropical cubic surfaces. Throughout, X is a smooth del Pezzo surface of degree 3, without Eckhart points, and X 0 the very affine surface obtained by removing the 27 lines from X. Recall that an Eckhart point is an ordinary triple point in the union of the (−1)-curves. Going well beyond the summary statistics of Theorem 1.1, we now offer an in-depth study of the combinatorics of the surface trop(X 0). We begin with the construction of trop(X 0) from six points in TP2, as in Section 4. The points P5 and P6 are general in R2 ⊂ TP2. The first four points are the coordinate points P1 = (0 : −∞ : −∞), P2 = (−∞ : 0 : −∞), P3 = (−∞ : −∞ : 0), P4 = (0 : 0 : 0). (5.1) Theorem 4.4 tells us that trop(X 0) is determined by the locations of P5 and P6 when the points are generically chosen. There are two generic types, namely (aaaa) and (aaab), as 21 Figure 9: Markings of a conic G1 which produce trees of type (aaab). 2 Figure 10: Markings of a conic G1 which produce trees of type (aaaa). shown in Figures 4 and 5. This raises the question of how the type can be decided from the positions of P5 and P6. To answer that question, we shall use tropical convexity [15, §5.2]. There are five generic types of tropical triangles, depicted here in Figures 11 and 12. The unique 2-cell in such a tropical triangle has either 3, 4, 5 or 6 vertices. Two of these have 4 vertices, but only one type contains a parallelogram. That is the type which gives (aaaa). Theorem 5.1. Suppose that the tropical cubic surface constructed as in Theorem 4.4 has one of the two generic types. Then it has type (aaaa) if and only if the 2-cell in the tropical triangle spanned by P4, P5 and P6 is a parallelogram. In all other cases, it has type (aaab). Note that the condition that the six points Pi are in general position is not sufficient to imply that the tropical cubic surface is generic. In some cases, the corresponding point in the Naruki fan trop(Y 0) will lie on the boundary of the subdivision induced by the map from trop(G0), as described in Section 3 and below. If so, the tropical cubic surface is degenerate. Proof of Theorem 5.1. The tree arrangements for the two types of generic surfaces consist of distinct combinatorial types, i.e. there is no overlap in Figures 4 and 5. Therefore, when the tropical cubic surface is generic, it is enough to determine the combinatorial type of a single tree. We do this for the conic G1. Given our choices of points (5.1) in TP2, the tropical conic G1 is dual to the Newton polygon with vertices (0, 0), (1, 0), (0, 1), and (1, 1). The triangulation has one interior edge, either of slope 1 or of slope −1. We claim the following: The tropical cubic surface trop(X 0) has type (aaaa) if and only if the following holds: 1. The bounded edge of the conic G1 has slope −1 and contains a marked point Pj, or 22 2. the bounded edge of the conic G1 has slope 1 and contains a marked point Pj, and the other two points Pj, Pk lie on opposite sides of the line spanned by the bounded edge. To show this, we follow the proof of Theorem 4.4. For each configuration of P4, P5, P6 on the conic G1, we draw lines with slope 1 through these points. These are the tropical lines F14, F15, F16. Each intersects G1 at one further point. These are the images of E4, E5, E6 under the tree involution, i.e. the points labeled F14, F15, F16 on the tree G1. Together with E2, E3, F12 and F13 lying at infinity of TP2, we can reconstruct a tree with 10 leaves. Then, we can identify the type of the tree arrangement. We did this for all possible configurations up to symmetry. Some of the results are shown in Figures 9 and 10. The claim follows. To derive the theorem from the claim, we must consider the tropical convex hull of the points P4, P5, P6 in the above cases. As an example, the 2-cells of the tropical triangle corresponding to the trees in Figures 9 and 10 are shown in Figures 11 and 12 respectively. The markings of G1 producing a type (aaaa) tree always give parallelograms. Finally, if the marking of a conic produces a type (aaab) tree then the 2-cell may have 3, 4, 5, or 6 vertices. However, if it has 4 vertices, then it is a trapezoid with only one pair of parallel edges. We next discuss some relations among the 27 boundary trees of a tropical cubic surface X. Any pair of disjoint (−1)-curves on X meets exactly five other (−1)-curves. Thus, two 10- leaf trees T and T (cid:48) representing disjoint (−1)-curves have exactly five leaf labels in common. Let t and t(cid:48) denote the 5-leaf trees constructed from T and T (cid:48) as in the proof of Proposition 2.4. Thus T double-covers t, and T (cid:48) double-covers t(cid:48). Given a subset E of the leaf labels of a tree T , we write TE for the subtree of T that is spanned by the leaves labeled with E. Proposition 5.2. Let T and T (cid:48) be the trees corresponding to disjoint (−1)-curves on a cubic surface X, and E the set of five leaf labels common to T and T (cid:48). Then t = T (cid:48) E and t(cid:48) = TE. Proof. The five lines that meet two disjoint (−1)-curves C and C(cid:48) define five points on C and five tritangent planes containing C(cid:48). The cross-ratios among the former are equal to the cross-ratios among the latter modulo C(cid:48), see [18, Section 4]. The proposition follows because the metric trees can be derived from the valuations of all the various cross ratios. Proposition 5.2 suggests a combinatorial method for recovering the entire arrangement of 27 trees on trop(X 0) from a single tree T . Namely, for any tree T (cid:48) that is disjoint from Figure 11: The tropical triangles formed by points on G1 as in Figure 9, giving type (aaab). 23 Figure 12: The tropical triangles formed by points on G1 as in Figure 10, giving type (aaaa). T , we can recover both t(cid:48) and T (cid:48) from both T and T (cid:48), with labels Ei common with T , we can determine T (cid:48) T (cid:48) is an amalgamation of t(cid:48), T (cid:48) reminiscent of a tree building algorithm in phylogenetics known as quartet puzzling [9]. E. Moreover, for any of the 10 trees Ti that are disjoint Ei as well. Then E, and the 10 subtrees T (cid:48) Ei. This amalgamation process is We next examine tropical cubic surfaces of non-generic types. These surfaces are obtained from non-generic fibers of the vertical map on the right in (3.1). We use the subdivision of the Naruki fan trop(Y 0) described in Lemma 3.3. There are five types of rays in this subdivision. We label them (a), (b), (a2), (a3), (a4). A ray of type (ak) is a positive linear combination of k rays of type (a). The new rays (a2), (a3), (a4) form the barycentric subdivision of an (aaaa) cone. With this, the maximal cones in the subdivided Naruki fan are called (aa2a3a4) and (aa2a3b). They are known as the generic types (aaaa) and (aaab) in the previous sections. A list of all 24 cones, up to symmetry, is presented in the first column of Table 1. The fiber of trop(G0) → trop(Y 0) over any point in the interior of a maximal cone is a tropical cubic surface. However, some special fibers have dimension 3. Such fibers contain infinitely many tropical cubic surfaces, including those with Eckhart points. Removing such Eckhart points is a key issue in [11]. We do this by considering the stable fiber, i.e. the limit of the generic fibers obtained by perturbing the base point by an infinitesimal. Alternatively, the tree arrangement of the stable fiber is found by setting some edge lengths to 0 in Remark 3.5. We computed representatives for all stable fibers. Our results are shown in Table 1. We explain the two simplest non-trivial cases. The 36 type (a) rays in the Naruki fan are Figure 13: The bounded complex of the tropical cubic surface of type (a) 24 QPE1F35E3F15E5F13F24G6F26G4F46G2 Figure 14: The 27 trees on the tropical cubic surface of type (a) Figure 15: The bounded complex of the tropical cubic surface of type (b) in bijection with the 36 positive roots of E6. Figure 13 shows the bounded cells in the stable fiber over the (a) ray corresponding to root r = d1 +d3 +d5. It consists of six triangles sharing a common edge. The two shared vertices are labeled by P and Q. Recall the identification of the roots of E6 involving d7 with the 27 (−1)-curves from (2.10). Then, considering Ei, Fij and Gi as roots of E6, exactly 15 of them are orthogonal to r. The other 12 roots are E1, F35; E3, F15; E5, F13; F24, G6; F26, G4; F46, G2. (5.2) These form a Schlafli double six. The 36 double six configurations on a cubic surface are in bijection with the 36 positive roots of E6. Each of the six pairs forms an A2 subroot system with d1 + d3 + d5. The non-shared vertices in the (a) surface are labeled by these pairs. The 12 rays labeled by (5.2) emanate from Q, and the other 15 rays emanate from P . Each other vertex has 7 outgoing rays, namely its labels in Figure 13 and the 5 roots orthogonal to both of these. Figure 14 shows the resulting 27 = 12 + 15 trees at infinity. The 40 type (b) rays in the Naruki fan are in bijection with the type A subroot systems ×3 2 Figure 16: The 27 trees on the tropical cubic surface of type (b) 25 ×15×12P1P2P3d1−d3d1+d2+d5d2+d3+d5d2−d5d2+d4+d6d4+d5+d6d4−d6d1+d3+d4d1+d3+d6×27 Type #cones in moduli Vertices Edges Rays Triangles Squares Flaps Cones 0 (a) (a2) (a3) (a4) (b) (aa2) (aa3) (aa4) (a2a3) (a2a4) (a3a4) (ab) (a2b) (a3b) (aa2a3) (aa2a4) (aa3a4) (a2a3a4) (aa2b) (aa3b) (a2a3b) (aa2a3a4) (aa2a3b) 1 36 270 540 1620 40 540 1620 540 1620 810 540 360 1080 1080 3240 1620 1620 1620 2160 3240 3240 3240 6480 1 8 20 37 59 12 23 43 68 43 71 68 26 45 69 46 74 74 74 48 75 75 77 78 0 13 37 72 118 21 42 82 133 82 138 133 48 86 135 87 143 143 143 91 145 145 148 150 27 69 108 144 177 81 114 156 195 156 201 195 123 162 198 162 207 207 207 168 210 210 213 216 0 6 14 24 36 10 13 22 33 22 32 33 16 24 34 21 31 31 31 23 32 32 30 31 0 0 4 12 24 0 7 18 33 18 36 33 7 18 33 21 39 39 39 21 39 39 42 42 0 42 81 117 150 54 87 129 168 129 174 168 96 135 171 135 180 180 180 141 183 183 186 189 135 135 135 135 135 135 135 135 135 135 135 135 135 135 135 135 135 135 135 135 135 135 135 135 Table 1: All combinatorial types of tropical cubic surfaces in E6. Figure 15 illustrates the stable fiber over a point lying on the ray corresponding to d1 − d3, d1 + d2 + d5, d2 + d3 + d5, d2 − d5, d2 + d4 + d6, d4 + d5 + d6, d4 − d6, d1 + d3 + d4, d1 + d3 + d6. (5.3) This is the union of three type A2 subroot systems that are pairwise orthogonal. The bounded complex consists of 10 triangles. The central triangle P1P2P3 has 3 other triangles attached to each edge. The 9 pendant vertices are labeled with the roots in (5.3). The 3 vertices in the triangles attached to the same edge are labeled with 3 roots in a type A2 subroot system. Each of P1,P2 and P3 is connected with 9 rays, labeled with the roots in E7\E6 that are orthogonal to a type A2 subroot system in (5.3). Each of the other vertices is connected with 6 rays. The labels of these rays are the roots in E7\E6 that are orthogonal to the label of that vertex but are not orthogonal to the other two vertices in the same group. All of the 27 trees are isomorphic, as shown in Figure 16. In each tree, the 10 leaves are partitioned into 10 = 4 + 3 + 3, by orthogonality with the type A2 subroot systems in (5.3). The bounded part of the tree is connected by two flaps to two edges containing the same Pi. 26 We close this paper with a brief discussion of open questions and future directions. One obvious question is whether our construction can be extended to del Pezzo surfaces of degree d = 2 and d = 1. In principle, this should be possible, but the complexity of the algebraic and combinatorial computations will be very high. In particular, the analogues of Theorem 4.4 for 7 and 8 points in TP2 are likely to require rather complicated genericity hypotheses. For d = 4, we were able compute the Naruki fan trop(Y 0) without any prior knowledge, by just applying the software gfan to the 45 trinomials in Proposition 2.1. We believe that the same will work for d = 3, and that even the tropical basis property [15, §2.6] will hold: Conjecture 5.3. The 270 trinomial relations listed in Proposition 2.2 form a tropical basis. This paper did not consider embeddings of del Pezzo surfaces into projective spaces. However, it would be very interesting to study these via the results obtained here. For cubic surfaces in P3, we should see a shadow of Table 1 in TP3. Likewise, for complete intersections of two quadrics in P4, we should see a shadow of Figures 1 and 2 in TP4. One approach is to start with the following tropical modifications of the ambient spaces TP3 resp. TP4. Consider a graded component in (2.1) with L very ample. Let N + 1 be the number of monomials in Ei, Fij, Gk that lie in H 0(X,L). The map given by these monomials embeds X into a linear subspace of PN . The corresponding tropical surfaces in TPN should be isomorphic to the tropical del Pezzo surfaces constructed here. In particular, if L = −K is the anticanonical bundle, then the subspace has dimension d, and the ambient dimensions are N = 44 for d = 3, and N = 39 for d = 4. In the former case, the 45 monomials (like E1F12G2 or F12F34F56) correspond to Eckhart triangles. In the latter case, the 40 monomials (like E1E2F12G or E1F12F13F45) are those of degree (4, 2, 2, 2, 2, 2) in the grading (2.5). The tropicalizations of these combinatorial anticanonical embeddings, X ⊂ P3 ⊂ P44 for d = 3 and X ⊂ P4 ⊂ P39 for d = 4, should agree with our surfaces here. This will help in resolving remaining issues surrounding the excess of lines in tropical cubic surfaces. Examples of the superabundance of tropical lines on generic smooth tropical cubic hypersurfaces were first found by Vigeland [27] and these examples were later considered in [6] and [8]. One last consideration concerns cubic surfaces defined over R. A cubic surface equipped with a real structure induces another involution on the 27 metric trees corresponding to real (−1)-curves. These trees already come partitioned by combinatorial type, depending on the type of tropical cubic surface. One could ask which trees can result from real lines, and whether the tree arrangement reveals Segre’s partition of real lines on cubic surfaces into hyperbolic and elliptic types [22]. For example, for the (aaaa) and (aaab) types, if the involution on the trees from the real structure is the trivial one, then the trees with combinatorial type occurring exactly three times always correspond to hyperbolic real lines. Acknowledgements. This project started during the 2013 program on Tropical Geometry and Topology at the Max-Planck Institut fur Mathematik in Bonn, with Kristin Shaw and Bernd Sturmfels in residence. Qingchun Ren and Bernd Sturmfels were supported by NSF grant DMS-0968882. Kristin Shaw had support from the Alexander von Humboldt Foundation in the form of a Postdoctoral Research Fellowship. We are grateful to Maria Angelica Cueto, Anand Deop- urkar and also an anonymous referee for helping us to improve this paper. 27 References [1] L. Allermann and J. Rau: First steps in tropical intersection theory, Mathematische Zeitschrift, 264(3) (2010), 633–670. [2] F. Ardila and C. Klivans: The Bergman complex of a matroid and phylogenetic trees, Journal of Combinatorial Theory. Series B, 96(1) (2006) 38–49. [3] F. Ardila, R. Reiner and L. Williams: Bergman complexes, Coxeter arrangements, and graph associa- hedra, S´eminaire Lotharingien de Combinatoire, 54A (2006) Article B54Aj. [4] V. Batyrev and O. Popov: The Cox ring of a del Pezzo surface, In Arithmetic of higher-dimensional algebraic varieties, Progress in Mathematics, Birkhauser, Boston, 226 (2004) 85–103. [5] B. Bertrand, E. Brugall´e and G. Mikhalkin: Tropical open Hurwitz numbers, Rendiconti del Seminario Matematico della Universit`a di Padova, 125 (2011) 157–171. [6] T. Bogart and E. Katz: Obstructions to lifting tropical curves in surfaces in 3-space, SIAM Journal on Discrete Mathematics, 26(3) (2012) 1050–1067. [7] E. Brugall´e and L. Lopez de Medrano. Inflection points of real and tropical plane curves, Journal of Singularities 3 (2012) 74–103. [8] E. Brugall´e and K. Shaw: Obstructions to approximating tropical curves in surfaces via intersection theory, Canadian Journal of Mathematics, to appear, arXiv:1110.0533. [9] D. Bryant and M. Steel: Constructing optimal trees from quartets, Journal of Algorithms 38 (2001) 237–259. [10] E. Feichtner and B. Sturmfels: Matroid polytopes, nested sets and Bergman fans, Portugaliae Mathe- matica 62 (2005) 437–468. [11] P. Hacking, S. Keel and J. Tevelev: Stable pair, tropical, and log canonical compactifications of moduli spaces of del Pezzo surfaces, Inventiones Mathematicae 178 (2009) 173–227. [12] S. Herrmann, A. Jensen, M. Joswig and B. Sturmfels: How to draw tropical planes, Electronic Journal of Combinatorics 16(2) (2009) R6. [13] A. Jensen: Gfan, a software system for Grobner fans and tropical varieties, Available at http://home. imf.au.dk/jensen/software/gfan/gfan.html. [14] S. Keel: Intersection theory of moduli space of stable n-pointed curves of genus zero, Transactions of the American Mathematical Society 330(2) (1992) 545–574. [15] D. Maclagan and B. Sturmfels: Introduction to Tropical Geometry, Graduate Studies in Mathematics, Vol 161, American Mathematical Society, 2015. [16] G. Mikhalkin: Tropical geometry and its applications, International Congress of Mathematicians. Vol. II, European Mathematical Society, Zurich, (2006) 827–852. [17] G. Mikhalkin and J. Rau: Tropical Geometry, in preparation, preliminary version available at https://www.dropbox.com/s/g3ehtsoyy3tzkki/main.pdf. [18] I. Naruki: Cross ratio variety as a moduli space of cubic surfaces, Proceedings of the London Mathe- matical Society. Third Series, 45(1) (1982) 1–30. [19] F. Rincon: Computing tropical linear spaces, Journal of Symbolic Computation 51 (2013) 86–98. [20] Q. Ren, S. Sam, G. Schrader and B. Sturmfels: The universal Kummer threefold, Experimental Math- ematics 22 (2013) 327–362. [21] Q. Ren, S. Sam and B. Sturmfels: Tropicalization of classical moduli spaces, Mathematics in Computer Science, Special Issue on Computational Algebraic Geometry, 8 (2014) 119–145. [22] B. Segre: The Non-Singular Cubic Surfaces. A new method of Investigation with Special Reference to Questions of Reality, Oxford University Press, London, 1942. [23] K. Shaw: A tropical intersection product on matroidal fans, SIAM Journal on Discrete Mathematics 27 (2013) 459–491. [24] W. A. Stein et al.: Sage Mathematics Software (Version 5.9), The Sage Development Team, 2013, http://www.sagemath.org. 28 [25] M. Stillman, D. Testa and M. Velasco: Grobner bases, monomial group actions, and the Cox rings of del Pezzo surfaces, Journal of Algebra 316(2) (2007) 777–801. [26] B. Sturmfels and Z. Xu: Sagbi bases of Cox-Nagata rings, Journal of the European Mathematical Society 12 (2010) 429–459. [27] M. Vigeland: Smooth tropical surfaces with infinitely many tropical lines, Arkiv for Matematik 48 (2010) 177–206. Authors’ addresses: Qingchun Ren, Google Inc, Mountain View, CA 94043, USA, [email protected] Kristin Shaw, Technische Universitat Berlin, MA 6-2, 10623 Berlin, Germany, [email protected] Bernd Sturmfels, University of California, Berkeley, CA 94720-3840, USA, [email protected] 29
1105.1188
3
1105
2011-06-07T21:56:00
Inverses of monomial Cremona maps
[ "math.AG", "math.AC" ]
We show that monomial Cremona maps of degree d on P^n can have inverses whose degree d' is quite large (for d > 2, d' = ((d-1)^n - 1)/(d-2) occurs), and that the full list of degrees d' does not always form an interval. An easy method for inverting the maps is presented.
math.AG
math
INVERSES OF MONOMIAL CREMONA TRANSFORMATIONS PETER M. JOHNSON Abstract. We show that monomial Cremona transformations of degree d in Pn can have inverses whose degree d′ is quite large (for d > 2, d′ = (d−1)n −1 d−2 occurs), and that the full list of possible degrees d′ for fixed d and n does not always form an interval. An easy method for inverting the maps is presented. 1. Introduction: The problem of inverse degrees A Cremona transformation is a birational morphism ϕ : Pn 99K Pn, where we fix n ≥ 2 to avoid trivialities and sometimes write Pn k to emphasize the field k. There are homogeneous polynomials gi of the same degree d in variables x0, x1, . . . , xn such that ϕ induces the partial function (x0 : · · · : xn) → (g0 : · · · : gn) . If nonconstant factors common to all the gi are canceled, the representation of ϕ is unique up to a nonzero constant factor, thus giving a well-defined degree d, even under independent coordinate changes in the domain and codomain. The inverse of ϕ also has a degree, here called the inverse degree d′ of ϕ, so d′′ = d. The notation d′ emphasizes that we are interested in which values of inverse degrees occur as ϕ ranges over Cremona transformations of degree d, in some fixed Pn In what follows it can be assumed that k (after extending) is k . algebraically closed. Classical tools, in a setting described for example in [12], but going back to the earliest investigations by Cremona, Nother and Cayley, easily yield a bound d′ ≤ dn−1 from the study of intersections with certain sets of hyperplanes. In P2 this gives d′ = d. In P3 it gives √d ≤ d′ ≤ d2, a fact Cremona regarded as evident -- see p. 278, l. 6 of [3]. Cremona [3], [4] showed that all such d′ occur here when d = 2 or d = 3. This is now established for all d in Pan [11], with examples that modify ones in [10] that were found to be flawed. In Pn, n > 3, many related questions remain open. We focus almost entirely on the much simpler monomial Cremona transforma- tions, where the defining polynomials gi are monomial. These have been studied actively during the last decade -- see for example Costa and Simis [2], Simis and Villarreal [14], and their references. As described in [8], with fuller details in Theorem 2.2 of [14], the inverse of any monomial Cremona transformation is monomial. Even in this special case, little is known about which inverse degrees are possible. Our results dash hopes for simple answers. A recent preprint [2] of Costa and Simis, based on [1], provides some definitive results on rational maps defined by quadratic monomials. One can see when such a map is Cremona, and if so quickly read off the degree of its inverse, by examining a graph whose vertices correspond to the variables, with the same number of edges 1MSC-class: 14E07 (Primary); 15A29, 11C20 (Secondary). 2Keywords: monomial Cremona transformation, birational morphism, degree of inverse. 1 2 PETER M. JOHNSON (allowing loops) which correspond to the quadratic monomials. Note that some articles such as [2] use n variables where we, following others, use n + 1. As a trivial application of the main result of [2], one can see that a monomial Cremona transformation of degree 2 in Pn, where n ≥ 2, has inverse degree at most n, with the bound attained only when the graph is either a triangle, giving the classic Cremona transformation in P2, or a linear tree (chain) with a loop added at one of the two extremities, giving a map ϕ : (x0 : · · · : xn) → (x2 0 : x0x1 : · · · : xn−1xn). In particular, when n = 3 and d = 2, the maximum value of d′ is 3 for monomial Cremona transformations, whereas without monomiality the maximum is 4. An example will be provided where the monomial Cremona transformations have inverse degrees fairly close to the upper bound of dn−1, which in this special case ceases to be optimal for n > 2. Except for very small values of n and d, it is not practical to examine all monomial Cremona transformations to see which inverse degrees d′ actually occur. Still, the few results observed from programs reveal that the d′ do not always form a sequence: gaps can appear. This and other aspects relating to computation are discussed in the last section. 2. Methods for obtaining inverses To prepare the ground for a formula for calculating inverse degrees, simplifying a method presented in Th. 2.2 of [14], we establish notation and briefly explain easy ways to establish some of the most relevant results. The notation used to define ϕ will be changed slightly to admit m+1 monomials g0, . . . , gm in n+1 variables, so ϕ becomes a rational map from Pn to the closure of its image in Pm. In many sources monomials are taken to be monic, but we shall temporarily allow constant factors, to indicate how little effect this generalization has on the maps. Only the main case m = n will be described, roughly following [8]. The group of monomial Cremona transformations in Pn k , k an arbitrary field, is a split extension. First, the maps (x0 : · · · : xn) 7→ (c1x0 : · · · : cnxn), ci ∈ k∗, where constant factors act trivially, form a normal subgroup T, a torus (k∗)n. Multiplying by elements of T does not affect degrees of maps. In the group of monomial Cremona transformations, the monic monomials, which are the real focus of interest, form a complementary subgroup to T. As determined in [8], or from discussions about lattices below, this subgroup can be identified with Aut(Zn), or GLn(Z). It acts on T as the full group of rational automorphisms. Henceforth, monomials will be monic. Given systems of coordinates, every rational map ϕ : Pn 99K Pm defined by monomials gi can be represented by its n + 1× m + 1 log-matrix A, whose jth column lists the exponents of the monomial gj. A log-matrix A is stochastic or, for emphasis, d-stochastic: every column has the same sum, which is d. Since only exponents of the gj are used, the field k is now irrelevant. The rational map ϕ is unchanged if, in its log-matrix, all entries in a row are adjusted by the same amount, remaining in N. Thus it can usually be assumed that the minimal entry in each row of A is 0, in which case its column sums give the degree of ϕ, and we say that A is reduced. For greater flexibility, matrix entries will now be allowed to range over Z, and the gi may be Laurent monomials. For each index i, let Ri be the 1-stochastic matrix whose entries in the ith row are 1 and whose other entries are 0. Stochastic INVERSES OF MONOMIAL CREMONA TRANSFORMATIONS 3 matrices will be called equivalent if they have the same size and their difference is an integral linear combination of the Ri or, in other words, if they induce the same rational map. Matrices are supposed be n + 1 × m + 1, except when monomial rational maps are composed, which corresponds to multiplying matrices of compatible sizes, then usually passing to the reduced matrix. To say A is d-stochastic means 1A = d1, where 1 denotes an all 1 row vector, here of size n + 1. Then multiplying by A on the left induces a map of Z- 0 denotes the sublattice of Zm+1 lattices (free abelian groups) Λm consisting of column vectors whose entries sum to 0. Stochastic matrices are equivalent precisely when they induce the same lattice map. 0 , where Λm 0 → Λn There are many criteria for deciding if a map ϕ is birational. Some that are specific to the monomial situation appear in [13] and [15]. If the above lattice map is not an isomorphism, the associated monomial map ϕ has no inverse, not even among rational maps. A proof can be extracted from Sections 1 and 2 of [13], but uses more machinery than would be expected for such a basic fact, so the following approach seems worth recording. The key result is: Theorem 1. If a birational map ϕ : Pn 99K W , W closed in Pm, is defined by monomials, the induced lattice map Λm 0 must be surjective. 0 → Λn Proof. By assumption, ϕ has a rational inverse ψ, which must define an injective function on some nonempty open subset V of W via n + 1 homogeneous poly- nomials in variables y0, . . . ym coordinatizing Pm, with open image U ⊂ Pn on which ϕ is defined. All sets here are irreducible. If desired, one can assume that the coordinate functions xi or yj vanish nowhere on U or V , to work with rings RU , RV of functions on U, V , generated by Laurent monomials of degree 0 in the coordinates. Recall that Λn 0 induced by the log-matrix of ϕ. A suitable change of basis for Λn 0 , and a shift to multiplicative language, produce Laurent monomials zi that are free generators of RU , such that certain powers zdi i generate the same multiplicative group as that generated by the monomials ϕ∗(yj). Although ψ induces an isomorphism between k(U) and k(V ) (quotient fields of RU and RV ), the images of the zdi i already generate k(V ). This forces di = 1 for all i, so the above lattice map is surjective. (cid:3) 0 contains the image of the map on Λm A rational map ϕ : Pn 99K Pn defined by monomials is birational precisely when it induces an automorphism of the lattice Λn 0 , and the inverse must be monomial. Such maps have been dealt with before, most notably in [14]. Our main aim is to provide practical methods for obtaining inverse maps and their degrees, where the matrices can have entries in Z. Let ϕ be birational, with d-stochastic log-matrix A acting on Zn+1. The re- striction to Λn 0 has determinant ±1 and, for every column vector v, Av − dv lies 0 . Thus det(A) = ±d (cf. Lemma 1.2 of [13]). Now assume d 6= 0. Since A−1 in Λn 0 , the difference of any two columns of A−1 is integral (cf. Lemma 2.1 of acts on Λn [14]). This inverse is of the form d−1A∗, where A∗ has entries in Z. In each row i of A−1, let ri be the least element, and define ki = −dri. All ki lie in N, since A−1A = I and n + 1 ≥ 2. Using again the matrices Ri with ith row 1 to adjust rows, define B = (I +Pi kiRi)A−1. Then B is the reduced matrix that represents 4 PETER M. JOHNSON the inverse of ϕ, so its column sums give the inverse degree. This yields a formula that may have theoretical use: Theorem 2. Let A be a d-stochastic matrix obtained from Laurent monomials that define a Cremona transformation. Then the inverse map is also monomial and, assuming d 6= 0, its degree is d−1 − P ri, where ri denotes the least value occurring in the ith row of A−1. To clarify this result, and to eliminate the apparent need to work with fractions, one should work within the framework of Laurent monomials in affine coordinates Xi = xi/x0. We present this indirectly via an easily applicable matrix-based approach, in which d and d′ are regarded as functions on the group G of auto- morphisms of Λn 0 , with d′(g) = d(g−1) ∈ Z+. Using the ordered basis ei − e0 (0 < i ≤ n + 1), G is identified with the matrix group GLn(Z). Starting with the matrix of any g ∈ G, one prepends a row and column in the zero-th position, so that all entries of this column, and all column sums, are zero. Adjusting rows as before yields a reduced matrix A = Ag which, as should be clear from earlier discussions, is the log-matrix of a monomial Cremona transformation ϕ whose column sums give the degree d(ϕ). Thus d(g) is the sum of n+1 nonnegative terms, one for each row of A, using the negative of the least term in the row, or 0 if no negative terms are present. Similar formulas appear in Remark 3.2 of [8] in a loosely related context. Example 1. We pass back and forth between some matrices in GL2(Z) and the related log-matrices for monomial Cremona transformations of P2 (so d′ = d). g = (cid:20) 1 24 g−1 = (cid:20) 25 24 −g = (cid:20) −1 −24 gt = (cid:20) 1 −1 1 25 (cid:21) ⇋   1 (cid:21) ⇋   −1 −25 (cid:21) ⇋   24 25 (cid:21) ⇋   1 0 −2 −49 24 0 25 0 1 1 23 25 −24 1 0 −24 0 0 −1 2 0 49 0 −1 −24 25 0 −1 0 −25 −26 1 0 0 25 1 24         ⇋ Ag =   ⇋ Ag−1 =  ⇋ A−g =  ⇋ Agt =   49 47 0 0 0 1 24 1 25 24 24 49 1 0 0 47 0 2 0 24 23 25 24 2 49 0 0 1 1 0 26 0 1 0 24 25  , d(g) = 49.  , d(g−1) = 49.  , d(−g) = 49.  , d(gt) = 26. Permutations of coordinate vectors induce a subgroup Sn+1 of G = GLn(Z), and d is constant on each double coset Sn+1gSn+1, since this involves permuting rows and columns of stochastic matrices. By considering Ag1Ag2, one sees that d(g1g2) ≤ d(g1)d(g2). Also, for g /∈ Sn+1, d(g) ≥ 2, and d(−I) = n. Such ideas, developed further, may well yield useful information. INVERSES OF MONOMIAL CREMONA TRANSFORMATIONS 5 3. Examples with large inverse degrees There are in the literature many examples of monomial Cremona transforma- tions with diverse features of interest -- see for example Section 5 of [15]. An attempt to construct maps in Pn of degree d with relatively large inverse degrees d′ led to the following examples, which may be those with the largest possible d′. Extensive tests produced no other example (up to permutations) that attained or exceeded the value given in this section. We will proceed as in Theorem 2. This should be compared with the later approach. Example 2. For d, n ≥ 2, define ϕ : (x0 : · · · : xn) → (xd n−1xn), which has an upper-triangular log-matrix A whose nonzero entries are a11 = d, and, for i > 1, ai−1,i = d − 1, aii = 1. Let B be the upper-triangular matrix with entries bij = (1 − d)j−i for i ≤ j. It is easy to verify that dividing the first line of B by d produces A−1. For convenience, write c = d − 1. To illustrate, the matrices A and B (almost inverses) are shown below when n is 2 or 3. The degree of ϕ is d and the above criterion will, after simplification, give the inverse 0 x1 : · · · : xd−1 0 : xd−1 degree d′ = 1 + c + · · · + cn−1, which is (d−1)n d−2 For n of the form 2m, the raw sum for d′ is 1 i=0 c2i+1+0, whose fractional parts yield an alternating sum that meshes with the last sum. The case n = 2m + 1, d′ = 1 c+1 +c2m−1+2Pm−2 i=0 c2i+1 + 0, is almost identical. c+1 + c2m−1 c+1 + c2m+1 if d > 2. −1 For n = 2, A =   c + 1 c 0 1 c 0 1 0 0 For n = 3, A =   c + 1 c 0 0 1 c 0 0 1 c 0 0 1 0 0 0 c+1 + 2Pm−1  , B =  c2 1 −c 0 1 −c  1 0 0    . c2 −c3 1 −c c2 0 1 −c 0 0 1 −c 1 0 0 0   , B = .   4. Computational aspects All monomial Cremona transformations of degree 5 in P3 can be generated from the 367,290 combinations (unordered selections) of 4 out of 56 monomials in x0, . . . , x3, discarding those that fail to define a birational map of degree 5. Only 11,496 maps survive. By using combinations, the effect of permuting the columns (four coordinates in the image) has been factored out, so that only one out of each class of 24 maps is counted. The S4 that permutes rows acts on these classes, in orbits (often regular) consisting of maps with the same inverse degree. The frequency of occurrence of each d′, but not the orbits, of these 11,496 maps, is shown below. The d′ for the 48,042 monomial Cremona transformations of degree 3 in P4 are also shown. In that case, four kinds of maps are square-free, with descriptions given in Prop. 5.5 of [15]. 6 PETER M. JOHNSON Table 1. All monomial Cremona transformations with d = 5 in P3. Value of d′: Freq. of d′: 4 5 6 7 3 12 120 672 1932 1044 1584 1440 1248 696 816 552 13 14 480 168 16 240 17 168 15 240 19 24 20 48 9 10 11 8 18 0 21 24 Table 2. All monomial Cremona transformations with d = 3 in P4 . Value of d′: Freq. of d′: 2 432 9 1080 3 4 5 6 7 8 8670 14640 10920 5820 3720 1200 15 10 840 120 13 240 11 360 12 0 14 0 One sees from orbit sizes that, in both cases, the only examples giving the max- imum value of d′ are those described in the previous section, up to permutations of rows and columns. More importantly, even for these small values of d and n, the list for d′ contains gaps instead of forming an interval. Similar exhaustive calculations show that for monomial Cremona transformations in P3, the next case d = 6 has values of d′ that do form an interval [3, 31], and here each d′ from 28 to 31 arises from an essentially unique map. When d = 7, d′ can assume any value in [3, 43] except 39 or 40, while for d = 8, d′ ∈ [4, 57], excluding only 54. A double gap first appears for d = 10, where d′ cannot be 84, 87 or 88. Moving to P4, it is unlikely that theory will help explain the results. For example, when d = 4, the d′ lie in [2, 40], with gaps at 32, 34 -- 36, and 38 -- 39. When d = 5, the d′ lie in [3, 85], with gaps at 63, 70, 72, 74 -- 75, 77 -- 80, and 82 -- 84. The calculations were performed in SINGULAR [6], with about 100 lines of code, available on request. On an ordinary computer, this took a few minutes, a few hours, or (for d = 5 in P4) about two days. Results from a similar program that randomly generated large numbers of Cremona transformations with fixed d and n had already hinted strongly where the smaller gaps lie. Since this seems to be the first record of systematic calculations made in this area, it seems appropriate to mention what data can and cannot be expected to be obtainable. Tables for values of d up to 12 in P3 have been produced, with 13 expected to be the last, as the naive algorithm used becomes too slow. An implementation to count the number of monomial Cremona transformations in P4 with given multidegrees [10] d1 = d, d2, d3 = d′, and giving data for a few values of d, should soon follow. The diversity and richness of examples obtained already provide a counter- part and challenge to theory. To go further, it seems worthwhile to use random sampling to assemble extensive collections of certain kinds of Cremona transfor- mations, not just monomial ones, then search for new phenomena. This needs to be done with care. We point out where some simple approaches run into difficulties. The idea used above for random generation of monomial Cremona transforma- tions quickly becomes useless as d and n grow, as d-stochastic matrices giving Cremona transformations are rare, and those giving large values of d′ are even rarer. Also, the model used to generate columns (d approximately independent INVERSES OF MONOMIAL CREMONA TRANSFORMATIONS 7 and uniform selections of a position) is strongly biased against outcomes where the nonzero values are concentrated in few entries, the ones most likely to give a d′ that is unusually large or small. By repeated sampling until many1 monomial Cremona transformations were obtained, the maximum d′ observed started to fall well below the bound given by the examples of the previous section, blocking attempts to clarify where further gaps might lie. To show just one small exam- ple, by generating only 10,000 Cremona transformations with d = 6 in P3 we did come fairly close to the maximum value d′ = 31 but, as typically happened, failed to observe the minimum, here d′ = 3, although values just above this appeared frequently. Table 3. 10,000 samples (from 3,226,875 attempts) with d = 6 in P3. Value of d′: Freq. of d′: 5 6 7 8 9 13 14 4 11 260 270 797 838 1169 1737 1031 1342 19 12 15 56 792 704 224 311 27 23 20 64 6 1 17 174 25 9 16 158 24 1 10 18 36 26 0 22 3 21 17 A better approach is to work in a fixed Pn, using matrices in GLn(Z) to gen- erate monomial Cremona transformations with variable parameters (d, d′) (or multidegrees) whose range of values should become clearer over time. A stream of matrices can be obtained from the identity matrix by making random choices to replace some line with that line plus a multiple of another line. Any matrix in GLn(Z) is reachable by this process: just consider how to row-reduce nonsingular matrices when working over the integers. For a slightly different explanation, see [16]. In practice, best results were obtained using multiples restricted to have small absolute value (even 5 seems too large), which helped postpone arriving at a case with d or d′ beyond the range considered, at which point the generation process was started afresh. This idea has the great advantage that it produces a usable example at each step. However, building a table of pairs (d, d′) that occur in a certain range is a slow process, as the cases of most interest tend to appear rarely -- in some cases, only once in more than 330 million trials. Several days of computation were enough to fill most of the possible positions repeatedly, but in no case was there compelling evidence to suggest whether some new apparent gap was real or illusory. Acknowledgment. A conversation with Ivan Pan helped clarify details about bounds for d′. He also sent an early version of [11]. 120,000 down to 100, as parameters grew. 8 PETER M. JOHNSON References [1] B. Costa, Transforma¸coes de Cremona definidas por monomios, Ph.D. thesis, UFPE, Re- cife, Brazil, 2011. [2] B. Costa, A. Simis, Cremona maps defined by monomials, 21 pp., arXiv:1101.2413v1 [math.AC]. [3] L. Cremona, Sulla transformazione razionale di 2.◦ grado nello spazio, la cui inversa `e di 4.◦ grado, Memorie dell'Accademia delle Scienze dell'Istituto di Bologna, Ser. III, 1 (1871), 365 -- 386. Reprinted in [5]. [4] L. Cremona, Sulle transformazioni razionali nello spazio, Annali di Matematica pura ed applicata, ser. II, V (1871), 131 -- 162. Reprinted in [5]. [5] Opere matematiche di Luigi Cremona, Vol. 3, Ulrico Hoepli, Milan, 1917. [6] W. Decker, G.-M. Greuel, G. Pfister, H. Schonemann, Singular 3-1-2 -- A computer algebra system for polynomial computations. http://www.singular.uni-kl.de (2010). [7] W. Fulton, Intersection theory, Springer, New York, 1984. [8] G. Gonzalez-Sprinberg and I. Pan, On the monomial birational maps of the projective space, An. Acad. Brasil. Cienc. 75 (2003), no. 2, 129 -- 134. [9] G. Gonzalez-Sprinberg and I. Pan, On characteristic classes of determinantal Cremona transformations, Pr´epublication de l'Institut Fourier no 681 (2005), 9 pp. [10] I. Pan, Sur le multidegr´e des transformations de Cremona, C. R. Acad. Sci. Paris, S´erie I, 330 (2000), 297 -- 300. [11] I. Pan, On Cremona transformations of P3 with all possible multidegrees, preprint (2011), 4 pp. [12] J. G. Semple, J. A. Tyrrell, Specialization of Cremona transformations, Mathematika 15 (1968), 171 -- 177. [13] A. Simis and R. H. Villarreal, Constraints for the normality of monomial subrings and birationality, Proc. Amer. Math. Soc. 131 (2003), 2043 -- 2048. [14] A. Simis and R. H. Villarreal, Combinatorics of Cremona monomial maps, 10 pp., arXiv:0904.4065v2 [math.AG], to apppear in Math. Comp. [15] A. Simis and R. H. Villarreal, Linear syzygies and birational combinatorics, Results Math. 48 (2005), no. 3 -- 4, 326 -- 343. [16] S. Trott, A pair of generators for the unimodular group, Canad. Math. Bull. 5 (1962), 245 -- 252. Departamento de Matem´atica Universidade Federal de Pernambuco 50740-540 Recife -- PE Brazil [email protected]
1101.1305
1
1101
2011-01-06T21:01:10
Quantum Sheaf Cohomology, a pr\'ecis
[ "math.AG", "hep-th" ]
We present a brief introduction to quantum sheaf cohomology, a generalization of quantum cohomology based on the physics of the (0,2) nonlinear sigma model.
math.AG
math
Quantum Sheaf Cohomology, a pr´ecis Josh Guffin Abstract. We present a brief introduction to quantum sheaf cohomology, a generalization of quantum cohomology based on the physics of the (0,2) nonlinear sigma model. This paper is based on a talk given on 13 December, 2010 during the Second Latin Congress on Symmetries in Geometry and Physics at the Universidade Federal do Paran´a in Curitiba, Brazil. Throughout, we will consider X to be a Kahler manifold of complex dimension n. In addition, we will consider E → X to be a complex Hermitian vector bundle of rank k satisfying (i) c2(E) = c2(TX ), (ii) det E ∨ ∼= ωX. As these conditions imply the usual Green-Schwarz anomaly cancellation condi- tions, we will call such a bundle omalous 1. One may consider (ii) to be an analogue of the usual condition for existence of the B-model. A bundle satisfying these con- ditions may be obtained by, for example, selecting a deformation of the tangent bundle when X is a projective toric variety. Quantum Cohomology Ordinary cohomology. We now give some elementary facts about the coho- mology of X, stated in a way that will facilitate our point of view on quantum sheaf cohomology. Since X is Kahler, there is a Hodge decomposition on H •(X, C), H •(X, C) ∼=Mp,q H p(X,VqT ∨ X). By a slight abuse of language, we will refer to elements of the sheaf cohomology X ) as (p, q)-forms -- clearly H •(X, C) possesses a basis consisting of such forms. The cup/wedge product on cohomology furnishes this vector space groups H p(X,VqT ∨ 1991 Mathematics Subject Classification. Primary 32L10, 81T20; Secondary 14N35. 1That is, not anomalous -- this delightful terminology is due to Ron Donagi. 2 JOSH GUFFIN with the structure of a bigraded C-algebra. Finally, integration of forms induces a trace on the algebra; in terms of a basis element ω, ZX 0 tr(ω) = ω ω ∈ H n(X,VnT ∨ otherwise. X) The pairing (α, β) 7→ tr(α ∧ β) induced by this trace is a non-degenerate bilinear form satisfying (α, β ∧ γ) = (α ∧ β, γ), so that H •(X, C) is a bigraded Frobenius algebra. Physics. The relationship between ordinary cohomology and quantum coho- mology may be elucidated by appealing to physics -- in particular to a topologically- twisted (2,2) nonlinear sigma model of maps P1 → X. Of the many intriguing as- pects of this quantum field theory, we will be most interested in its algebra of mass- less supersymmetric operators2. Using elementary physics arguments, one identifies a basis for the set of such operators that may be set into one-to-one correspondence with (p, q)-forms on X. The (2,2) supersymmetry of the model forces the product of two massless super- symmetric operators to be massless and supersymmetric. The particular form of the product obtains by considering three-point correlation functions in the quantum field theory: the quantum product of two massless operators O1 and O2 is defined to be the unique operator (O1 ∗ O2) such that for all massless operators O3, (1) h1(O1 ∗ O2)O3i = hO1O2O3i . Here, 1 denotes the operator corresponding to 1 ∈ H 0(X,V0T ∨ function is computed using the instanton expansion X ). Such a correlation hO1O2O3i = Xβ∈H2(X,Z) hO1O2O3iβ qβ. Although they have intrinsic meaning in physics, we will consider the expressions qβ to comprise a set of formal variables endowed with the structure of a monoid via the product qαqβ = qα+β. We denote by CJqK the ring of formal power series with complex coefficients in these variables -- one sometimes insists on convergence or other properties, but such subtleties are beyond the scope of this review. Mathematically, one defines the expression hO1O2O3iβ as the Gromov-Witten in- variant3 hI0,3,βi(ω1, ω2, ω3), where ωi are the forms corresponding to the operator Oi. Physically, one says that hO1O2O3iβ qβ denotes the contribution of instantons of degree β to the correlation function hO1O2O3i. This expression is morally the integral of induced forms on some compactification M (X, β) of the moduli space of holomorphic maps f : P1 → X of class β = f∗[P1]. We will write the induced forms schematically using maps ζβ : H p(X,VqT ∨ X) → H p(cid:16)M (X, β),VqT ∨ M(X,β)(cid:17) . 2More precisely, it is the algebra of local, scalar, supersymmetric operators [Wit91]. 3See equation 7.4 of [CK00] for a precise definition of Gromov-Witten invariants. (2) (3) QUANTUM SHEAF COHOMOLOGY, A PR ´ECIS 3 If ωi are the forms corresponding to operators Oi, modulo the subtleties of obstruc- tion bundles we have that hO1O2O3iβ =ZM(X,β) ζβ(ω1) ∧ ζβ(ω2) ∧ ζβ(ω3). Depending on the compactification, there may be more than one such map -- in the case of the stable maps compactification, pullbacks via evaluation maps play the role of ζβ. For toric varieties, one often uses the Morrison-Plesser compactification [MRP95] wherein -- as indicated in (3) -- one map suffices for each β. The three-point correlation functions in (2) induce a non-degenerate bilinear pairing X )JqK, leading to the (ω1, ω2) = h1O1O2i on the unital algebra Lp,q H p(X,VqT ∨ following definition. Definition 1. The quantum cohomology of X is the Frobenius algebra QH •(X) :=Mp,q H p(X,VqT ∨ X)JqK, with the product and bilinear pairing induced by (2,2) three-point functions. Here, the (2,2) correlation functions are defined either via Gromov-Witten invari- ants or as correlation functions in the quantum field theory, depending on whether your tastes tend to the mathematical or to the physical. Example 2. The "classical sector" is the set of maps homotopic to a point, β = 0, and the moduli space of such maps is simply X itself. Thus, in this sector, the quantum product reduces to the wedge product on forms; ordinary cohomology is the "classical limit" of quantum cohomology. This sector may be isolated by setting q = 0. For example, the ordinary and quantum cohomology of Pn are respectively H •(Pn, C) ∼= QH •(Pn) ∼= C[H] hH n+1i , C[H]JqK hH n+1 − qi . Here H denotes the hyperplane class. For Pn × Pm, the equivalent expressions are (4) H •(Pn × Pm, C) ∼= QH •(Pn × Pm) ∼= C[H1, H2] , H m+1 2 (cid:11) , 1 (cid:10)H n+1 (cid:10)H n+1 C[H1, H2]Jq1, q2K 1 − q1, H m+1 2 − q2(cid:11) . Quantum Sheaf Cohomology As in our study of the passage from ordinary cohomology to quantum cohomology, we first consider the "ordinary sheaf cohomology" -- in particular that of an omalous bundle E → X. Here, by ordinary sheaf cohomology we mean cohomology valued in polysections, (5) Mp,q H p(X,VqE ∨). (6) 4 JOSH GUFFIN Again by a slight abuse of language, we will refer to elements of H p(X,VqE ∨) as (p, q)-forms -- clearly the vector space (5) possesses a basis consisting of such forms, and the cup/wedge product furnishes it with the structure of a bigraded C-algebra. The trace on this algebra is slightly more subtle and follows from the omality of E. In particular, one uses the existence of an isomorphism ψ : H n(X,VkE) → H n(X, ωX ) to define, for a basis element ω, ZX 0 tr(ω) = ψ(ω) ω ∈ H n(X,VkE ∨) otherwise. The pairing (α, β) 7→ tr(α ∧ β) induced by this trace endows Lp,q H p(X,VqE ∨) with the structure of a bigraded Frobenius algebra. Physics. To understand the relationship between sheaf cohomology and quan- tum sheaf cohomology we again appeal to physics -- in particular a topologically- twisted (0,2) nonlinear sigma model of maps P1 → X. A recent physics review of this and related models may be found in [McO10]. We will again be most inter- ested in its algebra of massless supersymmetric operators4. The same elementary physics arguments used for the (2,2) theory identify a basis for this set of operators that may be placed into one-to-one correspondence with (p, q)-forms (that is, ele- ments of (5)), and the quantum product of two massless operators is defined using three-point correlation functions of the (0,2) in analogy to (1). Unlike the (2,2) case, however, there is no mathematical definition of hO1O2O3iβ in a (0,2) theory so the following definition is purely physical. Definition 3. The quantum sheaf cohomology of an omalous bundle E → X is the Frobenius algebra QH •(X, E) :=Mp,q H p(X,VqE ∨) ⊗ CJqK with the product and bilinear pairing induced by (0,2) three-point functions. As in the case of ordinary quantum cohomology, the classical limit of quantum sheaf cohomology is precisely the ordinary sheaf cohomology with the Frobenius structure induced by (6). Unlike the case in (2,2) theories, (0,2) supersymmetry is not enough to guarantee that the product of massless operators is massless: one needs to work harder to show that the algebra closes in the set of all operators. Existence. The (modern) history of quantum sheaf cohomology begins with the observation in [ABS04] of an analogue of QH •(X) for (0, 2) theories. Therein, the quantum sheaf cohomology of a one-parameter family of deformations of the tangent bundle of P1 × P1 was computed using a conjectured form of mirror sym- metry for (0, 2) models. Their calculations were confirmed in a sheaf-cohomology- based computation by Katz and Sharpe [KS06]. Inspired by these results, Adams, 4As explained in [ADE06], we are actually interested in local, scalar, supersymmetric op- erators with vanishing holomorphic conformal weight, but for continuity we will refer to them as massless supersymmetric or simply massless. QUANTUM SHEAF COHOMOLOGY, A PR ´ECIS 5 Distler, and Ernebjerg [ADE06] gave a physics definition of quantum sheaf coho- mology and found a physics proof of two sufficient conditions for its existence. We restate these conditions here as conjectures. Conjecture 4. Let E and E ′ be omalous elements of a family of bundles U . Let γ : [0, 1] → U continuous with γ(0) = E, γ(1) = E ′, γ(t) omalous for all t ∈ [0, 1]. Then QH •(X, E) exists iff QH •(X, E ′) exists. Conjecture 5. If E → X is omalous and rk E < 8, then QH •(X, E) exists. Since QH •(X, TX ) = QH •(X), the former condition implies the existence of quan- tum sheaf cohomology for all omalous one-parameter families of tangent-bundle deformations. The latter is likely an artefact of the technique used in the physics proof -- there are no known examples of omalous bundles of rank eight or higher for which the massless operators fail to close under the quantum product, and there are no physical reasons to expect such a bundle to exist. Computation example. Although there is no definition for the invariants hO1O2O3iβ, a number of physics-inspired techniques exist to compute them when the omalous bundle is a deformation of the tangent bundle of a toric variety [KS06, GK07, MM08] or a complete intersection therein[MM09]. One of the advantages of using a toric variety X is that deformations of TX are easily obtained by deforming the Euler exact sequence: 0 Or E0 X Mρ∈∆ OX (Dρ) TX 0. Here, r is the rank of the Picard group, ∆ denotes the set of torus-invariant divisors corresponding to one-cones in the fan of X, and E0 is a collection of sections of OX (Dρ), which are toric analogues of OPn (1). Taking X = P1 × P1, for example, the sequence becomes 0 E0 O2 X OX (1, 0)2 ⊕ OX (0, 1)2 TX 0, where the map is E0 = x0 x1 0 0 0 0 x2 x3  . A deformation of TX may be obtained by choosing a different collection of sections for the map. For example, selecting the map to be (7) E = x0 x1 γ1x2 + γ2x3 γ3x2 ǫ1x0 + ǫ2x1 ǫ3x0 x2 x3  as in [GK07] gives a convenient basis for the space of deformations of the tangent bundle (ǫi, γi ∈ C). Therein, several of the invariants hO1O2O3iβ were computed 6 JOSH GUFFIN for the bundle E → P1 × P1 defined as the cokernel of (7). These were then used to deduce the quantum sheaf cohomology of E; (8) QH •(P1 × P1, E) ∼= C[ψ,eψ]Jq1, q2K eψ2 + γ1ψeψ − γ2γ3ψ2 − q2 + . * ψ2 + ǫ1ψeψ − ǫ2ǫ3eψ2 − q1, These computations were confirmed in [MM08] using physics techinques. Note that as ǫi, γi → 0, the quantum sheaf cohomology in (8) limits to the ordinary quantum cohomology of P1 × P1 in (4). This material is based on work supported by the National Science Foundation under DMS grant no. 0636606. References [ABS04] Allan Adams, Anirban Basu, and Savdeep Sethi. (0,2) duality. Adv. Theor. Math. Phys., 7:865 -- 950, 2004, hep-th/0309226. [ADE06] Allan Adams, Jacques Distler, and Morten Ernebjerg. Topological heterotic rings. Adv. Theor. Math. Phys., 10:657 -- 682, 2006, hep-th/0506263. [CK00] D. A. Cox and S. Katz. Mirror symmetry and algebraic geometry. American Mathe- [GK07] [KS06] matical Society, 2000. Providence, USA 469 p. Josh Guffin and Sheldon Katz. Deformed quantum cohomology and (0,2) mirror sym- metry. 2007, arXiv:0710.2354 [hep-th]. Sheldon H. Katz and Eric Sharpe. Notes on certain (0,2) correlation functions. Commun. Math. Phys., 262:611 -- 644, 2006, hep-th/0406226. [McO10] Jock McOrist. The Revival of (0,2) Linear Sigma Models. 2010, 1010.4667. [MM08] Jock McOrist and Ilarion V. Melnikov. Half-Twisted Correlators from the Coulomb Branch. JHEP, 04:071, 2008, 0712.3272. [MM09] Jock McOrist and Ilarion V. Melnikov. Summing the Instantons in Half-Twisted Linear Sigma Models. JHEP, 02:026, 2009, 0810.0012. [MRP95] David R. Morrison and M. Ronen Plesser. Summing the instantons: Quantum coho- mology and mirror symmetry in toric varieties. Nucl. Phys., B440:279 -- 354, 1995, hep- th/9412236. [Wit91] Edward Witten. Mirror manifolds and topological field theory. 1991, hep-th/9112056. Department of Mathematics, University of Pennsylvania, Philadelphia, PA 19104 E-mail address: [email protected]
1008.1790
1
1008
2010-08-10T20:03:44
A characterization of the symmetric square of a curve
[ "math.AG", "math.DG" ]
In this paper a new intrinsic geometric characterization of the symmetric square of a curve and of the ordinary product of two curves is given. More precisely it is shown that the existence on a surface of general type S of irregularity q of an effective divisor D having self-intersection D^2>0 and arithmetic genus q implies that S is either birational to a product of curves or to the second symmetric product of a curve.
math.AG
math
A CHARACTERIZATION OF THE SYMMETRIC SQUARE OF A CURVE MARGARIDA MENDES LOPES, RITA PARDINI AND GIAN PIETRO PIROLA Abstract. In this paper a new intrinsic geometric characterization of the symmetric square of a curve and of the ordinary product of two curves is given. More precisely it is shown that the existence on a surface of general type S of irregularity q of an effective divisor D having self- intersection D2 > 0 and arithmetic genus q implies that S is either birational to a product of curves or to the second symmetric product of a curve. Keywords: surface of general type, irregular surface, curves on surfaces, symmetric product. 2000 Mathematics Subject Classification: 14J29 1. Introduction In this paper we give a new intrinsic geometric characterization of the symmetric square of a curve and of the ordinary product of two curves. The symmetric products Sk(C) of a curve C of genus g > 0 give, together with the ordinary products, the simplest examples of irregular varieties. To give some perspective, we recall that the cohomology ring of Sk(C) approxi- mates the cohomology ring of the Jacobian J(C) of C. In particular one has ([M]) H 1(C, Z) ≡ H 1(J(C), Z) ≡ H 1(Sk(C), Z) and, for k > 1, H 2(Sk(C), Z) ≡ H 2(J(C), Z) ⊕ Z. Notice that the Torelli-type theorem for k < g (see [Ra2]) provides an equivalence between curves and symmetric products. In view of their simple cohomological structure, one expects the symmet- ric products to have a very important place in the classification of irregular varieties. Some progress has been done using Fourier-Mukai transform and generic vanishing in the case k = g − 1 by C. Hacon ([H]), giving a cohomo- logical characterization of the theta divisor of a principally polarized abelian variety (PPAV). For 1 < k < g − 2 there is a famous conjecture due to O. Debarre ([De2]) that claims that the symmetric products Sk(C) and the Fano surface F of the lines of a smooth cubic 3-fold are the only varieties giving the minimal cohomological class of a PPAV (A, θ). The Debarre conjecture is proved for q = 4, by Z. Ran ([Ra1]), when A = J(X) is a Jacobian of a curve by 1 2 MARGARIDA MENDES LOPES, RITA PARDINI AND GIAN PIETRO PIROLA Debarre ([De2]), and when A is the intermediate Jacobian of a generic cubic 3-fold by A. Horing ([Ho]). Coming back to surfaces, it is a basic problem to give some geometrical or cohomological characterization of the second symmetric product of a curve. In the spirit of surface classification theory, a conjecture, not equivalent, and in some sense even stronger than Debarre's, is the following: Conjecture: X) ≡ V2 H 0(Ω1 The only minimal surfaces X of irregularity q > 2 with H 0(Ω2 are the symmetric products S2(C) and the Fano surfaces F of the lines of a smooth cubic 3-fold. X ), The above conjecture was proven for q = 3 in [HP] and independently in [Pi]. The proof of [HP] uses in a crucial way the fact that the image of the Albanese map is a divisor and it seems difficult to generalize. The proof of [Pi] is based on the geometric characterization of S2(C) given in [CCM] using the geometry of families of curves. Namely, in [CCM] S2(C) is proven to be the only minimal algebraic surface with irregularity q covered by curves of genus q and self-intersection 1. Analysing the curves of small genus on a surfaces of general type, we discovered a surprisingly precise geometric characterization of the second symmetric product (and of the ordinary product). Theorem 1.1. Let S be a smooth surface of general type with irregularity q containing a 1-connected divisor D such that pa(D) = q and D2 > 0. Then the minimal model of S is either: (a) the product of two curves of genus g1, g2 ≥ 2 (and g1 + g2 = q) or (b) the symmetric product S2(C) where C is a smooth curve of genus q (and C 2 = 1). Furthermore, if D is 2-connected, only the second case occurs. This result is in some sense very atypical of the theory of algebraic sur- faces, because we obtain a complete classification of the surface from the ex- istence of a single divisor with certain properties. The only similar instance we know of is the characterization of rational surfaces from the existence of a smooth rational curve with positive self-intersection. The proof of Theorem 1.1 consists of two main steps, that we describe in the symmetric product case. First we reduce to the case when the curve C is smooth. This step requires a very careful numerical analysis of the effective divisors contained in C. Then we observe that the Albanese variety Alb(X) of the surface is isomorphic to the Jacobian J(C) of C and, combining the Brill-Noether theory on C with the generic vanishing theorem of Green and Lazarsfeld, we show that the curve C moves in a positive dimensional family. This allows us to use the results in [CCM] and complete the proof. A CHARACTERIZATION OF THE SYMMETRIC SQUARE OF A CURVE 3 It appears that a sort of weak Brill-Noether theory can be performed for line bundles on irregular surfaces. We will come back to these topics in a forthcoming paper (cf. [MPP3]). Notation: A surface is a smooth projective complex surface. The irregu- larity of a surface S, often denoted by q or q(S), is h1(OS ) = h0(Ω1 S) and the geometric genus pg(S) is h0(KS) = h2(OS ). A curve on a surface is a nonzero effective divisor. A fibration of genus b of a surface S is a map f : S → B with connected fibers, where B is a smooth curve of genus b; f is relatively minimal if its fibers contain no −1-curve. A quasi-bundle is a fibration such that all the smooth fibers are isomorphic and the singular fibers of are multiples of smooth curves. 2. Auxiliary facts In this section we collect some auxiliary facts. We start by recalling the well known equality for reducible curves on smooth surfaces: • If a curve D decomposes as D = A + B where A, B > 0 and AB = m then pa(A) + pA(B) + m − 1 = pa(D). The following results are needed in the sequel: Lemma 2.1. Let S be a surface and let f : S → B a relatively minimal fibration of genus ≥ 2 with general fibre of genus ≥ 2. If C is a section of f (i.e. an irreducible curve C such that CF = 1), then: (i) C 2 ≤ 0; (ii) if C 2 = 0, then f is a quasi-bundle and there exists m > 0 such that mC is a fiber of a quasi-bundle fibration of S. Proof. (i) Let us notice first that any section C of f is smooth of genus b. Denote by F the general fiber of f ; by Arakelov's theorem (cf. [Se, Theorem 3.1]) the relative canonical class KS/B ∼ KS − (2b − 2)F is nef, hence KSC ≥ (2b − 2) and, by the adjunction formula, C 2 ≤ 0. (ii): follows by Theorem 3.2 in [Se]. (cid:3) Lemma 2.2. Let D be a 1-connected curve on a surface. Then every sub- curve A < D satisfies pa(A) ≤ pa(D). Proof. Set B := D − A and let m := AB. Then pa(A) + pa(B) + m − 1 = pa(D). Suppose for contradiction that pa(A) > pa(D). Then we obtain pa(B) + m − 1 < 0, i.e. m < 1 − pa(B). Since 1 − pa(B) = h0(B, OB) − h1(B, OB) we conclude that m < h0(B, OB). This contradicts [KM, Lemma 1.4], that states that any subcurve D′ of a 1-connected curve D satisfying D′(D − D′) = b verifies h0(D′, OD′) ≤ b . (cid:3) Lemma 2.3. Let S be a surface of general type and let D be an irreducible curve of S. If D2 > 0, then: (i) pa(D) ≥ 2; 4 MARGARIDA MENDES LOPES, RITA PARDINI AND GIAN PIETRO PIROLA (ii) if pa(D) = 2, then the minimal model T of S has K 2 T = 1, q(T ) = 0. Proof. Let η : S → T be the morphism onto the minimal model, so that KS := η∗KT +E, where E is the exceptional divisor. Since D2 > 0, the curve D is not contracted by η, hence KSD ≥ η∗KT D > 0 and the adjunction formula gives immediately pa(D) ≥ 2, with equality holding if and only if D2 = η∗KT D = 1. Hence, if pa(D) = 2 the index theorem gives K 2 T = 1 and, since by [De1] irregular surfaces have K 2 ≥ 2pg, T is regular. (cid:3) 3. Curves with pa = q As a preparation for the proof of Theorem 1.1, in this section we make a detailed numerical analysis of the following situation: • S is a surface of general type with irregularity q; • D is a 1-connected curve of S such that pa(D) = q and D2 > 0. Proposition 3.1. Let S be a surface of general type with irregularity q and let D be a 1-connected curve of S such that pa(D) = q and D2 > 0. If D = A + B, where A, B are curves such that AB = 1 and pa(A) ≥ 1, pa(B) ≥ 1, then S is birational to the product of two curves. Proof. The equality AB = 1 implies that both A and B are 1-connected (see [CFM, Lemma A.4]) and that q = pa(D) = pa(A) + pa(B). Since, by assumption, pa(A) ≥ 1, pa(B) ≥ 1, one has also pa(A) < q and pa(B) < q. So both A2 ≤ 0, B2 ≤ 0 (cf. [R], [Ca, Remark 6.8] and also [MPP2]). By the hypothesis D2 = A2 + 2 + B2 > 0, at least one of the inequalities is in fact an equality. Suppose then that B2 = 0. Then (cf. ibidem) there exists a fibration f : S → E where E is a curve of geometric genus g(E) ≥ q − pa(B) and such that mB is a fibre F of f for some m > 0. Since AB = 1 and B is nef, there is a unique irreducible curve θ ≤ A such that θB 6= 0. Of course θ is not contained in a fibre of f and, by Lemma 2.2, pa(θ) ≤ pa(A). From pa(A) = q − pa(B) we obtain pa(θ) ≤ g(E). Since f induces a surjective morphism θ → E of degree m and pa(θ) ≤ g(E), we conclude, by the Hurwitz formula, that g(θ) = pa(θ) = g(E) and, in addition, m = 1 or g(E) = 1 and θ → E is unramified. In the latter case, if m > 1 we have a contradiction because the existence of a multiple fibre of f means that the cover is ramified. So m = 1 and the fibration f : S → E satisfies g(E) + g(F ) = q. By [Be, Lemme], S is birational to the product F × E. (cid:3) Next we show that, when S is not birational to the product of two curves, we can assume that D is 2-connected. Lemma 3.2. Let S be an irregular surface of general type that is not bira- tional to the product of two curves. Let D be a 1-connected curve of S such that pa(D) = q, D2 > 0. Then there is a birational morphism p : S → S0, where S0 is a smooth surface such that p(D) contains a 2-connected curve D0 satisfying D2 0 > 0 and pa(D0) = q. A CHARACTERIZATION OF THE SYMMETRIC SQUARE OF A CURVE 5 Proof. If D is 2-connected there is nothing to prove. Otherwise we have a decomposition D = A + B where AB = 1 and, by Lemma 3.1, pa(A) = 0, pa(B) = q. By [CFM, Lemma A.4], A and B are 1-connected. Since pa(A) = 0, A is contracted by the Albanese map and so A2 < 0. If A2 < −1, then from D2 > 0 and AB = 1 we must have B2 > 0 and we consider now the curve B. If B is 2-connected we take D0 := B. If B is not 2-connected, then it has a decomposition B = A1 + A2 with A1A2 = 1. As above we can suppose pa(A1) = 0 and pa(A2) = q and we can now restart the reasoning. If A2 = −1, then A contains an irreducible (−1)-curve E such that ED = If p : S → S1 is the blow down of E and D1 := p(D), then we have 0. D = p∗D1, hence D1 is 1-connected, D2 1 = D2 > 0 and pa(D1) = pa(D) = q. Hence we may replace D and S by D1 and S1 and, if D1 is not 2-connected, repeat the previous step. Since D has a finite number of components, the process described above must stop and so in the end we get a surface S0 birational to S contained a 2-connected D0 as wished. (cid:3) 4. The 2-connected case Here we examine the situation of §3 under the additional assumption that D be 2-connected. Lemma 4.1. Let S be an irregular surface of general type and let D be a 2-connected curve of S such that pa(D) = q and D2 > 0. Then: (i) D is contained in the fixed part of KS + D; (ii) D is smooth irreducible; (iii) h0(S, D) = 1; (iv) the Albanese image of S is a surface. Proof. (i) Since D2 > 0 and, because D is 1-connected, h0(D, OD) = 1, one has h1(S, KS + D) = 0. Hence the cokernel of the restriction map H 0(S, KS + D) → H 0(D, ωD) is H 1(S, KS ) and therefore by pa(D) = q, the image of r must be zero. (ii) Assume by contradiction that D is not smooth. Then D has multiple points. Since D is contained in the fixed part of KS + D, any multiple point P of D is a base point of KS + D and so, by [ML, Thm. 3.1], D is not 2-connected. This contradicts the hypothesis. Finally, D, being smooth and 1-connected, is necessarily irreducible. (iii) Assume that h0(S, D) > 1. Then the irreducible curve D moves. Since, by hypothesis, pg(S) > 0, we conclude that D is not a fixed component of KS + D, contradicting (i). (iv) Since D is irreducible we have q ≥ 3 by Lemma 2.3. Assume for contradiction that the Albanese image of S is a curve E. Then E is a smooth curve of genus q and we have thus a fibration f : S → E. Since the smooth curve D satisfies D2 > 0, D is not contained in any fibre F of f , and 6 MARGARIDA MENDES LOPES, RITA PARDINI AND GIAN PIETRO PIROLA so f induces a degree m morphism f D : D → E, where m := DF . Since g(D) = g(E) = q and q ≥ 3, we have m = 1 by the Hurwitz formula. Now notice that we can assume that S is minimal, and thus f is relatively minimal. In fact every (−1)-curve θ is contained in the fibres of S and thus, since DF = 1, the image of D in the minimal model of S is still a smooth curve with geometric genus q. Then D is a section of f , but this contradicts Lemma 2.1. So the Albanese image of S must be a surface. (cid:3) Corollary 4.2. Let S be an irregular surface of general type that is not birational to a product of curves and let D be a nef 2-connected curve of S such that pa(D) = q and D2 > 0. Assume also that there is no (−1)-curve θ such that Dθ = 0. Then any curve C numerically equivalent to D is smooth irreducible. Proof. Since D is nef, also C is nef. Then it is well known that C nef and big implies that C is 1-connected (see [ML, Lemma 2.6]). If C is not 2-connected there is a decomposition C = A + B where AB = 1 and, by Proposition 3.1, pa(A) = 0, pa(B) = q. As in Lemma 3.2 one has A2 < 0 and since C is nef one must have A2 = −1, but this contradicts the hypothesis on D. Hence C is 2-connected and so, by Lemma 4.1, C is smooth irreducible. (cid:3) Proposition 4.3. Let S be a surface of general type with irregularity q and let D be a smooth irreducible curve D such that d := D2 > 0 and g(D) = q. Then the set {L ∈ Pic0(S)h0(D + η) > 0} has dimension ≥ min{q, d}. Proof. By Lemma 4.1, the Albanese image of S is a surface. So, by the generic vanishing theorem of Green-Lazarsfeld ([GL1],[GL2]), the set V 1(S) = {η ∈ Pic0(S)h1(η) > 0} is the union of finitely many translates of proper abelian subvarieties of A. Since D2 > 0, the map Pic0(S) → Pic0(D) is in- jective and thus an isomorphism, hence we can identify Pic0(S) and Pic0(D). Denote by Wd(D) ⊂ Pic0(D) the image of the natural map Sd(D) → Pic0(D) defined by ∆ 7→ [OD(∆ − D)]. If d ≥ q then Wd(D) = Pic0(D), otherwise Wd(D) is d-dimensional. Then Wd(D) generates Pic0(D) and it is irreducible, hence it cannot be contained in V 1(S). Note that for η /∈ V 1(S) the restriction sequence 0 → η → η + D → (η + D)D → 0 gives an isomorphism H 0(η + D) ∼= H 0((η + D)D). Hence for every η ∈ Wd(D) \ V 1(S) (and, by semicontinuity, for every η ∈ Wd(D)) we have h0(D + η) > 0. (cid:3) 5. Proof of the main result Proof of Theorem 1.1. By Proposition 3.1, if there exists a decomposition D = A + B, with A, B > 0, AB = 1 and pa(A), pa(B) > 0, then S is birational to a product of curves, namely we have case (a). A CHARACTERIZATION OF THE SYMMETRIC SQUARE OF A CURVE 7 So assume that no such decomposition exists. Then, up to replacing S by a surface in the same birational class, by Proposition 3.2 we may assume that D is 2-connected, hence smooth and irreducible by Lemma 4.1. Write d := D2. By Proposition 4.3, there exist a d-dimensional system C of curves numerically equivalent to D. Since we can obviously assume that there is no (−1)-curve A such that DA = 0, any curve C of C is smooth by Lemma 4.2. The Jacobian of every curve of C is isomorphic to Pic0(D), hence the smooth elements of C are all isomorphic to D. If d > 1, by [GP, Lemma 2.2.1] (cf. also [CCM, §0]) S is not of general type, against the assumptions. So we have d = 1 and the result follows by [CCM, Thm. 0.20]. (cid:3) Acknowledgments. The first author is a member of the Center for Mathemat- ical Analysis, Geometry and Dynamical Systems (IST/UTL). The second and the third author are members of G.N.S.A.G.A.–I.N.d.A.M. This research was par- tially supported by FCT (Portugal) through program POCTI/FEDER and Project PTDC/MAT/099275/2008 and by the italian PRIN 2008 project Geometria delle variet`a algebriche e dei loro spazi di moduli. References [Be] [Ca] [CaP] A. Beauville, L'in´egalit´e pg ≥ 2q − 4 pour les surfaces de type g´en´eral, appendix to [De1]. F. Catanese, On the moduli space of surfaces of general type, J. differential Ge- ometry, 19 (1984), 483–515. A. Causin and G. Pirola, Hermitian matrices and cohomology of Kaehler vari- eties, Manuscripta Math. 121 (2006), 157-168. [CFM] C. Ciliberto, P. Francia, M. Mendes Lopes, Remarks on the bicanonical map for surfaces of general type, Math. Z. 224 3 (1997), 137–166. [CCM] F. Catanese, C. Ciliberto, M. Mendes Lopes, On the classification of irregular surfaces of general type with non birational bicanonical map, Transactions of the AMS 350, No.1, (January 1998), 275-308. O. Debarre, In´egalit´es num´eriques pour les surfaces de type g´en´eral, with an appendix by A. Beauville, Bull. Soc. Math. France 110 3 (1982), 319–346. O. Debarre, Minimal cohomology classes and Jacobians, J. Algebraic Geom. 4 no. 2 (1995), 321–335. [De2] [De1] [GL1] M. Green, R. Lazarsfeld, Deformation theory, generic vanishing theorems and some conjectures of Enriques, Catanese and Beauville, Invent. Math. 90 (1987), 389–407. [GL2] M. Green, R. Lazarsfeld, Higher obstructions to deforming cohomology groups of [GP] [H] [HP] [Ho] line bundles, J. Amer. Math. Soc. 4 (1991), 87–103. L. Guerra, G.P. Pirola, On rational maps from a general surface in P3 to surfaces of general type, Advances in Geometry, 8 no. 2 (2008), 289-307. C.D. Hacon, Fourier transforms, generic vanishing theorems and polarizations of abelian varieties, Math. Z. 235 no.4 (2000), 717–726. C.D. Hacon, R. Pardini, Surfaces with pg = q = 3, Trans. Amer. Math. Soc. 354 no. 7 (2002), 2631-2638. A. Horing, Minimal classes on the intermediate Jacobian of a generic cubic three- fold, Communications in Contemporary Mathematics 12, no. 1 (2010), 55-70. 8 MARGARIDA MENDES LOPES, RITA PARDINI AND GIAN PIETRO PIROLA [KM] [M] [ML] K. Konno, M. Mendes Lopes, The base components of the dualizing sheaf of a curve on a surface, Arch. Math. 90 no. 5 (2008), 395–400. I.G. MacDonald, Symmetric products of an algebraic curve, Topology 1, no. 4 (1996), 319–343. M. Mendes Lopes, Adjoint systems on surfaces, Boll. Un. Mat. Ital. A (7) 10 (1996), no. 1, 169–179. [MP1] M. Mendes Lopes, R. Pardini, On surfaces with pg = 2q − 3, Advances in Geom- etry, 10 3 (2010), 549–557. [MP2] M. Mendes Lopes, R. Pardini, The geography of irregular surfaces, arXiv:0909.5195. [MP3] M. Mendes Lopes, R. Pardini, Severi type inequalities for surfaces with ample canonical class, to appear in Comment. Math. Helv. [MPP1] M. Mendes Lopes, R. Pardini, G.P. Pirola, On surfaces of general type with q = 5. arXiv:1003.5991 [MPP2] M. Mendes Lopes, R. Pardini, G.P. Pirola, On the canonical map of surfaces with q ≥ 6, preprint. [MPP3] M. Mendes Lopes, R. Pardini, G.P. Pirola, Curves on irregular surfaces and [Mu] [PP] [Pi] [R] [Ra1] [Ra2] [Se] [Xi] Brill-Noether theory, in preparation. S. Mukai, Curves and Grassmannians, Algebraic geometry and related topics (Inchon, 1992), 19–40, Conf. Proc. Lecture Notes Algebraic Geom., I, Int. Press, Cambridge, MA, 1993. G. Pareschi, M. Popa, Strong generic vanishing and a higher dimensional Castelnuovo-de Franchis inequality, Duke Math. J. 150, 2 (2009), 269-28 G.P. Pirola, Algebraic surfaces with pg = q = 3 and no irrational pencils, Manuscripta Math. 108 no. 2 (2002), 163–170. C.P. Ramanujam, Remarks on the Kodaira vanishing theorem, J. Indian Math. Soc 36 (1972), 121–124. Z. Ran, On subvarieties of abelian varieties, Invent. Math. 62 no.3 (1981), 459– 479. Z. Ran, On a theorem of Martens, Rend. Sem. Mat. Univ. Politec. Torino 44 no.2 (1987), 287–291. F. Serrano, Fibrations on algebraic surfaces, Geometry of complex projective varieties (Cetraro, 1990), 289–301, Sem. Conf., 9, Mediterranean, Rende, 1993 G. Xiao, Irregularity of surfaces with a linear pencil, Duke Math. J. 55 3 (1987), 596–602. Rita Pardini Dipartimento di Matematica Universit`a di Pisa Largo B. Pontecorvo, 5 56127 Pisa, Italy [email protected] Margarida Mendes Lopes Departamento de Matem´atica Instituto Superior T´ecnico Universidade T´ecnica de Lisboa Av. Rovisco Pais 1049-001 Lisboa, PORTUGAL [email protected] Gian Pietro Pirola Dipartimento di Matematica Universit`a di Pavia Via Ferrata, 1 27100 Pavia, Italy [email protected]
1603.00787
1
1603
2016-03-02T16:42:54
Computing jumping numbers in higher dimensions
[ "math.AG" ]
The aim of this paper is to generalize the algorithm to compute jumping numbers on rational surfaces described in [AAD14] to varieties of dimension at least 3. Therefore, we introduce the notion of $\pi$-antieffective divisors, generalizing antinef divisors. Using these divisors, we present a way to find a small subset of the `classical' candidate jumping numbers of an ideal, containing all the jumping numbers. Moreover, many of these numbers are automatically jumping numbers, and in many other cases, it can be easily checked.
math.AG
math
COMPUTING JUMPING NUMBERS IN HIGHER DIMENSIONS HANS BAUMERS AND FERRAN DACHS-CADEFAU Abstract. The aim of this paper is to generalize the algorithm to compute jumping numbers on rational surfaces described in [AAD14] to varieties of dimension at least 3. Therefore, we introduce the notion of π-antieffective divisors, generalizing antinef divisors. Using these divisors, we present a way to find a small subset of the 'classical' candidate jumping numbers of an ideal, containing all the jumping numbers. Moreover, many of these numbers are automatically jumping numbers, and in many other cases, it can be easily checked. 1. Introduction To an ideal sheaf a on a smooth algebraic variety X, one can associate its multiplier ideals J (X, ac), where c ∈ Q>0. They form a nested family of ideals in OX , which decreases when c increases. The values of c where the ideal changes are called the jumping numbers of the pair (X, a). They are very interesting geometric invariants, that were studied in [ELSV04], but also appeared earlier in [Lib83], [LV90], [Vaq92] and [Vaq94]. The jumping numbers determine in some sense how bad a singularity is. For example, if a is the ideal corresponding to a smooth hypersurface, then the jumping numbers are just the positive integers. When the ideal represents a more singular variety - or when it takes more blow-ups to obtain a log resolution - the jumping numbers are in general smaller and more numerous. The smallest jumping number is called the log canonical threshold. Koll´ar (see [Kol97]) proved that, if a is a principal ideal, it corresponds to the smallest root of the Bernstein- Sato polynomial. Ein, Lazarsfeld, Smith and Varolin (see [ELSV04]) generalized this result for all jumping numbers in the interval (0,1]. We are interested in ways to compute jumping numbers. For monomial ideals, Howald ([How01]) showed a combinatoric description of the multiplier ideal, which also allows to determine the jumping numbers. In [Tuc10], Tucker presents an algorithm to compute jumping numbers on surfaces with rational singularities. Alberich-Carraminana, `Alvarez Montaner and the second author ([AAD14]) introduce another algorithm in that setting. Shibuta ([Shi11]) constructed an algorithm to compute multiplier ideals and jumping numbers in arbitrary dimensions using D-modules, which was simplified by Berkesch and Leykin ([BL10]). The first author is supported by a PhD fellowship of the Research Foundation - Flanders (FWO). The second author was partially supported by Generalitat de Catalunya SGR2014-634 project, Spanish Min- isterio de Econom´ıa y Competitividad MTM2015-69135-P and by the KU Leuven grant OT/11/069. 1 2 H. BAUMERS AND F. DACHS-CADEFAU In this paper, we present an algorithm that can be used for computing jumping numbers in arbitrary dimensions, based on the algorithm in [AAD14]. The idea is to start with computing the so-called supercandidates, and then checking whether they are jumping numbers. The supercandidates can be computed in arbitrary dimensions, as long as we have enough understanding of the Picard groups of the exceptional divisors in a chosen resolution of a. For checking that they are jumping numbers, we give some possible criteria. Although we do not present a technique that works in full generality, we are able to compute jumping numbers of ideals where previous algorithms seemed to be insufficient (or got stuck while computing them). In Section 2 we introduce the basics on multiplier ideals and jumping numbers, together with some elementary results that we need. We also recall some of the notions introduced by Tucker in [Tuc10]. In Section 3, we define the π-antieffective divisors, which are a generalization of the notion of antinef divisors. We also construct a method to compute the π-antieffective closure of a divisor, generalizing the unloading procedure presented in [Lau72] and [EC15]. Lipman's correspondence between antinef divisors and integrally closed ideals (see [Lip69]) does not hold anymore in higher dimensions, but we present a weaker alternative in Section 3.2. Section 4 contains the core of the paper. Here we present our algorithm to compute the supercandidates, and ways to check whether they are jumping numbers. Finally, in Section 5 we present some illustrative examples. 2. Preliminaries Through this section, let X be a smooth algebraic variety with dim X = n > 2 over an algebraically closed field k of characteristic zero. Let a be a sheaf of ideals on X. We define a log resolution of the pair (X, a) as a birational morphism π : Y → X, such that • Y is smooth, • the pre-image of a is locally principal, i.e., a · OY = OY (−F ) for some effective Cartier divisor F , and • F + Exc(π) is a simple normal crossing divisor. By Hironaka's resolution of singularities (see [Hir64]), log resolutions always exist.Note that if a = OX (−D) for an effective divisor D, then F = π∗D, and π will be called a log resolution of (X, D) instead of (X, OX (−D)). Given a birational morphism π : Y → X, the relative canonical divisor measures in some way the difference between X and Y . Definition 2.1. Let π : Y → X be a birational morphism of smooth varieties, the relative canonical divisor of π is the divisor class Kπ := KY − π∗KX . It is important to notice that, even though KX and KY are only defined as divisor classes, the relative canonical divisor can be chosen to be an effective divisor, supported on the exceptional locus of π. Indeed, if X and Y are smooth, there is a unique way to write with ki ∈ Z>0, where the Ei are the irreducible components of Exc(π). Kπ =X kiEi , Another notion that we need to introduce is Q-divisors. Definition 2.2. A Q-divisor D on an algebraic variety Y is a formal finite sum D = 3 P aiDi, where the Di are irreducible codimension one subvarieties of Y , and ai ∈ Q. For any Q-divisor D =P aiDi, one denotes its round-down and round-up as and respectively. ⌊D⌋ =X⌊ai⌋Di ⌈D⌉ =X⌈ai⌉Di , 2.1. Multiplier Ideals. Having introduced these basic notions, we define multiplier ideals. Definition 2.3. Let a be a sheaf of ideals on X, π : Y → X a log resolution of a, and F the divisor satisfying a · OY = OY (−F ). For c ∈ Q>0, we define the multiplier ideal associated to a with coefficient c as where Kπ is the relative canonical divisor. J (X, ac) := π∗OY (Kπ − ⌊cF ⌋) , If a = OX (−D) for an effective divisor D on X, we will denote the multiplier ideals by J (X, cD). For simplicity, if no confusion can arise, we will simply write J (ac) or J (cD). Remark 2.4. It is clear from the definition of Kπ that π∗OY (Kπ) = OX , and therefore for any effective divisor N, we have π∗OY (Kπ − N) ⊆ OX . It is due to this fact that J (ac) ⊆ OX are subsheaves of OX, which justifies the name multiplier ideal. It is easy to see that J (ac) ⊆ J (ac′) if c > c′, and that J (a(c+ε)) = J (ac) if 0 6 ε ≪ 1. This yields the following result. Proposition-Definition 2.5. Let a be an ideal sheaf on X. There exists an increasing sequence of rational numbers 0 = λ0 < λ1 < λ2 < λ3 < . . . satisfying • J (aλi) ! J (aλi+1) for i ∈ N, • J (ac) = J (aλi) for c ∈ [λi, λi+1). The numbers λi, i > 0, are called the jumping numbers of a. 4 H. BAUMERS AND F. DACHS-CADEFAU The jumping numbers of a divisor D are defined analogously. The smallest jumping number is called the log canonical threshold of a or D, and is denoted by lct(X, a) or lct(X, D), respectively. This is a very important invariant of the pair (X, a) or (X, D), that appears in different branches of algebraic geometry. For a nice overview, we refer to [Kol97]. Now we fix some notations. For a pair (X, a), we fix a log resolution π : Y → X. We denote by F the divisor satisfying a · OY = OY (−F ). The irreducible components of F are denoted Ei, i ∈ I, and we write eiEi , F =Xi∈I where ei ∈ Z>0. The divisor F has an exceptional part and a non-exceptional part, also called the affine part. The affine part is sometimes denoted Faf f , and the exceptional components are denoted E1, . . . , Er. So we also have Note that Faf f = 0 whenever the support of a has codimension at least 2. F = Faf f + eiEi . r Xi=1 Let a ⊆ OX be an ideal on X, D be an effective divisor on X, F the divisor on Y defined as before and c a positive rational number. The multiplier ideals associated to a or D and c satisfy the following properties (see [Laz04]). • The definition of multiplier ideal does not depend on the resolution (Esnault- Viehweg in [EV92]). • (Local Vanishing) Riπ∗OY (Kπ − ⌊cF ⌋) = 0 for all i > 0 and c > 0. • We have that J (ac) is integrally closed for all c > 0. • The integers are jumping numbers for the pair (X, D). • For c > 0, we have that J ((c + 1)D) = π∗OY (Kπ − ⌊cπ∗D⌋ − π∗D) = J (cD) ⊗OX OX (−D). It follows that λ > 0 is a jumping number if and only if λ + 1 is a jumping number. • (Skoda's theorem) If m ∈ N with m > n, then J (am) = aJ (am−1) . Therefore, for any λ > n, one has that λ is a jumping number if and only if λ − 1 is a jumping number. • From the proof of Skoda's theorem, one can actually deduce a stronger result. If a is an ideal generated by ℓ elements and m > ℓ, then J (am) = aJ (am−1) . Therefore, for any λ > ℓ, one has that λ is a jumping number if and only if λ − 1 is a jumping number. 5 Remark 2.6. Lipman and Watanabe (see [LW03]), and independently Favre and Jonsson (see [FJ05]), proved that every integrally closed ideal in a two-dimensional regular local ring is a multiplier ideal. However, this is no longer true in higher dimensions. Lazarsfeld and Lee showed in [LL07] that if dim X > 3, integrally closed ideals need to satisfy certain conditions in order to be realized as multiplier ideals. The conditions allow them to give examples of integrally closed ideals that cannot be realized as multiplier ideals. 2.2. Contributing divisors. One can easily see from the definition of multiplier ideals that, with the notations above, the jumping numbers are contained in the set (cid:26) ki + n ei i ∈ I, n ∈ Z>0(cid:27) . (cid:12)(cid:12)(cid:12)(cid:12) These numbers are the candidate jumping numbers. It is important to notice that the smallest candidate is always a jumping number, and hence it equals the log canonical threshold. So we have and similar for a divisor D. Furthermore, if Ei is not exceptional, then ki = 0 and the candidates lct(X, a) = min(cid:26) ki + 1 ei i ∈ I(cid:27) , (cid:12)(cid:12)(cid:12)(cid:12) m ∈ Z>0(cid:27) (cid:26) m ei (cid:12)(cid:12)(cid:12)(cid:12) are always jumping numbers. For the exceptional ones, in general many candidate jumping numbers are not a jumping number. Definition 2.7. Let G be a reduced divisor supported on the exceptional part of π and λ a positive rational number. We will say that λ is a candidate jumping number for G if and only if λ can be expressed as ki+ni for each Ei 6 G with ni ∈ Z>0. ei A notion that is stronger and more interesting than being a candidate for a divisor, is being contributed by a divisor. This notion was introduced by Smith and Thompson in [ST07], and developed further by Tucker in [Tuc10]. Definition 2.8. [Tuc10, Definition 3.1] Let G be a reduced divisor supported on the exceptional part of π, and λ a candidate jumping number for G. We say that G contributes λ as a jumping number if π∗OY (Kπ − ⌊λF ⌋ + G) ! J (X, aλ) . We will say that this contribution is critical if moreover for any non-zero divisor G′ < G one has π∗OY (Kπ − ⌊λF ⌋ + G′) = J (X, aλ) . As an illustration of these concepts, we consider some examples in dimension two. 6 H. BAUMERS AND F. DACHS-CADEFAU Example 2.9. Let X be the affine plane and D = {y2 = x3}. Let π : Y → X be the minimal log resolution of D, and E1, E2 and E3 the exceptional divisors. Then we have π∗D = Daf f + 2E1 + 3E2 + 6E3, where Daf f is the strict transform of D. Moreover, we have that Kπ = E1 + 2E2 + 4E3, so the candidate jumping numbers are Hence (cid:26) 0 + n 1 , 1 + n 2 , 2 + n 3 , 4 + n 6 lct(X, D) = 5 6 n ∈ Z>0(cid:27) . (cid:12)(cid:12)(cid:12)(cid:12) is the smallest jumping number. Moreover, since we are in the case of a divisor, the integers are always jumping numbers, and the jumping numbers are periodic. Then one concludes that (cid:26) 5 6 , 1, , 2, 11 6 17 6 , 3, . . .(cid:27) is the set of jumping numbers. Clearly, all jumping numbers of the form 5 are contributed by E3, and all integers are contributed by Daf f . 6 + m for m ∈ N Example 2.10. [AAD14, Example 3.9] Let X be the affine plane again and consider the ideal a = (x2y2, x5, y5, xy4, x4y) ⊆ OX . Let π : Y → X be its minimal log resolution. Then Kπ = E1 + 2E2 + 4E3 + 2E4 + 4E5, and a · OY = OY (−F ), where F = 4E1 + 5E2 + 10E3 + 5E4 + 10E5. Using the fact that the log canonical threshold is the minimal candidate jumping number, one can see that lct(X, a) = 1 2. In order to compute all the jumping numbers in this case, we use the algorithm presented in [AAD14]. This yields that the set of jumping numbers associated to a is We can see that 7 by E3 and also by E5, 7 10 is contributed by E3 + E5. However, since it is (critically) contributed 10 is not critically contributed by E3 + E5. , (cid:26) 1 2 7 10 , n > 9(cid:27) . n 10(cid:12)(cid:12)(cid:12)(cid:12) The following result is a nice characterization of contribution which will be used in the following sections. It appears in [ST07] and [Tuc10, Proposition 4.1] for surfaces, but it holds in a more general setting. We repeat it here for completeness. First, we introduce a notation. Notation 2.11. If E is a subscheme of a scheme Y , ι : E → Y is the embedding, and F is a sheaf of OY -modules, then we denote by F E the OE-module ι∗F . Sometimes, if G is a sheaf of OE-modules, we will consider it as a sheaf on Y by simply writing G instead of ι∗G. Proposition 2.12. Suppose λ is a candidate jumping number for the reduced divisor G. Suppose that G is mapped onto an affine subscheme of X. Then λ is realized as a jumping number for (X, D) or (X, a) contributed by G if and only if H 0(G, OY (Kπ − ⌊λF ⌋ + G)G) 6= 0 . 7 Furthermore, this contribution is critical if and only if we have H 0(G′, OY (Kπ − ⌊λF ⌋ + G′)G′) = 0 for all divisors G′ on Y with 0 6 G′ < G. Proof. Consider the exact sequence 0 → OY (Kπ − ⌊λF ⌋) → OY (Kπ − ⌊λF ⌋ + G) → OY (Kπ − ⌊λF ⌋ + G)G → 0 . After pushing forward through π, we get 0 → J (X, aλ) → π∗OY (Kπ − ⌊λF ⌋ + G) → π∗(OY (Kπ − ⌊λF ⌋ + G)G) → 0 , since by local vanishing we have R1π∗OY (Kπ − ⌊λF ⌋) = 0. So we see that λ is a jumping number contributed by G if and only if π∗(OY (Kπ −⌊λF ⌋+ G)G) 6= 0. Since G is mapped onto an affine scheme, this is equivalent to H 0(G, OY (Kπ −⌊λF ⌋+ G)G) 6= 0. The second statement follows immediately from the definition of critical contribution. (cid:3) Remark 2.13. The condition that G is mapped onto an affine subscheme of X is a gener- alization of the two-dimensional case, where all the exceptional divisors are contracted to a point, and is also sufficient for our purposes, since we will only consider affine varieties X. Corollary 2.14 ([Tuc10, Corollary 4.2]). If G critically contributes a jumping number λ, then G is connected. Proof. Suppose that G is disconnected, so G = G′ + G′′ with 0 < G′, G′′ < G and G′ and G′′ disjoint. Then H 0(G, OY (Kπ − ⌊λF ⌋ + G)G) = H 0(G′, OY (Kπ − ⌊λF ⌋ + G′)G′) ⊕ H 0(G′′, OY (Kπ − ⌊λF ⌋ + G′′)G′′) . So if λ is contributed by G, it is also contributed by G′ or G′′, contradicting critical contribution. (cid:3) 3. π-antieffective divisors and integrally closed ideals From now on, we consider a regular local ring R over k of dimension at least 2, such that X = Spec R is the germ of a smooth algebraic variety over k. Let a be an ideal sheaf on X, fix a log resolution π : Y → X, and let F be the divisor satisfying a · OY = OY (−F ). We denote the relative canonical divisor by Kπ, this divisor is supported over the exceptional divisors Ei for i = 1, ..., r. If dim X = 2, recall the following definition. Definition 3.1. If Y is a surface, then a divisor D on Y is called antinef (or π-antinef ) if −D · Ei > 0 for all i ∈ {1, . . . , r}. 8 H. BAUMERS AND F. DACHS-CADEFAU This notion is introduced in [Lip69], and is also explained in [Tuc09]. A generalization of this concept to higher dimensions is given in [CGL96]. In section 3.1, we define π-antieffective divisors, which is a generalization to higher dimensions of antinef divisors. It is different from the one in [CGL96], but more useful for our purposes. We prove the existence of the π-antieffective closure and present a way to compute it. Lipman proved that in the two-dimensional case there is a one-to-one correspondence between antinef divisors on Y and integrally closed ideals in R defining invertible sheaves on Y . In higher dimensions, this correspondence does not hold anymore. We will prove a weaker version in Section 3.2. Before introducing π-antieffective divisors, we give a definition. Definition 3.2. We say that two divisors D1 and D2 on Y are equivalent if and only if they define the same ideal in R, i.e., if and only if π∗OY (−D1) = π∗OY (−D2). From now on, if we want to refer to linear or numerical equivalence, we will state it clearly, so no confusion will arise. 3.1. Unloading. The following definition is the generalization we want of the notion of antinef divisor. Definition 3.3. Let D be a divisor on Y with integral coefficients. We say that D is π-antieffective if and only if H 0(E, OY (−D)E) 6= 0 for every π-exceptional prime divisor E, i.e., if and only if −DE is a divisor class on E containing an effective divisor. In general, for any divisor, we can find a π-antieffective divisor equivalent to the given divisor. Theorem 3.4. Let D be a divisor on Y , then there exists a unique integral effective π-antieffective divisor D satisfying • D > D, and • for any π-antieffective divisor D′ such that D′ > D, we have D′ > D. Moreover, this divisor is equivalent to D. This leads us to the following definition. Definition 3.5. The π-antieffective divisor D satisfying • D > D, and • for any π-antieffective divisor D′ such that D′ > D, we have D′ > D, is called the π-antieffective closure of D. The proof of the theorem will be divided in several results. In the forthcoming lemma, we prove that such a minimal divisor exists. Later on (see Propositions 3.7 and 3.8), we prove that a divisor and its π-antieffective closure are equivalent. 9 Lemma 3.6. Let D be a divisor, then there exists a unique integral π-antieffective divisor D satisfying • D > D, and • for any π-antieffective divisor D′ such that D′ > D, we have D′ > D. Moreover, D − D is supported on the exceptional locus of π. Proof. This proof is based partially on Paragraph 2.2 in [Tuc09]. Let D be the set of all π-antieffective divisors D′ such that D′ > D. In the first part of the proof, we show that D is non-empty, while the second part is devoted to prove that D has a unique minimal element. We start with showing that D is non-empty. Take g ∈ R such that νi(g) > 0 for all divisorial valuations νi associated to one of the exceptional divisors Ei. Let G be the exceptional part of div(g), and take D0 := π∗π∗D + mG for sufficiently large m ∈ N. We claim that D0 ∈ D. Clearly D0 > D. Moreover, D0 is π-antieffective. Indeed, it equals π∗(π∗D + mdiv(g)) − mGaf f , where Gaf f is the affine part of π∗div(g), and hence, since Pic X = 0, we have that −D0E is linearly equivalent to mGaf f E, which is clearly effective. In order to check the unicity of a minimal element in D, assume that there exist i Ei in D. Now, define i }. Take an exceptional divisor E, and suppose that two different minimal divisors D1 = Pi d1 D′ = Pi diEi with di = min{d1 d1 E 6 d2 E. Then If Ei 6= E, then EiE is an effective divisor on E, and therefore PEi6=E(d1 i − di)EiE = D1E − D′E is an effective divisor on E. Since D1 is π-antieffective, −D1E defines the class of an effective divisor, so also −D′E defines an effective divisor class. By repeating this argument for any E, we conclude that D′ is also π-antieffective and satisfies D′ 6 D1 and D′ 6 D2, contradicting the minimality in D of D1 and D2. Finally, since D0 − D is supported on the exceptional locus of π, where D0 is the divisor constructed above, the same holds for D − D, where D is the unique minimal element of D. (cid:3) The following result tells us how to find equivalent divisors. Proposition 3.7. Let D be a divisor on Y and E an exceptional divisor of π. H 0(E, OY (−D)E) = 0, then If π∗OY (−D − E) = π∗OY (−D) . Proof. Denote by ι : E → Y the embedding of E in Y . Consider the short exact sequence 0 → OY (−E) → OY → ι∗OE → 0 . After tensoring with OY (−D) and pushing forward through π, we get the exact sequence (3.1) 0 → π∗OY (−D − E) → π∗OY (−D) → π∗(ι∗OE ⊗OY OY (−D)) . i , d2 −D′E = −d1 i Ei and D2 = Pi d2 EEE − XEi6=E diEiE . 10 H. BAUMERS AND F. DACHS-CADEFAU By the projection formula, we know that ι∗OE ⊗OY OY (−D) = ι∗(OE ⊗OE ι∗OY (−D)) = ι∗ι∗OY (−D) . Denote the image of E through π by C, and name the morphisms as follows. E ι−−−→ Y π′=πEy C πy ιC−−−→ X ∗ι∗OY (−D). Since OY (−D) is Then we obtain that the last sheaf in (3.1) equals ιC∗π′ invertible, we know that π′ ∗ι∗OY (−D) is a quasi-coherent OC-module (see [Har77, Propo- sition II.5.8]). Hence, since C is affine, this OC-module is determined by its global sections. Since by assumption ι∗OY (−D) has no global sections, also ιC∗π′ ∗ι∗OY (−D) has no global sections so we conclude that the sheaf equals to the zero sheaf. Hence by (3.1), π∗OY (−D − E) = π∗OY (−D) . (cid:3) This result gives a constructive way to find the π-antieffective closure of a divisor D, called the unloading procedure. It goes as follows. Let D be a non-π-antieffective divi- sor. Then there exists at least one π-exceptional divisor E such that H 0(E, OY (−D)E) = 0. We replace D by D′ := D + E and repeat until the obtained divisor is π-antieffective. Note that we cannot accidentally 'miss' the π-antieffective closure by chosing a specific order, since the order of adding Ei's does not matter. Indeed, if −DE1 and −DE2 are both not linearly equivalent to an effective divisor, then −(D + E1)E2 = −DE2 − E1E2 is not an effective divisor class either, so E2 still has to be added somewhere in the process. Therefore, we obtained the following proposition. Proposition 3.8. Let D be a divisor, then after finitely many steps of unloading we reach the π-antieffective closure of D. Together with the fact that we only encounter equivalent divisors during the unloading procedure, this finishes the proof of Theorem 3.4. Our unloading procedure is based on work of Enriques in [EC15]. It is also Laufer's It has also been described by algorithm to compute the fundamental cycle [Lau72]. Casas-Alvero [CA00] and Reguera [Reg97]. An improved version of it is used in [AAD14] to compute jumping numbers on surfaces with rational singularities. Here, we generalized the algorithm in [Lau72] to higher dimensions. The main difference is that checking positivity of an intersection number is replaced by checking effectivity of a divisor class. In Section 4, the unloading procedure will be used in an algorithm to compute jumping numbers. 11 3.2. A correspondence between globally generated invertible sheaves and in- tegrally closed ideals. The main goal of this section is to generalize the results of Lipman about the correspondence between integrally closed ideals and antinef divisors (see [Lip69, §18]). Lipman proves that in the two-dimensional case there is a one-to- one correspondence between antinef divisors and m-primary integrally closed ideals that determine invertible sheaves on Y . A first generalization of this result to higher dimensions is [CGL96, Proposition 1.20]. In this paper, the authors prove a similar relation between finitely supported integrally closed ideals and globally generated divisors on varieties obtained by finitely many point blow- ups. They also prove that if −D is globally generated, then −D·C > 0 for any exceptional curve C, as well as a counterexample for the reverse implication in dimensions higher than 2. We will prove a similar relation, which works for ideals that are not necessarily finitely supported, but which determine an invertible sheaf in a fixed birational morphism. The proof is essentially the same as the corresponding part of Lipman's proof in the two- dimensional setting. We repeat it here for completeness. Theorem 3.9. The mapping D 7→ ID = Γ(Y, OY (−D)) is a one-to-one correspondence between the set of effective divisors D on Y such that OY (−D) is generated by its global sections, and integrally closed ideals I of R such that I · OY is an invertible sheaf. The inverse is given by I 7→ DI, where DI is the divisor satisfying I · OY = OY (−DI ). Proof. If D is an effective divisor, then D = Pi diEi, where Ei runs over all prime divisors on Y . Moreover, di > 0 for all i, and di = 0 for all but finitely many i. We have OY (−D) ⊆ OY , and hence ID = Γ(Y, OY (−D)) ⊆ Γ(Y, OY ) = R . So ID is an ideal of R. Moreover, since OY (−D) is generated by its global sections, ID · OY = OY (−D) is invertible. Now we prove that ID is integrally closed. It is clear that ID = {f ∈ R ∀i : νi(f ) > di} , where νi is the divisorial valuation corresponding to Ei. Take f ∈ R and suppose f satisfies an equation f n + a1f n−1 + · · · + an−1f + an = 0 , where aj ∈ (ID)j for j = 1, . . . , n. Then for any i the properties of valuations yield nνi(f ) > min j {νi(aj) + (n − j)νi(f )} . So there exists j0 ∈ {1, . . . , n} such that nνi(f ) > νi(aj0) + (n − j0)νi(f ) , and hence j0νi(f ) > νi(aj0) > j0di, meaning that f ∈ ID. This implies that ID is integrally closed. Conversely, take an integrally closed ideal I such that I · OY is invertible. Then DI is such that I · OY = OY (−DI), so OY (−DI) is clearly generated by its global sections. 12 H. BAUMERS AND F. DACHS-CADEFAU Indeed, a set of generators of I determines global sections of I · OY , and their restrictions generate the stalks. By [Lip69, Proposition 6.2], IDI = Γ(Y, OY (−DI)) = I, since I is integrally closed. Conversely, if OY (−D) is generated by its global sections, then OY (−D) = ID · OY , and also ID · OY = OY (−DID ), which implies DID = D. (cid:3) In our setting we are interested in the relation between π-antieffective divisors and integrally closed ideals. However, no one-to-one correspondence between them is known. We do know that the set of π-antieffective divisors contains the set of divisors associated to an integrally closed ideal. Proposition 3.10. If OY (−D) is generated by global sections, then D is π-antieffective. Proof. Let E ⊂ Y be an exceptional divisor, and denote ι : E → Y and L = OY (−D). We have to show that H 0(E, ι∗L) 6= 0. Since L is globally generated, there exists an exact sequence of sheaves on Y . Pulling back by ι, we get an exact sequence 0 → K → Om Y → L → 0 ι∗K → Om E → ι∗L → 0 on E, so ι∗L is also generated by its global sections. In particular, H 0(E, ι∗L) 6= 0. (cid:3) The converse of the previous proposition is true in dimension 2 (see [Lip69, §18]), but in higher dimensions it does not hold anymore, as is clear from the following example, which is inspired heavily on Remark 1.24 in [CGL96]. Example 3.11. Let X = Spec R be a smooth affine three-dimensional variety. Consider the blowing-up at a point 0 on X, followed by blowing up at nine points on the exceptional divisor E0 in general position. This means that they lie on a non-singular cubic curve C0, and that C0 is the only cubic curve on E0 passing through these nine points. Denote the new exceptional divisors by E1, . . . E9, and the morphism by π : Y → X. We will show i=1 Ei is π-antieffective, but that OY (−D) is not generated by its that D := 3E0 + 4P9 global sections. Since −DE0 is linearly equivalent to the strict transform of the curve C0, it is effective in Pic E0. Furthermore, −DEi for i ∈ {1, . . . , 9} is the class of a line on Ei, hence is also effective in Pic Ei. So we see that D is π-antieffective. If OY (−D) would be globally generated, then also its restriction to E0 should be globally generated, as is clear from the proof of Proposition 3.10. But this is the sheaf defined by the strict transform of the curve C0. Since this is the only cubic curve on P2 through the nine points, this divisor cannot be moved, and hence the sheaf is not globally generated. 4. An algorithm to compute jumping numbers In this section, we discuss a technique that can be used to compute jumping numbers, based on the algorithm of Alberich-Carraminana, `Alvarez Montaner and the second author described in [AAD14]. ei (cid:12)(cid:12)(cid:12) minn ki+1 will be (4.1) where Dλ =Pi∈I eλ min(cid:26) ki + 1 + eλ ei i i ∈ I(cid:27) , (cid:12)(cid:12)(cid:12)(cid:12) i Ei is the π-antieffective closure of ⌊λF ⌋ − Kπ. 4.1. Computation of supercandidates. We will construct a set S ⊂ Q that contains all the jumping numbers, but is in general much smaller than the set of candidate jumping numbers. Following [AAD14], we work as follows. Let S be the empty set. One by one, we will add numbers to this set. First, we add the log canonical threshold λ1 = i ∈ Io to S. Assume that we added the value λ, then the next value we add 13 Definition 4.1. The elements of the set S are called supercandidates. Following Definition 4.3 in [AAD14], we define the minimal jumping divisor. Definition 4.2. If λ is a supercandidate, then the minimal jumping divisor associated to λ is the reduced divisor Gλ, supported on those components Ei where the minimum in (4.1) is reached. Remark 4.3. The name minimal jumping divisor is in contrast to the maximal jumping divisor, which is the reduced divisor supported on all the Ei for which λ is a candidate. We will not use this notion in this paper. Theorem 4.4. All jumping numbers are supercandidates. Proof. Suppose λ is a jumping number, which is not a supercandidate. Let λ′ be the largest supercandidate smaller than λ. Note that this λ′ always exists since lct(X, a) is a supercandidate smaller than λ, and the supercandidates are a discrete set. Then λ is strictly smaller then the supercandidate following λ′, i.e., λ < minn ki+1+eλ′ where Dλ′ =Pi∈I eλ′ is the π-antieffective closure of ⌊λ′F ⌋ − Kπ. But then −Dλ′ 6 Kπ − ⌊λF ⌋ 6 Kπ − ⌊λ′F ⌋ , ei i i i ∈ Io, (cid:12)(cid:12)(cid:12) and hence1 J (X, aλ) = J (X, aλ′ ) , contradicting the fact that λ is a jumping number. (cid:3) Remark 4.5. Note that we cannot conclude as in [AAD14] that every supercandidate is a jumping number, since there is no correspondence between π-antieffective divisors and integrally closed ideals in higher dimensions. However, no examples are known where not all supercandidates are actual jumping numbers. In the next subsection, we discuss some techniques to check whether supercandidates are jumping numbers. 1Indeed, if D1 6 D2, then OY (D1) ⊆ OY (D2), pushing forward preserves inclusion, and Dλ′ and ⌊λ′F ⌋ − Kπ are equivalent. 14 H. BAUMERS AND F. DACHS-CADEFAU 4.2. Checking supercandidates. Once we have our supercandidates, we have to check whether they are actual jumping numbers. An important tool is the following proposition. Proposition 4.6. If λ is a jumping number, then Gλ contributes λ. In particular, there is a divisor G critically contributing λ, satisfying G 6 Gλ. Moreover, if λ′ is the super- candidate previous to λ, we have π∗OY (Kπ − ⌊λF ⌋ + Gλ) = J (aλ′ ) . Kπ − ⌊λF ⌋ + Gλ > −Dλ′. In fact, for any fixed Ei, one has by formula (4.1) that i Ei be the π-antieffective closure of ⌊λ′F ⌋ − Kπ. We claim that Proof. Let Dλ′ =Pi∈I eλ′ so λ 6 ki + 1 + eλ′ i ei , ki − λei + 1 > −eλ′ i . If Ei 6 Gλ, this is an equality. Otherwise the inequality is strict, and then, since −eλ′ i an integer, after rounding up we get is ⌈ki − λei + 1⌉ > −eλ′ i + 1 , and hence ki − ⌊λei⌋ > −eλ′ i We conclude that indeed Kπ − ⌊λF ⌋ + Gλ > −Dλ′. . Note that also Kπ − ⌊λF ⌋ + Gλ 6 Kπ − ⌊λ′F ⌋, because λ is a candidate for Gλ, and hence π∗OY (−Dλ′) ⊆ π∗OY (Kπ − ⌊λF ⌋ + Gλ) ⊆ π∗OY (Kπ − ⌊λ′F ⌋) . By Theorem 3.4, all these ideals must be the same, and equal to J (aλ′). Since λ is a jumping number, we can conclude that J (aλ) π∗OY (Kπ − ⌊λF ⌋ + Gλ) , so Gλ contributes λ. In particular, there exists a G 6 Gλ critically contributing λ. (cid:3) Remark 4.7. Here it is important to notice that Gλ is not necessarily connected, unlike the critically contributing divisors (see Corollary 2.14). So in order to check whether a supercandidate λ is a jumping number, it suffices to check whether some G 6 Gλ contributes λ. By Corollary 2.14, we know that if we can find such a G, we can even find a connected one. The following proposition shows that a supercandidate is a jumping number if its min- imal jumping divisor has an irreducible connected component. Proposition 4.8. If λ is a supercandidate such that Gλ has a connected component that is irreducible, then λ is a jumping number. Proof. From the proof of the previous proposition, using the same notations, we see that 15 Kπ − ⌊λF ⌋ + Gλ = −Dλ′ + XEi66Gλ aiEi , where ai > 0. So for E 6 Gλ we have (Kπ − ⌊λF ⌋ + Gλ)E = −Dλ′E + XEi66Gλ ai EiE , which is effective in Pic E because Dλ′ is π-antieffective and E is different from the Ei that appear in the sum. Now if E is an irreducible connected component of Gλ, we have (Kπ − ⌊λF ⌋ + Gλ)E = (Kπ − ⌊λF ⌋ + E)E , so the fact that this divisor is effective implies by Propositon 2.12 that λ is a jumping number contributed by E. (cid:3) So if Gλ has an irreducible connected component, there is nothing to check anymore. This is actually a very important case, since in general many supercandidates have an irreducible minimal jumping divisor. In the other case, it would suffice to check whether Gλ contributes λ as a jumping number. However, in practice, this seems to be hard. Therefore, we suggest the following approach, which is more likely to work. Start by checking contribution by an irreducible E 6 Gλ. This can be done using Proposition 2.12, if we have enough understanding of Pic E. If this does not give a positive answer, check the connected G 6 Gλ consisting of two irreducible components, and proceed in this manner up to the maximal connected divisors G 6 Gλ. As soon as we find a G 6 Gλ contributing λ, we know that λ is a jumping number. If there is no such (connected) G, λ is not a jumping number. If G is reducible it is not very clear how to check whether it contributes a supercandidate λ as a jumping number. However, we have some tools that can be useful. Proposition 4.9. Suppose that λ is a supercandidate, such that λ is a candidate for ∼= OEi for i ∈ {1, 2}. Then OY (Kπ − G = E1 + E2. Suppose that OY (Kπ − ⌊λF ⌋ + G)Ei ⌊λF ⌋ + G)G ∼= OG, and hence λ is a jumping number contributed by E1 + E2. Proof. By Corollary 2.14, we can assume that E1 and E2 intersect transversally. Then by the following lemma, we see that OY (Kπ − ⌊λF ⌋ + G)G ∼= OG, and hence H 0(G, OY (Kπ − ⌊λF ⌋ + G)G) 6= 0. By Proposition 2.12, this means that G contributes λ as a jumping number. (cid:3) Lemma 4.10. Let G = E1 ∪ E2 be a closed reducible connected subvariety of a variety Y such that E1 and E2 intersect transversally, and D := E1 ∩ E2 is connected. Let L be a sheaf of OY -modules such that LEi ∼= OEi for i ∈ {1, 2}. Then LG ∼= OG. 16 H. BAUMERS AND F. DACHS-CADEFAU Proof. Consider the short exact sequence 0 → OG → OE1 ⊕ OE2 → OD → 0 on G. Here we consider sheaves OE, where E is equal to E1, E2 or D, as a sheaf on G by the pushforward through the inclusion morphism E ⊂ G. After tensoring with the restriction of L to G, we get 0 → LG → LE1 ⊕ LE2 → LD → 0. To compute the second and third term, we used the projection formula: if ι : E → G is the inclusion, where E is again equal to E1, E2 or D, then ι∗OE ⊗OG LG = ι∗(OE ⊗OE LE) = ι∗LE. By assumption, there are isomorphisms φi : OEi → LEi for i ∈ {1, 2}, and restricting D ◦ φ2D is an automorphism of D) = k∗. Composing to D gives isomorphisms φiD : OD → LD. Hence φ1−1 OD, so it corresponds to multiplication with an element c ∈ Γ(D, O∗ φ1 with multiplication with this constant yields φ1D = φ2D. This means that we have a commutative diagram 0 −−−→ OG −−−→ OE1 ⊕ OE2 −−−→ OD −−−→ 0 0 −−−→ LG −−−→ LE1 ⊕ LE2 −−−→ LD −−−→ 0. φ1⊕φ2y φDy Then we see that LG must be isomorphic to OG, since they are the kernel of the same morphism. (cid:3) Remark 4.11. The condition in Proposition 4.9 seems quite special, but it is the gener- alization of the analogous result in the two-dimensional case ([Tuc10, Proposition 4.1]). Moreover, we did not spot any other behaviour in concrete examples. Proposition 4.12. If G = E1 + · · · + Em is a connected divisor critically contributing λ as a jumping number, then H 0(Ei, OY (Kπ − ⌊λF ⌋ + G)Ei) 6= 0 for all i ∈ {1, . . . , m}. Proof. It suffices to prove the statement for E1. Denote G′ = E2+· · ·+Em, and D = E1G′. Then we have a short exact sequence 0 → OG′(−D) → OG → OE1 → 0. As before, we consider all the sheaves as sheaves on G by the pushforward. After tensoring with OY (Kπ − ⌊λF ⌋ + G)G, we get 0 → OY (Kπ − ⌊λF ⌋ + G′)G′ → OY (Kπ − ⌊λF ⌋ + G)G → OY (Kπ − ⌊λF ⌋ + G)E1 → 0, 17 and taking global sections gives 0 → H 0(G′, OY (Kπ − ⌊λF ⌋ + G′)G′) → H 0(G, OY (Kπ − ⌊λF ⌋ + G)G) → H 0(E1, OY (Kπ − ⌊λF ⌋ + G)E1 Since G contributes critically, H 0(G′, OY (Kπ − ⌊λF ⌋ + G′)G′) = 0 and H 0(G, OY (Kπ − ⌊λF ⌋ + G)G) 6= 0. Therefore H 0(E1, OY (Kπ − ⌊λF ⌋ + G)E1) 6= 0, which proves the proposition. (cid:3) Remark 4.13. This proposition can be used to decide that a divisor G 6 Gλ does not contribute λ as a jumping number. Remark 4.14. Our algorithm does not work in general, since we need to have enough understanding of the Picard groups of the exceptional divisors, and it is not always clear how to check the existence of global sections of sheaves on reducible varieties. However, it appears to be not realistic to develop a practical algorithm in full generality, since there is a wide range of possible singularities. Nonetheless, our technique seems to suffice in many situations. Remark 4.15. One could also determine the jumping numbers by computing all the candidate jumping numbers, and checking whether they are jumping numbers or not. This can be done by checking whether they are contributed by some divisor for which they are a candidate. However, our algorithm is more efficient in general, since if there are many exceptional divisors in a resolution, the set of candidate jumping numbers is much bigger than the set of supercandidates. Moreover, the minimal jumping divisor can be much smaller than the maximal one, which reduces the amount of possible contributing divisors. Also, Proposition 4.8 implies that several supercandidates do not need to be checked anymore. On the other hand, algorithms as [BL10] and [Shi11] have to pass by computations of generalized Bernstein-Sato polynomials, which is not easy in general. 5. Examples 5.1. Example 1. Let D be the germ of the surface given by x(yz − x4)(x4 + y2 − 2yz) + yz4 − y5 = 0 in the local ring at the origin in A3. We consider an embedded resolution given by six point blow-ups. We start by blowing up at the origin. Then we blow up at the singular point of the strict transform of D. In a third step we blow up at the singular point of the strict transform of D on the intersection of E1 and E2, followed by the intersection point of E2, E3 and the strict transform of D. The last step consists of blowing up at the two remaining singular points. Denoting the resolution by π : Y → X, we have Kπ = 2E1 + 4E2 + 8E3 + 14E4 + 6E5 + 6E6 and F = π∗D = Daf f + 5E1 + 9E2 + 16E3 + 27E4 + 11E5 + 11E6. 18 H. BAUMERS AND F. DACHS-CADEFAU All the exceptional divisors are projective planes, blown up at at most 4 additional points, so one can completely understand their Picard groups and effective cones. Applying the algorithm of section 4, we find the following supercandidates and their minimal jumping divisors. Gλ E2 + E4 E2 + E4 λ 5 9 2 3 20 27 7 9 23 27 8 9 25 27 26 27 1 E1 + E2 + E3 + E4 + E5 + E6 + Daf f . E2 + E4 E2 + E4 E4 E4 E4 E4 We verify one supercandidate to illustrate the algorithm. It is easy to see that lct(X, D) = 5 9. We have to compute the π-antieffective closure of (cid:4) 5 First, note that E1 is isomorphic to P2 blown up at two additional points, the second one infinitely near to the first one. As generators of the Picard group, we denote the pullback of a line by ℓ, the pullback of the first exceptional curve by e1, and the second exceptional curve by e2. Then −(E2 + E4)E1 = −e1 + e2, which is not effective. Hence, in the next step of the unloading, we consider E1 + E2 + E4. 9 F(cid:5) − Kπ = E2 + E4. Now we see that E3 is isomorphic to P2 blown up at one point. If we denote the generators of the Picard group by ℓ (pullback of a line) and e (the exceptional curve), we find −(E1 + E2 + E4)E3 = −2ℓ. Since E3E3 = −ℓ−e, we see that we need to add E3 twice to achieve an effective divisor on E3. So in the next step, we consider E1 + E2 + 2E3 + E4. Continuing in this manner, we add E4 twice, E5 and E6, and we can check that the obtained divisor E1 + E2 + 2E3 + 3E4 + E5 + E6 is π-antieffective. This implies that the next supercandidate is min(cid:26)0 + 1 + 0 1 , 2 + 1 + 1 5 , 4 + 1 + 1 9 , 8 + 1 + 2 16 , 14 + 1 + 3 27 , 6 + 1 + 1 11 (cid:27) = 6 9 = 18 27 = 2 3 , and this minimum is achieved for the terms coming from E2 and E4. All the supercandidates with irreducible minimal jumping divisor are jumping numbers contributed by E4, by Proposition 4.8. For the numbers 7 9 , one can see that (Kπ −⌊λF ⌋+E2)E2 is effective, so these numbers are jumping numbers contributed by E2. One can check that the supercandidates λ = 5 3 are not contributed by a single exceptional divisor. However, for these numbers, we have (Kπ − ⌊λF ⌋ + E2 + E4)Ei = 0 in Pic Ei for i ∈ {2, 4}, so, by Proposition 4.9, they are jumping numbers contributed 9 and λ = 2 9 and 8 19 by E2 + E4. Note that in fact we didn't need to check whether 5 9 is a jumping number, since this is the log canonical threshold. Finally, since we are in the case of a divisor, 1 is always a jumping number. Existing implemented algorithms ([BL10]) did not give a result after several days of computation. 5.2. Example 2. In this example we show that we don't need to understand the Picard groups and effective cones of all exceptional divisors completely in order to compute the jumping numbers. Take X = Spec C[x, y, z](x,y,z) and D the zero locus of (xd + yd + zd)2 + g(x, y, z), with d > 3 and g(x, y, z) a generic homogeneous polynomial of degree 2d + 1. In the first step of the resolution, we blow up at the origin of X. We denote the exceptional divisor by E1. Step 2 consists in blowing up at k = d(2d + 1) singular points of the strict transform of D; the exceptional divisors are denoted by Ep i . In step 3, we blow up at C, the intersection of E1 and Daf f . This is a curve of genus g = 1 2(d−1)(d−2). The exceptional divisor is denoted by E2. The final step of the resolution is blowing up at the intersection of E1, E2 and Daf f , which is also isomorphic to C. We denote the exceptional divisor by E3. We denote the composition of these blow-ups by π : Y → X. We have π∗D = Daf f + 2dE1 + (2d + 2) k k Xi=1 Ep i + (2d + 1)E2 + (4d + 2)E3, Kπ = 2E1 + 4 Ep i + 3E2 + 6E3. Xi=1 In order to compute the jumping numbers using the unloading procedure, we need to know the mutual intersections of all components of π∗D in the Picard groups of the exceptional divisors, as well as the self-intersections of the exceptional divisors, and we need to determine when a divisor of the form − a1E1 + ap k Xi=1 Ep i + a2E2 + a3E3!(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)E is effective in Pic E, where E varies over all the exceptional divisors. (Note that we take one coefficient for all the Ep i , since everything we will encounter will be symmetrical in the Ep i .) In order to do this, we have to know more about the exceptional divisors. The divisor E1 ⊂ Y is a projective plane, blown up at k = d(2d + 1) additional points. The Ep i are projective planes, blown up at two additional points, the second center lying on the exceptional curve of the first blow-up. The divisors E2 and E3 are ruled surfaces over C. We show how we can obtain conditions on the aj from the information on E3. The other exceptional divisors are treated similarly. 20 H. BAUMERS AND F. DACHS-CADEFAU Since E3 is a ruled surface over C, its Picard group equals Z ⊕ p∗(Pic C), where Z is generated by a section, say C ′. Here p denotes the canonical morphism E3 → C. It is not obvious to give a complete description of the effective cone, but we know that aC ′ + p∗d is effective if a > 0 and d is effective, or if a > 0 and deg d > d − 2 (in fact the first one suffices for our purposes), and that it is not effective if a < 0. It will turn out that this is enough information. We denote E1E3 = C1, Ep i E3 = fi, E2E3 = C2, Daf f E3 = C0, where all of the Cj's are sections, and the fi are fibres. From [Vey91, Proposition 2.1] we know that (4d + 2)E3E3 = −Daf f E3 − 2dE1E3 − (2d + 2) Ep i E3 − (2d + 1)E2E3 k Xi=1 = −C0 − 2dC1 − (2d + 2) k Xi=1 fi − (2d + 1)C2. In order to have more information about the self-intersection of E3, we compute the pullbacks of some additional divisors. First, if D1 is the divisor given by the zero locus of i + (2d + 1)E3. Analogously, for a generic plane H through the origin, we have π∗H = Haf f + E1 + E2 + i + 2E3. Since D1,af f does not intersect E3, and Haf f intersects E3 in d fibers, we xd + yd + zd, we see that π∗D = D1,af f + dE1 + (d + 1)E2 + (d + 1)Pk Pk i=1 Ep i=1 Ep find (2d + 1)E3E3 = −dC1 − (d + 1)C2 − (d + 1) k d fi, k Xi=1 fi − f ′ j, 2E3E3 = −C1 − C2 − Xi=1 i=1 fi − (2d + 1)Pd j are fibres. One easily verifies that Pk where the f ′ j = 0 in Pic E3, since it is the pullback of the principal divisor on C defined by the rational function g ℓ2d+1 , where ℓ is the linear polynomial defining H. (We considered the curve C embedded in the projective plane E1, before the other blow-ups, with its standard coordinates.) From this observation, together with the equalities above, one can conclude that C0 = C1 = C2 in j is the pullback of adivisor of degree Xj=1 j=1 f ′ j=1 f ′ Pic E3 and hence that E3E3 + C1 = −(d + 1)Pd −d(d + 1). So the remarks about the effective divisors on E3 above yield that a divisor 21 − a1E1 + ap k Xi=1 Ep i + a2E2 + a3E3!(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)E3 is effective if a3 > a1 + a2 and (d + 1)a3 > (2d + 1)ap, and that it is not effective if a3 < a1 + a2. A similar analysis on the other divisors yields the following conclusions. For E1 we obtain sufficient conditions (cid:26) (2d + 1)a1 > da3, (d + 1)a1 > dap, the first of which is a necessary condition. For the Ep i we obtain sufficient and necessary conditions The exceptional divisor E2 is also a ruled surface. Here we have sufficient conditions ap > a1, ap > a2, 2ap > a3.   (cid:26) 2a2 > a3, 2(d + 1)a2 > (2d + 1)ap, the first of which is necessary. All together, this gives the following set of sufficient conditions for a divisor a1E1 + ap k Xi=1 Ep i + a2E2 + a3E3! to be π-antieffective: a1 6 1 2 a3 6 a2 6 ap a3 6 2a1 + a3 > a1 + a2 1 d 1 d ap 6 a1 + a1 a1 ap < a2 + 2ap < a3 + 1 2d + 1 1 2d + 1 a2 a3.   The first three inequalities are necessary. Proposition 5.1. In this example, the set of jumping numbers in (0, 1] is A := A1 ∪ A2 ∪ A3, 22 where H. BAUMERS AND F. DACHS-CADEFAU A1 =n n 2d(cid:12)(cid:12)(cid:12) 3 6 n < do , 4d + 2 (cid:12)(cid:12)(cid:12)(cid:12) d 6 n 6 2d(cid:27) , A2 =(cid:26) 2n + 1 A3 =n n 2d(cid:12)(cid:12)(cid:12) d + 3 6 n 6 2do . Proof. Since d > 3, the log canonical threshold is λ1 = min(cid:26) 3 2d , 5 2d + 2 , 4 2d + 1 , 7 4d + 2 , 1(cid:27) = 3 2d . This is indeed the minimal value in A. Moreover, Gλ1 = E1 if d > 3, and E1 + E3 if d = 3, in which case λ1 = 1 2. The rest of the proof goes by inductively computing the next supercandidates. Given a supercandidate λ in one of the three sets, we compute ⌊λπ∗D⌋ − Kπ, and we unload using the set of necessary and sufficient conditions on the coefficients given above. Then the formula (4.1) gives the next supercandidate. We treat the case where λ ∈ A1 to show how the proof works. The other cases are similar. Suppose we know that λ = n 2d is a supercandidate, where 3 6 n < d. We have ⌊λπ∗D⌋ − Kπ = (n − 2)E1 + (n − 4) Ep i + (n − 3)E2 + (2n − 6)E3. k Xi=1 Using the conditions for π-antieffective divisors described above, we find that the π- antieffective closure is (n − 2)E1 + (n − 2) Then the next supercandidate is Ep i + (n − 2)E2 + (2n − 4)E3. k Xi=1 λ′ = min(cid:26) n + 1 2d , n + 3 2d + 2 , n + 2 2d + 1 , 2n + 3 4d + 2 , 1(cid:27) = n + 1 2d , which is indeed the next value in A. For the minimal jumping divisor, we have 2 or λ = 1, E1 + E3 if λ = 1 E1 if λ ∈ A1 or A3, E3 otherwise. Gλ =  By Proposition 4.8 and checking that (cid:0)Kπ −(cid:4) 1 supercandidates are jumping numbers contributed by E1 or E3. 2 π∗D(cid:5) + E3(cid:1)(cid:12)(cid:12)E3 is effective in Pic E3, all (cid:3) References 23 [AAD14] Maria Alberich-Carraminana, Josep Alvarez Montaner, and Ferran Dachs-Cadefau, Mul- local rings with rational singularities, ArXiv e-prints, tiplier ideals in two-dimensional arXiv:1412.3605 (to appear in Michigan Math. J.) (2014). Christine Berkesch and Anton Leykin, Algorithms for Bernstein-Sato polynomials and multi- plier ideals, ISSAC 2010 -- Proceedings of the 2010 International Symposium on Symbolic and Algebraic Computation, ACM, New York, 2010, pp. 99 -- 106. MR 2920542 Eduardo Casas-Alvero, Singularities of plane curves, London Mathematical Society Lecture Note Series, vol. 276, Cambridge University Press, Cambridge, 2000. [BL10] [CA00] [CGL96] Antonio Campillo, G´erard Gonzalez-Sprinberg, and Monique Lejeune-Jalabert, Clusters of [EC15] infinitely near points, Math. Ann. 306 (1996), no. 1, 169 -- 194. MR 1405323 (97k:14004) Federigo Enriques and Oscar Chisini, Lezioni sulla teoria geometrica delle equazioni e delle funzioni algebriche, N. Zanichelli, 1915. [ELSV04] Lawrence Ein, Robert Lazarsfeld, Karen E. Smith, and Dror Varolin, Jumping coefficients of [EV92] [FJ05] multiplier ideals, Duke Math. J. 123 (2004), no. 3, 469 -- 506. H´el`ene Esnault and Eckart Viehweg, Lectures on vanishing theorems, DMV Seminar, vol. 20, Birkhauser Verlag, Basel, 1992. Charles Favre and Mattias Jonsson, Valuations and multiplier ideals, J. Amer. Math. Soc. 18 (2005), no. 3, 655 -- 684. [Har77] Robin Hartshorne, Algebraic geometry, Springer-Verlag, New York, 1977, Graduate Texts in [Hir64] [How01] [Kol97] Mathematics, No. 52. Heisuke Hironaka, Resolution of singularities of an algebraic variety over a field of character- istic zero. I, II, Ann. of Math. (2) 79 (1964), 109 -- 203; ibid. (2) 79 (1964), 205 -- 326. Jason A. Howald, Multiplier ideals of monomial ideals, Trans. Amer. Math. Soc. 353 (2001), no. 7, 2665 -- 2671. J´anos Koll´ar, Singularities of pairs, Algebraic geometry -- Santa Cruz 1995, Proc. Sympos. Pure Math., vol. 62, Amer. Math. Soc., Providence, RI, 1997, pp. 221 -- 287. MR 1492525 (99m:14033) [Lau72] Henry B. Laufer, On rational singularities, Amer. J. Math. 94 (1972), 597 -- 608. MR 0330500 [Laz04] [Lib83] [Lip69] [LL07] [LV90] [LW03] (48 #8837) Robert Lazarsfeld, Positivity in algebraic geometry. II, Ergebnisse der Mathematik und ihrer Grenzgebiete., vol. 49, Springer-Verlag, Berlin, 2004. Anatoly S. Libgober, Alexander invariants of plane algebraic curves, Singularities, Part 2 (Arcata, Calif., 1981), Proc. Sympos. Pure Math., vol. 40, Amer. Math. Soc., Providence, RI, 1983, pp. 135 -- 143. MR 713242 (85h:14017) Joseph Lipman, Rational singularities, with applications to algebraic surfaces and unique fac- torization, Inst. Hautes ´Etudes Sci. Publ. Math. (1969), no. 36, 195 -- 279. MR 0276239 (43 #1986) Robert Lazarsfeld and Kyungyong Lee, Local syzygies of multiplier ideals, Invent. Math. 167 (2007), no. 2, 409 -- 418. Fran¸cois Loeser and Michel Vaqui´e, Le polynome d'Alexander d'une courbe plane projective, Topology 29 (1990), no. 2, 163 -- 173. MR 1056267 (91d:32053) Joseph Lipman and Kei-ichi Watanabe, Integrally closed ideals in two-dimensional regular local rings are multiplier ideals, Math. Res. Lett. 10 (2003), no. 4, 423 -- 434. [Reg97] Ana-Jos´e Reguera, Curves and proximity on rational surface singularities, J. Pure Appl. Al- [Shi11] gebra 122 (1997), no. 1-2, 107 -- 126. MR 1479350 (99g:14046) Takafumi Shibuta, Algorithms for computing multiplier ideals, J. Pure Appl. Algebra 215 (2011), no. 12, 2829 -- 2842. 24 [ST07] H. BAUMERS AND F. DACHS-CADEFAU Karen E. Smith and Howard M. Thompson, Irrelevant exceptional divisors for curves on a smooth surface, Algebra, geometry and their interactions, Contemp. Math., vol. 448, Amer. Math. Soc., Providence, RI, 2007, pp. 245 -- 254. [Tuc09] Kevin Tucker, Integrally closed ideals on log terminal surfaces are multiplier ideals, Math. Res. Lett. 16 (2009), no. 5, 903 -- 908. MR 2576706 (2011c:14055) [Tuc10] , Jumping numbers on algebraic surfaces with rational singularities, Trans. Amer. Math. Soc. 362 (2010), no. 6, 3223 -- 3241. [Vaq92] Michel Vaqui´e, Irr´egularit´e des revetements cycliques des surfaces projectives non singuli`eres, Amer. J. Math. 114 (1992), no. 6, 1187 -- 1199. MR 1198299 (94d:14015) [Vaq94] , Irr´egularit´e des revetements cycliques, Singularities (Lille, 1991), London Math. Soc. Lecture Note Ser., vol. 201, Cambridge Univ. Press, Cambridge, 1994, pp. 383 -- 419. MR 1295085 (95f:14030) [Vey91] Willem Veys, Congruences for numerical data of an embedded resolution, Compositio Math. 80 (1991), no. 2, 151 -- 169. MR 1132091 (93d:14027) KU Leuven, Department of Mathematics, Celestijnenlaan 200B box 2400, BE-3001 Leu- ven, Belgium E-mail address: [email protected], [email protected]
1904.07520
1
1904
2019-04-16T08:01:17
An action of the Polishchuk differential operator via punctured surfaces
[ "math.AG" ]
For a family of Jacobians of smooth pointed curves there is a notion of tautological algebra. There is an action of $\mathfrak{sl}_2$ on this algebra. We define and study a lifting of the Polishchuk operator, corresponding to $f\in \mathfrak{sl}_2$, on an algebra consisting of punctured Riemann surfaces. As an application we prove that a collection of tautological relations on moduli of curves, discovered by Faber and Zagier, come from a class of relations on the universal Jacobian.
math.AG
math
AN ACTION OF THE POLISHCHUK DIFFERENTIAL OPERATOR VIA PUNCTURED SURFACES GABRIEL C. DRUMMOND-COLE AND MEHDI TAVAKOL Abstract. For a family of Jacobians of smooth pointed curves there is a notion of tautological algebra. There is an action of sl2 on this algebra. We define and study a lifting of the Polishchuk operator, corresponding to f ∈ sl2, on an algebra consisting of punctured Riemann surfaces. As an application we prove that a collection of tautological relations on moduli of curves, discovered by Faber and Zagier, come from a class of relations on the universal Jacobian. 1. Introduction The study of algebraic cycles on moduli spaces of curves was initiated by Mumford in the influential article [8]. He developed intersection theory on such moduli spaces and defined the notion of tautological classes. These are the most natural algebraic cycles on the moduli space and many geometric constructions lead to tautological classes. The collection of tautological cycles generate a distinguished subring of the Chow ring, known as the tautological ring. A fundamental open question concerning tautological rings is to understand the space of all relations among tautological classes. The purpose of this note is to study the connection between two classes of tautological relations. The first class of relations was discovered by Faber and Zagier around 2000 in an unpublished work. The second class is based on the study of tautological classes on the universal Jacobian by Yin [19]. The method of Yin gives a powerful tool to produce a large class of tautological relations on moduli of curves. Conjecturally, his method should give a complete description of tautological rings. The main ingredient in his approach is the Polischuck differential operator D which acts on the tautological ring of the universal Jacobian. We will show that there is a natural lifting of this differential operator to an algebra built from punctured Riemann surfaces. The lifting of the operator D corresponds to gluing Riemann surfaces along the punctures and closing punctures with open discs. Using this combinatorial interpretation of D we are able to find closed formulas for a certain class of tautological relations 1 2 G. C. DRUMMOND-COLE AND M. TAVAKOL from the Jacobian side. We analyze these relations and show that they match with a class of Faber -- Zagier relations. Conventions 1.1. Throughout this note we consider algebraic cycles modulo rational equivalence. Chow rings and cohomology rings are taken with Q-coefficients. Acknowledgments. The first author was supported by the Institute for Basic Science under IBS-R003-D1. The second author was sup- ported by the Institute for Basic Science under IBS-R003-S1, by the Max Planck institute for mathematics, and by the Australian Research Council grant DP180103891. 2. Tautological classes on the universal Jacobian The tautological ring of a fixed Jacobian variety under algebraic equivalence was defined and studied by Beauville [1]. Tautological rings of families of Jacobian varieties under rational equivalence have been studied extensively since then. For more details see the references [6, 7, 13, 14, 15]. Here we consider the relative version of this story under rational equivalence following Yin [19] to which we refer the reader for precise definitions. Let π : C → S be a family of smooth In this curves of genus g ≥ 2 which admits a section s : S → C. article we will assume that the base scheme S is the universal curve Cg = Mg,1. Consider the relative Picard group Jg = Pic0(C/S) of divisors of degree zero. It is an abelian scheme of relative dimension g over the base S. The section s induces an injection ι : C → Jg from C into the universal Jacobian Jg. The geometric point x on a curve C is sent to the line bundle OC(x− s) via the morphism ι. For an integer k consider the associated endomorphism of Jg induced by multiplication with k on fibers of the family Jg → S. Definition 2.1. For integers i, j the subgroup CHi (j)(Jg) of the Chow group CHi(Jg) is defined as all degree i classes on which the morphism k∗ acts via multiplication with k2i−j. Equivalently, the action of the morphism k∗ on CHi Proposition 2.2. The Beauville decomposition of the Chow group of (j)(Jg) is multiplication by k2g−2i+j. CH Jg has the form CH∗(Jg) = Li,j CH(i,j)(Jg), where CH(i,j)(Jg) := (j) (Jg) for i ≡ j mod 2. In the following the Chow group of the universal Jacobian equipped with the intersection product is denoted by (CH∗(Jg), .) and we usually i+j 2 THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 3 drop the product sign to simplify the notation. But there is another product on the Chow group of Jg: Definition 2.3. The Pontryagin product x ∗ y of two algebraic cycles x, y ∈ CH∗(Jg) is defined as µ∗(π∗1x·π∗2y), where π1, π2 : Jg×SJg → Jg are the natural projections and µ : Jg ×S Jg → Jg is the addition morphism. Recall that to an abelian scheme A/S there is an associated Poincar´e line bundle P on A ×S At trivialized along the zero sections. Here, At denotes the dual abelian scheme Pic0 A/S. Definition 2.4. A polarization of A/S is a symmetric isogeny λ : A → At such that the pullback of the Poincar´e bundle via the morphism (idA, λ) : A → A ×S At is relatively ample over S. We say that λ is a principal polarization when λ defines an isomorphism. Any such polarization induces a line bundle Lλ in the rational Picard group of A with the following properties: • The line bundle Lλ is relatively ample over S, • It is symmetric, i.e. [−1]∗Lλ = Lλ, • It is trivialized along the zero section. The first Chern class of Lλ is called the universal theta divisor and it will be denoted by θ. For more details we refer the reader to [19, Chapter 2]. Let ℓ be the first Chern class of the Poincar´e bundle. Definition 2.5. The Fourier -- Mukai transform F is defined as F (x) = π2,∗(π∗1x · exp(ℓ)). It gives an isomorphism between (CH∗(Jg), .) and (CH∗(Jg),∗). We now recall the definition of the tautological ring of Jg: Definition 2.6. The tautological ring R∗(Jg) of Jg is defined as the smallest Q-subalgebra of the rational Chow ring CH∗(Jg) which con- tains the class of ι∗[C] and is stable under the Fourier -- Mukai transform and all maps k∗ for integers k. Remark 2.7. It follows that for an integer k the tautological algebra becomes stable under k∗ as well. The generators of R∗(Jg) are expressed in terms of the components of the curve class in the Beauville decomposition. Define the following classes: pi,j := F(cid:16)θ j−i+2 2 · ι∗[C](j)(cid:17) ∈ CH(i,j)(Jg). 4 G. C. DRUMMOND-COLE AND M. TAVAKOL We have that p2,0 = −θ and p0,0 = g[Jg]. The class pi,j vanishes for i < 0 or j < 0 or j > 2g − 2. The tautological class ψ is defined as ψ := s∗(K), where K is the first Chern class of the relative dualizing sheaf of the morphism C → S. The pullback of ψ via the natural map Jg → S is denoted by the same letter. The following fact is proved in [19, Theorem 3.6]: Theorem 2.8. The tautological ring of Jg is generated by the classes {pi,j} and ψ. In particular, it is finitely generated. 2.9. Lefschetz decomposition of Chow groups. The action of the Lie algebra sl2 on the Chow groups of a fixed abelian variety was studied by Kunnemann [4]. Polishchuk [13] studied the sl2 action for abelian schemes which works over families. We follow the standard convention that sl2 is generated by elements e, f, h satisfying: [e, f] = h, [h, e] = 2e, [h, f] = −2f. With this notation the action of sl2 on Chow groups of Jg is given by e : CHi (j)(Jg) → CHi+1 (j) (Jg) x → −θ · x, f : CHi h : CHi (j)(Jg) → CHi−1 (j)(Jg) → CHi (j) (Jg) (j)(Jg) θg−1 x → − (g − 1)! ∗ x, x → −(2i − j − g)x, The operator f restricted to the tautological ring of Jg is given by the following differential operator: D = 1 2 Xi,j,k,l(cid:18)ψpi−1,j−1pk−1,l−1 −(cid:18)i + k − 2 +Xi,j pi−2,j∂pi,j . i − 1 (cid:19)pi+k−2,j+l(cid:19) ∂pi,j ∂pk,l 3. Faber -- Zagier relations Let g ≥ 2 and consider the moduli space Mg of smooth curves of genus g. Consider the universal curve π : Cg → Mg and denote by ωπ its relative dualizing sheaf. The first Chern class of ωπ is denoted by K. In [8] Mumford defined the kappa class κi as the push-forward π∗(K i+1). It is an algebraic cycle of degree i. Notice that κ0 = 2g − 2. Definition 3.1. The tautological ring R∗(Mg) of Mg is defined as the Q-subalgebra of the rational Chow ring CH∗(Mg) of Mg generated by kappa classes. THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 5 In unpublished work Faber and Zagier studied the Gorenstein quo- tient of the tautological ring of Mg. Recall that there is an isomorphism Φ : Rg−2(Mg) ∼= Q. This follows from a result of Looijenga [5] which states that Rg−2(Mg) is at most one-dimensional and from the result of Faber [2] which shows that κg−2 is nonzero. There is a natural way to extend Φ to a group homomorphism by requiring that any element is sent to zero unless it is of degree g− 2. Each element x of the tautological ring R∗(Mg) defines a linear map Φ : R∗(Mg) → Q, Φx : R∗(Mg) → Q that sends an element y ∈ R∗(Mg) to Φ(x · y) ∈ Q. Definition 3.2. The Gorenstein quotient of the ring R∗(Mg), denoted G∗(Mg), is the quotient of R∗(Mg) by the ideal generated by all ele- ments x for which Φx defines the zero map. Let p = {p1, p3, p4, p6, p7, p9, p10, . . .} be a variable set indexed by the positive integers not congruent to 2 mod 3. The formal power series Ψ is defined by the formula: Ψ(t, p) =(cid:18)1 + tp3 + t2p6 + t3p9 + . . .(cid:19) · ∞Xn=0 +(cid:18)p1 + tp4 + t2p7 + . . .(cid:19) · (6n)! (3n)!(2n)! tn ∞Xn=0 (6n)! (3n)!(2n)! 6n + 1 6n − 1 tn. Let σ be a partition of σ with parts not congruent to 2 modulo 3. For such partitions define rational numbers αn(σ) as follows: log(Ψ(t, p)) =Xσ ∞Xn=0 αn(σ)tnpσ, where for σ the partition [1a13a34a4 . . . ], we use pσ to denote the mono- mial (pa1 4 . . . ). Define 1 pa3 3 pa4 then the relation γ :=Xσ ∞Xn=0 αn(σ)κntnpσ; [exp(−γ)]tn σ = 0 p 6 G. C. DRUMMOND-COLE AND M. TAVAKOL holds in the Gorenstein quotient G∗(Mg) of R∗(Mg) when g−1+σ < 3n and g ≡ n + σ + 1 (mod 2). In 2013 Pandharipande and Pixton [9] proved that Faber -- Zagier relations hold in the tautological ring of Mg. It is an open question whether all relations in the tautological ring follow from Faber -- Zagier relations. 3.3. Relations on moduli of curves from the universal Jaco- bian. The differential operator D provides a powerful tool to produce tautological relations. The crucial property of D is that it preserves the rational equivalence of algebraic cycles. Therefore, if we start from a collection of tautological relations we can produce a larger class by applying the differential operator D. These relations can be used to produce tautological relations on moduli of curves. Assume that the base scheme S is the universal curve Cg = Mg,1 as before. Definition 3.4. The tautological ring R∗(S) of S is defined to be the Q-subalgebra of the Chow ring CH∗(S) generated by kappa classes and the class of the relative dualizing sheaf ωπ of π : S → Mg. Remark 3.5. Consider the natural morphism π : Jg → S. According to [19, Corollary 3.8] the pullback homomorphism π∗ identifies R∗(S) with i=0 R∗(0,2i)(Jg) of R∗(Jg). Therefore, we also obtain relations on the universal curve Cg using the method explained above. These relations can be pushed down to Mg via the canonical map Cg → Mg to give relations in R∗(Mg) as well. All tautological relations on Cg for g ≤ 19 and on Mg for g ≤ 23 can be recovered using this method. the subspace L3g−2 Definition 3.6. Let ~ = (j0, j1, . . .) be an ordered set of indeterminates. Then the Q-linear bracket operation on power series is {xn}~ := xnjn Example 3.7. Let A(z) be the power series (3n)!(2n)!(cid:16) z 72(cid:17)n A(z) = (6n)! ∞Xn=0 . The top Faber -- Zagier relation of genus 3k − 1 for ~κ is the relation (3.8) [exp (−{log(A)}~κ)]zk = 0 This relation corresponds to the empty partition in the previous sec- tion. Remark 3.9. Replacing z with αz in the definition of A merely multi- plies the expression on the right side of Equation (3.8) by αk, so the choice of normalization factor 72 is unimportant. THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 7 Theorem 3.10. Let g = 3k−1. The vanishing of the tautological class p2k 3,1 on the universal Jacobian Jg gives the top Faber -- Zagier relation. Remark 3.11. The precise statement of Theorem 3.10 is that by Re- mark 3.5 the vanishing of p2k 3,1 on Jg gives a relation on the universal curve Cg over Mg. We obtain a relation on Mg by multiplying with K and pushing down. In general this would involve a further layer of complication. But in our case, we will show that our relation on Cg is in fact the pullback of the top Faber -- Zagier relation to Cg via the pro- jection Cg → Mg. This implies that the multiplication and pushdown procedure again yields the top Faber -- Zagier relation. The remainder of the paper will be devoted to the proof of this the- orem. We will begin, in Section 4, by giving an interpretation of the Polishchuk differential operator and several related operators on R∗(Jg) and related rings in terms of punctured surfaces. The Polishchuk op- erator itself is difficult to analyze directly, but some of the related operators are more amenable to enumerative combinatorics. One such modified operator will be shown to yield the top Faber -- Zagier relation. Then in Section 5, we will show that in the case of interest, the out- put of the modified operator in fact coincides with the output of the Polishchuk operator. As a roadmap, we have the following schematic chain of proportionalities and equalities: left hand side of (3.8) (q2k 3,1) ∝ ∂3k − =cpb ◦ ∂3k c− = pb◦∂3k ∝ (q2k 3,1)ξ0=6k−4 Poli(q2k 3,1)ξ0=6k−4 pushforward to Mg of the image of p2k 3,1 under the Polishchuck opera- tor Corollary 4.19 Section 5.7 via Lemma 5.6 Lemma 4.11 Remarks 4.5 and 4.10 which together assemble to the proof of Theorem 3.10. Remark 3.12. The conclusion of Theorem 3.10 is not itself surprising. We know from [19] that this method should give a relation in the tau- tological ring of Mg. According to [2] it is expected that the space of degree k relations in the tautological ring of Mg when g = 3k − 1 should be one dimensional. Therefore, we expect to get the unique known relation up to a scalar multiple, and in that sense the content is that the scalar multiple is not zero. Our emphasis instead is that the 8 G. C. DRUMMOND-COLE AND M. TAVAKOL method of proof is new -- this is a proof of concept that Yin's theory can be used effectively to extract concrete relations. 4. Lifting of the Polishchuk differential operator using punctured surfaces In this section we study several polynomial algebras related to the tautological ring of the universal Jacobian R∗(Jg). Notation 4.1. We use the notation Λq for the polynomial ring Q[qi,j, φ] on φ and variables qi,j with i ≡ j (mod 2) and i, j ≥ 0. We use the notationcΛq for the further extension of Λq by two more variables: cΛq = Λq[q0,−2, q1,−1]. Remark 4.2. We think of the generator qi,j as representing an orientable surface of Euler characteristic −j with i boundary components, and passing to the extended ring corresponds to allowing generators for the disk and the sphere as well as all other orientable surfaces of finite type. The evident inclusion and projection are maps of rings between Λq andcΛq. There is a projection π from Λq to R∗(Jg) defined by (4.3) pi,j i+j−2 φ 7→ ψ 4 qi,j 7→ 2 i! 2 (killing generators that are out of range). The projection π realizes will be our main players. R∗(Jg) as a quotient of Λq orcΛq. The following differential operators Definition 4.4. We define the following differential operators on Λq: • the one-component gluing operator 2(cid:19)qi−2,j∂qi,j , • the two-component gluing operator ∂1 :=Xi,j (cid:18)i ∂2 := Xi,j,k,l ∂′ψ :=Xi,j (cid:18)i 1 2 ikqi+k−2,j+l∂qi,j ∂qk,l, 2(cid:19)φqi−2,j−2∂qi,j , • the one-component capping operator and THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 9 • the two-component capping operator ikφqi−1,j−1qk−1,l−1∂qi,j ∂qk,l. ∂ψ := Xi,j,k,l 1 2 We also give names and notation for linear combinations of these atomic operators: ∂Poli := ∂1 − ∂2 + ∂ψ ∂± := ∂1 ± ∂2 ∂c± := ∂1 ± ∂2 + ∂ψ + ∂′ψ Polishchuk operator gluing operators surface operators We call the gluing and surface operators positive or negative according to the sign of ∂2. Finally, we use the same formulas, implicitly extending the indexing of the summations for operators acting on the ringcΛq, adding the word "extended" to the terminology and a hat to the notation. Remark 4.5. Change of basis reveals that the Polishchuk operator ∂Poli passes to the quotient R∗(Jg) as the Polishchuk differential operator D (see Section 2.9). Remark 4.6 (Warning). The differential operators which act on Λq in Definition 4.4 are not naively compatible with their extended versions acting oncΛq under inclusion and projection between Λq andcΛq. While the operators are not naively compatible, there is a compati- bility relation related to some further quotients. Definition 4.7. We use the notation Λξ for the ring Q[ξi, φ] where i varies over non-negative integers. We use the notation cΛξ for the ring Q[ξi, φ], where i ≥ −1. The evaluation from cΛξ and Λξ to R∗(Cg) The inverse pullback map pb : Λq → Λξ is the Q-linear map 4i , and ξ−1 to 0. projects φ to ψ 4 , ξi to κi (4.8) q0,2n 7→ nXr=0(cid:18)n + 1 r + 1(cid:19)φn−rξr + 2n+1φn. The extended inverse pullback map cpb :cΛq →cΛξ[φ−1] is Remark 4.10. As we saw in Remark 3.5 according to [19, Corollary 3.8] the pullback map π∗ : CH∗(Jg) → CH∗(Cg) descends to an isomorphism q0,2n 7→ ξn + φn. (4.9) 10 G. C. DRUMMOND-COLE AND M. TAVAKOL between the space L3g−2 universal curve Cg. If we map our ring Λξ to R∗(Cg) via i=0 R∗(0,2i)(Jg) and the tautological ring of the ξn 7→ κn 4n , φ 7→ ψ 4 , then our pb descends to his "π∗". Our formula (4.8) and Yin's formula (identity (3.7) in op. cit.): p0,2n 7→ 1 2n+1 nXr=0(cid:18)n + 1 r + 1(cid:19)ψn−rκr + ψn differ only by our change of basis (4.3). Lemma 4.11. LetfΛq be the subalgebra of Λq containing only qi,j with i = j + 2. The following diagram commutes: P∞ pb r=0 ∂r Poli Λq Λξ fΛq cΛq Λξ[φ−1] P∞ r=0 b∂r c− cpb cΛq cΛξ[φ−1] Proof. It suffices to check on a monomial ν in the variables qi,j. Write I for half of the total sum of all i indices in the variables of ν; then the only nonzero contribution comes from ∂I along the bottom. Then we are checking that c− Poli along the top and b∂I pb(∂I Poliν) =cpb(b∂I ν)ξ−1=0. c− It will be convenient to consider the sums involved in testing this equal- ity as occurring over surfaces `a la Remark 4.2, where boundary compo- nents are matched up with one another with cylinders (for applications of ∂1 and ∂2) or with pairs of caps (for ∂ψ and ∂′ψ). Then the sum c−ν can be indexed over all ordered perfect pairings of boundary components, along with a choice of "cylinder" or "caps" for each pair. Note that the condition i = j+2 means that each component at the beginning is of genus zero. making up b∂I Many such ordered labeled perfect pairings index the sum on the left side as well, but some are missing. We will consider a dichotomy of two types of such missing or mismatched terms. The first type consists of ordered perfect pairings where at some point in the iterated gluing procedure a disk (q1,−1) arises. The second type consists of surfaces THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 11 where there is never a disk and where a pair of caps is applied to a eventually in the gluing procedure the disk (q1,−1 term) must be glued to some other connected component. This is done either with a cylin- connected surface (i.e., the operator b∂′ψ is applied). For the first type, we note that b∂1 and b∂′ψ do not act on q1,−1, so der (via the two-component gluing operator b∂2) or a pair of caps (via the two-component capping operator b∂ψ); then this kind of missing term arises in pairs. That is, consider pairs which consist of the same operators in the same order except we swap the first application of a cylinder or caps between a disk and another surface. Then the eventual contribution from this pair consists of α(1 − q0,−2φ) for some α. But zero when κ−1 = 0. then evaluating via cpb yields cpb(α)(1 − κ−1φ − 1) which evaluates to involved trick, replacing every extended one-component capping (b∂′ψ) with an application of a cylinder (b∂1). The modified ordered labeled partial pairing which results indeed appears in the indexing set on the left hand side of the equation in ∂I Poliν. This is because no disks arise by assumption and no sphere can arise without either a disk or a pair of caps on the same connected surface. Call this assignment (starting with an ordered labeled perfect pairing and replacing extended one- component capping with extended one-component gluing) ζ. For the second type in our dichotomy, we perform a similar but more Then it will suffice to show that for each ordered labeled perfect pairing Υ and the corresponding monomial νΥ, we have the equality pb(νΥ) = XbΥ∈ζ −1(Υ)cpb(bνbΥ)(cid:12)(cid:12)(cid:12)(cid:12)ξ−1=0 . The connected components of the surface which arises from the gluing indexed by the ordered labeled perfect pairing Υ are in canonical bi- jection with the connected components of the surface for each bΥ in the summation. Then it suffices to check the above equality for a single connected component. So now let Υ− be an ordered pairing (not necessarily perfect, not labeled) on the boundary components of a connected surface of genus zero with (n + 1) pairs. These pairs correspond to one component gluings. Let bΥ− be a labeled version, where we label each pair either "cylinder" or "caps". Let α(bΥ−) be the number of "cylinder labels". Since we began with components of genus zero, and (extended) two- component gluings and cappings can never create genus, this eventu- ally correspnods to a surface with no boundary components and Euler 12 G. C. DRUMMOND-COLE AND M. TAVAKOL characteristic −2n. Then we must only show for all Υ− that r + 1(cid:19)φn−rξr + 2n+1φn nXr=0(cid:18)n + 1 =X φn+1−α(bΥ−)(cid:16)ξα(bΥ−)−1 + φα(bΥ−)−1(cid:17) , where the sum on the right is over all bΥ− that become Υ− by forgetting Υ−. This 2n+1 polarizes into(cid:0)n+1 r+1(cid:1) choices of bΥ− with α(bΥ−) − 1 = r. the labeling. Then this final equation follows from noticing that there are 2n+1 elements indexing the sum on the right hand side correspond- ing to a choice between cylinders and caps for each term in the pairing There is one term left over on the right, namely φn+1ξ−1, which evalu- ates to zero. (cid:3) We will also want to pick out a particular case. Notation 4.12. We use the notation β0,2k to denote the coefficient of q0,2k in ∂3k 3,1). For consistency in our formulas we let β0,2k+1 be zero. + (q2k +(q2 3,1). The monomial q2 Example 4.13. To illustrate the method we look at the case g = 2 and compute the expression ∂3 3,1 corresponds to two vertices and there are three leaves attached to every vertex of the graph. Every application of the operator ∂1 or ∂2 corresponds to gluing two half edges. We obtain two distinct isomorphism classes of graphs, depending on whether there are any self-gluings. In the first case, we have no self-gluings, and the coefficient of q0,2 is 36 = (3· 3)(2· 2)(1· 1). In the second case, we have two self-gluings, and the order of gluings matters, so the coefficient is 54 = 3(3 · 3)2, and so β0,2 is 90. See Figure 1 Figure 1. The graphs corresponding to ∂3 2 and ∂2∂2 1 − (q2k Our next goal is to count ∂3k 3,1) in order to relate it to the top Faber -- Zagier relation. It is too hard to directly obtain an explicit closed form for the coefficients involved in this expression, so we perform a kind of trick. First we calculate directly a particular evaluation of ∂3k + (q2k 3,1), which is easier both because the positive gluing operator is easier than the negative and because we're only interested in a special case. Then we formally express ∂3k 3,1) in terms of the (non-explicit) ± (q2k THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 13 coefficients β0,2k. Finally, we use the calculation in the special case to get an explicit formula for our case of actual interest. Let us turn to the positive gluing operator. It behooves us to further investigate the metaphor of Remark 4.2. all surfaces obtained by gluing two boundary components of Σ together. phism classes of (possibly disconnected) orientable surfaces with finite (possibly empty) labeled boundary, and let ΛΣ be the subspace spanned only by those surfaces with nonpositive Euler characteristic. Definition 4.14. Let cΛΣ be the Q-vector space spanned by homeomor- Let the gluing operator ∂glue on cΛΣ take a surface Σ to the sum of Let the capping operator ∂cap on cΛΣ take a surface Σ to the sum of all There are linear evaluationsbρ : cΛΣ →cΛq and ρ : ΛΣ → Λq which take disjoint unions to products and connected surfaces with i boundary components and Euler characteristic −j to qi,j. Lemma 4.15. The map ρ intertwines surfaces obtained by capping off two distinct boundary components of Σ with disks. (1) the operators ∂glue on ΛΣ and ∂+ on Λq; that is, ρ∂glue = ∂+ρ, and (2) the operators ∂glue + ∂cap on cΛΣ and ∂c,+φ=1 on cΛq, i.e., bρ(∂glue + ∂cap) = evφ=1 ∂c+bρ. Proof. Because gluing preserves Euler characteristic, the operator ∂glue restricts as follows ΛΣ ∂glue ΛΣ cΛΣ ∂glue cΛΣ so the first statement makes sense. Applying ∂1 on the right (in either case) corresponds to gluing two boundary components of the same connected surface, while applying ∂2 corresponds to gluing boundary components of two different connected surfaces. The coefficient(cid:0)i 2(cid:1) corresponds to the choice of two boundary components of a surface with i boundary components; the coefficient ik corresponds to choosing one boundary component each from surfaces with i and k respectively. The half is there because the formula is symmetric and otherwise would count each pair twice. Similarly, the 14 G. C. DRUMMOND-COLE AND M. TAVAKOL one-component and two-component capping operators correspond to the operations that cap two boundary components of a single connected component or of two distinct connected components, respectively. (cid:3) Corollary 4.16. We have the following identity: ∂3k + q2k 3,1(cid:12)(cid:12)(cid:12)(cid:12)q0,n=1 = (6k)! 8k (the evaluation happening for all n). Proof. Let Σ be the disjoint union of 2k pairs of pants. Then ∂3k + q2k 3,1 = ∂3k + ρ(Σ) = ρ∂3k glue(Σ). But ∂3k glue(Σ) is a sum over the perfect pairings on the 6k boundary components of Σ along with an ordering of the 3k pairs of the perfect pairing. Then there are (6k − 1)!!(3k)! = (6k)!(3k)! (3k)!23k = (6k)! 8k of these. (cid:3) Let P(n) denote the unordered partitions of the set {1, . . . , n} (see Appendix B for details on our notation). Lemma 4.17. We have the following identity: (4.18) z2k ∂3k ± (q2k 3,1) (3k)!(2k)! ∞Xk=0 = exp ∞Xn=1 ±β2n (3n)!(2n)! q0,2nz2n! . Proof. Since ∂± acts on the bigraded ring Λq by lowering the first grad- ing by two and preserving the second grading, necessarily ∂3k 3,1) is ± of the form (q2k Xp∈P(2k) αpYbi∈p q0,bi for some coefficient αp depending on the partition p (the partition keeps track of which copies of q3,1 have been glued together). Each bi must be even for the coefficient to be non-zero, so we may assume bi = 2ki now (i.e., αp = 0 if p has an odd length block). Each individual summand must arise by applying the two-component gluing operator ∂2 precisely 2ki − 1 times and the one-component gluing operator ∂1 precisely ki+1 times in some order to the monomial q2ki 3,1 . The sum of all the terms that arise in this way is then ±β2kiq0,2ki by Definition 4.12. The sign is negative for the negative gluing operator because ∂2 is necessarily applied an odd number of times. To get the coefficient αp, we need to combine these calculations over the blocks bi in the partition THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 15 p. This entails distributing the 3k applications of ∂± into individual (not necessarily contiguous) blocks of size 3ki. The number of ways of doing that is so we get ∂3k ± (q2k (3k)! , 2bi)! Qbi∈p( 3 3,1) = Xp∈P(2k) (3k)!Ybi∈p ±βbi ( 3 2bi)! q0,bi. By the exponential compositional formula (reviewed as Corollary B.3), the exponential generating function for 3,1) is the formal ex- ponential of the exponential generating function for ±βbi 2bi)! q0,bi. These are precisely the left and right sides of Equation (4.18). (3k)! ∂3k ± (q2k ( 3 (cid:3) 1 Corollary 4.19. Write ~q = (q0,0, q0,2, q0,4, . . .). Then (q2k 3,1) = 0 ∂3k − is the top Faber -- Zagier relation of genus 3k − 1 for ~q. Proof. First, evaluate the positive version of Equation (4.18) at q0,j = 1. For the right side use Corollary 4.16. Taking formal logarithms we then get the equation β2n (3n)!(2n)! ∞Xn=1 z2n = log ∞Xn=0 (6n)! (3n)!(2n)!(cid:18)z2 8(cid:19)n! = log(A(9z2)) where the rightmost expression uses the series of Example 3.7. Then the vanishing of the z2n term of the right hand side of Equa- tion (4.18) for the negative gluing operator is equivalent to the vanish- ing of the z2n term of the right-hand side, i.e., the zn term (dividing powers of z by two in the series) of β2n (3n)!(2n)! zn)~q  exp ∞Xn=1 −β2n (3n)!(2n)! q0,2nzn! = exp−( ∞Xn=1 = exp(cid:16)−{log(A(9z))}~q(cid:17) (see Definition 3.6 for notation). By Remark 3.9 the factor of 9 does not affect the equation, and so the vanishing of the zn term here coincides with the relation of Equation 3.8. (cid:3) 16 G. C. DRUMMOND-COLE AND M. TAVAKOL 5. The cancellation of contributions from ψ-classes The goal of this section is to complete the proof of Theorem 3.10, which says that the vanishing of the tautological class p2k 3,1 on Jg gives the top Faber -- Zagier relation on the tautological ring of Mg by the method explained in Remark 3.5. According to Corollary 4.19 we know that the 3k-fold application of the negative gluing operator ∂3k 3,1 gives the top Faber -- Zagier relation of genus 3k − 1. We also know via Lemma 4.11 that the relation arising from the Polishchuk operator coincides with that arising from the 3k-fold application of the negative surface oper- ator ∂3k − to the monomial p2k c− to the same monomial. However, the negative gluing and negative surface operators differ. The difference is given by the operator ∂ψ + ∂′ψ. To complete the proof, in this section we will show that all contributions of ψ classes cancel 5.1. Enumerative combinatorics for surface gluing. Our current and final goal is to perform the enumerative combinatorics for the ap- (p2k 3,1). c− with one other when we consider cpb ◦ ∂3k plicationcpb ◦ ∂3k Lemma 5.2. The expression ∂3k c− graphs as follows. c−(p2k 3,1). q2k 3,1 can be written as a sum over ∂3k c− q2k 3,1 = (3k)! 3kXr=0 (2r − 1)!!φr Xχ(Γ)=2r−2kYΓc −q0,−χ(Γc), where Γ runs over isomorphism classes of possibly disconnected ordered trivalent graphs of Euler characteristic 2r−2k with precisely 2k vertices and 2r leaves and Γc runs over connected components of Γ. Proof. This essentially follows from Lemma 4.15, which is written in terms of ∂c+ and surfaces involves the evaluation at φ = 1. However, of the four constituent operators of the negative surface operator ∂c−, only the two-component gluing operator changes the number of connected components (always reducing it by one) and only the capping operators change the Euler characteristic (always reducing it by two). Therefore we can recover the overall sign of a term as well as its power of φ from the combinatorics of the surface. Then isomorphism classes of closed orientable surfaces equipped with a fixed decomposition into pairs of pants and disks in which each connected component contains at least one pair of pants are in natural bijection with isomorphism classes of trivalent graphs with leaves -- vertices correspond to pants and leaves to disks. The Euler characteristic of a surface made of gluing 2k pairs of THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 17 pants and 2r disks is 2r−2k, while the Euler characteristic of a trivalent graph with 2k vertices and 2r leaves is r− k. The coefficients (3k)! and (2r − 1)!! come from choosing an order for the gluings and for choosing a perfect pairing between the caps for the capping operators. (cid:3) c−q2k Recall the extended inverse pullback map, (4.9). We want to apply this to our calculation of ∂3k 3,1 from Lemma 5.2. The extended inverse pullback map takes the term −q0,2n corresponding to the connected component Γc of Euler characteristic −2n to the sum −ξn − φn. We will reorganize this application of the extended inverse pullback map into a summation over powers of φ, which will turn out eventually to have little reliance on the monomial in ξn variables. For this purpose we introduce several generating functions so that we can perform the computation in formal series. See appendix A for our conventions on graphs (in particular the definition of an ordered graph). Definition 5.3. Let G+(n, m) be the set of isomorphism classes of (possibly disconnected) trivalent ordered graphs with n vertices and m leaves, such that each connected component has positive Euler char- acteristic. Similarly define G0(n, m) and G−(n, m) with the indicated Euler characteristic restrictions on each connected component. We use the superscripts c and ℓf to further restrict to graphs that are connected and leaf free, respectively. Let G+(x, y) denote the following generating function for G+(n, m): G+(x, y) = Xm,n≥0 #G+(n, m) xn n! ym. Define G0(x, y), G−(x, y), and the connected variants similarly. The leaf-free variant is obtained from the general formula by evaluating y = 0 and we will think of it as a single variable series instead. So G+ consists of forests, G0 of disjoint unions of "hairy loops", and G− of "everything else". We will also need to count another kind of tree. Definition 5.4. Let Trr(n, m) be the set of isomorphism classes of ordered trivalent trees T with n trivalent vertices and m + 2 leaves equipped with an ordered pair of distinct distinguished leaves, called the roots. Let Trr(x, y) denote the following generating function: Trr(x, y) = Xm,n≥0 #Trr(n, m) xn n! ym. 18 G. C. DRUMMOND-COLE AND M. TAVAKOL Now we will combine these generating functions into a single gener- ating function in five variables which will keep track of a complicated weighted count of graphs. The behavior of these series will give us a 3,1) to the simpler expression key to relate the desired quantitycpb ∂3k ∂3k − Definition 5.5. The master series Ω(x, y, z, w, u) is the following for- mal series: 3,1) analyzed in Corollary 4.19. (q2k (q2k c− Ω(x, y, z, w, u) = ezTrr(x,y) ewGc 0(x,y)G+(x, y)Gℓf −(xu) . The following technical lemma providing a kind of evaluation of the master series constitutes the promised key. Lemma 5.6. Given a non-negative integer n, construct a one-variable series Ωev(x) by performing the following R-linear substitution of mono- mials on Ω: xn1yn2zn3wn4un5 7→(0 (n2 − 1)!!n3!(cid:0) 3 2 n5+3n n3 n2 odd; (cid:1)(3n1 + 6n − 3)n4xn1 n2 even. Then Ωev is well-defined and equal to 1 for all n. We will defer the proof of this lemma, which is a somewhat involved exercise in formal series, to Appendix C, and meanwhile use it to prove the main theorem. 5.7. Proof of the main theorem. We are now ready to complete the proof of Theorem 3.10. As per the discussion at the beginning of the section, it is enough to show (q2k 3,1)ξ0=6k−4 = ∂3k − (q2k 3,1) c− cpb ◦ ∂3k for all k. The evaluation at 6k−4 corresponds to the identity κ0 = 2g−2 noted at the beginning of Section 3 and the assignment g = 3k − 1 of Example 3.7. Lemma 5.8. Let g = 3k − 1 for k ≥ 1. Then c− cpb ◦ ∂3k q2k 3,1ξ0=6k−4 = (3k)! (2r − 1)!! 3kXr=0 Xχ(Γξ⊔Γ0⊔Γ±)=r−k φr−χ(Γ±)(3 − 6k)#π0(Γ0)(−1)#π0(Γ±) YΓc⊂Γξ −ξ−χ(Γc), THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 19 where the summation on the second line is over isomorphism classes of ordered trivalent graphs of Euler characteristic r − k with 2k vertices and 2r leaves equipped with a decomposition as a disjoint union into (1) a graph Γξ all of whose connected components have negative Euler characteristic, (2) a graph Γ0 all of whose connected components have zero Euler characteristic, and (3) a graph Γ± all of whose connected components have nonzero Euler characteristic and the product is over connected components of Γξ. Proof. By Lemma 5.2, we can write ∂3k c− (q2k 3,1) as a sum over graphs. Then applying cpb corresponds, for such a graph Γ, to choosing some subgraph Γξ consisting of a collection of connected components of Γ to evaluate via q0,2n 7→ ξn and the complementary subgraph Γ \ Γξ to evaluate via q0,2n 7→ φn. This needs a little modification because of the exceptional values when n = −1 and n = 0. That is, when n = −1, we evaluate φ−1 + ξ−1 to φ−1 and when n = 0 we evaluate φ0 + ξ0 to 6n − 3 as described above. So we should separate out cases according to the sign of the Euler characteristic of connected components, and only include connected components with negative Euler characteristic in Γξ. (cid:3) We would like to more or less "hold Γξ fixed" in Lemma 5.8 and show that the coefficients are individually zero except for the trivial power of φ, i.e., the case r − χ(Γ±) = 0. But there is a relationship between Γξ and the summation index r which makes this slightly awkward. To get around this issue, we will use the decomposition of Appendix A. This decomposition starts from a conneced trivalent graph Γ of negative Euler characteristic and outputs a connected leaf-free trivalent graph of the same Euler characteristic, called the core of Γ, along with a collection of trivalent doubly rooted trees, the insertion forest. Lemma 5.9. Let g = 3k + 1 for k ≥ 1. Then 3,1 = (3k)!XΓ′ −)(cid:18) 3 ξ −ξ−χ(Γc)  YΓ′ #π0(Γins) (cid:19)(3 − 6k)#π0(Γ0)(−1)#π0(Γ′ (2r − 1)!! 3kXr=0 2 (#V ′ξ + #V ′ − c⊂Γ′ ξ ) −⊔Γ+) q2k c− cpb ◦ ∂3k XΓ′ −,Γ0,Γ+,Γins⊂Γ where φr−χ(Γ+)−χ(Γ′ 20 G. C. DRUMMOND-COLE AND M. TAVAKOL (1) Γ′ξ and Γ′ − (2) Γ0 varies over ordered trivalent graphs where every component vary over leaf-free ordered trivalent graphs, has Euler characteristic zero, (3) Γ+ varies over ordered trivalent forests, (4) Γins varies over ordered trivalent forests where each connected component is given an ordered pair of distinct roots, such that the total number of vertices of all five graphs is 2k and the total number of non-root leaves is 2r, and (5) all of the orders are subordinate to an order on the disjoint union Γ of all of the five graphs. Proof. Beginning with the situation of Lemma 5.8, we can further de- compose Γ± into Γ− (its components with negative Euler characteristic) and Γ+ (its components with positive Euler characteristic), and per- form the reduction procedure of the appendix on Γc and Γ−. Then we have cores Γ′ξ of Γξ and Γ′ − of Γ−) and insertion forests for them. If we like, we can think of a single insertion forest Γins for the disjoint union Γ′ξ ⊔ Γ′ . Then the number of graphs with insertion forest Γins and − #π0(Γins)(cid:1). Since the core is leaf-free and trivalent, we have core Γ′ is(cid:0) E(Γ′) 3V = 2E, so we can write the edge count as 3 2 the vertex count. (cid:3) Now given the equation of Lemma 5.9, write n1 for the total number of vertices of Γ′ − ⊔ Γ0 ⊔ Γ+ ⊔ Γins and write 2n for the total number of vertices of Γ′ξ. Note that χ(Γ+)+χ(Γ′ −) = n1−r. Another simplification comes from noticing that r ≤ 3k is not a necessary constraint to specify because it follows automatically from graph combinatorics, which can be seen as follows. Leaf-free graphs, connected trivalent graphs of Euler characteristic zero, and doubly rooted trees all have at least as many vertices as non-root leaves. On the other hand, a tree has 2 more leaves than vertices. Then for a fixed k, the maximum number of non-root leaves that can occur with a set of graphs (Γ′ξ, Γ′ , Γ0, Γ+, Γins) as above − is when Γins has 2k connected components, each a single vertex. In this case there are 6k = 2(3k) leaves so r > 3k is not possible. n1 the binomial coefficient comes from choosing the which 2n vertices of the 2k lie in Γ′ξ. The coefficient of this term is then: Then we can separate out a factor of (3k)!(cid:0)2k ∞Xr=0 (2r − 1)!! XΓ′ −,Γ0,Γ+,Γins(cid:18) 2n(cid:1)PQ−ξ−χ(Γc) where #π0(Γins)(cid:19)· 2#V ′ − #V ′−, #V0, #V+, #Vins(cid:19)φn1(cid:18)3n + 3 (3 − 3(n1 + 2n))#π0(Γ0)(−1)#π0(Γ′ −⊔Γ+). THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 21 Now the sum is being taken over tuples of ordered trivalent graphs (Γ′ , Γ0, Γ+, Γins) with Euler characteristic restrictions and extra root − data for Γins but without any ambient graph Γ. This sum in turn is obtained from the formal series in x, y, z, w, and u (5.10) XΓ′ −,Γ0,Γ+,Γins n1!(−1)#π0(Γ′ −⊔Γ0⊔Γ+) #V ′−!#V0!#V+!#Vins! xn1y2rz#π0(Γins)w#π0(Γ0)u#V ′ − by linearly replacing xn1yn2zn3wn4un5 as in Lemma 5.6 and evaluating at x = φ. Lemma 5.11. The coefficients of xn1yn2zn3wn4un5 in (5.10) and in the master series Ω(x, y, z, w, u) of Definition 5.5 agree up to a nonzero scalar multiple depending only on n1 which is 1 for n1 = 0. Proof. Rescale the coefficients of (5.10) by dividing by n1!. Then we can decompose the rescaled series, using the fact that Γ′ has no leaves, − Γ0 and Γins have as many non-root leaves as vertices, and Γ+ has two leaves more than its number of vertices: XΓ′ −,Γ0,Γ+,Γins (−1)#π0(Γ′ −⊔Γ0⊔Γ+) #V ′−!#V0!#V+!#Vins! xn1y2rz#π0(Γins)w#π0(Γ0)u#V ′ − =XΓ′ − (−1)#π0(Γ′ #V ′−! XΓ+ −) (xu)#V ′ −XΓ0 (−w)#π0(Γ0) #V0! (xy)#V0 (−y2)#π0(Γ+) #V+! (xy)#V+XΓins z#Vins #Vins! (xy)#Vins. These are all exponential generating functions for the appropriate types of ordered trivalent graphs, so this product is exp(−Gℓf,c 0(x, y)) exp(−Gc which is the master series. − (xu)) exp(−wGc +(x, y)) exp(zTrr(x, y)), (cid:3) c−q2k no dependence on φ. Proof of Theorem 3.10. By Lemma 5.6, for any n the coefficient of xn1 in Ωev(x) is 1 if n1 = 0 and 0 otherwise. Then by Lemma 5.11, the same is true for the "evaluation" `a la Lemma 5.6 of the expres- sion (5.10). This implies that for each choice of Γ′ξ, the coefficient in 3,1ξ0=6k−4 of the corresponding expression in ξj variables has 3,1ξ0=6k−4 differ 3,1? The operator ∂c− differs from ∂− by terms with a coeffi- cpb ◦ ∂3k But how does the φ0 term of the expressioncpb◦ ∂3k cient of φ. On the other hand, the operatorcpb differs from the identity from ∂3k c−q2k − q2k 22 G. C. DRUMMOND-COLE AND M. TAVAKOL by terms with positive powers of φ and special terms with power φ0 for q0,0 and φ−1 for q0,−2. But to arrive at q0,0 we must have a surface of genus 1, implying at least one cap, and to arrive at q0,−2 we must have a surface of genus 0, implying at least three caps. Since caps only arise (in pairs) from the application of ∂ψ and ∂′ψ, this means that for terms containing qi such operators, yielding a total power of φ of at least 3j+i 2 − j. This is greater than zero unless i = j = 0. But in the i = j = 0 case we are looking at summands where we have applied only ∂− and then avoided the two q2k 3,1ξ0=6k−4 is special cases of cpb. Then the coefficient of φ0 in cpb ◦ ∂3k 0,−2 there must have been at least 3j+i 0,0qj exactly ∂3k − q2k 3,1. 2 c− (cid:3) 6. Final remarks The following conjecture was proposed by Yin: Conjecture 6.1. Every relation in the tautological ring of Cg comes from a relation on the universal Jacobian Jg. Yin further conjectured that the sl2 action is the only source of all tautological relations. For more details and precise statements we refer to [19, Conjecture 3.19]. In a similar way one can use relations on the universal Jacobian and produce relations on all powers of the universal curve. Conjecturally these relations are enough. In [17] it is proved that all tautological relations on Cn 2 can be obtained from the universal Jacobian. From the results in [11] the same is true for the moduli spaces Cn g for g = 3, 4 and n ∈ N as well as Cn 5 when n ≤ 7. By the results in [10, 18] this procedure leads to a conjectural description of the space of relations on the moduli space Mrt g,n of stable n-pointed curves of genus g with rational tails. The analogous version of Faber -- Zagier relations on the moduli space Mg,n are introduced by Pixton [12]. These relations are conjectured to be all tautological relations. There is a natural way to restrict these relations on the space Mrt g,n and push them forward to Cn g which contracts all rational components. It is a reasonable hope that the method studied in this article can be used to prove more Faber -- Pixton -- Zagier relations on Cn g from the universal Jacobian. This would give a partial positive answer to [18, Conjecture 3.24]. g via the morphism Mrt g,n → Cn Appendix A. Trivalent graphs and cores In this appendix, we describe a structure theorem for trivalent graphs (i.e., all vertices are of valence 1 or 3 -- but the graph may have leaves) THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 23 which realize such graphs as being built by inserting a collection of trees into a closed trivalent core graph. A.1. Conventions on graphs. We consider a graph Γ as being a tuple (V, H, τ, ι) where V is the set of vertices, H is the set of half- edges, τ : H → V is the target map, and ι : H → H is an involution. An edge is a non-fixed orbit of ι, and the set of edges is denoted E or E(Γ). A leaf is a fixed orbit of ι. A graph is leaf-free if it has no leaves. If v is a vertex then τ−1(v) is the set of half-edges of v. The valence of v is the cardinality τ−1(v); in particular trivalent means valence 3. The geometric realization Γ of a graph Γ is the topological space (([0, 1] × H) ⊔ V )/ {(0, h) ∼ τ (h), (t, h) ∼ (1 − t, ι(h))} . If h is a half-edge then τ (h) is the vertex of h. A connected component of a graph Γ is a subgraph whose geometric realization is a connected component of Γ. A graph is a tree if its geometric realization is simply connected and a forest if its connected components are trees. The Euler characteristic of Γ is χ(Γ) = V −E. Lemma A.2. The Euler characteristic of a trivalent graph is L−V where L is the set of leaves. Proof. We know 3V = H and H = 2E + L so E = 3V −L . (cid:3) 2 2 An ordered graph is a graph equipped with an order on its vertices and an order on the half-edges of each vertex. Remark A.3. Graphs, and even ordered graphs, can have automor- phisms, but a leaf-free ordered graph has only the identity automor- phism. A.4. The core of a trivalent graph. Definition A.5. Let Γ be a trivalent graph. A half-edge h of Γ is superfluous if either (1) the half-edge h is a leaf or (2) the half-edge h "points toward a tree" in the sense that there is a simply connected component of Γ \ {h, ι(h)} which does not contain τ (h) but is in the same connected component as τ (h) in the larger ambient graph Γ. A vertex is superfluous if it has a superfluous half-edge and is otherwise core. Definition A.6. Let Γ be a (possibly ordered) trivalent graph. The core of Γ is the graph Γ′ = (V ′, H′, τ′, ι′) (ordered if Γ is) whose 24 G. C. DRUMMOND-COLE AND M. TAVAKOL (1) vertices are the core vertices of Γ, (2) half-edges are the half-edges of the core vertices, (3) whose target map τ′ is induced, (4) whose half-edge bijection ι′ is induced by the bijection ι of Γ in the sense explained below, and (5) whose orders on vertices and half-edges are induced by those of Γ. For h a half-edge of a core vertex v0, construct a sequence (h = h0, h1, . . . , hn = ι′(h)) of half-edges recursively as follows. Suppose (h0, . . . , h2k) is constructed. Let h2k+1 = ι(h2k). If τ (h2k+1) is a core vertex, then n = 2k + 1 and we are done. Otherwise, τ (h2k+1) is superfluous and thus has at least one super- fluous half-edge. The half-edge h2k+1 cannot be superfluous because then v0 would be superfluous (it would imply v0 is a vertex of a sub- tree in Γ \ {h2k, h2k+1}, which in turn would imply that both half-edges of v0 other than h0 were superfluous). On the other hand, if both the remaining half-edges of τ (h2k+1) were superfluous, then h2k would be superfluous as well. Then since τ (h2k+1) is superfluous, h2k+1 is not superfluous, and at least one of the other two half-edges of τ (h2k+1) is not superflous, we conclude that τ (h2k+1) has precisely one super- fluous half-edge. Let h2k+2 be the (necessarily unique) non-superfluous half-edge of v different from h2k+1. Lemma A.7. The core is a well-defined leaf-free trivalent graph. Proof. The procedure to generate hn cannot repeat a half-edge and so terminates. If (h0, . . . , hn) is the involution sequence for h0 then (hn, . . . , h0) is the sequence for hn, so ι′ is an involution. Because of this reflection property, for hn to equal h0 would either require some hk to be a leaf or h2k+2 = h2k+1 for some index, both of which are false by construction. Therefore ι′ is fixed-point free. (cid:3) Lemma A.8. Let Γ be a connected trivalent graph with Euler char- acteristic χ. Then the core of Γ is empty if χ ≥ 0 and has Euler characteristic χ otherwise. A graph with m leaves such that every con- nected component has non-positive Euler characteristic must have m superfluous vertices. Proof. If Γ is a tree then every vertex is trivially superfluous. If Γ has Euler characteristic 0 then there is a single cycle in Γ and so every vertex must have a half-edge pointing "away" from the cycle. So every vertex is superfluous. THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 25 Figure 2. A graph Γ, the same graph Γ with superflu- ous vertices and half-edges in thinner gray, the core of Γ, and the insertion forest of Γ. The realization of any trivalent graph of negative Euler characteris- tic contains either an embedded "theta" or an embedded "handcuff". In either case the trivalent vertices of these subspaces must be core. Therefore a graph with negative Euler characteristic contains a core vertex. Now divide the superfluous vertices into two types, according to whether the vertex has half-edges appearing in a sequence to generate the involution ι′ or not. If v is of the first type, then it has precisely two half-edges h1 and h2 in such a sequence. The third half-edge h3 is necessarily superfluous, and as a result is attached to a tree (whose ver- tices are necessarily of the second type). Since there are core vertices, this must exhaust all superfluous vertices by connectedness. 26 G. C. DRUMMOND-COLE AND M. TAVAKOL Then in this case, the subgraph spanned by superfluous vertices is a forest subgraph of Γ, and the procedure to generate the core replaces each tree in the union with a "half-edge gluing". Removing a subtree from a graph increases the Euler characteristic by one. Identifying two leaves into an edge decreases the Euler characteristic by one. Therefore these two operations together preserve the Euler characteristic. (cid:3) Definition A.9. Let v be a superfluous vertex of a trivalent graph Γ which is in a connected component with nonempty core. The insertion forest of v is a graph built as follows. The vertices are of the superflu- ous vertices of Γ. The half-edges are the half-edges of the superfluous vertices of Γ. The target map is induced. For the involution ιif , let h be a half-edge of the insertion forest. If ι(h) is also in the insertion forest, then ιif (h) = ι(h). Otherwise we declare that ιif (h) = h, i.e., that h is a leaf of the insertion forest. We call a connected component of the insertion forest an insertion tree. Lemma A.10. Let T be an insertion tree of a trivalent graph Γ. Then T has precisely two leaves which are not leaves of Γ. Proof. These are the leaves that are paired under ι with half-edges in the core. If there were none this would violate connectedness. If there were only one, h, then ι(h) would be superfluous so τ (ι(h)) would If there were more than two, then be superfluous, a contradiction. superfluity would fail for some vertex of T . (cid:3) Appendix B. Generating functions This appendix is concerned with calculating generating functions for classes of graphs involved in the master series Ω and the key lemma 5.6. We use P(n) to denote the set of unordered partitions of the finite set {1, . . . , n}. That is, an element of P(n) is a set p of pairwise disjoint nonempty subsets (called blocks) {bi} of {1, . . . , n} whose union is {1, . . . , n}. We start with a counting lemma and a standard combinatorial tech- nique that we will use repeatedly. Lemma B.1. The number of isomorphism classes of rooted ordered trivalent trees with n vertices (for n ≥ 1) is (2n)! (n + 1)! 3n. Proof. It is a standard combinatorial identity that the number of planar binary trees with n vertices is each vertex are ordered, we have a choice of which half-edge points n(cid:1) for n ≥ 1. Since half-edges at 1 n+1(cid:0)2n THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 27 toward the root in each of the n vertices, which yields a factor of 3n. The planar structure of the tree induces an order on the other two half-edges of each vertex, and so specifying that this order and the given order must be compatible with respect to some convention uses the planarity to keep track of the distinction between the other two half-edges. The vertices must be ordered which gives another factor of n! and multiplying all of these together yields the result. (cid:3) Sometimes it is useful to include a "trivial tree" for the case n = 0 and sometimes it is not. Our default will follow our conventions and we will specify explicitly if we want to include a trivial tree in this count. The following "compositional formula" will be useful several times. Theorem B.2 (Compositional formula [16, 5.1.4]). Given two expo- nential generating functions F (z) = ∞Xn=1 G(z) = 1 + fn n! zn gn n! zn, ∞Xn=1 then the exponential generating function for the sequence h0 = 1 hn = Xp∈P(n) gpYbi∈p fbi is (G ◦ F )(z). One well-known special case of this theorem occurs when gn = 1 for all n. Corollary B.3 (Exponential compositional formula). Let F (z) be the exponential generating function for fn for n ≥ 1 as above. Then eF (z) is the exponential generating function for h0 = 1 hn = Xp∈P(n)Ybi∈p fbi. We will use this formulation to count possibly disconnected graphs of some sort in terms of a count of connected graphs of the same sort. Now we proceed to the various counting of the graphs involved. 28 G. C. DRUMMOND-COLE AND M. TAVAKOL Leaf-free, negative Euler characteristic. In this section, we prove the following count. Lemma B.4. The exponential generating function Glf − (x) for isomor- phism classes of ordered leaf-free trivalent graphs with n vertices such that each connected component has negative Euler characteristic is (B.5) Glf − (x) = (6n − 1)!! (2n)! x2n. ∞Xn=0 Proof. By Lemma A.2, asking for negative Euler characteristic is no condition on the graph, so this is counting perfect pairings of 3n half- edges. There are no perfect pairings if n is odd and (3n−1)!! if n is even. The denominator comes from making this an exponential generating function. (cid:3) Two-rooted trees. In this section, we prove the following count. Lemma B.6. The two variable generating function Trr(x, y) in Def- inition 5.4 for isomorphism classes of ordered trivalent trees with n vertices, two distinct distinguishable leaves called roots, and m non- root leaves is (B.7) Trr(x, y) = 1 √1 − 12xy − 1. Proof. We use Lemma B.1. A tree with n vertices has n + 2 leaves including the first root. Choosing a second root amounts to n + 1 choices. Then the number of vertices is the same as the number of non-root leaves. We want an exponential generating function, which introduces a factor of 1 n! . We arrive at the following series, as promised: (n + 1) (2n)! n! (n + 1)! (3xy)n = ∞Xn=1 ∞Xn=1(cid:18)2n n(cid:19)(3xy)n. (cid:3) Zero Euler characteristic. Next, we prove the following count. Lemma B.8. The generating function G0(x, y) for the isomorphism classes of ordered trivalent graphs with n and m leaves such that each connected component has Euler characteristic zero is (B.9) G0(x, y) = (1 − 12xy)− 1 4 Remark B.10. Having zero Euler characteristic means that the number of edges, 1 2(3n− m), must be equal to the number of vertices, n, which implies that the two indices n and m must be equal for the number to be non-zero. Then G0(x, y) can be written in terms of the single variable z = xy as an exponential generating function. THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 29 Proof. By Corollary B.3, it suffices to show that Gc 0(z) = − 1 4 log(1 − 12z), where Gc dered trivalent graphs with n vertices and zero Euler characteristic. 0(z) is the exponential generating function for connected or- A graph of Euler characteristic zero has a unique cycle, so we can decompose the count of such graphs by the length of this cycle. Then each vertex in the cycle is part of a unique maximal subtree containing no other vertex of the cycle, and this exhausts the vertices of the graph. This implies that the count of such graphs with n vertices is (B.11) Xp∈P(n) Ybi∈p #(TY (bi))! p − 1! 2 , where TY (n) counts ordered trivalent trees with n vertices including a special vertex and an ordered pair of two distinct leaves of the special vertex. This is the vertex in the unique cycle and the labeling corre- sponds to deciding "which half-edge goes which way" in that unique cycle. The p−1! corresponds to the dihedral symmetry involved in assembling these trees. 2 We will use Theorem B.2. To use the formula directly requires one of the exponential generating functions to have constant term 1 so we will artificially modify the series (B.11) by putting a dummy 1 in when n = 0, instead of a 0. Then Gc 0(z) + 1 is the composition of the exponential generating function (n − 1)! 1 + log(1 − z) 2n! ∞Xn=0 zn = 1 − with the exponential generating function TY (z) =X #TY (n) 1 2 zn n! for TY . Therefore to conclude we should compute TY (z). Counting TY (n) is the same as choosing a vertex and the ordered pair of half-edges -- this is a factor of 6n -- and then assembling the rest of the n−1 vertices into a rooted ordered tree (including the possibility of the trivial tree if n = 1). Then by Lemma B.1, there are (2n−2)! 3n−1 such trees (here this is valid for n ≥ 1), and combining with the 6n we get the exponential generating function which has exponential generating function n! TY (z) = 2 ∞Xn=1 (2n − 2)! (n − 1)!n! (3z)n. 30 G. C. DRUMMOND-COLE AND M. TAVAKOL Taking the derivative of this formal series yields 6√1−12z , so the expo- nential generating function is −√1 − 12z + c for a constant of integra- tion c which is 1 because there is no constant term in the exponential generating function TY (z). Then the composition formula of Theorem B.2 yields Gc 0(z) + 1 = 1 − 1 2 log(1 − (1 − √1 − 12z)) = 1 − 1 2 log(√1 − 12z) as desired. (cid:3) Positive Euler characteristic. In this section, we prove the follow- ing count. Lemma B.12. The two variable generating function Gc morphism classes of ordered trivalent trees with n vertices is +(x, y) for iso- (B.13) Gc +(x, y) = (1 − √1 − 12xy)3(1 + 3√1 − 12xy) 864x2 . Proof. Again we use Lemma B.1. Since ordered trees have only the identity automorphism, this means that we can count the unrooted ordered trivalent trees by dividing by the n + 2 possible choices of root, so that Gc +(x, y) = y2 (2n)! (n + 2)! (3xy)n. ∞Xn=1 It is standard that (2n)! n! ∞Xn=0 (3z)n = 1 √1 − 12z and this is 1 less than the second derivative of x2Gc +(x, y) viewed as a series in the single variable z = xy. Integrating twice with respect to z, we find that x2Gc +(x, y) = 3 2 (1 − 12xy) 108 − (xy)2 2 + xy 6 − 1 108 , where the constants of integration have been chosen to give the correct overall values (i.e., zero) for (xy)0 and (xy)1 on the right side. Rewriting in terms of √1 − 12xy (which will be more convenient later) yields the result. (cid:3) THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 31 Appendix C. Proof of the key lemma In this section we prove Lemma 5.6, which is about a certain "eval- uation" of the master series Ω. For this purpose, we will use the calculations of the constituent gen- erating functions of the master series Ω from Appendix B (i.e., Lem- mas B.4, B.6, B.8, and B.12). These computations in hand, we perform the evaluation of Lemma 5.6 variable by variable, starting with z and u. Notation C.1. We will use √1 − 12xy often, so we use Q as shorthand for it. Lemma C.2. For any n ≥ 0, the "evaluation" of the series in x, y, z, and u ezTr,r(x,y) Gℓf −(xu) obtained by linearly replacing zn3un5 with n3!(cid:0) 3 series in x and y: 2 n5+3n n3 (cid:1) is the following ezTr,r(x,y) = . zj Q−3n −(xQ− 3 Gℓf 2 ) j!(cid:18) 1 (cid:1) for un5zj yields (cid:19)(cid:18) 1 Q − 1(cid:19)j ∞Xj=0 Q − 1(cid:19)j Proof. First, by Lemma B.6, so substituting j!(cid:0) 3 ∞Xj=0(cid:18) 3 2 n5+3n j 2n5 + 3n j 2 n5+3n Q(cid:19) 3 =(cid:18) 1 . (cid:3) Next we turn to w, which is easy. Lemma C.3. The evaluation of the series w = 3v + 6n − 3 is Q Proof. By Lemma B.8, G0(x, y) = Q− 1 2 +3n. 2 , so 3v−3 1 0(x,y) in w, x, and y at ewGc e−w log(G0(x,y)) = G0(x, y)−w = Q w 2 . (cid:3) We can combine these to achieve the following: 32 G. C. DRUMMOND-COLE AND M. TAVAKOL Corollary C.4. Define a series in Ω′ev in x by starting from the fol- lowing series in x, y and v 3v−3 2 Q G+(x, y) and performing the following linear substitution: xn1yn2vn3 7→(0 n2 odd, (n2 − 1)!!xn1(n1)n3 n2 even. Then the simplified master series Ωev of Lemma 5.6 is related to Ω′ev via Ωev(x) = Ω′ev(x) Gℓf −(x) . Proof. By Lemmas C.2 and C.3, the simplified master series Ωev is obtained from 3v−3 2 Q by performing the indicated substitution. In order to compare terms, write G+(x, y)Gℓf −(xQ− 3 2 ) 1 Gℓf −(x) Then the quotient can be written =Xn γnxn. If we write v′ = v − n, then we are replacing v′ in the xn1+n term of the series expansion of the expression 3v′−3 γn 3(v−n)−3 2 Q G+(x, y) xn. Xn γn 2 Q G+(x, y) xn with n1 (and also doing a substitution for y). That's the same as replacing v in the xn1 term of the series expansion of the expression γn 3v−3 2 Q G+(x, y) with n1, doing the substitution for y, and then multiplying by xn at the end. Then the γn and xn terms have become decoupled and the overall effect is to do the substitution on the simpler quotient Q and then divide by Gℓf G+(x,y) (cid:3) 3v−3 2 −(x). THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 33 Now the key lemma can be derived from the following technical re- sult. Lemma C.5. The coefficient of (xy)2n in the expansion of Q is 6n−3 2 (1 + Q)2n+1(2Q + 1)− 2n+1 2 (6n)!n! (3n)!(2n)!(2n)! 2 3(cid:19)n √3(cid:18)1 . Assuming this lemma, we finish the main proof. Proof of Lemma 5.6. By Corollary C.4, it suffices to perform the sub- stitution of that corollary on the series expansion of the expression 3v−3 2 Q G+(x, y) and verify that the result is equal to Gℓf −(x). Lemma B.12 says G+(x, y) = exp(cid:18) (1 − Q)3(1 + 3Q) 864x2 (cid:19) . So the parity of x and y in the series in question will always be the same. Then after substituting for y, there will be no odd powers of x, so we may confine ourselves to the even powers. In fact, the power of y and the power of x differ only in terms of the x2 in the denominator of Gc +(x, y) so if we are interested in the 2n1 power of x in the substitution, it is the x2n1 term of the series ( 6n1−3 Q x 2 ) 1 n! (2n1 + 2n − 1)!!(cid:18)− where Qx = √1 − 12x. But (2n1 + 2n − 1)!! n! xn = ∞Xn=0 ∞Xn=0 (cid:19)n , (1 − Qx)(1 + 3Qx) 864x2 (2n1 − 1)!! (1 − 2x) 2 2n1 +1 so the series of interest is ( 6n1−3 Q x 2 ) (2n1 − 1)!! (cid:16)1 + (1−Qx)(1+3Qx) 432x2 . 2 (cid:17) 2n1+1 By direct manipulation we have the equation 1 + (1 − Qx)3(1 + 3Qx) 432x2 = 4 3 (1 + 2Qx) (1 + Qx)2 . 34 G. C. DRUMMOND-COLE AND M. TAVAKOL Then finally we are interested in the x2n1 coefficient of 6n1 −3 2 Q x 2 (1 + Qx)2n1+1 (1 + 2Qx) 2n1+1 2 . By Lemma C.5, this coefficient is 4(cid:19) 2n1 +1 (2n1 − 1)!!(cid:18)3 4(cid:19) 2n1+1 2 (6n1)!n1! (3n1)!(2n1)!(2n1)! (2n1 − 1)!!(cid:18)3 2 3(cid:19)n1 √3(cid:18)1 (6n1 − 1)!! = (2n1)! −(x), as desired. , (cid:3) which is by Lemma B.4 the x2n1 coefficient of Gℓf It only remains to prove Lemma C.5. Proof of Lemma C.5. Let us begin with an overview. The proof will use a method of Hautus and Klarner [3] for extracting the "diagonal" of an analytic series in two variables. One expositional option would be to build such a two-variable series whose diagonal was precisely the generating function of the numbers we care about. We have chosen instead to build a series whose even diagonal entries are the numbers we care about and whose odd entries are irrelevant for our purposes. This allows us a little more flexibility in the shape of the series we build and we can apply a special case of Hautus and Klarner. The price we pay for this is that after finishing the computation we have an extra step to extract the even degree coefficients. Let us begin. It is convenient to write our series in terms of U = 2 (i.e., the reciprocal of Qx). Consider the following series (1 − 12x)− 1 in x and t: F (x, t) = 3n−3 2 (cid:18)1 + 1 U ∞Xn=0 = U(U + 1)(U + 2)− 1 2 tn 2 1 U U(cid:19)n+1(cid:18) 2 + 1(cid:19)− n+1 ∞Xn=0(cid:18) U + 1 t(cid:19)n U 2√U + 2 Note that the even "diagonal coefficients", i.e., the coefficients of x2nt2n in this series, are indeed the quantities of interest to us. The auxiliary functions W (x) = Y (x) = , U(U + 1) √U + 2 U + 1 U 2√U + 2 = W (x) U 3 are themselves analytic in a neighborhood of x = 0. Then the special case of the result of Hautus and Klarner [3, Section 4] allows us to THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 35 extract the one variable diagonal series associated to F (x, t). That is, if F (x, t) = then we can extract αm,nxmtn, ∞Xm=0 ∞Xn=0 ∞Xn=0 F∆(z) = αn,nzn (the variable z here is unrelated to the variable z used in the master se- ries and other similar series earlier). The recipe of Hautus and Klarner says that F∆(z) is given via a residue computation which simplifies in the case of interest to be where x is defined implicitly by zY (x) = x as an analytic function of z in a neighborhood of the origin. It's convenient for us to rewrite this F∆(z) = F∆(z) = , Y (x) W (x) 1 − zY ′(x)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)z= x (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)z= x W (x)/Y (x) dz dx Y (x) . Now using x = U 2−1 12U 2 , we see that the equality zY (x) = x occurs at (C.6) 12z = U 2 − 1 U 2 U 2√U + 2 U + 1 = (U − 1)√U + 2. and so in our case, we can use the facts that W (x) = U 3Y (x) and that dU dx = 6U 3 along with the chain rule to get F∆(z) = dz U 3 dU · 6U 3(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)z= x = , 1 6 dz dU(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)z= x Y (x) Y (x) which is the formal value of U′(z)/6 where U and z are related by Equation (C.6). Now by trigonometric identities or by solving a cubic, we find that near (z = 0, U = 1) we can give an explicit form for the inverse function: U(z) = 2 cos(cid:18)2 3 cos−1(6z)(cid:19) , where we take the standard branch of the arccosine. 36 G. C. DRUMMOND-COLE AND M. TAVAKOL It is a tedious but elementary verification that both U′(z)/6 and the convergent series (6n)!n! are solutions to the second order linear differential equation ∞Xn=0 (3n)!(2n)!(2n)!(cid:16)z 3(cid:17)2n 20y + 108zy′ + (36z2 − 1)y′′ = 0. Because the coefficients of y and y′′ are even functions of z and the coefficient of y′ is an odd function, U′(−z)/6 is also a solution, so the average 1 12(U′(z) + U′(−z)) and the given series are both solutions with vanishing first derivative at 0. Then they agree up to a scalar multiple, which can be checked by hand to be 2√3 (cid:3) . References 1. A. Beauville, Algebraic cycles on Jacobian varieties, Compositio Mathematica 140 (2004), no. 03, 683 -- 688. 2. C. Faber, A conjectural description of the tautological ring of the moduli space of curves, Moduli of curves and abelian varieties, Aspects Math., E33, Vieweg, Braunschweig, 1999, pp. 109 -- 129. MR 1722541 (2000j:14044) 3. M. L. J. Hautus and D. A. Klarner, The diagonal of a double power series, Duke Math. J. 38 (1971), 229 -- 235. MR 0276641 4. K. Kunnemann, A Lefschetz decomposition for Chow motives of abelian schemes, Inventiones mathematicae 113 (1993), no. 1, 85 -- 102. 5. E. Looijenga, On the tautological ring of Mg, Inventiones mathematicae 121 (1995), no. 1, 411 -- 419. 6. B. Moonen, Relations between tautological cycles on Jacobians, Commentarii Mathematici Helvetici 84 (2009), no. 3, 471 -- 502. 7. B. Moonen and A. Polishchuk, Algebraic cycles on the relative symmetric pow- ers and on the relative Jacobian of a family of curves. II, Journal of the Institute of Mathematics of Jussieu 9 (2010), no. 4, 799 -- 846. 8. D. Mumford, Towards an enumerative geometry of the moduli space of curves, Arithmetic and geometry 2 (1983), no. 36, 271. 9. R. Pandharipande and A. Pixton, Relations in the tautological ring of the moduli space of curves, arXiv preprint arXiv:1301.4561 (2013). 10. D. Petersen, Poincar´e duality of wonderful compactifications and tautological rings, Int. Math. Res. Not. IMRN (2016), no. 17, 5187 -- 5201. MR 3556436 11. D. Petersen, M. Tavakol, and Q. Yin, Tautological classes with twisted coeffi- cients, arXiv preprint arXiv:1705.08875 (2017). 12. A. Pixton, Conjectural relations in the tautological ring of M g,n, arXiv preprint arXiv:1207.1918 (2012). 13. A. Polishchuk, Universal algebraic equivalences between tautological cycles on Jacobians of curves, Mathematische Zeitschrift 251 (2005), no. 4, 875 -- 897. 14. , Lie symmetries of the Chow group of a Jacobian and the tautological subring, Journal of Algebraic Geometry 16 (2007), no. 3, 459 -- 476. THE POLISHCHUK DIFFERENTIAL OPERATOR VIA SURFACES 37 15. , Algebraic cycles on the relative symmetric powers and on the relative Jacobian of a family of curves. I, Selecta Mathematica 13 (2008), no. 3, 531 -- 569. 16. R. Stanley, Enumerative combinatorics. vol. 2, volume 62 of Cambridge Studies in Advanced Mathematics, 1999. 17. M. Tavakol, The moduli space of curves and its invariants, Proceedings of 5th Frontiers in Mathematical Sciences, IPM Tehran, to appear. 18. , The connection between R∗(C n g ) and R∗(M rt g,n), Journal of Pure and Applied Algebra 222 (2018), no. 6, 1306 -- 1315. 19. Q. Yin, Tautological cycles on curves and Jacobians, Ph.D. thesis, Radboud Universiteit Nijmegen, 2013. Center for Geometry and Physics, Institute for Basic Science, Po- hang, Republic of Korea 37673 E-mail address: [email protected] School of mathematics and statistics, University of Melbourne, VIC 3010, Australia E-mail address: [email protected]
1509.03216
1
1509
2015-09-10T16:38:51
T-structures on elliptic fibrations
[ "math.AG" ]
We consider t-structures that naturally arise on elliptic fibrations. By filtering the category of coherent sheaves on an elliptic fibration using the torsion pairs corresponding to these t-structures, we prove results describing equivalences of t-structures under Fourier-Mukai transforms.
math.AG
math
T-STRUCTURES ON ELLIPTIC FIBRATIONS JASON LO Abstract. We consider t-structures that naturally arise on elliptic fibrations. By filtering the category of coherent sheaves on an elliptic fibration using the torsion pairs corresponding to these t-structures, we prove results describing equivalences of t-structures under Fourier-Mukai transforms. Contents Introduction 1. 1.1. Main Results 2. Notation 3. General constructions on fibrations 3.1. Base change formulas 3.2. Torsion pairs induced from fibers 3.3. More torsion classes 4. Elliptic fibrations 4.1. 4.2. References t-structures on elliptic surfaces t-structures on elliptic threefolds 1 3 5 7 7 9 11 14 15 22 28 1. Introduction This is the first in a series of articles on elliptic fibrations. In this article, we focus on t-structures that can be constructed using the geometry of an elliptic fibration, and using a relative Fourier-Mukai transform from the derived category of coherent sheaves on the elliptic fibration. In later articles, we will study the relations between these t-structures and those that appear in the study of Bridgeland stability conditions. We will also study the different notions of stabilities associated with them. The study of t-structures comes into various problems of active interest in alge- braic geometry: for instance, explicitly in the study of Bridgeland stability condi- tions, and implicitly in the construction of stable sheaves. In the study of Bridgeland stability conditions (see [Bri2, Bri3, ABL, MM, Mac, Sch, MP1, MP2, BMS]), t-structures are part of the definition for a stability con- dition. A Bridgeland stability condition on a smooth projective variety X is a pair (Z, A) where A is the heart of a t-structure on D(X) := Db(Coh(X)), the bounded derived category of coherent sheaves on X, and Z a group homomorphism from the 2010 Mathematics Subject Classification. Primary 14D06; Secondary: 14F05, 14J27, 14J60. Key words and phrases. t-structure, torsion pair, elliptic fibration, cohomology, Fourier-Mukai transform. 1 2 JASON LO Grothendieck group K(X) to C, with A, Z satisfying certain properties. There- fore, undertanding t-structures on D(X) has implications on the types of stability conditions that can arise, as well as on which objects in D(X) can arise as stable objects. In the construction of stable sheaves on varieties, t-structures come into play in the method of spectral construction (see [FMW] and, for instance, [BBR, CDFMR, RP, ARG]), albeit implicitly. For instance, when X admits the structure of an elliptic fibration π : X → S, if we have a 'dual' fibration bπ : Y → S where Y is another elliptic fibration with a Fourier-Mukai transform Φ : D(Y ) → D(X), then the spectral construction produces stable sheaves on X of the form Φ(F ) where F is a coherent sheaf on Y supported in codimension 1. In our notation in Section 3.2 below, we can view the sheaf Φ(F ) as lying in the category Φ({Coh≤0(Ys)}↑), where {Coh≤0(Ys)}↑ is contained in the heart of a t-structure that is a tilt of the standard t-structure on D(Y ). In this paper, we consider various t-structures on D(X) where X is a smooth projective variety that comes with an elliptic fibration π : X → S and a Fourier- Mukai partner bπ : Y → S (see Section 2 for the precise assumptions). These t-structures arise from the geometry of X itself, the geometry of the fibration π, and also from the Fourier-Mukai transform between D(X) and D(Y ). Each of these t-structures corresponds to a torsion pair in Coh(X) or Coh(Y ), which in turn is determined by the torsion class in the torsion pair. When X, Y are elliptic threefolds, the torsion classes in Coh(X) from which we build all the other torsion classes in this article (by taking intersections and extensions) can be summarised in the following diagram, where each arrow denotes an inclusion of categories: W0,X {Coh≤0(Xs)}↑ TX . / W ′ 0,X / Coh≤2(X) 8q q q q q q q q q q Coh≤1(X) 8♣ ♣ / Coh(π)≤1 ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ Coh≤0(X) BX,∗ / Coh(π)≤0 Here, • Coh≤d(X) is the category of coherent sheaves E on X supported in dimen- sion at most d; • Coh(π)≤d is the category of coherent sheaves E on X such that the dimen- sion of π(supp(E)) is at most d; / O O O O / O O 8 / O O O O / / 8 / O O T-STRUCTURES ON ELLIPTIC FIBRATIONS 3 • {Coh≤0(X)}↑ is the category of coherent sheaves E on X such that the restriction Es to the fiber over any closed point s ∈ S is supported in dimension 0; • W0,X is the category of coherent sheaves E on X such that Ψ(E) is isomor- phic to a coherent sheaf on Y sitting at degree 0 (i.e. the category of the 'Ψ- WIT0 sheaves'), with Ψ being the Fourier-Mukai trasform D(X) → D(Y ) coming from the construction of Y ; • W ′ 0,X is the category of coherent sheaves E on X such that, for a general closed point s ∈ S, the restriction Es to the fiber over s is taken by Ψs (the base change of Ψ under the closed immersion {s} ֒→ S) to a sheaf sitting at degree 0; • TX is the extension closure of W ′ • BX,∗ is the extension closure of Coh≤0(X) and a category of fiber sheaves 0,X and Φ({Coh≤0(Ys)}↑). - see (3.9). Using these subcategories, we filter the category of coherent sheaves on X or Y into smaller pieces with distinct geometric properties, the behaviors of which under the Fourier-Mukai transform Ψ : D(X) → D(Y ) are tractable. As a consequence, we obtain results that explicitly describe how t-structures on D(X) and D(Y ) are equivalent, or related by tilts, under a relative Fourier-Mukai transform Ψ : D(X) → D(Y ). Another underlying motivation for studying the t-structures that appear in this paper is, to some extent, to mitigate the problem that stability is not always well- behaved under base change. Given a flat morphism π : X → S of smooth projective varieties and a torsion-free coherent sheaf E on X, there are various results on how the stability (in the sense of slope stability or Gieseker stability) of E is related to the stability of the restriction of E to the generic fiber of π - see [Bri1, Proposition 7.1], [BM, Lemma 2.1], and [ARG, Proposition 3.3]. In general, however, there seems to be few results relating the stability of E and the stability of its restriction to a special fiber. To this end, we propose to replace the notion of stability for coherent sheaves on X by the notion of being Ψ-WIT0. As observed in Lemma 3.6, being WIT0 is compatible with base change when π is of relative dimension 1, and so, compared to slope stability, it gives us a better behaved notion under the Fourer-Mukai transform Ψ. More precisely, when π is an elliptic fibration and F is a sheaf supported on a fiber of π, that F is WIT0 is equivalent to all its Harder-Narasimhan (HN) factors having strictly positive slopes [BBR, Corollary 3.29, Proposition 6.38]. Therefore, when a sheaf E on X is Ψ-WIT0, we can restrict it to a special fiber and still understand the HN filtration after restriction. As a result, even though special cases of the torsion classes Coh≤d(X) and Coh(π)≤d have appeared in the author's earlier article [Lo1] (as the torsion classes TX and BX), by taking the above perspective in this article, we obtain finer results on the behavior of these torsion classes under Fourier-Mukai transforms. 1.1. Main Results. Let us write CX to denote the heart of the t-structure on D(X) given by D≤0 D≥0 C := {E ∈ D(X) : H 0(E) ∈ TX , H i(E) = 0 ∀ i > 0}, C := {E ∈ D(X) : H −1(E) ∈ T◦ X , H i(E) = 0 ∀ i < −1}, 4 JASON LO where T◦ and write DY to denote the heart of the t-structure on D(Y ) given by X is the torsion-free class in Coh(X) corresponding to the torsion class TX, D≤0 D≥0 D := {E ∈ D(Y ) : H 0(E) ∈ BY,∗, H i(E) = 0 ∀ i > 0}, D := {E ∈ D(Y ) : H −1(E) ∈ B◦ Y,∗, H i(E) = 0 ∀ i < −1}, where B◦ BY,∗. That is, Y,∗ is the torsion-free class in Coh(Y ) corresponding to the torsion class CX = hT◦ X [1], TXi and DY = hB◦ Y,∗[1], BY,∗i. Let us also write Λ to denote the composition of the Fourier-Mukai functor Ψ(−) with the derived dual functor −∨, i.e. Λ(−) = (Ψ(−))∨. Then we have: Theorem 4.12. When X is a smooth elliptic surface, the functor Λ induces an equivalence between the t-structure (D≤0 C ) on D(X), and the t-structure (D≤0 D ) on D(Y ). Equivalently, Λ induces an equivalence of hearts D , D≥0 C , D≥0 (1.1) CX ∼→ DY [−1]. Theorem 4.12 for elliptic surfaces can be considered as a special case of a re- sult of Yoshioka's [Yos3, Proposition 3.3.5] - see the end of Section 4.1, including Lemma 4.13, for a precise explanation of this. Yoshioka's result is more general in the case of elliptic surfaces, as it is stated for tilts of categories of perverse sheaves, as opposed to tilts of categories of coherent sheaves. The result [Yos3, Proposi- tion 3.3.5] was obtained by Yoshioka as part of an argument towards showing an isomorphism between moduli spaces of twisted stable sheaves on elliptic fibrations that are Fourier-Mukai partners (see [Yos3, Proposition 3.4.4] or [Yos1, Theorem 3.15]). When X and Y are elliptic threefolds, the functor Λ no longer induces an equiv- D ). Instead, the hearts C ) and (D≤0 alence between the t-structures (D≤0 of these two t-structures differ by a tilt (up to a shift): D , D≥0 C , D≥0 Theorem 4.26. Suppose X is a smooth elliptic threefold. Then the heart Λ(CX ) differs from the heart DY [−2] by one tilt. We prove Theorem 4.26 by explicitly studying the effects of Λ on various sub- categories of Coh(X), which arise from a nested sequence of Serre subcategories of Coh(X) - see Theorem 4.24 and its proof. The implications of our results on moduli spaces will be addressed in a later paper. The relationships between the t-structures considered in this paper, as well as other t-structures that come up in the study of Bridgeland stability conditions, will also be explored in a later article. Some of the t-structures considered in this article were already considered in [Lo2] and [CL], which grew out of an attempt to understand the results in [Bri1] and [BM]. Acknowledgements. The author would like to thank the National Center for Theoretical Sciences in Taipei, as well as the Max Planck Institute for Mathematics in Bonn, for their hospitality during the author's stay from March through May 2014, when most of this project took place. He would like to thank Ziyu Zhang for many invaluable comments on various versions of this article, and Tom Nevins and Sheldon Katz for comments on a later version. He would also like to thank the T-STRUCTURES ON ELLIPTIC FIBRATIONS 5 referee for a careful reading of the manuscript, and many helpful comments leading to improvement on the exposition in this article. 2. Notation The setup considered in the rest of this article (except for Section 3.1, where the setup is slightly more general) is as follows: we will assume that we have a pair of morphisms of smooth projective varieties π : X → S and π : Y → S that satisfies the following conditions: (i) There is a pair of relative integral functors that are quasi-inverse to each other, up to a shift (so that they are necessarily equivalences): Ψ : Db(X) ∼→ Db(Y ) and Φ : Db(Y ) ∼→ Db(X). (ii) The morphisms π, π are both flat. Note that, by property (i) and our assumption that X, Y, S are all projective, the kernels of the relative integral functors Ψ and Φ both have finite homological dimen- sions, as complexes of OX -modules or OY -modules, respectively [RMS, Proposition 2.10]. This ensures, that given any morphism of varieties S′ → S, the correspond- ing base changes ΨS ′ : D(XS ′) → D(YS ′ ) and ΦS ′ : D(YS ′ ) → D(XS ′ ) still take the bounded derived categories Db(XS ′ ) and Db(YS ′ ) into each other [BBR, Section 6.1.1]. (1) For any variety W , we will write D(W ) = Db(Coh(W )) to denote the bounded derived category of coherent sheaves on W unless otherwise stated. (2) For any closed point s ∈ S, we will write ιs (resp. js) to denote the closed Let π,bπ be as above. The following notations will be used throughout this article: immersion of the fiber Xs → X (resp. Ys → Y ) of π (resp.bπ) over s. When E is a coherent sheaf on X (resp. on Y ), we will write Es to denote the restriction ι∗ s to denote the derived restriction Lι∗ s E). sE (resp. the derived restriction Lj∗ sE (resp. the restriction j∗ s E), and write EL (3) We will write BX to denote the category of coherent sheaves E on X such that Es is zero for a general closed point s ∈ S. We similarly define BY . (4) For any Abelian category A and any E ∈ D(A), we will write H i(E) to denote the degree-i cohomology of E with respect to the standard t- structure on D(A). When B is the heart of a t-structure on D(A), for any E ∈ D(A), we will write H i B(E) to denote the degree-i cohomology of E with respect to the t-structure with heart B. We will also define, for any integers j, k, D[j,k] B := D[j,k] B (A) := {E ∈ D(A) : H i B(E) = 0 for all i /∈ [j, k]}. (5) For each integer i, we will write Wi,X to denote the category of coherent sheaves E on X such that (2.1) Ψ(E) ∼= bE[−i] for some bE ∈ Coh(Y ), and refer to objects in Wi,X as Ψ-WITi sheaves on X. For a Ψ-WITi sheaf E on X, we will refer to bE in (2.1) as the transform of E. We similarly define Wi,Y and Φ-WITi sheaves for any integer i, with Ψ replaced by Φ and X replaced by Y in the definitions above. 6 JASON LO (6) (WITi sheaves) For any integer i, we will write Ψi(−) to denote the compos- ite functor H i(Ψ(−)), where H i is the cohomology functor with respect to the standard t-structure on D(Y ). Similarly, we write Φi(−) := H i(Φ(−)). (7) If a coherent sheaf E on X is supported on a finite number of fibers of π, then we will refer to it as a fiber sheaf. Similarly for coherent sheaves on Y . (8) (Torsion pairs) Given an Abelian category A, recall that a torsion pair (T , F ) in A is a pair of full subcategories of A satisfying the following two conditions: (a) Every E ∈ A fits in a short exact sequence in A 0 → T → E → F → 0 where T ∈ T and F ∈ F ; (b) HomA(T, F ) = 0 for any T ∈ T , F ∈ F . Whenever we have a torsion pair (T , F ) in an Abelian category A, we will refer to T as the torsion class of the torsion pair, and F as the torsion-free class of the torsion pair. We will say that a subcategory C of an Abelian category A is a torsion class if it is the torsion class of a torsion pair in A. (9) Whenever we have a proper morphism f : V → W of noetherian schemes, we will write Coh(f )b to denote the category of coherent sheaves E on V such that dim (f (supp(E))) = b for any b ≥ 0. We will also write Coh(f )≤b to denote the category of coherent sheaves E on X with dim (f (supp(E))) ≤ b. For any b ≥ 0, the category Coh(f )≤b is a Serre subcategory (i.e. closed under subobjects, extensions and quotients in Coh(V )), and in particular, is the torsion class of a torsion pair in Coh(V ). The category Coh(f )0 = Coh(f )≤0 is precisely the category of fiber sheaves on V . Also, when f is flat of relative dimension 1 and W is irreducible of dimension d, we have Coh(f )≤d = BV . (10) For any variety W , when C1, · · · , Cn are subcategories of D(W ), we will write hC1, · · · , Cni to denote the extension closure generated by the Ci in D(W ). (11) For any noetherian scheme W and any integer d ≥ 0, we will write Coh≤d(W ) to denote the category of coherent sheaves on W supported in dimension at most d, write Coh=d(W ) to denote the category of pure d-dimensional coherent sheaves, and write Coh≥d(W ) to denote the category of coherent sheaves with no nonzero subsheaves supported in dimension d − 1 or lower. (12) When W is a smooth projective variety, we sometimes write TW to denote the category of coherent sheaves on W that are torsion, and write FW to denote the category of coherent sheaves on W that are torsion-free. (13) For a fixed variety W and a full subcategory C of Coh(W ), we will define C◦ := {E ∈ Coh(W ) : HomCoh(W )(C, E) = 0 for all C ∈ C}. (14) In a noetherian Abelian category A (such as Coh(W ) for a noetherian scheme W ), whenever we have a full subcategory C ⊆ A that is closed under extensions and quotients in A, we have a torsion pair (C, C◦) in A by [Pol, Lemma 1.1.3]. (15) Whenever we have a torsion pair (T , F ) in an Abelian category A, there is a corresponding t-structure (D≤0, D≥0) on the derived category D(A) T-STRUCTURES ON ELLIPTIC FIBRATIONS 7 given by D≤0 := {E ∈ D(A) : H 0(E) ∈ T , H i(E) = 0 for all i > 0}, D≥0 := {E ∈ D(A) : H −1(E) ∈ F , H i(E) = 0 for all i < −1 }. (16) (Elliptic fibrations) By an elliptic fibration, we will mean a proper flat morphism π : X → S such that the generic fiber of π is a smooth elliptic curve. (By the flatness of π, it follows that all fibers of π are 1-dimensional.) We will refer to π (or X) as an elliptic surface when dim X = 2, and as an elliptic threefold when dim X = 3. In addition, we will saybπ is a dual elliptic fibration, or say π andbπ are a pair of dual elliptic fibrations, when π,bπ are elliptic fibrations of the same dimension satisfying conditions (i) and (ii), where the kernels of Ψ and Φ in condition (i) are both coherent sheaves sitting at degree 0, flat over both X and Y , and we have ΦΨ = idD(X)[−1], ΨΦ = idD(Y )[−1]. Note that, since Ψ and Φ are assumed to be relative integral functors, all 0-dimensional sheaves on X are Ψ-WIT0, and are taken to pure 1- dimensional fiber sheaves on Y which are Φ-WIT1. Example 2.1. The prototypical examples of dual elliptic fibrations π : X → S andbπ : Y → S satisfying our definition above include: • Elliptic surfaces π : X → S considered by Bridgeland in [Bri1], or elliptic threefolds π : X → S considered by Bridgeland-Maciocia in [BM]. In both cases, the fibration bπ : Y → S is constructed as a relative moduli space of stable sheaves on the fibers of π, and the singular fibers of π are not necessarily irreducible. If P denotes the universal sheaf on Y × X for the above moduli problem, then the relative integral functor Ψ : D(X) → D(Y ) with kernel P is a Fourier-Mukai transform. • Weierstrass fibrations π : X → S (which are elliptic fibrations) in the sense of [BBR, Section 6.2], where all the fibers are Gorenstein and geometrically integral. In this case, Y can be taken as the Altman-Kleiman compactifi- cation of the relative Jacobian of X, and Ψ : D(X) → D(Y ) taken to be the relative Fourier-Mukai transform with the relative Poincar´e sheaf as the kernel. In both cases, a quasi-inverse Φ : D(Y ) → D(X) can always be constructed making Φ are both coherent sheaves sitting at degree 0, flat over both factors of X × Y . π,bπ a pair of dual fibrations in the sense above. In particular, the kernels of Ψ and 3.1. Base change formulas. Suppose π : X → S and bπ : Y → S are a pair of proper morphisms of varieties satisfying properties (i) and (ii) as in the beginning of Section 2. Then we have the following base change formula [BBR, (6.3)]: 3. General constructions on fibrations (3.1) js∗Ψs(E) ∼= Ψ(ιs∗E) for all E ∈ D(Xs). Note that, this is where condition (ii) comes in, since (3.1) depends on the morphism π being flat. Assuming additionally that the kernel for the relative integral functor Ψ (resp. Φ) has finite Tor-dimension as a complex of OX -modules (resp. OY -modules), we have the following well-known observation as a consequence of the base change (3.1); we omit its proof: 8 JASON LO Lemma 3.1. For every closed point s ∈ S, the induced integral functors Ψs : D(Xs) → D(Ys) and Φs : D(Ys) → D(Xs) are equivalences. The following is a second base change formula useful to us, which depends on π being flat [BBR, (6.2)]: with ιs, js as above, for any E ∈ D(X) we have (3.2) Lj∗ s Ψ(E) ∼= Ψs(Lι∗ sE) This leads to the following observation that we will use frequently: Lemma 3.2. For any E ∈ D(X), we have π(supp(E)) = π(supp(ΨE)). Proof. Take any s ∈ S \ π(supp(E)). Then 0 = EL (ΨE)L π(supp(ΨE)) ⊆ π(supp(E)). By symmetry, we have equality. s ) ∼= s by the base change (3.2), i.e. s ∈ S \ π(supp(ΨE)). In other words, we have (cid:3) s , and we have 0 = Ψs(EL An immediate consequence of the base change formula (3.1) is the following: Lemma 3.3. If l, m are integers such that Ψi(E) = 0 for all i /∈ [l, m] and for all E ∈ Coh(X), then we have Ψi s(F ) = 0 for any closed point s ∈ S, i /∈ [l, m] and any F ∈ Coh(Xs). As a result, if Ψ is a relative integral functor that takes coherent sheaves on X to n-term complexes, then for any closed point s ∈ S, the base change Ψs also takes coherent sheaves on the fiber Xs to n-term complexes. Remark 3.4. From [BBR, Corollary 6.3], we know that, if n := p+m0 where p is the dimension of the fibers of π, and m0 is the largest index m such that H m(K) = 0, where K ∈ D(X ×S Y ) is the kernel of the relative integral functor Ψ, then we have, for any E ∈ Coh(X), the base change (3.3) Ψn(E)s ∼= Ψn s (Es) for any closed point s ∈ S. Given Lemma 3.3, we can think of the base change (3.3) as saying: for a coherent sheaf E on X, 'the right-most cohomology of Ψ(E) vanishes if and only if the same holds on each fiber'. Borrowing notation from [Lo2], we define, for any base change morphism S′ → S, the subcategory of Coh(XS ′) Bi,XS′ := {E ∈ Coh(XS ′ ) : Ψi S ′(E) = 0} for any integer i. We similarly define Bi,YS′ (using the vanishing of Φi S ′ ) for any morphism S′ → S. The interpretation of (3.3) at the end of Remark 3.4 can now be stated precisely as follows: Lemma 3.5. Let n be as in Remark 3.4. Then for any E ∈ Coh(X), we have E ∈ Bn,X if and only if Es ∈ Bn,Xs for every closed point s ∈ S. Proof. From Lemma 3.1, we know Ψs is an equivalence. The claim then follows from (3.3). (cid:3) When the morphism π has relative dimension 1 and the kernel of Ψ is a sheaf sitting at degree 0, we have n = 1 where n is as in Remark 3.4. It then follows that B1,X = W0,X (and similarly for Y ). In other words, we have the following interpretation of WIT0 sheaves on fibrations of relative dimension 1: T-STRUCTURES ON ELLIPTIC FIBRATIONS 9 Lemma 3.6. Suppose π has relative dimension 1, and the kernel of the integral functor Ψ is a sheaf (sitting at degree 0). Then for any E ∈ Coh(X), we have that E is Ψ-WIT0 if and only if Es is Ψs-WIT0 for every closed point s ∈ S. Proof. This follows from Lemma 3.3, together with Lemma 3.5 with n = 1. (cid:3) Remark 3.7. Though innocuous-looking, Lemma 3.6 is a key lemma in this article. On an elliptic fibration π : X → S, the stability of a sheaf F on X is related to the stability of the restriction of F to the generic fiber of π - but this relation often depends on the Chern classes of F (see [Bri1, Section 7.1] or [BM, Lemma 2.1]). By replacing stability with WITi properties, we obtain a framework that is more compatible with base change. For a fiber sheaf on an elliptic fibration that possesses a dual fibration, being WITiis inherently related to the structure of its HN filtration with respect to slope stability on the fibers (see [BBR, Corollary 3.29]). 3.2. Torsion pairs induced from fibers. Given any morphism of algebraic va- rieties π : X → S, we describe here two recipes for constructing torsion pairs: one restricts a torsion pair on X to torsion pairs on the fibers of π, while the other gives a torsion pair on X induced from torsion pairs on the fibers of π. Take any subcategory T of Coh(X), and fix any closed point s ∈ S. Let ιs denote the inclusion of the fiber Xs ֒→ X. Consider the following two subcategories of Coh(Xs): T s := {F ∈ Coh(Xs) : F ∼= Es for some E ∈ T }; T ′ := {F ∈ Coh(Xs) : there exists E ։ ιs∗F in Coh(X) for some E ∈ T }. The inclusion T s ⊆ T ′ is clear. When T is closed under taking quotients in Coh(X), we also have the inclusion T ′ ⊆ T s: for any F ∈ T ′, suppose E is an object in Coh(X) such that we have a surjection E ։ ιs∗F in Coh(X). Then ιs∗F lies in T , and we have F ∼= (ιs∗F )s. In particular, when T is the torsion class of a torsion pair in Coh(X), the two subcategories T s and T ′ coincide. Lemma 3.8. Let T be a torsion class in Coh(X). Then, for each closed point s ∈ S, the category T s is a torsion class in Coh(Xs). Proof. To show that T s is a torsion class, it suffices to check that it is closed under quotients and extensions in Coh(Xs). That T s is closed under quotients is clear from the description T s = T ′ above. Now, suppose we have F1, F2 ∈ T s and Fi ∼= Eis for some Ei ∈ T for i = 1, 2. Consider any extension in Coh(Xs) 0 → F1 → F → F2 → 0, which pushforwards to a short exact sequence in Coh(X) 0 → ιs∗F1 → ιs∗F → ιs∗F2 → 0. For each i, we have ιs∗Fi ∈ T , and so ιs∗F also lies in T . Then F ∈ T s since F ∼= (ιs∗F )s. (cid:3) Definition 3.9. Suppose that, for each closed point s ∈ S, we are given a subcat- egory Ts of Coh(Xs). Then we set (3.4) {Ts}↑ := {E ∈ Coh(X) : Es ∈ Ts for all closed points s ∈ S}. 10 JASON LO Lemma 3.10. Suppose that, for each closed point s ∈ S, we have a torsion class Ts in Coh(Xs). Then the category {Ts}↑ is the torsion class of a torsion pair in Coh(X). Proof. Let us write T to denote {Ts}↑. It suffices to check that T is closed under quotients and extensions in Coh(X). That T is closed under quotients is clear. Now, take any E1, E2 ∈ T and consider the extension 0 → E1 → E → E2 → 0 in Coh(X). Fixing any closed point s ∈ S and restricting the above short exact sequence to Xs, we get the exact sequence E1s α→ Es → E2s → 0 in Coh(Xs). Since Ts is closed under quotients, the image of α lies in Ts. Then, because Ts is closed under extensions, we have Es ∈ Ts. (cid:3) Remark 3.11. Given the constructions in Lemmas 3.8 and 3.10, it is natural to ask: are the two constructions described mutually inverse? In other words: (a) Given a torsion class T on X, do we have {T s}↑ = T ? (b) Given a torsion class Ts in Coh(Xs) for each closed point s ∈ S, do we have ({Ts}↑)s = Ts for each closed point s ∈ S? In Lemma 3.12 below, we show that the answer to the question (b) is 'yes'. On the other hand, even though we always have the inclusion T ⊆ {T s}↑, without further assumptions on the varieties X, S or the morphism π, the answer to the question (a) is a priori a 'no'. Lemma 3.12. Suppose that, for each closed point s ∈ S, we have a torsion class Ts in Coh(Xs). Then ({Ts}↑)s = Ts for each closed point s ∈ S. Proof. For any closed point s ∈ S and F ∈ Ts, we have ιs∗F ∈ {Ts}↑. Hence F ∼= s(ιs∗F ) lies in ({Ts}↑)s. The other inclusion follows directly from the definitions. ι∗ (cid:3) Example 3.13. Let π : X → S and π : Y → S be a pair of proper morphisms satisfying conditions (i) and (ii) laid out in the beginning of Section 2, and let n be as in Remark 3.4. Then we have: (a) Bn,X s = Bn,Xs for any closed point s ∈ S, and (b) {Bn,Xs}↑ = {Bn,Xs}↑ = Bn,X. To see why (a) holds, fix any closed point s ∈ S. That Bn,Xs ⊆ Bn,Xs follows from (3.3). To show the other inclusion, take any F ∈ Bn,Xs. By (3.1), we have ιs∗F ∈ Bn,X, and so F ∼= ι∗ sιs∗F ∈ Bn,Xs, giving us (a). In part (b), the first equality follows from part (a), while the second equality follows from Lemma 3.5. Therefore, T = Bn,X is an example of a torsion class in Coh(X) for which the answer to question (a) in Remark 3.11 is 'yes'. Remark 3.14. Suppose π,bπ satisfy conditions (i) and (ii) in the beginning of Section 2, that they both have relative dimension 1, and that the kernels of Ψ and Φ are both sheaves sitting at degree 0. Then by Lemma 3.10, the category {Coh≤0(Ys)}↑ T-STRUCTURES ON ELLIPTIC FIBRATIONS 11 is a torsion class in Coh(Y ), and by Lemma 3.6, every E ∈ {Coh≤0(Ys)}↑ is Φ- WIT0. As a result, the category Φ({Coh≤0(Ys)}↑) (which will be used frequently later on) is contained in W1,X . We briefly return to the question (a) in Remark 3.11. Let us write H i to denote the i-th cohomology functor with respect to the standard t-structure on either D(X) or D(Xs), for any closed point s ∈ S. Given a torsion class T in Coh(X), let us also write Hi to denote the i-th cohomology functor with respect to the t-structure on D(X) with heart hT ◦[1], T i, or the i-th cohomology functor with respect to the t-structure on D(Xs) with heart h(T s)◦[1], T si, for any closed point s ∈ S. Then, for any coherent sheaf E on X, we have E ∈ T if and only if Hi(E) = 0 for all i 6= 0, and for any closed point s ∈ S E ∈ {T s}↑ if and only if Hi(Es) = 0 for all i 6= 0. The condition that {T s}↑ ⊆ T , which is equivalent to having a 'yes' to question (a) in Remark 3.11, is now equivalent to the condition For any coherent sheaf E ∈ Coh(X), if Hi(Es) = 0 for all i 6= 0 and all closed points s ∈ S, then Hi(E) = 0, which can be thought of as a 'Nakayama's Lemma-type' statement. The following observation on 1-dimensional closed subschemes of the total space of a fibration will be used from time to time: Lemma 3.15. Let π : X → S be a proper morphism of varieties of relative dimen- sion 1. Let Z be any irreducible, 1-dimensional closed subscheme of X. Then Z is either of the following two types: (i) Z is contained in a fiber of π; (ii) for any s ∈ S, the intersection Z ∩ π−1(s) is 0-dimensional if nonempty. Proof. Consider the locus D := {s ∈ S : Z ∩ π−1(s) is 1-dimensional}. If D is empty, then Z is of type (ii). Therefore, let us suppose D is nonempty. Then π−1(D) is a closed subset of X by semicontinuity. Note that, the dimension of D must be exactly zero, or else Z would have dimension at least 2, a contradiction. That is, D is a finite number of closed points. Now, the intersection π−1(D)∩Z is a 1-dimensional closed subset of Z. By the irreducibility of Z, we have π−1(D) ∩ Z = Z, and D consists of a single point, i.e. Z is of type (i). (cid:3) 3.3. More torsion classes. In this section, we introduce a few more torsion classes in Coh(X) that depend on the geometry of the fibration. Suppose π : X → S and π : Y → S are a pair of dual elliptic fibrations. We define W ′ W ′ 0,X := {E ∈ Coh(X) : Es is Ψs-WIT0 for a general closed point s ∈ S}, 1,X := {E ∈ Coh(X) : Es is Ψs-WIT1 for a general closed point s ∈ S}, TX := hW ′ 0,X , (Φ({Coh≤0(Ys)}↑))i. (3.5) Note that, by Lemma 3.3, for any closed point s ∈ S and any coherent sheaf F on Xs, the functor Ψs takes F to a complex of length at most 2, sitting at degrees 0 and 1. Also, we have defined W ′ 1,X so that they contain sheaves that restrict 0,X , W ′ 12 JASON LO to zero on a general fiber of π. That is, we have BX ⊆ W ′ In addition, we have TX ⊆ W ′ 0,X contains all the torsion sheaves on X - this is because for a torsion sheaf T on X and a general closed point s ∈ S, the restriction T s must be 0-dimensional, which is Ψs-WIT0. i,X for i = 0, 1. 0,X , i.e. W ′ We can similarly define W ′ 0,Y , W ′ 1,Y and TY , with X replaced with Y and Ψ replaced with Φ. Lemma 3.16. We have (i) W0,X ⊆ W ′ (ii) T◦ 0,X ⊆ TX ; X ⊆ (W0,X )◦ = W1,X ⊆ W ′ 1,X . Proof. Part (i) follows immediately from Lemma 3.6 and the definition of TX . In part (ii), the first inclusion follows from part (i), while the second inclusion follows from (W0,X , W1,X ) being a torsion pair in Coh(X) (see, for instance, [BM, Lemma 9.2]). To show the last inclusion of part (ii), take any E ∈ W1,X . By generic flatness, there exists a dense open subscheme U ⊆ S such that both EU and (bE)U are flat. By base change [BBR, Proposition 6.1], the restriction EU is ΨU -WIT1. Then by [BBR, Corollary 6.3(iii)], we have that Es is Ψs-WIT1 for every closed point s ∈ U , i.e. E ∈ W ′ (cid:3) 1,X . Remark 3.17. Since (W0,X , W1,X ) is a torsion pair in Coh(X), every coherent sheaf E on X fits in a short exact sequence in Coh(X) (3.6) 0 → E0 → E → E1 → 0 where E0 ∈ W0,X and E1 ∈ W1,X . By Lemma 3.16, we can also regard E0, E1 as objects in W ′ 1,X , respectively. 0,X , W ′ Lemma 3.18. For any E ∈ Coh(X), let E0, E1 be as in (3.6). Then: (i) If E ∈ W ′ (ii) W ′ 1,X ∩ B◦ i,X (where i = 0 or 1), then E1−i ∈ BX . X ⊆ W1,X . Proof. For part (i), consider the short exact sequence (3.6): (3.7) 0 → E0 α→ E → E1 → 0. For any closed point s ∈ S, we have the exact sequence (3.8) E0s To begin with, suppose E ∈ W ′ αs→ Es → E1s → 0. 0,X . Then for a general s, the restriction E1s is s(E1s). Thus Next, suppose E ∈ W ′ Ψs-WIT0. On the other hand, the base change (3.3) gives (cE1)s ∼= Ψ1 (cE1)s = 0 for a general s ∈ S, i.e. cE1 ∈ BX . 1,X . By Lemma 3.6, the restriction E0s is Ψs-WIT0 for every closed point s ∈ S. However, Es is Ψs-WIT1 for a general closed point s ∈ S by assumption. Therefore, the map αs in (3.8) must be zero for a general closed point s ∈ S. In other words, the injection α in Coh(X) vanishes when we restrict to a general fiber over S. Hence E0 ∈ BX , and the lemma is proved. For part (ii), take any E ∈ W ′ 1,X ∩ B◦ we know that E0 ∈ BX , and hence must vanish, implying E ∈ W1,X . X. Let E0, E1 be as in (3.6). From part (i), (cid:3) Lemma 3.19. We have (i) The category W ′ (ii) W ′ 1,X ∩ FX = (W ′ 0,X )◦. 0,X is closed under quotients and extensions. T-STRUCTURES ON ELLIPTIC FIBRATIONS 13 (iii) The category TX is closed under quotients and extensions in Coh(X). 0,X )◦. Proof. Part (i) is clear. For part (ii), let us first show that W ′ Take any morphism α : F → E in Coh(X) where E ∈ W ′ 0,X . We want to show that α is the zero map. Since W ′ 0,X is closed under quotients by part (i), we can assume that α is an injection in Coh(X). Since E is torsion-free, we can also assume that F is torsion-free and nonzero. Then, for a general closed point s ∈ S, F s is Ψs-WIT0 while Es is Ψs-WIT1, implying αs = 0 for a general s. Thus F must be torsion, a contradiction. Hence α must be zero. 1,X ∩ FX , F ∈ W ′ 1,X ∩ FX ⊆ (W ′ For the other inclusion in (ii), take any E ∈ (W ′ 3.17, we have E ∈ W ′ know E is torsion-free. Thus (ii) is proved. 1,X . Since W ′ 0,X )◦ that is nonzero. By Remark 0,X contains all the torsion sheaves on X, we also Given part (i), in order to show part (iii), it is enough to show that any quotient α ։ F ′ 0,X )◦. Therefore, from part (ii), we 1,X . Now, consider the short exact sequence of an object F in Φ({Coh≤0(Ys)}↑) lies in TX . Consider any surjection F in Coh(X). We can assume that F ′ lies in (W ′ know F ′ is torsion-free and lies in W ′ in Coh(X) 0 → K → F α→ F ′ → 0 where K = ker (α). From Lemma 3.18(ii), we also know F ′ ∈ W1,X . On the other hand, that F ∈ W1,X implies K ∈ W1,X . The short exact sequence above is therefore taken by Ψ to the short exact sequence in Coh(Y ) 0 → bK → bF →cF ′ → 0. Since bF ∈ {Coh≤0(Ys)}↑, the same holds for cF ′, and so F ′ ∈ Φ({Coh≤0(Ys)}↑) ⊆ TX. Thus (iii) holds. (cid:3) By Lemma 3.19(iii), we now have a torsion pair (TX , T◦ X ) in Coh(X). We set CX := hT◦ X [1], TXi. For any subcategory C of Coh(X), where X is a smooth projective variety, we will define CD to be the subcategory of Coh(X) consistant of all sheaves of the form E xtc X (F, OX ), where F ∈ C and c is the codimension of the support of F . smooth, Now we define, for an elliptic fibration bπ of any dimension n and when Y is (3.9) BY,∗ := hCoh≤0(Y ), (W1,Y ∩ Coh=1(Y ))Di. Remark 3.20. In [HL, Definition 1.1.7], for a coherent sheaf E of codimension c on a smooth projective variety X, the notation ED denotes the sheaf E xtc X (E, ωX ) where ωX is the canonical sheaf on X. Lemma 3.21. The category BY,∗ is closed under quotients and extensions in Coh(Y ). Proof. Since BY,∗ is defined to be the category generated by Coh≤0(Y ) and (W1,Y ∩ Coh=1(Y ))D via extensions, we only need to verify that it is closed under quotients. Take any nonzero surjection α : E ։ T in Coh(Y ), where E ∈ (W1,Y ∩ Coh=1(Y ))D. We want to show that T ∈ BY,∗ as well. Note that, it suffices to assume T is pure 1-dimensional. By definition, E ∼= E xtn−1(F, OY ) for some 14 JASON LO F ∈ W1,Y ∩Coh=1(Y ). That F is pure 1-dimensional implies E ∼= F ∨[n− 1]. Hence we have a short exact sequence in Coh(Y ) (3.10) 0 → K → F ∨[n − 1] α→ T → 0 where K := ker (α). Assuming that α is not an isomorphism, we have that K is a pure 1-dimensional sheaf. That is, all the terms in (3.10) are pure 1-dimensional sheaves. Considering (3.10) as an exact triangle in D(Y ), then dualising and shift- ing, we obtain the exact triangle in D(Y ) (3.11) T ∨[n − 1] α∨[n−1] −→ F → K ∨[n − 1] → T ∨[n] where all the terms T ∨[n−1], F and K ∨[n−1] are again pure 1-dimensional sheaves. As a result, we have a short exact sequence in Coh(Y ) 0 → T ∨[n − 1] α∨[n−1] −→ F → K ∨[n − 1] → 0. Thus T ∨[n−1] ∈ W1,Y ∩Coh=1(Y ), and T ∼= E xtn−1 Hence BY,∗ is closed under quotients, and we are done. Y (T ′, OY ) where T ′ := T ∨[n−1]. (cid:3) Remark 3.22. Since Coh(Y ) is a Noetherian abelian category, we have another torsion pair (BY,∗, (BY,∗)◦) in Coh(Y ) by [Pol, Lemma 1.1.3]. Let us define, for an elliptic fibrationbπ : Y → S of any dimension, Y,∗[1], BY,∗i. DY := hB◦ Remark 3.23. When Y is an elliptic surface, the objects of BY are exactly the fiber sheaves. Also, since 0-dimensional sheaves on Y are always Φ-WIT0, the objects of W1,Y ∩ BY are exactly the pure 1-dimensional Φ-WIT1 fiber sheaves in this case. Therefore, we have the following equivalent description of BY,∗ when Y is an elliptic surface: (3.12) BY,∗ = hCoh≤0(Y ), (W1,Y ∩ BY )Di. Remark 3.24. On the other hand, on an elliptic fibration of any dimension, if E is a Φ-WIT1 pure 1-dimensional sheaf, then E cannot have any subsheaf lying in {Coh≤0(Ys)}↑, which is contained in W0,Y . That is, a Φ-WIT1 pure 1-dimensional sheaf on an elliptic fibration of any dimension is a fiber sheaf. 4. Elliptic fibrations In this section, we will consider a pair of dual elliptic fibrations π : X → S and bπ : Y → S. We first prove for elliptic surfaces, that the t-structure on D(X) with heart TX is equivalent to the t-structure on D(Y ) with heart DY (up to a shift) via a derived equivalence from D(X) to D(Y ) - see Theorem 4.12. This result is a special case of Yoshioka's result [Yos3, Proposition 3.3.5]. We then extend the above result to the case of elliptic threefolds - see Theorem 4.24 and Theorem 4.26 for the precise statements; in this case, the two t-structures differ by a tilt (in the sense of [HRS, Chap. I, Sec. 2]). Below, we choose to discuss elliptic surfaces and elliptic threefolds separately because the two cases are interesting in their own rights. Our central idea is to filter coherent sheaves on X or Y using the following torsion classes in Coh(X) (some of which are Serre subcategories) and their counterparts T-STRUCTURES ON ELLIPTIC FIBRATIONS 15 in Coh(Y ), to the point that we understand the image under the Fourier-Mukai transform Ψ of any subfactor in the filtration: (4.1) W0,X , W ′ 0,X , {Coh≤0(Xs)}↑, TX , BX,∗, Coh≤d(X), Coh(π)≤d where d ≥ 0. In Yoshioka's work [Yos1], he considered an elliptic surface π : X → S with a zero section σ, where all the fibers of π are integral. After identifying a compactification bπ : Y → S of the relative Jacobian with π itself and using the Poincar´e sheaf as the kernel, he proceeded to consider a Fourier-Mukai transform Ψ : D(X) → D(X). In [Yos1, Theorem 3.15] and [Yos1, Remark 3.5], Yoshioka proved an isomorphism between two moduli spaces of semistable sheaves on X where: • one of the two moduli spaces parametrises pure 1-dimensional sheaves (so they have rank zero); • the semistability is with respect to σ + kf, k ≫ 0 where f is a fiber class for the fibration π; • the isomorphism is induced by the composite functor (Ψ(−))∨, i.e. the Fourier-Mukai transform Ψ followed by the dualising functor on D(X). Later, in [Yos3, Proposition 3.4.5], he generalised the above results to the case of twisted semistable perverse coherent sheaves on dual elliptic surfaces that arise as resolutions of singularities. Following Yoshioka's idea, we will study the functor that is the composition of the Fourier-Mukai transform from an elliptic fibration to its dual, followed by the dualising functor. From now on, we will write Λ(−) to denote the composite functor (Ψ(−))∨, i.e. the Fourier-Mukai functor Ψ(−) : D(X) → D(Y ) followed by the derived dual −∨ : D(Y ) → D(Y ), irrespective of the dimensions of X and Y . We will also write Λi(−) to denote H i(Λ(−)), where H i(−) is the degree-i cohomology functor with respect to the standard t-structure on D(Y ). 4.1. t-structures on elliptic surfaces. In this section, we assume that π : X → S andbπ : Y → S are a pair of dual elliptic surfaces. Lemma 4.1. Suppose X is an elliptic surface, and E ∈ BX. Let E0, E1 be as in (3.6). Then ΛE ∈ D[0,1] Coh(Y ), and (i) Λ0E ∼= E xt1(cE1, OY ); (ii) there is a short exact sequence in Coh(Y ) Proof. Take any E ∈ BX. From (3.6), we obtain the exact triangle in D(Y ) 0 → E xt2(cE1, OY ) → Λ1E → E xt1(cE0, OY ) → 0. (4.2) (4.3) (4.4) Taking derived duals, we obtain the exact triangle cE0 → Ψ(E) → cE1[−1] → cE0[1]. (cE1)∨[1] → ΛE → (cE0)∨ → (cE1)∨[2]. Since cE0 is Φ-WIT1, it has no 0-dimensional subsheaves. Besides, since E0 is a fiber sheaf, its transformcE0 remains a fiber sheaf. Hence cE0 is pure 1-dimensional, and thus E xti(cE0, OY ) = 0 for all i 6= 1, meaning (cE0)∨ ∼= E xt1(cE0, OY )[−1] is a 1-dimensional sheaf sitting at degree 1. 16 JASON LO On the other hand, since E1 is a fiber sheaf, the same holds for cE1, and so E xt0(cE1, OY ) = 0. The lemma then follows by taking the long exact sequence of cohomology of (4.4). (cid:3) Lemma 4.2. Suppose X is an elliptic surface, and E ∈ TX ∩ B◦ D[0,1] Coh(Y ). Furthermore: X . Then ΛE ∈ is as in (3.6). Proof. Take any E ∈ TX ∩ B◦ is a locally free sheaf. Thus part (i) holds. X, then E has no fiber subsheaves, and in particular, (i) If E ∈ W0,X , then bE is a locally free sheaf. (ii) Λ1E is a 0-dimensional sheaf that is a quotient of E xt2(cE1, OY ), where E1 is pure 1-dimensional. Let Ei be as in (3.6). By [Lo1, Corollary 5.4], we know cE0 By [Lo1, Lemma 2.6], we know E1 is a fiber sheaf, and so (cE1)∨ is a 2-term complex sitting at degrees 1 and 2, where the degree-2 cohomology is E xt2(cE1, OY ), by part (i), we know cE0 is locally free, and so E xt1(cE0, OY ) = 0. Taking the long exact sequence of cohomology of (4.4) then gives us part (ii) of the lemma. That ΛE ∈ D[0,1] (cid:3) which is 0-dimensional. Since E0 is a subsheaf of E, we have E0 ∈ TX ∩ B◦ Coh(Y ) also follows from the long exact sequence. X . And Before we consider the images of sheaves supported in dimension 2 under the functor Λ, we prove: Lemma 4.3. Suppose π : X → S is an elliptic surface or an elliptic threefold. Suppose E is a pure d-dimensional, Ψ-WIT0 sheaf on X. (i) If E ∈ Coh(π)d−1, then bE is a pure sheaf of dimension d. (ii) If E ∈ Coh(π)d and E has no subsheaves E′ in Coh(π)d−1, then bE is a pure sheaf of dimension d + 1. Proof. (i): suppose E is a pure d-dimensional Ψ-WIT0 sheaf lying in Coh(π)d−1. When d = 3, E is torsion-free, and the result is just [BM, Lemma 9.4]. When d = 1, of part (i). T ∈ W1,Y as well. In particular, T cannot have any subsheaf in {Coh≤0(Ys)}↑ by E is a fiber sheaf, and so bE is a Φ-WIT1 fiber sheaf, which is necessarily pure. Now, suppose d = 2. Then bE ∈ Coh(π)1 by Lemma 3.2, and so dimbE ≤ 2. Suppose there is a nonzero subsheaf T of bE where T ∈ Coh≤1(Y ). Since bE ∈ W1,Y , we have Lemma 3.6. Hence T is forced to be a fiber sheaf. The injection T ֒→ bE is then transformed under Φ to a nonzero map bT → E, implying E has a fiber subsheaf, contradicting its purity. Hence bE must be pure when d = 2. This finishes the proof suppose E ∈ Coh(π)d. Let us write Z := π(suppbE) = π(supp E). In this case, that, for a general closed point s ∈ Z, we have dim (bEs) = 1. If we write bEZred for the pullback of bE along the base change Zred ֒→ Z ֒→ S, then it is enough to show dim ((bEZred )s) = 1 for a general s ∈ Z. That is, we can assume that Z is reduced. Applying generic flatness to E and bE with respect to the morphism the fiber Es is 0-dimensional for a general closed point s ∈ Z. We will now show X ×S Z → Z together with [BBR, Corollary 6.3(3)], we obtain that, for a general (ii): suppose E is a pure d-dimensional Ψ-WIT0 sheaf, except that now we T-STRUCTURES ON ELLIPTIC FIBRATIONS 17 s ∈ Z, the restriction bEs ∼= dEs is 1-dimensional, as wanted. Therefore, we have dimbE = d + 1. Now, suppose we have an injection T ֒→ bE where 0 6= T ∈ Coh≤d(Y ). Then T is Φ-WIT1. We consider the different cases: • When d = 2, and X is of dimension 3: if π(supp(T )) is 2-dimensional (and hence equal to S), then T itself is 2-dimensional. This implies that T s is 0-dimensional for a general closed point s ∈ S, and so T ∈ W ′ 0,Y . However, we also have T ∈ W ′ 1,Y , 1,Y by Lemma 3.16(ii). Thus T lies in ∈ W ′ 0,Y ∩ W ′ contradicting the assumption that E has no subsheaves in Coh(π)d−1. • When d = 1, and X is of dimension 2 or 3: by Lemma 3.15 and Remark 3.14, that T ∈ W1,Y implies T cannot have any subsheaf in {Coh≤0(Ys)}↑ which is contained in BY by Lemma 3.18, i.e. dim (bπ(supp(T ))) ≤ 1. The injection T ֒→ bE is then taken by Φ to a nonzero morphism bT → E, and must be a fiber 1-dimensional sheaf. Then the injection T ֒→ bE is taken by Φ to a nonzero morphism bT → E, contradicting the assumption that E has no subsheaves lying in Coh(π)d−1. Hence bE must be pure of dimension d + 1, finishing the proof of (ii). Lemma 4.3 also yields the following results on reflexivity of sheaves under Fourier- (cid:3) Mukai transforms: Lemma 4.4. Let π : X → S be an elliptic surface or an elliptic threefold. Suppose (4.5) exact sequence Ext1(W0,X ∩ Coh(π)≤0, E) = 0. of [HL, Definition 1.1.7]), is torsion-free and reflexive, while T is a coherent sheaf of 0 → bE → (bE)DD → T → 0 E is a Ψ-WIT0 torsion-free sheaf. Then bE is torsion-free, and is reflexive whenever Proof. By Lemma 4.3(i), we know bE is pure of codimension 0, so we have a short where (bE)DD, being the double dual of bE (where the 'dual' of a sheaf is in the sense codimension at least 2 (which implies T ∈ BY ). Note that Ext1(W0,Y ∩ BY , bE) ∼= we have Hom(W0,Y ∩ BY , T ) = 0. This implies that T ∈ W1,Y , and so (bE)DD is together imply that T must be a fiber sheaf. That is, we have T ∈ W1,Y ∩Coh(bπ)≤0. Hom(W1,X ∩ BX , E) = 0 since E is torsion-free. Now, take any A ∈ W0,Y ∩ BY . Applying the functor Hom(A, −) to (4.5), we get Hom(A, T ) = 0. In other words, also Φ-WIT1. On the other hand, since dim T ≤ 1, Lemma 3.15 and Remark 3.14 The lemma then follows from Ext1(W1,Y ∩ Coh(bπ)≤0, bE) ∼= Ext1(W0,X ∩ Coh(π)≤0, E). (cid:3) Corollary 4.5. Suppose X is an elliptic threefold where all the fibers are Cohen- Macaulay with trivial dualising sheaves. If E is a Ψ-WIT0 reflexive torsion-free sheaf, then bE is also a reflexive torsion-free sheaf. Proof. Since E is reflexive and torsion-free, we have Ext1(Coh≤1(X), E) = 0 by [CL, Lemma 4.21]. The corollary then follows from Lemma 4.4. (cid:3) Lemma 4.6. Let X be an elliptic threefold. Suppose that: 18 JASON LO • E ∈ W0,X ; • Hom(Φ(Coh≤0(Y )), E) = 0; and • bE is a pure sheaf of dimension at least 2. Then E xt2(bE, OY ) = 0, and Λ(E) ∈ D[0,1] sion 2, then bE is reflexive. Proof. Consider the canonical short exact sequence Coh(Y ). In particular, if bE is pure of dimen- (4.6) 0 → bE → (bE)DD → T → 0 in Coh(Y ) where T ∈ Coh≤1(Y ). Since (bE)DD is a reflexive sheaf on a threefold, we have E xti((bE)DD, OY ) = 0 for i ≥ 2 regardless of whether bE is of dimension 2 or 3 (see [HL, Proposition 1.1.6(ii), Proposition 1.1.10(4')]). On the other hand, from (4.6) we have the exact sequence E xt2((bE)DD, OY ) → E xt2(bE, OY ) → E xt3(T, OY ) → 0, vanishes by showing that T is pure 1-dimensional. which gives E xt2(bE, OY ) ∼= E xt3(T, OY ). We will now show that E xt3(T, OY ) Let Q be the maximal 0-dimensional subsheaf of T ; we can pull back the short ex- act sequence (4.6) along the inclusion Q ֒→ T , to obtain the following commutative diagram of short exact sequences in Coh(Y ): 0 0 / bE / bE / F / Q / 0 . / T / 0 / (bE)DD Then F is necessarily pure of dimension at least 2, since it is a subsheaf of (bE)DD. However, we have Ext1(Q,bE) ∼= Hom(Q, bE[1]) ∼= Hom(Φ(Q), E), which vanishes since Hom(Φ(Coh≤0(Y )), E) = 0 by assumption, implying F ∼= bE⊕Q, contradicting the purity of F unless Q = 0. Hence T is pure 1-dimensional, and we obtain Λ(E) ∈ D[0,1] Coh(Y ). The last assertion of the lemma follows from [HL, Proposition 1.1.10]. (cid:3) Corollary 4.7. Suppose π : X → S is an elliptic threefold. Suppose E lies in Coh≤1(X) ∩ Coh(π)1 and has no fiber subsheaves. Then E is Ψ-WIT0, and its Proof. By Lemma 3.15, we have E ∈ Φ({Coh≤0(Ys)}↑), which implies E is Ψ-WIT0 transform bE is a 2-dimensional reflexive sheaf. by Remark 3.14. That E has no fiber subsheaves implies bE is pure of dimension 2 by Lemma 4.3(ii). Lemma 4.6 gives us the reflexivity of bE. Lemma 4.8. Suppose X is an elliptic surface, and E ∈ Coh=2(X) ∩ W ′ ΛE ∈ D[0,1] Coh(Y ), and Λ1E is a 0-dimensional sheaf. 0,X . Then (cid:3) Proof. By Lemma 3.18, we have E1 ∈ BX . On the other hand, cE0 is pure 2- dimensional by Lemma 4.3. Hence in the exact triangle (4.4), the complex (cE0)∨ / /   /   / / / / / T-STRUCTURES ON ELLIPTIC FIBRATIONS 19 lies in D[0,1] result, we have ΛE ∈ D[0,1] Coh(Y ), while ΛE1 = (cE1)∨[1] also lies in D[0,1] Coh(Y ). In particular, we have the exact sequence Coh(Y ) by Lemma 4.1. As a (4.7) E xt2(cE1, OY ) → Λ1E → E xt1(cE0, OY ) → 0. Since cE0 is pure 2-dimensional, the sheaf E xt1(cE0, OY ) is 0-dimensional, as is E xt2(cE1, OY ). Hence Λ1E itself is 0-dimensional. Lemma 4.9. Let X be an elliptic surface, and suppose E ∈ Φ({Coh≤0(Ys)}↑). Then (cid:3) (i) ΛE ∈ D[0,1] Coh(Y ) and Λ1E is a 0-dimensional sheaf; sheaf (lying in {Coh≤0(Ys)}↑) sitting at degree 0. Proof. By Lemma 3.6, the category Φ({Coh≤0(Ys)}↑) is contained in W1,X . Take (ii) E is torsion-free if and only if ΛE ∼= E xt1(bE, OY ) is a pure 1-dimensional any E ∈ Φ({Coh≤0(Ys)}↑). Then bE is Φ-WIT0. Consider the short exact sequence in Coh(Y ) where F0 is the maximal 0-dimensional subsheaf of bE. Then both F0, F1 are Φ-WIT0, and the dimension of F1 is 1 if F1 6= 0; we also have the exact triangle in D(Y ) (4.8) 0 → F0 → bE → F1 → 0 1 → (bE)∨ → F ∨ 0 → F ∨ F ∨ 1 [1]. Here, F ∨ 0 is a 0-dimensional sheaf sitting at degree 2, while F ∨ 1 On the other hand, we have Ψ(E) ∼= bE[−1], and so ΛE ∼= (bE)∨[1] ∈ D[0,1] Besides, the exact triangle above gives Λ1E ∼= H 2((bE)∨) ∼= H 2(F ∨ Now, the transform of (4.8) is a short exact sequence in Coh(X) 0-dimensional sheaf. Thus part (i) holds. ∼= E xt1(F1, OY )[−1]. Coh(Y ). 0 ), which is a (4.9) 0 →cF0 → E →cF1 → 0. sitting at degree 0. Since cF0 is a fiber sheaf, it must be zero when E is torsion-free, in which case the argument in part (i) shows that ΛE ∼= E xt1(bE, OY ) is a pure 1-dimensional sheaf For the 'if' part of part (ii), suppose that ΛE ∼= E xt1(bE, OY ) is a pure 1- see that F0 vanishes and E ∼=cF1. However, if cF1 had a torsion subsheaf F ′, then it would be a Ψ-WIT1 fiber sheaf by [Lo1, Lemma 2.6]. This means that F1 itself would also have a fiber subsheaf, contradicting that F1 is a pure 1-dimensional sheaf lying in {Coh≤0(Ys)}↑. Thus part (ii) is proved. (cid:3) dimensional sheaf sitting at degree 0. Then from the computation of part (i), we Lemma 4.10. Suppose X is an elliptic surface and E ∈ Coh(X) is a pure 2- dimensional sheaf such that: • E ∈ W ′ • Hom(Φ({Coh≤0(Ys)}↑), E) = 0. 1,X ; Then Λ(E[1]) ∈ D[0,1] form Coh(Y ) and Λ1(E[1]) sits in an exact sequence in Coh(Y ) of the E xt1(A, OY ) → Λ1(E[1]) → E xt1(B, OY ) → 0 where A is pure 2-dimensional (hence E xt1(A, OY ) is 0-dimensional), and B ∈ W1,Y ∩ BY (in particular, B is a pure 1-dimensional fiber sheaf ). D(Y ) (4.10) and 1, and we have the exact sequence in Coh(Y ) 20 JASON LO Proof. Let E0, E1 be as in (3.6). By Lemma 3.18, we know E0 ∈ BX . However, we are assuming E to be pure, and so E = E1, and Λ(E[1]) ∼= (cE1)∨. Suppose T is the maximal torsion subsheaf of cE1. Note that, cE1 has no 0- dimensional subsheaves, or else there would be a nonzero map from a fiber sheaf to E1, contradicting the purity of E. Therefore, T must be a pure 1-dimensional sheaf if it is nonzero, and we have the short exact sequence in Coh(Y ) 0 → T →cE1 → cE1/T → 0 we have E xt0(T, OY ) = 0 = E xt2(T, OY ). Hence T ∨ is a pure 1-dimensional sheaf (cE1/T )∨ → (cE1)∨ → T ∨ → (cE1/T )∨[1]. where cE1/T is a pure 2-dimensional sheaf. We thus obtain an exact triangle in Since cE1/T is pure 2-dimensional, the sheaf E xt2(cE1/T, OY ) vanishes, and so (cE1/T )∨ has cohomology only in degrees 0 and 1. Also, since T is pure 1-dimensional, sitting at degree 1. Thus Λ(E[1]) ∼= (cE1)∨ is a 2-term complex sitting in degrees 0 where E xt1(cE1/T, OY ) is a 0-dimensional sheaf. composition T ′ ֒→ T ֒→ cE1 = bE under Φ would give us a nonzero element in W0,Y ,cE1) = 0, meaning the Φ-WIT0 part of T vanishes, i.e. T is Φ-WIT1 as Hom(Φ({Coh≤0(Ys)}↑), E), contradicting our assumption. Thus T must be a fiber sheaf by Lemma 3.15. Since E = E1 is pure 2-dimensional, we have Hom(BY ∩ E xt1(cE1/T, OY ) → H 1((cE1)∨) → E xt1(T, OY ) → 0. To finish off the proof, observe that T has no nonzero subsheaves lying in the category {Coh≤0(Ys)}↑. For, if T had such a subsheaf T ′, then the image of the claimed. (cid:3) Pulling the above results together, we can now characterise the image of the heart CX under the functor Λ: Proposition 4.11. Let X be an elliptic surface. Then for any E ∈ CX , we have: (i) ΛE ∈ D[0,1] (ii) Λ1E ∈ BY,∗. Coh(Y ); Proof. We have the following inclusions of torsion classes in Coh(X): BX ⊆ TX = Coh≤1(X) ⊆ W ′ 0,X . Given any E ∈ W ′ 0,X , we can first find a short exact sequence in Coh(X) 0 → E0 → E → E1 → 0 where E0 ∈ Coh≤1(X) and E1 ∈ Coh=2(X) ∩ W ′ sequence in Coh(X) 0,X , and then another short exact 0 → E0,0 → E0 → E0,1 → 0 where E0,0 ∈ BX and E0,1 ∈ Coh≤1(X) ∩ B◦ we obtain a filtration in Coh(X) X. Setting E′′ := E0,0 and E′ := E0, E′′ ⊆ E′ ⊆ E T-STRUCTURES ON ELLIPTIC FIBRATIONS 21 where E′′ ∈ BX , E′/E′′ ∈ TX ∩ B◦ is defined as the extension closure hW ′ and 4.9 together imply Λ(TX ) ⊂ D[0,1] Coh(Y ). X and E/E′ ∈ W ′ 0,X ∩ Coh=2(X). Since TX 0,X , Φ({Coh≤0(Ys)}↑)i, Lemmas 4.1, 4.2, 4.8 Now, by the definition of TX and Lemma 3.16(ii), we see that any object in (TX )◦ satisfies the hypotheses of Lemma 4.10. This gives us Λ(T◦ Coh(Y ). Together with the last paragraph, we now have Λ(CX) ⊂ D[0,1] Coh(Y ), i.e. we have part (i) of the proposition. Part (ii) of the proposition follows from the computations in Lemmas 4.1, 4.2, 4.8, 4.9 and 4.10. (cid:3) X [1]) ⊂ D[0,1] We now have the following theorem: Theorem 4.12. When X is a smooth elliptic surface, the functor Λ(−) := (Ψ(−))∨ induces an equivalence between the t-structure (D≤0 C ) on D(X), and the t- structure (D≤0 D ) on D(Y ). Equivalently, Λ induces an equivalence of hearts D , D≥0 C , D≥0 (4.11) CX ∼→ DY [−1]. Proof. By Remark 3.22 and Lemma 3.19, the two categories CX , DY are both hearts of t-structures. Proposition 4.11(i) shows that the functor Λ takes CX to a heart of the form hF [1], T i[−1], for some torsion pair (T , F ) in Coh(Y ). Moreover, Propo- sition 4.11(ii) shows that T ⊆ BY,∗. Therefore, to prove the theorem, it remains to show that BY,∗ ⊆ T . That is, it remains to show that every object E ∈ BY,∗ appears as the degree-1 cohomology of ΛE′ for some E′ ∈ CX . Furthermore, from the construction of BY,∗ and Remark 3.23, it is enough to consider the following two cases: (1) when E ∼= Oy is a skyscraper sheaf of length 1 supported at a closed point y ∈ Y ; (2) when E ∈ (W1,Y ∩ BY )D. y [1] ∼= Oy[−1]. This shows that Oy ∈ T . In case (2), suppose E ∈ (W1,Y ∩ BY )D. Then E ∼= E xt1 In case (1), observe that cOy ∈ BX ⊂ CX , and Λ(cOy) ∼= (Oy[−1])∨ ∼= O ∨ F ∈ W1,Y ∩ BY . Then bF ∈ BX ⊂ CX , and ΛbF ∼= F ∨ ∼= E xt1(F, OY )[−1] = E[−1], showing E ∈ T . This completes the proof of the theorem. In [Yos3, Section 3.3], Yoshioka considered a torsion pair (T, F) in a category of perverse sheaves. When the category of perverse sheaves coincides with Coh(X), the torsion class T in Coh(X) is the category of objects E ∈ Coh(X) such that in its relative Harder-Narasimhan filtration with respect to π (see, for instance, [Yos3, (3.5)]), all the subfactors satisfy the inequality µf ≥ 0. Here, µf is a slope function defined as follows: writing f to denote the fiber class of the morphism π, and for any F ∈ Coh(X), we set Y (F, OY ) for some (cid:3) µf (F ) :=( c1(F )·f rk F +∞ if rk (F ) > 0, if rk (F ) = 0. When π : X → S is a smooth elliptic surface with a section and integral fibers, it has a relative compactified Jacobian bπ : Y → S that is a fine moduli space and a universal 'Poincar´e' sheaf. For instance, bπ can be taken to be the Altman- Kleiman compactification [BBR, Remark 6.33]. Under this setting, Theorem 4.12 22 JASON LO can be considered as a special case (the 'untwisted' case, and where there are no singularities to resolve) of [Yos3, Proposition 3.3.6], in the following precise sense: Lemma 4.13. Suppose π : X → S is a smooth elliptic surface with a section and integral fibers, and bπ : Y → S is a compactification of the relative Jacobian of X, and Ψ : D(X) → D(Y ) is the relative Fourier-Mukai transform with the Poincar´e sheaf as the kernel. Then the torsion classes T and TX in Coh(X) coincide. Proof. From our definition (3.5) of TX , it is clear that µf (F ) ≥ 0 for any F ∈ TX. Since TX is closed under quotients, the right-most subfactor in the relative HN filtration of F must have µf ≥ 0. Hence µf (F ′) ≥ 0 for all the subfactors F ′ in the relative HN filtration of F , and so TX ⊆ T. For the other inclusion, take any E ∈ T. Consider the short exact sequence in Coh(X) 0 → E′ → E → E′′ → 0 0,X and E′′ ∈ (W ′ where E′ ∈ W ′ 0,X )◦. Then E′′ is torsion-free and Φ-WIT1. Note that the left-most subfactor in the relative HN filtration of E′′ must have µf ≤ 0, for otherwise that subfactor would lie in W ′ 0,X by [BBR, Corollary 3.29], a contradiction. Since T is a torsion class in Coh(X), we have E′′ ∈ T. Therefore, all a 0-dimensional sheaf for a general closed point s ∈ S (again by [BBR, Corollary the subfactors in the relative HN filtration of E′′ have µf = 0. As a result, cE′′s is 3.29]). In particular, we have cE′′ ∈ Coh≤1(Y ). By Lemma 3.15, we can then fit cE′′ in a short exact sequence in Coh(Y ) where (cE′′)0 ∈ {Coh≤0(Ys)}↑ and (cE′′)1 ∈ BY . Note that, all the terms in the above short exact sequence are Φ-WIT0, and so we obtain E′′ ∈ hBX , Φ({Coh≤0(Ys)}↑)i ⊆ TX. Thus E ∈ TX . (cid:3) 0 → (cE′′)0 → cE′′ → (cE′′)1 → 0 4.2. t-structures on elliptic threefolds. In this section, we prove an analogue of Theorem 4.12 on dual elliptic threefolds π : X → S andbπ : Y → S. The strategy is the same as that on elliptic surfaces - we analyse the images of various categories of coherent sheaves under the functor Λ(−) := (Ψ(−))∨. Note that, since we defined a fiber sheaf on X to be a sheaf supported on a finite number of fibers of π, now that π : X → S is an elliptic threefold, the category BX coincides with Coh(π)≤1, and is strictly larger than the category of fiber sheaves on X, which is precisely Coh(π)≤0. Lemma 4.14. Suppose X is an elliptic threefold. Let E be any fiber sheaf on X. Then: (i) If E is 0-dimensional, then Λ(E) ∼= E xt2(bE, OY )[−2]. (ii) If E is 1-dimensional and Ψ-WIT0, then Λ(E) ∼= (bE)∨ ∼= E xt2(bE, OY )[−2]. (iii) If E is 1-dimensional and Ψ-WIT1, then Λ(E) ∼= (bE)∨[1] lies in D[1,2] with Λ1E ∼= E xt2(bE, OY ) being a pure 1-dimensional fiber sheaf (if nonzero), and Λ2E ∼= E xt3(bE, OY ) being a 0-dimensional sheaf. Overall, if E is an arbitrary fiber sheaf on X, then Λ(E) only has cohomology at degrees 1 and 2, and Λ2E ∈ BY,∗. Coh(Y ), T-STRUCTURES ON ELLIPTIC FIBRATIONS 23 Proof. The proofs of statements (i), (ii) and (iii) are all straightforward, and we omit them here. Given any fiber sheaf E on X, i.e. E ∈ Coh(π)0, we can find a short exact sequence in Coh(X) 0 → E0 → E → E1 → 0 where E0 ∈ Coh(π)0 ∩ W0,X and E1 ∈ Coh(π)0 ∩ W1,X . We can then find another short exact sequence in Coh(X) 0 → E0,0 → E0 → E0,1 → 0 where E0,0 ∈ Coh≤0(X) and E0,1 ∈ Coh(π)0 ∩ W0,X ∩ Coh=1(X). As a result, we obtain a filtration in Coh(X) E0,0 ⊆ E0 ⊆ E where E0,0 ∈ Coh≤0(X), E0/E0,0 ∈ Coh=1(X) ∩ W0,X and E/E0 ∈ Coh=1(X) ∩ W1,X . The final claim of the lemma then follows from statements (i) through (iii). (cid:3) Lemma 4.15. Suppose X is an elliptic threefold. Let E ∈ Coh≤1(X) be such that E has no fiber subsheaves. Then: (i) E ∈ Coh(π)1. (ii) E ∈ W0,X , and bE is a 2-dimensional reflexive sheaf. (iii) Λ(E) ∼= (bE)∨ ∼= E xt1(bE, OY )[−1] ∈(cid:16)Coh=2(Y ) ∩ Ψ({Coh≤0(Xs)}↑)(cid:17) [−1]. Proof. Part (i) follows from Lemma 3.15. Part (ii) is just Corollary 4.7, and part (iii) follows easily from part (ii) and [HL, Proposition 1.1.10]. (cid:3) Lemma 4.16. Suppose X is an elliptic threefold. Let E ∈ Coh=2(X) ∩ Coh(π)1. Then, with E0, E1 as in (3.6), we have: Overall, Λ(E) ∈ D[0,2] Coh≤0(Y ). (i) If E0 6= 0, then cE0 is pure 2-dimensional and reflexive, and Λ(E0) ∼= E xt1(cE0, OY )[−1]; (ii) cE1 ∈ Coh≤2(Y ), and Λ(E1) ∈ D[0,2] Coh(Y ) with H 0(Λ(E1)) ∼= E xt1(cE1, OY ) (which is pure of dimension 2 if nonzero) and H 2(Λ(E1)) ∼= E xt3(cE1, OY ). Coh(Y ) with Λ0E ∼= E xt1(cE1, OY ), Λ1E ∈ Coh≤2(Y ) and Λ2E ∈ Proof. If E0 6= 0, then E0 is also pure of dimension 2. Then by Lemma 4.3(i), cE0 Hom(Φ(Coh≤0(Y )), E0) = 0. Hence cE0 is reflexive by Lemma 4.6, and part (i) is pure 2-dimensional. That E0 is pure 2-dimensional also implies the vanishing For part (ii), note that Lemma 3.2 gives holds. dim (π(supp(cE1))) = dim (π(supp(E1))) = 1, and socE1 ∈ Coh≤2(Y ) and Λ(E1) ∈ D[0,2] Coh(Y ). IfcE1 is 2-dimensional, then since it is of codimension 1, the sheaf E xt1(cE1, OY ) is nonzero and is also pure 2-dimensional. The rest of the lemma is clear. (cid:3) 24 JASON LO Lemma 4.17. Suppose X is an elliptic threefold. Let E ∈ Coh=2(X) ∩ Coh(π)2, and suppose E has no subsheaves lying in Coh(π)1. Let E0, E1 be as in (3.6). Then cE0 ∈ Coh=3(Y ), and ΛE ∈ hCoh≥2(Y ), Coh≤2(Y )[−1], BY,∗[−2]i. Proof. Since dim E = dim (π(supp(E))) = dim S, we know Es is 0-dimensional for a general closed point s ∈ S. As a result, E ∈ W ′ 0,X and if we let E0, E1 be as in (3.6), then Lemma 3.18 says that E1 ∈ BX. That E has no subsheaves in Coh(π)1 implies E0 also has no subsheaves in sion 3. On the other hand, since E0 is pure 2-dimensional, Lemma 4.6 gives us Coh(Y ) with Λ0(E0) ∈ Coh=3(Y ) and Coh(π)1. Therefore, by Lemma 4.3(ii), we obtain that cE0 is pure of dimen- E xt2(cE0, OY ) = 0. Thus ΛE0 ∼= (cE0)∨ ∈ D[0,1] Now, since E1 ∈ BX = Coh(π)≤1, we can fit E1 in a short exact sequence in Λ1(E0) ∈ Coh≤1(Y ). Coh(X) (4.12) 0 → T1 → E1 → E1/T1 → 0 where T1 ∈ Coh≤1(X) and E1/T1 ∈ Coh=2(X) ∩ Coh(π)1. We can further fit T1 in a short exact sequence in Coh(X) (4.13) 0 → T1,f → T1 → T1/T1,f → 0 where T1,f ∈ Coh(π)0, and T1/T1,f ∈ Coh=1(X) ∩ (Coh(π)0)◦. Now, by Lemmas 4.14, 4.15 and 4.16, as well as the filtrations (4.12) and (4.13), we see that ΛE1 ∈ hCoh=2(Y ), Coh≤2(Y )[−1], BY,∗[−2]i. This, together with the previous paragraph, gives us ΛE ∈ hCoh≥2(Y ), Coh≤2(Y )[−1], BY,∗[−2]i as claimed. (cid:3) Lemma 4.18. Suppose X is an elliptic threefold. Let E ∈ W ′ let E0, E1 be as in (3.6). Then: 0,X ∩ Coh=3(X), and (i) E1 ∈ BX ; (ii) E0 6= 0 and cE0 ∈ Coh=3(Y ); (iii) Λ(E0) ∈ D[0,2] Λ2(E0) ∈ Coh≤0(Y ). Coh(Y ) with Λ0(E0) ∼= E xt0(cE0, OY ), Λ1(E0) ∈ Coh≤1(Y ) and (iv) Λ0(E) is nonzero, supported in dimension 3 and lies in Coh≥2(Y ). Overall, we have ΛE ∈ D[0,2] Coh(Y ) with Λ1 ∈ Coh≤2(Y ) and Λ2E ∈ BY,∗. Proof. Part (i) follows from Lemma 3.18. If E0 = 0, then E = E1 would be torsion by part (i), a contradiction; thus E0 6= 0. That cE0 is pure 3-dimensional follows Now, by part (i), we know E1 ∈ BX . From the second half of the proof of Lemma from Lemma 4.3(i). Thus part (ii) holds, and part (iii) follows easily. 4.17, we also know that ΛE1 ∈ hCoh=2(Y ), Coh≤2(Y )[−1], BY,∗[−2]i, T-STRUCTURES ON ELLIPTIC FIBRATIONS 25 and so Λ0(E1) is a pure 2-dimensional sheaf if nonzero. From (3.6), we have the exact triangle in Db(Y ) Λ(E1) → Λ(E) → Λ(E0) → Λ(E1)[1] and the associated long exact sequence of cohomology: 0 → Λ0(E1) → Λ0(E) α→ Λ0(E0) → Λ1(E1) → · · · . From parts (ii) and (iii), we know that Λ0(E0) 6= 0 is nonzero and pure 3- dimensional. Since Λ1(E1) ∈ Coh≤2(Y ) from above, we see that α is nonzero. Thus im (α) is nonzero and is pure 3-dimensional. Now, we also know Λ0(E1) ∈ Coh=2(Y ) from above, and so Λ0(E) must be nonzero, supported in dimension 3, and lies in Coh≥2(Y ). (cid:3) Lemma 4.19. Suppose X is an elliptic threefold, and E ∈ Φ({Coh≤0(Ys)}↑). Then Λ(E) ∈ D[0,2] Coh(Y ) where Λ0(E) ∼= E xt1(T2, OY ) is a pure 2-dimensional sheaf Coh≤1(Y ) and Λ2(E) ∈ Coh≤0(Y ). if nonzero (here, T2 denotes the pure 2-dimensional component of bE), Λ1(E) ∈ Proof. By Lemma 3.6, the sheaf E is Ψ-WIT1. Thus Λ(E) ∼= (bE[−1])∨ ∼= (bE)∨[1]. Since bE ∈ {Coh≤0(Ys)}↑ ⊆ Coh≤2(Y ), we know (bE)∨ ∈ D[1,3] Let T1 be the maximal 1-dimensional subsheaf of bE, and let T2 := bE/T1. From the short exact sequence 0 → T1 → bE → T2 → 0 in Coh(Y ), we obtain an exact Coh(Y ), and so Λ(E) ∈ triangle in D(Y ) D[0,2] Coh(Y ). T1 → bE → T2 → T1[1]. (cid:3) Dualising this exact triangle and taking the long exact sequence of cohomology, 2 ) ∼= E xt1(T2, OY ). The rest of the lemma we obtain Λ0(E) ∼= H 1((bE)∨) ∼= H 1(T ∨ follows easily from the long exact sequence. Lemma 4.20. Suppose X is an elliptic threefold. Let E ∈ Coh=3(X) ∩ W ′ suppose Hom(Φ({Coh≤0(Ys)}↑), E) = 0. Then E ∈ W1,X , and Λ(E[1]) ∈ D[0,2] 1,X and Coh(Y ) with Λ0(E[1]) ∼= E xt0(E′, OY ), where E′ is the torsion-free part of bE and is Φ- WIT0, and Λ1(E[1]) ∈ Coh≤2(Y ) while Λ2(E[1]) ∈ BY,∗. Proof. With E as given, let E0, E1 be as in (3.6). Then E0 ∈ BX by Lemma 3.18. Since E is pure 3-dimensional, we have E0 = 0 and so E ∈ W1,X . Hence (4.14) That E is torsion-free implies Λ(E[1]) ∼= (bE)∨. Now, let T be the maximal torsion subsheaf of bE. Note that T could well be the same as cE0. Also, let T1 be the maximal 1-dimensional subsheaf of T and let can consider the images of T1, T2 and bE/T under derived dual (−)∨ separately: T2 := T /T1, which is pure 2-dimensional if nonzero. By the vanishing (4.14), T1 is pure 1-dimensional, and must be Φ-WIT1 (by Remark 3.14 and Lemma 3.15). We Hom(BY ∩ W0,Y , bE) = 0. ∼= E xt2(T1, OY )[−2] ∈ BY,∗[−2]. • T ∨ 1 26 JASON LO • T ∨ 2 ∈ D[1,2] H 2(T ∨ 2 ) ∼= E xt2(T2, OY ) is 0-dimensional. Coh(Y ) where H 1(T ∨ 2 ) ∼= E xt1(T2, OY ) is pure 2-dimensional, and • (bE/T )∨ ∈ D[0,2] Coh(Y ) where H 0((bE/T )∨) ∼= E xt0(bE/T, OY ) (and bE/T , being the quotient of a Φ-WIT0 sheaf, is again Φ-WIT0). Also, H 1((bE/T )∨)) ∼= E xt1(bE/T, OY ) is 1-dimensional while H 2((bE/T )∨)) ∼= E xt2(bE/T, OY ) is From above, we see that H 0(T ∨) = 0. From the short exact sequence in Coh(Y ) 0-dimensional. we obtain the long exact sequence 0 → T → bE → bE/T → 0, 0 → H 0((bE/T )∨) → H 0((bE)∨) → H 0(T ∨) → · · · . Thus H 0((bE)∨) ∼= H 0((bE/T )∨) ∼= E xt0(bE/T, OY ) where bE/T is Φ-WIT0 and pure 3-dimensional if nonzero. The rest of the lemma is clear. (cid:3) Lemma 4.21. If E ∈ T◦ X, then E satisfies the hypotheses of Lemma 4.20. Proof. Let E ∈ T◦ X . Since TX contains all the torsion sheaves and contains Φ({Coh≤0(Ys)}↑), we see that E is pure 3-dimensional and satisfies the vanish- ing Hom(Φ({Coh≤0(Ys)}↑), E) = 0. Also, since W ′ 0,X ⊂ TX by definition, we have E ∈ (W ′ 1,X . Thus E satisfies all the hypotheses of Lemma 4.20. (cid:3) 0,X )◦. From Remark 3.17, we get E ∈ W ′ Remark 4.22. For ease of reference, let us list here the conclusions of some of the threefolds. Then lemmas above. Suppose that π : X → S andbπ : Y → S are a pair of dual elliptic • Lemma 4.14: for fiber sheaves E, we have ΛE ∈ hCoh(bπ)0[−1], BY,∗[−2]i. (cid:16)Coh=2(Y ) ∩ Ψ({Coh≤0(Xs)}↑)(cid:17) [−1]. • Lemma 4.15: for E ∈ Coh≤1(X) without fiber subsheaves, we have ΛE ∈ • Lemma 4.16: for E ∈ Coh=2(X) ∩ Coh(π)1, we have ΛE ∈ hCoh=2(Y ), Coh≤2(Y )[−1], Coh≤0(Y )[−2]i. • Lemma 4.17: for E ∈ Coh=2(Y ) ∩ Coh(π)2 ∩ (Coh(π)≤1)◦, we have ΛE ∈ hCoh≥2(Y ), Coh≤2(Y )[−1], BY,∗[−2]i. • Lemma 4.18: for E ∈ W ′ 0,X ∩ Coh=3(X), we have ΛE ∈ hCoh≥2(Y ), Coh≤2(Y )[−1], BY,∗[−2]i. • Lemma 4.19: for E ∈ Φ({Coh≤0(Ys)}↑), we have ΛE ∈ hCoh=2(Y ) ∩ {Coh≤0(Ys)}↑, Coh≤1(Y )[−1], Coh≤0(Y )[−2]i. • Lemma 4.20: for E ∈ Coh=3(X) ∩ W ′ 1,X with Hom(Φ({Coh≤0(Ys)}↑), E) = 0, we have Λ(E[1]) ∈ hCoh=3(Y ), Coh≤2(Y )[−1], BY,∗[−2]i. T-STRUCTURES ON ELLIPTIC FIBRATIONS 27 Lemma 4.23. Give an exact triangle E′ → E → E′′ → E′[1] in D(X), if both ΛE′, ΛE′′ lie in the extension closure hCoh≥2(Y ), Coh≤2(Y )[−1], BY,∗[−2]i, then so does ΛE. Proof. From the long exact sequence of cohomology for the exact triangle E′ → E → E′′ → E′[1], we have the following exact sequences: 0 → Λ0E′′ →Λ0E → Λ0E′, Λ1E′′ →Λ1E → Λ1E′, and Λ2E′′ →Λ2E → Λ2E′ → 0. Since Λ0E′′, Λ0E′ ∈ Coh≥2(Y ), we have Λ0E. Also, that Λ1E′′, Λ1E′ ∈ Coh≤2(Y ) implies Λ1E since Coh≤2(Y ) is a Serre subcategory of Coh(Y ). And finally, that Λ2E′′, Λ2E′ ∈ BY,∗ implies Λ2 since BY,∗ is closed under quotients and extensions in Coh(Y ). (cid:3) Combining the above lemmas together, we now have: Theorem 4.24. Suppose X is an elliptic threefold. We have (4.15) Λ(CX ) ⊆ hCoh≥2(Y ), Coh≤2(Y )[−1], BY,∗[−2]i. Proof. We begin by showing that any sheaf in W ′ 0,X can be filtered by sheaves of the types from the various lemmas above. The idea is to use the fact that W ′ 0,X is a torsion class (hence closed under quotients in Coh(X)), and that we have the following nested sequence of Serre subcategories in Coh(X): Fix any E ∈ W ′ Coh(π)≤0 ⊂ Coh≤1(X) ⊂ Coh(π)≤1 ⊂ Coh≤2(X). 0,X . Then E fits in a short exact sequence in Coh(X) 0 → E0,0 → E → E0,1 → 0 where E0,0 ∈ Coh≤2(X) and E0,1 ∈ Coh=3(X) ∩ W ′ short exact sequence in Coh(X) 0,X . We can then fit E0,0 in a 0 → E1,0 → E0,0 → E1,1 → 0 where E1,0 ∈ Coh≤2(X) ∩ Coh(π)≤1 = Coh(π)≤1 and E1,1 lies in Coh≤2(X) ∩ (Coh(π)≤1)◦, which is the same as Coh=2(X) ∩ Coh(π)2 ∩ (Coh(π)≤1)◦. In turn, we can fit E1,0 in a short exact sequence in Coh(X) 0 → E2,0 → E1,0 → E2,1 → 0 where E2,0 ∈ Coh(π)≤1 ∩ Coh≤1(X) = Coh≤1(X) and E2,1 lies in Coh(π)≤1 ∩ (Coh≤1(X))◦, which is the same as Coh=2(X) ∩ Coh(π)1. Next, we can fit E2,0 in a short exact sequence in Coh(X) 0 → E3,0 → E2,0 → E3,1 → 0 where E3,0 ∈ Coh≤1(X) ∩ Coh(π)0 and E3,1 ∈ Coh≤1(X) ∩ (Coh(π)0)◦. Overall, we have constructed a filtration of E E3,0 ⊆ E2,0 ⊆ E1,0 ⊆ E0,0 ⊆ E where the subfactors are: • E3,0, which satisfies the hypotheses of Lemma 4.14; 28 JASON LO • E2,0/E3,0 ∼= E3,1, which satisfies the hypotheses of Lemma 4.15; • E1,0/E2,0 ∼= E2,1, which satisfies the hypotheses of Lemma 4.16; • E0,0/E1,0 ∼= E1,1, which satisfies the hypotheses of Lemma 4.17; • E/E0,0 ∼= E0,1, which satisfies the hypotheses of Lemma 4.18. As a result (see Remark 4.22, for instance), each of the subfactors of E listed above is contained in the category hCoh≥2(Y ), Coh≤2(Y )[−1], BY,∗[−2]i. (4.16) Therefore, Λ(W ′ 0,X ) is contained in the category (4.16). Besides, the two categories Λ(Φ({Coh≤0(Ys)}↑)) and Λ(T◦ X [1]) are also contained in the category (4.16) by Lemmas 4.19, 4.20 and 4.21. The inclusion (4.15) thus follows from Lemma 4.23. (cid:3) Remark 4.25. Since BY,∗ ⊆ TX , Theorem 4.24 immediately gives (4.17) Λ(CX ) ⊆ hCY , CY [−1], CY [−2]i. Now we have the following theorem, which can be considered as the analogue of Theorem 4.12 on elliptic threefolds: Theorem 4.26. Suppose X is a smooth elliptic threefold. Then the heart Λ(CX ) differs from the heart DY [−2] by one tilt. Proof. Since BY,∗ ⊆ Coh≤1(Y ), we have Coh≥2(Y ) ⊆ B◦ follows from Theorem 4.24 and [BMT, Proposition 2.3.2(b)]. Y,∗. The theorem then (cid:3) References [ABL] D. Arcara, A. Bertram and M. Lieblich, Bridgeland-stable moduli spaces for K-trivial surfaces, J. Eur. Math. Soc., Vol. 15 (1), pp. 1-38, 2013. [ARG] B. Andreas, D. Hern´andez Ruip´erez and D. S´anchez G´omez, Stable sheaves over K3 fibra- tions, Internat. J. Math., Vol. 21 (1), pp. 25-46, 2010. [BBR] C. Bartocci, U. Bruzzo, D. Hern´andez-Ruip´erez, Fourier-Mukai and Nahm Transforms in Geometry and Mathematical Physics, Progress in Mathematics, Vol. 276, Birkhauser, 2009. [BMS] A. Bayer, E. Macr`ı and P. Stellari, The space of stability conditions on abelian threefolds, and on some Calabi-Yau threefolds, preprint, 2014. arXiv:1410.1585 [math.AG] [BMT] A. Bayer, E. Macr`ı and Y. Toda, Bridgeland stability conditions on threefolds I: Bogomolov-Gieseker type inequalities, J. Algebraic Geom., Vol. 23, pp. 117-163, 2014. [Bri1] T. Bridgeland, Fourier-Mukai transforms for elliptic surfaces, J. Reine Angew. Math., Vol. 498, pp. 115-133, 1998. [Bri2] T. Bridgeland, Stability conditions on triangulated categories, Ann. Math., Vol. 166, pp. 317-345, 2007. [Bri3] T. Bridgeland, Stability conditions on K3 surfaces, Duke Math. J., Vol. 141, pp. 241-291, 2008. [BM] T. Bridgeland and A. Maciocia, Fourier-Mukai transforms for K3 and elliptic fibrations, J. Algebraic Geom., Vol. 11, pp. 629-657, 2002. [CL] W.-Y. Chuang and J. Lo, Stability and Fourier-Mukai transforms on higher dimensional elliptic fibrations, 2013. Preprint. arXiv:1307.1845 [math.AG] [CDFMR] P. Candelas, D.-E. Diaconescu, B. Florea, D. R. Morrison and G. Rajesh, Codimension-three bundle singularities in F-theory, J. High Energy Phys., Vol. 014, 2002. arXiv:hep-th/0009228v2 [FMW] R. Friedman, J. Morgan and E. Witten, Vector bundles over elliptic fibrations, J. Alge- braic Geom., Vol. 8, pp. 279-401, 1999. [HRS] D. Happel, I. Reiten and S. O. Smalø, Tilting in abelian categories and quasitilted algebras, Mem. Amer. Math. Soc., Vol. 120, 1996. [HL] D. Huybrechts and M. Lehn, The Geometry of Moduli Spaces of Sheaves, Aspects of Math- ematics, Vol. 31, Vieweg, Braunschweig, 1997. T-STRUCTURES ON ELLIPTIC FIBRATIONS 29 [Lo1] J. Lo, Stability and Fourier-Mukai transforms on elliptic fibrations, Adv. Math., Vol. 255, pp. 86-118, 2014. arXiv:1206.4281 [math.AG] [Lo2] J. Lo, Torsion pairs and filtrations in abelian categories with tilting objects, 2013. To appear in J. Algebra Appl. arXiv:1302.2991 [math.AG] [MM] A. Maciocia and C. Meachan, Rank one Bridgeland stable moduli spaces on a principally polarized Abelian surface, Int. Math. Res. Not. IMRN, Vol. 9, pp. 2054-2077, 2013. [MP1] A. Maciocia and D. Piyaratne, FourierMukai transforms and Bridgeland stability condi- tions on Abelian threefolds, preprint, 2013. arXiv:1304.3887 [math.AG] [MP2] A. Maciocia and D. Piyaratne, Fourier-Mukai Transforms and Bridgeland Stability Con- ditions on Abelian Threefolds II, preprint, 2013. arXiv:1310.0299 [math.AG] [Mac] E. Macr`ı, A generalized Bogomolov-Gieseker inequality for the three-dimensional projective space, Algebra Number Theory, Vol. 8, pp. 173-190, 2014. [Pol] A. Polishchuk, Constant families of t-structures on derived categories of coherent sheaves, Mosc. Math. J., Vol. 7, pp. 109-134, 2007. [RMS] D. Hern´andez Ruip´erez, A. C. L´opez Mart´ın and F. Sancho de Salas, Relative integral functors for singular fibrations and singular partners, J. Eur. Math. Soc., Vol. 11, pp. 597- 625, 2009. [RP] D. Hern´andez Ruip´erez and J. M. Munoz Porras, Stable sheaves on elliptic fibrations, J. Geom. Phys., Vol. 43, pp. 163-183, 2002. [Sch] B. Schmidt, A generalized Bogomolov-Gieseker inequality for the smooth quadric threefold, Bull. Lond. Math. Soc., Vol. 46 (5), pp. 915-923, 2014. [Yos1] K. Yoshioka, Moduli spaces of stable sheaves on Abelian surfaces, Math. Ann., Vol. 321, pp 817-884, 2001. [Yos2] K. Yoshioka, Perverse coherent sheaves and FourierMukai transforms on surfaces I, Kyoto J. Math., Vol. 53 (2), pp. 261-344, 2013. [Yos3] K. Yoshioka, Perverse coherent sheaves and Fourier-Mukai transforms on surfaces II, preprint. http://www.math.kobe-u.ac.jp/HOME/yoshioka/preprint/PerverseII.pdf Department of Mathematics, University of Illinois at Urbana-Champaign, 1409 W Green St, Urbana IL 61801, USA E-mail address: [email protected]
0801.0261
4
0801
2012-10-09T23:02:04
An Abelian Category of Motivic Sheaves
[ "math.AG", "math.NT" ]
The goal of this paper is to construct a category of motivic "sheaves" on an algebraic variety defined over a subfield of C, using Nori's method. This categoryis abelian and it possesses faithful exact realization functors to the categoriesof constructible sheaves for the classical and etale topologies. Moreover, there is a tannakian subcategory of motivic local systems with a realization functor into the category of variations of mixed Hodge structures. Conversely, all basic geometric examples of the latter come from this motivic category.
math.AG
math
An abelian category of motivic sheaves Donu Arapura∗ Abstract A category of motivic "sheaves" is constructed over a variety in char- acteristic 0 using Nori's method. Although the relationship with many alternative constructions remains to be clarified, it does have many of the properties one expects. For example, it is abelian and has Betti, Hodge and ℓ-adic realizations, and it has a Tannakian subcategory of motivic local systems. Contents 1 Introduction 2 Representations of graphs 2.1 Endomorphism coalgebras . . . . . . . . . . . . . . . . . . . . . . 2.2 Nori's construction . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Enriched model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Premotivic sheaves . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Constructible sheaves 3.2 Cohomology of pairs . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Premotivic sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Realizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Base change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Motivic Sheaves 4.1 Zariski Descent . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Extension by zero . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Cellular decompositions . . . . . . . . . . . . . . . . . . . . . . . 4.4 Tensor products . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ind objects ∗Partially supported by the NSF 2 7 7 9 15 16 17 17 21 22 25 27 30 30 33 35 39 42 1 5 Direct Images 5.1 Direct Images (abstract construction) . . . . . . . . . . . . . . . 5.2 Direct Images (conclusion) . . . . . . . . . . . . . . . . . . . . . . 5.3 Direct image with compact support . . . . . . . . . . . . . . . . . 6 Motivic Local Systems 6.1 Local Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Pure Objects and Weights . . . . . . . . . . . . . . . . . . . . . . 6.4 Andr´e's category of motives . . . . . . . . . . . . . . . . . . . . . 7 Nori's Hodge conjecture 7.1 Conjecture over C . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Conjecture over general bases . . . . . . . . . . . . . . . . . . . . A 2-categories B Comparison theorem C Classical t-structure for mixed Hodge modules 43 43 45 51 52 52 54 56 60 62 62 63 64 67 67 1 Introduction The basic homological invariants of a fibration of topological spaces f : X → S, are the local systems Rif∗Q. When this is a family of complex algebraic varieties defined over a subfield k of C, there are many related invariants, such as the Gauss-Manin connection, the associated variation of mixed Hodge structure, and the action of the algebraic fundamental group on ´etale cohomology of the fibres. According to Grothendieck's philosophy, all of these structures should come from the motive of the family. My goal here is to make this idea precise in the following way. Given a field F , and a variety S, as above, I will construct an abelian category M(S; F ) of motivic "sheaves" of F -modules. The above local systems can be promoted to objects in M(S; Q), and the associated structures can be obtained by applying appropriate realizations functors. Before explaining what I will do, let me say a few words about what I won't. The usual approach to building a category of motives is to start with a category of varieties and algebraic correspondences and modify and complete this in some way. This stays very close to the underlying geometry which is good. On the other hand, it is usually very hard to prove for example that what one gets is (derived from) an abelian category. A more pragmatic approach is to take a system of compatible realizations. This usually has good categorical properties, but is somewhat ad hoc in nature; and in the relative setting, it would be appear that any such approach would be necessarily very technical (e.g. [S3]). Here I want to take a middle path first blazed by Nori while building a category of motives over a field [N2, L2]. The approach appeals to a particular 2 realization at the outset, but is essentially geometric in its character. Since the construction is not that widely known, I will indicate the basic idea starting with a toy model and then refining it. In fact, one of the goals of this paper is to give an exposition of some, although not all, aspects of Nori's construction. Consider the category of k-algebraic varieties V ark. Since we assume that k ⊆ C, we may apply singular (or Betti) cohomology H i to obtain a contravariant functor from V ark to the category Q-mod of finite dimensional Q-vector spaces. The key point is that H i(X) is not just a vector space, but a module over the ring of natural transformations End(H i), or a comodule over the "dual" coalgebra End∨(H i) (section 2.1 ) which is technically better behaved. The category M′ i(k) of finite dimensional comodules over this coalgebra forms an approximation to Nori's category. It is abelian, and the objects M ∈ M′ i(k) are not too wild, in that they admit presentations of the form Mj H i(Xj) →Mk H i(Yk) → M → 0 Furthermore, each object M also carries a canonical mixed Hodge structure and (after tensoring with Qℓ) an action of Gal(¯k/k) as one would hope. So far so good, but it would be better to include the various M′ i into a single category M, so that standard exact sequences respect the M-structure. Toward this end, it is necessary to modify the basic construction by incorporating boundary operators into the foundations. Thus instead of starting with V ark, the source category ∆ consists of triples (X, Y ⊂ X, i ∈ N) and the appropriate notion of morphism, which includes abstract boundary maps (X, Y, i) → (Y, ∅, i − 1). This is really a partial category in the sense that the composition law is only partially defined; nevertheless the basic constructions go though. The category of comodules over End∨(H), where H : ∆ → Q-mod is the functor sending (X, Y, i) 7→ H i(X, Y ), yields a rational version of Nori's category of effective cohomological motives. Following the usual pattern, the category M(k) is obtained by inverting the Tate motive. This step can be built in from the beginning, and we find it convenient to do so. Now turning to the general case, the building blocks for M(S; F ) are quadru- ples consisting of a quasiprojective family f : X → S, a closed subvariety Y ⊂ X and indices i ∈ N, w ∈ Z. This is subject to a further technical admissibility con- dition (definition 3.2.1) which will be satisfied if f is projective. When Y = ∅, this data represents the motivic version of Rif∗F (w) denoted here by hi S(X)(w). The parameter w keeps track of Tate twists, which although extraneous for or- dinary sheaves are nontrivial in the Hodge and ´etale realizations. For nonempty Y , the associated motive hi S(X, Y )(w) roughly corresponds to the fiberwise co- homology of the pair. In essence, M(S; F ) is set up as the universal theory for which: (M1) M(S; F ) is an F -linear abelian category with a faithful exact functor RB to the category of sheaves of F -modules on S with its classical topology. (M2) A morphism X ′ → X over S, taking Y ′ to Y would give rise to a morphism 3 S(X, Y )(w) → hi S(X ′, Y ′)(w) compatible with the usual pullback map of hi under RB. (M3) Whenever Z ⊆ Y ⊆ X, there are connecting morphisms hi S(X, Y )(w) → hi+1 S (Y, Z)(w) compatible with the usual pullback maps. S (X × P1, X × {0} ∪ Y × P1)(w) ∼= hi S(X, Y )(w − 1). (M4) hi+2 (M5) Objects and morphisms of M(S; F ) can be patched on a Zariski open cover. The actual construction is obtained by modifying the framework discussed in the previous paragraph. Given stratification Σ and a collection of base points on the strata, let ∆(Σ) be the collection of quadruples such that the cohomology sheaf is constructible with respect to Σ. We can make this into a partial category by adding morphisms corresponding to items (M2), (M3) and (M4). The functor HΣ : ∆(Σ) → Q-mod is defined by sending (X, Y, i, w) to the product H i(Xs, Ys) at the given set of base points. The category PM(S, Σ; F ) of Σ-constructible premotivic sheaves is constructed explicitly as the category of comodules over End∨(HΣ). Note that contrary to initial appearances, this is not simply a product of M(k) over the base points because ∆(Σ) does not decompose this way (see example 3.5.3). The trivial exception is when S is a finite set of points. The category PM(S; F ) of premotivic sheaves is given as the direct limit of these categories as Σ gets finer. This will satisfy (M1) to (M4). The category M(S; F ) is obtained from PM(S; F ) by forcing (M5) by passing to the associated stack. This last step can be made explicit. In fact a weak form of (M5) holds for PM. So it is not quite clear to me whether this axiom is redundant, nevertheless it is included for completeness. Here are the precise properties: Theorem 1.0.1. To every k-variety, there is an F -linear abelian category M(S; F ) such that: 1. These are defined over the prime field F0, i.e. M(S, F ) ∼= M(S, F0)⊗F0 F . 2. There is an exact Betti realization functor RB : M(S; F ) → Cons(San, F ) to the category of constructible sheaves of F -modules for the classical topol- ogy. 3. There is an exact Hodge realization functor RH : M(S; Q) → Cons-M HM (S) ⊂ DbM HM (S) to the heart of the classical t-structure of the derived category mixed Hodge modules (see appendix C). 4 4. There is an exact ´etale realization functor Ret : M(S; F ) → Cons(Set, F ) to the category of constructible sheaves of F -modules for the ´etale topology. (In this case, F should be finite or Qℓ.) 5. When f satisfies a suitable admissibility condition (of being controlled in the sense of definition 3.2.1), there exist motives in M(S; F ) correspond- ing to Rif∗F (n) under realization. 6. There are inverse images compatible with realizations. 7. There are higher direct images for projective or constant maps compatible with realizations. Many of the above items are formal consequences of the definitions, but the last is not. The construction of direct images is technically the most difficult part of this paper. General arguments give the existence of an adjoint to inverse image which ought to play the role of the direct image. Proving that this has reasonable properties for projective maps requires work, which uses a refinement of the method of [Ar]. This earlier paper was really the starting point for this entire project. This ultimately hinges on Nori's insight that Beilinson's "basic lemma" can be used to construct cell decompositions which reduce the homolog- ical complexities. In the relative setting, there are few additional complications. For instance, these decompositions are only obtained locally over the base, so patching issues of the sort given in (M5) comes into play. But modulo these technicalities, the basic strategy of using cell decompositions does work. The objects of M(S) play the role of constructible sheaves. Inside this, we have a subcategory of "local systems" arising from particularly nice families (X → S, Y, i, w). The precise condition is that X can be completed to a smooth projective map so that Y together with the boundary is a divisor with relative normal crossings. These enjoy the following good properties. Theorem 1.0.2. There is an abelian full subcategory Mls(S; Q) ⊂ M(S, Q) of motivic local systems such that: 1. The images of Mls(S; Q) ( respectively Mls(S; Qℓ)) under RB (respec- tively Ret) is contained in the category of locally constant (respectively lisse sheaves). The image under RH is contained in the category V M HS(San) of admissible variations of mixed Hodge structures. 2. There are tensor products on Mls(S; F ) compatible with realizations. With this structure it is a Tannakian category. 3. The subcategory Mpure(S, Q) ⊂ Mls(S, Q) generated by smooth projective families is a semisimple Tannakian category. 4. Objects in Mls(S, Q) carry a weight filtration such that the associated graded objects are pure. 5 A number of the arguments again rely on the existence of cellular decom- positions. Regarding item 4, I do not have a good notion of weight in M(S) at present. I expect that it would require the development of an analogue of perverse sheaf in DbM(S), since the pure objects are almost surely of this form. A natural question, that is only partially solved here, is the relation of this approach to motives to the others. Andr´e [A1] defines the class of motivated cycles on smooth projective variety to be the cycles which would be expected to be algebraic assuming Grothendieck's standard conjectures. Andr´e showed that the category of pure motives over a field constructed with such correspondences has all the expected properties. This construction can be extended to more general bases without much difficulty [AD]. We show that this category is precisely Mpure(S). In his unpublished work, Nori has constructed a functor from Voevodsky's category Dgm(k) to DbM(k). It seems reasonable to expect that this generalizes over a base, but such matters will be postponed for the future. In the final section, I discuss Nori's Hodge conjecture which says that M(C) embeds fully faithfully into the category of mixed Hodge structures. This would imply that the canonical mixed Hodge structure on cohomology is "Galois invariant". The relative case can be reduced to this by rather formal argument involving direct image and restriction functors. Acknowledgement: I would like to thank Madhav Nori for giving me his permission to include some of his beautiful constructions (of course, I take re- sponsibility for errors). Also thanks to M. de Cataldo, F. D´eglise, M. Levine, J. Lipman, D. Patel, M. Saito and J. Schurmann for some useful conversations. This research was started at the Max Planck Institute in the fall of 2007 and continued at the Tata Institute in the winter of 2008; I am grateful to these institutes as well as the NSF for their support. Notation: Since the notation will tend to get rather heavy, I will routinely suppress subscripts, superscripts and others symbols whenever they can be un- derstood from context. Given a ring R, let R-Mod (respectively R-mod) stand for the category of (finitely presented) left R-modules. Fix a field k embeddable into C and another field F . For most of the paper, I will work with a fixed em- bedding ι : k ֒→ C. A k-variety is simply a reduced separated k-scheme of finite type. Let V ark be the category of these. Given a k-variety X, the word point x ∈ X generally refers to a ¯k-rational point. I will denote the analytic space (X ×ι Spec C)an by Xι,an or Xan or sometimes just X, in keeping with the pre- vious comment regarding notation. A quasi-projective morphism is a morphism which can be factored as a composition of an open immersion followed by a pro- jective morphism. I will usually write H i(X; F ) for H i(Xι,an; F ). Given a map f : X → S of spaces and a sheaf F on X, I will often denote the higher direct image Rif∗F by H i S(X, F ). Since this will never be used to denote cohomology with support in this paper, there should be no danger of confusion. 6 2 Representations of graphs 2.1 Endomorphism coalgebras In the next couple of sections, we set up the basic foundation for the rest of the paper. Let F be a field. Suppose we are given an F -linear abelian category A with an exact faithful embedding H into the category of finite dimensional vector spaces F -mod. Then the ring End(H) = EndF (H), of F -linear natural transformations of H to itself, will act naturally on H(A) for any A ∈ ObA. This suggests that one might be able to reconstruct A as the category of finite dimensional modules over this ring. However, this does not generally work (example 2.2.8). The right thing to consider is the category of comodules over the dual object End∨(H) whose construction we learned from [JS]. Before getting into the construction, we should explain how to characterize it. Given a commutative F -algebra R, we can form new category A ⊗ R with the same objects as A, but HomA⊗R(−, −) = HomA(−, −) ⊗ R. The functor H extends to an R-linear functor H ⊗ R : A ⊗ R → R-mod. In this way, we have an algebra valued functor R 7→ EndR(H ⊗ R). Lemma 2.1.1. This functor is represented by a colagebra End∨(H), i.e. HomF (End∨(H), R) ∼= EndR(H ⊗ R) This implies that End∨(H)∗ = End(H), but usually End(H)∗ 6= End∨(H). Nevertheless, most of the statements become easier to follow if one formulates them for End(H) and dualizes. The lemma tells us how to make sense of this. Note that we can express End∨(H) or any coalgebra as a directed union of finite dimensional subcoalgebras ∪Ei. Thus the correct dual object to End∨(H) is not End(H) but the pro-algebra " lim←− "E∗ i . Moreover, A can described as 2- colimit of the categories of Ei-modules. We will find this viewpoint convenient later on, but for the moment, it seems simpler to work with the coalgebra. Given pair of functors G, H : C → D, with D F -linear, Hom(G, H) is an F -vector space. More explicitly, we can identify Hom(G, H) with ker[ YM∈ObC Hom(G(M ), H(M )) −→ Yf :N→P ∈MorC Hom(G(N ), H(P ))] (1) where the map takes the collection (ηM )M to (H(f ) ◦ ηN − ηP ◦ G(f ))f . Compo- sition makes End(H) = Hom(H, H) into an algebra as noted above. Following [JS], it is convenient to introduce a smaller "predual" object, which means that Hom∨(G, H)∗ = Hom(G, H). Let F -Lin be the collection of F -linear abelian categories with finite dimen- sional Hom's. Suppose that we now have a pair of functors G, H : C → D, with D ∈ F -Lin. Define Hom∨(G, H) to be the cokernel of Mf :N→P ∈C Hom(G(N ), H(P ))∗ S−→ MM∈ObC Hom(G(M ), H(M ))∗ (2) 7 where the map S is defined so that Hom∨(G, H)∗ = Hom(G, H). More ex- plicitly, S sends η∗ N ∈ Hom(G(N ), H(N ))∗ plus P ∈ Hom(G(P ), H(P ))∗ where η∗ f ∈ Hom(G(N ), H(P ))∗ to η∗ hη∗ N , ηN i = hη∗ P , ηP i = −hη∗ hη∗ f , H(f ) ◦ ηN i f , ηP ◦ G(f )i Upon setting End∨(H) := Hom∨(H, H), we see that this satisfies lemma 2.1.1 as a vector space, and we can use this formula to define the colagebra structure. However, it will be useful to describe this more explicitly. The sum of the maps End(H(M ))∗ → F, dual to the identity, is easily seen to factor through End∨(H) and this defines the counit 1∨ H : End∨(H) → F Given functors G, H, L : C → D with D ∈ F -Lin we have a comultiplication ◦∨ : Hom∨(G, L) → Hom∨(G, H) ⊗ Hom∨(H, L) dual to composition ◦. More precisely, ◦ is given by product of compositons cM : Hom(G(M ), H(M )) ⊗ Hom(H(M ), L(M )) → Hom(G(M ), L(M )) Then ◦∨ is given by the sum of the dual maps c∗ M Hom(G(M ), L(M ))∗ → Hom(G(M ), H(M ))∗ ⊗ Hom(H(M ), L(M ))∗ (3) Given G, G′ : C → D and H, H ′ : D → E with D, E ∈ F -Lin, there is composi- ton ⋄∨ : Hom∨(G′ ◦ G, H ′ ◦ H) → Hom∨(G, H) ⊗ Hom∨(G′, H ′) dual to the operation ⋄ defined in appendix A. The operation ⋄ is a product of maps dM : Hom(G(M ), H(M ))×Hom(G′(H(M )), H ′(H(M ))) → Hom(G′◦G(M ), H ′◦H(M )) and ⋄∨ = P d∗ following duality principle: M . To simplify arguments with these operations, we use the Lemma 2.1.2. Suppose we are given an identity in +, ◦, ⋄, 1G, which amounts to the commutivity of a finite diagram with arrows labelled by these opera- tions. Then the dual identity, obtained by reversing arrows and relabelling by +, ◦∨, ⋄∨, 1∨ G, also holds. Proof. Suppose we have a finite diagram with vertices given as finite tensor products of spaces Hom∨(−, −), and edges labelled by +, ◦∨, ⋄∨, 1∨ G. Then commutivity can be established by chasing elements. Given an element of one of the vertices, we can find a subdiagram of finite dimensional vector spaces which contains it. Duality for finite dimensional vector spaces implies that the commutivity of the subdiagram would then follow from commutivity of the dual diagram. 8 Using this principle, we can see that: Lemma 2.1.3. Given composable functors H and G 1. End∨(G) is a colagebra over F with respect to ◦∨, 1∨. 2. The map p given by End∨(H ◦ G) p End∨(G) ⋄∨ 1∨⊗1 6❧❧❧❧❧❧❧❧❧❧❧❧❧ End∨(H) ⊗ End∨(G) is a colagebra homomorphism. Proof. The first part is clear, since the dual statement is that End(G) is an algebra. For the second, we have to establish that p preserves comultiplication. Dually, by identities given in the appendix, 1 ⋄ (α ◦ β) = (1 ◦ 1) ⋄ (α ◦ β) = (1 ⋄ α) ◦ (1 ⋄ β) One can now readily verify lemma 2.1.1. We also leave the formulation and proof of the corresponding statement for Hom∨ to the reader. 2.2 Nori's construction Any category can be regarded as a directed graph (or diagram in Nori's termi- nology) by forgetting the composition law. This forgetful functor admits a left adjoint: given a directed graph ∆, we can form a category P aths(∆), whose objects are vertices of ∆ and morphisms are finite (possibly empty) connected paths between vertices. The adjointness amounts to the obvious fact that given a graph ∆ and a category C, there is a one to one correspondence between graph morphisms ∆ → C and functors P aths(∆) → C. In view of this, we may apply category theoretic terminology and results to directed graphs. Let H : ∆ → F -mod be a functor, i.e. a quiver. We can define End∨(H) by the formula (2), which simplifies to coker[ Mf :N→P ∈Mor∆ Hom(H(P ), H(N )) S−→ MM∈Ob∆ End(H(M ))] (4) where S takes ηf ∈ Hom(H(P ), H(N )) to the difference of ηf ◦H(f ) ∈ End(H(N )) and H(f ) ◦ ηf ∈ End(H(P )). We note the following, which is easily checked. Lemma 2.2.1. 9 / /   6 1. The collection of functors from graphs to F -mod forms a category where the morphisms are commutative diagrams H / F -mod H′ ;✇✇✇✇✇✇✇✇✇ ∆ π ∆′ 2. End∨(H) is isomorphic to End∨( H), where H is the extension of H to P aths(∆). 3. The assignment (∆, H) 7→ End∨(H) is functorial. In particular, there is an induced map End∨(H) → End∨(H ′) of colagebras where H and H ′ are as in 1. 4. If ∆ is a category then End∨(H) ∼= End∨(H ′), where H ′ is the induced functor on the category H(∆) with the same objects as ∆ but morphisms given by its image under H. 5. The functor (∆, H) 7→ End∨(H) preserves finite coproducts. In more explicit terms, if ∆ decomposes into a disjoint union of ∆1` ∆2, then End∨(H) = End∨(H∆1) × End∨(H∆2) (which is the coproduct of coal- gebras). Proof. The first statement is clear. For the second, we have that End∨(H) and End∨( H) are the quotients of L End(H(M )) by and IH = S( Mf ∈Mor∆ I H = S( Mf ∈MorP aths(∆) Hom(H(P ), H(N ))) Hom(H(P ), H(N ))) respectively. Clearly IH ⊆ I H . So we have to check the reverse inclusion. We first note that S(1) = 0, so it suffices to check that S(ηf1...fn ) ∈ IH for n ≥ 2. For n = 2, this follows from the identity S(ηf1f2 ) = S(ηf1f2 ◦ H(f1)) + S(H(f2) ◦ ηf1f2 ) ∈ IH The general case is similar. Although the third statement is similar to lemma 2.1.3, the previous for- malism will not apply without modification. So it is easier to prove directly. An element of End∨(H) is represented by a finite sum P hM of elements hM ∈ End(H(M ))∗. Define π(hM ) = hπ(M) ∈ End(H ′(M ))∗. To see that 10 /   ; this is compatible with comultiplication ◦∨, observe that ◦∨(hM ) = c∗ where c∗ M is given in (3). Then M (hM ), π(◦∨(XM hM )) = π(X c∗ M (hM )) =X c∗ π(M)(hπ(M)) = ◦∨(π(XM hM )) The fourth and fifth statement follows immediately from the formulas (2) and (4). We let End∨(H)-comod denote the category of right comodules over this coalgebra in F -mod. Corollary 2.2.2. A morphism (∆, H) → (∆′, H ′) as above induces a faithful exact functor End∨(H)-comod → End∨(H ′)-comod. Proof. This isn't so much a corollary as a statement of the fact that both cate- gories can be viewed as subcategories of F -mod. We can therefore view End∨(H)-comod as a subcategory of End∨(H ′)-comod. We will often apply this, without comment, when ∆ ⊂ ∆′ is a subgraph and H is the restriction of H ′. Corollary 2.2.3. If H ′ : ∆ → F -mod is another functor with a natural isomor- phism Γ : H → H ′, then End∨(H)-comod and End∨(H ′)-comod are isomorphic. Corollary 2.2.4. Let π : ∆ → ∆ be a morphism of graphs such that it is surjective on objects and such that every fiber is connected. Then End∨(H) ∼= End∨(H ◦ π). Proof. The assumption guarantees that H(P aths(∆)) and H ◦ π(P aths( ∆)) are equivalent. Given M ∈ Ob∆, H(M ) is naturally a left End(H(M ))-module, and hence by transpose an End(H(M ))∗-comodule. Via the map End(H(M ))∗ → End∨(H), M becomes a End∨(H)-comodule, which we usually denote by h(M ). This is a functor ∆ → End(H)-comod. The structure of a general comodule is clarified by the following. Lemma 2.2.5. Any object V of End∨(H)-comod fits into an exact sequence h(Mi) → m Mi=1 n Mj=1 h(Nj) → V → 0 for some Mi, Nj ∈ Ob∆. Proof. The lemma will follow from the claim that any comodule is the image of finite sum of the form L h(Mi). Set E∨(D) = End∨(HD) for any subgraph. When D is finite, E∨(D) is quotient of a finite sum of comodules of the form End(H(M )) ∼= H(M )dim H(M), so the claim follows when V = E∨(D). In 11 general, the matrix coefficients of the E∨(∆) coaction on V lie in some E∨(D) with D finite. Thus V has a quotient of a finite sum of copies of E∨(D). So the claim holds in general. Remark 2.2.6. There is a dual description of End∨(H)-comod which is closer to what Nori originally used [L2, N2]. As in the previous argument, we can express E-comod = ∪E∨(D) as D ⊂ ∆ runs over finite subgraphs. Therefore as explained in appendix A, we have equivalences End∨(H)-comod ∼ 2- lim−→ ∼ 2- lim−→ D D E∨(D)-comod End(HD)-mod Theorem 2.2.7. If U : A → F -mod is an exact faithful F -linear functor on an F -linear abelian category, then A is equivalent to End∨(U )-comod. Proof. The proof given in [JS, §7, thm 3] for the complex field works in general. It is instructive to observe that the corresponding statement for End(U )-mod will usually fail. Example 2.2.8. Let A be the category of finite dimensional Z-graded C-vector spaces. This can be identified with the category of comodules over End∨(U ) = is much bigger. For example, C with the End(U )-action arising from a nontriv- C[T, T −1] in the usual way. However, the category of End(U ) =QZ C modules ial ultraproduct End(U ) → (Q C/U) ∼= C gives an End(U )-module which does not arise from a graded module. We can use this theorem to deduce a version of Nori's Tannakian theorem. (The original statement, which is stronger, can be found in [Br, thm 1], [L2, §3.3] or [N2].) Corollary 2.2.9 (Nori). Suppose that A is an F -linear abelian category equipped with a faithful exact functor U : A → F -mod. If G : ∆ → A is a morphism of directed graphs such that H is equivalent to U ◦ G, then there is a functor G : End∨(H)-comod → A (called the extension of G) rendering the diagram ∆ h G H 7♥♥♥♥♥♥♥ 'PPPPPPPPPPPPP G A U End∨(H)-comod /❴❴❴ / F -mod U commutative up to natural equivalence. It is convenient to prove a slightly stronger statement, where F -mod is re- placed by, for example, the category of finite dimensional modules over an F - algebra. 12 / /   '   / 7 / Corollary 2.2.10. Let R be an F -linear abelian category with a faithful exact functor ρ : R → F -mod. Suppose that H : ∆ → F -mod factors as H1 ◦ ρ (up to natural equivalence). Suppose that A is an F -linear abelian category equipped with a faithful exact functor U : A → R. If G : ∆ → A is a morphism of directed graphs such that H1 is equivalent to U ◦ G, then there are functors End∨(H)-comod → R, G : End∨(H)-comod → A rendering the diagram End∨(H)-comod ∆ h G H1 G 7♥♥♥♥♥♥♥ 'PPPPPPPPPPPPPPP 'PPPPPPPPPPP /❴❴❴❴ U A U R ρ F -mod commutative up to natural equivalence. Proof. We obtain a commutative diagram of coalgebras End∨(H) End∨(U ◦ ρ) w♦♦♦♦♦♦♦♦♦♦♦ End∨(ρ) Thus we get a functor between the categories of finite dimensional comodules End∨(H)-comod G / End∨(U ◦ ρ)-comod ∼ A t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ End∨(ρ)-comod ∼ R The equivalences of categories, indicated by ∼, follow from the theorem and the above assumptions. There is also a naturally statement, which we give only in the situation of corollary 2.2.9. Corollary 2.2.11. Given a diagram ∆ G / A π U H )❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘ #●●●●●●●●● 6❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧ ;✇✇✇✇✇✇✇✇✇ H′ U ′ A′ ∆′ G′ F -mod 13 / /   '   / 7 '   / /   w /   t   / ) #   / / 6 ; which commutes up to natural isomorphism, the diagram End∨(H)-comod G / A π End∨(H ′)-comod G′ / A′ commutes up to natural isomorphism. The case of particular interest to us in corollary 2.2.10 is the category R = (F -mod)n of finite dimensional vector spaces admitting gradings of the form V = V1 ⊕ V2 ⊕ . . . Vn. This can be identified with the category of finite modules over the ring F n. A natural example arises as follows. Given functors Hi : ∆i → F -mod, we can define a new functor H1 × H2 × . . . Hn on the cartesian product ∆1 × ∆2 × . . . ∆n in the category of graphs, to (F -mod)n by H1 × H2 × . . . Hn(M1, . . . Mn) = H1(M1) ⊕ . . . Hn(Mn) We have an induced functor End∨(H1 × H2 × . . . Hn)-comod →Yi (End∨(Hi)-comod) (5) where to be clear, on the left we really mean End∨(ρ ◦ (H1 × . . . Hn)), where ρ : (F -mod)n → F -mod is the forgetful functor. We will need a criterion for when this is an equivalence. It usually isn't. Example 2.2.12. Let ∆ be a graph consisting of a single vertex pt and no morphisms. Let H(pt) = F . Since this is a finite graph, we can work with en- domorphism rings rather than coalgebras. One has End(H) = F and End(H × H) = M2(F ) the ring of 2 × 2 matrices. Therefore by Morita's theorem, End(H ×H)-mod ∼ F -mod ≁ (F -mod)2. The natural map End(H ×H)-mod → (F -mod)2 is the diagonal embedding F -mod → (F -mod)2. Lemma 2.2.13. Suppose that each object in ∆i has maps to an object ∅i sat- isfying Hi(∅i) = 0. Then (5) is an equivalence. Proof. It is enough to check this for n = 2 graphs. By taking limits, we can reduce to the case where ∆i are both finite. From equation (1), the ring End(H1 × H2) consists of families (fP1,P2) ∈Y End(H1(P1) ⊕ H2(P2)) compatible with composition along morphisms. Choose maps τi : Pi → ∅i. By considering compatibility along the morphisms 1 × τ2 : (P1, P2) → (P1, ∅2) and τ1 × 1 : (P1, P2) → (∅1, P2), we see that fP1,P2 must be of the form (cid:18)fP1,∅2 0 0 f∅1,P2(cid:19). Thus End(H1 × H2) = End(H1) × End(H2) 14 /     / 2.3 Enriched model Let H : ∆ → F -mod be a functor on a graph as above. Although functors on End∨(H)-comod can be constructed with the help of corollary 2.2.10, it is sometimes difficult to apply. It will be convenient to give an alternative de- scription (up to equivalence) of End∨(H)-comod which allows us to incorporate extra structure. Fix a finite dimensional commutative F -algebra with an alge- bra homomorphism p : R → F such that F is flat over R. These rather strong assumptions are valid in the case of principal interest to us, where R = F × F with p projection onto the first factor. Suppose that H # : ∆ → R-mod is a functor. We can define the R-coalgebra End∨ R(H #) by replacing Hom by HomR in (4). This is not the same as the coalgebra End∨(H # ◦ ρ) considered earlier. Let End∨ R(H #)-comod denote the R-linear category of comodules in R-mod. Then we wish to describe the relationship between End∨(H #)-comod and End∨(H)-comod. First, we make a brief digression. Given an F -linear abelian category C, an ideal I is a collection of subspaces I(c1, c2) ⊆ Hom(c1, c2) such that Hom(c2, c3) ◦ I(c1, c2) ⊆ I(c1, c3), I(c1, c2) ◦ Hom(c0, c1) ⊆ I(c0, c2) Given an ideal I, C/I is the category with the same objects as C and HomC/I (c1, c2) = HomC (c1, c2)/I(c1, c2) For example, if G : C → D is an exact functor, ker G = {f ∈ M orC G(f ) = 0} is an ideal. Note, however, that the quotient C/ ker G should not be confused with the quotient of C by the thick subcategory generated by {c ∈ ObC G(c) = 0}. Lemma 2.3.1. If G : C → D is an exact functor between essentially small abelian categories such that (a) G is essentially surjective. (b) G is surjective on Homs. Then D is equivalent to C/ ker G. Proof. G induces an equivalence C/ ker G ∼ D since HomC (c, c′)/ ker G ∼= HomD(G(c), G(c′)) Returning to the set up describe earlier. We have an isomorphism p : End∨(H #) ⊗R F ∼= End∨(H) and hence an exact functor p : End∨(H #)-comod → End∨(H)-comod given by M 7→ M ⊗R F . The conditions of the above lemma are easily verified in general. In the case, where R = F 2, this is almost immediate. End∨(H #)- comodule decomposes into a sum of two factors corresponding to the idempo- tents of R, and p is projection on the first factor. Thus: 15 Corollary 2.3.2. End∨(H)-comod is equivalent to End∨(H #)-comod/ ker p. Corollary 2.3.3. An F -linear functor on End∨ maps to zero, induces a functor on End∨(H)-comod. R(H #)-comod such that ker(p) 2.4 Products We need to incorporate tensor products into our story. The category of functors from graphs to F -mod forms a category with tensor product given as follows. Let H : ∆ → F -mod and H ′ : ∆′ → F -mod be two such functors. Then H ⊗ H ′ : ∆ × ∆′ → F -mod is given by (M, N ) 7→ H(M ) ⊗ H ′(N ). The one point graph {∗} with ∗ 7→ F gives the unit making this into a tensor category, where for our purposes a tensor category over F is an F -linear additive category with a bilinear symmetric monoidal structure. We have End∨(H ⊗ H ′) ∼= End∨(H) ⊗ End∨(H ′) [JS, §8, prop 1]. This yields a product End∨(H)-comod × End∨(H ′)-comod → End∨(H ⊗ H ′)-comod When H = H ′ is equipped with a symmetric associative pairing H ⊗H → H and a unit ∗ ∈ Ob∆, H(∗) = F . Then End(H) becomes a commutative bialgebra. Thus End(H)-comod becomes a tensor category with a tensor preserving functor to F -mod given by the forgetful functor. With minor modifications to the proof of corollary 2.2.9, we have Corollary 2.4.1. Suppose that H has a product as above. If in the hypothesis of corollary 2.2.10, R = F -mod, A is an F -linear abelian tensor category, and the functors ρ, U, G are product preserving. Then G is also product preserving. Recall [De3, sect 2] [L1, chap IV, sect 1] that a dual of an object M in a tensor category, with unit 1, is an object M ∨ equipped with morphisms δ : 1 → M ∨ ⊗ M and ǫ : M ⊗ M ∨ → 1 such that the compositions M id⊗δ−→ M ⊗ M ∨ ⊗ M ǫ⊗id−→ M M ∨ δ⊗id−→ M ∨ ⊗ M ⊗ M ∨ id⊗ǫ−→ M ∨ (D1) (D2) yield the identities. Alternatively, M ∨ is characterized by the natural isomor- phisms Hom(X ⊗ M, Y ) ∼= Hom(X, M ∨ ⊗ Y ) Hom(X ⊗ M ∨, Y ) ∼= Hom(X, M ⊗ Y ) (D3) (D4) In particular, the dual is unique up to isomorphism if it exists. A map f : M → N yields a dual or transpose map f ∨ : N ∨ → M ∨ if M, N both possess duals. 16 Lemma 2.4.2. Given an exact sequence M1 → M2 → M3 → 0 i exists for i = 1, 2 then M ∨ 2 → M ∨ 3 = ker(M ∨ if M ∨ Proof. Set M ∨ diagram 3 exists. 1 ). Condition (D3) is a consequence of the 0 0 / Hom(X ⊗ M3, Y ) Hom(X ⊗ M2, Y ) Hom(X ⊗ M1, Y ) ∼= ∼= ∼= / Hom(X, M ∨ 3 ⊗ Y ) / Hom(X, M ∨ 2 ⊗ Y ) / Hom(X, M ∨ 1 ⊗ Y ) and (D4) is similar. A neutral Tannakian category over F is an abelian tensor category over F , with a faithful exact tensor preserving functor to F -mod, such that every object possesses a dual. Such a category can be realized as the category of comodules over a commutative Hopf algebra. Proposition 2.4.3. Suppose that H : ∆ → F -mod is equipped with a symmetric associative product as above. Assume that for every object M ∈ Ob∆, h(M ) has a dual. Then End∨(H)-comod is neutral Tannakian. Proof. The proposition follows from lemmas 2.2.5 and 2.4.2 3 Premotivic sheaves 3.1 Constructible sheaves We recall: Definition 3.1.1. If X is a complex variety (defined over k ⊂ C), a sheaf F on Xan is called constructible (or k-constructible) if it has finite dimensional stalks and there exists a partition Σ of X into Zariski locally closed (defined over k) so that Fσ are locally constant for each σ ∈ Σ. In this case, F is also called Σ-constructible. Let Cons(X) or Cons(X, Σ) denote the category of these. The definition of constructibility for sheaves on the ´etale topology Xet is similar [Mi, chap V], [SGA4, exp IX]. Basic examples of constructible sheaves in- clude the direct images Rif∗F and more generally direct images of constructible sheaves [V1, cor. 2.4.2]. We give a slight refinement below (theorem 3.1.10). Definition 3.1.2. Given a morphism f : X → S and a sheaf F on Xan or Xet, we say that H i S(X, F ) = Rif∗F commutes with base change if for any quasi-projective morphism g : T → S the canonical base change map g∗H i S(F ) → H i T (X ×S T, g∗F ) is an isomorphism. 17 / / /   ✤ ✤ ✤ / /     / / / Definition 3.1.3. Given a morphism f : X → S and a sheaf F on X, if H i S(X, F ) commutes with base change for all i, we will say that F has the base change property (with respect to f ). The condition implies that H i S(X, F )s → H i(Xs, FXs ) is an isomorphism for every s ∈ S. Open immersions X → S give examples where this will fail for s ∈ S − X. We review some (known) criteria for this to hold. A morphism f : X → S will be called locally trivial if it is a topological (although not necessarily an analytic) locally trivial fibre bundle with respect to the the analytic topology. More generally: Definition 3.1.4. Let say that the pair (f : X → S, F ∈ Cons(X)) is locally trivial if there exists an open cover {Ui} of S and a stratified space Φ with a constructible sheaf G, such that there are stratified homeomorphisms f −1Ui ∼= Φ × Ui compatible with f such that G pulls back to the restriction of F . Theorem 3.1.5. 1. Given a short exact sequence 0 → F1 → F2 → F3 → 0 if any two of the Fi have the base change property, then so does the third. 2. If (f : X → S, F ) is locally trivial, then F has the base change property. 3. (Proper base change) If f : X → S is proper, then any sheaf has the base change property. 4. (Locally trivial base change) If T → S is locally trivial, then the base map for f : X → S with respect to T is an isomorphism for any F and i. Proof. The first statement is obvious. The second is clear once we observe that it can be reduced to the case of a product S × Φ → S, with F pulled back from Φ. For the third, when f : X → S is proper and T is a point, the base change property follows from [I, thm 6.2]. Therefore g∗H i S(F ) → H i T (X ×S T, g∗F ) is an isomorphism on stalks. For the fourth statement, we can reduce to the case of product, and then apply the Kunneth formula. We can combine these criteria into one convenient notion. 18 Definition 3.1.6. Given a quasi-projective morphism f : X → S and a sheaf F ∈ Cons(X), we will say that the pair (f, F ) is controlled, or that f is con- trolled with respect to F , if there exists a commutative diagram S = / S f 7♦♦♦♦♦♦♦♦♦♦♦♦♦♦ ?⑦⑦⑦⑦⑦⑦⑦ T h ¯g X g j ¯X such that h and ¯g are projective, j is an open immersion, and such that ( ¯X, j!F ) is locally trivial over T . It is worth observing that the condition is automatic if S is point because everything is locally trivial over a point. Also note that in general the conditions imply that ( ¯X, ¯X − X) is a relative fibre bundle over T . Such a diagram, which need not be unique, will be called a control diagram for the pair (f, F ). Lemma 3.1.7. If (f : X → S, F ) is controlled, then F has the base change property with respect to f . Proof. Choose a control diagram as above. Let q : S′ → S be a morphism, and consider the diagram X q2 X ′ g g′ / T q1 T ′ h h′ / S q S′ where the squares are Cartesian. We have to prove that q∗Ri(h ◦ g)∗F ∼= Ri(h ◦ g)∗F . It is enough to check isomorphism on the E2 terms of Leray spectral sequence. We have q∗Rah∗Rbg∗F ∼= Rah′ ∗q∗ 1Rbg∗F because h is proper, and we have Rah′ ∗q∗ 1Rbg∗F ∼= Rah∗Rbg′ ∗q∗ 2F because g is locally trivial with respect to F . Proposition 3.1.8. Suppose that (f : X → S, F ) is a controlled pair with a locally closed S-embedding X → PN × S. Given a nonempty Zariski open set P ⊂ PN , there exists a Zariski open cover {Sα} of S and elements Hα ∈ P , such that (Hα ∩ f −1Sα → Sα, FHα ) are controlled, where qα : f −1Sα − Hα → f −1Sα is the inclusion. (f −1Sα → Sα, qα!q∗ αF ) 19   / /   7 / ? / / / / O O / / O O O O Proof. Choose a control diagram X g T / S h ?⑦⑦⑦⑦⑦⑦⑦ ¯g j ¯X for F . Then by assumption, j!F is construcible with respect to a stratification { ¯X•} of ¯X which is locally trivial over S. Given s ∈ S, we may choose H ∈ P transverse to ¯X• ∩ ¯Xs. It remains transverse to ¯X• ∩ ¯Xt, for t in a neighbourhood Ss of s. It follows that the stratification generated by ¯X• and H and ¯X• ∩ H are locally trivial over Ss. In order to proceed, we will need Whitney stratifications. For our purposes a stratification of a variety X is a finite partition Σ of X into Zariski locally closed sets such that the closure of any stratum σ ∈ Σ is a union of strata. When X is complex, we will say that this is Whitney if any stratum is smooth and if the Whitney conditions hold for any x ∈ σ ⊂ ¯τ [Li1, Mr, T, V2]. This means that given sequences xi ∈ σ and yi ∈ τ , both converging to x, the limit of the secants xiyi (with respect to a local embedding into CN ) lies in the limit of tangent spaces Tτ,yi when the limits exist. These conditions may appear strange at first glance, but their importance comes from the fact that they imply local triviality of the topology of X along each stratum σ. In more precise terms, there exists a tubular neighourhood σ ⊂ Tσ ⊂ X an [Mr] with a retraction π : Tσ → σ which makes it a locally trivial fibre bundle with a contractible fibre. A number of authors have observed that the Whitney conditions can be reformulated in more algebraic language; a simple description can be found in the proof of [Li1, thm 3.2] for instance. So in particular, given k-variety X, a stratification which is Whitney for one embedding k ⊂ C will be Whitney for all. Concerning existence in this generality, we have Lemma 3.1.9. If X is a k-variety with a filtration X = Y 0 ⊃ Y 1 ⊃ . . . Y n = ∅ by closed sets, there exists a Whitney stratification defined over k such that each Y i is a union of strata. Sketch. This is a modification of Teissier's method for constructing canonical Whitney stratifications [T]. For simplicity, assume that X is irreducible. Define X 0 = X, X 1 = Xsing ∪ Y 1 and inductively set X i+1 = {x ∈ X i ∃j < i, the Whitney conditions fail at x ∈ X i ⊂ X j} ∪ Y i+1 This gives a chain of closed sets decreasing to ∅ by [T, p 477]. The partition {X i − X i+1} can be seen to give a Whitney stratification by arguing as in [T, pp 478-480]. The following gives a version of Deligne's generic base change theorem [SGA4h, thm 1.9, p 240] for complex varieties. 20 / /   / ? Theorem 3.1.10. Given a morphism f : X → Y defined over k and a k- constructible sheaf F on X, the sheaves Rif∗F are k-constructible. There exists a dense Zariski open U ⊂ S such that the restriction F has the base change property with respect to f −1U → U . Proof. Let j : X ֒→ ¯X be an open immersion such that there is a proper map ¯f : ¯X → Y extending f . Let Σ be a Whitney stratification of Y with connected strata, and let Λ be a Whitney stratification of X refining f −1Λ such that j!F is Σ-constructible. By lemma 3.1.9, we may assume that Σ and Λ are defined over k. We may also assume that ¯X − X is a union of strata, and that ¯f is a submersion on each stratum. Each σ ∈ Σ possesses a tubular neighourhood σ ⊂ Tσ ⊂ Y with a retraction π : Tσ → σ which makes it a locally trivial fibre bundle with a contractible fibre G. The preimage f −1Tσ inherits a stratification from X, such that f −1σ is a union of strata. Thom's isotopy theorem [Mr, V2] implies that f −1Tσ → Tσ is a stratfied fibre bundle. That is, there exists an open cover {Vi} of Tσ and stratfied space Φ such that there are homeomorphisms f −1Vi ∼= Φ × Vi of stratified spaces compatible with projection. One can see that there is no loss in generality in assuming that each Vi = π−1Ui for an open subset Ui ⊂ σ. We may assume that the Ui are contractible. It follows that (f −1Tσ, f −1σ) is a (relative) fibre bundle over σ with fibre say (Φ × G, Φ′). We can see that Φ carries a constructible sheaf G which pulls back to the restriction of F under the homeomorphisms f −1Ui ∼= Φ × G × Ui. This implies that Rif∗F is locally constant along σ, and hence k-constructible. Applying the above argument to a Zariski dense stratum σ, shows that (X → Y, F ) is a locally trivial over σ. Therefore the base change property holds over σ. 3.2 Cohomology of pairs Let S be a k-variety. Let V ar2 S be the category whose objects are pairs (X → S, Y ) with Y ⊆ X closed. A morphism from (X → S, Y ) → (X ′ → S, Y ′) is a morphism of S-schemes X → X ′ such that f (Y ) ⊆ Y ′. For such an object and a sheaf F on Xan, set H i S(X, Y ; F ) = Rif∗jX,Y !FX−Y where f : X → S is the projection and jX,Y : X − Y → X is the inclusion. We revert to writing this as H i S(X, F ) when Y is empty. When F = F is constant, H i S(X, Y ; F ) is k-constructible by the theorem 3.1.10, and we can describe this as the sheaf associated to U 7→ H i(f −1U, f −1U ∩ Y ; F ) The map (X, Y ) 7→ H i S(X, Y ; F ) is easily seen to give a contravariant func- S(X ′, Y ′) are induced by the S(X, Y ) → H i tor on V ar2 homomorphisms S. The morphisms H i H i(f ′−1U, f ′−1U ∩ Y ′; F ) → H i(f −1U, f −1U ∩ Y ; F ) 21 Definition 3.2.1. A pair (f : X → S, Y ) in V ar2 S is controlled with respect to F if f is controlled with respect to the sheaf jX,Y !FX−Y . The pair is said to be controlled it if it so with respect to the constant sheaf F . The control condition for a pair, with respect to F , amounts to requiring that both ( ¯X, ¯X − X) and ( ¯Y , ¯Y − Y ) are relative fibre bundles over an intermediate projective family T → S, where ¯Y is the closure of Y . Lemma 3.2.2. If f : (X → S, Y ) is controlled with respect to F , then jX,Y !FX−Y has the base change property with respect to f . Proof. This is a consequence of lemma 3.1.7. Therefore if (X → S, Y ) is controlled then H i S(X, Y ; F )s ∼= H i(Xs, Ys; F ) (6) for every s ∈ S. From proposition 3.1.8, we obtain. Lemma 3.2.3. Suppose that (X → S, Y ) is controlled. Then for a general hyperplane H (with respect to a locally closed embedding X ⊂ PN × S), (H → S, H ∩ Y ) is controlled. Given a chain of closed sets X ⊃ Y ⊃ Z and a sheaf F on X, we get an exact sequence 0 → jX,Y !FX−Y → jX,Z!FX−Z → i∗jY,Z!FY −Z → 0 where i : Y → X is the inclusion. This induces a long exact sequence . . . H i S(X, Y ; F ) → H i S(X, Z; F ) → H i S(Y, Z; F ) → H i+1 S (X, Y ; F ) . . . which reduces to the usual exact sequence for pairs, when S is point and F is constant. 3.3 Premotivic sheaves Let S be a k-variety. The category PM(S) of premotivic sheaves is constructed as a direct limit of categories PM(S, Σ). Each PM(S, Σ) is obtained by apply- ing Nori's construction to an appropriate graph ∆(S, Σ) and functor HΣ given below. Let S ∈ ObV ark be connected. Then we construct a graph ∆(S) as follows. The objects (i.e. vertices) are quadruples (X → S, Y, i, w) consisting of (1) a quasi-projective morphism X → S. (2) a closed subvariety Y such that the pair (X → S, Y ) is controlled (defini- tion 3.2.1), (3) a natural number i ∈ N and an integer w. 22 The set of morphisms (edges) is the union of the three following sets: Type I: Geometric morphisms (X → S, Y, i, w) → (X ′ → S, Y ′, i, w) where (X → S, Y ) → (X ′ → S, Y ′) is a morphism in V ar2 S. Type II: Connecting morphisms (f : X → S, Y, i + 1, w) → (f Y : Y → S, Z, i, w) for every chain Z ⊆ Y ⊆ X of closed sets. Type III: Twisted projection morphisms (X × P1, Y × P1 ∪ X × {0}, i + 2, w + 1) → (X, Y, i, w) for every (X, Y, i, w) ∈ ObΓ(S). components. Thus the parameters i and w are locally constant. For arbitrary S ∈ V ark, set ∆(S) = Q ∆(Si), where Si are the connected By a good stratification or simply stratification of S, we mean a finite par- tition Σ into connected locally closed sets (defined over k) such that Σ contains the closure of every element. Given a stratification Σ, let ∆(S, Σ) ⊂ ∆(S) be the full subgraph consisting of objects such that H i S(X, Y ; F ) is constructible the sheaves H i with respect to the stratification Σ. We have that S ∆(S, Σ) = ∆(S) because Given Σ as above, let s = (sσ ∈ σ(¯k)) denote a collection of base points, one S(X, Y ; F ) are constructible. for each σ ∈ Σ. Let Σ be the cardinality of Σ. Define HΣ,s,F (X, Y, i, w) = Yσ∈Σ [H i S(X, Y ; F )]sσ = Yσ∈Σ H i(Xsσ , Ysσ ; F ) to be the product of stalks. We usually suppress the symbols Σ, s, F . We want to extend H = HΣ,s,F to a functor ∆(S, Σ)op → F -mod. We do this case by case. Type I: A morphism g : (f : X → S, Y, i, w) → (f ′ : X ′ → S, Y ′, i, w) of type I gives rise to the natural homorphism H i(f ′−1U, f ′−1U ∩ Y ′; F ) → H i(f −1U, f −1U ∩ Y ; F ) for each U ⊂ S. Since this is clearly a morphism of presheaves, it in- duces a morphism of sheaves H i S(X, Y ). Thus we get the desired map H(X ′, Y ′, i, w) → H(X, Y, i, w) by taking the product of this sheaf map over stalks. We give a second description which is a bit more S(X ′, Y ′) → H i 23 complicated, although better for comparing to the ´etale case. We have a commutative diagram jX ′Y ′!FX ′−Y ′ (PPPPPPPPPPPPP FY ′ 8♣♣♣♣♣♣♣♣♣♣♣♣ [1] FX ′ Rg∗jXY !FX−Y Rg∗g∗FY ′ 8♣♣♣♣♣♣♣♣♣♣♣ [1] (PPPPPPPPPPPP Rg∗g∗FX ′ where the triangles are distinguished, and the solid vertical arrows are the adjunction homomorphisms. Thus we get the dotted arrow above. From which we obtain Rf ′ ∗jX ′Y ′!F → Rf ′ ∗Rg∗jXY !F ∼= Rf∗jXY !F So we get a map of sheaves Rif ′ ∗jX ′Y ′!F → Rif∗jXY !F which is easily seen to coincide with the previous map. Type II: A morphism (X, Y, i + 1, w) → (Y, Z, i, w) of type II gives rise to a con- S (X, Y ) induced from the exact S(Y, Z) → H i+1 necting homomorphism H i sequence 0 → jX,Y! F → jX,Z!F → jY,Z!F → 0 Taking a product over stalks yields H(Y, Z, i, w) → H(X, Y, i + 1, w) . Type III: Finally a morphism (X × P1, Y × P1 ∪ X × {0}, i + 2, w + 1) → (X, Y, i, w) corresponds to the isomorphism H i S(X, Y ; F ) → H i+2 S (X × P1, Y × P1 ∪ X × {0}; F ) given by exterior product with the fundamental cycle of (P1, {0}). This gives rise to H(X, Y, i, w) → H(X × P1, Y × P1 ∪ X × {0}, i + 2, w + 1) Thus we can apply the construction from the previous section to obtain: Definition 3.3.1. The category PM(S, Σ, s; F ) of Σ-constructible premotivic sheaves of F -modules on S is the category of finite dimensional right comodules over End∨(HΣ,s). For any finite commutative F -algebra R, let PM(S, Σ, s; F )⊗F R denote the category with finitely generated right comodules over End∨(H) ⊗F R. 24 (   ✤ ✤ ✤ ✤ ✤ ✤ ✤ o o   8   ( o o 8 3.4 Realizations By definition there is a faithful exact forgetful functor U : PM(S, Σ; F ) → F -mod. We can see immediately from the universal coefficient theorem that PM(S, R) = PM(S, F ) ⊗F R, whenever F ⊆ R is a field extension. The matrix coefficients of the End∨(H)-coaction of any object V of PM(S, Σ) lie in some End∨(HD) for a finite subgraph D. Thus V can be regarded as an End∨(HD)-comodule, or equivalently an End(HD)-module. In fact, we can describe PM(S, Σ, s) as the direct limit of the categories of finite dimensional End(HD)-modules, as D ⊂ ∆(S) varies over finite subgraphs (cf [Br]). This dual description was employed by Nori in his work, and it would appear that PM(Spec k, Spec C) is just Nori's category of cohomological motives tensored with F . We write this as PM(k; F ) or simply PM(k) from now on. Given M = (X → S, Y, i, w) ∈ Ob∆(S), H(M ) is naturally an End(H(M ))- S(X, Y )(w) S(X, Y ) if w = 0 (we will see shortly that this independent of Σ and s in a module, and hence by transpose an End∨(H)-comodule denoted by hi or hi suitable sense). When S = Spec k, we omit the subscript. By definition we have Proposition 3.4.1. PM(S, Σ, s; F ) is an F -linear abelian category with an exact faithful functor to F -mod. In view of the following, we may suppress base points. Lemma 3.4.2. Suppose that tσ is another collection of base points, then PM(S, Σ, s) and PM(S, Σ, t) are isomorphic. Proof. Given a homotopy class of paths γσ in σ joining sσ to tσ, parallel trans- port along these curves yields an isomorphism of fiber functors Hs ∼= Ht. Remark 3.4.3. This business of choosing base points and then suppressing them is a bit clumsy. A more elegant approach is to simply redefine HΣ,F (X, Y, i, w) = Yσ∈Σ Γ(σ, π∗ σH i S(X, Y ; F )) where πσ : σ → σ(C) are the universal covers, and then build PM(S, Σ; F ) ac- cordingly. However, the original approach does make certain things more trans- parent, and will generally be preferred. We have the following consequences of corollary 2.2.10. Construction 3.4.4. Let Cons(Sι,an, Σ; F ) denote the category of sheaves of F -modules which are constructible with respect to the stratification Σ. The fibre functor Φ : Cons(San, Σ) → F -mod given the product of stalks at the base points provides a faithful exact functor. The discussion from the previous section shows that (X, Y, i, w) 7→ H i S(X, Y ; F ) is a functor on ∆(S, Σ)op and that H is a composition of this with Φ. Thus corollary 2.2.10 yields an extension functor Rι,B = RB : PM(S, Σ) → Cons(Sι,an, Σ) that we call Betti realization. RB coincides with the forgetful functor U on PM(k). 25 Construction 3.4.5. The map tn(X, Y, i, w) = hi S(X, Y )(w + n) extends to a functor ∆(S, Σ)op → PM(S, Σ) satisfying tntm = tn+m. When composed with the forgetful functor to F -mod, we obtain H. Thus this extends to an endofunctor T n : PM(S, Σ) → PM(S, Σ) satisfying T n(hi S(X, Y )(w)) = hi S(X, Y )(w + n) and T nT m = T n+m; in particular, it is an automorphism. Construction 3.4.6. Let F be finite or Qℓ. Let ¯k ⊆ C denote the algebraic closure of k. Consider the map (f : X → S, Y, i, w) 7→ Ri ¯f et ∗ jet ¯X, ¯Y !F ¯X− ¯Y (w) where the sheaves and operations are on the ´etale topology, ¯f : ¯X → S etc. are the base changes to ¯k, j ¯X ¯Y : ¯X − ¯Y → ¯X is the inclusion, and (w) represents the Tate twist. This is easily seen to be a functor by modifying the above discus- sion. Thanks to the comparison theorem between ´etale and classical cohomology (appendix B). (Ri ¯f et ∗ jet ¯X, ¯Y !F ¯X− ¯Y (w))s ∼= (Ri ¯f∗jX,Y !FX−Y )s in F -mod for s ∈ S(C). Thus we can obtain an extension which is the ´etale realization functor Ret from PM(S; F ) to the category Cons(Set, Σ, ; F ) of sheaves of F - modules on Set constructible with respect to Σ. Construction 3.4.7. Let Cons-M HM (Sι,an, Σ; Q) denote the heart of the classical t-structure on the category M HM (S, Σ) of Σ-constructible mixed Hodge modules (appendix C). We have an embedding rat : Cons-M HM (San, Σ; Q) ֒→ Cons(San, Σ) which can be composed with the above functor Φ to obtain a fibre functor. Con- sider the map (f : X → S, Y, i, w) 7→ cH i ◦ ¯Rf∗j ¯X, ¯Y !F ¯X− ¯Y (w) where the operations are in the derived categories of mixed Hodge modules, and cH i = cτ≤i cτ≥i is cohomology with respect to the classical t-structure. When composed with rat, we obtain H i S(X, Y ). Thus we obtain a Hodge realization functor Rι,H = RH : PM(S, Σ) → Cons-M HM (Sι,an, Σ) A special given in section 6.1 can be made more explicit. We fixed an embedding ι : k ֒→ C at the outset. Let write PM(S; F )ι for the resulting category. We now show that the category PM(S; F ) is independent of this. 26 Proposition 3.4.8. For any two embeddings of ι, µ : k ⊂ C, the categories PM(S, F )ι and PM(S, F )µ are equivalent. Proof. It suffices to show that PM(S, Σ, F )ι and PM(S, Σ, F )µ are equivalent for every stratificaion. We note that PM(S, Σ, ; F ) = PM(S, Σ; F0) ⊗F0 F for any subfield. Thus it suffices to assume that F is the prime field. Suppose that F = Z/pZ. Then, we can see this immediately by the comparison theorem [SGA4, exp XVI, thm 4.1] HΣ,s(X, Y, i, w; F ) = Yσ∈Σ ∼= Yσ∈Σ [Rif∗j!F ]sσ [Ri ¯f et ∗ jet ! F ]sσ = H et Σ,s(X, Y, i, w; F ) This description is independent of the embedding. Therefore PM(S, Σ, F )ι and PM(S, Σ, F )µ are equivalent. The remaining case F = Q follows Nori's argument [N2] quite closely. Write H ι and H µ for the functors corresponding to the embeddings. Define the cat- egory T of triples (A, B, h) A, B ∈ ObF -mod, h : A ⊗ Qℓ ∼= B ⊗ Qℓ, where morphisms are compatible pairs of linear maps. If p denote the first projection (A, B, h) 7→ A, then p is easily seen to be fully faithful and essentially surjec- tive. Therefore it is an equivalence. So there is functor q : Q-mod → T and natural isomorphisms γ : q ◦ p ∼= 1T and η : p ◦ q ∼= 1Q-mod. We get a functor H T : ∆(S, Σ) → T by taking H ι and H µ as the first and second component. For the third, we use the composition of the comparison maps h : H ι ⊗ Qℓ ∼= H et ∼= H µ ⊗ Qℓ The map p induces a homomorphism End∨(H ι) → End∨(H T ). We claim that this gives an isomorphism. Here we use the duality principle given in lemma 2.1.2. The dual of p is given by p∗(f ) = 1p ⋄ f . The map is injective as it has a left inverse given by g 7→ (γ ⋄ 1H ) ◦ (1q ⋄ g) ◦ (γ−1 ⋄ 1H) The map p∗ is also surjective, because p∗(1q ⋄ [(η ⋄ 1) ◦ g ◦ (η−1 ⋄ 1)]) = g By an identical argument End∨(H µ) ∼= End∨(H T ) 3.5 Base change Lemma 3.5.1. Let S = S1 ∪ S2 ∪ . . . Sn be a decomposition into connected components. Choose stratifications Σi of Si and let Σ =S Σi. Then PM(S, Σ) = PM(S1, Σ1) × PM(S2, Σ2) × . . . PM(Sn, Σn) 27 Proof. This is an immediate consequence of lemma 2.2.13, which applies because the empty families (∅, ∅, i, w) ∈ ∆(Si, Σi)op map to 0 under H. A morphism of (pointed) stratified varieties is a morphism of varieties such that a nonempty preimage of any stratum is a union of strata (and base points go to base points). If the underlying morphism of varieties is the identity, we say that the first stratification refines the second. Construction 3.5.2. Suppose that f : (T, Λ, t) → (S, Σ, s) is a morphism of pointed stratified varieties. First, suppose (*) that the map on base points is surjective. Applying corollary 2.2.10 to (X → S, Y, i, w) 7→ hi T (X ×S T, Y ×S T )(w) yields an extension, which is the base change functor f ∗ : PM(S, Σ, s) → PM(T, Λ, t). map If f (t) 6= s, set T ′ = T ` s′ where s′ = s − f (t). Then the f ′ : (T ′, Λ′ = Λ ∪ s′, t′ = t ∪ s′) → (S, Σ, s), which is f on T and identity on s′, satisfies (*). We now define f ∗ as the composite PM(S, Σ, s) f ′∗ → PM(T ′, Λ′, t′) = PM(T, Λ, t) × PM(s′, s′, s′) p → PM(T, Λ, t) where p is the projection. We can always extend a morphism of stratified varieties to a morphism of pointed stratified varieties, and this way obtain base change functor f ∗ : PM(S, Σ) → PM(T, Λ). Alternately, we could define this directly in the spirit of remark 3.4.3 without recourse to base points. We mention two important special cases of this construction: 1. The construction applies when S = T and Λ refines Σ. This leads to faithful exact embeddings ρΣ,Λ : PM(S, Σ) → PM(S, Λ). 2. When T = s is the of set of, say n, base points, we get an embedding PM(S, Σ) → PM(k)n. The last map is usually not an equivalence. Example 3.5.3. Let S = A1 with Σ = {0, A1 − {0}} and base points s = {0, 1}. Now consider the motive QS represented by (id : S → S, ∅, 0, 0). By passing to sheaves under RB, we see that this is not isomorphic to Q0 ⊕ Q1, so the base point map PM(S) → PM(k) × PM(k) cannot be an equivalence. When f : T → S is an inclusion, we often denote f ∗M by M T . Lemma 3.5.4. The assignment (S, Σ) 7→ PM(S, Σ), f 7→ f ∗ yields a con- travariant pseudofunctor from the category of stratified varieties to the 2-category of abelian categories. This commutes with RB. 28 Proof. The functor id∗ S : PM(S, Σ) → PM(S, Σ) is the extension of (X → ∼= S, Y, i, w) 7→ hi idPM(S). Suppose that f : (T, Λ) → (S, Σ) and g : (V, Θ) → (T, Λ) are given, and assume the maps surject on base points. Then (f ◦ g)∗ is the extension of S(X, Y )(w). So clearly we have a natural isomorphism id∗ S (X → S, Y, i, w) 7→ hi V (X ×S V, Y ×S V )(w) ∼= g∗hi T (X ×S T, Y ×S T )(w) So we obtain a natural isomorphism (f ◦ g)∗ ∼= g∗ ◦ f ∗. These natural transfor- mations can be seen to have the necessary compatibilities (see appendix A) to define a pseudofunctor. The last statement follows from the isomorphism RB(hi T (X ×S T, Y ×S T )(w)) ∼= H i T (X ×S T, Y ×S T ) and corollary 2.2.11. Definition 3.5.5. The category of motivic sheaves of F -modules is given by the 2-colimit PM(S; F ) = 2- lim−→ Σ PM(S, Σ; F ) (see appendix A). It follows from the discussion in appendix A that PM(S; F ) is abelian, and the natural maps PM(S, Σ; F ) → PM(S; F ) are exact. Note that hi S(X, Y ) ∈ PM(S, Σ) maps to the same symbol hi S(X, Y ) under refinement. We denote the common value in the colimit by hi S(X, Y ) as well. Observe that RB provides a faithful exact embedding of PM(S, Σ) into the category Sh(S) of sheaves of F -vector spaces on S. This is compatible with refinement. So it passes to the limit. Therefore in more concrete terms, we can identify PM(S) (up to equivalence) with the directed union of subcategories RB(PM(S, Σ)) ⊂ Sh(S) [Σ Note that this lies in the subcategory Cons(S) of constructible sheaves. As a corollary to lemma 3.5.4. Corollary 3.5.6. S 7→ PM(S) is a contravariant pseudofunctor. There are a number of useful variations of this construction. Let ∆c(S, Σ) ⊂ ∆(S, Σ) denote the full subgraph consisting of tuples (X → S, Y, i, w) where X → S is projective. Then set PMc(S, Σ) be the category of comodules over End∨(HΣ∆c(S,Σ)). The category of compact motives PMc(S) = 2- lim−→ Σ PMc(S, Σ; F ) This can be regarded as an abelian subcategory of PM(S). It contains motives hi S(X) of projective families. 29 4 Motivic Sheaves 4.1 Zariski Descent In the next section, we will see that motivic sheaves in PMc(S) can be patched on a Zariski open cover. This is not yet evident for PM(S), and may be an indication of a defect in the definition. Since this issue plays a relatively minor role (so far), we just sketch a construction of a new category of motivic sheaves M(S) which removes this defect. We can repackage M(S) in the language of fibred categories [Gi, Vi]. Given a functor π : F → C, the fibre π−1(A) over A ∈ Ob(S) is the category with objects π−1A and morphism π−1idA. Fibres need not behave as expected, for example fibres over isomorphic objects need not be equivalent, unless further conditions are imposed. An arrow φ ∈ M orF is cartesian if for any commutative diagram consisting of solid arrows A ● π(A) ● ● ● ψ #● )❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ )❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ #❋❋❋❋❋❋❋❋ π(ψ) B φ / C π(B) / π(C) π(φ) the dotted arrow can be filled in uniquely. The functor π : F → C is fibred if any arrow of C can be lifted to a cartesian arrow of F with a specified target. It is sometimes convenient to fix a collection of specific cartesian lifts. Such a collection is called a cleavage. Given a cleavage, there is a well defined way to define a pullback functor f ∗ : π−1(B) → π−1(A) for any f : A → B ∈ M orC. These form a pseudofunctor. Conversely, any pseudofunctor determines a fibred category. Define M to be the category whose objects are pairs (S ∈ V ark, M ∈ M(S)), and morphisms are pairs (f : T → S, M → f ∗N ∈ M orM(T )). This is fibred over V ark via the natural projection π : M → V ark. The categories M(S) are just the fibres π−1(S), and the original pseudofunctor is determined by the cleavage {(f, f ∗N = f ∗N )}. Definition 4.1.1. Let π : M → V ark denote that stack associated to PM → V ark for the Zariski site [Gi, chap 2 §2]. The category of motivic sheaves M(S) over S ∈ ObV ark is the fibre over S. Unravelling all of this, we see that: 1. M → V ark is a fibred category. Fix a cleavage for it, so that there are functors f ∗ : M(S) → M(T ) for each f : T → S. If f is an inclusion, we denote f ∗M by M S. 2. For any M, N ∈ M(S), U 7→ HomM(S)(M U , N U ) is a sheaf on the Zariski topology, i.e. M is a prestack. 30 ) # ) # / / 3. Given an open cover {Ui} of S, Mi ∈ PM(Ui) with isomorphisms gji : ∼= MjUij satisfying the usual cocycle condition gii = id, gik = MiUij gijgjk, there exists M ∈ M(S) such that Mi ∼= M Ui. 4. There is a functor ι : PM → M of fibred categories. This means that it fits into a commutative diagram PM ι #❍❍❍❍❍❍❍❍❍ M ①①①①①①①① V ark and takes cartesian arrows to cartesian arrows. The functor ι is univeral among all such functors from PM to stacks. So that any functor M → M′ to a fibred category satistfying conditions (2) and (3) must factor through M in an essentially unique way, or more precisely, there is an equivalence between the categories Homf ibcat(M, M′) and Homf ibcat(M, M′). A concrete construction of the associated stack is given in [LM, chap 3] (although stated for groupoids, it works in general). It is a two step process. First, one constructs a prestack PM+. The objects of PM+(S) are the same as for PM(S). For HomPM+(S)(M, N ) we take the global sections of the sheaf associated to U 7→ HomPM(S)(M U , N U ). This forces condition (2) to hold. Now construct M(S) as the category whose objects consists of descent data, i.e., collections (Mi ∈ PM+(Ui), gij) as in (3). Given two objects (Mi ∈ PM+(Ui), gij), (M ′ ij), after passing to a common refinement we can assume that Ui = U ′ i . A morphism is then given by a compatible collection of morphisms Mi → M ′ i . We have obvious functors PM(S) → PM+(S) → M(S). It is easy to see from this description that M(S) is abelian, and PM(S) → M(S) is exact. To summarize: i ∈ PM+(U ′ i ), g′ Theorem 4.1.2. M is a stack of abelian categories over V ark. There is an exact functor PM → M which is universal among all functors from PM to stacks. S(X, Y )(w) as the image of hi As corollary, the realization functors RB et cetera extend to M. We define hi S(X, Y )(w) in M(S). Since the collection of sheaves Sh(S) forms a stack over Szar, we can see that RB factors through M. With the help of the above explicit description, we get a slightly sharper result. Corollary 4.1.3. The functor RB : M(S) → Sh(S) extends to an exact faithful functor RB : M(S) → Sh(S). In particular, RB(hi S(X, Y )(w)) = H i S(X, Y ). By a very similar argument, we have: Corollary 4.1.4. The functor Ret : M(S, F ) → Sh(Set, F ) extends to an exact faithful functor Ret : M(S, F ) → Sh(Set, F ). 31 / / # We want give an alternative concrete construction. It will be convenient to start with some generalities. It will be convenient to "sheafify" the discussion of section 2.2. Let D = {DU } be a collection of graphs indexed Zariski open subsets of S, such that for each inclusion ι : V ⊂ U we have a restriction morphism ι−1 : DV → DU satisfying (ι ◦ µ)−1 = µ−1 ◦ ι−1. Let H : DU → F -mod be a collection of morphisms compatible with restriction. We refer (D, H) as a compatible collection. Given such a collection, let E ∨(H) to be the sheaf associated to the presheaf U 7→ End∨(HDU ) Let E ∨(H)-comod denote the category of Zariski sheaves of finite dimensional F -vector spaces with a coaction by E ∨(H). When D consists of finite graphs, this is isomorphic to the category of modules over the sheaf of rings E(H) = Hom(E ∨(H), FS) Let Coh(E(H)) denote the category of coherent i.e. modules. The colimit locally finitely presented, Coh(E ∨(H)) = 2- lim−→ D′⊂D finite Coh(E(HD′ )) can be realized as a subcategory of E ∨(H)-comod. Suppose that we are given compatible collections D and D and morphisms π : DU → Du compatible with restriction. Then a functor H on D induces a functor Coh(E ∨(H)) → Coh(E ∨(H ◦ π)) An analogue of corollary 2.2.4 is: Lemma 4.1.5. This functor is an equivalence if for any two objects of the fiber π : DU → Du are connected by a chain of morphisms on some cover of U . Given a functor H # : ∆(S, Σ) → F -mod2 compatible with restriction and such that H = p1 ◦ H #, we can define a category M#(S, Σ) by mimicing the procedure used to define M′. We have refinement of corollary 2.3.2 and which follows by the same argument. Lemma 4.1.6. M#(S, Σ)/ ker p1 ∼ M(S, Σ). Returning to the initial set up of a stratified variety We now give the explicit construction. Fix a stratified variety (S, Σ) with a given set of base points s for Σ. Given Zariski open sets V ⊂ U ⊂ S, let s′ denote the intersection of s with U − V . As in construction 3.5.2, we have a map given by composition End∨(H∆(U,Σ)) → End∨(H∆(V,Σ)) × End∨(H∆(s′)) → End∨(H∆(V,Σ)) where the last map is projection. This makes U 7→ End∨(H∆(U,Σ)) 32 into a presheaf of coalgebras. Let E ∨(S, Σ) denote the associated sheaf on the Zariski topology Szar. Then D = {∆(U, Σ)} gives a compatible collection with restrictions given by (X → V, Y, i, w) 7→ (X ×V U, Y ×V U, i, w) Set Coh(E ∨(S, Σ)) = Coh(E ∨(H)) as above. A premotivic sheaf M in PM(S, Σ) determines an object of U 7→ M U of E ∨(S, Σ)-comod, which can be seen to lie in some Coh(E(HD′ )), with D′ ⊂ D finite, and therefore in Coh(E ∨(S, Σ)). This gives a fully faithful exact functor PM(S, Σ) → Coh(E ∨(S, Σ)). Lemma 4.1.7. Coh(E ∨(S, Σ))-comod coincides with the full subcategory of co- modules M which are locally isomorphic to objects of PM(−, Σ) Proof. An object of Coh(E ∨(S, Σ))-comod lies in some Coh(E(D)). So it is locally of the form coker(E(D)n → E(D)m) which lies in PM(S, Σ). Set M′(S) = 2- lim−→Σ M′(S) induced from the one above. Coh(E ∨(S, Σ)). Then we have a functor PM(S) → Lemma 4.1.8. M(S) is equivalent to M′(S). Proof. The functor PM(S) → M′(S) induces a functor PM+(S) → M′(S) which is fully faithful. This extends to a functor M(S) → M′(S) which is again fully faithful. It also essentially surjective by the previous lemma. In view of this result, we will generally denote M′(. . .) simply by M(. . .) from henceforth. We also define M(S, Σ) = Coh(E ∨(S, Σ))-comod. 4.2 Extension by zero Let j : S ֒→ ¯S be an open immersion with boundary ∂ ¯S = ¯S − S. Suppose that Σ is a stratification of ¯S such that S is union of strata. Let ∆( ¯S, ∂ ¯S, Σ) ⊂ ∆( ¯S, Σ) denote the full subgraph consisting of tuples ( ¯X → ¯S, ¯Y , i, w) where ¯Y ⊇ f −1∂ ¯S. We construct the category M( ¯S, ∂ ¯S, Σ) as in §3.3, as the category of End∨(HΣ∆( ¯S,∂ ¯S,Σ))-comodules, and M( ¯S, ∂ ¯S) is the colimit of these over Σ. There is an exact faithful forgetful functor ι ¯S : M( ¯S, ∂ ¯S) → M( ¯S). We define ∆ex(S, Σ) ⊆ ∆(S, Σ) be the full subgraph of tuples (X → S, Y, i, w) which extend to ∆( ¯S, ∂ ¯S, Σ). Then we can define the subcategory Mex(S, Σ) ⊆ M(S, Σ) by taking comodules over HΣ restricted to ∆ex. Let Mex(S) = 2- lim−→ Mex(S, Σ) ⊂ M(S) Lemma 4.2.1. The category : M( ¯S, ∂ ¯S) is equivalent to a subcategory Mex(S) ⊂ M(S) via j∗. Proof. We start by proving that j∗ : M( ¯S, ∂ ¯S, Σ) → Mex(S, Σ) is an equiva- lence. We have a morphism π : ∆( ¯S, ∂ ¯S, Σ) → ∆ex given by restriction. This is surjective by definition, and the fibres of π are clearly connected. For ( ¯X → 33 ¯S, ¯Y , i, w) ∈ ∆( ¯S, ∂ ¯S, Σ), Rij ¯X, ¯Y !F = 0 outside of S. Thus H( ¯X → ¯S, ¯Y , i, w) coincides with the value of H on the restriction. Therefore we have a morphism of pairs (∆( ¯S, ∂ ¯S, Σ), H) → (∆ex, H), and so j∗ : M( ¯S, ∂ ¯S, Σ) → Mex(S, Σ) is an equivalence by corollary 2.2.4. Passing to the limit yields the lemma. Lemma 4.2.2. Mc(S) ⊆ Mex(S). Proof. Given a projective family X ⊂ PN × S over S, we get an extnesion over ¯S by taking the closure ¯X ⊂ PN × ¯S. Given Y ⊂ X, we may take ¯Y ⊂ ¯X to be the union of the closure of Y with the preimage of ∂S. Thus we see that ∆c(S, Σ) ⊂ ∆ex(S, Σ). Definition 4.2.3. Define j! : Mex(S) → M( ¯S) by j! = ι ¯S′ ◦ j′∗−1. Lemma 4.2.4. This is compatible with extension by zero for sheaves j! : Sh(S) → Sh( ¯S) in the sense that RB(j!F ) = j!RB(F ). We have j!Mc(S) ⊂ Mc( ¯S). Proposition 4.2.5. There exists a natural transformation j!j∗ → 1 on M( ¯S) compatible with the usual adjunction map for sheaves. Proof. The proof is similar to the proof of proposition 5.1.4. Let M or′ ⊂ M orM( ¯S) denote the subcategory of morphisms M2 → M1 such that there is a commutative diagram j!j∗RBM1 RBM1 ∼= = RBM2 / RBM2 commutes. The functor (M2 → M1) 7→ M1 is clearly faithful and exact. There- fore by corollary 2.2.10, we get a functor M( ¯S) → M or′ such that hi ¯S(X, Y )(w) 7→ [hi ¯S(X, Y ∪ f −1∂ ¯S)(w) → hi ¯S(X, Y )(w)] for each (f : X → ¯S, Y, i, w) ∈ ∆(S). Proposition 4.2.6. PMc(S) forms a stack in the Zariski topology, i.e. objects and morphisms can be patched on Zariski open covers. Proof. By compactness and induction, it suffices to treat covers S = U0 ∪ U1 consisting of two open sets. Given Mi ∈ ObPMc(Ui) with an isomorphism ∼= M1U0∩U1. Let ji : Ui ֒→ S and j01 : U0 ∩ U 1 → S denote the f : M0U0∩U1 inclusions. We define a morphism g : j01!M0 → j0!M0 ⊕ j1!M1 extending 1 on the first factor and −f on the second. Then M = coker(g) gives an object of PMc(S) which restricts to Mi. 34 / /     / 4.3 Cellular decompositions A number of constructions will be based on the existence "cellular" decomposi- tions. The use of such decompositions plays a key role in Nori's work and also [Ar]. This hinges on a result of Beilinson [B2] that Nori calls the basic lemma. We need a slight modification of this result. Define a map f : X → S to be equidimensional (respectively uniform) if dim Xs is constant (respectively, if for every s, all irreducible components of Xs have the same dimension). Proposition 4.3.1. Let X → S be a uniform affine morphism. Suppose that F is a constructible sheaf on X such that (X → S, F ) is controlled. Then there exists a Zariski open cover {Sβ} of S, dense affine open subsets gβ : U β ֒→ X β = f −1Sβ such that (1) (U β → Sβ, gβ ! gβ∗F ) is controlled. (2) For every s ∈ Sβ, H i Sβ (X β, X β − U β; F )s = H i Sβ (X β, gβ ! gβ∗F )s = 0 unless i = dim Xs. Proof. Most of the argument is pretty much identical to the proof of [B2, lemma 3.3]. Nevertheless, we spell this out since the result is central. Let h / S g T ¯g 7♦♦♦♦♦♦♦♦♦♦♦♦♦♦ ?⑦⑦⑦⑦⑦⑦⑦ ¯f X k ¯X be a control diagram. After replacing ¯X by the blow up along ¯X − X, we can assume that k is affine. We can choose a divisor Z ′ ⊂ T such that F is constant on the complement of Z = g−1Z ′. Let ℓ : X − Z ֒→ X be the inclusion, then M = ℓ!FX−Z is also controlled by the above diagram. Set ¯M = Rk∗M . Note that F [dim Xs], M [dim Xs] and ¯M [dim Xs] restrict to perverse sheaves on the fibres of f or ¯f , because ℓ and k are affine embeddings [BBD, cor. 4.1.3]. Fix an embedding ¯X ⊂ PN × S over S. Given a hyperplane H, let V = ¯X − H j k′ X − H = V ∩ X / ¯X k / X j′ i i′ H k′′ H ∩ X denote the inclusions. Let js : ¯Xs − H ֒→ ¯Xs etc. denote the restrictions of these inclusions to the fibre over s ∈ S. We claim that j!Rk′ ∗M V ∩X ∼= Rk∗j′ ! M V ∩X (7) 35 / /   / ? 7 / o o O O / O O o o O O holds for a dense open set P of hyperplanes H in the dual projective space P = PN . The argument which we sketch is from [B2, pp 35-36]. First note that (7) equivalent to (8) by the argument indicated in [loc. cit.]. Let HP ⊂ ¯XP = ¯X × P denote the universal hyperplane. Let MP denote the pullback of M to XP = X × P. Let iP : HP → ¯XP denote the canonical morphism. Similarly the other morphism i′, . . . have obvious extensions denoted by (−)P such that they can be recovered by taking the fibre at H ∈ P. With this notation, it suffices to prove i∗Rk∗M ∼= Rk′′ ∗ i′∗M P RkP ∗MP ∼= Rk′′ i∗ P ∗i′ P ∗MP by virtue of theorem 3.1.10 (2). But this is a consequence of the 3.1.10 (3), because iP is locally trivial. By combining this with proposition 3.1.8, we can find a cover {Sβ} and hyperplanes H β ∈ P so that j′ ! M V ∩X is controlled over Sβ. In order to simplify notation, replace S by Sβ etc. below, for a fixed β. We set U = X − H − Z with inclusion g. Then g!g∗F = j′ ! M V ∩X . So the first item of the proposition holds, and in particular, this sheaf has the base change property. It remains to prove item (2). Since each fibre Xs is affine, we have H i S(g!g∗F ) = 0 for i > dim Xs by Artin's vanishing theorem. The remaining half also follows from the affineness of X. As fs is affine, Rfs! is left t-exact for the perverse t-structure [BBD, cor. 4.1.2]. Therefore H i c(V ∩ Xs, ¯M V [dim Xs]) = Hi(Rfs! ¯M V [dim Xs]) = 0 for i < 0. Then from this, (7) and the proper base change theorem we obtain, Rif∗(j′ ! M V ∩X )s ∼= Hi(Rf∗j′ ! M V ∩X )s ∼= Hi(R ¯f∗Rk∗j′ ! M V ∩X )s ∼= Hi(R ¯f∗j!Rk∗M V ∩X )s ∼= Hi(R ¯f∗j′ ∼= Hi(R ¯f∗j! ¯MV )s ∼= (Ri( ¯f ◦ j)! ¯M V )s = 0 ! M V ∩X )s for s ∈ S and i < dim Xs. Corollary 4.3.2. The schemes U β → Sβ can be assumed to be smooth. Corollary 4.3.3. If X → S is equidimensional of relative dimension n. Then H i Sβ (X, gβ ! gβ∗F ) = 0 unless i = n. 36 We define ∆eq(S, Σ) ⊂ ∆(S, Σ) by requiring X → S and Y → S to be equidimensional. The categories Meq(S, Σ) and Meq(S) are defined by the same procedure as before by restricting H to ∆eq. These can be viewed as subcategories of M(S). Lemma 4.3.4. Let X → S be an affine equidimensional morphism to a variety which is controlled with respect to a sheaf F . Then there is a Zariski open cover {Sβ} of S and filtrations X β 0 ⊂ X β 1 ⊂ . . . X β n = X β = f −1Sβ such that (1) X β i → Sβ is equidimensional of pure relative dimension i. (2) The pairs (X β a → S, X β a−1) are controlled with respect to F , and H i Sβ (X β a , X β a−1; F ) = 0 for i 6= a. If in addition, X ′ 0 ⊂ X ′ 1 ⊂ . . . X ′ n = X is a given chain of closed sets each of pure relative dimension i. Then we can choose X ′ i ∩ X β ⊆ X β i . Proof. This follows from the previous proposition and induction on dim X. Definition 4.3.5. Suppose that we are given a morphism X → X of S-schemes, a Zariski open cover {Sβ} of S, filtrations by closed sets of the preimage of each Sβ X β = X β d ⊃ X β We refer to the collection (. . . , X • thing as a quasi-filtered X-variety. 0 ⊃ X β d−1 ⊃ . . . X β • ) as a quasi-filtration on X, and the whole −1 = ∅ These objects form a category QV arS, where a morphism φ : ({Sβ}β∈B, X ′ → X, X ′ β • ) → ({T γ}γ∈G, X → X, X γ • ) is given by a map r : B → G, such that Sβ ⊆ T r(β) plus commutative squares of S-schemes X ′ X X ′ f / X β with X ′ . We say that φ covers f . Let say that the quasi- filtration is simple if the cover {Sβ} consists of {S} alone. A filtered variety is • mapping to X r(β) • 37 / /     / the special case of a simple quasi-filtered variety, where X → X is the identity. Let F V arS be the full subcategory of filtered varieties. We give a relative version of Jouanolou's trick [Jo, lemma 1.5] below. To simply the statement, let us say that X → X is bundle of affine spaces if X = T ×Aff(n) An k , where T → X is a torsor for the affine group in the Zariski topology Lemma 4.3.6. If f : X → S is a quasi-projective morphism then there exists a commutative diagram X π f ❅❅❅❅❅❅❅❅ X f S such that f is affine, and π is a bundle of affine spaces. Proof. When X = PN × S, X can be taken to be product of S with the com- plement of the incidence variety StN = {(x, H) ∈ PN × PN x /∈ H}. For the general case, let X ⊂ ¯X ⊂ PN × S be a relative compactification. After blowing up, we can assume that ¯X − X is a divisor. Then the preimage X of X in StN × S will do the job. When f : X → S is projective, we see that the pullback of any constructible sheaf π∗F is necessarily controlled. We say that a quasi-filtration (π : X → X, X•) is cellular with respect to a controlled constructible sheaf F if 1. π is a bundle of affine spaces, 2. π∗F is controlled, 3. X• satisfies condition (2) of lemma 4.3.4 with respect to π∗F , i.e. H i Sβ ( X β a , X β a−1; π∗F ) = 0 Note that the first assumption implies that Rπ∗π∗F = F . The second assump- tion can be seen to be redundant, but there is no harm in including it. Lemma 4.3.7. A bundle π : X → X of affine spaces over an affine scheme admits a section. Proof. The bundle X/X is a homogeneous space associated to a torsor for the affine group Aff(n). Using the exact sequence 1 → Gn a → Aff(n) → GL(n) → 1 and the fact that X is affine, we conclude that H 1(X, Gn that X/X is a vector bundle. So it has a section. a ) = 0, and therefore Proposition 4.3.8. 38 / /    (1) Every equidimensional quasiprojective morphism possesses a cellular quasi- filtration with respect to a given controlled sheaf F . (2) Every morphism of equidimensional quasiprojective schemes over S can be lifted to a morphism of cellular quasi-filtered varieties with respect to a controlled constructible sheaf on the target. (3) The category of cellular quasi-filtrations of a fixed pair (X, F ) is connected i.e. any two objects can be connected by a chain of morphisms. Proof. The first two statements follow immediately from lemmas 4.3.4 and 4.3.6. The remaining part, take a bit more work. •, lemma 4.3.4 shows that there is third cellular filtration X ′′ First we treat the special case of (3) for cellular filtrations. Given two such filtrations X•, X ′ • ⊇ X•∪X ′ •. Now we prove it general. Suppose that we have cellular quasi-filtrations (Sβ, X → X, X β • ). Take the fibre product Z = X ×X Y . By lemma 4.3.7, Z → X and Z → Y admit sections σ and τ . Lemma 4.3.4 shows that we can refine σ( X β • of Z. By refining X • • ) to a cellular filtration Z • • , we obtain a diagram of cellular quasi-filtrations • ) and (T γ, Y → X, Y γ • and Y • • ) ∪ τ ( Y γ ( X, X • • ) → ( X, X ′ • •) ← ( Z, Z • • ) → ( Y ′, Y ′ • •) ← ( Y , Y • • ) Given a filtration X• ⊂ X by closed sets and a sheaf F , we have a spectral sequence Epq 1 = H p+q S (Xp, Xp−1; F ) ⇒ H p+q S (X, F ) cf [Ar, (10)]. When this is cellular, this reduces to an isomorphism at E2. Then putting this remark together with the above results yields Lemma 4.3.9. Suppose that (π : X → X, X•) is cellular with respect to jXY !F . H i S(X, Y ; F ) is isomorphic to the ith cohomology of the complex . . . H i S( Xi, (π−1Y ∩ Xi) ∪ Xi−1) → H i+1 S ( Xi+1, (π−1Y ∩ Xi+1) ∪ Xi) . . . 4.4 Tensor products We have a product structure on ∆(S, Σ) (and ∆eq(S, Σ)) given by (X → S, Y, i, w)×(X ′ → S, Y ′, i′, w′) = (X×SX ′ → S, X×SY ′∪X ′×SY, i+i′, w+w′) which makes it into a monoid in the category of graphs with unit (idS, ∅, 0, 0). Unfortunately, this does not immediately lead to a product on M(S, Σ). The problem has to do with the Kunneth formula. To remedy this, we define a full subgraph ∆cell(S, Σ) ⊂ ∆eq(S, Σ) 39 The objects of ∆cell consist of quadruples (X → S, Y, i, w) such that X → S is affine and such that H j S(X, Y ) = 0 unless j = i, and such that X − Y → S is smooth. Thanks to Kunneth's formula, we have a commutative diagram ∆cell(S, {S}) × ∆cell(S, Σ) ∆cell(S, Σ) H{S}×HΣ F -mod × F -mod HΣ ⊗ / F -mod leading to a product End∨(H∆cell(S,{S}))-comod×End∨(H∆cell(S,Σ))-comod → End∨(H∆cell(S,Σ))-comod With this, PMcell(S) = End∨(H∆cell(S,{S}))-comod becomes a tensor cate- gory. We form the associated stack Mcell(S, Σ) as before. The tensor product extends to this. To summarize Lemma 4.4.1. There are tensor products Mcell(S, {S}) × Mcell(S, Σ) → Mcell(S, Σ) compatible, via the forgetful functor U , with the vector space tensor product. With this structure Mcell(S, {S}) becomes a tensor category. The key point is: Theorem 4.4.2. The category Mcell(S) is equivalent to Meq(S). Before giving the proof, we give a construction. Let C[0,∞)(M(S, Σ)) be the category of bounded complexes supported in nonnegative degrees. Let Hi : C[0,∞)(M(S, Σ)) → M(S, Σ) denote the ith cohomology functor. Then composition gives a functor RB ◦ H∗ from C[0,∞)(M(S, Σ)) to the category Gr Sh(S((C)) of [0, ∞)-graded sheaves. Let C(S, Σ) be the so called comma category whose objects are triples (K •, M, φ : RB(M ) → RB ◦ H0(K •)) 1 → K • where K • ∈ ObC[0,∞)(M(S, Σ)) and M ∈ ObM(S, Σ). Morphisms are pairs K • 2 , M1 → M2 satisfying obvious compatibilities. Let Ciso(S, Σ) be the full subcategory consisting of triples for which φ is an isomorphism. We can identify F 2-mod with F -mod × F -mod. There is a faithful exact functor U2 : C(S, Σ) → (F -mod)2 given by (K •, M, φ) 7→ (Qi U (K i)) × U (M ). Given a simple quasi-filtration (T → S, T•) and a stratification Σ, choose base points s for (S, Σ) and t• ∈ T•. We define a functor H # : ∆(S, Σ)op → C(S, Σ)iso as follows. On objects H #(X → S, Y, i, w) = ((h0 S(XT0 , YT0 ∪ XT0−1 ) → h1 S(XT1 , YT1 ∪ XT1−1 ) → . . .)[i]; hi S(X, Y ); φ) (9) where the differentials of the complex are connecting maps and φ is given by lemma 4.3.9. 40 / /     / Definition 4.4.3. PM#(f, T → S, T•, Σ) = End∨ F 2(H # ◦ U2)-comod. This carries an exact faithful embedding into F 2-mod. The categories PM#(f, T → S, T•, Σ) fibred over S-schemes. We can form the associated stack M#(f, T → ST•, Σ) for the Zariski topology. This can be constructed explicitly by following the procedure outlined at the end of § 4.1. In order to simplify notation, we usually just write this as M#(T•), when the rest of the data is understood. We let hT• (X, Y )(w) denote the object of this category associated to (X, Y, i, w). We have a functor M#(T•) → C(S, Σ)iso, and a functor p : M#(T•) → M(S, Σ) given as a composition of this with the functor C(S, Σ)iso → M(S, Σ) given by projection onto the second factor. From lemma 4.1.6, we obtain Lemma 4.4.4. M(S, Σ) is equivalent to M#(T•)/ ker p. Proof of theorem 4.4.2. Restriction gives a functor ι : PMcell(S, Σ) → PMeq(S,Σ) which is necessarily exact and faithful. It suffices to show that this is an equiv- alence, because it will then induce an equivalence of the corresponding stacks Mcell(S, Σ) ∼ Meq(S,Σ). We show that ι is essentially surjective and full, and for this it suffices to have a right inverse up to natural equivalence. This is induced by the functor PM#(T•) → Mcell(S) given by (X → S, Y, i, w) 7→ H0(h0 S(XT0, YT0 ∪ XT0−1 ) → h1 S(XT1, YT1 ∪ XT1−1) → . . .)[i] Corollary 4.4.5. There is a Kunneth decomposition for motives associated to objects in ∆eq(S, {S}): hi S(X ×S X ′, X ×S Y ′ ∪ X ′ ×S Y ) ∼= Mj+j′=i S(X, Y ) ⊗ hj′ hj S (X ′, Y ′) Proof. This follows from the theorem and lemma 4.4.1. For objects in ∆eq(S, {S}), we get exterior products S(X, Y ) ⊗ hj′ hj S (X ′, Y ′) → hj+j′ S (X ×S X ′, X ×S Y ′ ∪ X ′ ×S Y ) and cup products S(X, Y ) ⊗ hj′ hj S (X, Y ) → hj+j′ S (X, Y ) by composing this with the restriction to the diagonal. Corollary 2.4.1 shows that these products are compatible with the standard tensor products on the categories of classical and ´etale local systems. 41 4.5 Ind objects Let Ind-A denote the category of Ind-objects of a category A obtained by for- mally adjoining filtered colimits [KS]. This is abelian, when A is [loc. cit.]. This can be given a concrete description in many cases. For example, it is well known that Ind-F -mod can be identified with F -Mod. We extend this to sheaves. Re- call that an object c of an additive category is compact or finitely presented if Hom(c, −) commutes with arbitrary small coproducts. For example, a finite dimensional vector space V is seen to be compact in the category of all vector spaces, because an element of Hom(V, −) is determined by its value on a finite basis. For essentially the same reasons, we have: Lemma 4.5.1. A constructible sheaf is compact in the category Sh(S) of sheaves of F -modules. Proof. For any collection of sheaves F , Gi, we have canonical map κ :Mi∈I Hom(F , Gi) → Hom(F ,Mi∈I Gi) Injectivity can be checked on stalks. Consider the projection pj :M Gi →Y Gi → Gj One checks that φ ∈ Hom(F , ⊕Gi) lies in im(κ) precisely when it has finite support in the sense that there exists a finite subset J ⊂ I such that the germs pi(φs) = 0 for all s ∈ S and i /∈ J. Suppose F is constructible with respect to a necessarily finite Zariski stratifi- cation Σ. Let πσ : σ → σ denote the universal cover of a stratum. Choose bases of cardinality say nσ for each H 0(π∗ σF ). Then φ ∈ Hom(F , ⊕Gi) is determined by its image r(φ) = (π∗ H 0(π∗ σHom(F , ⊕Gi)), σφ) ∈Yσ the map r is injective. In explicit terms, r(φ) is given by a collection i.e. of nσ sections of ⊕π∗ σGi for each σ. The projections r(pj (φ)) are given by simply projecting these sections to Gj. It should now be clear that φ has finite support. Corollary 4.5.2. There is a fully faithful exact embedding of Ind-Cons(S) into Sh(S). Proof. This follows from [KS, prop 6.3.4] and the exactness of filtered colimits. Therefore we have an exact faithful functor Ind-M(S) → Ind-Cons(S) → Sh(S) given by composition. This is also denoted by RB. 42 5 Direct Images 5.1 Direct Images (abstract construction) We start by giving a general construction of direct images. Set DM (S) = D(Ind-M(S)). Fix a morphism f : S → Q. Since the functor f ∗ is exact, it extends to an exact functor on Ind-M(Q) → Ind-M(S). Thus we have an extension f ∗ : DM (Q) → DM (S) as a triangulated functor. Theorem 5.1.1. If f : S → Q is a morphism of quasiprojective varieties, then there is a triangulated functor rf∗ : DM (S) → DM (Q) which is right adjoint to f ∗. Proof. The theorem will be deduced from a form of Brown's representability theorem due to Franke [F]. Note that the extension f ∗ to Ind-M(−) com- mutes with filtered direct limits and therefore coproducts. Since Ind-M(S) is a Grothendieck category by [KS, thm 8..6.5], Franke's theorem [F, thm 3.1] implies that M 7→ Hom(f ∗M, N ) is representable by an object rf∗N . The map N → rf∗N extends to a functor which is necessarily the right adjoint, cf. [N, p223]. Moreover, this is automat- ically triangulated by [Li2, prop 3.3.8]. The abstract construction is not terribly useful by itself. We would really like more: Definition 5.1.2. Let us say that a morphism f : S → Q possesses a good direct image if 1. rf∗(Db(M(Q)) ⊆ Db(M(S)), where we identify DbM with a triangulated subcategory of DM . 2. For each M ∈ M(S). The map RBrf∗M → Rf∗RBM adjoint to the canonical map f ∗RBrf∗M ∼= RBf ∗rf∗M → RBM is an isomorphism. The following lemma gives a criterion for checking this. Lemma 5.1.3. Suppose that r′f∗ : Db(M(S)) → Db(M(Q)) is a functor equipped with natural transformations and such that: η : 1 → r′f∗f ∗ ǫ : f ∗r′f∗ → 1 43 (1) The map adjoint to is an isomorphism. (2) The composition RBr′f∗M → Rf∗RBM RBǫ : f ∗RBr′f∗M → RBM RBM RB ǫ−→ RBr′f∗f ∗M (1) −→ Rf∗f ∗RBM coincides with the adjunction map 1 → Rf∗f ∗. Then r′f∗ is right adjoint to f ∗. So, in particular, f has a good direct image. Proof. It is enough to check that the compositions and f ∗ → f ∗r′f∗f ∗ → f ∗ r′f∗ → r′f∗f ∗r′f∗ → r′f∗ are both identity [M, chap IV]. Since the realizations RB are embeddings, this follows from the compatibility of η, ǫ with the usual adjunctions on the categories of sheaves. As a prelude to a more general result proved later, we show that f : S → Q has a good direct image when it is a closed immersion. By 2.2.10, the map ∆(S, Σ) → M(Q) given by (X → S, Y, i, w) 7→ hi Q(X, Y )(w) induces an exact functor f∗ : M(S) → M(Q). Proposition 5.1.4. If f is a closed immersion, then f∗ satisfies the conditions of lemma 5.1.3. Therefore, f∗ is right adjoint to f ∗. Proof. We have to construct natural transformations η : 1 → f∗f ∗ and ǫ : f ∗f∗ → 1 satisfying the conditions of lemma 5.1.3. We can see from the construction that f ∗f∗hi S(X, Y )(w) is equal to hi S(X, Y )(w). Thus we have a canonical isomorphism, which gives the required map ǫ. This clearly satisfies lemma 5.1.3 (1). Let M orM(Q) denote the category whose objects are morphisms of M(Q), and whose morphisms are commutative squares. Let M or′ ⊂ M or(M(Q)) de- note the subcategory of morphisms M1 → M2 such that there is a commutative diagram RBM1 f∗f ∗RBM1 = ∼= RBM1 / RBM2 44 / /     / commutes. The morphisms are squares M1 M2 h1 h2 M ′ 1 / M ′ 2 such that RBh2 is given by f∗f ∗h1. The functor (M1 → M2) 7→ M1 is clearly faithful and exact. Therefore by corollary 2.2.10, we get a functor M(Q) → M or′ such that hi Q(X, Y )(w) 7→ [hi Q(X, Y )(w) → hi Q(XS, YS)(w)] This gives the canonical adjunction η : 1 → f∗f ∗. Combining this with proposition 4.2.5 yields: Lemma 5.1.5. Let j : S → ¯S be an open immersion with complement i : ¯S − S → S. Then for any F ∈ M( ¯S), there is a canonical exact sequence 0 → j!j∗F → F → i∗i∗F → 0 where i : ∂ ¯S → ¯S is the inclusion. 5.2 Direct Images (conclusion) We come to the main technical result of this paper. Theorem 5.2.1. A morphism f : S → Q possesses a good direct image if either f is projective or Q is a point. The proof, which will be broken into a series of lemmas, is quite messy, although the basic idea is rather simple. The hypothesis of the theorem is used in the following way: a controlled pair (g : X → S, Y ) determines a controlled pair (f ◦ g : X → Q, Y ) when either f is projective or trivially when Q is a point. Let us say that (X → S, Y, i, 0) ∈ ∆(S) is a f -cellular if H j S(X, Y )) is zero for all but one value of j, say j = m. Then the proof will show that hm+i Q (X, Y )[m] will give a model for rf∗(hi S(X, Y ) under RB, so "goodness" is verified in this case. In general, we will realize rf∗M as an explicit complex of f -cellular motives, which maps to Rf∗RBM . Since this construction depends on auxiliary choices, it is necessary work on a bigger category M#, lying over M, in order to get a functor temporarily called q. The final step is to show that q descends to a functor on the derived categories, and that this is indeed the adjoint to f ∗. S(X, Y )). This clearly maps to Rf∗H i Q(H i By factoring f through a closed immersion followed by a projection, and applying proposition 5.1.4, we can see that to prove theorem 5.2.1, we can assume that f : S → Q is flat and therefore equidimensional. Let g : X → S be a quasi-projective morphism with Y ⊂ X closed, such that (X → S, Y ) 45 / /     / • ) which is cellular with respect to the sheaves H ∗ is controlled. By proposition 4.3.8, we can find a quasi-filtration ({Q•}, T → S, T • S(X, Y ). To simplify the discussion, let us suppose that this is simple, i.e. that the cover {Q•} = {Q}. Consider the commutative diagram XT• XT X g f ◦g ❄❄❄❄❄❄❄❄ f T• / T / S / Q where the squares are cartesian. Then we can form a complex of sheaves K• i = H i Q(XT0 , YT0 ∪ XT0−1 ) → H i+1 Q (XT1, YT1 ∪ XT1−1 ) → . . . where the differentials are the connecting maps. Proposition 5.2.2. With the previous assumptions, there is a canonical iso- morphism H j Q(S, H i S(X, Y )) ∼= Hj(K• i ) where Hj stands for the jth cohomology sheaf. Proof. When Q is a point and Y = ∅, this was originally proved in [Ar, thm 3.1]. The general case can be proved by the same method. However, a slightly cleaner alternative is to deduce it from [CM]. Since the model case (Q = pt, Y = ∅) is spelled out in detail in [C], we will be content to give the broad outline. Since both sides of the purported isomorphism are stable under base change to T , we can assume that T = S. Let L = Rf∗Rg∗jXY !F . We consider two filtrations on L. The first is defined by truncations P •(L) = Rf∗τ≤−•Rg∗jXY !F , so that Gr• S (X, Y ). The second F is the filtration on L associated to T•, P L = Rf∗H −• F •L = Rf∗jST•!j∗ ST• Rg∗jXY !F Then Gr• F L ∼= Rf∗jT•T•−1!j∗ T•T•−1 Rg∗jXY !F ∼= Rf∗Rg∗jXT• ,YT• ∪XT•−1 !F holds. Then the cellularity of T• implies that the assumptions of [CM, prop 5.6.1] are satisfied. Therefore we have a natural filtered quasi-isomorphism (L, P ) ∼= (L, Dec(F )) where Dec(F ) is the shifted filtration associated to F . This has the prop- erty that E1(Dec(F )) = E2(F ) (c.f. [De1]). Thus it follows that there is an isomorphism of the spectral sequences associated to P and Dec(F ). The spectral sequence for P is Leray with a shift in indices, and in particular E1(P ) = H j S(X, Y )). On the other hand, we can identify Q(S, H i E1(F ) = H • Q(XT• , YT• ∪ XT•−1), using the fact that H • are controlled. Hence E1(Dec(F )) = E2(F ) = Hj(Ki). Q(XT• , . . .) commutes with base change because the maps 46 / /   / /      / / / We now construct an auxiliary category M# by a variation of the method used in section 4.4. A pair of stratifications Σ of S and Λ of Q will be called f - admissible if each σ ∈ Σ is a fibre bundle over some λ ∈ Λ. Given a stratification Σ′ of S, there exists an admissible pair (Σ, Λ) for which Σ refines Σ′. Any pair of stratifications can always be refined so that admissibility holds. Choose admissible stratifications Σ and Λ. Then composition gives a functor RB ◦ H∗ from C[0,∞)(M(Q, Λ)) to the category Gr Sh(Q((C)) of [0, ∞)-graded sheaves. We also have a functor H ∗ Q ◦ RB : M(S, Σ) → Gr Sh(Q(C)). Let C(S, Σ) be the so called comma category whose objects are triples (K •, M, φ : H ∗ Q ◦ RB(M ) → RB ◦ H∗(K •)) 1 → K • where K • ∈ ObC[0,∞)(M(Q, Λ)) and M ∈ ObM(S, Σ). Morphisms are pairs K • 2 , M1 → M2 satisfying obvious compatibilities. Let Ciso(S, Σ) be the full subcategory consisting of triples for which φ is an isomorphism. We can identify F 2-mod with F -mod × F -mod. There is a faithful exact functor U2 : C(S, Σ) → (F -mod)2 given by (K •, M, φ) 7→ (Qi U (K i)) × U (M ). Given a simple cellular quasi-filtration (T → S, T•) and a stratification Σ, choose base points s for (S, Σ) and t• ∈ T•. We define a functor H # : ∆(S, Σ)op → C(S, Σ)iso as follows. On objects H #(X → S, Y, i, w) = (hi Q(XT0 , YT0 ∪ XT0−1 )(w) → hi+1 Q (XT1 , YT1 ∪ XT1−1 )(w) → . . . ; hi S(X, Y )(w); φ) (10) where the differentials of the complex are connecting maps and φ is given by proposition 5.2.2. We can extend this to nonsimple cellular quasi-filtrations ({Q•}, T → S, T • • ) as follows. For notational simplicity, we assume that the cover consists of two sets {Q0, Q1} and that T 0 • on Q01 = Q0 ∩ Q1. Let j0, j1, j01 denote the inclusions of Q0, Q1 and Q01 into Q respectively. Since the varieties T • • can be extended over Q by taking closures, the extensions by zero • can be refined to T 01 • , T 1 M j α = jα!hi+j Qα (XT α j , YT α j ∪ XT α j−1 )(w) are defined. We can now define H #(X → S, Y, i, w) by taking hi S(X, Y )(w) as the second component as above. For the first component, we use the complex ker[M • 0 ⊕ M • 1 → M • 01] where the map is given the difference of restrictions. This complex is quasi- isomorphic to M • α on Qα. The quasi-isomorphism φ can be thus extended, so that H #(X → S, Y, i, w) ∈ Ciso We let PM#(f, {Q•}, T → S, T • F 2 (H # ◦ U2)-comod, and let M#(f, {Q•}, T → S, T • • , Σ) denote the associated stack. This carries an ex- act faithful embedding into F 2-mod. In order to simplify notation, we usu- ally just write these as PM#(T • • ). We let hT• (X, Y )(w) de- note the object of this category associated to (X, Y, i, w). We have a functor • ) and M#(T • • , Σ) = End∨ 47 M#(T • tors p : PM#(T • extend to functors on M#(T • 2.3.2, we obtain • ) → C(S, Σ)iso. We can compose this with the projections to get func- • ) → C[0,∞)(M(Q, Λ)). These • ) denoted by the same symbols. From corollary • ) → M(S, Σ) and q : M#(T • Lemma 5.2.3. M(S, Σ) is equivalent to M#(T • • )/ ker p. Let • ) → Db(M(Q, Λ)) denote the composition of q with the canonical map. ¯q : M#(T • Lemma 5.2.4. There is a functor r′f∗ fitting into the commutative diagram M#(T • • ) ¯q (PPPPPPPPPPPP r′f∗ / Db(M(S, Σ)) / Db(M(Q, Λ)) such that there is a natural isomorphism RBr′f∗ ∼= Rf∗RB. This functor is independent of the choice of quasi-filtration. Proof. We note that q and ¯q can be extended to Cb(M#(T • • )) by taking the total complex associated to the double complex induced by these functors. By lemma 5.2.3, M(S) is equivalent to M#(T • • )/ ker p. This extends to an equivalence Cb(M(S)) ∼ Cb(M#(T • • )/ ker p). From the definition of Ciso(S), it follows that RB(Hi(q(M ))) ∼= H i Q(S, RB(p((M ))) (11) as functors in M ∈ Cb(M#(T • that Hi ◦ q factors through Db(M#(T • by the commutative diagram • )). Since RB is faithful, this isomorphism implies • )/ ker p). We can summarize all of this Cb(M#(T • • )) ¯q Db(M(Q, Λ)) 5❧❧❧❧❧❧❧❧❧❧❧❧❧ h p Db(M#(T • • )/ ker p) RB )❘❘❘❘❘❘❘❘❘❘❘❘❘ u❧❧❧❧❧❧❧❧❧❧❧❧❧ ∼ e Db(M(S, Σ)) Rf∗◦RB / Db(Sh(Q)) Since e is an equivalence, it has an inverse. The desired functor r′f∗ would be given by h ◦ e−1, but since it depends, a priori, on the the choice of the quasi- filtration, we temporarily denote it by r′f∗,Q,T• . Given a second cellular quasi- filtration (T ′ •), we wish to show that there is a canonical isomorphism r′f∗,Q,T• • . By proposition 4.3.8, we can assume that there is a morphism (T ′ • ) in QV arQ covering the identity id : • → S, T ′ ∼= r′f∗,Q,T ′ •) → (T• → S, T • • → S, T ′ 48   ( / /   )   5 u / S → S. Therefore the complexes defining r′f∗,Q,T• and r′f∗,Q,T ′ • become quasi- isomorphic since the constructions factor through C(id)iso. So we can now omit the second subscript. To finish the proof of the theorem, we will verify the conditions of lemma 5.1.3 which entails constructing adjunctions. Lemma 5.2.5. Given maps g : X → S and f : S → Q of topological spaces, consider a commutative diagram X ×Q S π Γ p X f ◦g S f / Q where Γ is the inclusion of the graph of g. Then the adjunction map f ∗Rf∗Rg∗ → Rg∗ is the composition of the base change map f ∗R(f ◦ g)∗ → Rp∗π∗ and the adjunction Rp∗π∗ → Rp∗RΓ∗Γ∗π∗ = Rg∗ Proof. This follows by applying [Li2, prop 3.7.2ii] to the diagram id X Γ X ×Q S π X id X f ◦g p S f / Q Corollary 5.2.6. If the first base change map is an isomorphism, f ∗Rf∗Rg∗ → Rg∗ can be identified with the map Rp∗π∗ → Rg∗ induced by Γ. Lemma 5.2.7. There is a morphism ǫ : f ∗r′f∗ → 1 compatible with the ad- junction ǫ : f ∗Rf∗ → 1. Proof. The previous corollary implies that Γ∗ : H i S(X ×Q S, Y ×Q S) → H i S(X, Y ) is precisely the adjunction map ǫ. We have to lift this to a morphism ǫ : f ∗r′f∗ → 1. Let Σ and Λ be an f -admissible pair of stratifications. Choose a quasi-filtration T• → S. Define a functor H ♠ : ∆(S, Σ) → F -mod3 49 / /     m m / / /     / /     / by sending (X → S, Y, i, w) to the direct sum of the three vertices of the diagram H0(H i S(XT0 ×Q S, YT0 ×Q S ∪ XT0−1 ×Q S) → . . .) ∼ H i S(X ×Q S, Y ×Q S) Γ∗ H i S(X, Y ) We build a category M♠(S) = EndF 3∨(H ♠)-comod. From the universal prop- erty, we have a functor h♠ which assigns to (X → S, Y, i, w) the diagram of motives H0(hi S(XT0 ×Q S, YT0 ×Q S ∪ XT0−1 ×Q S) → . . .) ∼ hi S(X ×Q S, Y ×Q S) Γ∗ hi S(X, Y ) In particular, we obtain a projection M♠(S) → M(S). As above one can argue that M♠(S)/ ker(p) ∼ M(S). We can see that h♠ factors through ker p. This determines a functor ǫ : M(S) → M orM(S) compatible with adjunction. Lemma 5.2.8. There is a morphism η : 1 → r′f∗f ∗ compatible with the ad- junction η : 1 → Rf∗f ∗ Proof. The strategy is similar to the previous argument. Let Σ and Λ and T• → S be as in the above argument. Define H ♣ : ∆(Q, Λ) → F -mod2 by sending (X → Q, Y, i, w) to the sum of vertices of the diagram H i Q(X, Y ) H i Q(XT0 , YT0 ∪ XT0−1 ) / H i+1 Q (XT1 , YT1 ∪ XT1−1 ) / . . . Q(X, Y ) corresponds to the first component. Let M♣(S) = F 3(H ♣)-comod. As above, we have a functor h♣ sending (X → Q, Y, i, w) partitioned so that H i End∨ to hi Q(X, Y ) hi Q(XT0 , YT0 ∪ XT0−1) / hi+1 Q (XT1 , YT1 ∪ XT1−1 ) / . . . This yields the map η once we observe that M♣(S)/ ker(p) ∼ M(S) where p is the natural projection. 50 o o   o o     / /   / / Proof of theorem. To finish the proof of the theorem, it is enough to observe that by the previous lemmas, we can apply lemma 5.1.3 to conclude that rf∗ = r′f∗. 5.3 Direct image with compact support Theorem 5.3.1. If f : S → Q is an morphism of quasiprojective varieties, then c,Q = rif! : Mc(S; F ) → M(Q; F ), such that RB(rif!(M )) ∼= there is a functor hi If g : S′ → S is a morphism, there is an isomorphism Rif!(S, RB(M )). g∗rif!(M ) ∼= rif!(g∗M ) compatible with the base change isomorphism on re- alizations. Proof. Choose a relative compactification ¯S j ¯f S f Q Then we can define rif! = (r ¯f i defined. First observe that we have ∗) ◦ (j!), provided that we can show that it is well RB(r ¯f i ∗ ◦ j!M ) = R ¯f i ∗ ◦ j!RB(M ) = Rif!RB(M ) as required. To see that it is independent of the choice, first observe that given a second compactification S → S, we can find a third compactification S which dominates both ¯S and S. For example, we can take S equal the closure of diagonal in ¯S ×Q S. Thus we can assume that S = S, and so we can assume that we have a commutative diagram ¯S S π j ❃❃❃❃❃❃❃❃ ✏✏✏✏✏✏✏✏✏✏✏✏✏✏✏ ¯f f f j S Q Therefore, we get a morphism ri ¯f∗j!M → ri f π∗j!M ∼= ri fj!M which induces identity on realization. So it must be an isomorphism in M(Q). Remark 5.3.2. The above construction can be lifted to a functor on derived categories, rf! : DbMc(S; F ) → DbM(Q; F ), by defining rf! = (r ¯f∗) ◦ (j!) using the notation of the proof. 51 / /   / /     O O 6 Motivic Local Systems 6.1 Local Systems Call an object of ∆(S) tame if it is of the form ( ¯X − D → S, E ∩ ( ¯X − D), i, w) such that ¯X → S is smooth and projective, and D + E is a divisor such that any intersection of components is smooth over S (we will refer to this condition as having relative normal crossings). We usually just write E instead of E∩( ¯X −D) above. Such a pair ( ¯X − D → S, E) is a fibre bundle, so it is controlled. It is easy to see that ∆tame(S) ⊂ ∆eq(S, {S}). Definition 6.1.1. The category of premotivic local systems PMls(S; F ) = End∨(H∆tame(S))-comod. The category of motivic local systems Mls(S; F ) is obtained by forming the associated stack as in section 4.1. We note the following properties which are either immediate consequences of what has been said or easily checked. 1. Mls(S) ⊂ Meq(S, {S}) is an abelian subcategory. 2. The realizations RB and Ret take Mls(S) to the categories of locally constant sheaves for the classical and ´etale topologies, and they factor through Mls(S). 3. The tensor product given earlier restricts to a product Mls(S)×Mls(S) → Mls(S). (The key point is that Mls is equivalent to comodules over the restriction of End∨ to ∆cell ∩∆tame.) This induces a product on the stacks Mls(S) × Mls(S) → Mls(S) By item 2 above, we see that Mls(S) is strictly contained in M(S) in general. However, we do observe the following: Theorem 6.1.2. When S = Spec k, M(S; F ) and Mls(S; F ) are equivalent. Proof. By theorem 4.4.2, M(Spec k; F ) is equivalent to Mcell(S; F ). Given (X, Y, i, w) ∈ ∆cell(S), by resolution of singularities we can find a tame object such that ( X, E, i, w) and a map π : X → X which is an isomorphism over X − Y and such that E = π−1Y . Therefore hi(X, Y )(w) ∼= hi( X, E)(w) ∈ Mls. Again by resolution of singularities, any morphism in ∆cell can be lifted to a morphism in ∆tame. This together with lemma 2.2.5 implies the theorem. We outline the construction of Gysin maps, which will be needed later. Given a smooth subscheme ¯Y → S of ¯X transverse to D + E with relative dimension m. Set c = n − m. Then the Gysin homomorphism on cohomology H i S( ¯Y − D, E) → H i+2c S ( ¯X − D, E) can be defined simply by dualizing the restriction under Poincar´e duality. How- ever, this description is not very convenient. A better alternative is to define this via a deformation to the normal bundle as in [BFM]. Let X be the blow 52 up of ¯X × A1 along ¯Y × {0}. Let Y be the strict transform of ¯Y × A1. Let D, E be the preimages of D, E in X. The fibre of the natural map π : X → A1 over t 6= 0 is X. While the fibre π−10 is the union of the projectivized nor- mal bundle p : P(N ⊕ O ¯Y ) → ¯Y and the blow up B of ¯X along ¯Y . Let τ = c1(OP(N ⊕O)(1))c ∈ H 2c(P(N ⊕ O), P(N )). The Gysin map can then be realized as the composition of the given maps H i S( ¯Y − D, E) S(p−1Y − p−1D, p−1E) S p∗ −→ H i ∪τ−→ H i+2c ∼=←− H i+2c ∼=←− H i+2c π∗ t−→ H i+2c S S S (p−1Y − p−1D, p−1E) (π−1(0) − D, π−1(0) ∩ E ∪ B) ( X − D, E ∪ B) ( ¯X − D, E) The second description yields a motivic Gysin map S( ¯Y − D, E) → hi+2c hi S ( ¯X − D, E)(c) (12) We can define the Gysin morphism S( ¯Y − f −1D, f −1E) → hi+2c hi S ( ¯X − D, E)(c) for an arbitrary map f : ¯Y → ¯X as the composition of the Gysin morphism associated to the inclusion of graph of Γf ⊂ ¯Y × ¯X followed by a Kunneth projection. When, S is smooth let V M HS(Sι,an) denote the category of rational vari- ations of mixed Hodge structures on Sι,an, which are admissible in the sense of Steenbrink and Zucker [SZ] and Kashiwara [K]. In a nutshell, an object of this category consists of a filtered local system (V, W ) together with a compatible bifiltered vector bundle with connection (V ∼= V ⊗OS, W, F, ∇) subject to the ap- propriate axioms (Griffith's transversality...). For the precise conditions, see [PS, sect. 14.4.1] or the above references. Given (X = ¯X − D → S, E, i, 0) ∈ ∆tame, we can construct an admissible variation as follows: S(X, E ∩ X; Q) ¯X/S(log D + E)(−E) ¯X/S(log D + E)(−E) ¯X/S(log D + E)(−E) V = H i V = Rf∗Ω• F p = im Rf∗Ω≥p Wq = im Rf∗WqΩ• ∇ = Gauss-Manin connection   jX,E!QX−E → QX →M QEi →M QEi∩Ej . . . 53 This is given in [SZ], when E = ∅. The general case is easily reduced to this via the resolution where E = ∪Ei is the decomposition into irreducible components. We can extend this to arbitrary objects (X = ¯X − D → S, E, i, w) ∈ ∆tame by tensoring the above variation with Q(w). This construction is easily checked to yield a functor ∆tame(S)op → V M HS(S). Thus we get Example 6.1.3. an exact faithful Hodge realization functor Rι,H = RH : Mls(S; Q) → V M HS(Sι,an) This functor is compatible with tensor product. This coincides with the Hodge re- alization constructed earlier, restricted Mls(S; Q), once we identify V M HS(S) ⊂ Cons-M HM (S). One of the consequences of the admissibility conditions mentioned above is the following removable singularities theorem: An admissible variation extends from a Zariski open to the whole variety if the underlying local system extends. Using this, it is possible to prove a stronger statement that RH extends to all of Mls(S). We can define a system of realizations on S by following the usual pattern [De2, J1]. Here we outline the construction. A "locally constant" or more cor- rectly lisse ℓ-adic sheaf V on Set corresponds to a representation of the algebraic fundamental group πet 1 (S) → GLN (Qℓ). Composing this with the canonical map from the topological fundamental group κ : π1(Sι,an) → πet 1 (S) results in a local system κ∗ ι V of Qℓ-modules on Sι,an. By a system of realizations we will mean 1. A collection of locally constant ℓ-adic sheaves Vℓ on Set, for each prime ℓ. Each Vℓ should be mixed in the sense that they carry weight filtrations. 2. A collection of variations of mixed Hodge structures Vι on Sι,an indexed by embeddings of ι : k ֒→ C. 3. Compatibility isomorphisms κ∗ ι Vℓ ∼= Vι ⊗ Qℓ respecting weight filtrations. These form a Q-linear abelian category SR(S). An appeal to corollary 2.2.10 and the comparison theorem (appendix B) yields a realization functor RSR : Mls(S, Q) → SR(S) which combine all of the previous realizations into one. Thus Mls gives a finer theory than motives built from systems of realizations. 6.2 Duality The goal of this section is to prove: Theorem 6.2.1. Mls(S; F ) is a neutral Tannakian category over F Corollary 6.2.2. Mls(S, Q) is equivalent to the category of representations of a proalgebraic group (which we refer to as the Tannakian dual of this category). To be more explicit, after choosing a base point s ∈ S(¯k), we obtain a so called fibre functor Fs : Mls(S, Q) → Q-mod given as the composition of RB with the stalk at s. Setting πmot (X, s) to the group of tensor automorphisms of Fs, we have that πmot (X, s) is proalgebbraic and that Mls(S, Q) is equivalent 1 1 54 to the category of representations of it. The methods of [Ar2] show that this carries more structure, but the details will be spelled out elsewhere. As to the theorem's proof, we know that Mls(S; F ) is a tensor category over F with a tensor preserving fibre functor. What remains to be proven is that every object has a dual. By proposition 2.4.3 it is enough to construct duals for objects of the graph ∆tame(S). We will show that S( ¯X − D, E)(w)∨ = h2n−i hi S ( ¯X − E, D)(−w + n) (Dual) where n is the relative dimension of ¯X → S. As first step, we note the following form of Poincar´e duality. Lemma 6.2.3. There is a pairing H i S( ¯X − D, E) ⊗ H 2n−i S ( ¯X − E, D) → FS which is perfect in the sense that it induces an isomorphism of local systems H i S( ¯X − D, E) ∼= H 2n−i S ( ¯X − E, D)∗ Proof. This follows from Verdier duality [I] H i S( ¯X, Rj( ¯X,D)∗j( ¯X,E)!F ) ∼= H −i S ( ¯X, DRj( ¯X,D)∗j( ¯X,E)!F )∗ ∼= H −i S (j( ¯X,D)!Rj( ¯X,E)∗F [2n])∗ ∼= H 2n−i S ( ¯X − E, D)∗ The next task is to realize the above pairing geometrically by a morphism of Mls. When D = E = ∅, we can take the cup product pairing which is induced by the diagonal embedding into the product. In general, we need to blow up the product to get a well defined diagonal. Set Y = ¯X × ¯X, D1 = D × ¯X, D2 = ¯X × D, E1 = E × ¯X and E2 = ¯X × E. Let Y be obtained by blowing up Y along D1 ∩ D2 and then along the intersection of the strict transforms of E1 and E2. Let G be the exceptional divisor of Y → Y . Denote the strict transforms of Di, Ej by Di, Ej. The diagonal embedding ¯X → Y extends to an embedding of d : ¯X → Y (it is not necessary to blow up X since D and E are already divisors). The image of d is disjoint from Di, Ej and d−1G ⊆ D ∪ E. We define ǫ : hi S( ¯X − D, E) ⊗ h2n−i ( ¯X − E, D)(n) → FS S by the composition of S( ¯X − D, E) ⊗ h2n−i hi S ( ¯X − E, D) → h2n → h2n ∼=← h2n → h2n ∼=→ h2n S (Y − (D1 ∪ E2), E1 ∪ D2) S ( Y − ( D1 ∪ E2 ∪ G), E1 ∪ D2) S ( Y − ( D1 ∪ E2), E1 ∪ D2 ∪ G) S ( ¯X, E ∪ D) S ( ¯X) ∼= F (−n) 55 after twisting by F (n). The middle isomorphism is excision. For the last isomor- phism, by projection we can reduce to the case X = Pn S)n, where it follows from Kunneth. S and then to X = (P1 To construct δ, we dualize the above description using Gysin maps in place of pull backs: FS = h0 S( ¯X, E ∪ D) → h2n ∼=← h2n → h2n → hi S ( Y − ( D1 ∪ E2), E1 ∪ D2 ∪ G)(n) S ( Y − ( D1 ∪ E2 ∪ G), E1 ∪ D2)(n) S (Y − (D1 ∪ E2), E1 ∪ D2)(n) S( ¯X − D, E) ⊗ h2n−i ( ¯X − E, D)(n) S To prove (Dual), we have to establish equations (D1) and (D2). It is enough ( ¯X − to verify these on the corresponding vector spaces H i E, D)s, and this becomes an exercise in linear algebra. If ej is a basis of the first space, and ej the dual basis of the second, then S( ¯X − D, E)s, H 2n−i S eℓ ⊗ eℓ δ(1) =Xℓ ǫ(X ajℓej ⊗ eℓ) =X ajj Therefore (ǫ ⊗ id) ◦ (id ⊗ δ)(X ajej) = (ǫ ⊗ id)(Xjℓ ajej ⊗ eℓ ⊗ eℓ) =X aℓeℓ proves (D1). The remaining equation is similar. 6.3 Pure Objects and Weights We work in M(S, Q) throughout this section. Let f : X → S be a smooth pro- jective map of relative dimension n. Fix an embedding X ⊂ PN S . The standard generator c1(O(1)) ∈ H 2(PN ) induces an isomorphism QS(0) ∼= h2 S )(1). This yields a map QS(0) → h2 S(X)(1) by restriction. Cupping with this induces the Lefschetz operator ℓ : hi S (X)(1). The isomorphism S(X) → hi+2 S(PN ℓi : hn−i S (X) ∼−→ hn+i S (X)(i) follows from the usual hard Lefschetz theorem on the corresponding sheaves. Therefore we get, as usual, the Lefschetz decomposition hi S(X) ∩ ker ℓn−i+1. This allows us to define the Hodge involu- where pi(X) = hi tion ∗ = ∗H on h∗ S(X) = ⊕hi S(X) by the formula in [A1, pp 10-11]. Note that the induced involution on the cohomology of a fiber H ∗(Xs, C) coincides with the Hodge star operator with respect to the Fubini-Study metric (up to a factor and complex conjugation) [loc. cit.] S(X) = ⊕ℓkpi−2k(X)(−k), Proposition 6.3.1. The algebra End(h∗ S(X)) is semisimple. 56 Proof. Set a′ = ∗at∗, where at is the transpose (c.f. [K, 1.3]). With the help of the Hodge index theorem, we see that the bilinear form trace(ab′) is positive definite (compare [K, p. 381]). Then the criterion of [K, 3.13] shows that the algebra is semisimple. Definition 6.3.2. Call an object of Mls(S) pure (of weight i) if it is a finite sum of summands of motives h∗ S(X)) with X → S smooth and projective. Let Mpure(S) ⊂ Mls(S) (Mpure,i(S) ⊂ Mls(S)) be the full subcategory of pure objects. S(X) (or hi Theorem 6.3.3. Mpure(S, Q) and Mpure,i(S, Q) are semisimple abelian sub- categories of Mls(S, Q). The Hodge realization RH takes Mpure(S, Q) (respec- tively Mpure,i(S, Q)). to the category of pure polarizable variations of Hodge structure HS(S) (of weight i respectively) There is a direct sum decomposition Mpure(S, Q) = Li Mpure,i(S, Q), i.e. every object and morphism on the left decomposes into a sum as indicated. Furthermore Mpure(S, Q) is a Tannakian subcategory. Proof. These are abelian and semisimple by [J2, lemma 2] and the previous proposition. The second statement is clear. The third statement follows im- mediately from the previous two. It is easy to see from the constructions that Mpure(S, Q) is closed under tensor product and duals. Corollary 6.3.4. The Tannakian dual of Mpure(S, Q) is proreductive. Theorem 6.3.5. There are exact functors grj : Mls(S, Q) → Mpure,j(S, Q) which splits the inclusions Mpure,j(S, Q) ⊂ Mls(S, Q). These are compatible with the Hodge realizations in the sense that RH grj = GrW j RH . Proof. It is enough to define gr =Lj grj, and then set grj to the composition of this with the projection Mpure(S) → Mpure,j(S). Fix a smooth projective map ¯X → S with a relative normal crossing divisor S( ¯X − D, E, C) can be computed by taking the direct image of D + E. Then H ∗ the double complex Ω• ¯X/S(log D) →Mi Ω• Ei/S(log D) →Mi<j Ω• Ei∩Ej/S(log D) → . . . When restricted to the fibres, this complex forms part of a differential graded co- homological mixed complex [De1]. From this we can deduce a spectral sequence associated to the diagonal filtration (c.f. [De1, 7.1.6, 8.1.19.1]) S(Y (r) q ) ⇒ H b−a S ( ¯X − D, E, Q) (13) E−a,b 1 where H p = Mp+2r=b,q−r=−a q =(D(r) Y (r) E(q−1) ∩ D(r) if q = 0 if q > 0 57 and D(r) and E(r) are disjoint unions of r + 1-fold intersections of components of D and E. We see also that (13) degenerates at E2, and the induced filtration on the abutment is the weight filtration. For later use, we record the precise formula for Wi+kH i S( ¯X − D, E, C) im H i(WkΩ• ¯X/S(log D) →M Wk+1Ω• Ei/S(log D) → . . .) (14) The differentials of (13) are sums of restrictions and Gysin maps. So we can regard this as a spectral sequence of variations of Hodge structures E−a,b 1 = Mp+2r=b,q−r=−a H p S(Y (r) q )(−r) ⇒ GrW H b−a S (X, Q) (15) As noted above, the differentials are sums of restrictions and Gysin maps. The motivic versions of Gysin maps were defined in (12) of section 6.1. Thus we can form a graded complex of motives in Mpure e−a,b = Mp+2r=b,q−r=−a hp S(Y (r) q )(−r) which maps to the left side of (15) under RH . We would like to take gr(hi( ¯X − D, E)) to be the sum ⊕h−a(e•,i+•), but at the moment this not well defined. It depends on the choice of compactification, so we denote it by Gi( ¯X, D, E). We build a graph ∆(S), with a forgetful functor π : ∆(S) → ∆(S), whose objects consist of such compactifications, along with labels i, w ∈ Z. Any two compactifications are dominated by a third. Therefore the fibres of π are connected. From corollary 2.2.4, it follows that End∨(H ◦π) ∼= End∨(H). Let C be the category of triples (A, B, φ), where A ∈ Mls(S), B ∈ ∗ RH A ∼= RH B. There is an exact faithful functor U2 : Mpure(S), and φ : GrW C → Q2-mod taking (A, B, φ) to U (A) × U (B), where U : Mls(S) → Q-mod is the forgetful functor. We have a functor from H # : ∆(S) → C sending a labelled compactification ( ¯X, i, w) to (hi( ¯X − D, E)(w), Gi( ¯X, D, E)(w), φ), where φ is the natural isomorphism of Hodge realizations from (15). We can form the category PM# of comodules over End∨(U2◦H #). This has a natural projection p : PM# → End∨(H ◦ π)-comod. There is an equivalence PM#/ ker p ∼ PM(S). We have a functor G : PM# → Mpure(S) which factors through this equivalence, and this yields gr. This leads to a theory of weights in Mls. Let us say that an object M ∈ Mls(S, Q) has weight(s) in I ⊂ Z if grjM = 0 for j /∈ I. For M ∈ Mls(S, Q). Define WkM to be the maximal subobject of M with weights ≤ k. Note that this exists because Mls(S, Q) embeds into Q-mod, so it is noetherian. Theorem 6.3.6. For all k, one has 1. WkM/Wk−1M ∼= grkM , 2. Wk is strictly preserved by morphisms, 58 3. RH (WkM ) = WkRH (M ). We first need: Proposition 6.3.7. Given M ∈ M(S) and j ∈ Z, there exists N ⊆ M and N ′ ⊆ M so that N ′ has weights < j and N/N ′ ∼= grjM . S( ¯X −D, E) with ¯X smooth, Proof. By lemma 2.2.5, we can assume that M = hi and D + E a divisor with relative normal crossings. In principle, the proof amounts to realizing the formula for W given in (14) by a motive. When E = 0, this is easy to do directly. Let D(r) denote union of (dim(D/S) − r + 1)- fold intersections of components of D, with D(−1) = ∅. The point is that dim(D(r)/S) = r. Then (14) reduces to Wi+kH i( ¯X − D, C) = im H i(WkΩ• X (log D)) = im H i(X − D(k), C) Therefore N (respectively N ′) may be taken as im[hi( ¯X − Dk−1) → hi( ¯X − D)] with k = j − i (respectively k = j − i − 1). The general case, while feasible, is rather messy to write explicitly. So in- stead, we finish the proof by induction on d = dim( ¯X/S). Once we have estab- lished this for a given d, it follows the proposition holds for all motives generated as an abelian category by varieties of dimension at most d. So now consider the sequence hi−1(E − D) → hi( ¯X − D, E − D) → hi( ¯X − D) By induction, we can find NE, N ′ this motive . Then E ⊂ hi−1(E − D) satisfying the proposition for N = im NE + im[hi( ¯X − D(j−i−1), E) → hi( ¯X − D, E)] N ′ = im N ′ E + im[hi( ¯X − D(j−i−2), E) → hi( ¯X − D, E)] will satisfy the proposition for M . Proof of theorem. Let k be the least weight of M . The canonical map ι : grk(WkM ) → grkM is a monomorphism, since grk is exact. The previous proposition shows that ι is also an epimorphism and hence an isomorphism. Ap- plying the same argument to the quotients M/WjM establishes part 1. The fil- tration Wk is functorial by construction. Strictness follows by what has just been proved (cf [De1, §1]). Finally part 3 follows immediately from theorem 6.3.5. From the construction, we can deduce the following: Proposition 6.3.8. The total functor gr : Mls(S) → Mpure(S) is an exact tensor functor. Corollary 6.3.9. The Tannkian dual of Mls(S) is a semidirect product of the Tannakian dual of Mpure(S) with another group. 59 6.4 Andr´e's category of motives Andr´e [A1] has given an entirely different construction of pure motives over a field k that we recall. Given a smooth projective variety X ∈ V ark, a class in H 2n(X, Q) is called motivated cycle of degree n if it can be expressed as p∗(α ∪ ∗β), where α, β are algebraic cycles on a product X × Y , with Y smooth and projective, and p : X×Y → X is the projection. Let An mot(X) denote the set of these classes. It contains the space of algebraic cycles and would coincide with it assuming Grothendieck's standard conjectures. Andr´e's category of motives MA(k) is built by taking as objects triples (X, n, p) with X smooth projective, n ∈ Z, and p an idempotent in the ring of motivated cycles on X×X. Morphisms are given by HomMA((X, n, p), (Y, m, q)) = pAn−m mot (X × Y )q Andr´e proved that this category is semisimple Tannakian. The construction of MA and this result was extended to more general smooth bases in [AD]. For simplicity, we concentrate on the case of S = Spec k, Given a smooth projective variety X, let hA(X) ∈ X denote the object represented by the A(X) by (X, n, πi), where πi : H 2 dim X−i(X) ⊗ H i(X) is triple (X, 0, id) and hi Kunneth component of the diagonal. Then hA(X) = ⊕hi A(X). By construction, we have an exact faithful embedding RB : MA(k) → Q-mod, sending (X, n, p) to pH ∗(X). In particular, RB(hi(X)) = H i(X). The category MA can be also be constructed by first forming the category M otCor of smooth projective varieties and motivated correspondences, and then taking the pseudo-abelian (or idempotent) completion, and then formally inverting Tate. Theorem 6.4.1. For any smooth S, the categories MA(S) and Mpure(S, Q) are equivalent. We will give the proof when S = Spec k for simplicity, even though the arguments works in general. We will refer to a cohomology class in H i(X, Q) as a Nori cycle of weight 2j if it lies in the image of HomM(k)(Q(−j), hi(X)). Proof. This is broken into a series of steps. 1. Motivated cycles are Nori cycles. Proof. Algebraic cycles are certainly Nori cycles. In general, motivated cycles are built from algebraic cycles by applying ∗, ∪, p∗. Each of these operations preserves the space of Nori cycles. 2. There is an functor ι : MA → Mpure taking hi A(X) to hi(X). This functor commutes with RB. Proof. By 1, the map on objects X → ⊕hi(X) gives a functor ι′ : M otCor → Mpure. Since M (k) is abelian and the Tate motive is invertible, ι′ extends uniquely to a functor ι : MA → Mpure. The final statement follow more or less automatically from RB(hi A(X)) = H i(X) = RB(hi(X)). 60 3. There is a functor grA : M → MA taking hi(X) → hi A(X). This satisfies grA ◦ ι = id and it commutes with RB. Proof. The functor grA is constructed by substituting MA for Mpure in the proof of theorem 6.3.5. 4. The functor ι takes simple objects to simple objects. Proof. Suppose that M ∈ MA is simple, but ι(M ) is not. We may write ι(M ) = N1 ⊕ N2 with Ni 6= 0. Then M = grA(N1) ⊕ grA(N2) which leads to a contradiction. 5. The set of simple objects of MA and Mpure are in one to one correspon- dence via ι and grA. Proof. Write hi A(X) = ⊕Mj, with Mj simple. Then hi(X) = ⊕ι(Mj) gives a decomposition into simple objects by 4. Since every simple object of Mpure is a summand of some hi(X), with X smooth and projective, this proves the claim. 6. If M ∈ MA is simple, then End(M ) ∼= End(ι(M )). Proof. The map g : End(ι(M )) → End(M ), induced by grA, is a sur- jective homomorphism because ι gives a splitting. Since End(ι(M )) is a division ring, g is necessarily an injection as well. 7. Given M, N ∈ MA, Hom(M, N ) ∼= Hom(ι(M ), ι(N )). Proof. Decompose M = L M mj j ⊕ N ′ , such that Mj are distinct simple objects and M ′, N ′ have no simple factors in common. Let Dj = End(Mj). Then j ⊕ M ′ and N = L M nj Hom(M, N ) =Y M atnjmj (Dj) = End(ι(M )) 61 7 Nori's Hodge conjecture 7.1 Conjecture over C As is well known, the usual form of the Hodge conjecture is equivalent to the statement that the Hodge realization of homological pure motives is a full faith- ful embedding [Sv, p 405]. In a nutshell, this comes down to the observation that given smooth projective varieties X and Y , a morphism HomMHS(H i(X), H i(Y )) is a Hodge cycle on X × Y and therefore a correspondence, assuming the con- jecture. The analogous statement in the present setting is due to Nori. Conjecture 7.1.1 (Nori). The Hodge realization RH : M(C, Q) → M HS is a full faithful embedding. This would imply that the canonical mixed Hodge structure is "Galois invari- ant" in the following sense: if H i(X) ∼= H i(Y ) in MHS, then H i(X σ) ∼= H i(Y σ) for any σ ∈ Aut(C). Since M(C) = Mls(C) and M HS are Tannakian, we can rewrite Hom(A, B) = Hom(Q, A∗ ⊗ B), and reformulate the last conjecture as Conjecture 7.1.2. The map HomM(C)(Q, M ) → HomMHS(Q, RH (M )) is surjective for each M ∈ M(C). In particular, a Hodge cycle on any complex algebraic variety X is a Nori cycle. When M lies in Mpure(C), this is implied by the usual Hodge conjecture, but in general, it is neither weaker nor stronger than the Hodge conjecture. It should be viewed as refinement of Deligne's conjecture that Hodge cycles are absolute [DMOS]. To understand this from a different perspective, let us recall that the original form of Beilinson's Hodge conjecture [B1] would imply that the regulator map on the higher Chow group reg : CH a(X, b) ⊗ Q → HomMHS(Q(−a), H 2a−b(X)) is surjective for all a, b. The conjecture is known to be overly optimistic in general (cf [J2]), but it is expected for instance when X is defined over ¯Q. The map can be made explicit as follows. An element α on the left is given a cycle in Bloch's complex [Bl], and so it possesses a fundamental class in reg(α) ∈ H 2a(X × Ab, X × ∂Ab)(a) ∼= H 2a−b(X)(a) where Ab is thought of as a simplex with boundary ∂Ab. It is clear from this, that we may factor reg through HomM(C)(Q(−a), H 2a−b(X)) → HomMHS (Q(−a), H 2a−b(X)) Thus the truth of Beilinson's conjecture, in cases where it is expected, would imply the truth of Nori's. 62 Theorem 7.1.3 (Andr´e [A1]). Hodge cycles on abelian varieties are motivated. So we deduce: Corollary 7.1.4. Conjecture 7.1.2 holds for any variety whose motive lies in the tensor category generated by abelian varieties. 7.2 Conjecture over general bases We can formulate an ostensibly stronger form of Nori's conjecture. Conjecture 7.2.1. Given a smooth complex variety S, (a) the Hodge realization RH : Mls(S, Q) → V M HM (S) is a full faithful embedding. (b) if M ∈ Mls(S), the map HomMls(S)(QS, M ) → HomV MHS (QS, RH (M )) is surjective. As above, we note that (a) and (b) are equivalent because Mls(S) is Tan- nakian. We will prove that the conjectures for a general base follows from the earlier ones. As a first step, let us suppose that we have motive M ∈ Mls(S), then by theorem 5.2.1, we have a good direct image p∗M = r0p∗M ∈ M(C), where p : S → Spec C is the canonical map. Restricting the adjunction map p∗p∗M → M to s ∈ S, yields map p∗M → Ms. Under Betti realization RB(p∗M ) → RB(Ms) can be identified with the inclusion of the subspace of π1(S, s)-invariants of the fibre of the local system RB(M ). As an aside, we observe that this leads to a direct construction for basic examples: Lemma 7.2.2. If M = hi S(X, E), then p∗M ∼= im[hi(X, E) → hi(Xs, Es)] Proof. Let I denote the right hand side. Consider the inclusion X → X × S of S-schemes given by the graph of X → S. This induces a map hi(X, E) = p∗hi It is suffices to prove that this is an isomorphism of Betti realizations. For this, apply the global invariant cycle theorem [BBD, 6.2.8]: S(X × S, E × S) → p∗M , which is a morphism I → p∗M . RB(I) = im[H i(X, E) → H i(Xs, Es)] = H i(Xs, Es)π1(S,s) Theorem 7.2.3. Suppose that S is smooth and connected with a point s ∈ S. Given M ∈ Mls(S), if conjecture 7.1.2 holds for p∗M then conjecture 7.2.1(b) holds for M . 63 Proof. Let us suppose that conjecture 7.1.2 holds for I = p∗M Any mor- phism γ ∈ Hom(QS, RH(M )) gives a π1(S, s) invariant weight Hodge cycle on H i(Xs, Es). Thus γ lies in Hom(QS, RH (I)). So it must come from a mor- phism Q → I by our assumption. This induces a map of pullbacks QS → p∗I, which when composed with the adjunction map ǫ : p∗I → M gives the desired lift of γ to γ′ ∈ HomMls(S)(QS, M ). Corollary 7.2.4. Conjecture 7.1.2 implies conjecture 7.2.1. Corollary 7.2.5. Conjecture 7.2.1(b) for any motive that lies in the tensor category generated by relative smooth curves over S. By a similar argument we obtain an analogue of Deligne's "principle B" [DMOS] in the theory of absolute Hodge cycles. Proposition 7.2.6. Given a tame family (f : X → S, E), a π1(S, s)-invariant Nori cycle in H i(Xs, Es) yields, under parallel transport, a Nori cycle in every fibre H i(Xt, Et). Proof. A π1(S, s)-invariant Nori cycle on H i(Xs, Es) induces a morphism Q(j) → p∗hi(X, E). This can specialized to any fibre. Appendices A 2-categories We will generally take the view that a category is the same as any other category equivalent to it. This needs some elaboration. The category Cat of all small categories is a 2-category [M]. Among other things, this means that the set of functors HomCat(C, D) is itself a category, where the morphisms are natural transformations. In this setting, functors are usually called 1-morphisms and natural transformations 2-morphisms. Each kind of morphism can be composed as usual. This is denoted by ◦. There are identities denoted by 1X etc. There is another kind of composition for adjacent 2-morphisms, denoted here by ⋄. Given objects A, B, C, 1-morphisms F, F ′, G, G′ and 2-morphisms α, β as indicated below F F ′ A α B B β C G G′ 64   D D    D D  The composition β ⋄ α sits as follows F ′◦F A β⋄α C G′◦G It is simply given as the composition F ′(F (x)) F ′(αx) −→ F ′(G(x)) βG(x)−→ G′(G(x)) This operation is associative, and the additional identities α ⋄ 11A = 11A ⋄ α = α, 1F ⋄ 1G = 1F ◦G (β′ ◦ β) ⋄ (α′ ◦ α) = (β′ ⋄ α′) ◦ (β ⋄ α) are satisfied. In Cat, we can either require that equations (or diagrams) hold (or commute) strictly, i.e. on the nose, or only up to a natural isomorphism. The latter is frequently the more usual occurence. Recall, for example, that categories C and D are equivalent (respectively isomorphic) if there are functors F : C → D and G : D → C such that F ◦ G ∼= 1D and G ◦ F ∼= 1G (respectively F ◦ G = 1D and G ◦ F = 1G). Given a category C, a pseudofunctor F : C → Cat is an assigment of objects to objects, and morphisms to 1-morphisms, together with natural isomorphisms, ǫc : F (1c) ∼= 1F (c) and ηf,g : F (f ) ◦ F (g) ∼= F (f ◦ g). These are required to satisfy certain commutivities that ensure that any two isomorphisms F (f1) ◦ F (f2) ◦ . . . F (fn) ∼= F (f1 ◦ f2 ◦ . . . fn) built from η, ǫ coincide. pseudofunctors are simply pseudofunctors on Cop. It suffices to check this for n ≤ 3. Contravariant There are number of related notions of colimit (= direct limit) of categories. We single out the notion that is most useful for us and refer to it as a 2-colimit, although "pseudo-colimit" or something like that may conform better to current usage. To simplify matters, we discuss 2-colimits in the filtered setting where things are easier (cf. [SGA4, exp VI §6]); this reference also gives a more general construction. A category D is filtered if for any two objects d1, d2, there exists an object d3 and morphisms d1 → d3, d2 → d3, and for any two parallel morphisms f, g : d1 → d2, there exists a morphism h : d2 → d3 such that hf = hg. For example, a partially ordered set is filtered precisely when it is directed. Given a pseudofunctor F : D → Cat with D filtered, the 2-colimit L = 2- lim−→d F (d) is the category whose set of objects is the disjoint union ObL =ad ObF (d) 65   D D  The set of morphisms from A ∈ ObF (d1) to B ∈ ObF (d2) is given by the filtered colimit lim−→ f :d1→d3,g:d2→d3 HomF (d3)(F (f )(A), F (g)(B)) We can see that there is a family of 1-morphisms {F (d) → L} such that for f ∈ Hom(d, d′) F (d) L =④④④④④④④④ F (f ) F (d′) commutes up to natural isomorphism. Moreover L would be universal in the sense that for any category L′ with a family {F (d) → L′} as above, there is a unique 1-morphism L → L′ such that the appropriate diagrams nonstrictly commute. We really want to consider 2- lim−→d F (d) only up to equivalence of categories. In practice, there may be other representations of the colimit which are more natural than the original construction. Lemma A.0.7. Suppose that Ci ⊂ C is a directed family of subcategories of a given category. Then 2- lim−→ Ci is equivalent to the directed union S Ci which the category having S ObCi and S M orCi as its set of objects and morphisms. Example A.0.8. Suppose E is a coalgebra over a field F . We can express it as directed union, and therefore a 2-colimit, of finite dimensional coalgebras E = lim−→ Ei . From the earlier descripition, it is not difficult to deduce the following: Proposition A.0.9. Let F be a pseudofunctor from a filtered category D to the 2-category of abelian categories. Suppose that that F (f ) is exact for each f ∈ M orD. Then 2- lim−→d F (d) are exact. F (d) is abelian and the functors F (d′) → 2- lim−→d F (d) and the functors F (d′) → L are additive. Sketch. It is clear that L = 2- lim−→d Given a morphism in L represented by f : A → B in F (d), ker(f ), coker(f ) and A ֌ im(f ) ։ B represents the kernel, cokernel and image factorization in L. In a similar vein: Proposition A.0.10. Let F be a pseudofunctor from a filtered category D to the 2-category of triangulated categories (with t-structure) and (exact) triangulated functors. Then 2- lim−→d F (d) carries the structure of a triangulated category (with t-structure) so that the functors F (d′) → 2- lim−→d F (d) are triangulated (and exact). 66 / /   = B Comparison theorem Let X be a C-variety. Define the site Xcl with objects given by local home- omorphisms U → Xan and coverings are surjective families {Ui → U }. Then there is an obvious map of sites Xcl → Xan , which induces an equivalence of the categories of sheaves [SGA4, exp XI §4]. In particular, the cohomologies are the same. There is a canonical morphism of sites ǫ : Xcl → Xet which induces a map from ´etale to classical cohomology. Since ´etale cohomology does not work properly for nontorsion coefficients, we start with finite coefficients, and then take the limit. Choose N > 0. A sheaf of Z/N Z-modules is constructible for either topology if there is a decomposition of X into Zariski locally closed sets, for which the restrictions are locally constant. The pullback ǫ∗ preserves constuctibility. The following comparison theorem is given in [SGA4, exp XVI, thm 4.1; exp XVII, thm 5.3.3]: Theorem B.0.11. Suppose that f : X → Y is a morphism of C varieties, and that F is a constructible sheaf of Z/N Z-modules there are isomorphisms ǫ∗Rf et,i ǫ∗Rf et,i ∗ F ∼= Rf an,i ! F ∼= Rf an,i ∗ ! ǫ∗F ǫ∗F Fix a prime ℓ. A constructible ℓ-adic sheaf is a system . . . Fn → Fn−1 . . . of sheaves on Xet, such that each Fn is a constructible Z/ℓnZ-module and the maps induce isomorphisms Fn ⊗ Z/ℓn−1Z ∼= Fn−1. Standard sheaf theoretic operations can essentially be defined componentwise, and they work as expected [SGA5, SGA4h]. Given a constructible sheaf F = {Fn} on Xet, define ǫ∗F = lim←− n ǫ∗Fn ǫ∗(F " ⊗ "Qℓ) = (lim←− n ǫ∗Fn) ⊗ Qℓ on Xcl. Then with this notation, the above theorem extends to the case where F is an ℓ-adic sheaf (⊗Qℓ) [BBD, §6.1]. C Classical t-structure for mixed Hodge mod- ules Saito [S1, S2] has introduced a category of mixed Hodge modules1 M HM (S). When S is nonsingular, an object M of this category consists of a filtered per- verse sheaf (K, W ) of Q-vector spaces on San together with compatible bifiltered regular holonomic DS-module (M, W, F ). These are subject to a rather deli- cate set of conditions that we will not attempt to spell out. The definition is inductive. In particular, when S is a point, M HM (S) is nothing but the 1To avoid confusion, we note that we are following the conventions of §4 of [S2]. 67 category of polarizable mixed Hodge structures. One has a forgetful functor rat : M HM (S) → P erv(San) to the category of perverse sheaves given by M 7→ K. Saito has established the following properties: 1. There is an exact faithful functor rat : M HM (S) → P erv(San) for any S. It extends to a triangulated functor DbM HM (S) → DbCons(San). 2. M HM (S) contains the category of admissible variations of mixed Hodge structure. In fact, M is a variation if and only if rat(M ) is a local system up to shift. 3. Standard sheaf theoretic operations extend to DbM HM (S) including Grothendieck's "six operations" and vanishing cycles functors. These are compatible with the corresponding operations on DbCons(San) via rat. The most natural t-structure on DbM HM (S) has M HM (S) as its heart. This corresponds to the perverse t-structure on the constructible derived cat- egory, so we refer to this as the perverse t-structure on M HM . Saito [S2, remark 4.6] has pointed out that there this a second t-structure that we call the classical t-structure which lifts the standard t-structure on DbCons(San) with Cons(San) as its heart. Theorem C.0.12. There exists a nondegenerate t-structure (cD≤0, cD≥0) on DbM HM (S) which is compatible with the standard t-structure on DbCons(San). Proof. Let ix : x → X denote the inclusion of a point. Define M ∈ Ob(cD≤0) (respectively ∈ Ob(cD≥0)) if i∗ xM = 0 for all x ∈ X and k > 0 (respectively k < 0). To see that this is a t-structure, note that it is enough to check this on each step of the filtered union DbM HM (S) = S DbM HM (S, Σ), where M HM (S, Σ) ⊂ M HM (S) denotes the full subcategory consisting of mixed Hodge modules such that rat(M) is Σ-construcible. One can now use induction on the cardinality Σ. If Σ = 1, the purported t-structure is in fact what it is claimed to be since it is the perverse t-structure up to shift. When Σ > 1, let T be a closed stratum and U = S − T . By induction, (cD≤0, cD≥0) determine t-structures on T and U . For S this follows by verifying the conditions of [BBD, thm 1.4.10] using [S2, 4.4.1]. (cD≤0, cD≥0) is clearly compatible with the standard t-structure on DbCons(San). Let Cons-M HM (S) denote the heart cD≤0M HM (S)∩cD≥0M HM (S), and likewise for Cons-M HM (S, Σ). Lemma C.0.13. The functor rat : Cons-M HM (S, Σ) → Cons(San, Σ) yields an exact faithful embedding. 68 Proof. Exactness is already clear from the theorem, so the only issue is faithful- ness. This can be proved by induction on Σ. This holds when Σ = 1 because Cons-M HM (S, Σ) is the category of variations of Hodge structures. In general, let i : T → S be a closed stratum and j : U → S be the complement. Suppose that f ∈ Hom(M, N ) is morphism in Cons-M HM (S, Σ) such that rat(f ) = 0. We need to prove that f = 0. By induction f T = 0 and f U = 0. From the distinguished triangle j!j∗M → M → i∗i∗M → j!j∗M [1] and adjointness we obtain an exact sequence Hom(i∗i∗M, N ) Hom(M, N ) / Hom(j!j∗M, N ) Hom(i∗M, i∗N ) Hom(j∗M, j∗N ) Therefore f = 0. References [A1] Y. Andr´e, Pour une th´eorie inconditionelle de motifs, Inst. Hautes ´Etudes Sci. Publ. Math. No. 83 (1996) [A2] Y. Andr´e, Une introduction aux motifs, Soc. Math. France (2004) [Ar] D. Arapura,The Leray spectral sequence is motivic, Invent. Math. 160 (2005) [Ar2] D. Arapura ,The Hodge theoretic fundamental group and its cohomology. The geometry of algebraic cycles, 322, Clay Math. Proc., 9, AMS( 2010) [AD] D. Arapura, A. Dhillon, The motive of the moduli stack of G-bundles over the universal curve, Proc. Ind. Acad. Sci 118 (2008) [AK1] D. Arapura, M. Kumar, Beilinson-Hodge cycles on semiabelian varieties, Math. Res. Lett. 16 (2009) [SGA4] M. Artin, A. Grothendieck, J-L Verdier, Theorie de topos et cohomolo- gie ´etale de sch´emas Springer LNM 269, 270, 305 (1972-1973) [BFM] P. Baum, W. Fulton, R. Macpherson, Riemann-Roch for singular vari- eties, Inst. Hautes ´Etudes Sci. Publ. Math. No. 45 (1975) [BBD] A. Beilinson, J. Bernstein, P. Deligne, Faiseux pervers Asterique 100 (1982) [B1] A. Beilinson, Notes on absolute Hodge cohomology, Appl. Alg. K-theory to Alg. Geom, Comtemp. Math, AMS (1986) 69 / /   /   [B2] A. Beilinson, On the derived category of perverse sheaves, K-theory, arithmetic and geometry, Springer 1289 (1987) [Bl] S. Bloch, Algebraic cycles and higher K-theory , Adv. Math. 61 (1986) [Br] A. Brugui´eres, On a tannakian theorem due to Nori, preprint (2004) [C] M. de Cataldo, The standard filtration on cohomology with compact sup- ports..., Contemp. Math. 496, AMS (2009) [CM] M. de Cataldo, L. Migliorini, The perverse filtration and the Lefschetz hyperplane theorem, Annals of Math 171 (2010) [CD] D-C. Cisinski, F. D´eglise, Triangulated categories of mixed motives, arXiv:0912.2110 [Dg] F. D´eglise, Finite correspondences and transfers over a regular base, Al- gebraic cycles and motives, Lond. Math. Soc. LN 343, (2007) [De1] P. Deligne, Theorie de Hodge II,III, Inst. Hautes ´Etudes Sci. Publ. Math 40, 44, (1971,1974) [De2] P. Deligne, Le groupe fondemental de la droite projective moins trois points, Galois groups over Q, MSRI 16, Springer (1987) [De3] P. Deligne, Categories tannakiennes, Grothendieck Festschrift, Birkhauser (1990) [SGA4h] P. Deligne et. al., Cohomologie Etale, SGA4 1 2 Springer LNM 569 (1977) [DMOS] P. Deligne, J. Milne, A. Ogus, K. Shih, Hodge cycles, motives, and Shimura varieties, Springer LNM 900 (1982) [F] J. Franke, On the Brown representability theorem for triangulated cate- gories, Topology 40 (2001) [Gi] J. Giraud, Cohomologie non abelienne, Springer-Verlag (1971) [Gd] R. Godement, Topologie algebrique et theorie des Faisceaux, Hermann (1958) [Gr] P. Griffiths, Hodge theory and geometry, Bull. Lond. Math. Soc. 36 (2004) [SGA5] A. Grothendieck, et. al. Cohomologie ℓ-adique et Fonctions L , Springer LNM 589, (1977) [I] B. Iversen, Cohomology of sheaves, Springer-Verlag (1986) [J1] U. Jannsen, Mixed motives and algebraic K-theory Springer LNM 1400 (1990) 70 [J2] U. Jannsen, Motives, numerical equivalence and semisimplicity, Invent Math. 107 (1991) [Jo] J.P. Jouanolou, Un suit exacte de Mayer-Vietoris en K-theorie alge- brique,in Algebraic K-theory, Lect. Notes Math 341, Springer-Verlag (1973) [K] S. Kleiman, Algebraic cycles and the Weil conjectures Dix Expos´es, North-Holland (1968) [JS] A. Joyal, R. Street, Introduction to Tannaka duality and quantum groups, in Springer LNM 1488 (1991) [K] M Kashiwara, A study of a variation of mixed Hodge structure, Publ. Res. Inst. Math. Sci. 22 (1986) [KS] M. Kashiwara, P. Schapira, Categories and Sheaves, Springer-Verlag (2006) [LM] G. Laumon, L. Moret-Bailly, Champ Algebriques, Springer-Verlag (2000) [L1] M. Levine, Mixed Motives, Math. Surveys 57, AMS (1998) [L2] M. Levine, Mixed Motives, Handbook of K-theory, Springer-Verlag (2005) [Li1] [Li2] [M] J. Lipman, Equisingularity and simultaneous resolution of singularities, Resolution of singularities (Obergurgl 1997), Birkhauser, (2000) J. Lipman, Notes on derived functors and Grothendieck duality, SLN 1960, Springer-Verlag (2009) S. MacLane, Categories of the working mathematician, 2nd ed. Springer- Verlag (1998) [Mr] J. Mather, Notes on topological stability, Harvard (1970) [Mi] J. Milne, Etale cohomology, Princeton U. Press (1980) [N] A. Neeman, The Grothendieck duality theorem via Bousfield's techniques and Brown representability. J. Amer. Math. Soc. 9 (1996) [N1] M. Nori, Constructible sheaves, Proc. Int. Conf. on Algebra, Mumbai (2000), TIFR (2002) [N2] M. Nori, Unpublished notes on motives [PS] C. Peters, J. Steenbrink, Mixed Hodge structures, Springer-Verlag (2008) [Sv] N. Saavedra, Cat´egories Tannakiennes LNM 265, Springer (1972) [S1] M. Saito, Module Hodge polarizables, Publ. Res. Inst. Math. Sci. 25 (1989) 71 [S2] M. Saito, Mixed Hodge modules, Publ. Res. Inst. Math. Sci. 26 (1990) [S3] M. Saito, Arithmetic mixed sheaves, Invent Math 44 (2001) [SZ] J. Steenbrink, S. Zucker, Variations of mixed Hodge structures, Invent. Math. 80 (1983) [T] B. Teissier, Varietes Polaires II, pp 314-491, LNM 961, Springer [V1] [V2] J-L. Verdier, Class d'homologies d'un cycle, Sem. Geom. An. Ast´erisque no. 36-37, 101-151 (1976) J-L. Verdier, Stratification de Whitney et th´eor'eme de Bertini-Sard. In- vent 36 (1976) [Vi] A. Vistoli, Notes on Grothendieck topologies, fibered categories and de- scent theory , Fundamental algebraic geometry. Grothendieck's FGA ex- plained. AMS (2005) [VSF] V. Voevodsky, A. Suslin, E. Friedlander, Cycles, transfers, and motivic homology theories, Ann. Math Stud. 143, Princeton (2000) Department of Mathematics, Purdue University, West Lafayette IN 47907, U.S.A. [email protected] 72
1011.1752
3
1011
2015-09-14T06:59:08
On the dimension of spline spaces on planar T-meshes
[ "math.AG" ]
We analyze the space of bivariate functions that are piecewise polynomial of bi-degree \textless{}= (m, m') and of smoothness r along the interior edges of a planar T-mesh. We give new combinatorial lower and upper bounds for the dimension of this space by exploiting homological techniques. We relate this dimension to the weight of the maximal interior segments of the T-mesh, defined for an ordering of these maximal interior segments. We show that the lower and upper bounds coincide, for high enough degrees or for hierarchical T-meshes which are enough regular. We give a rule of subdivision to construct hierarchical T-meshes for which these lower and upper bounds coincide. Finally, we illustrate these results by analyzing spline spaces of small degrees and smoothness.
math.AG
math
ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES BERNARD MOURRAIN, GALAAD, INRIA M´EDITERRAN´EE [email protected] m,m ′ (T ) of bivariate functions that are Abstract. We analyze the space S r ′) and of smoothness r along the in- piecewise polynomial of bi-degree 6 (m, m terior edges of a planar T-mesh T . We give new combinatorial lower and upper bounds for the dimension of this space by exploiting homological techniques. We relate this dimension to the weight of the maximal interior segments of the T-mesh, defined for an ordering of these maximal interior segments. We show that the lower and upper bounds coincide, for high enough degrees or for hierarchical T-meshes which are enough regular. We give a rule of subdivision to construct hierarchical T-meshes for which these lower and upper bounds co- incide. Finally, we illustrate these results by analyzing spline spaces of small degrees and smoothness. Introduction Standard parametrisations of surfaces in Computer Aided Geometric Design are based on tensor product B-spline functions, defined from a grid of nodes over a rectangular domain [11]. These representations are easy to control but their re- finement has some drawback. Inserting a node in one direction of the parameter domain implies the insertion of several control points in the other directions. If for instance, regions along the diagonal of the parameter domain should be refined, this would create a fine grid in some regions where it is not needed. To avoid this problem, while extending the standard tensor product representation of CAGD, spline functions associated to a subdivision with T-junctions instead of a grid, have been studied. Such a T-mesh is a partition of a domain Ω into axis-aligned boxes, called the cells of the T-mesh. The first type of T-splines introduced in [21, 20], are defined by blending func- tions which are products of univariate B-spline basis functions associated to some nodes of the subdivision. They are piecewise polynomial functions, but the pieces where these functions are polynomial do not match with the cells of the T-subdivision. Moreover, there is no proof that these piecewise polynomial functions are linearly independent. Indeed, [3] shows that in some cases, these blending T-spline func- tions are not linearly independent. Another issue related to this construction is that there is no characterization of the vector space spanned by these functions. For this reason, the partition of unity property which is useful in CAGD is not avail- able directly in this space. The spline functions have to be replaced by piecewise rational functions, so that these piecewise rational functions sum to 1. However, this construction complexifies the practical use of such T-splines. 1 2 BERNARD MOURRAIN Being able to describe a basis of the vector space of piecewise polynomials of a given smoothness on a T-mesh is an important but non-trivial issue. It yields a construction of piecewise polynomial functions on the T-subdivision which form a partition of unity so that the use of piecewise rational functions is not required. It has also a direct impact in approximation problems such as surface reconstruction [5] or isogeometric analysis [14], where controlling the space of functions used to ap- proximate a solution is critical. In CAGD, it also provides more degrees of freedom to control a shape. This explains why further works have been developed to under- stand better the space of piecewise polynomial functions with given smoothness on a T-subdivision. To tackle these issues special families of splines on T-meshes have been stud- ied. In [6], [8], these splines are piecewise polynomial functions on a hierarchical T-subdivision. They are called PHT-splines (Polynomial Hierarchical T-splines). Dimension formulae of the spline space on such a subdivision have been proposed when the degree is high enough compared to the smoothness [6], [13], [16] and in some cases for biquadratic C1 piecewise polynomial functions [7]. The construction of a basis is described for bicubic C1 spline spaces in terms of the coefficients of the polynomials in the Bernstein basis attached to a cell. When a cell is subdivided into 4 subcells, the Bernstein coefficients of the basis functions of the old level are modified and new linearly independent functions are introduced, using Bernstein bases on the cells at the new level. In this paper, we analyse the dimension of the space S r m,m′ (T ) of bivariate func- tions that are piecewise polynomial of bi-degree 6 (m, m′) of smoothness r along the interior edges of a general planar T-mesh T , where r is a smoothness distribution on T . As we will see, computing this dimension reduces to compute the dimension of the kernel of a certain linear map (namely the map ∂2 introduced in Section 2). Thus for a given T-mesh, a given smoothness distribution r and a given bi-degree (m, m′), it is possible to compute the dimension of S r m,m′(T ) by linear algebra tools (see eg. a software implementation developed by P. Alfed1 for such computations). We would like to avoid a case-by-case treatment and to describe this dimension in terms of combinatorial quantities attached to T and easy to evaluate. As shown in [17] or [1], the dimension may also depend on the geometry of the T-mesh and not just on its topology. This explains why it is not always possible to provide a purely combinatorial formula for the dimension of S r m,m′ (T ). The main results in this paper are • a description of the dimension in terms of a combinatorial part that depends only on the topology of the T-mesh and an homological part that takes into account the fact that the dimension may depend on the geometry of the T-mesh (Theorem 3.1); • combinatorial lower and upper bounds on the dimension that are easy to evaluate (Theorem 3.7); • sufficient conditions under which the lower and upper bounds coincide so that the dimension depends only on the topology of the T-mesh (Theorem 3.9). 1http://www.math.utah.edu/∼pa/MDS/index.html ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES 3 We proceed as follows. By extending homological techniques developed in [2] and [19], we obtain combinatorial lower and upper bounds on the dimension of these spline spaces for general T-subdivisions. We relate the upper bound to the maximal interior segments and their weights and show that the lower and upper bounds coincide for T -meshes which are enough regular. Namely, if a T-mesh is (m + 1, m′ + 1)-weighted, the dimension depends directly on the number of faces, interior edges and interior points. In particular, we obtain the dimension formula for a constant smoothness distribution r = (r, r′) with m ≥ 2r + 1 and m′ ≥ 2r′ + 1, providing a new proof of a result also available in [6], [13], [16] for a hierarchical T-mesh. The algebraic approach gives an homological interpretation of the method called Smoothing Cofactor-Conformality method in [23]. It allows us to generalize the dimension formulae obtained by this technique [16], [13]. We also give a rule of subdivision to construct hierarchical T-meshes for which the lower and upper bounds coincide. As a consequence, we can recover the dimension of the space of Locally Refined splines described in [9]. We do not consider the problem of constructing explicit bases for these spline spaces, which will be analyzed separately. In the first section, we recall the notations and the polynomial properties which are needed in the following. Section 2 describes the chain complex associated to the spline space and analyzes its homology. In Section 3, we give lower and upper bounds on the dimension of the spline space and analyze cases where these bounds are coincide. Section 4 deals with the properties of hierarchical T -meshes, obtained by recursive subdivisions of cells. In the last section, we analyse some examples for small degree and smoothness. 1. Planar T-splines In the following, we will deal with notions which are of topological and algebraic nature. We start by the topological definitions. 1.1. T-meshes. For any set S ⊂ R relative interior, ∂S its boundary. We define a T-mesh T of R 2 as: 2, S is its closure for the usual topology, S ◦ its • a finite set T2 of closed axis-aligned rectangles of R • a finite set T1 of closed axis-aligned segments included in ∪σ∈T2 ∂σ, • a finite set of points T0 = ∪τ ∈T1∂τ , 2, such that • For σ ∈ T2, ∂σ is the finite union of elements of T1. • For σ, σ′ ∈ T2 with σ 6= σ′, σ ∩ σ′ = ∂σ ∩ ∂σ′ is the finite union of elements of T1 ∪ T0. • For τ, τ ′ ∈ T1 with τ 6= τ ′, τ ∩ τ ′ = ∂τ ∩ ∂τ ′ ⊂ T0. We denote by Ω = ∪σ∈T2 σ ⊂ R 2 and call it the domain of the T-mesh T . The elements of T2 are called 2-faces or cells and their number is denoted f2. The elements of T1 are called 1-faces or edges. An element of T1 is called an interior edge if it intersects Ω◦. It is called a boundary edge otherwise. The set of interior edges is denoted by T o 1 . The number of edges in T1 is f1 and the number of interior edges is f o 1 . An edge parallel to the first (resp. second) axis of R 2 is called horizontal (resp. ) be the set of horizontal (resp. vertical) interior vertical). Let T o,h 1 (resp. T o,v 1 4 BERNARD MOURRAIN edges and f h Then, the number of interior edges is f o 1 (resp. f v 1 = f h 1 + f v 1 . 1 ) the number of interior horizontal (resp. vertical) edges. The elements of T0 are called 0-faces or vertices. A vertex is interior if it is in Ω◦. 0 . We 0 be the number of interior It is a boundary vertex otherwise. The set of interior vertices is denoted T o denote by f0 be the number of vertices of T0 and by f o vertices. A vertex is a crossing vertex if it is an interior vertex and belongs to 4 distinct elements of T1. A vertex is a T-vertex if it is an interior vertex and belongs to exactly 3 distinct elements of T1. Let f + 0 ) be the number of crossing (resp. T) vertices. A boundary vertex is a vertex in T0 ∩ ∂Ω. The number of boundary vertices is f b 0 . A vertex is a corner vertex if it belongs to ∂Ω and to a vertical and a horizontal boundary edge. 0 (resp. f T To simplify the definitions and remove redundant edges, we will assume that any vertex γ ∈ T0 belongs at least to one horizontal edge τh(γ) ∈ T1 and one vertical edge τv(γ) ∈ T1. We denote by νh(T ) = {s1, . . . , sl} ⊂ R (resp. νv(T ) = {t1, . . . , tm} ⊂ R) the set of first (resp. second) coordinates of the points in vertical (horizontal) segments ∈ T v 1 ). The elements of νh(T ) (resp. νv(T )) are called the horizontal (resp. vertical) nodes of the T-mesh T . 1 (resp. ∈ T h Example 1.1. Let us illustrate the previous definitions on the following T-mesh: γ1 γ2 γ3 4 3 2 1 0 0 1 2 3 4 5 In this example, there are f2 = 7 rectangles, f o are horizontal and f v γ1, γ3 are T-vertices and γ2 is a crossing vertex. There are f b vertices and 12 corner vertices. 1 = 5 are vertical. There are f o 1 = 9 interior edges such that f h 1 = 4 0 = 3 interior points γ1, γ2, γ3; 0 = 15 boundary The horizontal nodes are νh(T ) = {0, . . . , 5} and the vertical nodes are νv(T ) = {0, . . . , 4}. For our analysis of spline spaces on T-meshes, we assume the following: Assumption: The domain Ω is simply connected and Ωo is connected. This implies that Ω has one connected component with no "hole" and that the number of boundary edges through a boundary vertex is 2. 1.2. T-splines. We are going now to define the spaces of piecewise polynomial functions on a T-mesh, with bounded degrees and given smoothness. An element in such a space is called a spline function. Definition 1.2. A smoothness distribution on a T-mesh T is a pair of maps (rh, rv) from (νh(T ) × νv(T )) to  × . ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES 5 1 (resp. τ ∈ T h For convenience, we will define the smoothness map r on T1 as follows: for any τ ∈ T v 1 ), r(τ ) = rh(s) (resp. r(τ ) = rv(t)) where s ∈ νh(T ) (resp. t ∈ νv(T )) is the first (resp. second) coordinate of any point of τ . We will also define the horizontal and vertical smoothness on T0 as follows: for any γ = (s, t) ∈ T0, rh(γ) = rh(s) and rv(γ) = rv(t). For r, r′ ∈ , we say that r is a constant smoothness distribution equal to (r, r′) if ∀s ∈ νh(T ), r(s) = r, ∀t ∈ νv(T ), r(t) = r′. Let R = R[s, t] be the polynomial ring with coefficients in R. For m, m′ ∈ , we denote by Rm,m′ = R(m,m′) the vector of polynomials in R of degree ≤ m in s and ≤ m′ in t. An element of Rm,m′ is of bi-degree 6 (m, m′). The goal of this paper is to analyse the dimension of the space of splines of bi-degree ≤ (m, m′) and of smoothness r on a T-mesh T , that we define now. Definition 1.3. Let T be a T-mesh and r a node smoothness distribution. We denote by S r m,m′ (T ) the vector space of functions which are polynomials in Rm,m′ on each cell σ ∈ T2 and of class r(τ ) in s (resp. in t) at any point of τ ∩ Ωo if τ is a vertical (resp. horizontal) interior edge. We will say that f ∈ S r m,m′(T ) is of (continuity) class C r on T . We notice that the boundary edges and their smoothness are not involved in the characterization of a spline function. Example 1.4. We consider again the T-mesh of Example 1.1. If we take the node smoothness distribution rh(1) = 1, rh(2) = 0, rh(3) = rh(4) = rh(5) = 1, and rv 3,3(T ) is the vector space of bicubic piecewise polynomial constant equal to 1, then S r functions on T which are C1 in s for s < 2 and s > 2, continuous for s = 2 and C1 in t in Ωo. 1.3. Polynomial properties. We recall here basic results on the dimension of the vector spaces involved in the analysis of S r m,m′(T ): For any τ ∈ T1, let lτ ∈ R be a non-zero polynomial (of degree 1) defining the line supporting the edge τ . Let ∆r(τ ) = lr(τ )+1 . We denote by Ir(τ ) = (∆r(τ )) the ideal generated by the polynomial ∆r(τ ) ∈ R and by Ir m,m′(τ ) = Ir(τ ) ∩ Rm,m′ its part of bi-degree ≤ (m, m′). Notice that Ir(τ ) defines the line supporting the egde τ with multiplicity (r(τ ) + 1). By definition, two horizontal (resp. vertical) edges τ1, τ2 which share a point define the same ideal Ir(τ1) = Ir(τ2). We define the bi-degree δ for any egde τ ∈ T1 as follows: τ • δ(τ ) = (r(τ ) + 1, 0) if τ is vertical, • δ(τ ) = (0, r(τ ) + 1) if τ is horizontal. Let Ir(γ) = Ir(τv) + Ir(τh) = (∆r(τv), ∆r(τh)) where τv, τh ∈ T1 are vertical and horizontal edges such that τv ∩ τh = {γ}. The ideal Ir(γ) defines the point γ with multiplicity (rh(γ) + 1) × (rv(γ) + 1). We denote by Ir m,m′(τv) + m,m′(τh). Notice that these definitions are independent of the choice of the vertical Ir edge τv and horizontal edge τh which contain γ. The bi-degree of a vertex γ ∈ T0 is δ(γ) = (rh(γ) + 1, rv(γ) + 1). m,m′(γ) = Ir Here are the basic dimension relations that we will use to analyse the spline functions on a T-mesh. Lemma 1.5. • dim Rm,m′ = (m + 1) × (m′ + 1). 6 BERNARD MOURRAIN • dim(cid:0)Rm,m′/Ir • dim(cid:0)Rm,m′/Ir γ ∈ T0. m,m′(τ )(cid:1) =(cid:26) (m + 1) × (min(r(τ ), m′) + 1) m,m′(γ)(cid:1) = (min(rh(γ), m) + 1) × (min(rv(γ), m′) + 1) for all (min(r(τ ), m) + 1) × (m′ + 1) if τ ∈ T h 1 if τ ∈ T v 1 Proof. To obtain these formulae, we directly check that • a basis of Rm,m′ is the set of monomials sitj with 0 ≤ i ≤ m, 0 ≤ j ≤ m′; m,m′(τ ) is the set of monomials sitj with 0 ≤ i ≤ m and • a basis of Rm,m′/Ir 1 (resp. 0 ≤ i ≤ min(r(τ ), m), 0 ≤ j ≤ m′ if 0 ≤ j ≤ min(r(τ ), m′) if τ ∈ T h τ ∈ T v 1 ); • a basis of Rm,m′/Ir m,m′(γ) is the set of monomials sitj with 0 ≤ i ≤ min(rh(γ), m), 0 ≤ j ≤ min(rv(γ), m′). since the ideal of an edge τ ∈ T1 is up to a translation (sr(τ )+1) or (tr(τ )+1). (cid:3) An algebraic way to characterise the C r-smoothness is given by the next lemma: Lemma 1.6 ([2]). Let τ ∈ T1 and let p1, p2 be two polynomials. Their derivatives coincide on τ up to order r(τ ) iff p1 − p2 ∈ Ir(τ ). In the following, we will need algebraic properties on univariate polynomials. We denote by U = R[u] the space of univariate polynomials in the variable u with coefficients in R. Let Un denote the space of polynomials of U of degree ≤ n. For a polynomial g ∈ U of degree d and an integer n ≥ d, g Un−d is the vector space of multiples of g which are of degree ≤ n. For polynomials g1, . . . , gk ∈ U respectively i=1 gi Un−di is the vector of degree d1, . . . , dk and an integer n ≥ maxi=1,...,k di, Pk space of sums of multiples of gi of degree ≤ n. We will use the apolar product defined on Un by: hf, gin = n Xi=0(cid:18)n i(cid:19)figi where f = Pn i=0 fiui, g = Pn need is the following [18], [15], [10]: i=0 giui ∈ R[u]n. One of the properties that we will Lemma 1.7. Let g ∈ Un, d < n and a ∈ R. Then g is orthogonal to (u − a)dUn−d for the apolar product iff Proposition 1.8. Let a1, . . . , al ∈ R be l distinct points and d1, . . . , dl ∈  . Then ∂kg(a) = 0, k = 0, . . . , n − d. l dim l Xi=1 (u − ai)diUn−di! = min(n + 1, Xi=1 Proof. In order to compute the dimension of V := Pl i=1(u − ai)dUn−d ⊂ Un, we compute the dimension of the orthogonal V ⊥ in Un of V for the apolar product. Let g ∈ Un be an element of the orthogonal V ⊥ of V . By lemma 1.7, ∂kg(ai) = 0, k = 0, . . . , n − di, i = 1, . . . , l. In other words, g is divisible by (u − ai)n−di+1 for i = 1, . . . , l. As the points ai are distinct, g is divisible by n − di + 1). l Conversely, any multiple of Π of degree 6 n is in V ⊥. Thus V ⊥ = (Π) ∩ Un. Π := (u − ai)n−di+1. Yi=1 ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES 7 This vector space V ⊥ of multiples of Π in degree n is of dimension max(0, n + 1 − deg(Π)), so V is of dimension n + 1 − max(0, n + 1 − deg(Π)) = min(n + 1, deg(Π)) = min(n + 1, n − di + 1). l Xi=1 (cid:3) We are going to use an equivalent formulation of this result: (1.1) dim Un/ l Xi=1 (u − ai)d i Un−di! = (n + 1 − l Xi=1 n − di + 1)+ where for any a ∈ Z, a+ = max(0, a). A similar result is proved in [16][Lemma 2] when all di are equal to d, by analyzing the coefficient matrix of generators of i=1(u − ai)dUn−d. Pl 1.4. Maximal interior segments. In order to simplify the analysis of S r we introduce the following definitions: m,m′(T ), For any interior edge τ ∈ T o 1 , we define ρ(τ ) as the maximal segment made of edges ∈ T o 1 of the same direction as τ , which contains τ and such that their union is connected. We say that the maximal segment ρ(τ ) is interior if it does not intersect the boundary of Ω. As all the edges belonging to a maximal segment ρ have the same supporting line, we can define ∆r(ρ) = ∆r(τ ) for any edge τ belonging to ρ = ρ(τ ). The set of all maximal interior segments is denoted by mis(T ). The set of horizontal (resp. vertical) maximal interior segments of T is denoted by mish(T ) (resp. misv(T )). The degree of ρ ∈ mis(T ) is by definition δ(ρ) = δ(τ ) for any τ ⊂ ρ. For each interior vertex γ ∈ T o 0 , which is the intersection of an horizontal edge τh ∈ T o 1 , let ρh(γ) (resp. ρv(γ)) is the corresponding horizontal (resp. vertical) maximal segment. We denote by ∆r v(γ)) the equations of the corresponding supporting lines to the power rh(γ) + 1 (resp. rv(γ) + 1). 1 with a vertical edge τv ∈ T o h(γ) (resp. ∆r Notice that the intersection of two distinct maximal interior segments is either a T-vertex or a crossing vertex. We say that ρ ∈ mis(T ) is blocking ρ′ ∈ mis(T ) if one of the end points of ρ′ is in the interior of ρ. Example 1.9. In the figure below, the maximal interior edges are indicated by plain segments. 8 BERNARD MOURRAIN ρ3 ρ4 ρ1 ρ2 In this example, ρ1 is blocking ρ4 and ρ2 is blocking ρ3. 2. Topological chain complexes In this section, we describe the tools from algebraic topology, that we will use. For more details, see eg. [22], [12]. 2.1. Definitions. We consider the following complexes: Ir m,m ′ (T o) : 0 ↓ 0 ↓ → Lτ ∈T o 1 ↓ [τ ]Ir m,m ′ (τ ) 0 ↓ → Lγ∈T o 0 ↓ [γ]Ir m,m ′ (γ) R m,m ′ (T o) : Lσ∈T2 [σ]Rm,m ′ → Lτ ∈T o 1 [τ ]Rm,m ′ → Lγ∈T o 0 [γ]Rm,m ′ ↓ ↓ ↓ → 0 → 0 Sr m,m ′ (T o) : Lσ∈T2 [σ]Rm,m ′ → Lτ ∈T o 1 [τ ]Rm,m ′ /Ir m,m ′ (τ ) → Lγ∈T o 0 [γ]Rm,m ′ /Ir m,m ′ (γ) → 0 ↓ 0 ↓ 0 ↓ 0 The different vector spaces of these complexes are obtained as the components in bi-degree 6 (m, m′) of R-modules generated by (formal) independent elements [σ], [τ ], [γ] indexed respectively by the faces, the interior edges and interior points of T . An oriented edge τ ∈ T1 is represented as: [τ ] = [ab] where a, b ∈ T0 are the end points. The opposite edge is represented by [ba]. By convention, [ba] = −[ab]. The maps of the complex Rm,m′(T o) are defined as follows: • for each face σ ∈ T2 with its counter-clockwise boundary formed by edges τ1 = a1a2, . . . , τl = ala1, ∂2(σ) = [τ1] ⊕ · · · ⊕ [τl] = [a1a2] ⊕ · · · ⊕ [ala1]; • for each interior edge τ ∈ T o 1 from γ1 to γ2 ∈ T0, ∂1([τ ]) = [γ2] − [γ1] where [γ] = 0 if γ 6∈ T o 0 ; • for each interior point γ ∈ T o 0 , ∂0([γ]) = 0. By construction, we have ∂i ◦ ∂i+1 = 0 for i = 0, 1. The maps of the complex m,m′(T o) are obtained from those of Rm,m′(T o) by restriction, those of the com- Ir m,m′(T o) are obtained by taking the quotient by the corresponding vector plex Sr m,m′(T o). The corresponding differentials of the complex are denoted spaces of Ir ¯∂i. For each column of the diagram, the vertical maps are respectively the inclusion map and the quotient map. The complex Rm,m′(T o) is also known as the chain complex of T relative to its boundary ∂T . ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES 9 Example 2.1. We consider the following subdivision T of a rectangle Ω: σ1 γ1 β1 β3 σ2 σ3 β2 We have • ∂2([σ1]) = [γ1β1] + [β3γ1], ∂2([σ2]) = [γ1β2] + [β1γ1], ∂2([σ3]) = [γ1β3] + [β2γ1], • ∂1([β1γ1]) = [γ1], ∂1([β2γ1]) = [γ1], ∂1([β3γ1] = [γ1], • ∂0([γ1]) = 0. This defines the following complex: i=1[βiγ1]Rm,m′ → [γ1]Rm,m′ → 0 The matrices of these maps in the canonical (monomial) bases are Rm,m′(T ) : L3 i=1[σi]Rm,m′ → L3 [∂2] = 0 −I   −I I 0 −I I 0 I  , [∂1] =(cid:0) I I I (cid:1) where I is the identity matrix of size (m + 1) × (m′ + 1) (ie. the dimension of Rm,m′). Let us consider the case where γ1 = (0, 0), (m, m′) = (2, 2) and r is the constant distribution (1, 1) on T . The matrices of the complex S1,1 2,2(T ) are [ ¯∂2] =  −[Π1] 0 [Π3] [Π1] −[Π2] 0 0 [Π2] −[Π3]   , [ ¯∂1] =(cid:0) [P1] [P2] [P3] (cid:1) where [Πi] (resp. [Pi]) are the matrices of the projections Π1 = Π3 : R2,2 → R2,2/(s2) p 7→ p mod s2 Π2 : R2,2 → R2,2/(t2) p 7→ p mod t2 P1 = P3 : R2,2/(s2) → R2,2/(s2, t2) p mod s2 7→ p mod (s2, t2) P2 : R2,2/(t2) → R2,2/(s2, t2) p mod t2 7→ p mod (s2, t2). The matrices Πi are of size 12 × 16 and the matrices Pi are of size 9 × 12. 2.2. Their homology. In this section, we analyse the homology of the different complexes. The homology of a chain complex of a triangulation of a (planar) domain is well-known [22][Chap. 4], [12][Chap. 2]. Since, it is not explicit in the litterature, we give in the appendix a simple proof of the exactness of Rm,m′(T o). 10 BERNARD MOURRAIN 2.2.1. The 0-homology. We start by analysing the homology on the vertices. Lemma 2.2. H0(Rm,m′ (T o)) = H0(Sr m,m′(T o)) = 0. Proof. By Proposition D.1 in the appendix, we have H0(Rm,m′(T o)) = 0. Taking m,m′ (T o)) = the quotient by Ir 0. (cid:3) m,m′(τi), we still get that ¯∂1 is surjective so that H0(Sr Let us describe in more details H0(Ir m,m′ (T o)) = Mγ∈T o 0 [γ]Ir(γ)/∂1(Mτ ∈T o 1 [τ ]Ir(τ )). We consider the free R-module generated by the half-edge elements [γτ ], for all interior edges τ ∈ T o 1 and all vertices γ ∈ τ . By convention [γτ ] ≡ 0 if γ ∈ ∂Ω. For γ ∈ T o 0 , let Eh(γ) (resp. Ev(γ)) be the set of horizontal (resp. vertical) interior edges that contain γ and let E(γ) = Eh(γ) ∪ Ev(γ). We consider first the map ϕγ : Mτ ∈E(γ) [γτ ] R(m,m′)−δ(τ ) → [γ] Ir m,m′(γ) [γτ ] 7→ [γ] ∆r(τ ) m,m′(γ), this map is surjective. Its kernel is denoted Kr By definition of Ir m,m′(γ). Let Ph(γ) (resp. Pv(γ)) be the set of pairs (τ, τ ′) of distinct horizontal (resp. vertical) interior edges which contain γ (with (τ, τ ′) identify to (τ ′, τ )). We denote by P (γ) = Ph(γ) ∪ Pv(γ). If γ is a T-junction, one of the two sets is empty and the other is a singleton containing one pair. If γ is a crossing vertex, each set is a singleton. The following proposition gives an explicit description of the kernel Kr m,m′(γ), that we will exploit hereafter. Proposition 2.3. Kr m,m′ (γ) = X(τ,τ ′)∈P (γ) ([γτ ] − [γτ ′])R(m,m′)−δ(τ ) + Xτ ∈Eh(γ),τ ′∈Ev(γ) ([γτ ] ∆r(τ ′) − [γτ ′] ∆r(τ ))R(m−r−1,m′−r′−1) Proof. Let us suppose first that γ is a crossing vertex. We denote by τ1, τ2 the horizontal edges, τ3, τ4 the vertical edges at γ. The matrix of the map ϕγ in the basis [γτi] is [ϕγ] =(cid:0) ∆ ∆ ∆′ ∆′ (cid:1) where ∆ = ∆r(τ1) = ∆r(τ2 ), ∆′ = ∆r(τ3) = ∆r(τ4). Since ∆ and ∆′ have no common factor, the kernel of the matrix [ϕγ] is generated by the elements [γτ1] − [γτ2], [γτ2] ∆′ − [γτ3] ∆, [γτ3] − [γτ4], which give the description of m,m′(γ) in bi-degree ≤ (m, m′). A similar proof applies when there is no horizontal Kr or vertical pair of distinct edges at γ. This proves the result. (cid:3) We use the maps (ϕγ)γ∈T o 0 to define ϕ : Mγ∈T o 0 Mτ ∈Eγ [γτ ] R(m,m′)−δ(τ ) → Mγ∈T o 0 [γ] Ir m,m′(γ), ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES 11 so that we have the following exact sequence: 0 Kr 0 Mτ ∈E(γ) m,m′(γ) → Mγ∈T o 0 → Mγ∈T o Using this exact sequence, we can now identifyLγ∈T o [γτ ] R(m,m′)−δ(τ )/ Xγ∈T o Mγ∈T o 1 Mτ ∈Eγ [γτ ] R(m,m′)−δ(τ ) → Mγ∈T o 0 1 1 [γ] Ir m,m′(γ) → 0 [γ] Ir m,m′(γ) with the quotient Kr m,m′ (γ). The next proposition uses this identification and Proposition 2.3 to describe more explicitly H0(Ir m,m′ (T o)): Proposition 2.4. We have H0(Ir m,m′(T o)) = Mγ∈T o ,  [γτ ] R(m,m′)−δ(τ ) ([γτ ] − [γτ ′])R(m,m′)−δ(τ ) ([γτ ] − [γ ′τ ])R(m,m′)−δ(τ ) 0 Mτ ∈E(γ)  X(τ,τ ′)∈P (γ) + Xτ =(γ,γ ′)∈T o + Xτ ∈Eh(γ),τ ′∈Ev(γ) 1 ([γτ ]∆r(τ ′) − [γτ ′]∆r(τ ))R(m,m′)−δ(γ)  . Proof. The application ∂1 : Mτ ∈T o 1 lift to an application: [τ ]Ir m,m′ (τ ) → Mγ∈T o 0 [γ]Ir m,m′(γ) ∂1 : Mτ ∈T o 1 [τ ]R(m,m′)−δ(τ ) → Mγ∈T o 0 Mτ ∈E(γ) τ 7→ [γτ ] − [γ ′τ ] [γτ ]R(m,m′)−δ(τ ) so that the image of ∂1 lift in Lγ∈T o im ∂1 = Xτ ∈T o 1 Consequently, 0 Lτ ∈E(γ)[γτ ]R(m,m′)−δ(τ ) to ([γτ ] − [γ ′τ ])R(m,m′)−δ(τ ). H0(Sr m,m′(T o)) = Mγ∈T o 0 Mτ ∈E(γ) which yields the desired description of H0(Sr m,m′ (T )). [γτ ] R(m,m′)−δ(τ ), im ∂1 + Xγ∈T o 0 Kr m,m′ (γ)  , (cid:3) In the next proposition, we simplify further the description of H0(Ir m,m′ (T o)): Proposition 2.5. H0(Ir m,m′ (T o)) = ⊕ρ∈mis(T ) [ρ] R(m,m′)−δ(ρ) (cid:14) (cid:16)Pγ∈T o 0 ([ρv(γ)] ∆r h(γ) − [ρh(γ)] ∆r v(γ))R(m,m′)−δ(γ)(cid:17) . 12 BERNARD MOURRAIN Proof. Let B = Lγ∈T o and 0 Lτ ∈E(γ)[γτ ] R(m,m′)−δ(τ ), K = im ∂1 +Pγ∈T o 0 Kr,r′ m,m′(γ) K ′ = (cid:16)P(τ,τ ′)∈P (γ)([γτ ] − [γτ ′])R(m,m′)−δ(τ ) ([γτ ] − [γ ′τ ])R(m,m′)−δ(τ )(cid:17) . +Pτ =(γ,γ ′)∈T o 1 As K ′ ⊂ K ⊂ B, we have B/K ≡ (B/K ′)/(K/K ′). Taking the quotient by K ′ means, that we identify the horizontal (resp. vertical) edges which share a vertex. Thus all horizontal (resp. vertical) edges which are contained in a maximal segment ρ of T are identify to a single element, that we denote [ρ]. As [γτ ] = 0 if γ ∈ ∂Ω, we also have [ρ] = 0 if the maximal segment ρ intersects the boundary ∂Ω. This yields the desired description of H0(Ir (cid:3) m,m′(T )). Definition 2.6. Let hr m,m′(T ) = dim H0(Ir,r′ m,m′ (T o)). 2.2.2. The 1-homology. We consider now the homology on the edges. We use the property that H1(Rm,m′(T o)) = 0 (see Proposition D.2 in the appendix). Proposition 2.7. H1(Sr m,m′(T o)) = H0(Ir m,m′ (T o)). Proof. As H0(Rm,m′(T o)) = 0 and H1(Rm,m′(T o)) = 0, we deduce from the long exact sequence (see Appendix B) · · · → H1(Rm,m′(T o)) → H1(Sr m,m′(T o)) → H0(Ir m,m′ (T o)) → H0(Rm,m′(T o)) → · · · that H1(Sr m,m′(T o)) ∼ H0(Ir m,m′(T o)). (cid:3) 2.2.3. The 2-homology. Finally, the homology on the 2-faces will give us informa- tion on the spline space S r m,m′ (T ). We have the following result (proved in Proposition D.3 in the appendix): Proposition 2.8. H2(Rm,m′(T o)) = Rm,m′. The following proposition relates the spline space S r m,m′(T ) to an homology module. Proposition 2.9. H2(Sr m,m′(T o)) = ker ∂2 = S r m,m′(T ). m,m′(T o)) = ker ∂2 is a collection of polynomials (pσ)σ∈T2 Proof. An element of H2(Sr τ (τ ) if σ and σ′ share the (internal) edge τ . with pσ ∈ Rm,m′ and pσ ≡ pσ′ mod Ir By Lemma 1.6, this implies that the piecewise polynomial function which is pσ on σ and pσ′ on σ′ is of class C r across the edge τ . As this is true for all interior edges, (pσ)σ∈T2 ∈ ker ∂2 is a piecewise polynomial function of Rm,m′ which is of class C r, that is an element of S r (cid:3) m,m′(T ). 3. Lower and upper bounds on the dimension In this section, are the main results on the dimension of the spline space S r m,m′(T ). ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES 13 Theorem 3.1. Let T be a T-mesh and let r be a smoothness distribution on T . Then (3.1) dim S r (m + 1)(m′ + 1) (m + 1)(r(τ ) + 1) − Xτ ∈T o,v 1 (m′ + 1)(r(τ ) + 1) (rh(γ) + 1)(rv(γ) + 1) m,m′(T ) = Xσ∈T2 − Xτ ∈T o,h + Xγ∈T o 1 0 where hr m,m′(T ) = dim H0(Ir + hr m,m′(T ). m,m′ (T o)). Proof. The complex Sr m,m ′ (T o) : ⊕σ∈T2[σ]Rm,m ′ −→ ⊕τ ∈T o 1 [τ ]Rm,m ′ /Ir m,m ′ (τ ) −→ ⊕γ∈T o 0 [γ]Rm,m ′ /Ir m,m ′ (γ) −→ 0 induces the following relations dim(⊕σ∈T2[σ]Rm,m′ )−dim(⊕τ ∈T o 1 [τ ]Rm,m′ /Ir,r′ = dim(H2(Sr m,m′ (T o))) − dim(H1(Sr,r m,m′(T o)) = S r m,m′(T ), H0(Sr As H2(Sr H0(Ir m,m′(T o)), we deduce that dim S r m,m′(T ) = dim(⊕σ∈T2[σ]Rm,m′ ) − dim(⊕τ ∈T o m,m′(τ ))+dim(⊕γ∈T o m,m′ (T o))) + dim(H0(Sr m,m′(T o)) = 0 and H1(Sr 0 [γ]Rm,m′/Ir m,m′ (T o))) m,m′(T o)) = m,m′(γ)) + dim(⊕γ∈T o 0 [γ]Rm,m′/Ir 1 [τ ]Rm,m′ /Ir m,m′(γ)) + dim(H0(Ir,r′ m,m′(τ )) m,m′(T ))) which yields the dimension formula (3.1) using Lemma 1.5. (cid:3) As an immediate corollary of this theorem and of Proposition 2.5, we deduce the following result: Corollary 3.2. If the T-mesh T has no maximal interior segments then hr 0. m,m′(T ) = In the case of a constant smoothness distribution, Theorem 3.1 is written as follows: Theorem 3.3. Let T be a T-mesh and let r = (r, r′) be a constant smoothness distribution on T . Then (3.2) dim S r m,m′(T ) = (m + 1)(m′ + 1)f2 − (cid:0)(m + 1)(r′ + 1)f h + (r + 1)(r′ + 1)f0 + hr m,m′(T ). 1 + (m′ + 1)(r + 1)f v 1(cid:1) where 1 ) is the number of horizontal (resp. vertical) interior edges • f2 is the number of 2-faces ∈ T2, • f h 1 (resp. f v ∈ T o 1 , • f0 is the number of interior vertices ∈ T o 0 , • hr m,m′(T ) = dim H0(Ir m,m′ (T o)). We are going now to bound hr m,m′(T ) for general T-meshes. 14 BERNARD MOURRAIN Definition 3.4. Let ι be an ordering of mis(T ) that is a map from mis(T ) to N. For ρ ∈ mis(T ), let Γι(ρ) be the set of vertices γ of ρ which are not on a maximal interior segment ρ′ ∈ mis(T ) with ι(ρ′) > ι(ρ). The number of elements of Γι(ρ) is denoted λι(ρ). We define now the weight of a maximal interior segment. Definition 3.5. For ρ ∈ mis(T ), let • ωι(ρ) =Pγ∈Γι(ρ)(m − rv(γ)) if ρ ∈ mish(T ). • ωι(ρ) =Pγ∈Γι(ρ)(m′ − rh(γ)) if ρ ∈ misv(T ). We called it the weight of ρ. As in the usual spline terminology, for an interior point γ ∈ T o 0 , we call γ m − rh(γ) (resp. m′ − rh(γ) ) the horizontal (resp. vertical) multiplicity of γ. If ρ is horizontal (resp. vertical), the weight of ρ is the sum of the vertical (resp. horizontal) multiplicities of the vertices γ ∈ Γι(ρ). Notice that if r = (r, r′) is a constant smoothness distribution on T , then ωι(ρ) = (m − r)λ(ρ) for ρ ∈ mish(T ) (resp. ωι(ρ) = (m′ − r′)λ(ρ) for ρ ∈ misv(T )). Example 3.6. We consider the T-mesh of Example 1.9 with (m, m′) = (2, 2), the constant smoothness distribution r = (1, 1) and the ordering of the maximal interior segments ι(ρi) = i for i = 1, . . . , 4. Then we have • ωι(ρ1) = (2 − 1) × 2 = 2, • ωι(ρ2) = (2 − 1) × 2 = 2, • ωι(ρ3) = (2 − 1) × 3 = 3, • ωι(ρ4) = (2 − 1) × 3 = 3, since the multiplicity of a vertex is 2 − 1 = 1 and the interior points of ρ1, ρ2 are not in Γι(ρ1) or Γι(ρ2). In the following, we will drop the index ι to simplify the notations, assuming that the ordering ι is fixed. Then we have the following theorem: Theorem 3.7. Let T be a T-mesh and let r be a smoothness distribution on T . Then 0 ≤ hr (m + 1 − ω(ρ))+ × (m′ − r(ρ)) m,m′(T ) ≤ Xρ∈mish(T ) + Xρ∈misv (T ) (m − r(ρ)) × (m′ + 1 − ω(ρ)))+ . Proof. Let ρ1, . . . , ρl be the maximal interior segments of T . By Proposition 2.5, m,m′(T ) is the dimension of the quotient in bi-degree ≤ (m, m′) of the module hr M := ⊕l i=1[ρi] R by the module K generated by the following relations: for each vertex γ ∈ T o 0 which is on a maximal interior segment, • ∆r(ρj)[ρi ] − ∆r(ρi)[ρj ] if γ is the intersection of the maximal interior seg- ments ρi and ρj, • ∆r(ρ)[ρi ] if γ is the intersection of the maximal interior segment ρi with another maximal segment ρ intersecting ∂Ω. To compute the dimension of M/K in bi-degree 6 (m, m′), we use a graduation on M given by the indices of the segments. For r := Pi pi[ρi] ∈ M (with pi ∈ R(m,m′)−δ(ρi)), let In(r) be the element pi0 [ρi0 ] where i0 is the maximal index such ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES 15 that pi 6= 0. We denote it by In(r) and called it the initial of r. Let In(K) = {In(k) k ∈ K}. The dimension hr m,m′(T ) is then hr m,m′(T ) = dim(M/K) = dim(M/ In(K)). Notice that In(K) contains the multiples in bi-degree 6 (m, m′) of • ∆r(ρj)[ρi ] if γ is the intersection of two maximal interior segments ρi and ρj with i > j, • ∆r(ρ)[ρi ] if γ is the intersection of the maximal interior segment ρi with a maximal segment ρ intersecting Ω. Let Li be the vector space spanned by these initials in bi-degree 6 (m, m′), which are multiples of [ρi]. By definition, for each γ ∈ Γ(ρi), we have a generator ∆r(ρ)[ρi] in Li for {γ} = ρi ∩ ρ. By Proposition 1.8, Li is of dimension • min(m + 1, ω(ρi)) × (m′ − r(ρi)) if ρi ∈ mish(T ), • min(m′ + 1, ω(ρi)) × (m − r(ρi)) if ρi ∈ misv(T ). Thus the dimension of [ρi]R(m,m′)−δ(ρi)/Li is • (m + 1 − ω(ρi))+ × (m′ − r(ρi)) if ρi ∈ mish(T ), • (m − r(ρi)) × (m′ + 1 − ω(ρi))+ if ρi ∈ misv(T ). As In(K) ⊃Pi Li, we have m,m′(T ) = dim(M/ In(K)) hr ≤ dim([ρi] R(m,m′)−δ(ρi)/(Xi Li)) =Xi dim(cid:0)[ρi ] R(m,m′)−δ(ρi)/Li(cid:1) . This gives the announced bound on hr dim([ρi ] R(m,m′)−δ(ρi)/Li). m,m′(T ), using the previous computation of (cid:3) Definition 3.8. The T-mesh T with a smoothness distribution r is (k, k′)-weighted if • ∀ρ ∈ mish(T ), ω(ρ) ≥ k • ∀ρ ∈ misv(T ), ω(ρ) ≥ k′; Theorem 3.9. Let T be a T-mesh and let r be a smoothness distribution on T . If T is (m + 1, m′ + 1)-weighted, then hr m,m′(T ) = 0. Proof. By definition, ∀ρ ∈ mish(T ), ω(ρ) ≥ m + 1 (i.e. (m + 1 − ω(ρ))+ = 0) and ∀ρ ∈ misv(T ), ω(ρ) ≥ m′ + 1 (i.e. (m′ + 1 − ω(ρ))+ = 0). By Theorem 3.7, we directly deduce that hr (cid:3) m,m′(T ) = 0. Here is a direct corollary which generalizes a result in [16]: Corollary 3.10. Suppose that the end points of a maximal interior segment ρ ∈ mis(T ) are in Γ(ρ). If for each horizontal (resp. vertical) maximal interior segment ρ ∈ mis(T ), the sum of the vertical (resp. horizontal) multiplicities of the end points and of the vertices of ρ on a maximal segment connected to the boundary is greater than or equal to m + 1 (resp. m′ + 1), then hr m,m′(T ) = 0. Proof. Let ρ be a maximal interior segment of T . By hypothesis, the end points of ρ are in Γ(ρ). As any point γ ∈ ρ which is also on a maximal segment connected to the boundary is in Γ(ρ), the hypothesis implies that ω(ρ) ≥ m + 1 if ρ is horizontal and ω(ρ) ≥ m′ + 1 if ρ is vertical. We deduce by Theorem 3.7 that hr m,m′(T ) = 0. (cid:3) Another case where hr m,m′(T ) is known is described in the next proposition. 16 BERNARD MOURRAIN Proposition 3.11. If ∀ρ ∈ mish(T ), ω(ρ) ≤ m + 1 and ∀ρ ∈ misv(T ), ω(ρ) ≤ m′ + 1, then (3.3) (m + 1 − ω(ρ))+ × (m′ − r(ρ)) hr m,m′(T ) = Xρ∈mish(T ) + Xρ∈misv (T ) (m − r(ρ)) × (m′ + 1 − ω(ρ)))+ . Proof. In the case where ∀ρ ∈ mish(T ), ω(ρ) ≤ m + 1 and ∀ρ ∈ misv(T ), ω(ρ) ≤ m′ + 1, Proposition 1.8 implies that there is no relations in bi-degree 6 (m, m′) of the monomial multiples of ∆r(ρj )[ρi], ∆r(ρ)[ρi ] for i = 1, . . . , l, j < i using the same notation as in the proof of Theorem 3.7. This implies that In(K) = ⊕iLi, which shows that hr m,m′(T ) = dim(M/ In(K)) =Xi Thus the equality (3.3) holds. dim(R(m,m′)−δ(ρi)[ρi ]/Li). (cid:3) As a corollary, we have the following result for constant smoothness distribution: Theorem 3.12. Let T be a T-mesh and let r = (r, r′) be a constant smoothness distribution on T . Then 0 ≤ hr m,m′(T ) ≤ Xρ∈mish(T ) + Xρ∈misv(T ) (m + 1 − (m − r)λ(ρ))+ × (m′ − r′) (m − r) × (m′ + 1 − (m′ − r′)λ(ρ))+ . Moreover, equality holds in the following cases: • ∀ρ ∈ mish(T ), (m − r)λ(ρ) ≥ m + 1 and ∀ρ ∈ misv(T ), (m′ − r′)λ(ρ) ≥ m′ + 1; • ∀ρ ∈ mish(T ), (m − r)λ(ρ) ≤ m + 1 and ∀ρ ∈ misv(T ), (m′ − r′)λ(ρ) ≤ m′ + 1. 4. Hierarchical T-meshes We consider now a special family of T-meshes, which can be defined by recursive subdivision from an initial rectangular domain Ω. Their study is motivated by practical applications, where local refinement of tensor-product spline spaces are considered eg. in isogeometric analysis [14]. Definition 4.1. A hierarchical T-mesh is either the initial axis-aligned rectangle Ω or obtained from a hierarchical T-mesh by splitting a cell along a vertical or horizontal line. A hierarchical T-mesh will also be called a T-subdivision. It can be represented by a subdivision tree where the nodes are the cells obtained during the subdivision and the children of a cell σ are the cells obtained by subdividing σ. In a hierarchical T-mesh T , the maximal interior segments are naturally ordered in the way they appear during the subdivision process. This is the ordering ι that we will consider hereafter. Notice that a maximal interior segment ρ is transformed by a split either into a maximal segment which intersects ∂Ω or into a larger maximal interior segment with a larger weight. We remark that in a hierarchical T-mesh, if ρi is blocking ρj then i < j. ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES 17 Example 4.2. Here are an hierarchical T-mesh (case a) and a non-hierarchical T-mesh (case b): (a) (b) 4.1. Dimension formula for hierarchical T-meshes. As a corollary of Theo- rem 3.7, we deduce the following result, also proved in [6], [16], [13]: Proposition 4.3. Let T be a hierarchical T-mesh and let r = (r, r′) be a constant smoothness distribution on T . For m ≥ 2r+1 and m′ ≥ 2r′+1, we have hr,r′ m,m′(T ) = 0. Proof. We order the maximal interior segments in the way they appear during the subdivision. If a segment ρi ∈ mis(T ) is blocking ρj ∈ mis(T ), we must have i < j. This shows that the end points of ρi are in Γ(ρi). Thus, λ(ρi) ≥ 2. As m ≥ 2r + 1, we have (m − r)λ(ρi) ≥ 2(m − r) ≥ m + (m − 2r) ≥ m + 1. Thus, (m + 1 − (m − r)λ(ρi))+ = 0. Similarly (m′ + 1 − (m′ − r′)λ(ρi))+ = 0 holds since m′ ≥ 2r′ + 1. By Theorem 3.7, we deduce that hr (cid:3) m,m′(T ) = 0. Theorem 3.9 leads us to the following construction rule of a T-subdivision T for which hr m,m′(T ) = 0. Algorithm 4.4 ((k, k′)-weighted subdivision rule). For each 2-face σ of a T-mesh to be subdivided, (1) Split σ with the new edge τ ; (2) If the edge τ does not extend an existing segment, extend τ (on one side and/or the other) so that the maximal segment containing τ is either in- tersecting ∂Ω or horizontal (resp. vertical) and of weight ≥ k (resp. ≥ k′). If such a rule is applied in the construction of a T-subdivision, • either a new maximal interior segment is constructed so that its weight is ≥ k if it is a horizontal maximal interior segment (resp. ≥ k′ it it is a vertical maximal interior segment), • or an existing maximal interior segment is extended and its weight is also increased, • or a maximal segment intersecting ∂Ω is constructed. In all cases, if we start with a (k, k′)-weighted T-mesh, we obtain a new T-mesh, which is also (k, k′)-weighted. 18 BERNARD MOURRAIN By Theorem 3.9, if k ≥ m + 1 and k′ ≥ m′ + 1 then hr m,m′(T ) = 0 and the dimension of dim S r m,m′ (T ), given by formula (3.1), depends only on the number of cells, interior segments and interior vertices of T . From this analysis, we deduce the dimension formula of the space of Locally Refined splines described in [9]. 5. Examples In this section, we analyse the dimension formula of spline spaces of small bi- degree and small constant smoothness distribution r = (r, r′) on a T-mesh T . 5.1. Bilinear C0,0 T-splines. We consider first piecewise bilinear polynomials on T , which are continuous, that is m = m′ = 1 and r = r′ = 0. By Proposition 4.3, we have h0,0 1,1(T ) = 0. Using Theorem 3.1 and Lemma A.1 in the appendix, we obtain: (5.1) dim S0,0 1,1 (T ) = 4f2 − 2f o 1 + f o 0 = f + 0 + f b 0 . 5.2. Biquadratic C1,1 T-splines. Let us consider now the set of piecewise bi- quadratic functions on a T-mesh T , which are C1. For m = m′ = 2 and r = r′ = 1, Theorem 3.1 and Lemma A.1 again yield (5.2) dim S1,1 2,2 (T ) = 9f2 − 6f o 1 + 4f o 0 + h1,1 2,2(T ) = f + 0 − 1 2 f T 0 + 3 2 0 + 3 + h1,1 f b 2,2(T ). If the T-mesh T is (3, 3)-weighted, then by Theorem 3.9, we have a h1,1 but this is not always the case. 2,2(T ) = 0, Example 5.1. Here is an example where h1,1 there is one maximal interior segment ρ with ω(ρ) = 2 − 1 + 2 − 1 = 2: 2,2(T ) = 1 by Proposition 3.11, since t1 t0 s0 s1 s2 s3 The dimension of S1,1 2,2 (T ) is 9 × 4 − 6 × 5 + 4 × 2 + h1,1 2,2(T ) = 14 + 1 = 15. Notice that the dimension is the same without the (horizontal) interior segment. Thus a basis of S1,1 2,2 (T ) is the tensor product B-spline basis corresponding to the nodes s0, s0, s0, s1, s2, s3, s3, s3 in the horizontal direction and the nodes t0, t0, t0, t1, t1, t1 in the vertical direction. Example 5.2. Here is another example. We subdivide the T-mesh T1 to obtain the second T-mesh T2: ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES 19 σ1 T1 σ2 T2 Doing this, we increase the number of cells by 9 − 1 = 8, the number of interior edges by 24 − 4 = 20, the number of interior points by 16 − 4 = 12. The dimension of the spline space increases by 9 × 8 − 6 × 20 + 4 × 12 + h1,1 2,2(T1) = h1,1 2,2(T2) − h1,1 2,2(T1). Since there is no maximal interior segment in T1, by Corollary 3.2 we have h1,1 2,2(T1) = 0. Choosing a proper ordering of the interior segments, we deduce by Theorem 3.1 that h1,1 2,2(T2) ≤ 1. Suppose that σ1 = [a0, a3] × [b0, b3] and σ2 = [a1, a2] × [b1, b2]. For u0 ≤ u1 ≤ u2 ≤ u3 ∈ R, let N (u; u0, u1, u2, u3) be the B-spline basis function in the variable u of degree 2 associated to the nodes u0, . . . , u3 (see [5]). Then the piecewise polynomial function 2,2(T2) − h1,1 N (s; a0, a1, a2, a3) × N (t; b0, b1, b2, b3) 2,2 (T2), with support in σ1. It is not in S1,1 is an element of S1,1 is not polynomial on σ1. Thus we have dim S1,1 2,2 (T2) = dim S1,1 2,2 (T1), since the function 2,2 (T1) + 1. Notice that T2 is (2, 2)-weighted but not 3-weighted, since any new maximal segment intersects two of the other new maximal segments. For a general hierarchical T-mesh, we consider a sequence of T-meshes T0, . . . Tl where T0 has one cell, Tl = T and such that Ti+1 refines Ti by inserting new edges. We can assume that at each level i 6= 0 a new maximal interior segment ρi appears and that we number the maximal interior segments of T in the order they appear during this subdivision. Notice that any maximal interior segment of T extends one of the maximal interior segments ρi and thus its weight is bigger. Notice also that the maximal interior segment ρi introduced at level i extends to a maximal segment of T , which may intersect the boundary. In this case, it is not involved in the dimension upper bound. Then, we have the following corollary: Proposition 5.3. Let T be a hierarchical T-mesh. (5.3) 9f2 − 6f o 1 + 4f o 0 ≤ dim S1,1 2,2 (T ) ≤ 9f2 − 6f o 1 + 4f o 0 + σ where σ is the number of levels of the subdivision where a maximal interior segment with no-interior point is introduced. Proof. Consider the new maximal interior segment ρi of Ti appearing at level i. By construction, the end points of ρi are not on maximal interior segments of bigger index. Thus ω(ρi) ≥ 2. If ρi contains an interior vertex then ω(ρi) ≥ 3 and (3 − ω(ρi))+ = 0. Otherwise (3 − ω(ρi))+ = 1. As ρi extends to a maximal segment ρi which is either interior or intersecting the boundary, we have (3 − ω(ρi))+ ≤ (3 − ω(ρi))+ with the convention that ω(ρi) = 3 if ρi is intersecting the boundary. 20 BERNARD MOURRAIN By Theorem 3.7, we have 0 ≤ h1,1 2,2(T ) ≤ l l (3 − ω(ρi))+ ≤ (3 − ω(ρi))+ Xi=1 Xi=1 introduced at level i and ρi is its extension in T . By the previous remarks,Pl where l is the number of levels in the subdivision, ρi is the maximal interior segment i=1(3− ω(ρi))+ = σ is the number of levels of the subdivision where a maximal interior segment with no interior point is introduced. Using Theorem 3.1, this proves the bound on the dimension of S1,1 (cid:3) 2,2 (T ). Examples 5.1 and 5.2 show that the dimension can be given by the upper bound. On other other hand, for any (3,3)-weighted hierarchical subdivision, the lower bound is reached. This shows that the inequalities (5.3) are optimal for dim S1,1 2,2 (T ). Remark 5.4. In the case of a hierarchical subdivision where some cells of a given level are subdivided (as σ1 in Example 5.2) into 9 sub-cells which have the same length and height, it can be proved that the dimension is in fact: dim S1,1 2,2 (T ) = 9f2 − 6f o 1 + 4f o 0 + σ where σ is the number of isolated subdivided cells (i.e. the cell is subdivided, not touching the boundary and the adjacent cells sharing an edge are not subdivided) at some level of the subdivision. Indeed, any maximal interior segment subdividing a non-isolated cell contains an interior point and is not involved in the upper bound. As in Example 5.2, only the isolated cells have a maximal interior segment with no interior points. This example also shows that a new C(1,1) bi-quadratic basis element can be constructed for each isolated cell, proving that the upper bound is reached. This gives a dimension formula similar to the one in [7], for a slightly different subdivision strategy. 5.3. Bicubic C1,1 T-splines. For m = m′ = 3 and r = r′ = 1, that is for piecewise bicubic polynomial functions which are C1, Proposition 4.3 yields h1,1 3,3(T ) = 0. Using Theorem 3.1 and Lemma A.1 in the appendix, we obtain: 0 + f b 3,3 (T ) = 16f2 − 8f o 0 = 4(f + dim S2,2 1 + 4f o (5.4) 0 ). 5.4. Bicubic C2,2 T-splines. For m = m′ = 3 and r = r′ = 2, by Theorem 3.1 and Lemma A.1, we have: (5.5) dim S2,2 3,3(T ). If T is a hierarchical (4, 4)-weighted subdivision, then by Theorem 3.9, we have h2,2 3,3(T ) = 0. 3,3 (T ) = 16f2 − 12f o 3,3(T ) = f + 0 + 8 + h2,2 0 + h2,2 0 + 2f b 1 + 9f o 0 − f T References [1] Dmitry Berdinsky, Min-jae Oh, Tae-wan Kim, and Bernard Mourrain. On the problem of instability in the dimension of a spline space over a T-mesh. Comput. Graph., 36(5):507 -- 513, 2012. [2] L. Billera. Homology theory of smooth splines: generic triangulations and a conjecture of Strang. Trans. Amer. Math. Soc., 310:325 -- 340, 1988. [3] A. Buffa, D. Cho, and G. Sangalli. Linear independence of the T-spline blending functions associated with some particular T-meshes. Computer Methods in Applied Mechanics and Engineering, 199:14371445, 2010. [4] Henri Cartan and Samuel Eilenberg. Homological Algebra. Princeton University Press, 1999. ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES 21 [5] Carl deBoor. A Practical Guide to Splines. Springer, 2001. [6] Jiansong Deng, Falai Chen, and Yuyu Feng. Dimensions of spline spaces over T-meshes. J. Comput. Appl. Math., 194(2):267 -- 283, 2006. [7] Jiansong Deng, Falai Chen, and Liangbing Jin. Dimensions of biquadratic spline spaces over T-meshes. arXiv:0804.2533v1. [8] Jiansong Deng, Falai Chen, Xin Li, Changqi Hu, Weihua Tong, Zhouwang Yang, and Yuyu Feng. Polynomial splines over hierarchical T-meshes. Graph. Models, 70(4):76 -- 86, 2008. [9] Tor Dokken, Tom Lyche, and Kjell-Fredrik Pettersen. Locally refined splines. Preprint, 2010. [10] Richard Ehrenborg and Gian-Carlo Rota. Apolarity and canonical forms for homogeneous polynomials. Eur. J. Comb., 14(3):157 -- 181, 1993. [11] Gerald Farin. Curves and Surfaces for Computer Aided Geometric Design: A Practical Guide, 5th Edition. Morgan Kaufmann, San Mateo, CA, 2001 [12] Allen Hatcher. Algebraic Topology. Cambridge University Press, 2002. [13] Zhangjin Huang, Jiansong Deng, Yuyu Feng, and Falai Chen. New proof of dimension formula of spline spaces over T-meshes via smoothing cofactors. J. of Computational Mathematics, 24(4):501514, 2006. [14] Thomas J.R. Hughes, John A. Cottrell, Yuri Bazilevs. Isogeometric analysis: CAD, finite elements, NURBS, exact geometry, and mesh refinement. Computer Methods in Applied Mechanics and Engineering 194, 39-41, pp 4135-4195, 2005. [15] Joseph P. S. Kung and Gian-Carlo Rota. The invariant theory of binary forms. Bull. Amer. Math. Soc. (N.S.), 10(1):27 -- 85, 1984. [16] Chong-Jun Li, Ren-Hong Wang, and Feng Zhang. Improvement on the dimensions of spline spaces on T-mesh. Journal of Information & Computational Science, 3(2):235244, 2006. [17] Xin Li and Falai Chen. On the instability in the dimension of splines spaces over t-meshes. Comput. Aided Geom. Des., 28(7):420 -- 426, 2011. [18] G. Salmon. Modern Higher Algebra. G.E. Stechert and Co., New York, 1885. Reprinted 1924. [19] Hal Schenck and Mike Stillman. Local cohomology of bivariate splines. Journal of Pure and Applied Algebra, 117-118:535 -- 548, 1997. [20] Thomas W. Sederberg, David L. Cardon, G. Thomas Finnigan, Nicholas S. North, Jianmin Zheng, and Tom Lyche. T-spline simplification and local refinement. ACM Trans. Graph., 23(3):276 -- 283, 2004. [21] Thomas W. Sederberg, Jianmin Zheng, Almaz Bakenov, and Ahmad Nasri. T-splines and T-nurccs. ACM Trans. Graph., 22(3):477 -- 484, 2003. [22] Edwin H. Spanier. Algebraic Topology. MacGraw-Hill, 1966. [23] Ren-Hong Wang. Multivariate Spline Functions and Their Applications. Kluwer Academic Publishers, 2001. Appendix A. Combinatorial properties We recall some well-known enumeration results for a T-mesh of an axis-aligned rectangular domain Ω. Lemma A.1. • f2 = f + 1 = 2f + • f o 0 = f + • f o 0 + 1 0 + 1 2 f b 2 f T 0 − 1 0 + 1 0 + 3 2 f b 2 f T 0 + f T 0 0 − 2 Proof. Each face σ ∈ T2 is a rectangle with 4 corners. If we count these corners for all cells in T2, we enumerate 4 times the crossing vertices, 2 times the T -vertices which are interior or on the boundary and one time the corner vertices of Ω. This yields the relation 4f2 = 4f + 0 − 4)) + 4. 0 + 2(f T 0 + (f b Each interior edge τ ∈ T o 1 has two end points. Counting these end points for all interior edges, we count 4 times the crossing vertices, 3 times the T -vertices which 22 BERNARD MOURRAIN are interior and one time the T -vertices on the boundary: 2f o 1 = 4f + 0 + 3f T 0 + (f b 0 − 4). Finally, as an interior vertex is a crossing vertex or a T -vertex, we have 0 = f + f o 0 + f T 0 . (cid:3) Appendix B. Complexes and homology Let us recall here the basic properties that we will need on complexes of vector spaces. Given a sequence of K-vectors spaces Ai, i = 0, . . . , l and linear maps ∂i : Ai → Ai+1, we say that we have a complex A : Al → Al−1 → · · · Ai → Ai−1 → · · · A1 → A0 if im ∂i ⊂ ker ∂i−1. Definition B.1. The ith homology Hi(A) of A is ker ∂i−1/ im ∂i for i = 1, . . . , l. The complex A is called exact (or an exact sequence) if Hi(A) = 0 (ie. im ∂i = ker ∂i−1) for i = 1, . . . , l. If the complex is exact and Al = A0 = 0, we have l−1 (−1)i dim Ai = 0. Xi=1 Given complexes A = (Ai)i=0,...,l B = (Bi)i=0,...,l, C = (Ci)i=0,...,l and exact se- quences for i = 0, . . . , l, we have a long exact sequence [4], [22][p. 182]: 0 → Ai → Bi → Ci → 0 · · · → Hi+1(C) → Hi(A) → Hi(B) → Hi(C) → Hi−1(A) → · · · Appendix C. Dual topological complex The dual complex T ⋆ of the subdivision T , is such that we have the following properties. • a face σ ∈ T2 is a vertex of the dual complex T ⋆. • an edge of T ⋆ is connecting two elements σ, σ′ ∈ T2 if they share a common 1 . Thus it is identified with the edge τ of T between (interior) edge τ ∈ T o σ, σ′; • a face of T ⋆ corresponds to an elements γ ∈ T o 0 . It is either a triangle if γ is a T -junction or a quadrangle if γ is a crossing vertex. Notice that the boundary cells of T correspond to boundary vertices of T ⋆. They are connected by boundary edges which belong to a single face of T ⋆. Appendix D. Topological chain complex In this appendix section, we recall the main properties of the topological chain complex R m,m ′ (T o) : Lσ∈T2 where [σ]Rm,m ′ → Lτ ∈T o 1 [τ ]Rm,m ′ → Lγ∈T o 0 [γ]Rm,m ′ → 0 • ∀γ ∈ T o • ∀τ = [γ1, γ2] ∈ T o o , ∂0([γ]) = 0. 1 , ∂1([γ1, γ2]) = [γ1] − [γ2] with [γ] ≡ 0 iff γ ∈ ∂Ω. and ON THE DIMENSION OF SPLINE SPACES ON PLANAR T-MESHES 23 • ∀σ ∈ T o 2 with its counter-clockwise boundary formed by the edges [γ1, γ2], . . . , [γl, γ1], ∂2(σ) = [γ1, γ2] + · · · + [γl, γ1] with [γ, γ ′] ≡ 0 iff γ, γ ′ ∈ ∂Ω. We assume that Ω is simply connected and Ωo is connected. We prove that Rm,m′(T o) is acyclic on a T -mesh of Ω. Proposition D.1. H0(Rm,m′(T o)) = 0. Proof. Let γ ∈ T o τl = γlγl+1, such that τi ∈ T o 0 . There is a sequence of edges τ0 = γ0γ1, 0 and γl+1 = γ. Then τ1 = γ1γ2, . . ., 1 , γ0 6∈ T o ∂1([τ0] + · · · + [τl]) = [γ1] − [γ0] + · · · + [γl+1] − [γl] = [γ] since [γ0] = 0 and [γl+1] = [γ]. Multiplying by any element in Rm,m′, we get that [γ]Rm,m′ ⊂ im ∂1 and thus H0(Rm,m′(T o)) = 0. (cid:3) Proposition D.2. H1(Rm,m′(T o)) = 0. 1 the image of ∂2. For each γ ∈ T o Proof. Let p = Pτ ∈T o P ετ pτ = 0 with ετ = 1 if τ ends at γ, ετ = −1 if τ starts at γ and ετ = 0 [τ ]pτ ∈ ker ∂1 with pτ ∈ Rm,m′. Let us prove that p is in 0 and each edge τ which contains γ, we have 1 , we define εσ,τ = 1 if τ is oriented counter-clockwise on the boundary of σ, εσ,τ = −1 if τ is oriented counter-clockwise on the boundary of σ and εσ,τ = 0 otherwise. 2 and τ ∈ T o For any σ ∈ T o otherwise. For any oriented edge of the dual graph T ⋆ from σ′ to σ, let us define ∂⋆ 1 ([σ′, σ]) = 1 ([σ′, σ]), since the orien- εσ,τ pτ . Notice that ∂⋆ tation of τ on the boundary of σ and σ′ are opposite. 1 ([σ, σ′]) = εσ′,τ pτ = −εσ,τ pτ = −∂⋆ Let σ0 be the 2-face of T with the lowest left corner for the lexicographic ordering. We order the cells σ ∈ T2 according to their distance to σ0 in this dual graph T ⋆. We define an element q = Pσ∈T2 qσ[σ] where qσ ∈ Rm,m′ by induction using this order, as follows: • qσ0 = 0; • For any σ, σ′ ∈ T2, if σ > σ′ and σ and σ′ share a common edge τ , then qσ = qσ′ + ∂⋆ 1 ([σ′, σ]). Thus, if [σ0, σ1], [σ1, σ2], . . . , [σk−1, σk] is a path of T ⋆ connecting σ0 to σk = σ with 1 ([σi, σi+1]). Let us prove that this definition does not i=0 ∂⋆ σi+1 > σi then qσ =Pk−1 depend on the chosen path between σ0 and σ. We first show that for any face γ⋆ of T ⋆ attached to a vertex γ, if its counter- clockwise boundary is formed by the edges [σ, σ′], [σ′, σ′′], . . . , [σ′′′, σ] corresponding to the edges τ, τ ′, τ ′′, . . . of T containing γ, then 1 ([σ′′′, σ]) = εσ,τ pτ + εσ′,τ ′pτ ′ + εσ′′,τ ′′pτ ′′ + · · · = 0. 1 ([σ, σ′]) + ∂⋆ ∂⋆ By changing the orientation of an edge τ , we replace pτ by −pτ and εσ,τ by −εσ,τ so that the quantity εσ,τ pτ is not changed. Thus we can assume that all the edges τ, τ ′, τ ′′, . . . are pointing to γ. As p ∈ ker ∂1, we have pτ + pτ ′ + pτ ′′ + · · · = 0. 1 ([σ′, σ′′] + · · · + ∂⋆ Now as the cells σ, σ′, σ′′, . . . , σ are ordered counter-clockwise around γ and as the edges are pointing to γ, we have εσ,τ = εσ′,τ ′ = εσ′′,τ ′′ = · · · = 1, so that the 1 ([σ′′′, σ]) over a the boundary of a face γ⋆ of sum ∂⋆ T ⋆ is 0. 1 ([σ′, σ′′] + · · · + ∂⋆ 1 ([σ, σ′]) + ∂⋆ By composition, for any loop of T ⋆, the sum on the corresponding oriented edges is 0. This shows that the definition of qσ does not depend on the oriented path from σ0 to σ. 24 BERNARD MOURRAIN By construction, we have ∂2(q) = Xσ∈T2 (Xτ ∈T o 1 εσ,τ qσ[τ ]) = Xτ ∈T o 1 (Xσ∈T2 εσ,τ qσ)[τ ]. (Xσ For each interior edge τ ∈ T o τ. Thus, we have εσ1,τ = −εσ2,τ and qσ1 = qσ2 + εσ1,τ pτ . We deduce that 1 , there are two faces σ1 > σ2, which are adjacent to εσ,τ qσ) = εσ1,τ qσ1 + εσ2,τ qσ2 = εσ1,τ (qσ2 + εσ1,τ pτ ) + εσ2,τ qσ2 = pτ . This shows that ∂2(q) = p. In other words, im ∂2 = ker ∂1 and H1(Rm,m′ (T o)) = 0. (cid:3) Proposition D.3. If Ω has one connected component. Then H2(Rm,m′(T o)) = Rm,m′. Proof. An element of H2(Rm,m′(T o)) = ker ∂2 is a collection of polynomials (pσ)σ∈T2 such that pσ ∈ Rm,m′ and pσ = pσ′ if σ and σ′ share an (internal) edges. As T is a subdivision of a rectangle D0, all faces σ ∈ T2 share pairwise an edge. Thus pσ = pσ′ for all σ, σ′ ∈ T2 and H2(Rm,m′ (T )) = Rm,m′. (cid:3) Notice that by counting the dimensions in the exact sequence Rm,m′(T ), we recover the well-known Euler formula: f2 − f1 + f0 = 1 (the domain Ω has one connected component). T ε t3 t2 t1 γ s1 s2 s3 t5 t4 t3 t2 t1 σ s1 s2 s3 s4 t5 t4 t3 t2 t1 γ s1 s2 s3 s4 s5
1505.03108
2
1505
2017-03-27T09:16:51
Equivariantly uniformly rational varieties
[ "math.AG" ]
We introduce equivariant versions of uniform rationality: given an algebraic group G, a G-variety is called G-uniformly rational (resp. G-linearly uniformly rational) if every point has a G-invariant open neighborhood equivariantly isomorphic to a G-invariant open subset of the affine space endowed with a G-action (resp. linear G-action). We establish a criterion for Gm-uniform rationality of affine variety equipped with hyperbolic Gm-action with a unique fixed point, formulated in term of their Altmann-Hausen presentation. We prove the Gm-uniform rationality of Koras-Russel threefolds of the first kind and we also give example of non Gm-uniformly rational but smooth rational Gm-threefold associated to pairs of plane rational curves birationally non equivalent to a union of lines.
math.AG
math
EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES CHARLIE PETITJEAN Abstract. We introduce equivariant versions of uniform rationality: given an algebraic group G, a G-variety is called G-uniformly rational (resp. G-linearly uniformly rational) if every point has a G-invariant open neighborhood equivariantly isomorphic to a G-invariant open subset of the affine space endowed with a G-action (resp. linear G-action). We establish a criterion for Gm-uniform rationality of smooth affine varieties equipped with hyperbolic Gm-actions with a unique fixed point, formulated in term of their Altmann-Hausen presentation. We prove the Gm- uniform rationality of Koras-Russell threefolds of the first kind and we also give an example of a non Gm-uniformly rational but smooth rational Gm-threefold associated to pairs of plane rational curves birationally non equivalent to a union of lines. Introduction A uniformly rational variety is a variety for which every point has a Zariski open neighborhood isomorphic to an open subset of an affine space. A uniformly rational variety is in particular a smooth rational variety, but the converse is an open question [13, p.885]. In this article, we introduce stronger equivariant versions of this notion, in which we require in addition that the open subsets are stable under certain algebraic group actions. The main motivation is that for such varieties uniform rationality, equivariant or not, can essentially be reduced to rationality questions at the quotient level. We construct examples of smooth rational but not equivariantly uniformly rational varieties, the question of their uniform rationality is still open. We also establish equivariant uniform rationality of large families of affine threefolds. We focus mainly on actions of algebraic tori T. The complexity of a T-action on a variety is the codimension of a general orbit, in the case of a faithful action, the complexity is thus simply dim(X) − dim(T). Complexity zero corresponds to toric varieties, which are well-known to be uniformly rational when smooth. In fact they are even T-linearly uniformly rational in the sense of Definition 4 below. The same conclusion holds for smooth rational T-varieties of complexity one by a result of [21, Chapter 4]. In addition, by [3, Theorem 5] any smooth complete rational T-variety of complexity one admits a covering by finitely many open charts isomorphic to the affine space. In this article, as a step toward the understanding of T-varieties of higher complexity, we study the situation of affine threefolds equipped with hyperbolic Gm-actions. We use the general description developed by Altmann, Hausen and Süss (see [1, 2] ) in terms of pairs (Y, D), where Y is a variety of dimension dim(X) − dim(T) and D a so-called polyhedral divisor on Y . In our situation, Y is a rational surface and our main result, Theorem 16, allows us to translate equivariant uniform rationality into a question of birational geometry of curves on rational surfaces. The article is organized as follows. In the first section we introduce equivariant versions of uniform rationality and summarize A-H presentations of affine Gm-varieties. The second section explains how to use these presentations for the study of uniform rationality of these varieties. In the third section, we focus on families of Gm-rational threefolds, we show, for example, that all Koras-Russell threefolds of the first kind, and certain ones of the second kind (see [16, 17]) are equivariantly uniformly rational and therefore uniformly rational. In a fourth section we find examples of smooth rational Gm-threefolds including other Koras-Russell threefolds which are not equivariantly uniformly rational. It is not known if these varieties are uniformly rational, without any group action. In the last section, we introduce a weaker notion of equivariant uniform rationality and we illustrate differences between all these notions. The author would like to thank Karol Palka and Jérémy Blanc for helpful discussions concerning Subsection 3.2. 2000 Mathematics Subject Classification. 14L30, 14R20, 14M20, 14E08. Key words and phrases. uniform rationality, T-varieties, hyperbolic Gm-actions, birational equivalence of pairs of curves, Koras- Russell threefolds. 1 EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 2 1. Preliminaries 1.1. Basic examples of uniformly rational varieties. Recall that a a variety of dimension n is called uniformly rational if every point has a Zariski open neighborhood isomorphic to an open subset of An. Some partial results are known, for instance every smooth complete rational surface is uniformly rational. In fact it follows from [5, 6] that the blowup of a uniformly rational variety along a smooth subvariety is again uniformly rational. Since open subset of uniformly rational varieties are uniformly rational, it follows that every open subsets of the blowup of a uniformly rational variety along a smooth subvariety is again uniformly rational. In particular this holds for affine modifications of uniformly rational varieties along smooth subvarieties. Definition 1. [18, 9] Let (X, D, Z) be a triple consisting of a variety X, an effective Cartier divisor D on X and a closed sub-scheme Z with ideal sheaf IZ ⊂ OX (−D). The affine modification of the variety X along D with center Z is the scheme X ′ = XZ \ D′ where D′ is the proper transform of D in the blow-up XZ → X of X along Z. A particular type of affine modification is the hyperbolic modification of a variety X with center at a closed sub-scheme Z ⊂ X (see [27]): It is defined as the affine modification of X × A1 with center Z × {0}, and divisor X × {0}. As an immediate corollary of [5, Proposition 2.6], we obtain the following result: Proposition 2. Affine modifications and hyperbolic modifications of uniformly rational varieties along smooth centers are again uniformly rational. Example 3. Let An = Spec(C[x1, . . . , xn]) and let I = (f, g) be the defining ideal of a smooth subvariety in An. Then the affine modification of An with center I = (f, g) and divisor D = {f = 0} is isomorphic to the subvariety X ′ ⊂ An+1 defined by the equation: {g(x1, . . . , xn) − yf (x1, . . . , xn) = 0} ⊂ An+1 = Spec(C[x1, . . . , xn, y]). It is a uniformly rational variety. 1.2. Equivariantly uniformly rational varieties. Let G be an affine algebraic group and let X be a G-variety, that is an algebraic variety endowed with a G-action. We introduce equivariant versions, of uniform rationality. Definition 4. Let X be a G-variety and x ∈ X. i) We say that X is G-linearly rational at the point x if there exists a G-stable open neighborhood Ux of x, a linear representation of G → GLn(V ) and a G-stable open subset U ′ ⊂ V ≃ An such that Ux is equivariantly isomorphic to U ′. ii) We say that X is G-rational at the point x if there exists an open G-stable neighborhood Ux of x, an action of G on An and an open G-stable subset U ′ ⊂ An such that Ux is equivariantly isomorphic to U ′. iii) A G-variety that is G-linearly rational (respectively G-rational) at each point is called G-linearly uniformly rational (respectively G-uniformly rational ). iv) A G-variety that admits a unique fixed point x0 by the G-action is called G-linearly rational (respectively G-rational) if it is G-linearly rational (respectively G-rational) at x0. G-linearly uniformly rational or just G-uniformly rational varieties are always uniformly rational. The converse is trivially false: for instance the point [1 : 0] in P1 does not admit any Ga-invariant affine open neighborhood for the action defined by t · [u : v] → [u + tv : v]. For algebraic tori T, as already mentioned in the introduction, it is a classical fact that smooth toric varieties are T-linearly uniformly rational. Moreover, it is known that every effective T-action on An is linearisable for dim(T) ≥ n − 1 (see [14] for n = 2 and [4] for the general case), and in another direction every algebraic Gm-action on A3 is linearisable [17]. As a consequence, we obtain the following: Theorem 5. For T-varieties of complexity 0, 1 and for Gm-threefolds the properties of being T-linearly uniformly rational and T-uniformly rational are equivalent. 1.3. Hyperbolic Gm-actions on smooth varieties. By a Theorem of Sumihiro (see [25]) every normal Gm- variety X admits a cover by affine Gm-stable open subsets. This reduces the study of Gm-linearly uniformly rational varieties to the affine case. Recall that the coordinate ring A of an affine Gm-variety X is Z-graded in a natural way by its subspaces An := {f ∈ A/f (λ · x) = λnf (x), ∀λ ∈ Gm} of semi-invariants of weight n. In particular A0 is the ring of invariant functions on X. If X is smooth with positively graded coordinate ring, then by [20], X has the structure of a vector bundle over its fixed point locus X Gm, and hence the question whether X is Gm-linearly uniformly rational becomes intimately related to the uniform rationaly of X Gm. In this subsection, we consequently focus on hyperbolic Gm-actions. We summarize the correspondence between smooth affine varieties X EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 3 endowed with an effective hyperbolic Gm-action and pairs (Y, D) where Y is a variety, that we call A-H quotient, and D is a so-called segmental divisor on Y . All the definitions and constructions are adapted from [1]. Definition 6. A Gm-action is said to be hyperbolic if there is at least one n1 < 0 and one n2 > 0 such that An1 and An2 are nonzero. Definition 7. Let X = Spec(A) be a smooth affine variety equipped with a hyperbolic Gm-action. i) We denote by q : X → Y0(X) := X//Gm = Spec(A0) the categorical quotient of X. ii) The A-H quotient Y (X) of X is the blow-up π : Y (X) → Y0(X) of Y0(X) with center at the closed subscheme a normal semi-projective variety (see [1]). By virtue of [26, Theorem 1.9, proposition 1.4], Y (X) is isomorphic to defined by the ideal I = hAd · A−di, where d > 0 is chosen such that Ln∈Z Adn is generated by A0 and A±d. It is the fiber product of the schemes Y±(X) = ProjA0 (Ln∈Z≥0 In the remainder of the article, we use the notation π : VI → V to refer to the blow-up of an affine variety V A±n) over Y0(X). with center at the closed sub-scheme defined by the ideal I ⊂ Γ(V, OV ). Definition 8. A segmental divisor D on a normal algebraic variety Y is a formal finite sum D = P[ai, bi] ⊗ Di, where Di are prime Weil divisors on Y and [ai, bi] are closed intervals with rational bounds ai ≤ bi. The set of all closed intervals with rational bounds, admits a structure of abelian semigroup for the Minkowski sum, the Minkowski sum of two intervals [ai, bi] and [aj, bj] being the interval [ai + aj, bi + bj]. Every element n ∈ Z determines a map from segmental divisors to the group of Weil Q-divisors on Y : where for all i, qi ∈ Q is the minimum of nai and nbi. D =X[ai, bi] ⊗ Di → D(n) =X qiDi, Definition 9. A proper-segmental divisor (ps-divisor) D on a variety Y is a segmental divisor on Y such that for every n ∈ Z, D(n) satisfies the following properties: 1) D(n) is a Q-Cartier divisor on Y . 2) D(n) is semi-ample, that is, for some p ∈ Z>0, Y is covered by complements of supports of effective divisors linearly equivalent to D(pn). 3) D(n) is big, that is, for some p ∈ Z>0, there exists an effective divisor D linearly equivalent to D(pn) such that Y \ Supp(D) is affine. In the particular case of hyperbolic Gm-action, the main Theorem of [1] can be reformulated as follows: Theorem 10. For any ps-divisor D on a normal semi-projective variety Y the scheme S(Y, D) = Spec(Mn∈Z Γ(Y, OY (D(n)))) is a normal affine variety of dimension dim(Y ) + 1 endowed with an effective hyperbolic Gm-action, whose A-H quotient Y (S(Y, D)) is birationally isomorphic to Y . Conversely any normal affine variety X endowed with an effective hyperbolic Gm-action is isomorphic to S(Y (X), D) for a suitable ps-divisor D on Y (X). Remark 11. Alternatively, see [8, 10], any finitely generated Z-graded algebra A can be written in the form A =Mn<0 Γ(Y, OY (nD−)) ⊕ Γ(Y, OY ) ⊕Mn>0 Γ(Y, OY (nD+)) where (Y, D+, D−) is a triple consisting in a normal variety Y and suitable Q-divisors D+ and D− on it. These two presentations are obtained from each other by setting D− = D(−1), D+ = D(1) and conversely D = {1}D+ + [0, 1](−D− − D+). Remark 12. A method to determine a possible ps-divisor D such that X ≃ S(Y, D) is to embed X as a Gm-stable subvariety of an affine toric variety (see [1, section 11]). The calculation is then reduced to the toric case by considering an embedding in An endowed with a linear action of a torus T of sufficiently large dimension n. The inclusion of Gm ֒→ T corresponds to an inclusion of the lattice Z of one parameter subgroups of Gm in the lattice Zn = N of one parameter subgroups of T. We obtain the exact sequence: 0 / Z s F / N = Zn P N = Zn/Z / 0 , / / / / z z / EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 4 ∨ where F is given by the induced action of Gm on An and s is a section of F . Let vi, for i = 1, . . . , n, be the first integral vectors of the unidimensional cones generated by the i-th column vectors of P considered as rays in the lattice N ≃ Zn−1. Let Z be the toric variety of dimension dim(An) − dim(T), determined by the fan in N whose cones are generated the vi for i = 1, . . . , n. Then each vi corresponds to a T′-invariant divisor where T′ = Spec(C[N ]. By [1, section 11] Z contains the A-H quotient of X as a closed subset, and the support of Di is obtained by restricting the T′-invariant divisor corresponding to vi to Y . If X is the affine space endowed with a linear action of Gm, then Z is itself the A-H quotient of An. The segment associated to the divisor Di is equal to s(Rn ≥0 ∩ P −1(vi)). The section s can further be chosen so that the number of non zero coefficients in the associated matrix is minimal. The ps-divisor D from a such section will be called minimal. We would like to point out that this notion is more restrictive than that given in [1], in particular every minimal ps-divisor in our sense is also in the sense of [1]. 2. Algebro-combinatorial criteria for Gm-linear rationality Given a a smooth rational variety X endowed with a hyperbolic Gm-action which admits a unique fixed point x0, we develop in this section a method to test whether X is Gm-rational. ii) Let ϕ : Y → Y ′ be a proper dominant map. The polyhedral push-forward of D is defined by ϕ∗(D) := i) Let ϕ : Y → Y ′ be a morphism such that ϕ(Y ) is not contained in Supp(D′ Definition 13. [1, Definition 8.3] Let Y and Y ′ be normal semi-projective varieties and let D′ = P[a′ and D =P[ai, bi] ⊗ Di be ps-divisors on Y ′ and Y respectively. of D′ is defined by ϕ∗(D′) :=P[a′ P[ai, bi] ⊗ ϕ∗(Di), where ϕ∗(Di) is the usual push-forward of Di. Let ϕ : Y → Y ′ be a birational morphism and let D′ be a divisor on Y ′, then we decompose the pull-back of D′ by ϕ as follows: ϕ∗(D′) = (ϕ−1)∗(D′) + R where (ϕ−1)∗(D′) is the strict transform of D′ and R is supported in the exceptional locus of ϕ. i) for any i. The polyhedral pull-back i) is the usual pull-back of D′ i. i), where ϕ∗(D′ i] ⊗ ϕ∗(D′ i, b′ i, b′ i] ⊗ D′ i Definition 14. Two pairs (Yi, Di), i = 1, 2 consisting of a variety Yi and a Cartier divisor Di on Yi are called birationally equivalent if there exist a variety Z, and two proper birational morphisms ϕi : Z → Yi such that the strict transforms (ϕ−1 2 )∗(D2) of D1 and D2 respectively are equal. For ps-divisors, we extend this notion in the natural way to pairs (Yi, Di) consisting of a semi-projective variety Yi and a ps-divisor Di on Yi using the polyhedral pull-back defined above. 1 )∗(D1) and (ϕ−1 Since we consider hyperbolic Gm-actions with a unique fixed point, the construction of the A-H quotient Y as in Definition 7 ensures that Y (X) has only one exceptional divisor E over Y0(X). We denote by D the segmental divisor obtain from the ps-divisor D corresponding to X by removing all irreducible components whose support does not intersect E. The following example illustrate a situation for which D 6= D. Example 15. Let S be the affine surface defined by {x2y + x = z2} ⊂ A3 = Spec(C[x, y, z]) and let X := S × A1 be the cylinder over S, endowed with the hyperbolic Gm-action induced by the linear one λ(x, y, z, t) → (λ6x, λ−6y, λ3z, λ2t) on A4 = Spec(C[x, y, z, t]). Using the method described in Remark 12, we find that X is equivariantly isomorphic to S( A2 (u,v), D) with: D =(cid:26) 1 2(cid:27) D1 +(cid:26) 1 2(cid:27) D2 −(cid:26) 1 3(cid:27) D3 +(cid:20)0, 1 6(cid:21) E, where E is the exceptional divisor of the blow-up π : A2 D1, D2 and D3 are the strict transforms of the curves L1 = {v = 0}, L2 = {1 + v = 0} L3 = {u = 0} in A2 = Spec(C[u, v]). The divisor D2 does not intersect the exceptional divisor E, so: (u,v) → A2 ≃ Spec(C[u, v]) ≃ Spec(C[yt3, yx]) , and where D =(cid:26) 1 2(cid:27) D1 −(cid:26) 1 3(cid:27) D3 +(cid:20)0, 1 6(cid:21) E. Theorem 16. Let X be a smooth affine rational variety endowed with a hyperbolic Gm-action with a unique fixed point x0. Then X is Gm-rational if and only if the following holds: 1) There exists pairs (Y, D) and (Y ′, D′) such that S(Y, D) is equivariantly isomorphic to X and S(Y ′, D′) is equivariantly isomorphic to An endowed with a hyperbolic Gm-action. 2) The pairs (Y, D) and (Y ′, D′) are birationally equivalent. EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 5 Proof. Suppose that X is Gm-rational so that there exists an open Gm-stable neighborhood Ux0 of x0, an action of Gm on An, an open Gm-stable subvariety U ′ ⊂ An, and an equivariant isomorphism ϕ : Ux0 → U ′. We can always reduce to the case where Ux0 and U ′ are principal open sets. Indeed Ux0 is the complement of a closed stable subvariety of X determined by an ideal I = (f0, . . . , fk) where each fi ∈ Γ(X, OX ) is semi-invariant. As Ux0 contains x0, at least one of the fi does not vanish at x0. Denoting this function by f , the principal open subset Xf := X \ V (f ) is contained in Ux0. The restriction of ϕ to Xf induces an isomorphism between Xf and ϕ(Xf ). This yields a divisor U ′ \ ϕ(Xf ) on U ′. Since An is factorial, this divisor is the restriction of a principal divisor Div(f ′) on An for a certain regular function f ′. By construction ϕ induces an equivariant isomorphism between Xf and An f ′. Note that any non-constant semi-invariant function f ∈ Γ(X, OX ) such that f (x0) 6= 0 is actually invariant. Indeed, letting w be the weight of f , we have λ · f (x0) = λwf (x0) = f (λ−1 · x0) = f (x0) for all λ ∈ Gm, and so w = 0. Let (Y, D) be the pair corresponding to X with D minimal in the sense defined in Remark 12. We can identify every invariant function f on X non vanishing at x0 with an element f of Γ(Y, OY ) such that V (f ) ⊂ Y does not contain any irreducible component of Supp( D). Indeed, every such invariant function corresponds via the algebraic quotient morphism q : X → Y0 to a function on Γ(Y0, OY0 ) which does not vanish at q(x0). Since the center of the blow-up π : Y → Y0 is supported by q(x0) we can in turn identify f with a regular function on Y . We denote by Yf the corresponding open subset of Y where f does not vanish, so that with our assumption Y (Xf ) = Yf . By virtue of [2, proposition 3.3], for a ps-divisor D on a normal, semiprojective variety Y , let Df be the localization of D by f . Then Df is a ps-divisor on Yf , and the canonical map Df → D describes the open embedding Xf → X. Denoting i : Yf ֒→ Y the canonical open embedding, we said that the pair (Yf , Df = i∗(D)) describes the equivariant open embedding j : Xf ≃ S(Yf , i∗(D)) ֒→ X, and we have the following diagram: Xf q Xf //Gm = Y 0,f πY ′ Yf j i X = S(Y, D) q / Y0 = X//Gm π Y = BlI (Y0). A similar description holds for the principal invariant open subset An Gm-action. We denote the A-H quotient Y (An f ′ simply by Y ′ f ′ ) of An f ′ of An endowed with a hyperbolic f ′ and the corresponding ps-divisor by D′ f ′. By [1, Corollary 8.12.] Xf and An variety Y ′′, birational morphisms σ1 : Y f → Y ′′ and σ2 : Y ′ D ∼= σ∗ divisor E of Yf over Y0,f , or it is an isomorphism. But if σ1 contracts E then S(Y ′′, D′′) is not equivariantly isomorphic to Xf by Definition 7. Therefore σ1 is an isomorphism. The same holds for σ2. 2(D′′). Since σ1 is projective and birational, it either contracts the unique exceptional f ′ are equivariantly isomorphic if and only if there exist a normal semi-projective f ′ → Y ′′ and a ps-divisor D′′ on Y ′′such that 1 (D′′) and D′ f ′ ∼= σ∗ f ′ are minimal, the pairs (Yf , Df ) and (Y ′ Since Df and D′ isomorphism Φ : Yf → Y ′ birationally equivalent and we obtain a commutative diagram : f ′ such that (Φ−1)∗(D′ f ′ , D′ f ′ ) are equivalent, that is, there exists an f ′) = Df . This implies that the pairs (Y, D) and (Y ′, D′) are S(Y ′, D′) = An q An//Gm = Y ′ 0 Y ′ j′ i′ An f ′ ≃ Xf q Y ′ 0,f ′ ≃ Y 0,f Y ′ f ′ ≃ Yf j i X = S(Y, D) q X//Gm = Y0 Y. Conversely, assume that X = S(Y, D) and An = S(Y ′, D′) endowed with an hyperbolic Gm-action are such that the pairs (Y, D) and (Y ′, D′) are birationally equivalent. We can further assume that there exists a birational map   / /       /   / / O O O O     / /   ? _ o o     / / ? _ o o O O   / / O O ? _ o o O O EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 6 between Y and Y ′ which restricts to an isomorphism φ : Yg → Y ′ g′ between the principal open sets Yg of Y and Y ′ g′ of Y ′ corresponding to suitable functions g ∈ Γ (Y, OY ) and g′ ∈ Γ (Y ′, OY ′) whose zero loci do not intersect the exceptional divisors of Y → Y0 and Y ′ → Y ′ 0 respectively. Similarly as above the function g can be identified with an invariant function on X which does not vanish at x0. By virtue of [2, Proposition 3.3] the pair (Yg, Dg) describes the equivariant open embedding Xg ≃ S(Yg, Dg) ֒→ X. In the same way, g′ corresponds to an invariant function on Anand the pair (Yg′ , Dg′ ) describes the equivariant open embedding An g′ ) ֒→ An. This gives the result. (cid:3) g′ ≃ S(Y ′ g′ , D′ 3. Examples of Gm-uniformly rational threefolds In the particular case of affine threefolds, Gm-linear uniform rationality is reduced (by the previous section) to a problem of birational geometry in dimension 2. Indeed, using Theorem 16, the question may then be considered at the level of the A-H quotients which are rational semi-projective surfaces. 3.1. Hyperbolic Gm-action on A3. Using this presentation and the fact that every algebraic Gm-action on A3 = Spec(C[x, y, z]) is linearisable [17], we are able to characterize hyperbolic Gm-actions on A3 in terms of their A-H presentations. Indeed let Gm × A3 → A3 be an effective hyperbolic Gm-action given by λ · (x, y, z) → (λax, λby, λ−cz) with (a, b, c) ∈ Z3 >0. After a suitable Gm-invariant cyclic cover along coordinate axes, we can assume that A3//Gm ≃ A2, and the relation between such cyclic covers ans the A-H presentations of the T-varieties are controlled by [23]. Let (α, β, γ) ∈ Z3 be such that αa + βb − γc = 1. Let ρ(a, c) be the greatest common divisor of a and c, let ρ(b, c) be the greatest common divisor of b and c, and let δ be the greatest common divisor of a ρ(a,c) and b ρ(b,c) Then we have: Proposition 17. Up to equivariant cyclic covers, every A3 endowed with a hyperbolic Gm-action is equivariantly isomorphic to the Gm-variety S(Y, D) with Y and D defined as follows: i) Y is isomorphic to a blow-up π : A2 → A2 of A2 at the origin. ii) D is of the form: D =(cid:26) αρ(a, c) c (cid:27) ⊗ D1 +(cid:26) βρ(b, c) c (cid:27) ⊗ D2 +(cid:20) γ δ , γ δ + 1 δc(cid:21) ⊗ E, with D1, D2 are strict transforms of the coordinate axes and E is the exceptional divisor of π. Proof. Let A3 be endowed with a linear action of Gm, the A-H presentation is obtained from the following exact sequence: 0 / Z s F / Z3 P Z2 / 0 , by the method described in Remark 12, where F = t(a, b, −c), s = (α, β, γ) and P = c 0 ρ(a,c) 0 c ρ(b,c) a ρ(a,c) ρ(b,c) !. b The algebraic quotient of A3 for an hyperbolic Gm-action is isomorphic to A2//µ where µ is a finite cyclic group [12], thus the A-H quotient Y (A3) is by construction a blow-up of A2//µ. In this case Y (A3) is smooth and corresponds to the toric variety Z defined in Remark 12 that is a blow-up of A2 whose center is supported at the origin. Let now us consider each vi, for i = 1, . . . , 3, as in remark 12, that is the first integral vectors of the unidimensional cones generated by the i-th column vectors of P = c v2 =(cid:18) 0 0 1 (cid:19) as rays defining a toric variety correspond to the generators of A2 and thus the associated divisors are the strict transforms of the coordinate axes and the last on v3 corresponds to the exceptional divisor. To determine the associated coefficients, we used the formula [ai, bi] = s(Rn ≥0 ∩ P −1(vi)) given in remark 12. ρ(a,c) 0 c ρ(b,c) a ρ(a,c) ρ(b,c) !. The first two v1 =(cid:18) 1 0 (cid:19) and b (cid:3) / / / / / EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 7 Example 18. [24, example 1.4.8] The presentation of A3 = Spec(C[x, y, z]) equipped with the hyperbolic Gm- action λ · (x, y, z) = (λ2x, λ3y, λ−6z) is S( A2 (u,v) → A2 the blow-up of A2 = Spec(C[u, v]) at the origin and (u,v), D) with π : A2 D =(cid:26)− 1 3(cid:27) D1 +(cid:26) 1 2(cid:27) D2 +(cid:20)0, 1 6(cid:21) E, where E is the exceptional divisor of the blow-up, D1 and D2 are the strict transforms of the lines {u = 0} and {v = 0} in A2 respectively. Indeed, A3//Gm = Spec(C[u, v]) and d > 0 in Definition 7 has to be chosen so that Ln∈Z Adn is generated by A0 and A±d. This is the case if d is the least common multiple of the weights of the Gm-action on A3. Thus d = 6 and Y (X) is the blow-up of A2 = Spec(C[u, v]) with center at the closed sub-scheme with ideal (u, v), i.e. the origin with our choice of coordinates. 3.2. Gm-linear uniform rationality. In this subsection we will prove that some hypersurfaces of A4 are Gm- linearly uniformly rational. In particular, every Koras-Russell threefold of the first kind X is Gm-linearly uniformly rational. These varieties are defined by equations of the form: {x + xdy + zα2 + tα3 = 0} ⊂ A4 = Spec(C[x, y, z, t]), where d ≥ 2, and α2 and α3 are coprime. They are smooth rational and endowed with hyperbolic Gm-actions with algebraic quotients isomorphic to A2//µ where µ is a finite cyclic group. They have been classified by Koras and Russell, in the context of the linearization problem for Gm-actions on A3 [17]. These threefolds can be viewed as affine modifications of A3 = Spec(C[x, z, t]) along the principal divisor Df = {f = 0} with center I = (f, g) where f = −xd and g = x + zα2 + tα3. But since the center is supported on the cuspidal curve included in the plane {x = 0} and given by the equation : C = {x = zα2 + tα3 = 0} (see [27]) their uniformly rationality does not follow directly from Corollary 2. 3.2.1. A general construction. Here we give a general criterion to decide the Gm-uniform rationality of certain threefolds, arising as stable hypersurfaces of A4 endowed with a linear Gm-action. Since X is rational, its A-H quotient Y (X) is also rational. The aim is to use the notion of birational equivalence of ps-divisors to construct an isomorphism between a Gm- stable open set of the variety X with a corresponding stable open subset of A3. By Theorem 16 and Proposition 17, the technique is to consider a well chosen sequence of birational transformations Y (X) → A2 which maps the support of the ps-divisor corresponding to the threefolds X on to the strict transforms of the coordinate lines and the exceptional divisor in A2. Moreover, since we look for Gm-stable open subset of X containing the fixed point, it is enough to consider birational map Y0(X) → A2 which send a pair of curves on the coordinates axes of A2. Let p ∈ C[v] be a polynomial of degree k ≥ 1 such that p(0) = 0, let α2, α3 and d be integers such that dα3 and α2 are coprime. Let X be a hypersurface in A4 = Spec(C[x, y, z, t]) defined by one of the following equations: X = {yd−1zα2 + tα3 + p(xy)/y = 0} ⊂ A4 = Spec(C[x, y, z, t]). Every such X is endowed with a hyperbolic Gm-action induced by the linear action on A4 defined by λ·(x, y, z, t) = (λα2α3x, λ−α2α3y, λdα3 z, λα2t). The unique fixed point for this action is the origin of A4 and is a point of X. Theorem 19. With the notation above we have: 1) X is equivariantly isomorphic to S( A2 (u,vd), D) with D =(cid:26) a α2(cid:27) D1 +(cid:26) b α3(cid:27) D2 +(cid:20)0, 1 α2α3(cid:21) E, where E is the exceptional divisor of the blow-up π : A2 (u,vd) → A2 = Spec(C[u, v]), D1 and D2 are the strict transforms of the curves L1 = {u = 0} and L2 = {u + p(v) = 0} in A2 respectively, and (a, b) ∈ Z2 are chosen so that adα3 + bα2 = 1. 2) X is smooth if and only if L1 + L2 is a SNC divisor in A2. 3) Under these conditions, X is Gm-linearly rational at (0, 0, 0, 0). Proof. 1) The A-H presentation is obtained from the following exact sequence: 0 / Z s F / Z4 P Z3 / 0 , / / / / / EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 8 by the method described in Remark 12, where F = t(α2α3, −α2α3, dα3, α2), P =  s = (0, 0, a, b) chosen such that adα3 + bα2 = 1. 1 1 0 0 d α2 0 0 1 0 0 α3   and The corresponding toric variety Z is the blow-up of A3 = Spec(C[u, v, w]) along the sub-scheme with defining by the ideal I = (u, vd, vd−1w, . . . , vwd−1, wd), and Y corresponds to the strict transform by π : A3 I → A3 ≃ A4//Gm of the surface {u + w + p(v) = 0} ≃ Spec(C[u, v]), that is, Y ≃ A2 (u,vd) (see [23, section 3.1]). 1 of the toric divisor given by the ray v3 and D2 corresponds to the restriction to Y of the toric divisor given by the ray α2α3i E, where D1 corresponds to the restriction to Y The ps-divisor D is of the formn a v4, that is, the strict transforms of the curves(cid:8)u = ydzα2 = 0(cid:9) and {w = ytα3 = −u − p(v) = 0} in A2 respectively. The divisor E corresponds to the divisor given by v2, that is, the exceptional divisor of π : A2 (u,vd) → A2. α2o D1 +n b α3o D2 +h0, 2) As p(0) = 0, the equation of X takes the form : yd−1zα2 + tα3 + x (xy + αi) = 0, k Yi=1 and using the Jacobian criterion, we conclude that X is smooth if and only if αi 6= αj for i 6= j. (u,v) ֒→ P2 (u,v) and let A2 [u:v:w] be the embedding of A2 as the complement of the line at 3) Let D = L1 + L2 ⊂ A2 the infinity L∞ = {w = 0}. We denote by ¯D = ¯L1 + ¯L2 the closure of D in P2 (u:v:w) (see Figure 3.1). The only singularity is at the intersection of ¯L2 and L∞. After a sequence of elementary birational transformations we reach the k-th Hirzebruch surface Fk = P(OP1 ⊕ OP1(k)). The proper transform of ¯L2 is a smooth curve intersecting the section of negative self-intersection transversally (see Figure 3.2). The second step is the blow-up of all the intersection points between ¯L1 and ¯L2 except the point corresponding to the origin in A2, followed by the contraction of the proper transform of the fiber passing through each points of the blowup (see Figure 3.3). The final configuration is then the Hirzebruch surface F1 in which the proper transforms of ¯L1 and ¯L2 have self-intersection 1 and intersect each other in a unique point. Then P2 and the desired divisor are obtained from F1 by contracting the negative section (see Figure 3.4). Figure 3.1. Embedding in P2 of the divisor in P2. Figure 3.2. First sequence of blow-ups and contractions, to ob- tain a smooth normal crossing divi- sor divisor in Fk. EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 9 Figure 3.3. Intermediate step, resolution of the crossings, to ob- tain a divisor in Fk−2. Figure 3.4. Final resolution, to obtain a divisor in P2 This resolution gives a birational map from the A-H quotient of X to the A-H quotient of A3 which induces an isomorphism in a neighborhood of the origin of A2. By Theorem 16 this gives a Gm-equivariant isomorphism between an open neighborhood of the origin in X and on open neighborhood of the origin in A3. Let p(v) = v(1 + g(v)) be the polynomial which appears in the statement, and let φ be the birational map defined by: Its inverse is defined by φ : (u, v) → (−u′(g(v′ + u′) + 1), v′ + u′). φ−1 : (u′, v′) → (− u 1 + g(v) , v + u 1 + g(v) ). Then φ(u + p(v)) = v′(g(v′ + u′) + 1) and we obtain : Y (An) i Y ′ = A2 (u,vd) \ V (1 + g(v)) ≃ A2 (u′,v′d) \ V (g(v′ + u′) + 1) i′ Y (X) (u,vd), then S(Y ′, i∗(D)) = U is an equivariant open neighborhood of the fixed point in X, which is and i : Y ′ ֒→ A2 moreover equivariantly isomorphic to an open subset of A3 = Spec(C[Y, Z, T ]) endowed of the hyperbolic Gm-action. The action on A3 is defined by λ · (Y, Z, T ) = (λ−α2α3 Y, λdα3Z, λα2 T ) using Proposition 17. (cid:3) Remark 20. In the particular case where L1 + L2 is not a smooth normal crossing divisor in A2, that is the point 2 of the Theorem 19 is not satisfied, but the crossing of L1 and L2 at the origin is transversal, then S(Y, D) is equivariantly isomorphic to a normal but not smooth Gm-variety V with a unique fixed point contained in its regular locus. The same process as before can be applied, and the variety V admits an open Gm-stable neighborhood of the fixed point isomorphic to a Gm-stable neighborhood of the fixed point of A3 endowed with a linear hyperbolic Gm-action. In other words V is Gm-linearly rational, but not uniformly rational, since it is singular. 3.2.2. Applications. Specifying the coefficients of the polynomial p ∈ C[v] defined in the previous sub-section, we list below particular hypersurfaces of A4 which are Gm-uniformly rational. Proposition 21. The following hypersurfaces in A4 = Spec(C[x, y, z, t]) are Gm-linearly rational: X1 = {x + xkyk−1 + zα2 + tα3 = 0}, X2 = {x + yd−1(xd + zα2 ) + tα3 = 0}, considering the equivariant isomorphisms ψ1 and ψ2, respectively, in the proof.  / / ? _ o o EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 10 Proof. Applying Theorem 19, X1 corresponds to the choice d = 1 and p(v) = v + vk, and X2 corresponds to the choice d ≥ 2 and p(v) = v + vd. 1) An explicit isomorphism ψ1 : X1 \ V (1 + (xy)d−1) → A3 \ V (1 + (Y Z α2 + Y T α3)d−1) is given by: Its inverse ψ−1 1 is given by: ψ−1 1 2) An explicit isomorphism ψ2 : X2 \ V (1 + (xy)d−1) → A3 \ V (1 + (Y dZ α2 + Y T α3)d−1) is given by: →  Y Z T   =  −y 1+(xy)d−1 z t   .   =  − Zα2 +T α3 1+(Y Zα2 +Y T α3 )d−1 −Y (1 + (Y Z α2 + Y T α3)d−1) Z T     →  Y Z T   =  −y 1+(xy)d−1 z t   . T α3 :    x y z t Y Z x y z t ψ1 :   → T   ψ2 :   → T   x y z t Y Z Its inverse ψ−1 2 is given by: ψ−1 2 :  x y z t   =  −Y d−1Z α2 − 1+(Y dZα2 +Y T α3 )d−1 −Y (1 + (Y Z dα2 + Y T α3)d−1) Z T   (cid:3) Theorem 22. All Koras-Russell threefolds of the first kind {x + xky + zα2 + tα3 = 0} in A4 = Spec(C[x, y, z, t]) are Gm-linearly uniformly rational, considering the equivariant isomorphism ψ in the proof. Proof. Let X = {x + xky + zα2 + tα3 = 0} be a Koras-Russell threefold of the first kind, let U be the principal open subset of X where x does not vanish and let V be the principal open subset of X where 1 + yxd−1 does not vanish. The principal open subsets U = Xx and V = X1+yxd−1 form a covering of X by Gm-stable open subsets. Since Γ (U, OU ) = C[x, x−1, y, z, t]/(x + xky + zα2 + tα3) ≃ C[x, x−1, z, t], X is Gm-linearly rational at every point of U. By Proposition 21, we have an explicit Gm-equivariant isomorphism between an open neighborhood of the fixed point in X1 = {x + xkyk−1 + zα2 + tα3 = 0} and an open subset of A3. Moreover X1 admits an action of the cyclic group µk−1 given by ǫ · (x, y, z, t) → (x, ǫy, z, t) such that the action of µk−1 factor through that of Gm. Thus the quotient for the action of the cyclic group and the isomorphism obtained in Proposition 21 commute. In this case the quotient of A3 for the action of µk−1 is still isomorphic to A3. Since X1//µk−1 ≃ X, the Gm-equivariant map ψ1 given in Proposition 21 descends to a Gm-equivariant isomorphism ψ: X1 \ V (1 + (xy)k−1) //µk−1 X \ V (1 + yxk−1) ψ ψ A3 \ V (1 + (Y Z α2 + Y T α3)k−1) . //µd−1 / A3 \ V (1 + Y (Z α2 + T α3)). Remark 23. The variety X is endowed with an hyperbolic Gm-action, the Gm-stable principal open subset V = X1+yxd−1 is isomorphic to a principal open subset of A3 endowed with an hyperbolic Gm-action but the Gm-stable principal open subset U = Xx is isomorphic to a principal open subset of A3 endowed with a Gm-action with positive weights only. (cid:3) / /     / EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 11 Proposition 24. The Koras-Russell threefolds of the second kind given by the equations X = {x + y(xd + zα2 )l + tα3 = 0}, in A4 = Spec(C[x, y, z, t]) with l = 1 or l = 2 or d = 2 are Gm-linearly uniformly rational. Proof. In the case l = 1 we consider the Gm-uniformly rational variety: X2 = {x + yd−1(xd + zα2 ) + tα3 = 0}, given in Proposition 21. The cyclic group µd−1 on X2 via ǫ · (x, y, z, t) → (x, ǫy, z, t), and this action factors through that of Gm. Thus the quotient for the action of cyclic group and the isomorphism obtains in Proposition 21 commute. The conclusion follows by the same method as in the proof of Theorem 22. Let Xd−1 = {x + ydl−1(xd + zα2 )l + tα3 = 0} → X be the cyclic cover of order dl − 1 of X branched along the divisor {y = 0}. The A-H presentation of Xd−1(see [23]) is S( A2 (u,vd), D) with: where E is the exceptional divisor of the blow-up π : A2 (u,vd) → A2 ≃ Spec(C[u, v]) ≃ Spec(C[ydzα2, yx]) , and D =(cid:26) a α2(cid:27) Dα3 +(cid:26) b α3(cid:27) Dα2 +(cid:20)0, 1 α2α3(cid:21) E, where Dα2 and Dα3 are the strict transforms of the curves L1 = (cid:8)v + (u + vd)l) = 0(cid:9) and L2 = {u = 0} in A2 = Spec(C[u, v]) respectively, (a, b) ∈ Z2, being chosen so that adα3 + bα2 = 1. First of all, variables l and d can be exchanged, just considering the automorphism of A2 = Spec(C[u, v]) which send u on u − (v − ul)d and v on v − ul. Then v + (u + vd)l) is sent on v. From now it will be assume that l = 2. By showing that Xd−1 is Gm-linearly rational, one can explicit a birational map between X and A3. This map will be an equivariant isomorphism between an open subset of X containing the fixed point and an open subset of A3. The divisor D = L1 + L2 is birationally equivalent to D′ = {uv = 0} via be the birational endomorphism ϕ of A2 = Spec(C[u, v]) defined by u → u(1+(v−u2)2d−1) 1−u(v−u2)d−1 and v → v − u2. Thus Xd−1 is Gm-linearly rational. Moreover the application ϕ is µ2d−1-equivariant, considering the action of µ2d−1 given by ǫ · (u, v) → (ǫdu, ǫv). The desired result is now obtained by the same technique as in Theorem 22. (cid:3) 4. Examples of non Gm-rational varieties Since the property to be G-uniformly rational is more restrictive than being only uniformly rational, it is not surprising that there are smooth and rational G-varieties which are not G-uniformly rational. In this section we will exhibit some Gm-varieties which are smooth and rational but not not Gm-uniformly rational. However it is not known if these varieties are uniformly rational. This provides candidates to show that the uniform rationality conjecture has a negative answer. Proposition 25. Let C ⊂ A2 be a smooth affine curve of positive genus passing through the origin with multiplicity one and let X be a Gm-variety equivariantly isomorphic to S( A2 p ]E, where E is the exceptional divisor of the blow-up and D is the strict transform of C. Then X is a smooth rational Gm-variety but not a Gm-uniformly rational variety. (u,v), D) with D =n 1 po D + [0, 1 Proof. This is a direct consequence of the classification of hyperbolic Gm-actions on A3 given in Proposition 17. In this case the irreducible components of the support of the ps-divisors are all rational. But the variety S( A2 (u,v), D) given in [23, proposition 3.1] admits the support of D in the support of its ps-divisors. As the support of D is not rational it follows that the varieties obtained by this construction are not Gm-linearly rational and thus not Gm-uniformly rational since the two properties are equivalent in the case of Gm-varieties of complexity two (see Theorem 5). (cid:3) Example 26. Let V (h) be a smooth affine curve of positive genus passing through the origin with multiplicity one. Then the hypersurface {h(xy, zy)/y + tp = 0}, is stable in A4 = Spec(C[x, y, z, t]), for the linear Gm-action given by λ · (x, y, z, t) = (λpx, λ−py, λpz, λt). This variety is smooth using the Jacobian criterion and rational as its algebraic quotient is rational but not Gm-uniformly rational. 4.1. Numerical obstruction for rectifiability of curves. For a Gm-variety S(Y, D), the non-rationality of the irreducible components of the support of D (see Proposition 25) is not the only obstruction to being Gm-rational. There exist divisors D = L1 + L2 where Li is isomorphic to A1 for i = 1, 2 and such that D is not birationally equivalent to D′ = {uv = 0}. Such D can be used to construct Gm-variety S(Y, D) where the irreducible components of the support of the ps-divisors are all rational and such that S(Y, D) is not Gm-rational. To prove the existence of EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 12 such D, we will use an invariant, the Kumar-Murthy dimension (see [22]). Recall that a pair (X, D) is said smooth if X is a smooth projective surface and D is a SNC divisor on X. For every divisor D on a smooth projective variety, we define the Iitaka dimension, κ(X, D) := sup dimφnD(X) in the case where nD 6= ∅ for some n, and κ(X, D) := −∞ otherwise, where φnD : X 99K PN is the rational map associated to the linear system nD on X. Lemma 27. Let D0 =Pk i=1 Di be a reduced divisor on a complete surface X0, with Di irreducible for each i ≥ 0. For any birational morphism π : X → X0 such that the pair (X, DX ) is smooth, with DX the strict transform of D, the value κ(X, 2KX + DX ) does not depend on the choice of π. Proof. By the Zariski strong factorization Theorem, it suffices to show that this dimension is invariant under blow- ups. Let (X, DX ) be a resolution of the pair (X0, D0) such that X is smooth and DX is SNC. Let π : X → X be the blow-up of a point p in X. Since DX is SNC, there are three possible cases: p /∈ DX , p is contained in a unique irreducible component of DX , or p is a point of intersection of two irreducible components DX . We have then for any integer n: n(2K X + D X ) = π∗(n(2KX + DX)) + n(2 − m)E, m = 2, 1, 0 respectively. Therefore, Γ(X, O(n(2K X + D X ))) = Γ(X, O(π∗(n(2KX + DX ) + (2 − m)E))) = Γ(X, O(π∗(n(2KX + DX )))), and so, by the projection formula ([15, II.5]), Γ(X, O(π∗(n(2KX + DX )))) ≃ Γ(X, O(n(2KX + DX))) for any integer n. (cid:3) Definition 28. The Kumar-Murthy dimension kM (X0, D0) of (X0, D0) is the Iitaka dimension κ(X, 2KX + DX ) where π : X → X0 is any birational morphism such that the pair (X, DX) is smooth. Definition 29. A pair (Y, D) (as in definition 14) is birationally rectifiable if it is birationally equivalent to the union of k ≤ N = dim(Y ) general hyperplanes in PN . Note in particular that Y is rational and that the irreducible components of D are either rational or uniruled. Since, the Kumar-Murthy dimension of the pair (P2, D), where D is a union of two distinct lines, is equal to −∞, we obtain: Proposition 30. If a reduced divisor D = D1 + D2 in P2 is birationally rectifiable then kM (P2, D) = −∞. Example 31. Let C = {u + (v + u2)2 = 0} and C′ = {αv(v − β) + u = 0} be two curves in A2 = Spec(C[u, v]) where (α, β) ∈ C2 are generic parameters chosen such that C and C′ intersect normally. Let ¯C and ¯C′ be the closures in P2 of C and C′ respectively and let D = ¯C + ¯C′. Then: i) C and C′ are isomorphic to A1 ii) kM (P2, D) 6= −∞. Proof. The curve C′ is clearly isomorphic to A1. In the case of C, consider the following two automorphisms: ψ1 : (u, v) → (u, v + u2) and ψ2 : (u, v) → (u + v2, v) then the composition ψ2 ◦ ψ1 :(u → u + (u + v2)2 v → v + u2 C on a coordinate axe. A minimal log-resolution π : S7 → P2 of ¯C ∪ ¯C′ is obtained by performing a sequence of seven blow-ups, five of them with centers lying over the singular point of ¯C and the remaining two over the singular point of ¯C′. sends Figure 4.1. Resolution of (P2, ( ¯C + ¯C′), the divisors Ei and E′ blowing-up ¯C ∩ L∞ and ¯C′ ∩ L∞ numbered according to the order of their extraction. i are exceptional divisors obtained EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 13 Using the ramification formula for the successive blow-ups occurring in π, we find that the canonical divisor of S7 is equal to KS7 = −3l + E1 + 2E2 + 3E3 + 6E4 + 10E5 + E′ 2, where l denotes the proper transform of a general line in P2. The total transform of the divisor ¯C + ¯C′ is equal to π∗( ¯C + ¯C′) = ¯C + 2E1 + 4E2 + 6E3 + 11E4 + 18E5 + ¯C′ + E′ 2, where we have identified ¯C and ¯C′ with their proper transforms in S7. Since ¯C is of degree 4 and ¯C′ is of degree 2, the proper transform of ¯C + ¯C′ in S7 is linearly equivalent to 6l and 1 + 2E′ 1 + 2E′ we obtain 2KS7 + D = 2KS7 + π∗( ¯C + ¯C′) − (2E1 + 4E2 + 6E3 + 11E4 + 18E5 + E′ 1 + 2E′ 2, 1 + 2E′ 2) = E4 + 2E5 + E′ which is an effective divisor. Thus kM (P2, D) 6= −∞, and by Proposition 30, D is not birationally rectifiable. (cid:3) 4.2. Application. Let X be the subvariety of A5 = Spec(C[w, x, y, z, t]) defined by the two equations {w + y(x + yw2)2 + tα3 = 0} and {αx(yx − β) + w + zα2) = 0}, where (α, β) ∈ C2 are the same parameters as in Example 31. This variety is endowed with a hyperbolic Gm-action induced by the linear one on A5, λ · (w, x, y, z, t) = (λα2α3w, λα2α3 x, λ−α2α3 y, λα3z, λα2 t). Moreover is equivariantly isomorphic to the hypersurface in A4 = Spec(C[x, y, z, t]) defined by {zα2 − αx(xy − β) + y(x + y(zα2 − αx(xy − β))2)2 + tα3 = 0}. Theorem 32. The threefold X is a smooth rational Gm-variety but not a Gm-uniformly rational variety. Proof. The A-H presentation of X is given by S( A2 (u,v), D) with D =(cid:26) a α2(cid:27) D1 +(cid:26) b α3(cid:27) D2 +(cid:20)0, 1 α2α3(cid:21) E, where E is the exceptional divisor of the blow-up π : A2 (u,v) → A2, D1 and D2 are the strict transform of the curves C and C′ of Example 31, and (a, b) ∈ Z2 are such that aα3 + bα2 = 1. The presentation comes from the fact that X is endowed with an action of µα2 × µα3 factoring through that of Gm and given by (ǫ, ξ) · (x, y, z, t) → (x, y, ǫz, ξt). {✇✇✇✇✇✇✇✇✇ X//µα2 X #●●●●●●●●● X//µα3 By [23, Example 3.1], X//µα2 is equivariantly isomorphic to S( A2 equivariantly isomorphic to S( A2 A3 = Spec(C[x, y, t]) with the Gm-action defined via λ · (x, y, t) = (λα3 x, λ−α3 y, λt) and X//µα3 is equivariantly isomorphic to A3 = Spec(C[x, y, z]) with the Gm-action defined via λ · (x, y, z) = (λα2 x, λ−α2 y, λz). In particular X is a Koras-Russell threefolds (see [16, 17, 23]). Now the result follows from Proposition 30 and Example 31. α3i E) and X//µα3 is α2i E). In fact, X//µα2 is equivariantly isomorphic to α3o D2 +h0, 1 α2o D1 +h0, 1 (u,v),n 1 (u,v),n 1 (cid:3) 5. Weak equivariant rationality The property to be G-linearly uniformly rational is very restrictive. We will now introduce a weaker notion: Definition 33. A G-variety X is called weakly G-rational at a point x if there exist an open G-stable neighborhood Ux of x, an open subvariety V of An equipped with a G-action and a G-equivariant isomorphism between Ux and V . We said that X is weakly G-uniformly rational if it is weakly G-rational at every point. Note that, in contrast with Definition 4 ii) we only require that V ⊂ An is endowed with a G-action, in particular it need not be the restriction of a G-action on An. In summary we have a sequence of implications between these different notions of G-rationality: G-linearly uniformly rational implies G-uniformly rational which implies G-weakly uniformly rational, which finally implies uniformly rational. Theorem 34. Let S ⊂ A3 = Spec(C[x, y, z]) be the surface defined by the equation z2 + y2 + x3 − 1 = 0, equipped with the µ2-action τ · (x, y, z) → (x, y, −z) on A3, where τ is the non trivial element of µ2. Then S is weakly µ2-uniformly rational but not µ2-uniformly rational. # { EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 14 Proof. The surface S is the cyclic cover of A2 of order 2 branched along the smooth affine elliptic curve C = {y2 +x3 −1 = 0} ⊂ A2. By construction, the inverse image of C in S is equal to the fixed points set of the involution. It follows that S is not µ2 rational at the point p = (1, 0, 0). Indeed, every µ2-action on A2 being linearisable (see [19, Theorem 4.3]), its fixed points set is rational. Therefore there is no µ2-stable open neighborhood of p which is equivariantly isomorphic to a stable open subset of A2 endowed with a µ2-action. However, there is an open subset U of A2 which can be endowed with a µ2-action such that U is equivariantly isomorphic to a µ2-stable open neighborhood of p. Indeed, letting u = z + y and v = z − y, S is isomorphic to the surface defined in A3 = Spec(C[u, v, x]) by the equation {uv − x3 + 1 = 0}. The open subset V1 = S \ {1 + x + x2 = u = 0} is isomorphic to A2 with coordinates u and v/(1 + x + x2) = (x − 1)/u = w. Let V = S \ {1 + x + x2 = 0} be an open subset in V1, and let x = uw + 1, then V has the following coordinate ring: C(cid:20)u, w, 1 (uw + 1)2 + uw + 1 + 1(cid:21) = C(cid:20)u, w, 1 (uw)2 + 3uw + 3(cid:21) . The open subset V contains the point p and is stable by the action of µ2 defined by: τ · (u, v) = (w((uw)2 + 3uw + 3), u((uw)2 + 3uw + 3)−1). So S is µ2-weakly rational but not µ2-rational. (cid:3) References K. Altmann, J. Hausen. Polyhedral divisors and algebraic torus actions. Math. Ann. 334 (2006), no. 3, 557 -- 607. K. Altmann, J. Hausen, H. Suss. Gluing affine torus actions via divisorial fans. Transform. Groups 13 (2008), no. 2, 215 -- 242. I. Arzhantsev, A. Perepechko, H. Süss. Infinite transitivity on universal torsors. J. Lond. Math. Soc. (2) 89 (2014), no. 3, 762 -- 778. A. Białynicki-Birula. Remarks on the action of an algebraic torus on kn. II. Bull. Acad. Polon. Sci. Sér. Sci. Math. Astronom. Phys. 14 1966 177 -- 181. F. Bogomolov, C. Böhning. On uniformly rational varieties. Topology, geometry, integrable systems, and mathematical physics, 33 -- 48, Amer. Math. Soc. Transl. Ser. 2, 234, Amer. Math. Soc., Providence, RI, 2014. G. Bodnàr, H. Hauser, J. Schicho, O. Villamayor U. Plain varieties. Bull. Lond. Math. Soc. 40 (2008), no. 6, 965 -- 971. D. Cox, J. Little, H. Schenck. Toric varieties. Graduate Studies in Mathematics, 124. American Mathematical Society, Providence, RI, 2011. M. Demazure. Anneaux gradués normaux. Introduction à la théorie des singularités, II, 35 -- 68, Travaux en Cours, 37, Hermann, Paris, 1988. A. Dubouloz. Quelques remarques sur la notion de modification affine. arXiv:math/0503142 . H. Flenner, M. Zaidenberg. Normal affine surfaces with C∗-actions. Osaka J. Math. 40 (2003), no. 4, 981 -- 1009. W. Fulton. Introduction to toric varieties. Annals of Mathematics Studies, 131. The William H. Roever Lectures in Geom- etry. Princeton University Press, Princeton, NJ, 1993. R. Gurjar, M. Koras, P. Russell. Two dimensional quotients of Cn by a reductive group. Electron. Res. Announc. Math. Sci. 15 (2008), 62 -- 64. M. Gromov. Oka's principle for holomorphic sections of elliptic bundles. J. Amer. Math. Soc. 2 (1989), no. 4, 851 -- 897. A. Gutwirth. The action of an algebraic torus on the affine plane. Trans. Amer. Math. Soc. 105 1962 407 -- 414. R. Hartshorne. Algebraic geometry. Graduate Texts in Mathematics, No. 52. Springer-Verlag, New York-Heidelberg, 1977. M. Koras, P. Russell. Contractible threefolds and C∗-actions on C3. J. Algebraic Geom. 6 (1997), no. 4, 671 -- 695. S. Kaliman, M. Koras, L. Makar-Limanov, P. Russell, C∗-actions on C3 are linearisable. Electron. Res. Announc. Amer. Math. Soc. 3 (1997), 63 -- 71. S. Kaliman, M. Zaidenberg. Affine modifications and affine hypersurfaces with a very transitive automorphism group. Transform. Groups 4 (1999), no. 1, 53 -- 95. T. Kambayashi, Automorphism group of a polynomial ring and algebraic group action on an affine space. J. Algebra 60 (1979), no. 2, 439 -- 451. T. Kambayashi, P. Russell. On linearizing algebraic torus actions. J. Pure Appl. Algebra 23 (1982), no. 3, 243250. G. Kempf, F. Knudsen, D. Mumford, B. Saint-Donat . Toroidal embeddings. I. Lecture Notes in Mathematics, Vol. 339. Springer-Verlag, Berlin-New York, 1973. M. Kumar, P. Murthy. Curves with negative self-intersection on rational surfaces. J. Math. Kyoto Univ. 22 (1982/83), no. 4, 767 -- 777. C. Petitjean, Cyclic cover of affine article T-varieties, J. Pure Appl. Algebra 219 (2015), no. 9, 4265 -- 4277. C. Petitjean, Actions hyperboliques du groupe multiplicatif sur des variétés affines : espaces exotiques et structures locales, Thesis (2015). H. Sumihiro. Equivariant completion. J. Math. Kyoto Univ. 14 (1974), 1 -- 28. M. Thaddeus. Geometric invariant theory and flips. J. Amer. Math. Soc. 9 (1996), no. 3, 691 -- 723. M. Zaidenberg, Lectures on exotic algebraic structures on affine spaces. arXiv:math/9801075 [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] EQUIVARIANTLY UNIFORMLY RATIONAL VARIETIES 15 Charlie Petitjean, Institut de Mathématiques de Bourgogne, Université de Bourgogne, 9 Avenue Alain Savary, BP 47870, 21078 Dijon Cedex, France E-mail address: [email protected]
1305.7175
4
1305
2019-06-05T00:53:40
Nonlinear Traces
[ "math.AG", "math.CT", "math.QA", "math.RT" ]
We combine the theory of traces in homotopical algebra with sheaf theory in derived algebraic geometry to deduce general fixed point and character formulas. The formalism of dimension (or Hochschild homology) of a dualizable object in the context of higher algebra provides a unifying framework for classical notions such as Euler characteristics, Chern characters, and characters of group representations. Moreover, the simple functoriality properties of dimensions clarify celebrated identities and extend them to new contexts. We observe that it is advantageous to calculate dimensions, traces and their functoriality directly in the nonlinear geometric setting of correspondence categories, where they are directly identified with (derived versions of) loop spaces, fixed point loci and loop maps, respectively. This results in universal nonlinear versions of Grothendieck-Riemann-Roch theorems, Atiyah-Bott-Lefschetz trace formulas, and Frobenius-Weyl character formulas. On the one hand, we can then linearize by applying sheaf theories, such as the theories of coherent sheaves and D-modules, developed by Gaitsgory and Rozenblyum, as functors out of correspondence categories (in the spirit of topological field theory). This recovers the familiar classical identities, in families and without any smoothness or transversality assumptions. On the other hand, the formalism also applies to higher categorical settings not captured within a linear framework, such as characters of group actions on categories.
math.AG
math
NONLINEAR TRACES DAVID BEN-ZVI AND DAVID NADLER Abstract. We combine the theory of traces in homotopical algebra with sheaf theory in de- rived algebraic geometry to deduce general fixed point and character formulas. The formalism of dimension (or Hochschild homology) of a dualizable object in the context of higher algebra provides a unifying framework for classical notions such as Euler characteristics, Chern char- acters, and characters of group representations. Moreover, the simple functoriality properties of dimensions clarify celebrated identities and extend them to new contexts. We observe that it is advantageous to calculate dimensions, traces and their functoriality directly in the nonlinear geometric setting of correspondence categories, where they are directly identified with (derived versions of) loop spaces, fixed point loci and loop maps, respectively. This results in universal nonlinear versions of Grothendieck-Riemann-Roch theorems, Atiyah-Bott-Lefschetz trace formulas, and Frobenius-Weyl character formulas. We can then linearize by applying sheaf theories, such as the theories of ind-coherent sheaves and D-modules constructed by Gaitsgory-Rozenblyum [GR2]. This recovers the familiar classical identities, in families and without any smoothness or transversality assumptions. On the other hand, the formalism also applies to higher categorical settings not captured within a linear framework, such as characters of group actions on categories. Contents Inspirations and motivations 1. Introduction 1.1. Summary 1.2. 1.3. Acknowledgements 2. Overview 2.1. Traces in category theory 2.2. Traces in geometry 2.3. Trace formulas via sheaf theories 3. Traces in category theory 3.1. Preliminaries 3.2. Dualizability 3.3. Traces and dimensions 3.4. Functoriality of dimension 3.5. Functoriality of traces 4. Traces in Geometry 4.1. Categories of correspondences 4.2. Traces of correspondences 4.3. Geometric functoriality of dimension 4.4. Geometric functoriality of trace 5. Traces for sheaves 5.1. Dimensions and traces of sheaf categories: 1-categorical consequences 5.2. 5.3. Integration formulas for traces: 2-categorical consequences Ind-coherent sheaves and D-modules 1 2 2 6 6 7 7 9 12 16 16 17 19 21 23 24 24 25 27 29 29 30 31 33 2 DAVID BEN-ZVI AND DAVID NADLER References 36 1. Introduction This paper is devoted to traces and characters in homotopical algebra and their application to algebraic geometry and representation theory. We observe that many geometric fixed point and trace formulas can be expressed as linearizations of fundamental nonlinear identities, describing dimensions and traces directly in the setting of correspondence categories of varieties or stacks. This gives a simple uniform perspective on (and useful generalizations of) geometric character and fixed point formulas of Grothendieck-Riemann-Roch and Atiyah-Bott-Lefschetz type. In addition, one can also specialize the universal geometric formulas to higher categorical settings not captured within a linear framework, such as characters of group actions on categories. The paper is organized as follows: after a brief summary in Section 1.1, we give a detailed overview in Section 2, in three sections: first, the abstract functoriality of traces in higher category theory; second, their calculation in correspondence categories in derived algebraic geometry; and third, their specialization via sheaf theories. The rest of the paper follows the same structure with more details provided. We emphasize the formal nature and appealing simplicity of the constructions in any sufficiently derived setting. For example, in the second part, we work within derived algebraic geometry, but the statements and proofs should hold in any setting (for example, derived manifolds) with a suitable notion of fiber product to handle non-transversal intersections. The main objects appearing in trace formulas are the derived loop space (the self-intersection of the diagonal in its role as the nonlinear trace of the identity map) and more general derived fixed point loci. The importance of a derived setting also appears prominently in the third part, where the sheaf theories we apply must have good functorial properties with respect to fiber products. As a result, the theory of characters in Hochschild and cyclic homology is expressed directly by the geometry, resulting in simpler formulations. For example, the Todd genus in Grothendieck-Riemann-Roch and the denominators in the classical Atiyah-Bott formula arise naturally from derived calculations. 1.1. Summary. We now describe the main theorem, extending classical trace and dimension formulas to a very general setting in derived algebraic geometry (including equivariance for arbitrary Lie algebroids or affine algebraic groups) without any smoothness or transversality assumptions, while emphasizing that the main contribution of the paper is the simple geometric formalism underlying these formulas. For our general and formal nonlinear results, we need not assume anything about what classes of derived stacks and morphisms we work with. For applications, we need to be in a setting in which the powerful mechanism of sheaf theory is fully developed [G1, DG, GR2]. Setting 1.1. Throughout this paper, we work over a field k of characteristic zero. A stack X connotes either (1) a QCA derived stack in the sense of [DG]. In other words X is quasicompact with affine diagonal and the underived inertia of X is finite presentation over the underlying underived stack Xcl. (2) an ind-inf-scheme, in the sense of [GR2]. These include (derived) schemes of finite (or locally almost of finite) type and ind-schemes built out of unions of the former along closed embeddings, as well as their quotients by arbitrary Lie algebroids, or equivalently formal groupoids. NONLINEAR TRACES 3 By a proper map we indicate a proper and schematic map, while for ind-proper indicates ind- proper and ind-inf-schematic. All appearances of proper maps in this paper may be replaced by ind-proper ones. Thus the class of spaces we consider includes all k-derived schemes of finite type and their quotients by either Lie algebroids or finite type affine group schemes. Given a derived stack πX : X → Spec k, we denote by πLX : LX = Map(S1, X) → Spec k its derived loop space. In general, the derived loop space is a derived thickening of the inertia stack. For a map f : X → Y , we will denote by Lf : LX → LY the induced map on loops. Example 1.2. For many applications, the following two special cases are noteworthy. When X is a smooth scheme, LX ≃ TX [−1] is the total space of the shifted tangent space by the HKR theorem. The same holds for an arbitrary scheme, if we replace the tangent space by the tangent complex, see for example [BN10b]. For f : X → Y a map of schemes, Lf : TX [−1] → TY [−1] is (the shift of) the usual tangent map. When Y = BG is a classifying stack, LY ≃ G/G is the adjoint quotient. For X a G-scheme, and f : X/G → BG the corresponding classifying map, Lf : L(X/G) → LBG ≃ G/G is the universal family of derived fixed point loci. More precisely, for any element g ∈ G, the derived fixed point locus X g ⊂ X is precisely the derived fiber X g ≃ L(X/G) ×G/G {g} We will measure stacks X by differential graded (dg) enhancements of derived categories of sheaves. The most familiar is the assignment X 7→ Q(X) of the (unbounded) category of quasicoherent sheaves. However, we will make essential use of Grothendieck-Serre duality, in the guise of an adjunction (f∗, f !) between push-forward and extraordinary pullback for proper maps X. This duality is most naturally expressed in the setting of ind-coherent sheaves X 7→ Q!(X) as developed in [G1, GR2]. Ind-coherent sheaves agree with quasicoherent sheaves for smooth schemes but differ on singular schemes, where [bounded complexes of] coherent sheaves (the compact objects of Q!) differ from perfect complexes (the compact objects of Q). In other words, ind-coherent sheaves are to coherent sheaves and G-theory (the setting of Grothendieck- Riemann-Roch theorems) as quasicoherent sheaves are to perfect complexes and K-theory. Another sheaf theory to which the general formalism developed in [GR2] applies is the theory of D-modules X → D(X), which for smooth schemes agrees with the classical notion of quasicoherent complexes of modules for the sheaf of differential operators DX (i.e. with compact objects given by bounded coherent complexes of DX -modules). In general [GR1, GR2] D(X) is defined as the category of crystals, i.e., as ind-coherent complexes on the de Rham space of X D(X) := Q!(XdR). The compact objects in D(X) for X a scheme are the coherent D-modules, while for X a stack they form a smaller class, the safe D-modules of [DG]. The book [GR2] develops the theories Q! and D in particular as functors out of 2-categories of correspondences of schemes and stacks, with 1-morphisms from X to Y given by correspon- dences representable over Y and 2-morphisms given by ind-proper ind-schematic morphisms of correspondences. This theory encodes a huge amount of structure, including in particular pull- back and pushforward functors f ! and f∗ satisfying base change, as well as the (f∗, f !) adjunction for proper (or even ind-proper) maps. They also establish symmetric monoidal properties of the sheaf theory.1 (See Sections 2.3 and 5 for more background and precise statements.) Let S = Q! or S = D denote either of these sheaf theories. 1We will also need from [DG] the construction of continuous pushforward functors for all maps of QCA stacks (such as the non-representable projection πX : X → pt from a stack) and their base-change property. 4 DAVID BEN-ZVI AND DAVID NADLER We let ωX = π! X OSpec k ∈ S(X) denote the appropriate dualizing sheaf. Thus for ind- coherent sheaves, ωX ∈ Q!(X) is the algebraic dualizing sheaf, and for D-modules, ωX ∈ D(X) is the Verdier dualizing sheaf. Let ω(X) = πX∗ωX denote the corresponding complex of global volume forms: for ind-coherent sheaves, ω(X) ∈ k-mod consists of algebraic volume forms, and for D-modules, ω(X) ∈ k-mod consists of locally constant distributions (Borel-Moore chains) for X a scheme. For X a stack we use the continuous "renormalized" pushforward functor on D-modules of [DG], which roughly replaces equivariant cohomology (derived invariants) by a shift of equi- variant homology (derived coinvariants, see [DG, Example 9.1.6]), so that ω(Y /G) for a G- variety Y is given by a shift of G-coinvariants on Borel-Moore chains on Y . (Note that even for X = BG this differs from the standard definition of equivariant Borel-Moore homology, which is identified in this case with equivariant cohomology.) For a proper (or ind-proper) map f : X → Y , adjunction provides an integration map Rf : ω(X) → ω(Y ). Example 1.3. Let us continue with the special cases of Example 1.2, and focus in particular on algebraic distributions ωLX ∈ Q!(LX) on the loop space. When X is a smooth scheme, LX ≃ TX [−1] is naturally Calabi-Yau, and its global volume forms are identified with differential forms ω(TX [−1]) ≃ O(TX [−1]) ≃ Sym•(ΩX [1]). The canonical "volume form" on LX is given by the Todd genus (as explained by Markarian [Ma]): the resulting integration of functions on LX differs from the integration of differential forms on X by the Todd genus. When Y = BG is a classifying stack, LY ≃ G/G is naturally Calabi-Yau, and its global vol- ume forms are invariant functions ω(G/G) ≃ O(G/G) ≃ O(G)G. If G is reductive with Cartan subgroup T ⊂ G and Weyl group W , the naive invariants O(G)G ≃ O(T )W are equivalent to the derived invariants, but in general there may be higher cohomology. Theorem 1.4. Let S = Q! or S = D denote either the theory of ind-coherent sheaves or D-modules. Recall our conventions for stacks and morphisms, Setting 1.1. • For a stack X, there is a canonical identification HH∗(S(X)) ≃ ω(LX) of the Hochschild homology of sheaves on X with distributions (or renormalized Borel-Moore chains) on the loop space. • For a proper (or ind-proper) map of stacks f : X → Y the induced map HH∗(S(X)) → HH∗(S(Y )) is given by integration along the loop map Lf : LX → LY . • Grothendieck-Riemann-Roch: In particular, for any compact object M ∈ S(X) (co- herent sheaf or safe coherent D-module) with character [M ] ∈ HH∗(S(X)) ≃ ω(LX), there is a canonical identification [f∗M] ≃ZLf [M ] ∈ HH∗(S(Y )) ≃ ω(LY ) In other words, the character of a pushforward along a proper map is the integral of the character along the induced loop map. • Atiyah-Bott-Lefschetz: Let G be an affine group, and X a proper stack with G-action, or equivalently, a proper map f : X/G → BG. Then for any compact object M ∈ S(X/G) (G- equivariant coherent sheaf or safely equivariant coherent D-module on X), and element g ∈ G, there is a canonical identification [f∗M ]g ≃ZLf [M ]X g NONLINEAR TRACES 5 In other words, under the identification of invariant functions and volume forms on the group, the value of the character of an induced representation at a group element is given by the integral of the original character along the corresponding fixed point locus of the group element. • Extension to traces: The trace T r(S(Z)) of the endofunctor of S(X) given by a self- correspondence (e.g. a self-map) X ← Z → X is given by distributions on the fixed points ω(Z∆). • For a map f : (X, Z) → (Y, W ) of stacks with self-correspondences2, the induced map T r(S(Z)) → T r(S(W )) is given by integration along fixed points Z∆X → W∆Y . Example 1.5 (Frobenius-Weyl Character Formula). Here is a reminder of a well-known application of the Atiyah-Bott-Lefschetz formula in representation theory. • If G is a finite group, and X = G/K is a homogeneous set, and M = k[G/K] the ring of functions, one recovers the Frobenius character formula for the induced representation k[G/K]. • If G is a reductive group, X = G/B is the flag variety, X/G = pt/B → pt/G. The loop map L(X/G) = B/B ≃ eG/G → G/G is the (group) Grothendieck-Springer simultaneous resolution, with fibers giving fixed point loci on the flag variety. For M = L an equivariant line bundle on G/B, and g ∈ G runs over a maximal torus, one recovers the Weyl character formula for the induced representation H ∗(G/B, L). Remark 1.6. The reader will note no explicit appearance of the Todd genus in the above formulas. In other words as for K-theory, pushforward of sheaves naturally agrees with the pushforward in Hochschild homology. The Todd genus arises in comparing these natural push- forwards with the pushforward in cohomology, i.e., integration of forms. It arises when one unwinds the integration map RLf : ω(LX) → ω(LY ), given by Grothendieck duality, in terms of functions (or differential forms) using the Hochschild-Kostant-Rosenberg theorem. In par- ticular, the familiar denominators in the Atiyah-Bott formula are implicit in the integration measure on the fixed point locus. For instance, as mentioned above, when X is a smooth scheme, a geometric version of the HKR theorem asserts that the loop space is the total space of the shifted tangent complex LX ≃ TX [−1], and global volume forms are canonically functions ω(LX) ≃ O(TX [−1]) ≃ Sym•(ΩX [1]). Under this identification (as explained by Markarian [Ma]), the resulting inte- gration of functions on LX differs from the integration of differential forms on X by the Todd genus. Remark 1.7. The paper [KP] carries out the program described in this paper (i.e. recovering classical identities from non-linear ones) by calculating explicitly the derived contributions in the case of the Atiyah-Bott formula. Remark 1.8. The main contribution of this paper is hidden in the statement of this theorem: we establish nonlinear versions of character formulas in the setting of derived stacks, and deduce classical formulas and new higher categorical analogues formally by applying suitable sheaf theories. Thanks to the great generality of sheaf theory in derived algebraic geometry [GR2], the resulting applications hold with remarkably few assumptions. We are particularly interested in the higher categorical variants where one considers sheaves of categories, in particular Frobenius-Weyl character formulas for group actions on categories. Since the requisite foundations are not yet fully developed, we postpone details of this to future 2i.e., we lift f to an identification Z ≃ X ×Y W of correspondences from X to Y 6 DAVID BEN-ZVI AND DAVID NADLER works. Applications include an identification of the character of the category of D-modules on the flag variety with the Grothendieck-Springer sheaf, and of the trace of a Hecke functor on the category of D-modules on the moduli of bundles on a curve with the cohomology of a Hitchin space. 1.2. Inspirations and motivations. This work has many inspirations. It is heavily reliant on the ∞-categorical foundations of higher algebra, derived algebraic geometry and sheaf theory due to Lurie [L2, L4] and Gaitsgory-Rozenblyum [GR2]. It is also inspired by Lurie's cobordism hypothesis with singularities [L3], which provides a powerful unifying tool for higher algebra. Already in the setting of one-dimensional field theory, this result can be viewed as a vast gener- alization of the classical theory Hochschild and cyclic homology and characters therein [Lo], (in particular the natural cyclic symmetry of Hochschild homology is generalized to a circle action on the dimensions of arbitrary dualizable objects). In particular, the formal properties of traces we use are simple instances of the cobordism hypothesis with singularities on marked intervals and cylinders. The work of Toen and Vezzosi [TV] on traces and higher Chern characters of sheaves of categories (and in particular the role of the cobordism hypothesis therein) has also profoundly influenced our thinking. Another important inspiration is the categorical theory of strong duality, dimensions and traces introduced by Dold and Puppe in [DP] (see [M, PS] for more recent developments) with the express purpose of proving Lefschetz-type formulas. In [DP], dualizability of a space is achieved by linearization (passing to suspension spectra), while our approach is to pass to categories of correspondences (or spans) instead. We were also inspired by the preprint [Ma] and the subsequent work [Cal1, Cal2, Ram, Ram2, Shk]. There have been many recent papers [Pe, Lu, Po, CT] building on related ideas to prove Riemann-Roch and Lefschetz-type theorems in the noncommutative context of differential graded categories and Fourier-Mukai transforms; our work instead places these results in the context of the general formalism of traces in ∞- categories, and generalizes them to commutative but nonlinear settings. The Grothendieck-Riemann-Roch type applications in this paper concern the character map taking coherent sheaves to classes in Hochschild homology (or in a more refined version, to cyclic homology). This is significantly coarser than the well established theory of Lefschetz- Riemann-Roch theorems valued in Chow groups (see the seminal [Th], the more recent [Jo] and many references therein). Thus for schemes, the quantities compared are Dolbeault (or de Rham) cohomology classes rather than algebraic cycles or K- (or rather G-)theory classes. Our primary motivation is the development of foundations for "homotopical harmonic anal- ysis" of group actions on categories, aimed at decomposing derived categories of sheaves (rather than classical function spaces) under the actions of natural operators. This undertaking follows the groundbreaking path of Beilinson-Drinfeld within the geometric Langlands program and is consonant with general themes in geometric representation theory. The pursuit of a geometric analogue of the Arthur-Selberg trace formula by Frenkel and Ngo [FN] has also been a source of inspiration and applications. Remark 1.9. A companion paper [BN13] presents an alternative approach to Atiyah-Bott- Lefschetz formulas (and in particular a conjecture of Frenkel-Ngo) as a special case of the "secondary trace formula" identifying trace invariants associated to two commuting endomor- phisms of a sufficiently dualizable object. This is also applied to establish the symmetry of the 2-class functions on a group constructed as the 2-characters of categorical representations. 1.3. Acknowledgements. We would like to thank Dennis Gaitsgory and Jacob Lurie for pro- viding both the foundations and the inspiration for this work, as well as helpful comments and NONLINEAR TRACES 7 specifically D.G. for discussions of [GR2]. We would also like to thank Toly Preygel for many discussions about derived algebraic geometry. We gratefully acknowledge the support of NSF grants DMS-1103525 (D.BZ.) and DMS- 1319287 (D.N.). 2. Overview 2.1. Traces in category theory. We highlight structures arising in the general theory of dualizable objects in symmetric monoidal higher categories (see also [DP, M, PS]). For legibility, we suppress all ∞-categorical notations and complications from the introduction. We rely on [GR2] for the theory of symmetric monoidal (∞, 2)-categories, though only the formal outline of the theory is in fact needed for this paper. See [TV, HSS] for thorough treatments of the theory of traces in higher category theory. The basic notion in the theory is that of dimension of a dualizable object of a symmetric monoidal category A. By definition, for such an object A there exists another A∨ together with a coevaluation map ηA and evaluation map ǫA satisfying standard identities. By definition, the dimension of A is the endomorphism of the the unit 1A given by the composition 1A ηA A ⊗ A∨ dim(A) ǫA / 1A Example 2.1. For V a vector space, V ∨ = Homk(V, k) is the vector space of functionals, ǫV : V ⊗ V ∨ → k is the usual evaluation of functionals, ηV : k → End(V ) ≃ V ⊗ V ∨ is the identity map (which exists only for V finite-dimensional), and dim(V ) can be regarded as an element of the ground field (by evaluating it on the multiplicative unit). Remark 2.2 (Duality and naivet´e in ∞-categories.). It is a useful technical observation that the notion of dualizability in the setting of ∞-categories is a "naive" one: it is a property of an object that can be checked in the underlying homotopy category. As a result, all of the categorical and 2-categorical calculations in this paper are similarly naive and explicit (and analogous to familiar unenriched categorical assertions), involving only small amounts of data that can be checked by hand. The notion of dimension is a special case of the trace of an endomorphism Φ of a dualizable object A. By definition, the trace of Φ is the endomorphism of the unit 1A given by the composition 1A ηA / A ⊗ A∨ Φ⊗idA∨ / A ⊗ A∨ ǫA / / 1A which recovers the dimension for Φ = idA. Tr(Φ) A key feature of dimensions and traces is their cyclicity, which at the coarsest level is ex- pressed by a canonical equivalence m(Φ, Ψ) : Tr(Φ ◦ Ψ) ∼ / / Tr(Ψ ◦ Φ), see Proposition 3.15. At a much deeper level, an important corollary of the cobordism hy- pothesis [L3] is the existence of an S1-action on dim(A) for any dualizable object A (and an analogous structure for general traces, see Remark 3.28). Remark 2.3 (Dimensions and traces are local). It is useful for applications to note that the notion of dualizability and the definition of dimension and are local in the category A. Namely, they only require knowledge of the objects 1A, A, A∨, A ⊗ A∨, the morphisms ηA, ǫA, and / / 7 7 / / 6 6 / 8 DAVID BEN-ZVI AND DAVID NADLER standard tensor product and composition identities among them. Likewise, the notion of trace only requires the additional endomorphism Φ along with a handful of additional identities. 2.1.1. Functoriality of traces. Now suppose the ambient symmetric monoidal category A un- derlies a 2-category, so there is the possibility of noninvertible 2-morphisms. This allows for the notion of left and right adjoints to morphisms. Let us say a morphism A → B is contin- uous, or right dualizable, if it has a right adjoint. (The terminology derives from the setting of presentable categories, where the adjoint functor theorem guarantees the existence of right adjoints for colimit preserving functors.) Here are natural functoriality properties of dimensions and traces. Proposition 2.4. Let A, B denote dualizable objects of A and f∗ : A → B a continuous morphism with right adjoint f !. (1) There is a canonical map on dimensions dim(A) = / / Tr(IdA) / Tr(f !f∗) ∼ / / Tr(f∗f !) / Tr(IdB) = / / dim(B) compatible with compositions of continuous morphisms. (2) Given endomorphisms Φ ∈ End(A), Ψ ∈ End(B), and a commuting structure dim(f∗) α : f∗ ◦ Φ ∼ / / Ψ ◦ f∗ there is a canonical map on traces Tr(f∗, α) : Tr(Φ) / Tr(Ψ) compatible with compositions of continuous morphisms with commuting structures. We refer to the compatibility with compositions stated in the proposition as abstract Grothendieck- Riemann-Roch. To see its import more concretely, let us restrict the generality and focus on an object of A in the sense of a morphism V : 1A → A. Corollary 2.5. Let A, B denote dualizable objects of A and f∗ : A → B a continuous morphism. For V : 1A → A an object of A, we obtain a map on dimensions dim(V ) : 1A ≃ dim(1A) / dim(A) called the character of V and alternatively denoted by [V ]. It satisfies abstract Grothendieck- Riemann-Roch in the sense that the following diagram commutes [V ] 1A dim(A) [f∗V ] dim(f∗) / dim(B) Remark 2.6 (Functoriality of dimensions and traces is local). As in Remark 2.3, it is useful to note that the functoriality of dimension is local, depending only on a handful of objects, mor- phisms and identities, along with the additional adjunction data (f∗, f !). A similar observation applies to the functoriality of traces. Remark 2.7. It follows from the cobordism hypothesis with singularities [L3] (see [TV]) that the morphism dim(f∗) is S1-equivariant, and hence the character [V ] is S1-invariant, though we will not elaborate on this structure here. We refer to [HSS] for a thorough study of the functoriality and cyclicity of traces. 3 3 / / / / / / 6 6 / NONLINEAR TRACES 9 Example 2.8. Let dgCatk denote the symmetric monoidal ∞-category of presentable k-linear differential graded categories (or alternatively, stable presentable k-linear ∞-categories), see e.g. [GR2]. In this setting, any compactly generated category A is dualizable, and its dimension is the Hochschild chain complex dim(A) = HH∗(A). The S1-action on dim(A) corresponds to Connes' cyclic structure on HH∗(A), so that in particular, the localized S1-invariants of dim(A) form the periodic cyclic homology of A. More generally, the trace of an endofunctor Φ : A → A is the Hochschild homology Tr(Φ) = HH∗(A, Φ). For example, if A = R-mod for a dg algebra R, then Φ is represented by an R-bimodule M , and we recover the Hochschild homology HH∗(R, M ). Any compact object M ∈ A defines a continuous functor whose character is a vector 1dgCatk = dgV ectk M / / A dim(M ) ∈ HH∗(A) in Hochschild homology (with refinement in cyclic homology). The abstract Grothendieck- Riemann-Roch theorem expresses the natural functoriality of characters in Hochschild homology (or their refinement in cyclic homology). In fact, the construction of characters factors through the canonical Dennis trace map from the space Acpt of compact objects of A. Acpt / K(A) / HH∗(A) 2.2. Traces in geometry. To apply the preceding formalism to geometry, it is useful to orga- nize spaces and maps within a suitable categorical framework. We then arrive at loop spaces and fixed point loci as nonlinear expressions of dimensions and traces. This simple observa- tion provides the core of the paper. Throughout the discussion, we continue to suppress all ∞-categorical notations and complications. Our reference for the correspondence 2-category of stacks is Section V of [GR2] (see also [H1]). To begin, consider the general setup of the symmetric monoidal category Corr(C) of cor- respondences, or spans, in a category C such as stacks (or formally a symmetric monoidal ∞-category with finite limits; see [Ba, H1]). Here the objects X ∈ Corr(C) are the objects of C, the morphisms CorrC (X, Y ) are arbitrary spans in C, Z ~⑦⑦⑦⑦⑦⑦⑦ ❅❅❅❅❅❅❅ Y X (more generally one can require the left and right legs to live in specified subcategories of C as in [GR2]). The composition of morphisms Z ∈ CorrC (X, Y ) and W ∈ CorrC (Y, U ) is given by fiber product Z ×Y W z✈✈✈✈✈✈✈✈✈ $■■■■■■■■■■ Y Z ~⑦⑦⑦⑦⑦⑦⑦ X $■■■■■■■■■ z✉✉✉✉✉✉✉✉✉✉ W ❆❆❆❆❆❆❆❆ U / / ~  z $ ~ $ z 10 DAVID BEN-ZVI AND DAVID NADLER and the symmetric monoidal structure is given in terms of that on C (Cartesian product in the case of stacks). For the purpose of calculating dimensions and traces, we need not require any further properties of the spaces of Corr(C), since we need only the modest local data discussed in Remarks 2.3 and 2.6. (See [L3] and [FHLT], where the higher categories F amn of iterated correspondences of manifolds are constructed and applied.) With applications in mind, we will specialize to the correspondence category Corrk = Corr(Stk) of derived stacks over k. It would also be interesting to work with smooth manifolds instead, for example through the theory of C∞-stacks [J] (see Remark 2.20). It is natural to enhance Corr(C) to a 2-category Corr(C) by allowing non-invertible maps, or more generally correspondences, between correspondences (see Section V of [GR2] or [H1] for full details), so that maps from X to Y form the category of objects over X × Y . Our constructions naturally fit into the 2-category Corrprop with non-invertible 2-morphisms restricted to be proper (or more generally ind-proper) maps of correspondences. k Remark 2.9 (Correspondences are bimodules). It is useful to view the correspondence category Corr within the framework of coalgebras in symmetric monoidal categories. The diagonal map X → X × X makes any space or stack into a cocommutative coalgebra object with respect to the Cartesian product monoidal structure (or commutative coalgebra in the opposite category). Moreover, a map Z → X is equivalent to an X-comodule structure on Z. Thus correspondences from X to Y may be interpreted as X − Y -bicomodules, with composition of correspondences given by tensor product of bicomodules. Furthermore, it is natural to enhance Corr to a 2-category by allowing non-invertible maps between correspondences. This can be viewed as a special case of the Morita category of coalgebras in a symmetric monoidal category. The 2-category Corr of spaces, correspondences, and maps of correspondences is the Morita category on spaces regarded as coalgebra objects. (In particular, the cocommutativity of the coalgebra objects implies they are canonically self- dual, and the transpose of a correspondence is the same correspondence read backwards.) If we further keep track of the En-coalgebra structure of spaces and consider the corresponding Morita (n+1)-category, we recover the (n+1)-category of iterated correspondences of correspondences. (See [H1, H2] for a thorough treatment of categories of spans and Morita categories of En algebras; see also for example the category F amn of [L3] and [FHLT] in the topological setting.) 2.2.1. Geometric dimensions and loop spaces. A crucial feature of the category Corrk is that any object X ∈ Corrk is dualizable (in fact, canonically self-dual), thanks to the diagonal correspondence.3 Note for this it is crucial that we allow all maps, including the map πX : X → pt for any X, as possible legs in a span. We have the following calculations of dimensions and their functoriality. Note that the point pt = Spec k is the unit of Corrk. We keep track of properness of maps of correspondences for the later application of sheaf theory. Proposition 2.10. Let Corrk be the category of derived stacks and correspondences, and Corrprop the 2-category of derived stacks, correspondences, and proper maps of correspondences. (1) Any derived stack X is dualizable as an object of Corrk, and its dimension dim(X) is k identified with the loop space regarded as a self-correspondence of pt = Spec k. LX = X S 1 ≃ X ×X×X X 3Likewise, if we wish to make a space n-dualizable for any n we may simply consider it as an object of a higher correspondence category as in Remark 2.9, since En-(co)algebras are n + 1-dualizable objects of the corresponding Morita category. In other words, a space X defines a topological field theory of any dimension valued in the appropriate correspondence category. NONLINEAR TRACES 11 (2) A map f : X → Y regarded as a correspondence from X to Y is continuous in Corrprop k if and only if f is proper. Given a proper map f : X → Y , its induced map is identified with the loop map dim(f ) : dim(X) / dim(Y ) Lf : LX / LY Remark 2.11. All of the objects and maps of the proposition have natural S1-actions, on the one hand coming from loop rotation, on the other hand coming from the cyclic symmetry of dimensions. One can check that the identifications of the proposition are S1-equivariant (see Remark 4.2). Remark 2.12. Recall [To2, BN10b] that for a derived scheme X, the loop space LX ≃ TX[−1] is the total space of the shifted tangent complex. The action map of the S1-rotation action is encoded by the de Rham differential. For an underived stack X, the loop space is a derived enhancement of the inertia stack IX = {x ∈ X, γ ∈ Aut(x)}. The action map of the S1-rotation action is manifested by the "universal automorphism" of any sheaf on LX. Example 2.13. Let G denote an algebraic group and BG = pt/G its classifying space. There is a canonical identification LBG ≃ G/G of the loop space and adjoint quotient. Suppose we are given a G-derived stack X, or equivalently a morphism π : X/G → BG, from which one recovers X ≃ X/G ×BG pt. (Note that if we want π proper we should take X itself proper.) Let us explain how the loop map Lπ : L(X/G) → L(BG) captures the fixed points of G acting on X. For any self-map g : X → X, let us write X g for the derived fixed point locus given by the derived intersection X g = Γg ×X×X X. of the graph Γg ⊂ X × X with the diagonal. Then Lπ map fits into a commutative square L(X/G) Lπ / L(BG) ∼ ∼ {g ∈ G, x ∈ X g}/G p / G/G where p projects to the group element. In particular, fix a group element g ∈ G, with conjugacy class Og ⊂ G, and centralizer ZG(g) ⊂ G, so that Og/G ≃ BZG(g) ∈ G/G. Then the corresponding fiber of Lπ is the equivariant fixed point locus X g G = X g/ZG(g), or in other words we have a fiber diagram X g G / Og/G L(X/G) Lπ / / L(BG) Let us specialize to the case of a subgroup K ⊂ G, and the quotient X = G/K, so that we have a map of classifying stacks π : BK ≃ G\(G/K) → BG. Here the loop map Lπ realizes / /   /   /   /   12 DAVID BEN-ZVI AND DAVID NADLER the familiar geometry of the Frobenius character formula L(BK) ∼ Lπ / L(BG) ∼ K/K ≃ {g ∈ G, x ∈ (G/K)g}/G p / G/G The equivariant fixed point loci express the equivariant inclusion of conjugacy classes. Specializing further, for G a reductive group, B ⊂ G a Borel subgroup, and X = G/B the flag variety, we recover the group-theoretic Grothendieck-Springer resolution L(BB) ∼ B/B ≃ {g ∈ G, x ∈ (G/B)g}/G Lπ / L(BG) ∼ p / G/G 2.2.2. Geometric traces of correspondences. More generally, we have the following calculations of traces and their functoriality. Proposition 2.14. Let Corrk be the category of derived stacks and correspondences, and Corrprop the 2-category of derived stacks, correspondences, and proper maps of correspondences. (X, X) is its fiber product with the (1) The trace of a self-correspondence Z ∈ Corrprop k k diagonal In particular, for the graph Γf → X × X of a self-map f : X → X, its trace is the fixed point Tr(Z) ≃ Z∆ = Z ×X×X X ≃ Z ×X LX locus of the map Tr(Γg) ≃ X f = Γf ×X×X X (2) Given a proper map f : X → Y regarded as a correspondence from X to Y , and self- (Y, Y ), together with an identification (X, X) and W ∈ Corrprop correspondences Z ∈ Corrprop k k α : Z ∼ / / X ×Y W of correspondences from X to Y , the induced abstract trace map Tr(f, α) : Tr(Z) / Tr(W ) is equivalent to the induced geometric map τ (f, α) : Z∆X / W ∆Y 2.3. Trace formulas via sheaf theories. Given any sufficiently functorial method of mea- suring derived stacks, the preceding calculations of geometric dimensions, traces and their functoriality immediately lead to trace and character formulas. To formalize the functorial- ity needed, we will use the language of sheaf theories. Broadly speaking, a sheaf theory is a representation (symmetric monoidal functor out) of a correspondence category in the way a topological field theory is a representation of a cobordism category.4 It provides an approach to encoding the standard operations on coherent sheaves and D-modules, developed by Gaits- gory and Rozenblyum in the book [GR2] (following a suggestion of Lurie and previous versions in [FG, G1, DG, GR1]). 4Indeed a typical mechanism to construct "Lagrangian" field theories is as the composition of a sheaf theory with a "classical field theory" as in [FHLT, H1], a symmetric monoidal functor from a cobordism category to a correspondence category.   /   /   /   / / / NONLINEAR TRACES 13 We will take the target of our sheaf theories to be the linear setting of the symmetric monoidal (∞, 2)-category dgCat of presentable k-linear differential graded categories with continuous functors and natural transformations. (Recall from Setting 1.1 that for applications all stacks are assumed to be QCA or ind-inf-schemes. Also all proper maps can be replaced by ind-proper ones at no cost.) k Definition 2.15. A sheaf theory is a symmetric monoidal functor of (∞, 2)-categories S : Corrprop k / dgCat k from correspondences of stacks (with 2-morphisms given by proper maps of correspondences) to dg categories. We denote by S : Corrk / dgCatk the underlying 1-categorical sheaf theory, i.e. categories obtained by forgetting noninvertible morphisms. the symmetric monoidal functor on (∞, 1)- Let us first spell out some of the structure encoded in a 1-categorical sheaf theory S. The graph of a map of stacks f : X → Y provides a correspondence from X to Y and a correspondence from Y to X. We denote the respective induced maps by f∗ : S(X) → S(Y ) and f ! : S(Y ) → S(X). For π : X → pt = Spec k, we denote by ωX = π!k ∈ S(X) the S-analogue of the dualizing sheaf, and by ω(X) = π∗ωX ∈ S(pt) = dgV ectk the S-analogue of "global volume forms". We adopt traditional notations whenever possible, for example writing Γ(X, F ) = π∗(F ), for F ∈ S(X) The functoriality of S concisely encodes base change for f∗ and f !. Its symmetric monoidal structure provides equivalences S(X × Y ) ≃ S(X) ⊗ S(Y ), as well as a symmetric monoidal structure on S(X) for any X (using pullback along diagonal maps). The 2-categorical extension S further encodes an identification of f ! with the right adjoint of f∗ for f proper. Since a sheaf theory S is symmetric monoidal, it is automatically compatible with dimensions and traces: for any X ∈ Corrk, and any endomorphism Z ∈ Corrk(X, X), we have dim(S(X)) ≃ S(dim(X)) Tr(S(Z)) ≃ S(Tr(Z)) Let us combine this with the calculation of the right hand sides and highlight specific examples of interest. Proposition 2.16. Fix a sheaf theory S : Corrk → dgCatk. (1) The S-dimension dim(S(X)) = HH∗(S(X)) of any X ∈ Corrk is S1-equivariantly equivalent with S-global volume forms on the loop space dim(S(X)) ≃ ω(LX). In particular, for G an affine algebraic group, characters of S-valued G-representations are adjoint-equivariant S-global volume forms (2) The S-trace of any endomorphism Z ∈ Corrk(X, X) is equivalent to S-global volume forms on the restriction to the diagonal dim(S(BG)) ≃ ω(G/G) Tr(S(Z)) ≃ ω(Z∆) / / 14 DAVID BEN-ZVI AND DAVID NADLER In particular, the S-trace of a self-map f : X → X is equivalent to S-volume forms on the f -fixed point locus Tr(f∗) ≃ ω(X f ) Remark 2.17 (Local sheaf theory). To apply this proposition, far less structure than a full sheaf theory is required. We only need the data of the functor S on the handful of objects and morphisms involved in the construction of dimensions and traces as in Remark 2.3. In particular, we only need base change isomorphisms for pullback and pushforward along specific diagrams, rather than the general base change provided by a functor out of Corrk. This is often easy to verify in practice, in particular for the examples Q, Q! and D (see for example [BFN] for the quasicoherent setting). 2.3.1. Examples of sheaf theories. As we explain in Section 5, the work [GR2] (combined with essential results from [DG]) construct two sheaf theories Q! and D: • Theory Q!: the theory of ind-coherent sheaves Q!(X). This is the "large" version Q!(X) = Ind Coh(X) of the category of coherent sheaves, which by definition are the compact objects in Q!(X). (For smooth X, ind-coherent and quasicoherent sheaves are equivalent.) Maps are given by the standard pushforward f∗ and exceptional pullback f !. The Q!-dualizing sheaf is the usual dualizing complex ωX , and (for X proper) the Q!-global volume forms are its sections RΓ(X, ωX) = RΓ(X, OX )∗. The K-theory of Q!(X) is algebraic G-theory G(X), the homological version of algebraic K-theory for potentially singular spaces suited to Grothendieck- Riemann-Roch theorems. • Theory D: the theory of D-modules D(X) with the standard functors f∗ and f !. The compact objects are necessarily coherent D-modules (this suffices for X a scheme; see [DG] for a characterization in the case of a stack). The D-dualizing sheaf is the Verdier dualizing complex ωX , and the D-global volume forms (for X smooth) are the Borel-Moore homology RΓ(XdR, ωX ) = HdR(X)∗. Remark 2.18. More precisely, [GR2] construct the sheaf theories Q! and D as lax symmetric monoidal functors on a much broader class of stacks, with pullbacks allowed for arbitrary maps but pushforward only for schematic morphisms. The strictness follows from results of [DG], as we explain in Section 5.3.1, as does the definition of pushforwards for arbitrary maps of QCA stacks (without the functorial apparatus of [GR2] but sufficient for all the "local" constructions we need, in the sense of Remarks 2.3, 2.6 and 2.17). Remark 2.19 (Quasicoherent sheaves). The theory of quasicoherent sheaves X 7→ Q(X) be- haves similarly with respect to 1-categorical properties. It also defines a symmetric monoidal functor out of the (∞, 1)−category of correspondences of stacks, using standard pullback f ∗ and pushforward f∗ functors. Assuming X is perfect (in the sense of [BFN]), the compact objects of Q(X) form the subcategory of perfect complexes Perf (X), and we have Q(X) = Ind Perf (X). The analog of the dualizing sheaf is the structure sheaf OX , and the Q-"global volume forms" are the global functions RΓ(X, OX ). The K-theory of Q(X) is the usual algebraic K-theory K(X). However while we have (f ∗, f∗) adjunction and f ∗ preserves perfection for arbitrary mor- phisms, proper pushforward does not preserve perfection and we do not have proper adjunction of the form (f∗, f ∗) (unless we add smoothness and twisting by relative dualizing sheaves). In other words, Q and K-theory are better adapted to pullback, while Q! and G-theory are better adapted to integration and character formulas. Remark 2.20 (Sheaf theories in differential topology and elliptic operators). It is tempting to think of sheaf theories in algebraic geometry as analogues of elliptic operators or complexes NONLINEAR TRACES 15 in differential topology. In particular, the theory Q!(X) for a smooth variety X is a natural setting for the study of the Dolbeault ∂-operator coupled to vector bundles, while the theory D(X) is similarly a natural setting for the study of the de Rham operator d coupled to vector bundles. The pushforward operation is the analogue of the index. In this direction, it would be interesting to develop sheaf theories on derived manifolds, for example C∞-schemes and stacks. Quasicoherent sheaves in the sense of Joyce [J] are a natural candidate. Another interesting setting is categories of elliptic complexes on manifolds. The general results below would then provide an approach to generalizations of the classical Atiyah-Singer and Atiyah-Bott theorems. Let us spell out the main ingredients of Proposition 2.16 for our examples. Recall that for X a smooth scheme, LX ≃ SpecX Sym•(ΩX [1]), and for BG a classifying stack, L(BG) ≃ G/G. • Theory Q: For X a smooth scheme, we have the HKR identification of functions on the loop space (or the Hochschild chain complex) with differential forms, dim(Q(X)) ≃ Γ(X, Sym•(ΩX [1])), or more generally, Q-global volume forms on X f are the coherent cohomology O(X f ). For BG a classifying stack, Q-global volume forms on L(BG) are the coherent cohomology O(G/G), which for G reductive are the underived invariants O(T )W . • Theory Q!: For X smooth, we have Q(X) ≃ Q!(X), and so we recover the above de- scriptions. For X proper, Q-global volume forms on LX are the dual of the Hochschild chain complex (see [P]). • Theory D: For X a smooth scheme, D-global volume forms on LX are the de Rham dR(X), or more generally, D-global volume forms on X f are the de cochains dim(D(X)) ≃ C ∗ dR(X f ), or equivalently those of the underlying underived scheme of X f . For Rham cochains C ∗ BG a classifying stack, D-global volume forms on L(BG) are the Borel-Moore homology of G/G. 2.3.2. Integration formulas for traces. Now let us turn to the functoriality of dimensions and traces, which is reflected in integration of volume forms along proper maps. For a sheaf theory S : Corrprop k → dgCat , k the counit of the (f∗, f !) adjunction for a proper map f : X → Y gives rise to a canonical integration map Rf : ω(X) / ω(Y ) Theorem 2.21. Fix a sheaf theory S : Corrk → dgCat . k (1) For any proper map f : X → Y , the induced map on dimensions is identified (S1-equivariantly) with integration along the loop map dim(f∗) : dim(S(X)) / dim(S(Y )) (2) Given a proper map f : X → Y regarded as a correspondence from X to Y , and self- dim(f∗) ≃RLf : ω(LX) / ω(LY ) correspondences Z ∈ Corrk(X, X) and W ∈ Corrk(Y, Y ), together with an identification α : Z ∼ / / X ×Y W of correspondences from X to Y , the induced trace map is identified with integration along the natural map Tr(f∗, α) ≃Rτ (f,s) : ω(Z∆X ) / ω(W ∆Y ) / / / / 16 DAVID BEN-ZVI AND DAVID NADLER Remark 2.22. Similarly, in the case of the theory Q of quasicoherent sheaves, the standard adjunction (f ∗, f∗) leads to the evident contravariant functoriality of dimensions under arbitrary maps, given by pullback of functions on loop spaces. Remark 2.23 (Categorified version). For applications to categorical representation theory, in particular the geometric Langlands program, it is interesting to have character formulas for group actions on categories. Such formulas would follow from a good formalism of "stack theo- ries", the higher unstable analogs of sheaf theories, such as the assignment X → ShvCatk(X). Such stack theories could be formulated as symmetric monoidal functors S : Corrk → A out of a correspondence (∞, 2) (or more naturally (∞, 3)) category with values in a category A such as that of module categories for dgCatk. Namely, we are interested in categorified analogues of D and Q, taking values in the ∞-category PrL of presentable ∞-categories, in which we assign to a scheme or stack X the ∞-category of quasicoherent sheaves of module categories over D or Q. Since such theories have not been fully constructed yet, we will only briefly sketch the idea. For any stack X and sheaf theory S, the category of sheaves S(X) is naturally symmetric monoidal, and so we may consider its ∞-category of (presentable, stable) module categories S(X)-mod. To obtain a more meaningful geometric theory we should sheafify this construction. For example, strong or Harish-Chandra G-categories (in other words, module categories over D(G) with convolution) are identified with sheaves of categories over the de Rham stack of BG. However, in the quasicoherent case, the "1-affineness" theorem of Gaitsgory [G2] identifies Q(X)-modules with sheaves of categories on X for a large class of stacks (specifically, for X an eventually coconnective quasi-compact algebraic stack of finite type with an affine diagonal over a field of characteristic 0). In particular, Q(BG)-modules are identified with algebraic G-categories. In the quasicoherent case, the general formalism of this paper should provide an S1-equivariant equivalence dim(Q(X)-mod) = Q(LX), identifying the class [Q(X)] of the structure stack with the structure sheaf O(LX). In particular, the characters of quasicoherent G-categories are given by Q(G/G). The induced map on dimensions dim(f∗) : dim(Q(X)-mod) → dim(Q(Y )-mod) is identified S1-equivariantly with the morphism given by pushforward along the loop map dim(f∗) = Lf∗ : Q(LX) / Q(LY ) In particular, for an algebraic group G and G-space X with π : X/G → BG, the character of the G-category Q(X/G) is given by the pushforward Lπ∗O(LX/G) ∈ Q(G/G). Analogous results are expected for strong or Harish-Chandra G-categories (module categories for D(G) with convolution) using the sheafification of the theory of D(X)-module categories. We hope to return to these applications in future works. 3. Traces in category theory 3.1. Preliminaries. Our working setting is the higher category theory and algebra developed by J. Lurie [L1, L2, L4], see Chapter I.1 of [GR2] for an excellent overview. Throughout what follows, we will fix once and for all a symmetric monoidal (∞, 2)-category A with unit object 1A. By forgetting non-invertible 2-morphisms we obtain a symmetric monoidal (∞, 1)-category f (A), which we will abusively refer to as A whenever only invertible higher morphisms are involved. Conversely, given a symmetric monoidal (∞, 1)-category C, we can always regard it as a symmetric monoidal (∞, 2)-category i(C) with all 2-morphisms invertible.5 5One can understand the above two operations as forming an adjoint pair (i, f ). / NONLINEAR TRACES 17 Thus developments for higher ∞-categories equally well apply to the more familiar (∞, 1)- categories. In what follows, noninvertible 2-morphisms only play a significant role starting with Section 3.4. We will use ⊗ to denote the symmetric monoidal structure of A. We will write ΩA = EndA(1A) for the "based loops" in A, or in other words, the symmetric monoidal (∞, 1)- category of endomorphisms of the monoidal unit 1A. Note that the monoidal unit 1ΩA is nothing more than the identity id1A of the monoidal unit 1A. Example 3.1 (Algebras). Fix a symmetric monoidal (∞, 1)-category C, and let A = Alg(C) denote the Morita (∞, 2)-category of algebras, bimodules, and intertwiners of bimodules within C. The forgetful map A = Alg(C) → C is symmetric monoidal, and in particular, the monoidal unit 1A is the monoidal unit 1C equipped with its natural algebra structure. Finally, we have ΩA ≃ C. For a specific example, one could take C = k-mod = dgV ectk the (∞, 1)-category of com- plexes of k-modules (with quasi-isomorphisms inverted). Then A = Alg(C) is the (∞, 2)- category of k-algebras, bimodules, and intertwiners of bimodules. Example 3.2 (Categories). A natural source of (∞, 2)-categories is given by various theories , the (∞, 2)-category of k-linear of (∞, 1)-categories. For example, one could consider dgCat stable presentable ∞-categories (or k-linear presentable dg categories), k-linear continuous func- tors, and natural transformations. k Observe that Alg(k-mod) is a full subcategory of dgCat k , via the functor assigning to a k-algebra its stable presentable ∞-category of modules. The essential image consists of dg categories admitting a compact generator. 3.2. Dualizability. Definition 3.3. An object A of the symmetric monoidal (∞, 2)-category A is said to be dualizable (equivalently, A is dualizable in the (∞, 1)-category f (A)) if it admits a monoidal dual: there is a dual object A∨ ∈ A and evaluation and coevaluation morphisms ǫA : A∨ ⊗ A / 1A ηA : 1A / A ⊗ A∨ such that the usual compositions are naturally equivalent to the identity morphism A ηA⊗idA / A ⊗ A∨ ⊗ A idA ⊗ǫA / A A∨ idA∨ ⊗ηA / / A∨ ⊗ A ⊗ A∨ ǫA⊗idA∨ / A∨ Example 3.4. Any algebra object A ∈ Alg(C) is dualizable with dual the opposite algebra Aop ∈ Alg(C). The evaluation morphism ǫA : Aop ⊗ A / 1C is given by A itself regarded as an A-bimodule. The coevaluation morphism ηA : 1C / A ⊗ Aop is also given by A itself regarded as an A-bimodule. 3.2.1. Dualizable morphisms. Consider two objects A, B ∈ A, and a morphism Example 3.5. If A = Alg(C), then Φ is simply an Aop ⊗ B-module. Φ : A / B. / / / / / / / / 18 DAVID BEN-ZVI AND DAVID NADLER If B is dualizable with dual B∨, we can package Φ in the equivalent form of the morphism defined by eΦ : B∨ ⊗ A → 1A B∨ ⊗ A eΦ / / 1A ǫB ;✈✈✈✈✈✈✈✈✈ idB∨ ⊗Φ B ⊗ B∨ If A is dualizable with dual A∨, we can package Φ in the equivalent form of the morphism defined by uΦ : 1A → B ⊗ A∨ A ⊗ A∨ Φ⊗idA∨ ηA ;✈✈✈✈✈✈✈✈✈ uΦ / 1A / B ⊗ A∨ If both A and B are dualizable, we can also encode Φ by its dual morphism Φ∨ : B∨ / A∨ defined by B∨ idB∨ ⊗ηA / B∨ ⊗ A ⊗ A∨ idB∨ ⊗Φ⊗idA∨ / B∨ ⊗ B ⊗ A∨ ǫB∨ ⊗idA∨ / A∨ There is a natural composition identity Φ∨ (ΦΨ)∨ ≃ Ψ∨Φ∨ Note that for fixed A, B, the construction Φ 7→ Φ∨ naturally defines a covariant map (−)∨ : Hom(A, B) / Hom(B∨, A∨) and in particular a morphism Φ1 → Φ2 induces a natural morphism Φ∨ 1 → Φ∨ 2 . Let us record the canonical equivalences encoded by the following commutative diagrams (3.1) 1A ηA uΦ / ;✈✈✈✈✈✈✈✈✈ #❍❍❍❍❍❍❍❍❍ ηB ǫA∨ #❍❍❍❍❍❍❍❍❍ ;✈✈✈✈✈✈✈✈✈ eΦ / ǫB A ⊗ A∨ A∨ ⊗ A Φ⊗idA∨ Φ∨⊗idA / B ⊗ A∨ B∨ ⊗ A 1A idB ⊗Φ∨ idB∨ ⊗Φ B ⊗ B∨ B ⊗ B∨ Example 3.6. In the setting of algebras, bimodules and intertwiners, the morphisms Φ, uΦ, eΦ and Φ∨ are all different manifestations of the same bimodule Φ, making their various com- patibilities particularly evident.   ;   ; / 4 4 / / / /   # ; # O O /   O O ; NONLINEAR TRACES 19 Definition 3.7. (1) A morphism Φ : A → B is said to be left dualizable if it admits a left adjoint: there is a morphism Φℓ : B → A and unit and counit morphisms ηΦ : idB / Φ ◦ Φℓ ǫΦ : Φℓ ◦ Φ / idA satisfying the usual identities. (2) A morphism Φ : A → B is said to be right dualizable if it admits a right adjoint: there is a morphism Φr : B → A and unit and counit morphisms ηΦ : idA / Φr ◦ Φ ǫΦ : Φ ◦ Φr / idB satisfying the usual identities. Remark 3.8. If A and B are dualizable, and Φ : A → B is left (resp. right) dualizable, then Φ∨ : B∨ → A∨ is right (resp. left) dualizable with right adjoint (Φℓ)∨ : A∨ → B∨ (resp. left adjoint (Φr)∨ : A∨ → B∨). 3.3. Traces and dimensions. (We continue to refer to [TV, HSS] for thorough treatments of the theory of traces in higher category theory.) Let A ∈ A be a dualizable object with dual A∨. Consider an endomorphism Since A is dualizable, Φ has a trace defined as follows. Φ : A / A Definition 3.9. (1) The trace of Φ : A → A is the object Tr(Φ) ∈ ΩA defined by 1A ηA / A ⊗ A∨ Φ⊗idA / A ⊗ A∨ ǫA / / 1A . Tr(Φ) Given a natural transformation ϕ : Φ → Ψ, we define the induced morphism Tr(ϕ) : Tr(Φ) / Tr(Φ′) by applying ϕ ⊗ idA∨ to the middle arrow above. (2) The dimension (or Hochschild homology) of A is the trace of the identity or in other words, the object defined by dim(A) = Tr(idA) ∈ ΩA 1A ηA A ⊗ A∨ dim(A) ǫA / 1A Remark 3.10. Equivalently, we can describe the trace as the composition 1A Φ / / End(A) ∼ / / A ⊗ A∨ ǫA / / 1A where the middle arrow is the identification deduced from the dualizability of A. Remark 3.11. Observe that for fixed dualizable A ∈ A, taking traces gives a functor Tr : End(A) / ΩA Remark 3.12. Observe that for any dualizable endomorphism Φ, the standard identities encoded by Diagrams 3.1 give rise to an identification Tr(Φ) ≃ Tr(Φ∨) / / / / / / / / / / / / 20 DAVID BEN-ZVI AND DAVID NADLER Example 3.13. When A = 1A is the monoidal unit, and Φ : 1A → 1A is an endomorphism, we have an evident equivalence of endomorphisms Tr(Φ) ≃ Φ Theorem 3.14 ([L3]). There is a canonical S1-action on the dimension dim(A) of any dual- izable object A of a symmetric monoidal ∞-category A. 3.3.1. Cyclic symmetry. Proposition 3.15. Given two morphisms A Φ / Ψ / B between dualizable objects A, B ∈ A, there is a canonical equivalence m(Φ, Ψ) : Tr(Φ ◦ Ψ) ∼ / / Tr(Ψ ◦ Φ) functorial in morphisms of both Φ and Ψ. Proof. We construct m(Φ, Ψ) following the commutative diagram below: A ⊗ A∨ Φ⊗idA∨ B ⊗ A∨ Ψ⊗idA∨ A ⊗ A∨ 1A ηA ;✈✈✈✈✈✈✈✈✈ $❍❍❍❍❍❍❍❍❍ ηB idA ⊗Ψ∨ ✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿ idB ⊗Ψ∨ ✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿ ǫA $❍❍❍❍❍❍❍❍❍ ;✈✈✈✈✈✈✈✈✈ ǫB 1A B ⊗ B∨ / A ⊗ B∨ / B ⊗ B∨ Ψ⊗idB∨ Φ⊗idB∨ Following the top edge, we find the definition of Tr(Ψ ◦ Φ). Following the bottom edge, we find the definition of Tr(Φ ◦ Ψ). The identifications filling the left and right diamonds arise from the standard identities encoded by Diagrams 3.1. The identification filling the central square results from the symmetric monoidal structure. The construction is evidently functorial for morphisms Φ → Φ′. The functoriality for mor- phisms Ψ → Ψ′ is similar, once one recalls that the construction Ψ 7→ Ψ∨ is covariantly functorial in morphisms of Ψ. (cid:3) Example 3.16. Taking Φ = idA yields a canonical equivalence γ ′ : idTr(Φ′) ∼ / / m(idA, Φ′ A) and likewise, taking Φ′ = idA yields a canonical equivalence γ : idTr(Φ) ∼ / / m(ΦA, idA) Thus taking Φ = Φ′ = idA yields an automorphism of the identity of the Hochschild homology called the BV homotopy. (γ ′)−1 ◦ γ : idTr(idA) ∼ / / idTr(idA) Remark 3.17. The proposition is only the initial part of the full cyclic symmetry of trace (see Remark 3.28), and the example is the lowest level structure of the S1-action on Hochschild homology (see Theorem 3.14) defining cyclic homology. o o / /  / /  $ ; $ / / ; NONLINEAR TRACES 21 Lemma 3.18. Given morphisms A Φ / / B Ψ / / C Υ / / A between dualizable objects A, B, C ∈ A, there is a canonical commutative diagram Tr(ΨΦΥ) Tr(ΦΥΨ) m(Ψ,ΦΥ) )❘❘❘❘❘❘❘❘❘❘❘❘❘❘ m(ΨΦ,Υ) m(Φ,ΥΨ) Tr(ΥΨΦ) Proof. We construct the desired equivalence from the following diagram: C ⊗ C ∨ A ⊗ C ∨ B ⊗ C ∨ C ⊗ C ∨ ;✈✈✈✈✈✈✈✈✈ #❍❍❍❍❍❍❍❍❍ Υ / Ψ∨ %▲▲▲▲▲▲▲▲▲▲ %▲▲▲▲▲▲▲▲▲▲ Ψ / Φ∨ Φ / Ψ∨ %▲▲▲▲▲▲▲▲▲▲ %▲▲▲▲▲▲▲▲▲▲ Υ / Φ∨ Ψ / Ψ∨ %▲▲▲▲▲▲▲▲▲▲ %▲▲▲▲▲▲▲▲▲▲ Φ / Φ∨ ǫC #❍❍❍❍❍❍❍❍❍ ;✈✈✈✈✈✈✈✈✈ ǫA 1A B ⊗ B∨ C ⊗ B∨ A ⊗ B∨ B ⊗ B∨ / 1A A ⊗ A∨ Φ / / B ⊗ A∨ Ψ / / C ⊗ A∨ Υ / / A ⊗ A∨ The natural transformations m(Ψ, ΦΥ) and m(Φ, ΥΨ) describe passage from the top row to the middle row and from the middle to the bottom, respectively. The transformation m(ΨΦ, Υ) can then be identified with the transformation from the top row to the bottom given by inserting the diagonal morphisms id ⊗Φ∨ ◦ Ψ∨ and using standard composition identities. (cid:3) 3.4. Functoriality of dimension. Let Acont ⊂ A denote the (∞, 2)-subcategory of dualizable objects and continuous or right dualizable morphisms (morphisms that are left duals). Definition 3.19. Let Ψ : A → B denote a morphism in Acont with right adjoint Ψr : B → A. We define the induced morphism of dimensions to be the composition dim(Ψ) : dim(A) / dim(B) Tr(idA) ηΨ / / Tr(Ψr ◦ Ψ) m(Ψr ,Ψ) / Tr(Ψ ◦ Ψr) ǫΨ / / Tr(idB) Remark 3.20. In other words, the morphism dim(Ψ) is defined by the following diagram A ⊗ A∨ 1A B ⊗ A∨ ηA uΨ Ψ⊗idA∨ <②②②②②②②②②②②②②②②②②② "❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊ idB ⊗Ψ∨ ηB Ψr⊗idA ǫA "❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊ <②②②②②②②②②②②②②②②②② ǫB cΨ idB∨ ⊗Ψr∨ B ⊗ B∨ dim(A) 1A 1A Tr(ΨrΨ)≃Tr(ΨΨr) 1A dim(B) / / )   / % / % / % # ; / / # / % / % / % / ; / /   " < / / " O O / /   / / O O < 22 DAVID BEN-ZVI AND DAVID NADLER Following the top and bottom edge, we find the respective definitions of dim(A) and dim(B). The unit ηΨ defines a morphism from the top edge to the top zig-zag. The counit ǫΨ defines a morphism from the bottom zig-zag to the bottom edge. The passage from the top to bottom zig-zag is given by the construction m(Ψr, Ψ) and the identification Tr(Ψr∨ ◦ Ψ∨) ≃ Tr((Ψ ◦ Ψr)∨) ≃ Tr(Ψ ◦ Ψr) Proposition 3.21. For a diagram within Acont, there is a canonical equivalence A Φ / / B Ψ / / C dim(Ψ ◦ Φ) ≃ dim(Ψ) ◦ dim(Φ) : dim(A) / dim(C) Proof. The equivalence is given by filling in the following diagram dim(A) ηΦ Tr(ΦrΦ) m Tr(ΦΦr) ǫΦ dim(B) ηΨ ηΨ ηΨ ηΨ &◆◆◆◆◆◆◆◆◆◆◆ Tr(ΦrΨrΨΦ) m / m (PPPPPPPPPPPP Tr(ΨrΨΦΦr) ǫΦ / Tr(ΨrΨ) m m Tr(ΨΦΦrΨr) ǫΦ / Tr(ΨΨr) ǫΨΦ '❖❖❖❖❖❖❖❖❖❖❖ ǫΨ dim(C) Along the three boundary edges, we find the definitions of dim(Φ), dim(Ψ) and dim(ΨΦ) respectively. The two corner triangles are given by the composition identities for adjoints (for example, at the top left, relating the adjoint of ΦΨ with the composition of adjoints of Ψ and Φ). The middle triangle is given by the identity of Lemma 3.18. The top right square is given by taking traces of the evident commutative diagram of endo- morphisms ΦΦr ⊗ IdB∨ / IdB ⊗ IdB∨ ΦΦr ⊗ (ΨrΨ)∨ / IdB ⊗(ΨrΨ)∨ and using the canonical identification Tr(F ) = Tr(F ∨) for any dualizable morphism. Finally, the two remaining commuting squares are given by the functoriality of the cyclic rotation of the trace in its two arguments. For instance, in the top left square, we may either rotate Tr(Φr ◦ (IdA ◦Φ)) and then apply the unit ηΨ : IdA → ΨrΨ or first apply the unit and then rotate. This concludes the construction. (cid:3) Since we have an evident equivalence dim(1A) ≃ 1A for the unit 1A ∈ A, we have the following specialization of Proposition 3.21 in which we adopt suggestive notation. / / / & / /   / /     / ( /     / '     /   / NONLINEAR TRACES 23 Corollary 3.22 (Abstract Grothendieck-Riemann-Roch). Let A, B ∈ Acont and V : 1A → A and π∗ : A → B morphisms in Acont. Then the following diagram naturally commutes 1A / dim(A) dim(V ) #●●●●●●●●● dim(π∗V ) dim(B) dim(π∗) Remark 3.23. One can show along the same lines as the proposition that taking dimensions extends to a symmetric monoidal functor 3.5. Functoriality of traces. We would like to capture the functoriality for traces of arbitrary endomorphisms of dualizable objects. For this purpose we define a morphism between pairs dim : Acont / ΩA. of an object and an endomorphism to consist of a pair A ∈ Acont ΦA ∈ EndA(A) Ψ ∈ HomAcont (A, B) ψ : Ψ ◦ ΦA ≃ / / ΦB ◦ Ψ of a morphism and a commuting structure. Definition 3.24. For a morphism as above, we define the induced morphism of traces (Ψ, ψ) : (A, ΦA) → (B, ΦB) Tr(Ψ, ψ) : Tr(ΦA) / Tr(ΦB) to be the composition Tr(ΦA) ηΨ / / Tr(ΨrΨΦA) ψ / Tr(ΨrΦBΨ) m(Ψr,ΦB Ψ) / Tr(ΦBΨΨr) ǫΨ / / Tr(ΦB) Remark 3.25. Note that we could alternatively define a morphism Tr(Ψ, ψ) by applying the unit ηΨ to the right of ΦA, rotating the trace in the opposite direction, and again using the counit on the right. It is elementary to give a natural equivalence of the two constructions using nothing more than the dualizability of A. Remark 3.26. In parallel with Remark 3.20 about the functoriality of dimensions, it is enlight- ening to realize the functoriality of traces as a chase through the following diagram A ⊗ A∨ ΦA⊗idA∨ / A ⊗ A∨ Ψr ⊗idA∨ Ψ⊗idA∨ ηA uΨ Ψ⊗idA∨ <②②②②②②②②②②②②②②②②②② "❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊ idB ⊗Ψ∨ ηB Ψr ⊗idA∨ ǫA "❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊ <②②②②②②②②②②②②②②②②② ǫB cΨ idB∨ ⊗Ψr∨ Tr(ΦA) 1A 1A Tr(ΨrΨΦA)≃ Tr(ΨrΦB Ψ)≃Tr(ΦB ΨΨr) 1A Tr(ΦB ) 1A B ⊗ A∨ B ⊗ A∨ ΦB ⊗idA∨ idB∨ ⊗Ψr∨ idB ⊗Ψ∨ B ⊗ B∨ ΦB ⊗idB∨ / B ⊗ B∨ # /   / / / /   / "   < / / " O O / /   / /   O O / / O O / O O < 24 DAVID BEN-ZVI AND DAVID NADLER Proposition 3.27. Suppose given objects A, B, C ∈ Acont, endomorphisms ΦA, ΦB, ΦC , a commutative diagram of continuous morphisms ΨAC / B ΨBC / C ΨAB A and commuting structures sAB : ΨABΦA ∼ / / ΦBΨAB sBC : ΨBCΦB ∼ / / ΦC ΨBC sAC : ΨACΦA ∼ / / ΦC ΨAC with an identification sAC ≃ sBC sAB. Then there is a canonical equivalence Tr(ΨAC , sAC) ≃ Tr(ΨBC , sBC) ◦ Tr(ΨAB, sAB) : Tr(ΦA) / Tr(ΦC ) Proof. The construction is obtained from following a minor expansion of the diagram proving Proposition 3.21. The additional moves needed are commuting the commuting structures past the symmetry m of the trace and the unit and counits of the adjunctions. These all follow immediately from the 2-categorical interchange law for natural transformations. (cid:3) Remark 3.28. The full functoriality of the trace Tr takes roughly the following form, see [HSS] for a detailed treatment. Define the loop category LcontA to be the symmetric monoidal ∞- category with objects consisting of pairs (A, ΦA) of a dualizable object A ∈ A equipped with a (not necessarily continuous) endomorphism ΦA, and morphisms given by pairs (Ψ, ψ) as above with Ψ continuous. Taking traces to extend to a symmetric monoidal functor extending the dimension functor Tr : LcontA / ΩA dim : Acont / ΩA for constant loops ΦA = idA, and trivial commuting structures ψ = idΨ. In order to capture the full cyclic symmetry of the trace Tr, one should further extend it to a homotopical trace valued in ΩA, or in other words, to the appropriate full cyclic bar construction (of which the above forms only the one-simplices). . 4. Traces in Geometry 4.1. Categories of correspondences. For concreteness, we fix a base commutative ring, and work in the symmetric monoidal (∞, 1)-category Stacksk of derived stacks over Spec k. It is worth pointing out that the constructions of this section apply in any presentable ∞-category with the Cartesian symmetric monoidal structure. Let Corrk denote the symmetric monoidal ∞-category of correspondences in Stacksk. Thus morphisms are given by the classifying space of correspondences X Z / Y so all higher morphisms are isomorphisms. Composition of correspondences is given by the derived fiber product. The based loop category is again derived stacks, regarded as self-correspondences of the point Spec k. ΩCorrk = EndCorrk(Spec k) ≃ Stacksk % % / / / / / o o / NONLINEAR TRACES 25 We will also enhance Corrk to the symmetric monoidal (∞, 2)-category Corrk where we now allow noninvertible maps of correspondences Z X ~⑥⑥⑥⑥⑥⑥⑥⑥ `❇❇❇❇❇❇❇❇ W ❆❆❆❆❆❆❆❆ >⑥⑥⑥⑥⑥⑥⑥⑥ Y In other words, the morphisms Corrk(X, Y ) now form the ∞-category Stacks/X×Y of stacks over X × Y with arbitrary morphisms rather than isomorphisms as in Corrk(X, Y ). We will also have need to restrict the class of morphisms of correspondences to some sub- in which we category of Stacks/X×Y . In particular, we will consider the subcategory Corrprop only allow proper maps of correspondences. k 4.2. Traces of correspondences. Given a map Z → X, it is convenient to introduce the symmetric presentation of the based loop space LZ X = Z ×Z×X Z Note the two natural identification with the traditional based loop space LX ×X Z ≃ X ×X×X Z ∼o Z ×Z×X Z ∼ / / Z ×X×X X ≃ Z ×X LX There is a natural rotational equivalence LX ×X Z ≃ Z ×X LX that makes the above two identifications coincide. (It does not preserve base points and is not given by swapping the factors). Thus we can unambiguously identify all of the above versions of the based loop space. Proposition 4.1. (1) Any derived stack X is dualizable as an object of Corrk, with dual X ∨ identified with X itself, and dimension dim(X) identified with the loop space LX = X S 1 ≃ X ×X×X X regarded as a self-correspondence of pt = Spec k. (2) The transpose of any correspondence X ← Z → Y is identified with the reverse corre- spondence Y ← Z → X. The trace of a self-correspondence X ← Z → X is identified with the based loop space Tr(Z) ≃ Z∆X = Z ×X×X X ≃ LZ X regarded as a self-correspondence of pt = Spec k. In particular, the trace of the graph Γf → X × X of a self-map f : X → X is identified with the fixed point locus Tr(f ) ≃ Γf ∆X = Γf ×X×X X ≃ X f Proof. The evaluation and coevaluation presenting the self-duality of X are both given by X itself as a correspondence between pt = Spec k and X × X via the diagonal map. The standard identities follow from the calculation of the fiber product of the two diagonal maps X ∆12 ×X×X×X ∆23X ≃ X ~   ` > o 26 DAVID BEN-ZVI AND DAVID NADLER Thus the dimension of X is the loop space LX {✇✇✇✇✇✇✇✇✇ #●●●●●●●●● #●●●●●●●●● {✇✇✇✇✇✇✇✇✇ X × X X ❅❅❅❅❅❅❅❅ pt X ⑦⑦⑦⑦⑦⑦⑦⑦ pt By definition, the transpose of a correspondence X ← Z → Y is identified with Y ← Z → X by checking the definition Y × Z Z × X v♥♥♥♥♥♥♥♥♥♥♥♥♥ 'PPPPPPPPPPPP w♥♥♥♥♥♥♥♥♥♥♥♥ Z 'PPPPPPPPPPPPP w♥♥♥♥♥♥♥♥♥♥♥♥ 'PPPPPPPPPPPP &◆◆◆◆◆◆◆◆◆◆◆ x♣♣♣♣♣♣♣♣♣♣♣ #●●●●●●●●● X x♣♣♣♣♣♣♣♣♣♣♣ &◆◆◆◆◆◆◆◆◆◆◆ Y × X {①①①①①①①①① Y Y × Z × X Y × X The trace of a self-correspondence X ← Z → X is then calculated by the composition Y × X × X Y × Y × X Z ×Z×X Z xqqqqqqqqqqq &◆◆◆◆◆◆◆◆◆◆◆ xqqqqqqqqqqq Z {✇✇✇✇✇✇✇✇✇ #❋❋❋❋❋❋❋❋❋ Z × X Z &▼▼▼▼▼▼▼▼▼▼▼ x♣♣♣♣♣♣♣♣♣♣♣ &▼▼▼▼▼▼▼▼▼▼▼ #●●●●●●●●● {①①①①①①①①① X × X X × X X ❅❅❅❅❅❅❅❅ pt X ⑦⑦⑦⑦⑦⑦⑦⑦ pt Finally, the case of the graph Z = Γf of a self-map gives the fixed point locus by definition. (cid:3) Remark 4.2 (Cyclic version). The identification dim(X) ≃ LX above is naturally S1-equivariant for the standard loop rotation on LX and the cyclic symmetry of dim(X) provided by the cobordism hypothesis. To see this it is useful to consider X as an E∞-algebra object in Stacksop k via the diagonal map (or as an En-object for any n). In other words, for n = 1 we identify stacks and correspondences with objects and morphisms in the Morita category Alg(Stacksop k ). It follows from the properties of topological chiral homology [L2, Theorem 5.3.3.8] that for a (constant) commutative algebra A its topological chiral homology over a manifold is given by the tensoring of commutative algebras over simplicial sets RM A = M ⊗ A. In particular (passing back from the opposite category to stacks) we haveRS 1 X = X S 1 = LX. Moreover this identification holds not just for a fixed circle but over the moduli space of circles BDif f (S1) ∼ BS1, i.e. equivariantly for rotation. We also know from [L2, Example 5.3.3.14] or [L3, Example { #  # {  v ' x ' w & { & w ' x # x & { & x #  # x & {  NONLINEAR TRACES 27 4.2.2] that the S1-action on the dimension of an associative algebra A (given classically by the cyclic structure on the Hochschild chain complex) is given by the rotation S1-action on the topological chiral homology RS 1 A, i.e., the family of topological chiral homologies over the moduli space of circles. In our case this recovers the rotation action on the loop space. 4.3. Geometric functoriality of dimension. Proposition 4.3. The graph X ← Γf → Y of any proper morphism f : X → Y gives a continuous morphism F : X → Y in Corrprop , with right adjoint F r : Y → X identified with the opposite correspondence Y ← Γf → X. k Proof. We construct the unit and counit of the adjunction as follows. Consider the composition F rF : X → X of correspondences X ×Y X z✈✈✈✈✈✈✈✈✈ $❍❍❍❍❍❍❍❍❍❍ Y Γf ~⑥⑥⑥⑥⑥⑥⑥ X $❍❍❍❍❍❍❍❍❍ z✈✈✈✈✈✈✈✈✈✈ Γf ❆❆❆❆❆❆❆ X The unit ηf : idX = X → F rF ≃ X ×Y X is given by the relative diagonal map. Consider the opposite composition of correspondences X ×X X z✈✈✈✈✈✈✈✈✈ $■■■■■■■■■■ X Γf ~⑦⑦⑦⑦⑦⑦⑦ Y $❍❍❍❍❍❍❍❍❍ z✉✉✉✉✉✉✉✉✉✉ Γf ❅❅❅❅❅❅❅ Y The counit ǫf : F F r ≃ X → idY = Y is given by f itself. The standard identities are easily verified by identifying the resulting composite map Γf / Γf ×Y Γf ×X Γf / Γf of correspondences with the identity. (cid:3) Lemma 4.4. Let FZ : X → Y and FW : Y → X be morphisms in Corrk given by respective correspondences X ← Z → Y and Y ← W → X. Then the canonical equivalence m(FW , FZ ) : Tr(FW ◦ FZ ) ∼ / / Tr(FZ ◦ FW ) is given by the composition of evident geometric identifications (Z ×Y W ) ×X×X X ∼ / / W ×X×Y Z ∼ / / Z ×Y ×X W ∼ / / (W ×X Z) ×Y ×Y Y z $ ~ $ z z $ ~ $ z / / 28 DAVID BEN-ZVI AND DAVID NADLER Proof. Returning to the definition and using our previous identifications, observe that m(FZ , FW ) is calculated by commutativity of the diagram of correspondences X × X Z×X W ×X Y × X X × X ✽✽✽✽✽ ✽✽✽✽✽ pt X ;①①①①①①①①① #●●●●●●●●● Y ✽✽✽✽✽ X×W ✽✽✽✽✽ ✽ ✽✽✽✽✽ Y ×W ✽✽✽✽✽ ✽ Y × Y / X × Y W ×Y X×Y / Y × Y X #●●●●●●●●● ;①①①①①①①①① Y pt Following the top edge, we see Tr(FW ◦FZ ) ≃ (Z ×Y W )×X×X X. Following the bottom edge, we see Tr(FZ ◦FW ) ≃ (W ×X Z)×Y ×Y Y . Moving from the top to bottom edge via the successive equivalences of the three commuting squares, one finds the three successive equivalences in the assertion of the lemma. (cid:3) Proposition 4.5. Suppose f : X → Y is a proper morphism, and F : X → Y denotes the induced morphism in Corrprop given by the graph X ← Γf → Y . Then dim(F ) : dim(X) → dim(Y ) is canonically identified with the S1-equivariant morphism Lf : LX → LY . k Proof. Denote by F r : Y → X the right adjoint to F . We must calculate dim(X) / Tr(F rF ) m(F r,F ) / Tr(F F r) / dim(Y ) We have seen that the first and third morphisms correspond to the natural geometric maps LX ≃ X ×X×X X / (X ×Y X) ×X×X X X ×Y ×Y Y / Y ×Y ×Y Y ≃ LY induced by the relative diagonal X → X ×Y X and given map f : X → Y respectively. Furthermore, by Lemma 4.4, the middle map is the natural geometric identification (X ×Y X) ×X×X X ∼ / / X ×Y ×Y Y Altogether, the composition is easily identified with the loop map Lf : LX → LY . (cid:3) Remark 4.6. It follows from the proposition that the loop map Lf : LX → LY must be proper when the given map f : X → Y is proper. Let us note why this is true geometrically from the factorization LX → LX Y → LY appearing in the proof. First, the natural morphism LX → LX Y is the restriction along the diagonal X → X × X of the relative diagonal X → X ×Y X. The relative diagonal is a closed embedding since f is proper, and hence the natural morphism LX → LX Y is as well. Second, the natural morphism LX Y → LY is the restriction along the diagonal Y → Y ×Y of the proper morphism f : X → Y and thus is proper as well. Altogether, we see that Lf : LX → LY is itself proper. Remark 4.7. One can invoke the cobordism hypothesis with singularities to endow the mor- phism dim(F ) : dim(X) → dim(Y ) with a canonical S1-equivariant structure, and it will agree with the canonical geometric S1-equivariant structure on the map Lf : LX → LY under the identification of the proposition. / /  / /  # ; # / / ; / / / / / NONLINEAR TRACES 29 4.4. Geometric functoriality of trace. Consider a proper morphism f : X → Y and endo- morphisms FZ : X → X and FW : Y → Y in Corrk given by respective self-correspondences X ← Z → X and Y ← W → Y . By an f -morphism from the pair (X, FZ ) to the pair (Y, FW ), we mean an identification s : Z ∼ / / X ×Y W of correspondences from X to Y . This in turn induces an identification of what might be called relative traces Z ×Y ×Y Y ∼ / / X ×Y ×Y W generalizing the relative loop space LX Y from the case of the identity correspondences Z = X, W = Y . We thus obtain a map of traces τ (f, s) : Z∆X = Z ×X×X X / Z ×Y ×Y Y ∼ / / X ×Y ×Y W / Y ×Y ×Y W = W ∆Y Proposition 4.8. With the preceding setup, the trace map Tr(f, s) : Tr(FZ ) → Tr(FW ) is canonically identified with the geometric map τ (f, s) : Z∆X / W ∆Y Proof. Denote by F : X → Y the morphism given by the graph X ← Γf → Y , and by F r : Y → X its right adjoint. We must calculate Tr(FZ ) / Tr(F rF FZ ) s / Tr(F rFW F ) m(F r,FW F ) / Tr(FW F F r) / Tr(FW ) We have seen that the first and fourth morphisms correspond to the natural geometric maps Z∆X = Z ×X×X X / Z ×X×X (X ×Y X) X ×Y ×Y W / Y ×Y ×Y W = W ∆Y induced by the relative diagonal X → X ×Y X and given map f : X → Y respectively. Using associativity, the second map, induced by s, is the natural geometric identification Z ×X×X (X ×Y X) ≃ Z ×Y ×Y Y ∼ / / W ×Y ×Y X By Lemma 4.4, the third map, given by the cyclic symmetry, is nothing more than the natural identification W ×Y ×Y X ∼ / / X ×Y ×Y W Thus assembling the above maps we arrive at the composition defining τ (f, s). (cid:3) 5. Traces for sheaves In this section, we spell out how to apply the abstract formalism of traces of Section 3 and its geometric incarnation of Section 4 to categories of sheaves. Recall Definition 2.15: Definition 5.1. A sheaf theory is a symmetric monoidal functor of (∞, 2)-categories S : Corrprop k / dgCat k from correspondences of stacks (with 2-morphisms given by proper maps of correspondences) to dg categories. We denote by S : Corrk / dgCatk the underlying 1-categorical sheaf theory, i.e. categories obtained by forgetting noninvertible morphisms. the symmetric monoidal functor on (∞, 1)- / / / / / / / / / / / 30 DAVID BEN-ZVI AND DAVID NADLER Applying a sheaf theory to the geometric descriptions of traces of correspondences, one immediately deduces trace formulas for dg categories. We first spell out the consequences of the 1-categorial structure of a sheaf theory S, then the trace formulae arising from its 2-categorical enhancement S, and finally in Section 5.3 explain how to use the results of [DG, GR2] to deduce applications of this formalism. 5.1. Dimensions and traces of sheaf categories: 1-categorical consequences. The graph of a map of derived stacks f : X → Y provides a correspondence from X to Y and a correspondence from Y to X. We denote the respective induced maps by f∗ : S(X) → S(Y ) and f ! : S(Y ) → S(X). The functoriality of S concisely encodes base change for f∗ and f !. For π : X → pt = Spec k, we denote by ωX = π!k ∈ S(X) the S-analogue of the dualizing sheaf, and by ω(X) = π∗ωX ∈ S(pt) = dgV ectk the S-analogue of "global volume forms". Next we will record formal consequences of our prior calculations deduced from the fact that a sheaf theory is symmetric monoidal. Proposition 5.2. Fix a sheaf theory S : Corrk → dgCatk, and X, Y ∈ Corrk. (1) S(X) ∈ dgCatk is canonically self-dual, and for any f : X → Y , f ! : S(Y ) → S(X) and f∗ : S(X) → S(Y ) are canonically transposes of each other. (2) S(X) is canonically symmetric monoidal with tensor product F ⊗! G = ∆!(π! 1F ⊗ π! 2G) F , G ∈ S(X) (3) For any f : X → Y , the projection formula holds: f∗F ⊗! G ≃ f∗(F ⊗! f !G) F ∈ S(X), G ∈ S(Y ) (4) There is a canonical equivalence of functors and integral kernels (5) The functor q∗p! : S(X) → S(Y ) associated to a correspondence HomdgCatk (S(X), S(Y )) ≃ S(X × Y ) X p Z q / Y is represented by the integral kernel (p × q)∗ωZ ∈ S(X × Y ). Proof. (1) Follows immediately from Proposition 4.1. (2) Follows immediately from the commutative algebra structure on X ∈ Corrk (in fact commutative coalgebra structure on X ∈ Stacksk) provided by the diagonal map. (3) Follows from base change for the diagram f X id×f / Y ∆ X × Y f ×id / Y × Y (4) Since S is monoidal, we have S(X) ⊗ S(Y ) ≃ S(X × Y ) The self-duality of S(X) provides HomdgCatk (S(X), S(Y )) ≃ S(X)∨ ⊗ S(Y ) ≃ S(X) ⊗ S(Y ) By construction, the composite identification assigns the functor FK (F ) = π2∗ (π! 1F ⊗! K) K ∈ S(X × Y ) o o /   /   / (5) Follows from the projection formula: consider the diagram NONLINEAR TRACES Z p {✇✇✇✇✇✇✇✇✇ π1 Π q #●●●●●●●●● π2 where Π = p × q. Then we have X X × Y / Y q∗p!(−) ≃ π2∗Π∗Π!π! 1(−) ≃ π2∗Π∗(ωZ ⊗! Π!π! 1(−)) ≃ π2∗(Π∗ωZ ⊗! π! 1(−)) 31 (cid:3) Proposition 5.3. Fix a sheaf theory S : Corrk → dgCatk. (1) The S-dimension dim(S(X)) = HH∗(S(X)) of any X ∈ Corrk is S1-equivariantly equivalent with S-global volume forms on the loop space dim(S(X)) ≃ ω(LX). In particular, for G an affine algebraic group, characters of S-valued G-representations are adjoint-equivariant S-global volume forms dim(S(BG)) ≃ ω(G/G) (2) The S-trace of any endomorphism Z ∈ Corrk(X, X) is equivalent to S-global volume forms on the restriction to the diagonal Tr(S(Z)) ≃ ω(Z∆) In particular, the S-trace of a self-map f : X → X is equivalent to S-global volume forms on the f -fixed point locus Tr(f∗) ≃ ω(X f ) Proof. (1) Follows immediately from Proposition 4.1(1). To spell this out, using the previous proposition and base change, dim(S(X)) results from applying the composition π∗∆!∆∗π! ≃ π∗p2∗p! 1π! ≃ Lπ∗Lπ! : dgV ectk / dgV ectk to the unit 1dgV ectk = k. Here π : X → pt and Lπ : Lx → pt are the maps to the point, and p1, p2 : LX ≃ X ×X×X X → X are the two natural projections. Thus we find dim(S(X)) ≃ Lπ∗Lπ!(k) ≃ ω(LX). Furthermore, the S1-equivariance results from the one- dimensional cobordism hypothesis: the one-dimensional topological field theory defined by the dualizable object S(X) ∈ dgCatk factors through that defined by the dualizable object X ∈ Corrk. Moreover, we identified the S1-action on the dimension LX with loop rotation.6 (cid:3) (2) Similarly follows immediately from Proposition 4.1(2). 5.2. Integration formulas for traces: 2-categorical consequences. Now we turn to the functoriality of dimensions and traces. For this we require the 2-categorical enhanced version S of a sheaf theory, so as to take advantage of the resulting functorial adjunction S(X) f∗ f ! / S(Y ) for f : X → Y proper. In particular, applying the functor S on two-morphisms we find that 6One can also check directly that the cyclic structure on the cyclic bar construction of the dg category S(X) is induced by the cyclic structure of the loop space LX under the identification ω(LX) ≃ dim(S(X)). { #   o o / / / o o 32 DAVID BEN-ZVI AND DAVID NADLER • A proper map f : Z → W of correspondences X pZ ~⑥⑥⑥⑥⑥⑥⑥⑥ `❇❇❇❇❇❇❇❇ pW Z W f qZ ❆❆❆❆❆❆❆❆ >⑥⑥⑥⑥⑥⑥⑥⑥ qW Y induces a canonical integration morphism of integral transforms • In particular, when X = Y = pt, it induces a map of global volume forms Z Rf : qZ∗p! Rf : ω(Z) / qW ∗p! W / ω(W ) • There is a canonical composition identity Rg ◦Rf ≃Rg◦f Proposition 5.4. For f : X → Y proper, the unit and counit of the (f∗, f !) adjunction are given respectively by integration along the proper maps of self-correspondences ∆/Y : X → X ×Y X of X and f/Y : X → Y of Y : X z✉✉✉✉✉✉✉✉✉✉ d❍❍❍❍❍❍❍❍❍ p1 X ×Y X X ∆f $■■■■■■■■■■ :✈✈✈✈✈✈✈✈✈ p2 X Y f ~⑦⑦⑦⑦⑦⑦⑦⑦ _❅❅❅❅❅❅❅❅ Y X f f ❅❅❅❅❅❅❅❅ ?⑦⑦⑦⑦⑦⑦⑦⑦ Y Proof. The assertion follows immediately from the geometric description of the unit and counit in the correspondence category, Proposition 4.3, upon applying the functor S. (cid:3) Proposition 5.5. For any proper map f : X → Y , the induced map on dimensions is identified (S1-equivariantly) with integration along the loop map dim(f∗) : dim(S(X)) / dim(S(Y )) dim(f∗) ≃RLf : ω(LX) / ω(LY ) Proof. According to Definition 3.19, we must calculate the composition dim(f∗) : Tr(idS(X)) Tr(ηf ) / Tr(f !f∗) ∼ / / Tr(f∗f !) Tr(ǫf ) / Tr(idS(Y )) The equivalence of the middle arrow is given by the canonical identifications Tr(f !f∗) ≃ ω((X ×Y X) ×X×X X) ≃ ω(X ×Y ×Y Y ) ≃ Tr(f∗f !) By Proposition 5.4, the unit ηf : idS(X) → f !f∗ is given by the integration morphism R∆f : ∆f ∗ωX / ωX×Y X ~   ` > / / z   $ ~   d : _ ? / / / / / and hence its trace Tr(ηf ) : Tr(idS(X)) → Tr(f !f∗) is given by the induced integration map NONLINEAR TRACES 33 R∆f : ω(LX) / ω((X ×Y X) ×X×X X) Likewise, the counit ǫf : f∗f ! → idS(Y ) is given by by the integration morphism and hence its trace Tr(ǫf ) : Tr(f∗f !) → T r(idS(Y )) is given by the induced integration map / ωY / ω(LY ) Rf : f∗ωX Rf : ω(X ×Y ×Y Y ) RLf : ω(LX) Finally, by functoriality, their composition is given by the integration map / ω(LY ) (cid:3) Finally, we have the functoriality of traces in parallel with the previous theorem on the functoriality of dimensions. Let us recall the relevant setup. Consider a proper morphism f : X → Y and endomorphisms FZ : X → X and FW : Y → Y in Corrk given by respective self-correspondences X ← Z → X and Y ← W → Y . By an f -morphism from the pair (X, FZ ) to the pair (Y, FW ), we mean an identification s : Z ∼ / / X ×Y W of correspondences from X to Y . This in turn induces an identification of what might be called relative traces Z ×Y ×Y Y ∼ / / X ×Y ×Y W generalizing the relative loop space LX Y from the case of the identity correspondences Z = X, W = Y . We thus obtain a map of traces τ (f, s) : Z∆X = Z ×X×X X / Z ×Y ×Y Y ∼ / / X ×Y ×Y W / Y ×Y ×Y W = W ∆Y Proposition 5.6. With the preceding setup, the trace map Tr(f∗, s) : Tr(FX∗) → Tr(FY ∗) is canonically identified with the integration map Rτ (f,s) : ω(Z∆X ) / ω(W ∆Y ) Proof. The argument is parallel to the proof of Proposition 5.5. One calculates Tr(f∗, α) from Definition 3.24 using Proposition 4.8 and the compatibility of Proposition 5.2 and the integra- tion morphism for integral transforms from Section 2.3.2. (cid:3) 5.3. Ind-coherent sheaves and D-modules. We now apply the results of Gaitsgory-Rozenblyum [GR2] and Drinfeld-Gaitsgory [DG] establishing functoriality properties of categories of ind-coherent sheaves and D-modules. We first state a fundamental result of Gaitsgory-Rozenblyum [GR2]: Theorem 5.7. [GR2, Theorem III.3.5.4.3, III.3.6.3] There is a uniquely defined right-lax symmetric monoidal functor Q! from the (∞, 2)-category whose objects are laft (locally almost of finite type) prestacks, morphisms are correspondences with vertical arrow ind-inf-schematic, and 2-morphisms are ind-proper and ind-inf-schematic, to the (∞, 2) category dgCat of k-linear presentable dg categories with continuous morphisms. The functor Q! is strictly symmetric monoidal on the full subcategory of laft ind-inf-schemes. k / / / / / / / 34 DAVID BEN-ZVI AND DAVID NADLER The theorem encodes a tremendous amount of structure in great generality. Let us highlight some salient features useful in practice. The theorem assigns a symmetric monoidal dg category Q!(X) to any stack satisfying a reasonable finite type assumption. The symmetric monoidal structure, the !-tensor product, is induced by !-pullback along diagonal maps. For an arbitrary morphism p : X → Y there is a continuous symmetric monoidal pullback functor f ! : Q!(Y ) → Q!(X), while for f schematic (or ind-schematic) there is a continuous pushforward f∗ : Q!(X) → Q!(Y ), which satisfies base change with respect to !-pullbacks. Moreover for f proper (or ind- proper), (f∗, f !) form an adjoint pair. The theorem goes much further through the powerful formalism of inf-schemes: prototypical inf-schemes are quotients of schemes by infinitesimal equivalence relation. Thus one can treat on an equal footing ind-coherent sheaves that are equivariant for any formal groupoid. The most important example is the de Rham space XdR of a scheme, and one recovers D-modules on X as ind-coherent sheaves on the de Rham functor of X, D(X) = Q!(XdR). Thus by first applying the functor (−)dR the theorem encodes the theory of D-modules, as a functor out of the correspondence 2-category of stacks (or laft prestacks) with pullbacks for arbitrary maps and pushforward for (ind-)schematic maps. Corollary 5.8. The functors Q! and D define sheaf theories on laft ind-inf-schemes, i.e., define symmetric monoidal functors Q!, D : Corr(ind − inf − Schk)ind−prop −→ dgCat . k Thus the conclusions of Theorem 1.4 apply in this setting (in particular the Grothendieck- Riemann-Roch theorem for ind-proper maps of ind-inf-schemes). 5.3.1. The QCA setting. We are mostly interested in applications of sheaf theory on stacks, e.g. in an equivariant setting. That requires two features of the theories Q! and D that are not encoded in Theorem 5.7. Theorem 5.7 produces in general a right-lax symmetric monoidal functor -- in other words, we have the natural map Q!(X) ⊗ Q!(Y ) −→ Q!(X × Y ) satisfying the expected coherences, but it is not an equivalence in general (though it is for schemes). Also, while the theorem encodes arbitrary pullbacks, it does not encode a continuous push- forward functor p∗ : Q!(X) → Q!(Y ) for non-schematic morphisms (though a generalization to include QCA morphisms has been announced by the authors). This precludes an immediate application of our formalism to traces on stacks. However, since the full structure of a sheaf theory is far stronger than is needed for the "local" statements we discuss in this paper, we can get around this issue, by taking advantage of the following compilation of results of Drinfeld and Gaitsgory (specifically, see Section 3.6.1, Corollarys 3.7.14, 4.2.3, and 4.4.7, Proposition 4.4.11, Corollarys 8.3.4 and 8.4.3, Definition 9.3.2 and Proposition 9.3.12). Theorem 5.9. [DG] (1) For a QCA stack X, the categories Q!(X) and D(X) are dualizable and canonically self-dual. (2) The canonical functors define equivalences Q!(X) ⊗ Q!(Y ) ≃ Q!(X × Y ), D(X) ⊗ D(Y ) ≃ D(X × Y ) NONLINEAR TRACES 35 (3) For a morphism f : X → Y of QCA stack, the "renormalized pushforwards"7 f• : Q!(X) → Q!(Y ), f• : D(X) → D(Y ) defined as the transpose of f ! (by the self-duality of (1)) are continuous functors satis- fying base-change and the projection formula with respect to pullback. Let us now briefly indicate how the results of the previous two sections carry over to QCA stacks in the absence of a fully fledged sheaf theory, where we use the renormalized pushforward functors f• provided by Theorem 5.9 to carry out non-representable pushforwards. In particular, for X a general QCA stack, so that piX : X → pt is not representable, this means that the notation ω(X) has to be taken in a renormalized fashion, ω(X) := πX,•π! X k = πX,•ωX . For the theory of ind-coherent sheaves this produces the usual notion of derived global sections of the dualizing complex, but for the theory of D-modules this will differ in general from the nonrenormalized version, namely Borel-Moore chains on X: ω(X)non−renorm = RΓdR(ωX ) = CBM ∗ (X). In Proposition 5.2, the self-duality in assertion (1) for QCA stacks is the content of Theo- rem 5.9(1), and f• is defined so as to make it the transpose of f !. Assertion (2) follows from the sheaf theory construction (i.e. is independent of non-representable morphisms), the projection formula is asserted in item (3) of the theorem and the last two assertions are deduced from the first three. Proposition 5.3 is also deduced directly from Proposition 5.2. The general trace construction Rf discussed in Section 5.2 depends only on the functoriality of proper adjunction (which is part of the [GR2] formalism) and the definition of pullback and (renormalized) pushforward. The identities needed to verify Propositions 5.4, 5.5 and 5.6 only depend on base change, which is guaranteed by Theorem 5.9. We therefore get as a payoff that the conclusions of Theorem 1.4 hold for QCA stacks, in particular: Theorem 5.10. Let S = Q! or S = D denote either ind-coherent sheaves or D-modules. Let f : X → Y denote a proper morphism of QCA stacks. • For any compact object M ∈ S(X) (coherent sheaf or safe coherent D-module) with char- acter [M ] ∈ HH∗(S(X)) ≃ ω(LX), there is a canonical identification [f∗M] ≃ZLf [M ] ∈ HH∗(S(Y )) ≃ ω(LY ) • Assume Y = BG for an affine group and X = Z/G for Z a proper QCA stack. Then for any compact object M ∈ S(Z/G) (G-equivariant coherent sheaf or safely equivariant coherent D-module on X), and element g ∈ G, there is a canonical identification [f∗M ]g ≃ZLf [M ]X g • For a map f : (X, Z) → (Y, W ) of QCA stacks with self-correspondences, the induced map T r(S(Z)) → T r(S(W )) is given by integration along fixed points Z∆X → W∆Y . 7We note that for Q! the "renormalized" pushforward is the standard pushforward functor, while for D it differs for non-safe objects from the more familiar, but discontinuous, de Rham pushforward. 36 DAVID BEN-ZVI AND DAVID NADLER References [Ba] C. Barwick, On the Q-construction for exact ∞-categories. e-print arXiv:1301.4725. [BFN] D. Ben-Zvi, J. Francis and D. Nadler, Integral transforms and Drinfeld centers in derived algebraic geometry. Preprint arXiv:0805.0157. Jour. Amer. Math. Soc. 23 (2010), 909 -- 966. [BN09] D. Ben-Zvi and D. Nadler, The character theory of a complex group. e-print arXiv:math/0904.1247 [BN10b] D. Ben-Zvi and D. Nadler, Loop Spaces and Connections. arXiv:1002.3636. J. Topol. 5 (2012), no. 2, 377 -- 430. [BN10b] D. Ben-Zvi and D. Nadler, Loop Spaces and Representations. arXiv:1004.5120. Duke Math. J. 162 (2013), no. 9, 1587 -- 1619. [BN13] D. Ben-Zvi and D. Nadler, Secondary Traces. e-print arXiv:1305.7177. [Cal1] A. Caldararu and S. Willerton, The Mukai pairing. I. A categorical approach. arXiv:math/0308079. New York J. Math. 16 (2010), 61 -- 98. [Cal2] A. Caldararu, The Mukai pairing, II: The Hochschild-Kostant-Rosenberg isomorphism, Adv. Math. 194 (2005), no. 1, 3466, arXiv:math/0308080v3. [CT] D.-C. Cisinski and G. Tabuada, Lefschetz and Hirzebruch-Riemann-Roch formulas via noncommutative motives. arXiv:1111.0257. J. Noncommut. Geom. 8 (2014), no. 4, 1171 -- 1190. [DP] A. Dold and D. Puppe, Duality, trace, and transfer. Proceedings of the International Conference on Geometric Topology (Warsaw, 1978), pp. 81 -- 102, PWN, Warsaw, 1980. [DG] V. Drinfeld and D. Gaitsgory, On some finiteness questions for algebraic stacks. arXiv:1108.5351. Geom. Funct. Anal. 23 (2013), no. 1, 149 -- 294. [EKMM] A. Elmendorf, I. Kriz, M. Mandell and J.P. May, Rings, modules, and algebras in stable homotopy theory. With an appendix by M. Cole. Mathematical Surveys and Monographs, 47. American Mathematical Society, Providence, RI, 1997. [FG] J. Francis and D. Gaitsgory, Chiral Koszul Duality. Selecta Math. (N.S.) 18 (2012), no. 1, 27 -- 87. [FHLT] D. Freed, M. Hopkins, J. Lurie and C. Teleman, Topological Quantum Field Theories from Compact Lie Groups. arXiv:0905.0731. A celebration of the mathematical legacy of Raoul Bott, 367 -- 403, CRM Proc. Lecture Notes, 50, Amer. Math. Soc., Providence, RI, 2010. [FN] E. Frenkel and Ngo B.-C., Geometrization of trace formulas. Bull. Math. Sci. 1 (2011), no. 1, 129-199. [G1] D. Gaitsgory, Ind-coherent sheaves. arXiv:math/1105.4857 Mosc. Math. J. 13 (2013), no. 3, 399 -- 528, 553. [G2] D. Gaitsgory, Sheaves of categories and the notion of 1-affineness. Stacks and categories in geometry, topology, and algebra, 127 -- 225, Contemp. Math., 643, Amer. Math. Soc., Providence, RI, 2015. [GR1] D. Gaitsgory and N. Rozenblyum, Crystals and D-modules. Pure Appl. Math. Q. 10 (2014), no. 1, 57 -- 154. [GR2] D. Gaitsgory and N. Rozenblyum, A Study in Derived Algebraic Geometry. Draft available at http://math.harvard.edu/gaitsgde/GL/. Mathematical Surveys and Monographs 221, American Math- ematical Society, 2017. [H1] R. Haugseng, Iterated spans and "classical" topological field theories. arXiv:1409.0837. [H2] R. Haugseng, The higher Morita category of En-algebras. e-print arXiv:1412.8459. Geom. Topol. 21 (2017), no. 3, 1631 -- 1730. [HSS] M. Hoyois, S. Scherotzke and N. Sibilla, Higher traces, noncommutative motives, and the categorified Chern character. e-print arXiv:1511.03589. Adv. Math. 309 (2017), 97?154. [Jo] R. Joshua, Riemann-Roch for algebraic stacks. I. Compositio Math. 136 (2003), no. 2, 117 -- 169. [J] D. Joyce, An introduction to d-manifolds and derived differential geometry. Moduli spaces, 230 -- 281, London Math. Soc. Lecture Note Ser., 411, Cambridge Univ. Press, Cambridge, 2014. [Ke] B. Keller, On differential graded categories, International Congress of Mathematicians. Vol. II, 151-190, Eur. Math. Soc., Zurich, 2006. [KP] G. Kondyrev and A. Prihodko, Categorical proof of Holomorphic Atiyah-Bott formula. e-print arXiv:1607.06345 [Lo] J.-L. Loday, Cyclic homology. Appendix E by Mar´ıa O. Ronco. Second edition. Chapter 13 by the author in collaboration with Teimuraz Pirashvili. Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], 301. Springer-Verlag, Berlin, 1998. [Lu] V. Lunts, Lefschetz fixed point theorems for Fourier-Mukai functors and DG algebras. arXiv:1102.2884. J. Algebra 356 (2012), 230 -- 256. [L1] J. Lurie, Higher topos theory. arXiv:math.CT/0608040. Annals of Mathematics Studies, 170. Princeton University Press, Princeton, NJ, 2009 [L2] J. Lurie, Higher Algebra. Available at http://www.math.harvard.edu/lurie/ NONLINEAR TRACES [L3] J. Lurie, On the classification of topological http://www.math.harvard.edu/lurie/ Current developments Press, Somerville, MA, 2009. field Available in mathematics, 2008, 129-280, theories. 37 at Int. [L4] J. Lurie, Spectral Algebraic Geometry. Preprint, available at http://www.math.harvard.edu/lurie/ [Ma] N. Markarian, The Atiyah class, Hochschild cohomology and the Riemann-Roch theorem. arXiv:math/0610553. J. Lond. Math. Soc. (2) 79 (2009), no. 1, 129 -- 143. [M] J. P. May, Picard groups, Grothendieck rings, and Burnside rings of categories. Adv. Math. 163 (2001), no. 1, 1 -- 16. [Pe] F. Petit, A Riemann-Roch theorem for DG algebras, arXiv:1004.0361. Bull. Soc. Math. France 141 (2013), no. 2, 197 -- 223. [Po] A. Polishchuk, Lefschetz type formulas for dg-categories arXiv:1111.0728. Selecta Math. (N.S.) 20 (2014), no. 3, 885 -- 928. [PS] K. Ponto and M. Shulman, Traces in symmetric monoidal categories. arXiv:1107.6032 [P] A. Preygel, Thom-Sebastiani and duality for matrix factorizations. arXiv:math/1101.5834 [Ram] A. C. Ramadoss, The relative Riemann-Roch theorem from Hochschild homology, New York J. Math. 14 (2008) 643-717. [Ram2] A. C. Ramadoss, The Mukai pairing and integral transforms in Hochschild homology, Moscow Math. Journal, Vol 10, No 3, (2010), 629-645. [Sh] B. Shipley, Symmetric spectra and topological Hochschild homology. K-Theory 19 (2000), no. 2, 155 -- 183. [Shk] D. Shklyarov, Hirzebruch-Riemann-Roch theorem for DG algebras, arXiv:0710.1937v3. Proc. Lond. Math. Soc. (3) 106 (2013), no. 1, 1 -- 32. [Th] R.W. Thomason, Lefschetz-Riemann-Roch theorem and coherent trace formula. Invent. Math. 85 (1986), no. 3, 515 -- 543. [To] B. Toen, The homotopy theory of dg categories and derived Morita theory. arXiv:math.AG/0408337. Invent. Math. 167 (2007), no. 3, 615 -- 667. [To2] B. Toen, Higher and Derived Stacks: a global overview. Algebraic geometry, Seattle 2005. Part 1, 435 -- 487, Proc. Sympos. Pure Math., 80, Part 1, Amer. Math. Soc., Providence, RI, 2009. [TV] B. Toen and G. Vezzosi, Caract`eres de Chern, traces ´equivariantes et g´eom´etrie alg´ebrique d´eriv´ee. Selecta Math. (N.S.) 21 (2015), no. 2, 449 -- 554. Department of Mathematics, University of Texas, Austin, TX 78712-0257 E-mail address: [email protected] Department of Mathematics, University of California, Berkeley, CA 94720-3840 E-mail address: [email protected]
1504.03853
2
1504
2018-01-11T16:16:28
Stability of restrictions of cotangent bundles of irreducible Hermitian symmetric spaces of compact type
[ "math.AG" ]
It is known that the cotangent bundle $\Omega_Y$ of an irreducible Hermitian symmetric space $Y$ of compact type is stable. Except for a few obvious exceptions, we show that if $X \subset Y$ is a complete intersection such that $Pic(Y) \to Pic(X)$ is surjective, then the restriction $\Omega_{Y|X}$ is stable. We then address some cases where the Picard group increases by restriction.
math.AG
math
STABILITY OF RESTRICTIONS OF THE COTANGENT BUNDLE OF IRREDUCIBLE HERMITIAN SYMMETRIC SPACES OF COMPACT TYPE INDRANIL BISWAS, PIERRE-EMMANUEL CHAPUT, AND CHRISTOPHE MOUROUGANE ABSTRACT. It is known that the cotangent bundle ΩY of an irreducible Hermitian symmetric space Y of compact type is stable. We show that if X ⊂ Y is a subvariety whose structure sheaf has a short split resolution and such that the restriction map Pic(Y ) → Pic(X ) is surjective, then, apart from a few exceptions, the restriction ΩY X is stable. We then address some cases where the Picard group increases by restriction. CONTENTS Introduction 1. 2. Vanishing theorems 3. Restrictions with small Picard group 4. Restriction to a hypersurface with an increase of the Picard group References 1 3 12 15 22 1. INTRODUCTION Throughout this article, by (semi-)stability of a vector bundle we mean slope (semi-)stability with respect to some fixed polarization. The stable vector bundles with zero characteristic classes on a smooth projective variety are given by the irreducible unitary representations of the fundamental group of the variety. But, in the study of moduli spaces, it is a difficult and interesting question to produce explicit examples of stable vector bundles on algebraic varieties with non-zero charac- teristic classes. There are such vector bundles within the framework of homogeneous spaces: for example, an irreducible homogeneous bundle on a homogeneous space is stable with respect to any polarization [Ume78], [Ram66], [Bis04]. Once one has these stable vector bundles, theorems of Mehta-Ramanathan [MR84], Flenner [Fle84] and Langer [Lan04] assert the stability of the restriction of these bundles to a general hypersurface of high enough degree with respect to the given polariza- tion. As, Balaji and Kollár [BK12, question 4], asked if, having chosen a very ample polarization, a stable reflexive sheaf on a normal projective surface restricts to a stable bundle on a general curve of degree 4 or more, we in fact expect quite small optimal bounds on the degree. In this article, we address the following general questions: • Produce examples of large rank stable vector bundles on Fano manifolds of small degree. • More specifically, what can be said about the stability of the restriction of an irreducible ho- mogeneous bundle E, defined on a homogeneous space Y , to a subvariety X ⊂ Y with em- phasis on giving good bounds on the degree and a description of the required features of the general subvarieties ? We study these questions in the following setting. Recall that a Hermitian symmetric space is a Hermitian manifold in which every point is an isolated point of an isometric involution. It is ho- mogeneous under its isometry group. It is called irreducible if furthermore it cannot be written as the non trivial product of two Hermitian symmetric spaces. This property is equivalent to the fact Date: July 4, 2021. 2010 Mathematics Subject Classification. 32M15, 32Q26 (primary) 14J60, 14F17, 14N25 (secondary). Key words and phrases. Hermitian symmetric space, cotangent bundle, slope-stability, stability under restriction, van- ishing theorems for coherent cohomology, projective geometry. 1 that the isometry group is almost simple. Compact examples of irreducible Hermitian symmetric spaces consist of the usual Grassmannians, the quadric hypersurfaces1, the Lagrangian Grassman- nians parameterizing n-dimensional Lagrangian subspaces of C2n equipped with a symplectic form, the spinor Grassmannian parameterizing one family of n-dimensional isotropic subspaces of C2n equipped with a non degenerate quadratic form, and two exceptional manifolds. The 2-dimensional quadric has Picard number 2 and we consider its anti-canonical polariza- tion. All other examples have Picard number one, so that we do not have to specify the choice of a polarization. We will denote the ample generator of the Picard group of Y by OY (1). We will denote by deg Y the top degree self-intersection deg Y := OY (1)dim Y and by c1(Y ) the index of the Fano manifold Y . We recall that c1(Y ) is defined by the equality −KY = c1(Y )OY (1) as elements of the Néron-Severi group of Y . The degree degY X of a subvariety X of Y is defined such that OY (1)dim X = degY X deg Y . In particular, if X is given as the complete intersection of divisors in X Qc i =1di . i =1 OY (di ) then degY X = Πc We assume first that Y is a compact irreducible Hermitian symmetric space. Its cotangent bundle ΩY is an irreducible homogeneous, hence stable, vector bundle. We choose for X a locally factorial positive dimensional subvariety whose structure sheaf has a short split resolution (see Definition 3.1) and such that the restriction Pic(Y ) → Pic(X ) is surjective. By Lefschetz theorem [Lef21], this holds whenever X is a complete intersection of dimension dim X ≥ 3 (see also [Laz04, Example 3.1.25]) or whenever X is a very general complete intersection surface in Pn except if it is a degree d ≤ 3 surface in P3 or the intersection of two quadric threefolds in P4 (see also [Kim91, Theorem 1]). Theorem A (Theorem 2 and Theorem 3). Let Y be a compact irreducible Hermitian symmetric space, and let X be a locally factorial positive dimensional subvariety whose structure sheaf has a short split resolution in Y . Assume that the restriction homomorphism Pic(Y ) → Pic(X ) is surjective. If Y is a projective space or a quadric, assume moreover that X has no linear equation. Then, the restriction of ΩY to X is stable. Using a relative Harder-Narasimhan filtration, if Y be a compact irreducible Hermitian symmetric space, and X a general complete intersections, we infer the semi-stability of restriction of ΩY to X without assuming that the Picard group of X comes from that of Y (see Theorem 4). We then deal with the case of small dimensions where the Picard group increases by restriction. Recall that irreducible Hermitian symmetric spaces of dimension 2 or 3 are P2, P3 and Q3. We add the similar case of Q2 for completeness. Theorem B (Theorem 4, Theorem 5). Let Y be an irreducible compact Hermitian symmetric space of dimension 2 or 3. Let X ⊂ Y be a smooth divisor, and in the case of ambient dimension 3, let C be a complete intersection curve. • Take Y = P2. If degY X ≥ 2, then ΩY X is semi-stable and if degY X ≥ 3, then ΩY X is stable. • Take Y = P3. If degY X ≥ 2, then ΩY X is stable . If C is the complete intersection of two general non-linear hypersurfaces, then ΩY C is semi-stable. • Take Y = Q3. If degY X = 1, then ΩY X is semi-stable and if degY X ≥ 2, then ΩY X is stable. If C is the complete intersection of two general non-linear hypersurfaces, then ΩY C is semi- stable. Note in particular that we have a complete answer to the question of stability of the restriction to divisors: Corollary C. Let Y be an irreducible compact Hermitian symmetric space. Let X ⊂ Y be a smooth divisor, not of degree 1. Then ΩY X is stable with the only exception of a smooth conic in P2. Our arguments are very different in the two situations of theorem A and B. In the case of no in- crease of Picard groups, we use a new vanishing theorem that may be of independent interest: 1To deal with all quadrics, we handle the case of the two dimensional quadric, although it is reducible. 2 Theorem D (Theorem 1). Let Y be a compact irreducible Hermitian symmetric space, but not a pro- jective space. Let l , p, q be integers, with l > 0, p > 0, such that H q (Y , Ω p Y (l )) 6= 0. Then, If moreover q > 0, then l + q ≥ p c1(Y ) dim(Y ) . l + q ≤ p. The proof of this theorem relies on a combinatorial study of the Bott vanishing theorem [Bot57]. Note that by Akizuki-Kodaira-Nakano theorem [AN54], we have p + q ≤ dim Y under the assumption p H q(Y , Ω Y (l )) 6= 0 of the theorem. On the other hand, by Serre's asymptotic vanishing theorem, we also expect in the case q > 0, an upper bound on the value of l . Finally, by the work of Snow [Sno86, Sno88], we know that q ≤ p. However, to the best of our knowledge, it is the first time that the idea that a bound on the sum l +q could hold appears, addressing the question whether a similar bound could hold for some more general varieties Y . It should be mentioned that a somewhat related vanishing theorem was proved in [D-F04]. The connection between vanishing results and stability is as follows: by standard cohomological arguments, as used for example in [PW95], a subbundle F ⊂ ΩY X of the restriction of ΩY contradict- ing stability yields, under the assumption of the existence of a short resolution of OX by split vector bundles on Y , the non-vanishing of some cohomology group H q (Y , Ωp (l )), where q is related to the codimension of X and l to the degree of F . Our vanishing theorem implies the desired stability in- equality (see Section 3). Remarkably enough, the bounds in Theorem D are exactly the bounds we need to establish the stability inequality. In small dimensions, we use tools from projective geometry to make explicit the new line bundles that appear on the subvariety X . We found it noteworthy that the precise stability inequalities ulti- mately originate in some Bezout like theorems. Those cases with increase of the Picard group seem untractable either through vanishing theorems, or through the technique of covering rational curves as in [Hwa98]. In a particular case, we need some arguments of representation theory (see Section 4). Our initial motivation for the present work is its implication in terms of height inequalities: for example, starting with a manifold W with ample canonical bundle KW (e.g. an hypersurface of large degree in a projective manifold Y of Picard number one), if the restriction of its cotangent bundle ΩW to a curve C ⊂ X is semi-stable then, as ΩC is a quotient of ΩW C , the slope inequality gives a bound on the height of C with respect to the canonical polarization KW in terms of the genus of C . We hope to push further the present study of ΩY C to deal with the stability properties of ΩW C . Acknowledgements. We thank Frédéric Han and Laurent Gruson for their help in projective geome- try. We thank Michel Brion for pointing out a mistake in a previous version of this paper. We thank Jie Liu for pointing out the improvement in the use of Langer's technique for Q3. The first-named author is supported by a J. C. Bose Fellowship. Part of this project was done in the TATA Institute (Bombay); we thank it for hospitality. 2. VANISHING THEOREMS Let Y be a compact irreducible Hermitian symmetric space. We embed Y in some projective space PN thanks to a homogeneous ample line bundle L. For a sheaf F on Y and an integer ℓ, we will denote the tensor product F ⊗ L⊗ℓ by F (ℓ). 2.1. The statement. The current section is devoted to prove the Theorem 1. Let Y be a compact irreducible Hermitian symmetric space, but not a projective space. Let l , p, q be integers, with l ≥ 0 and p > 0, such that H q (Y , Ω p Y (l )) 6= 0. (1) Then, l + q ≥ p c1(Y ) dim(Y ) . Furthermore, equality holds if and only if • p = dim(Y ), q = 0 and l = c1(Y ), 3 • or Y is a quadric and l = 0, • or Y ≃ Q4, l = 2, p = 3, q = 1. (2) If moreover q > 0, then l + q ≤ p. The proof of Theorem 1 is given in the next subsections. Surprisingly enough, the first item is very intricate, and the proof of the vanishing theorem in this case entails involved combinatorial arguments. The second item is much easier. 2.2. The case of Grassmannians (type An). Fix positive integers a, b ≥ 2. Let Gab := G(a, a + b) be the Grassmannian that parametrizes a-dimensional linear subspaces of a fixed (a + b)-dimensional C-vector space V . It is the homogeneous space G(a, a + b) = SU(a + b)/[SU(a + b) ∩ U(a) × U(b)]. Let Oab(1) be the Plücker polarization on Gab, which is also the positive generator of Pic(Gab). The cotangent bundle of Gab is denoted by Ωab. In case Y is a Grassmannian, by [Sno86], the non-vanishing of H q(Y , Ω integers (λ1 ≥ λ2 ≥ · · · ≥ λn) such thatPn p Y (l )) implies the existence of a partition of p, which is l -admissible with cohomological degree q in the following sense. Recall first that given a non negative integer p, a partition of p is a sequence of non-increasing natural i =1 pi = p. It will be represented by its Young diagram: each part is represented by λi boxes on the i -th row (from top to bottom), and these rows are left-justified. The dual partition λ∨ of a partition λ is defined by the fact that its i -th part λ∨ i is equal to the number of boxes in the i -th column of the Young diagram of λ. The hook number hλ(i , j ) of a given box (i , j ) in a Young diagram is the number of boxes i + j − 1 that build the hook based at the given box, as illustrated in the following example for the partition (6,4,2,2) with dual partition (4,4,2,2,1,1): * * * 6 * * Definition 2.1 ([Sno86]). Let λ be a partition and l an integer. We say that λ is l -admissible if no hook number of λ is equal to l . The (l -)cohomological degree of an l -admissible partition is the number of hook numbers which are greater than l . The hook number of the partition λ at the box (i , j ) will be denoted by hλ(i , j ). Hence, in the case of Grassmannians, the first part of Theorem 1 is equivalent to the following combinatorial statement, which we now prove: Proposition 2.2. Assume that a, b ≥ 2. Let l be a non-negative integer. Let λ be a l -admissible parti- tion of p which Young diagram fits in a rectangle a × b, with l -cohomological degree q. Then, we have the inequality l + q ≥ + p a p b , with equality holding if and only if either λ = (0) and l = q = 0, or λ = (b a), l = a + b and q = 0, or λ = (2,1), l = 2 and q = 1. Proof. Given a partition λ, we denote by a(λ) the first part of λ, by b(λ) its length, by p(λ) the sum of its parts, and by q(λ) its cohomological degree. We assume that a = a(λ) and b = b(λ). If q = 0, we have p ≤ b + a ≤ l and the first inequality is an equality if and only if λ = (b a). We are in the first a case of the proposition. p b + We assume from now on that q > 0. Without loss of generality, we may also assume that a ≥ b and that the proposition is proved for partitions λ′ such that a(λ′) < a or b(λ′) < b. Given an l -admissible partition λ, we denote by ∆(λ) the number ∆(λ) := l + q(λ) − 4 p(λ) a(λ) − p(λ) b(λ) . Our aim is to prove that ∆(λ) ≥ 0. It will be convenient to prove this inequality repeatedly replacing λ by a combinatorially simpler l -admissible partition λ′ such that ∆(λ′) ≤ ∆(λ). The main steps of the proof are illustrated in Picture 1. Let us first assume that hλ(2,1) > l . In this case, we consider the partition µ obtained removing the first column: namely, we set µi = max(λi − 1,0). By induction, we know that ∆(µ) ≥ 0. Now, we have a(µ) = a − 1, b(µ) ≤ b, p(µ) = p − b and q(µ) ≤ q − 2. It follows: 0 ≤ ∆(µ) < l + (q − 2) − p − b a − p − b b = l + q − p a − p b + b a − 1 ≤ ∆(λ), where the last inequality follows from our assumption a ≥ b. From now on, assume that hλ(2,1) < l . Set k = λ∨ q+1. We consider the partition µ defined by (see Picture 1) µ1 = λ1 µi = λ2 µi = λk if if 2 ≤ i ≤ k k + 1 ≤ i ≤ b   For all integers i , j , we have hµ(i , j ) ≥ hλ(i , j ), and we have hµ(1, q +1) = hλ(1, q +1) < l and hµ(2,1) = hλ(2,1) < l . It follows that µ is also l -admissible. We have a(µ) = a, b(µ) = b, p(µ) ≥ p, q(µ) = q, so ∆(µ) ≤ ∆(λ), so that it is enough to prove the Proposition for µ. We now consider two cases. The first case is when µ2 = q. Then the Young diagram of µ is de- scribed by the three integers a, b and c with a ≥ b ≥ 2 and a ≥ c ≥ 1: a + b − 1 b + c − 2 · · · a − c a + b − c b − 1 1 c 1 Then, as all integers between 1 and a − c are hook numbers, and similarly for integers between 1 and b + c − 2, the integer l has to fulfill the following inequalities l ≥ a − c + 1, l ≥ b + c − 1. Since the cohomological degree q is positive, then, as all numbers between a + b − c and a + b − 1 are hook numbers, l ≤ a + b − c − 1. There are exactly c boxes with hook number bigger than l , all lying in the first row: hence, q = c and Writing p = a + (b − 1)c = ab − (b − 1)(a − c) we derive l + q ≥ (a − c + 1) + c ≥ a + 1. p a + = a + 1 + (b − 1)( c a − a − c b ). a(a − c) − bc − = b ab ≥ ac − bc ab ≥ 0, p b c a But a − c where we have used a − c ≥ l − b + 1 ≥ c. This proves p a + p b ≤ l + q. In case of equality, (b −1)(ac − bc) = 0 so that a = b. Furthermore, all previously used inequalities are equalities l = a − c + 1 = b + c − 1 = a + b − c − 1 5 leading to c = 1 and a = b = 2. The second case is when µ2 > q. Note that the inequalities hµ(2,1) < l and hµ(1,1) > l imply that µ2 < µ1. Similarly, µk+1 < µ2. If hµ(1, q) > l +1, we may consider the l -admissible partition ν obtained by setting ν1 = µ1−1, νk+1 = µ2, and for i 6∈ {1, k +1}, νi = µi . We have p(ν) ≥ p(µ), a(ν) = a −1, b(ν) = b and q(ν) = q, so ∆(ν) < ∆(µ) and the Proposition is proved by induction in this case. If hµ(1, q) = l +1, then hµ(1,1) = l + q, so l + q = a + b − 1. Since p(µ) ≤ ab − b, the inequality ∆(µ) > 0 is also proved in this case. • ←− hµ(1, q + 1) = hλ(1, q + 1) • ←− hµ(2,1) = hλ(2,1) • ←− hν(1, q + 1) = hλ(1, q + 1) • ↑ hν(1, q) = l + 1 λ µ ν P i c t ur e1 (cid:3) We now prove the second part of Theorem 1. Proposition 2.3 (Second part of Theorem 1 for Grassmannians). Let l be a positive integer. Let λ be a l -admissible partition of p with l -cohomological degree q > 0. Then, we have the inequality l + q ≤ p. Moreover, if the equality p = l + q holds, then λ is a hook (i.e., its shape is (a,1b−1)). Proof. Let Y (λ) := {(i , j ) j ≤ λi } ⊂ N2 be the Young diagram of λ. For x ∈ Y (λ), we abbreviate the hook number hλ(x) of λ at x by h(x). We have p = #Y (λ) and q = #{x ∈ Y (λ) h(x) > l }. Since q > 0, let x ∈ Y (λ) such that h(x) > l . Moreover, we can assume that x is minimal for this property, namely that h(y) < l if y is south-east from x. By definition of h(x), there are h(x)−1 elements z ∈ Y (λ) which are either on the same row as x on its right, or under x in the same column. For these elements, we have h(z) < l . This implies that p − q = #{y ∈ Y (λ) h(y) < l } ≥ h(x) − 1 ≥ l . We now deal with the case of equality (that will not be used in the sequel). If the equality p = l + q occurs, with q > 0 as above, then we first show that x is on the first row. 6 x ′′ x ′ x x x ′ x ′′ Case λi −1 = λi Case λi −1 > λi If x is not on the first row, then the hook number of the box x ′′, very right on the row over that of x not in the same row of x neither on the same column, is 2 or 1. Hence this box contributes to {y ∈ Y (λ) h(y) < l } and therefore, p − q > l . In the same way, we can show that x is on the first column and that all the boxes with hook number smaller than l are on the hook of x. Therefore λ is a hook. (cid:3) 2.3. The case of quadrics (type Bn or Dn). Let Y be a non singular quadric hypersurface of dimen- sion n with its natural polarization OY (1) = O Pn+1(1)Y . It is the homogeneous space Y = SO(n + 2)/(SO(n) × SO(2)). From the adjunction formula, c1(Y ) = n + 2 − 2 = dim Y . Recall a theorem of Snow. Theorem ([Sno86, page 174]). Let Y be a non singular quadric hypersurface of dimension n. H q(Y , Ω p Y (l )) 6= 0, then If • either p = q and l = 0, • or q = n − p and l = −n + 2p, • or q = 0 and l > p, • or q = n and l < −n + p. As when q = 0 and 0 < p < n, the inequality l > p holds, the cotangent bundle ΩY of the quadric is stable as soon as the Picard group of the quadric is the restriction of that of Pn+1, for example when n ≥ 3. Theorem 1 follows for quadrics by checking the above cases. 2.4. The case of Lagrangian Grassmannians (type Cn). In this case, Y parametrizes n-dimensional Lagrangian subspaces of C2n equipped with the standard symplectic form. It is the homogeneous p space Y = Sp(2n, C)/U(n). By [Sno88], the non-vanishing of H q(Y , Ω Y (l )) amounts to the existence of an l -admissible Cn-sequence of weight p and cohomological degree q, in the following sense: Definition 2.4. Fix l , n ∈ N with l > 0. A n-uple of integers (xi )1≤i ≤n will be called an l -admissible Cn-sequence if • ∀ 1 ≤ i ≤ n, xi = i • ∀ i ≤ j , xi + x j 6= 2l . Its weight is defined to be p := Xxi >0 xi and its cohomological degree is q := #©(i , j ) i ≤ j and xi + x j > 2lª. Notation 2.5. Given an integer x, we denote x + := M ax(0, x). Therefore, we have p = Pi x + i . The set of all (u, v) ∈ N2 such that u ≤ v and xu + xv > 2l will be denoted by Q. The cardinality #Q will be denoted by q. Moreover we adopt the following convention: if C is a condition on (u, v), then Q(C ) will denote the subset of Q consisting of pairs satisfying C . For example, given an integer v0, the set Q(v = v0) consists of all pairs (u, v) in Q such that v = v0. 7 Since this excludes the case of Y being a projective space or a quadric, which occurs when n ≤ 2, the first part of Theorem 1 amounts to the following proposition in this case: Proposition 2.6 (First part of Theorem 1 for Lagrangian Grassmannians). Let x = (xi ) be an l -admissible Cn-sequence of weight p and cohomological degree q, with n ≥ 3. Then l + q ≥ 2p n , with equality occurring if and only if xi = i , l = n + 1 and q = 0, or p = q = l = 0. Proof. Let t = #{i xi > l }. Snow classified the cases where l = 1 [Sno88, Theorem 2.2]. In fact, the combinatorics are quite simple in this case since for a 2-admissible sequence (xi ) we have x1 = −1 and xi < 0 ⇒ x2+i < 0. Thus, such a sequence satisfies x2i +1 = −(2i +1), x2i = 2i for i ≤ t, and x2i = −2i for i > t. We then have p = t(t + 1) and q = t 2. Since t ≤ n 2 , we have 2p n = 2t(t + 1) n ≤ t + 1 ≤ t 2 + 1 = q + l . Moreover, if the equality holds, then 2t = n and t = 0 or t = 1, contradicting n ≥ 3. Therefore, the proposition is true in this case. We now assume that l ≥ 2. Given i , j , if xi > l and x j > l , then evidently xi + x j > 2l . Therefore, q ≥ t(t + 1) 2 ≥ 2t − 1. On the other hand, we have p ≤ l(l −1) 2 + t n. If 2p ≥ (l + q)n, then l (l − 1) + 2t n ≥ (2t + l − 1)n . Since l > 1, this implies n ≤ l , and so q = 0. If l = n, then xn = −n, so we have 2p ≤ n(n − 1), therefore, 2p n ≤ n − 1 < l , and the proposition is true. p = n(n+1) . 2 If l > n, since 2p ≤ n(n + 1), we get that 2p n ≤ n + 1 ≤ l , and if the equality holds then l = n + 1 and (cid:3) We now prove the second part of Theorem 1: Proposition 2.7 (Second part of Theorem 1 for Lagrangian Grassmannians). Let (xi ) be an l -admissible Cn-sequence of weight p and cohomological degree q, with n ≥ 3 and q > 0. Then Moreover, the equality p = q + l holds if and only if l + q ≤ p . x = (−1, −2,... , −l , l + 1, −(l + 2), −(l + 3),... , −n) with p = l + 1 and q = 1. Proof. The proof is similar to that of Proposition 2.3. Let j be the minimal integer such that there exists i ≤ j with xi + x j > 2l . We have x j = j ≥ l + 1. We want to bound q = #Q. We observe that if (u, j ) ∈ Q with j −l ≤ u < j , then xu > 0. Otherwise, xu = −u and 0 < xu +x j = −u + j ≤ l , contradicting the assumption that xu + x j > 2l . Hence 1 ≤ x + u , and therefore #Q(v = j , j − l ≤ u < j ) ≤ Xj −l ≤ u< j x + u . Actually, a similar inequality holds with j − l replaced by j − 2l , but in the sequel we will use the inequality j − l ≥ 1. In fact, we have #Q(v = j , u < j − l ) ≤ j − l − 1. v . Therefore, by minimality of j , we have the inequality: q = #Q(v = j = u) + #Q(v = j , j − l ≤ u < j ) + #Q(v = j , u < j − l ) + #Q(v > j ) Finally, #Q(v > j ) ≤Pv > j x + ≤ 1 +P j −l ≤u< j x + ≤ Pu< j x + u + x j +Pv > j x + u + ( j − l − 1) +Pv > j x + v − l = p − l . v 8 This proves the inequality l + q ≤ p. We now deal with the case of equality. Assume that q = p − l . Then, asking for equalities in the previous estimates, we find that for j − l ≤ u < j , if xu > 0, then xu = 1, and for u < j − l , xu + j > 2l by the first inequality and xu < 0 by the second. In particular, j − l − 1 ≤ 0 and hence j = l + 1 for otherwise x j −l −1 + j = −( j − l − 1) + j = l + 1 > 2l . For v0 such that j = l + 1 < v0, from the equality Q(v = v0) = x + v0, we infer that if xv0 > 0 then for all u ≤ v0, xu + xv0 > 2l . In particular, xv0−1 > 0 and xv0−2 > 0. By decreasing induction, we find that xl = l , contradicting the l -admissibility. Hence, for j < v < 2l , we get xv < 0. Finally, x is of the form (−1, −2, −3, · · · , −l , l + 1, −(l + 2), −(l + 3),... , −n) or (1, −2, −3, · · · , −l , l + 1, −(l + 2), −(l + 3),... , −n). In the second case, one has p = l + 2 thus q = 2 thus x1 + xl +1 > 2l thus l = 1. But then x is not 1-admissible since x1 = 1. (cid:3) 2.5. The case of spinor Grassmannians (type Dn). In this case Y parametrizes one of the two fami- lies of n-dimensional isotropic subspaces of C2n equipped with a non-degenerate quadratic form. p It is the homogeneous space Y = SO(2n)/U(n). By [Sno88], the non-vanishing of H q(Y , Ω Y (l )) amounts to the existence of an l -admissible Dn-sequence of weight p and cohomological degree q in the following sense: Definition 2.8. Fix n, l ∈ N with l > 0. A n-uple of integers (xi )0≤i ≤n−1 will be called an l -admissible Dn-sequence if • xi = i for all 0 ≤ i ≤ n − 1, • xi + x j 6= l for all i < j . Its weight is defined to be p := Xxi >0 xi and its cohomological degree q := #©(i , j ) i < j and xi + x j > lª. Remark 2.9. Observe that the only 1-admissible Dn-sequence is the sequence (0, −1,... , −n) with p = q = 0. In fact, the 1-admissibility condition leads to the implication (xv > 0 =⇒ xv −1 > 0), and thus for other sequences to x1 = 1 so that x0 + x1 = 1. We continue to use Notation 2.5 except that now Q = {(i , j ) i < j and xi + x j > l }. Since Y is not a projective space or a quadric, we have n ≥ 5. The first part of theorem 1 amounts to the following proposition in this case: Proposition 2.10 (First part of Theorem 1 for spinor Grassmannians). Let (xi ) be an l -admissible Dn-sequence of weight p and cohomological degree q, with n ≥ 5. Then l + q ≥ 4p n , with equality occurring if and only if xi = i and l = 2(n − 1). Proof. First of all, if xn−1 = −(n − 1), let x ′ be the sequence of length n − 1 with x ′ = xi for i ≤ n − 2. i Then x ′ is evidently l -admissible. It has weight p and cohomological degree q. By induction on n, we get that l + q ≥ 4p n . Thus, in the rest of the proof, we assume that xn−1 = n − 1. 4p n−1 > Let us first assume that l > 2(n −1). In this case, we have q = 0. Since in any case p ≤ n(n−1) , we get 2 that 4p n ≤ 2(n − 1) < l + q. Let us now assume that n ≤ l ≤ 2(n − 1). Then, we denote by u the unique integer that satisfies the following condition Since the sequence (xi ) is l -admissible, if for l −n +1 ≤ i ≤ n −1 we have xi > 0, then l −n +1 ≤ l −i ≤ n − 1 and xl −i < 0. This implies that #{i xi > 0, l − n + 1 ≤ i ≤ n − 1} = u + 1. n − 1 − (l − n + 1) ≥ 2u , that is, l ≤ 2(n − u − 1). Since n ≤ l ≤ 2n − 2u − 2, we have n ≥ 2u + 2. The sum of positive xi 's with l − n + 1 ≤ i ≤ n − 1 can be at most (u + 1)(n − 1) − u(u+1) . Therefore, we have 2 4p ≤ 4(u + 1)(n − 1) − 2u(u + 1) + 2(l − n + 1)(l − n). 9 On the other hand q ≥ u, so introducing ∆ := (u + l )n − 2(l − n + 1)(l − n) − 4(u + 1)(n − 1) + 2u(u + 1), the proposition amounts to the positivity of ∆ whenever n ≤ l ≤ 2(n − u − 1). After fixing u and n, the above defined ∆ is a concave function on l , so we only need to consider the values of ∆ when l = n and when l = 2(n − u − 1). When l = n, we get that ∆ = n2 − (4 + 3u)n + 2u2 + 6u + 4. Fixing u, the two roots of this polynomial are n = u + 2 and n = 2u + 2. Since we know that n ≥ 2u + 2, we have ∆ ≥ 0 for l = n ≤ 2(n − u − 1). For l = 2(n − 1 − u), we have ∆ = 3un − 6u(u + 1) Since once again n ≥ 2u +2, we get that ∆ ≥ 0, and hence the inequality in the proposition follows for any l such that n ≤ l ≤ 2(n − u − 1). Moreover, we show that the equality l + q = 4p n can only occur if xi = i and l = 2(n − 1). Indeed, let us assume that ∆ = 0. By the concavity argument, we have either l = n or l = 2(n − u − 1). If l = n, we also get by the above argument that n = 2u + 2. Since the inequality 4p ≤ 4(u + 1)(n − 1) − 2u(u + 1) becomes an equality, we conclude that x is of the form (−0, −1, · · · , −u, u + 1, u + 2, · · · ,2u + 1). This implies that q = u(u+1) , and since q = u, we have u = 1 and n = 4, and the last equality contradicts the hypothesis of the proposition. If l = 2(n − u − 1), since ∆ = 3un − 6u(u + 1) = 0, we have n = 2(u + 1) or u = 0. The case of n = 2(u + 1), n = l , was already dealt with earlier. Thus we have u = 0 and l = 2(n − 1). The equality 2 amounts to p = (n−1)n 2 4p = 4(u + 1)(n − 1) − 2u(u + 1) + 2(l − n + 1)(l − n) , so that xi = i for all i , and we are in the case of the proposition. Let us now assume that l < n. We consider the sequence (x ′ n−1 = −(n −1). We observe that (x ′ i ) is l -admissible with weight p ′ = p −(n −1) and cohomological degree q ′ satisfying q ′ ≤ q −(n −l ). In fact, xi + xn−1 > l for i < n −1 −l , and xn−1−l = n −1 −l by l -admissibility, so that xn−1−l + xn−1 > l . = xi for i < n − 1 and x ′ i ) with x ′ i By our very first argument, we have 4p′ < l + q ′ +4. Therefore, if q ′ ≤ q −4, then we are done. This is indeed the case if n − l ≥ 4. Thus, we assume that q ′ ≥ q − 3, and so n ≤ l + 3. We now consider these cases. n−1 < l + q ′, so that 4p n If n = l + 3, we have xn−1 = l + 2, and so x2 = 2. Since q ′ ≥ q − 3, we get that xi < 0 for 3 ≤ i ≤ n − 2. Thus we have p ≤ l + 5. The inequality in the proposition is implied by the inequality l + 3 > 4(l +5) l +3 , which in turn is true for l ≥ 3. Observe that the value l = 2 is excluded because we would then have x4 = 4. Therefore, either x2 + x0 = 2 (if x2 = 2) or x2 + x4 = 2 (if x2 = −2). If n = l + 2, then there is at most one integer i such that 2 ≤ i ≤ n − 2 and xi > 0. By admissibility, x1 = 1, and therefore xl −1 = −l +1. Moreover, we have xl = −l . This implies that p ≤ 2l . The inequality of the proposition is implied by the inequality l + 2 > 8l l +2 , which in turn is true for l ≥ 3. The value n = l + 1 would contradict l -admissibility, since we would then have xl = l . (cid:3) We now prove the second part: Proposition 2.11 (Second part of Theorem 1 for spinor Grassmannians). Let (xi ) be an l -admissible Dn-sequence of weight p and cohomological degree q, with n ≥ 5 and q > 0. Then l + q ≤ p . Moreover, the equality p = q + l holds if and only if there are exactly two indices i , j such that xi > 0, x j > 0 and they satisfy the condition that either xi + x j = l + 1 (in this case q = 1) or x is equal to (0,1, −2, −3,... , −l , l + 1, −(l + 2), −(l + 3),... , −n) (then q = 2). 10 Proof. Let j be the smallest integer such that there exists i < j with xi + x j > l . Observe that x j > 0, and by Dn-admissibility, x0 + x j 6= l so that j 6= l . We first deal with the case j > l . The argument in this case is similar to the case of type Cn: q = #Q = #Q(v = j , j − l ≤ u < j ) + #Q(v = j , u < j − l ) + #Q(v > j ) ≤ P j −l ≤u< j x + ≤ Pu< j x + u + x + j u + ( j − l ) +Pv > j x + − l +Pv > j x + v v = p − l . If under the assumption j > l the equality p = q + l holds, then by the second inequality we have xu ≤ 0 for u < j − l , and by the first inequality we have x + u ≤ 1 for j − l ≤ u < j . If x j −l = −( j − l ), then x j −l + x j = l , contradicting l -admissibility. Therefore, x j −l = j − l ≤ 1, so j − l = 1 and x1 = 1. When v > j , the first inequality leads to the implication (xv > 0 =⇒ ∀ u < v, xu +xv > l ). Assuming the existence of a v > j such that xv > 0, we get that xv −1 > 0 and xv −2 > 0 because l > 1 (see Remark 2.9). By descending induction, this would lead to x j −1 > 0. Then j − 1 = 1, hence j = 2, l = 1, which is a contradiction. Thus, if v > j , then xv < 0. Hence x = (0,1, −2, −3, · · · , −l , l + 1, −(l + 2), −(l + 3),... , −n) (with p = l + 2 and q = 2). Let us now assume that j < l , and let i be the largest integer such that i < j and xi + x j > l . Note that xi > 0 and x j > 0, and by maximality of i we have xk < 0 for i < k < j . We have q = #Q = 1 + #Q(v = j , u < i ) + #Q(v > j ) ≤ 1 +Pu<i x + u +Pv > j x + ≤ (xi + x j − l ) +Pu<i x + v u +Pv > j x + v = p − l . The inequality l + q ≤ p is proved. In the case of q = p − l , for 1 ≤ u < i we have x + u ≤ 1 by the first inequality. By the second inequality we have xi +x j = l +1. Using the descending induction argument, if there is a v > j such that xv > 0, then xl > 0, and x0 + xl = l contradicting the admissibility. The only positive entries are among x1, xi , x j . In this case, since i + j = l + 1, we have Q = {(i , j )}, so q = 1, p = l + 1 and x1 < 0. (cid:3) 2.6. The exceptional cases (type E6 or E7). Now Y is homogeneous under a group of type E6 (case E I I I ) or E7 (case EV I I ). In the first case, we have dim(Y ) = 16 and c1(Y ) = 12. In the second case, we have dim(Y ) = 27 and c1(Y ) = 18. These values are well-known to the specialists; several arguments for the computation of c1 can be found at the end of Section 2.1 in [CMP08]. From Tables 4.4 and 4.5 in [Sno88] we conclude that the inequalities we are looking for hold: Proposition 2.12 (Theorem 1 for the exceptional cases). Let Y be a Hermitian symmetric space of type E I I I or EV I I . Let l , p, q be integers with l > 0, p > 0, and such that H q (Y , Ω p Y (l )) 6= 0. Then, p c1(Y ) dim(Y ) ≤ l + q. Equality implies that p = dim(Y ), l = c1(Y ) and q = 0. Assume moreover that q > 0. Then l + q ≤ p. 2.7. A cohomological property. We will need the following corollary of Theorem 1 to prove our sta- bility results. Proposition 2.13. Let Y be a compact irreducible Hermitian symmetric space, but not a projective space. Let l , p, q be integers, with q < dim Y , such that H q(Y , Ω p Y (l )) 6= 0. Then, l + q ≥ p c1(Y ) dim(Y ) , with equality holding if and only if • p = dim(Y ), q = 0 and l = c1(Y ), • or p = q = l = 0, • or Y is a quadric and l = 0, • or Y ≃ Q4, l = 2, p = 3, q = 1. 11 Proof. Assume that H q(Y , Ω p Y (l )) 6= 0. If l ≥ 0, then we are done by Theorem 1(1). If p = 0, then q = 0 if l > 0, while q = dim(Y ) if l < 0. Thus this case is also settled. Assume that q > 0 and l > 0. Let us prove that actually l + q < p c1(Y ) dim(Y ) + dim(Y ) − c1(Y ). (1) Firstly, the theorem states that l + q ≤ p, and we have p ≤ dim(Y ). Since (1 − c1(Y ) dim(Y ) )(l + q) ≤ dim(Y ) − c1(Y ), we find l + q ≤ c1(Y ) dim(Y ) (l + q) + dim(Y ) − c1(Y ) ≤ p c1(Y ) dim(Y ) + dim(Y ) − c1(Y ) . Note that the equality here would imply that p = dim(Y ), in which case, by Kodaira vanishing, we cannot have q > 0. Thus, the inequality (1) is proved. Now, coming back to the proof of Proposi- tion 2.13, if l < 0 then Serre duality leads to The relation (1) gives that (−l )+(dim(Y )−q) < (dim(Y )−p) c1(Y ) dim(Y ) +dim(Y )−c1(Y ), or in other words, dim(Y )−p H dim(Y )−q (Y , Ω Y (−l )) 6= 0. l + q > p c1(Y ) dim(Y ) . (cid:3) 3. RESTRICTIONS WITH SMALL PICARD GROUP 3.1. Short split resolution. We will prove stability of the restriction of ΩY to subschemes whose structure sheaf has a short split resolution in the following sense: Definition 3.1. A subscheme X ⊂ Y is said to have a short split resolution if there is a resolution 0 → Fk → Fk−1 → · · · → F1 → F0 → OX → 0, where F0 = OY , Fi := ⊕ j OY (−di j ), and the length k of the resolution satisfies k < dim(Y ). Example 3.2. The Koszul resolution of complete intersections is a short split resolution for a positive- dimensional complete intersection in Y . If, moreover, none of the equations are linear, then the integers di j in Definition 3.1 satisfy di j ≥ i + 1. Our second class of examples are some arithmetically Cohen-Macaulay subschemes. Let X ⊂ PN be a subscheme defined by a homogeneous ideal J in the homogeneous coordinate ring A := C[X0,..., XN ]. Recall that X is called arithmetically Cohen-Macaulay if the depth of A/J is equal to the dimension of A/J, namely dim X + 1. Let, moreover, I ⊂ A denote the homogeneous ideal of Y . The reason why we will consider arithmetically Cohen-Macaulay subschemes is the following: Lemma 3.3. Let X be an arithmetically Cohen-Macaulay subscheme of Y defined in PN by an ideal J . Assume that A/J has finite projective dimension over A/I . Then, the structure sheaf OX has a resolution by split vector bundles over Y of length k = dim Y − dim X 0 → Fk → Fk−1 → · · · → F1 → F0 → OX → 0, where F0 = OY and Fi := ⊕ j OY (−di j ) with di j ≥ i for i > 0. In particular, if dim(X ) > 0, then X has a short split resolution. Proof. We have I ⊂ J ⊂ A. By Auslander-Buchsbaum formula [Mat89, Theorem 19.1], we have the equality pdA/I (A/J) = depth(A/I ) − depth(A/J). Moreover, any homogeneous space embedded by a homogeneous ample line bundle is arithmeti- cally Cohen-Macaulay (see for example [BK05, Corollary 3.4.4]). Thus, depth(A/I ) = dim(A/I ) = dim Y + 1. Therefore, pdA/I (A/J) = dim Y − dim X . Hence a minimal free resolution of A/J over A/I has length k = dim Y − dim X . (cid:3) 12 Remark 3.4. Concretely, our assumption that A/J has finite projective dimension over A/I might be difficult to check. For example, Theorem 2 becomes obviously wrong if we take X ≃ P1 (since the restriction of ΩY to X will be split in this case). All our arguments hold for X ≃ P1, except that in this case the corresponding projective dimension is infinite. Lemma 3.5. Let X ⊂ Y admitting a short split resolution. Then, there is a short split resolution such that the integers di j in Definition 3.1 satisfy ∀i , j , di j ≥ i . If, moreover, X is not linearly degenerate in PN , then we may assume that the inequalities di j ≥ i + 1 hold for all (i , j ). Proof. We consider as in the proof of Fact 3.3 a minimal free resolution of A/J over A/I . The hypoth- esis implies that such a resolution will have length strictly less than dim(Y ). Moreover, since for such a resolution the differentials have positive degree, we have di j ≥ i for all i , j . If X is not included in any hyperplane, it has no equation of degree 1, so d1 j ≥ 2 for all j , and we deduce that di j ≥ i +1. (cid:3) 3.2. General argument. Recall that the slope of a coherent torsion-free sheaf F of positive rank on a polarized manifold (X , L) is µ(F ) := c1(F ) · Ldim X −1 . rankF Recall that a coherent torsion-free sheaf E on a polarized manifold (X , L) is said to be (slope-)stable if the slope of all its subsheaves of positive smaller rank is less than its slope. Note that it is enough to check the inequalities for saturated subsheaves. We can now prove the Theorem 2. Let Y be any compact irreducible Hermitian symmetric space excluding a projective space and a quadric. Let X be a locally factorial positive dimensional subvariety of Y having a short split resolution and such that Pic(X ) = Z · OY (1)X . Then the restriction of ΩY to X is stable. Note that if X is a complete intersection of dim X ≥ 3, the constraint on the Picard group of the complete intersection is ensured by Lefschetz theorem [Laz04, Example 3.1.25]. Proof. We will later prove a slightly weaker result for quadrics and projective spaces (Theorem 3), thus, for the moment Y is any compact irreducible Hermitian symmetric space. We first explain how, building on a classical argument, the assumption on the existence of a small split resolution reduces the check of the stability inequalities to vanishing theorems. Let F be a coherent subsheaf of ΩY X of rank 0 < p < dim Y . Since X is assumed to be locally factorial, the rank one reflexive subsheaf det F := (Vp F )∗∗ ofVp ΩY X is invertible [Har80, Proposition 1.9] and hence isomorphic to OY (−d)X =: OX (−d) for some integer d. We have c1(Y ) dim(Y ) OX (1)dim(X )−1 · KY µ(ΩY X ) = rank(ΩY X ) = − · degY X deg Y µ(F ) = OX (1)dim(X )−1 · det F rank(F ) = − d p · degY X deg Y . The inclusion F ⊂ ΩY X yields the non-vanishing of H 0(X , Hom(detF , Ω p Y X )) = H 0(X , Ω p Y (d)X ), from which we have to deduce the stability inequality µ(F ) < µ(ΩY X ), equivalently, d > p c1(Y ) dim(Y ) . Consider a resolution of OX as in Definition 3.1. The resolution 0 → Fk ⊗ Ω p Y (d) → · · · → F1 ⊗ Ω p Y (d) → F0 ⊗ Ω p Y (d) → Ω p Y (d)X → 0 translates the non-vanishing of H 0(X , Ω groups in the decomposition p Y (d)X ) into the non vanishing of one of the cohomology H i (Y , Fi ⊗ Ω p Y (d)) = ⊕ j H i (Y , Ω p Y (d − di j )), say of H i (Y , Ω p Y (d − di j )). 13 We now assume that Y is not a projective space. In our setting Proposition 2.13 reads (d − di j ) + i ≥ p c1(Y ) dim(Y ) . (2) It follows that d ≥ p c1(Y ) dim(Y ) . If the equality d = p c1(Y ) dim(Y ) holds, we get that di j = i , and the equality in (2) holds. Now assume that Y is not a quadric. Therefore, Proposition 2.13 gives that p = 0 or p = dim(Y ), equivalently, as we may assume F saturated (i.e. with torsion free quotient) either F = {0} or F = ΩY . Thus, ΩY is stable. (cid:3) Some remarks regarding the two excluded cases in Theorem 2. Remark 3.6. If X is contained in a linear subspace H in a projective space Y , then for the exact sequence, the slope − deg X even semi-stable. codimY H of N ∗ H Y X 0 → N ∗ H Y X → ΩY X → ΩH X → 0, is strictly bigger than the slope −(dim Y +1)deg X dim Y of ΩY X . Thus, ΩY X is not Remark 3.7. If X ⊂ Y is a linear section of the quadric Y (i.e. degY X = 1), then, as the above proof shows, the restriction of ΩY to X is semi-stable. Furthermore, all the vector bundles in the following exact sequence 0 → N ∗ X Y → ΩY X → ΩX → 0 have equal slope −1. So ΩY X is not stable. In fact, let us continue the end of the argument in the proof of Theorem 2: we have in this case i = 1 and d11 = 1 (since a resolution of X contains only one p term of degree 1, X being a hyperplane section). We get H 1(Y , Ω Y (d − 1)) 6= 0, which implies that p = 1 and d − 1 = 0 by Proposition 2.13. Thus the only destabilizing subsheaf is OX (−1). Thus, to get a result similar to Theorem 2 in these two cases, we exclude the case where X has a linear equation: Theorem 3. Let Y be a smooth quadric of dimension at least 3 or a projective space. Then ΩY is stable. Let X be a locally factorial subvariety in Y having a short split resolution and such that Pic(X ) = Z · OY (1)X . Assume that X is contained in no hyperplane section of Y . Then the restriction of ΩY to X is stable. Note that by Lefschetz theorem, this theorem applies to very general cubic hypersurfaces of Q3. Proof. We continue with the notation of Theorem 2. If Y is a quadric, by the above proof of Theorem p 2, we have the non-vanishing of some H i (Y , Fi ⊗ Ω Y (d)). If i > 0, by (2) we have for some j the inequality d − di j + i ≥ p c1(Y ) dim(Y ) (= p). Since, by Lemma 3.5, we get di j > i , we conclude that d > p, as wanted. If i = 0 and we assume d ≤ p p by contradiction, then H 0(Y , Ω Y (d)) 6= 0, and by a result due to Snow (see Section 2.3), we get that either p = 0 or p = dim(Y ). This implies stability as in the proof of Theorem 2. Assume now that Y is the projective space Pn. We may assume that 0 < p < n. We wish to prove that d p > n + 1 n . Since p+1 p if and only if one of the following hold: > n+1 n , it is enough to prove that d ≥ p + 1. For integers p, q, l , we have H q (Pn, Ω p Pn (l )) 6= 0 (1) l > 0, p < l and q = 0, (2) l = 0 and p = q, (3) l < 0, n − p < −l and q = n. 14 p Once again, we get that H i (Y , Fi ⊗ Ω Y (d)) 6= 0 for some i . If i = 0, since F0 = OY , this implies d > p or p = d = 0. If i > 0, since i < n, this implies that i = p and d = di j for some j . Thus we have d ≥ i + 1 = p + 1, as wanted. (cid:3) 4. RESTRICTION TO A HYPERSURFACE WITH AN INCREASE OF THE PICARD GROUP 4.1. Another argument for general complete intersection. In this section, we want to get rid of the assumption on the Picard group. This can be done at the cost of considering only general complete intersections. We get the following adaptation of theorems 2 and 3. Theorem 4. Let Y be a compact irreducible Hermitian symmetric space. Let X be a general positive- dimensional complete intersection in Y . If Y is neither a projective space nor a quadric, then the restriction of ΩY to X is semi-stable. If Y is a smooth quadric or a projective space, assume that none of the hypersurfaces Hi is linear. Then the restriction of ΩY to X is semi-stable. Proof. Let V = Γ(Y , OY (1))∗ be the minimal homogeneous embedding of Y , so that Y ⊂ PV . Let S = PS h1V ∗ × ... × PS hc V ∗. Let Z ⊂ Y × S be the universal family of complete intersections defined by (x,([H1],...,[Hc])) ∈ Z ⇐⇒ ∀i , Hi (x) = 0. Thus we have morphisms p : Z → Y and q : Z → S such that for general s = (Hi ) ∈ S (i.e. for s in a non empty Zariski open set), the inverse image Xs := q −1(s) = ∩c i =1(Hi = 0) is a complete intersection in Y of multi-degree (hi ). To proceed by contradiction, assume that semi-stability of the restriction of ΩY to X does not hold. We will use the relative Harder-Narasimhan filtration relative to q : Z → S [HL10, Theorem 2.3.2] (the idea of using this relative version appears e.g. in the proof of [HL10, Theorem 7.1.1]). After the choice of a suitable birational projective morphism f : T → S, we can build the following commutative diagram g ∗p ∗ΩY p ∗ΩY p / Y ZT T g f / Z q / S. Here, ZT is the fibered product of Z and T . The pulled-back sheaf g ∗p ∗ΩY has a filtration which induces for a general point s ∈ S the Harder-Narasimhan filtration of ΩY X s . We denote by F the first term of this filtration and by k its rank. The assumption that semi-stability fails amounts to saying Since S is smooth, f is an isomorphism in codimension 1; so the same holds for g , and since Z is also smooth, the line bundle det F on ZT defines a line bundle on Z denoted by L . It is a subsheaf of that 0 < k < dim Y . The rank one reflexive subsheaf det F := (Vk F )∗∗ of p ∗Vk ΩY is invertible. Vk p ∗ΩY . Now, p is a locally trivial morphism with fibers isomorphic to products of projective spaces, so Pic(Z ) ≃ Pic(Y ) × Pic(S), and L can be expressed as p ∗LY ⊗ q ∗LS, for some line bundles LY ∈ Pic(Y ) and LS ∈ Pic(S). This is the main feature of considering families: whereas the determinant of a single destabilizing subsheaf may fail to lie on Pic(Y )X , the determinant of the first term of the relative Harder-Narasimhan filtration does. Let d be the integer such that LY ≃ OY (−d), and let X = Xs. Given s ∈ S, we have LX ×{s} ≃ LY X , and hence for general s ∈ S, this yields an injection of sheaves OY (−d)X ⊂ p ∗ΩY . 15       /   / / Let h = h1 ... hc, we have: µ(ΩY X ) = µ(FX ) = OX (1)dim(X )−1 · KY rank(ΩY X ) = − c1(Y ) dim(Y ) · h · deg Y OX (1)dim(X )−1 · det FX rank(F ) = − d p · h · deg Y . Since OY (−d)X ⊂ ΩY X , it follows that H 0(X , Hom(OY (−d)X , Ωk Y X )) = H 0(X , Ωk Y X (d)) does not vanish. Using this and previous results, valid on Pic(Y )X , we deduce the inequality µ(FX ) < µ(ΩY X ), i .e., d > k c1(Y ) dim(Y ) . This contradicts the construction of L as the determinant of the first term of the relative Harder- Narasimhan filtration. (cid:3) 4.2. Variation on a result due to Langer. To state a general theorem due to Langer, we only assume in this subsection that Y is a smooth projective variety. We denote by OY (1) a polarization of Y . We set d := c1(OY (1))dim Y Definition 4.1. The discriminant of a rank r vector bundle E on Y is ∆(E) :=£2r c2(E) − (r − 1)c2 1(E)¤ · c1(OY (1))dim Y −2. Theorem ([Lan04, Theorem 5.2]). Consider a smooth projective variety Y . Consider a OY (1)-stable vector bundle E of rank r on Y . Let X be a smooth divisor in the complete linear system OY (h). If h > r − 1 r ∆(E) + 1 d r (r − 1) , (3) then EX is OY (1)X -stable. Remark 4.2. As noticed in [Lan04, Remark 5.3.2], when r > 2, the inequality (3) is equivalent to the inequality h > r −1 ∆(E), since h is an integer. Langer also points out that his Theorem "can be further r improved at the cost of simplicity". This is what is done in Proposition 4.3. The proof of the following result was communicated to us by Jie Liu: Proposition 4.3. Consider a smooth projective variety Y . Consider a OY (1)-stable vector bundle E of rank r on Y . Let X be a smooth divisor in the complete linear system OY (h). Let m be a common divisor of r and the degree of E . If h > r − 1 r m ∆(E) + m d r (r − 1) , (4) then EX is OY (1)X -stable. Proof. In the proof of [Lan04, Theorem 5.2], the term 1 (namely in Langer's notation µmax(G) − µ(G)) may be replaced by m r (r −1) that bounds the difference of two slopes r (r −1) . The inequality 0 ≤ d ∆(E) − ρ(r − ρ)d 2h2 + r 2( r − ρ r d h − 1 r (r − 1) )( ρ r d h − 1 r (r − 1) ) where ρ is the rank of a putative maximally destabilizing subsheaf of the restriction EX , may here be replaced by 0 ≤ d ∆(E) − ρ(r − ρ)d 2h2 + r 2( r − ρ r d h − m r (r − 1) )( ρ r d h − m r (r − 1) ) that is 0 ≤ d ∆(E) − + m2 (r − 1)2 . (cid:3) r md h r − 1 16 In particular, if Y = Q3 and E = ΩY , then d = c1(O Q3(1))3 = 2, r = 3, ∆(Ω Q3) = [6 × 4c1(O Q3(1))2 − 2((−3)c1(O Q3(1)))2] · c1(O Q3(1)) = 12 (here we have used the Euler sequence on P4 and the normal sequence for Q3 in P4 to compute 3 + 1 c2(Ω 4 . The inequality h ≥ 3 for h = degY X is therefore enough to derive the stability of the restriction Ω Q3 Q3(1))2). The inequality (4) reads h > 8 Q3 ) = 4c1(O X . Hence, the bounds of Langer are • for P2, h > 2, • for P3, h > 8/3, • for Q3, h ≥ 3. 4.3. Optimal bounds. With some obvious exceptions, we get the stability of ΩY X : Theorem 5. Let Y be a compact irreducible Hermitian symmetric space of dimension 2 or 3. Let X ⊂ Y be a smooth divisor. • Take Y = P2. Then ΩY X is semi-stable if degY X ≥ 2. If degY X ≥ 3, then ΩY X is stable. • If Y = P3, assume that degY X ≥ 2. Then ΩY X is stable. • If Y = Q2, then ΩY X is semi-stable but not stable. • Take Y = Q3. If degY X = 1, then ΩY X is semi-stable with respect to the anti-canonical polar- ization. If degY X ≥ 2, then ΩY X is stable. These cases will be considered in the next subsections. 4.4. The case of P2. Recall the Euler sequence on P2 0 −→ Ω P2(1) −→ O P2 ⊕3 −→ O P2(1) −→ 0. For a smooth conic C , since each section of H 0(C , O (1)C ) is a restriction of a section on P2, the rank 2 vector bundle Ω P2(1)C is isomorphic to the direct sum of two line bundles of degree −1 on the rational curve C . Consequently, it is semi-stable and not stable. P2(1)C of degree −2 has no sections. Therefore, Ω Let C be a curve of degree d ≥ 3 in P2. By Langer theorem (see Remark 4.2) we get the stability of P2(1)C . Ω P3(1)Q = O 4.5. The case of Q2. We consider a non degenerate quadric Q in P3. Recall that it is isomorphic to P1 × P1 in such a way that O P1×P1(1,1). Note that the tangent bundle T Q = O (2,0) ⊕ O (0,2), being the sum of two line bundles of the same O P3(1)Q-semi-stable. Its restriction T Q X to a smooth curve X ⊂ Q in any linear system O P3(d)Q = O (d, d) is the sum of two line bundles of degree O (d, d)·O (2,0) = O (d, d)·O (0,2) = 2d. Hence T QX is semi-stable for any d ≥ 1. 4.6. The case of Q3. We now consider a non degenerate quadric Q in P4 and a smooth hypersurface S ⊂ Q of degree d. By the results in Section 2.3, the vector bundle T Q is stable. The bound in Langer's theorem computed in Remark 4.2 is 3. Therefore, we conclude that the restriction T QS is stable if d ≥ 3. P3(1)Q-degree, is O We will study the degree 1 and 2 cases in the rest of this subsection. 4.6.1. Linear sections. For d = 1, the isomorphism Pic(S) = Z2 is due to the product structure S ≃ P1 × P1. Proposition 4.4. If S is a smooth linear section of the solid quadric Q, then T QS is semi-stable but not stable. Proof. Consider a putative destabilizing sheaf F ⊂ T QS. Assume that the rank of F is one. Replacing F by its reflexive hull, we get an exact sequence with L a line bundle and E a rank 2 vector bundle. Moreover, we have deg(F ) = deg(L), thus to prove semi-stability it suffices to show that the existence of such an exact sequence implies that deg(L) ≤ µ(T QS) = 2. 0 → L → T QS → E (5) 17 Let us write S = P1 × P1 π1,π2→ P1, and L = O (d1, d2) := π∗ O (d2). For example, for L = 1 1 T P1, we have an exact sequence as in (5), and L ≃ O (2,0) so deg(L) = 2. In particular T QS can not π∗ be stable. The semi-stability inequality is proved in the following Lemma 4.5. O (d1) ⊗ π∗ 2 Lemma 4.5. With the notation of the proof of Proposition 4.4, let O (d1, d2) be a subbundle of T QS. Then, we have d1 + d2 ≤ 2. Proof. Since L = O (d1, d2) is assumed to be a subbundle of T QS, there is a non vanishing section of L∗ ⊗ T QS. There is an exact sequence of sections on S: H 0(L∗ ⊗ T S) → H 0(L∗ ⊗ T QS) → H 0(L∗ ⊗ OS(1)) . We have L∗ ⊗ OS(1) ≃ O (1 − d1,1 − d2) and L∗ ⊗ T S ≃ O (2 − d1, −d2) ⊕ O (−d1,2 − d2). Assume that d1 > 1. Then H 0(L∗(1))) = 0 since π1∗L∗(1) = 0. Thus H 0(L∗ ⊗ T QS) = H 0(L∗ ⊗ T S) = H 0(O (2 − d1, −d2)). By the same argument, this space of sections is not equal to {0} if and only if d1 = 2 and d2 ≤ 0. Moreover, there will be non vanishing sections if and only if d2 = 0. We get (d1, d2) = (2,0) (so L is isomorphic to π∗ 1 T P1). Similarly, we can deal with the case d2 > 1. In the remaining cases we indeed have d1 + d2 ≤ 2 (cid:3) (asserted in the lemma). We now finish the proof of Proposition 4.4. Consider a rank two subsheaf F ⊂ T Q S. Its determi- nant is a line subbundle of ∧2T Q S. As T Y (−1) = O (−1)⊥/O (−1) is self-dual, we have ∧2T Q S = Ω2 Q S (4) = KQ ⊗ T Q S(4) = T Q S(1). Hence we get an inclusion detF ⊗ O (−1)S ⊂ T Q S. We already checked that line subbundles do not destabilize T Q S. Therefore, we infer that 2µ(F ) − 2 ≤ µ(T Q S) = 2, which implies the desired semi- stability inequality µ(F ) ≤ µ(T Q S). (cid:3) 4.6.2. Quadric sections. For d = 2, the surface S, intersection of two quadrics in P4, is a Del Pezzo surface of degree 4 meaning (−KS) · (−KS ) = 4 (see [Dol12, Definition 8.1.12]); it is known as a Segre quartic surface. Its Picard group Pic(S) is iso- morphic to Z6 with precise generators given by the abstract description of S as the projective plane P2 blown-up at 5 points in general position (see[GH94, page 550], [Dol12, Proposition 8.1.25]). Recall the diagram ι P3 Bl p(S) µ S φ π Σ  b / P2 where µ is the blow up of S at a point p on S not on a line of S, ι ◦ φ is given by the linear system of lines in P4 passing through p (its image is a smooth cubic Σ), b is the blow up of P2 at six points, ι is given by the linear system of cubics in P2 passing through the blown-up six points, and π is gotten i =1 Ei the sum of the five exceptional lines and  : E → S the natural inclusion, the main relations among sheaves in the two descriptions of S are from φ by the universal property of blow ups. With E =P5 and O P4(1)S = π∗O P2(3) ⊗ OS(−E) Before stating Proposition 4.7 which is the main result of this section, we start recalling a result of 0 → π∗Ω P2 → ΩS → ∗ΩE → 0. (6) Fahlaoui: 18      / Lemma 4.6. Let 0 6= ω ∈ H 0(P2, Ω P2(2)) ≃ C3 and let j ∈ {1,... ,5}. Consider the pull-back section π∗ω ∈ H 0(S, ΩS ⊗ π∗O (2)). If the class of ω in P2 is the point p j , then π∗ω vanishes at order exactly 2 along the exceptional divisor E j . Otherwise, it does not vanish along E j . Proof. Let us assume that the section ω of Ω such that the blown up point p j is [0 : 0 : 1] by P2(2) is given in homogeneous coordinates [X : Y : Z ] ω := X d Y − Y d X Z 2 Z¶2 =µ X X¶ . dµ Y In [Fah89, Exemple 1], it is shown that its pull-back π∗ω on S has poles only along the strict transform E0 of the line (Z = 0) with order two, and vanishes with multiplicity two along the exceptional divisor E j above p j : π∗ω ∈ H 0(S, ΩS ⊗ OS(2E0 − 2E j )). Thus the lemma is proved in this case. Since H 0(P2, Ω P2(2)) ≃ C3 and P2 is homogeneous under SL3, using the group action, we see that this result holds whenever the class of ω corresponds to p j . If this is not the case, then ω does not vanish at p j , which obviously implies that π∗ω does not vanish along E j . (cid:3) Proposition 4.7. If S is a smooth quadric section of the solid quadric Q, then ΩQ S is stable. The proof runs through the rest of this subsection. We denote by E0 the strict transform of a general line in P2 so that OS(E0) = π∗O P2(1). To begin with, consider a line bundle L = π∗O P2(−a) ⊗ OS(− 5Xj =1 b j E j ) with an inclusion L ⊂ ΩQ S which is seen as a non zero element of H 0(S, Hom(L, ΩQ S)). µ(ΩQ S) = P4(1)S KQ · O 3 = (−5 + 2)1 × 2 × 2 = −4 µ(L) = L · O 3 P4 (1)S = −3a −X b j . 3a +X b j ≥ 4, Firststep. We first intend to show the semi-stability inequality µ(L) ≤ µ(ΩQ S) that is and show that equality can occur only if L = OS(−2E0 +2E j ). By the general argument, this is ensured if L is the restriction of a line bundle on P4 i.e., a multiple of O P4(1)S. The conormal sequence for S in Q reads 0 → O P4(−2)S → ΩQ S → ΩS → 0. (7) If H 0(Hom(L, O proved, in its strict version. P4(−2)S)) 6= 0, then µ(L) ≤ µ(O From now on, we will assume that P4(−2)S) = −8 < µ(ΩQ S), and the desired inequality is H 0(Hom(L, O P4(−2)S)) = 0. Hence a non-zero element in H 0(S, Hom(L, ΩQ S)) gives a non-zero element in P2(a) ⊗ OS(X b j E j )). H 0(S, Hom(L, ΩS)) = H 0(S, ΩS ⊗ O In particular, we have an injection H 0(S, Hom(L, ΩS)) ,→ H 0(S \ ∪E j , ΩS ⊗ π∗O and from the Euler sequence with V := H 0(P2, O P2(a)) = H 0(P2 \ ∪p j , Ω P2(1)) P2(a)) = H 0(P2, Ω P2(a)) (8) 0 → O P2(a − 3) → V ∗ ⊗ O P2(a − 2) → TP2(a − 3) = Ω P2(a) → 0 this leads to a ≥ 2. 19 Let j be an integer between 1 and 5. To make use of the sequence (6), we consider a section ω j in H 0(P2, Ω P2(2)) ≃ C3 corresponding to p j , so that, by Lemma 4.6, π∗ω j ∈ H 0(S, ΩS ⊗ OS(2E0 − 2E j )). If H 0(Hom(L, OS(−2E0 + 2E j ))) 6= 0, then µ(L) ≤ µ(OS(−2E0 + 2E j )) = −4 = µ(ΩQ S) with equality if and only if L is isomorphic to OS(−2E0 + 2E j ). We assume from now on that H 0(S, Hom(L, ΩS)) 6= 0 and that for all j , H 0(Hom(L, OS(−2E0 + 2E j ))) = 0. The rational form π∗ω j yields the sequence and after a twist by L∗ a map 0 → OS(−2E0 + 2E j ) → ΩS → KS ⊗ OS(2E0 − 2E j ) H 0(Hom(L, ΩS)) → H 0(L∗ ⊗ KS ⊗ OS(2E0 − 2E j )) that is injective as H 0(Hom(L, OS(−2E0 + 2E j ))) = 0, and gives a curve C j in the linear system L∗ ⊗ KS ⊗ OS(2E0 − 2E j ) = π∗O P2(a − 1) ⊗ OS((b j − 1)E j ) ⊗ OS(Pk6= j (bk + 1)Ek). j the sum of irreducible components of C j that are not contracted by π and P4 (1)S = 3(a −1)+P b j −1+4 = µ(L∗). We hence have to show The curves C j are of degree d = C j ·O that d ≥ 4. Denote by C ′ by d ′ j ≤ d its degree. The class of C ′ π∗O P2(a′ j lies in some linear system j − 1) ⊗ OS((b′ j − 1)E j ) ⊗ OS(Xk6= j (b′ j k + 1)Ek). As C ′ j −C j consists of effective exceptional curves, a′ j = a, b′ j ≤ b j and b′ j k ≤ bk. As E j is a line that is not a component of C ′ The output is d = 3a +Pi bi ≥ 3a − 5d + 5 that is j , −b′ j d ≥ a 2 + 5 6 . +1 = C ′ j ·E j ≤ d ′ j . Hence, b j ≥ b′ j ≥ −d ′ j +1 ≥ −d +1. (9) Assume a = 2. Using the injection (8) and Lemma 4.6, a section in H 0(S, ΩS ⊗ π∗O P2(a)) can only vanish at order at most 2 along one exceptional divisor and does not vanish along the other divisors. Thus, for all i , it holds bi ≥ −2, and there is at most one negative coefficient bi . This bundle L = π∗O j =1 b j E j ) is of degree −6−Σbi , which is no more than −4, and we reach the desired inequality. P2(−2)⊗OS (−P5 We now assume a ≥ 3. It follows from (9) that 3 ≤ d. Assume that for some j , C ′ · E j ≥ 3. Then j ≥ 3. As S is the intersection of two quadrics and as a quadric does not contain any plane curve of j \ E j and d ′ j degree bigger than 2, the curve C ′ consider the plane P2 generated by the line E j and x. We get j cannot be a plane curve. Hence, choose a point x ∈ C ′ j ≥ #{P2 ∩C ′ d ′ j is a 3-plane then 〈C ′ j } ≥ C ′ j · E j + 1 ≥ 4. 〉 ∩ S contains C ′ j 〉 of C ′ If the linear span 〈C ′ + 1 ≥ 5, contra- j j is the whole P4. We can choose a line ℓ secant to C ′ dicting deg S = 4. Hence the linear span of C ′ j on two points and disjoint from E j , and we consider the 3-plane P3 generated by ℓ and E j . We infer d ′ j · E j + 2 ≥ 5 concluding for stability. ∪ E j , of degree d ′ j j 5(−1) = 4, with equality occurring if and only if L is isomorphic to π∗O However, we already know that this is not possible. · E j ≤ 2. Therefore −b′ j + 1 ≤ 2, b′ j ≥ −1. Hence, d = 3a +P bi ≥ 3 × 3 + P2(−3) ⊗ OS (P Ei ) = O P4(−1)S. ≥ #{P3 ∩C ′ Thus, for all j , we have C ′ j j } ≥ C ′ j Secondstep. We now prove stability, dealing with the only equality case we encountered. Namely we will show that L = OS(−2E0 + 2E j ) is not a subsheaf of ΩQ. Let C = 2E0 −P Ei be the class of the strict transform of the conic in P2 passing through the five points pi . Observe that L = O P4(−2) ⊗ OS(2C +2E j ), thus a section of L∗ ⊗ ΩQ is a section of ΩQ (2)S that vanishes at order 2 along E j and C . The proof of Proposition 4.7 will therefore be complete once the following lemma is proved: Lemma 4.8. Let s ∈ H 0(S, ΩQ(2)S) a non-vanishing section, and let ∆1, ∆2 be two secant lines. Then s does not vanish at order two along ∆1 and ∆2. 20 We will prove this lemma after some preliminary results. First, let us denote by Q2 a quadric cutting out S in Q. By simultaneous reduction of quadratic forms, we may assume that the quadric Q is defined by the identity matrix I and Q2 by some diagonal matrix D2. Since H 1(S, OS) = 0, the section s lifts to a section s ∈ H 0(S, Ω P4(2)S). We will have to consider affine cones: let U = C5 \ {0} and let p : U → P4. Whenever Z ⊂ P4 is a subvariety, we denote by bZ = p −1(Z ) its affine cone. The section s defines a section bs ∈ H 0(bS, Ω U bS), which can be written as bs =Pi , j ai , j Zi d Z j (Z j denotes the j -th coordinate function on C5). We denote by A the matrix (ai , j ). Sincebs is the pull-back of the section s, we have: Fact 4.9. A + t A belongs to the span of I and D2. We want to understand the scheme-theoretic vanishing locus of s. As a set, it is described by: Fact 4.10. Let u ∈ bS and x = [u] ∈ S. Then s(x) = 0 if and only if u is an eigenvector of A. Proof. The quadratic form Q yields an identification of C5 with its dual. Moreover, bs(u) identifies in Q is I , the tangent space of bQ at u has equation u itself. Thus s(x) vanishes if and only if these two the basis d Z j to the column vector Au. Since the coordinates have been chosen so that the matrix of linear forms define the same hyperplane, in other words if and only if Au is a multiple of u. (cid:3) At first order, the vanishing of s is characterized by: derivative d sx(X ) vanishes if and only if AU = 0. Fact 4.11. Let x = [u] ∈ S such that Au = 0 and u 6∈ Im(A). Let X = [U ] ∈ Tx S, with U ∈ TubS. Then, the bQ,x identifies with C5/C · u. The statement then follows from bQ,x. Proof. As the proof of Fact 4.10 shows, Ω the fact that d sx(X ) = AX ∈ C5/C · u ≃ Ω We now prove Lemma 4.8. Let π be the plane generated by ∆1 and ∆2. Since s vanishes along ∆1 (cid:3) Assume first that A has rank 2. Let x = [u] ∈ (∆1 ∪ ∆2) \ PImA. Since s vanishes at order two along and ∆2, by Fact 4.10, bπ must be included in an eigenspace of A. Replacing A by A − λ · I does not change the section s, thus we can assume thatbπ ⊂ ker A. Therefore the rank of A is at most 2. ∆1 ∪ ∆2, by Fact 4.11, we have TubS ⊂ ker A, and so equality of these subspaces. Since we may assume that x is not the intersection point ∆1 ∪ ∆2, we get a contradiction with the following fact: Fact 4.12. We have S ∩ π = ∆1 ∪ ∆2. For x ∈ ∆1 \ ∆2, Tx S 6= π, where Tx S ⊂ P4 denotes the embedded tangent space. Proof. Let Q ′ be any quadric containing S. We have Q ′ ∩ π = ∆1 ∪ ∆2 or Q ′ ∩ π = π, for degree reasons. The first point follows. Assume now that x ∈ ∆1 \ ∆2 and that Tx S = π. Let ℓ be a line through x and a point y in ∆2 \ ∆1. Once again, if Q ′ is a quadric containing S, then ℓ ∩Q ′ has multiplicity at least 2 at x (ℓ ⊂ π = Tx S ⊂ TxQ ′) and one at y, thus ℓ ⊂ Q ′. This implies that ℓ ⊂ S, contradicting the first point of the Fact. (cid:3) Assume now that A has rank 1. We will use the following observation: Fact 4.13. Let B be a square matrix which is the sum of an alternate matrix and a diagonal matrix. Assume that rankB = 1. Then, up to a permutation of the rows and columns, B can be written as a bloc-diagonal matrixµ β 0 0 ¶, with β a rank 1 matrix of order 2. 0 Proof. Write B = (bi , j ). Since a coefficient of B is non zero, a diagonal coefficient of B must be non zero, and assume that b1,1 6= 0. If all the other diagonal coefficients are 0, then we have bi , j = 0 for (i , j ) 6= (1,1) and the fact is true. In the other case, assume that b2,2 6= 0. We have b1,3 + b3,1 = b2,3 + b3,2 = b1,2 + b2,1 = 0 and b1,1b2,3 − b2,1b1,3 = b1,1b3,2 − b3,1b1,2 = 0, with b1,1, b2,2, b1,2 and b2,1 different from 0. It follows that b1,3 = b2,3 = b3,2 = b3,1 = 0. Similarly, all the coefficients bi , j are 0 except when i , j ≤ 2. (cid:3) 21 Now, A satisfies the hypothesis of Fact 4.13, and moreover the diagonal of A is a linear combination of I and D2. This implies that a linear combination of I and D2 has rank at most 2, contradicting the smoothness of S (in fact, S is smooth if and only if the quadrics in the pencil it defines all have rank at least 4). This ends the proof of Lemma 4.8. To complete the proof of Proposition 4.7, one has to consider rank 2 subsheaves in ΩQ. This case follows from the case of rank 1 subsheaves by the fact that ΩQ is self-dual (see the end of the proof of Proposition 4.4). REFERENCES [AN54] Yasuo Akizuki and Shigeo Nakano. Note on Kodaira-Spencer's proof of Lefschetz theorems. Proc. Japan Acad., 30:266 -- 272, 1954. I. Biswas. On the stability of homogeneous vector bundles. J. Math. Sci. Univ. Tokyo, 11(2):133 -- 140, 2004. [Bis04] [BK05] Michel Brion and Shrawan Kumar. Frobenius splitting methods in geometry and representation theory, volume [BK12] 231 of Progress in Mathematics. Birkhäuser Boston, Inc., Boston, MA, 2005. V. Balaji and János Kollár. Restrictions of stable bundles. In Compact moduli spaces and vector bundles, volume 564 of Contemp. Math., pages 177 -- 184. Amer. Math. Soc., Providence, RI, 2012. [Bot57] Raoul Bott, , Homogeneous vector bundles. Ann. of Math. (2), 66,203 -- 248, 1957. [CMP08] P. E. Chaput, L. Manivel, and N. Perrin. Quantum cohomology of minuscule homogeneous spaces. Transform. Groups, 13(1):47 -- 89, 2008. [D-F04] Carla Dionisi and Daniele Faenzi. A simple vanishing theorem for twisted holomorphic forms on Hermitian sym- metric varieties. http://dfaenzi.perso.math.cnrs.fr/publis/DF.vanishing.pdf. Igor V. Dolgachev. Classical algebraic geometry, a modern view. Cambridge University Press, Cambridge, 2012. [Dol12] [Fah89] Rachid Fahlaoui. Stabilité du fibré tangent des surfaces de del Pezzo. Math. Ann., 283(1):171 -- 176, 1989. [Fle84] Hubert Flenner. Restrictions of semistable bundles on projective varieties. Comment. Math. Helv., 59(4):635 -- 650, 1984. [GH94] Phillip Griffiths and Joseph Harris. Principles of algebraic geometry. Wiley Classics Library. John Wiley & Sons, Inc., New York, 1994. Reprint of the 1978 original. [Har80] Robin Hartshorne. Stable reflexive sheaves. Math. Ann., 254(2):121 -- 176, 1980. [HL10] Daniel Huybrechts and Manfred Lehn. The geometry of moduli spaces of sheaves. Cambridge Mathematical Li- brary. Cambridge University Press, Cambridge, second edition, 2010. [Hwa98] Jun-Muk Hwang. Stability of tangent bundles of low-dimensional Fano manifolds with Picard number 1. Math. Ann., 312(4):599 -- 606, 1998. [Kim91] Sung-Ock Kim. Noether-Lefschetz locus for surfaces. Transactions of the American Mathematical Society, [KO73] 324(1):369 -- 384, 1991. Shoshichi Kobayashi and Takushiro Ochiai. Characterizations of complex projective spaces and hyperquadrics. J. Math. Kyoto Univ., 13:31 -- 47, 1973. [Lan04] Adrian Langer. Semistable sheaves in positive characteristic. Ann. of Math. (2), 159(1):251 -- 276, 2004. [Laz04] Robert Lazarsfeld. Positivity in algebraic geometry, I. Classical setting: line bundles and linear series, volume 48 of Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics. Springer- Verlag, Berlin, 2004. Solomon Lefschetz. On certain numerical invariants of algebraic varieties with application to abelian varieties. Trans. Amer. Math. Soc., 22(3):327 -- 406, 1921. [Lef21] [Mat89] Hideyuki Matsumura. Commutative ring theory, volume 8 of Cambridge Studies in Advanced Mathematics. Cam- bridge University Press, Cambridge, second edition, 1989. Translated from the Japanese by M. Reid. [MR84] V. B. Mehta and A. Ramanathan. Restriction of stable sheaves and representations of the fundamental group. Invent. Math., 77(1):163 -- 172, 1984. [PW95] Thomas Peternell and Jarosław A. Wi´sniewski. On stability of tangent bundles of Fano manifolds with b2 = 1, J. Algebraic Geom., 4(2) :363 -- 384, 1995. [Ram66] S. Ramanan. Holomorphic vector bundles on homogeneous spaces. Topology, 5:159 -- 177, 1966. [Sno86] D. M. Snow. Cohomology of twisted holomorphic forms on Grassmann manifolds and quadric hypersurfaces. Math. Ann., 276(1):159 -- 176, 1986. [Sno88] D. M. Snow. Vanishing theorems on compact Hermitian symmetric spaces. Math. Z., 198(1):1 -- 20, 1988. [Ume78] H. Umemura. On a theorem of Ramanan. Nagoya Math. J., 69:131 -- 138, 1978. E-mail address: [email protected] SCHOOL OF MATHEMATICS, TIFR, HOMI BHABHA ROAD, BOMBAY 400005, INDIA E-mail address: [email protected] INSTITUT ÉLIE CARTAN DE LORRAINE, UNIVERSITÉ DE LORRAINE, 54506 VANDOEUVRE-LÈS-NANCY CEDEX, FRANCE E-mail address: [email protected] IRMAR, CAMPUS DE BEAULIEU, BÂT. 22, UNIVERSITÉ RENNES 1, 263 AV. GÉNÉRAL LECLERC, 35042 RENNES, FRANCE
1510.02731
3
1510
2017-01-02T20:52:01
On isolated singularities with noninvertible finite endomorphism
[ "math.AG" ]
We prove that if $\phi:(X,0)\to (X,0)$ is a finite endomorphism of an isolated singularity such that $\operatorname{deg}(\phi)\geq 2$ and $\phi$ is \'etale in codimension 1, then $X$ is $\mathbb{Q}$-Gorenstein and log canonical.
math.AG
math
ON ISOLATED SINGULARITIES WITH A NONINVERTIBLE FINITE ENDOMORPHISM YUCHEN ZHANG Abstract. We prove that if φ : (X, 0) → (X, 0) is a finite endomorphism of an isolated singularity such that deg(φ) ≥ 2 and φ is ´etale in codimension 1, then X is Q-Gorenstein and log canonical. 1. Introduction Let us start with an easy example. Let C be a smooth curve of genus g. By an argument using Riemann-Hurwitz Theorem, one can see that C has a finite endomorphism φ of degree ≥ 2 if and only if g ≤ 1. In this case, assume that there is an ample divisor H on C such that φ∗H is a multiple of H. Then φ induces a finite endomorphism on the cone X over C with polarization mH, where m is a sufficiently large integer. On the other hand, one can see easily by adjunction that a normal cone over a smooth curve of genus g has log canonical singularity if and only if g ≤ 1. This phenomenon is true in general. In this paper, we prove the following theorem: Theorem 1.1. Let (X, 0) be a normal projective variety with isolated singularity 0 ∈ X. Suppose that there exists a finite endomorphism φ : (X, 0) → (X, 0) such that deg φ ≥ 2 and φ is ´etale in codimension 1. Then X is Q-Gorenstein and log canonical. The assumption that X has isolated singularity is necessary. Otherwise, let X = E × V , where E is an elliptic curve and V is an arbitrary variety with a bad singularity. Then X has an induced noninvertible ´etale endomorphism from E. We briefly review the history of this problem. For the definitions of related terminologies, we refer to Section 2 and [BdFF12]. The surface case is studied in [Wahl90]. Let X be a normal surface and f : Y → X be the minimal resolution. The relative Zariski decomposition yields KY /X = P + N . Wahl's invariant is defined as the nonnegative intersection number −P 2, which is the key ingredient in the study of surfaces with noninvertible finite endomorphisms. A classification of such surfaces is given in [Favre10, FN05]. Wahl's invariant is generalized to higher Date: September 24, 2018. Key words and phrases. Isolated singularities, non-Q-Gorenstein, σ-decomposition, diminished base locus, movable modification. 1 2 YUCHEN ZHANG dimensions by Boucksom, de Fernex and Favre [BdFF12]. Due to the absence of minimal resolutions, they consider log discrepancy divisors on all birational models over X as Shokurov's b-divisor AX /X . The Zariski decomposition is replaced by the nef envelope EnvX (AX /X ). It can be shown that −(EnvX (AX /X ))n is a well-defined finite nonnegative number, which is called volBdFF(X). This volume behaves well under finite morphisms. In particular, they prove the following theorem: Theorem 1.2. [BdFF12, Theorem A and B][Fulger13, Proposition 2.12] For nor- mal isolated singularities (X, 0) with noninvertible finite endomorphism, volBdFF(X) = 0. Moreover, when X is Q-Gorenstein, volBdFF(X) = 0 if and only if X has log canonical singularity. The same theorem is obtained in [BH14] by analyzing the behavior of non-log- canonical centers under finite pullback. In [Fulger13], Fulger defines a courser volume volF(X) as the asymptotic order of growth of plurigenera, which coincides with volBdFF(X) when X is Q-Gorenstein. Unfortunately, in [yZhang14], the author produces a non-Q-Gorenstein isolated singularity (X, 0) such that volBdFF(X) = 0 while there is no boundary ∆ such that (X, ∆) is log canonical. We should remark that, in this example, X admits a small log canonical modification [OX12][BH14, Proposition 2.4]. The Q-Gorenstein case is further studied in [yZhang14, Section 3]. Specifically, the author shows that, like the surface case, volBdFF(X) can be calculated by an intersection number on a certain birational model f : Y → X, namely, the log canonical modification [OX12]. A key property of such a model is that KY +Ef is f - ample, where Ef is the reduced exceptional divisor. In the non-Q-Gorenstein case, the existence of the log canonical modification is conjectured to be true assuming the full minimal model program including the abundance conjecture, but has not yet been proved. In this paper, in order to show that a normal isolated singularity (X, 0) in The- orem 1.1 is indeed Q-Gorenstein, we consider a birational model over X called a movable modification (Theorem 3.1) where KY + Ef is f -movable. The existence of movable modifications is known to experts. However, we include a proof in Sec- tion 3. We introduce a new volume volmov(X) (Definition 4.6) using Nakayama's σ-decomposition. We show the following theorem: Theorem 1.3 (Proposition 3.5, Lemma 4.7, Theorem 4.8, 5.3 and 5.4). For normal isolated singularities (X, 0) and (Y, 0), (1) volmov(X) is a finite nonnegative number, and volmov(X) ≥ volBdFF(X) with equality when X is Q-Gorenstein. (2) If φ : (Y, 0) → (X, 0) is a finite morphism of degree d that is ´etale in codimension 1, then volmov(Y ) = d volmov(X). ON ISOLATED SINGULARITIES WITH A NONINVERTIBLE FINITE ENDOMORPHISM 3 (3) If φ : (X, 0) → (X, 0) is a finite endomorphism of degree ≥ 2 that is ´etale in codimension 1, then volmov(X) = 0. (4) If volmov(X) = 0, then X is numerically Q-Gorenstein. (5) Every numerically log canonical variety is Q-Gorenstein. As a byproduct while studying numerically Q-Gorenstein varieties, we obtain the following theorem which slightly generalizes [OX12, Theorem 1.1]. Theorem 1.4 (Theorem 3.6). Let X be a numerically Q-Gorenstein projective variety. Then there exists a log canonical model Y over X. The case that φ is not ´etale in codimension one is also interesting. Although X is not Q-Gorenstein in general (see example below), we still expect that there exists a boundary ∆ such that (X, ∆) is log canonical. It is known that X has klt singularities assuming that X is Q-Gorenstein [BdFF12, Theorem B]. 1H1 + p∗ Example 1.5. Let V = P1 × E where E is an elliptic curve. Let φ1 : P1 → P1 be raising to the fourth power and φ2 = [2] : E → E be multiplying by 2. Let H1 be a point in P1, H2 be a symmetric ample divisor on E (i.e., [−1]∗H2 = H2) and H = p∗ 2H2, where pi are the projections. Then φ = φ1 × φ2 is an endomorphism on V of degree 16 such that φ∗H ∼ 4H. We may take X be the cone over V with polarization H. Thus, φ induces an endomorphism on X of degree 16. But X is not Q-Gorenstein or log terminal. However, one may take ∆ be the 1(D1 +D2) for two different points D1 and D2 on P1 and see that KX +∆ cone over p∗ is Q-Cartier and log canonical. Question 1.6. Let (X, 0) be a normal isolated singularity. If φ : (X, 0) → (X, 0) is a noninvertible finite endomorphism that is not ´etale in codimension one, does there exist a boundary ∆ such that (X, ∆) is log canonical? The global counterpart of this problem is well studied. Let (V, H) be a nor- mal polarized projective variety. A finite endomorphism φ : V → V is called polarized if φ∗H is a multiple of H. The cone over a smooth variety V with the polarization H gives an isolated singularity as in the local case. The classifica- tion of polarized Kahler surfaces (also known as dynamic surfaces) is obtained in [FN05] and [sZhang06, Proposition 2.3.1]. The three dimensional case is studied in [Fujimoto02] and [FN07], where the classification of smooth projective 3-folds with κ(X) ≥ 0 that admit nontrivial endomorphism (which necessarily is ´etale) is given. Higher dimensions are studied in [NZ09, NZ10][GKP16, Theorem 1.21]. It is known that a Q-Gorenstein polarized projective variety with noninvertible endomorphism is log canonical [BH14]. We propose the global version of Question 1.6: 4 YUCHEN ZHANG Question 1.7. If φ is a noninvertible polarized finite endomorphism on (V, H), is V log Calabi-Yau? It is known that −KV is pseudo-effective when V is smooth [BdFF12, Theorem C]. There are also conditions that are weaker than being polarized that rise from dynamic systems, such as amplified and unity-free. We refer to [KR15] for the definitions and comparisons. Acknowledgement. The main work was done during Algebraic Geometry Sum- mer Research Institute organized by AMS and University of Utah. The author would like to thank to Tommaso de Fernex, Christopher Hacon, Lance Miller, Mircea Mustat¸a and Mihnea Popa for inspiring discussions, and the anonymous referee for pointing out several mistakes. The author is partially supported by NSF FRG Grant DMS-1265285. 2. Preliminaries Throughout this paper, we work over an algebraically closed field k of charac- teristic 0. 2.1. Movable divisors. Let f : Y → X be a projective morphism between normal varieties and D be an f -big Cartier divisors on Y . The base locus of D over X, Bs(D), is the co-support of the image of the following canonical map: f ∗f∗OY (D) ⊗ OY (−D) → OY . The stable base locus is defined to be B(D) = \ m≥1 Bs(mD)red. One can easily extend this definition to Q-Cartier divisors. We define the diminished base locus1 of D as B−(D) = [ ǫ>0 B(D + ǫH), where H is an f -ample divisor on Y . It can be shown that B−(D) is independent of the choice of H (see [ELMNP06, Section 1]. We say an f -pseudo-effective Q-Cartier divisor D on Y is f -mobile if codim(Bs(D)) ≥ 2. The f -movable cone Mov(Y /X) is the closure of the cone generated by f -mobile Cartier divisors in the finite dimensional space N 1(Y /X). We call an R-divisor D f -movable if D ∈ Mov(Y /X). Lemma 2.1. [Nakayama04, Theorem V.1.3] Let D be a Q-Cartier divisor on Y . Then D ∈ Mov(Y /X) if and only if B−(D) contains no divisor. (cid:3) 1This is called restricted base locus in [ELMNP06] and non-nef locus in [BDPP13] ON ISOLATED SINGULARITIES WITH A NONINVERTIBLE FINITE ENDOMORPHISM 5 Lemma 2.2. Let f : Y → X and g : Z → Y be two projective morphisms and φ = f ◦ g. If a Cartier divisor D on Z is φ-mobile, then D is also g-mobile. In particular, Mov(Z/X) ⊆ Mov(Z/Y ). Proof. Since the morphism g∗f ∗f∗g∗OZ (D) ⊗ OZ(−D) = φ∗φ∗OZ (D) ⊗ OZ (−D) ρφ−→ OZ factors through g∗g∗OZ(D) ⊗ OZ (−D) ρg−→ OZ , it follows that im ρφ ⊆ im ρg. Thus, if the co-support of im ρφ has codimension ≥ 2, then so is the co-support of im ρg. The lemma follows. (cid:3) The following lemma is a generalization of the well-known Negativity Lemma [KM98, Lemma 3.39]. Lemma 2.3. [Fujino11, Lemma 4.2] Let f : Y → X be a proper birational mor- phism where Y is a normal Q-factorial variety. Let E be an R-divisor on Y such that E is f -exceptional and E ∈ Mov(Y /X). Then E ≤ 0. (cid:3) 2.2. Shokurov's b-divisor. Let X be a normal variety. A Weil b-divisor W over X is the assignment to each birational model f : Y → X, a Weil divisor WY on Y that is compatible with pushforwards: if φ : Z → Y are two models over X, then φ∗(WZ) = WY . The Weil-divisor WY is called the trace of W on Y . We denote by Div(X ) the group of Weil b-divisors over X and define Q-Weil (R-Weil, resp.) b-divisor as elements of Div(X ) ⊗ Q (Div(X ) ⊗ R, resp.). We call the Weil b-divisor C a Cartier b-divisor over X if there exists a bi- rational model f : Y → X such that CY is Cartier and for every other model φ : Z 99K Y with common resolution W s ~⑥⑥⑥⑥⑥⑥⑥⑥ t ❆❆❆❆❆❆❆❆ Z /❴❴❴❴❴❴❴ Y we have CZ = s∗(t∗CY ). In this case, we say that f : Y → X is a determinant of C. Similarly, we define Q-Cartier b-divisor and R-Cartier b-divisor. A R-Cartier b-divisor C over X is called X-nef if there exists a (hence any) determinant f : Y → X such that CY is f -nef. We call a R-Weil b-divisor W X-nef if W is a limit of X-nef R-Cartier b-divisors, where the limit is taken in the numerical class of every model. The following lemma gives a characterization of X-nef R-Weil b-divisors. Lemma 2.4. [BdFF12, Lemma 2.10] A R-Weil b-divisor W is X-nef if and only if WY ∈ Mov(Y /X) on every Q-factorial model Y . (cid:3) ~ / 6 YUCHEN ZHANG Let W1 and W2 be two R-Weil b-divisors. We say W1 ≥ W2, if for every model Y , we have (W1)Y ≥ (W2)Y . 2.3. Nakayama's σ-decomposition. Let X be a Q-factorial projective normal variety over S and D be an S-big R-divisor on X. For every prime divisor Γ on X, we define σΓ(D) = inf{multΓ ∆∆ ≡S D, ∆ ≥ 0}. Since D is S-big, there exists an effective divisor ∆ that is numerically equivalent to D. Hence, the infimum above is not taken over an empty set. Remark 2.5. By the observation of Nakayama, this definition can be generalized to S-pseudo-effective R-divisors. Let D be an S-pseudo-effective divisor on X. Fix an S-ample divisor A on X. Since for every ǫ > 0, D + ǫA is S-big, we can define σΓ(D) = limǫ↓0 σΓ(D + ǫA). It is shown in [Nakayama04, Lemma III.1.4-5] that σΓ(D) is independent of the choice of A and only depends on the numerical class of D. However, an example is given in [Lesieutre15] that this limit can be ∞ when S is not a point. In this paper, we only use the σ-decomposition in the case that X is birational to S, hence every divisor is S-big (see [Nakayama04, Lemma III.1.4(2)]). It is shown in [Nakayama04, Lemma III.4.2] that there are finitely many prime divisors Γ on X such that σΓ(D) > 0, which leads us to the following definition. Definition 2.6. Let D be an S-big R-divisor on a Q-factorial projective variety X over S. We define Nσ(D) = X σΓ(D)Γ and Pσ(D) = D − Nσ(D), where Pσ(D) and Nσ(D) are called the positive and negative part of the σ-decomposition, respectively. Notation 2.7. To be precise, the σ-decomposition here should be written as Pσ/S and Nσ/S as we are using the relative version. However, in order to make our notation concise, we will omit the base S when it is clear from the context. We record some properties of the σ-decomposition as below. Proposition 2.8. Let D be an S-big R-divisor on a Q-factorial variety X. Let f : Y → X be a proper birational morphism, where Y is normal. (1) Nσ(D) = 0 if and only if D ∈ Mov(X/S). (2) Pσ(D) is the largest R-divisor in Mov(X/S) that is no greater than D. (3) f∗Pσ(f ∗D) = Pσ(D). In other words, Pσ(f ∗D) defines a R-Weil b-divisor over X. (4) σΓ is a continuous function on the big cone of X over S. So is Pσ, where the convergence is coefficiently-wise. ON ISOLATED SINGULARITIES WITH A NONINVERTIBLE FINITE ENDOMORPHISM 7 Proof. (1) and (2) are [Nakayama04, Proposition III.1.14]. Theorem III.2.5(1)]. (4) is [Nakayama04, Lemma III.1.7(1)]. (3) is [Nakayama04, (cid:3) Inspired by the above proposition, we give the definition of σ-closure. Definition 2.9. Let Y be a Q-factorial model over X and D be an S-big R-divisor on Y . The σ-closure Pσ(D) is an R-Weil b-divisor such that: for every model Z over X, let W be a model dominating Y and Z. W s ~⑥⑥⑥⑥⑥⑥⑥⑥ t ❆❆❆❆❆❆❆❆ Y /❴❴❴❴❴❴❴ Z Then the trace of Pσ(D) on Z is t∗Pσ(s∗D). It is well-defined by Proposition 2.8(3). Lemma 2.10. The σ-closure is X-nef. Proof. The lemma follows directly from Proposition 2.8(2) and Lemma 2.4. (cid:3) 2.4. Pullback of Weil divisors. We recall the definition of pullback of Weil di- visors from [dFH09, BdFF12] as below. Let X be a normal variety, D be a R-Weil divisor on X and E be a prime divisor over X. We define the valuation vE(D) = vE(OX (−D)) = vE(OX (⌊−D⌋)) as vE(D) = min{vE(φ) φ ∈ OU (⌊−D⌋), U ∩ cX (E) 6= ∅}. If D is Cartier, we see that vE(D) is the usual valuation. Suppose f : Y → X is a proper birational morphism and Y is normal. The natural pullback f ♮D is defined as f ♮D = X vE(D)E, where the sum runs through all prime divisors on Y . OY (−f ♮D) = (OX (−D)·OY )∨∨. The natural pullback is also known as Z(OX (−D))f in [BdFF12], where Z(OX (−D)) is viewed as a Weil b-divisor consisting of natural pullbacks. It is easy to see that In general, the natural pullbacks do not behave well under composition. Lemma 2.11. [dFH09, Lemma 2.7] Let f : Y → X and g : Z → Y be two proper birational morphisms between normal varieties, D be an R-Weil divisor on X. Then (f ◦ g)♮D − g♮(f ♮D) is not necessarily zero . However, it is effective and g-exceptional. Moreover, if OX (−D) · OY is an invertible sheaf on Y , then (f ◦ g)♮D = g♮(f ♮D). (cid:3) Lemma 2.12. [BdFF12, Lemma 1.8 and 2.10] ~ / 8 YUCHEN ZHANG (1) If OX (−D) · OY is an invertible sheaf, then −f ♮D is relatively globally generated over X. In other words, the following canonical homomorphism is surjective. f ∗f∗OY (−f ♮D) → OY (−f ♮D). (2) If −f ♮D is Q-Cartier, then −f ♮D ∈ Mov(Y /X). Proof. (1) Since OX (−D) · OY is the image of f ∗OX (−D) under the homo- morphism f ∗OX (−D) → f ∗OX , we have a surjection f ∗OX (−D) → OX (−D) · OY . It is obvious that f∗OY (−f ♮D) = OX (−D). The statement follows. (2) Suppose −mf ♮D is Cartier. Let g : Z → X be a log resolution of X such that OX (−D) · OZ is an invertible sheaf and g factors through f via φ : Z → Y . Such log resolution exists by [dFH09, Theorem 4.2]. By part (1), −mg♮D is relatively globally generated over X. Hence, −mf ♮D = φ∗(−mg♮D) is f -mobile. (cid:3) It is shown in [dFH09, Lemma 2.8] that for every positive integer m, we have f ♮D ≥ 1 m f ♮(mD). Thus, we can define the pullback of D as a R-Weil divisor, f ∗D = lim inf m→∞ f ♮(mD) m . It is not hard to see that the lim inf above is actually a limit ([BdFF12, Lemma 2.1]). Lemma 2.13. If Y is Q-factorial, then −f ∗D ∈ Mov(Y /X). Proof. This is obvious by Lemma 2.12 and the definition. (cid:3) Lemma 2.14. Let f : Y → X and g : Z → Y be two proper birational morphisms between normal varieties. Then for every Q-Weil divisor D on X, (f ◦ g)∗D − g∗(f ∗D) is an effective g-exceptional divisor. Proof. The lemma follows from Lemma 2.11 and definition. (cid:3) We will use the following notation from [BdFF12, Remark 2.4]. Definition 2.15. EnvX (D) is the R-Weil b-divisor whose trace on every model Y is −f ∗(−D). It is shown in [BdFF12, Corollary 2.13] that EnvX (D) is the largest X-nef R-Weil b-divisor W such that WX = D. Lemma 2.16. In the setting of Lemma 2.14, if Z is Q-factorial, then −(f ◦g)∗D = Pσ(−g∗(f ∗D)). Hence, EnvX (−D) = Pσ(−f ∗D). ON ISOLATED SINGULARITIES WITH A NONINVERTIBLE FINITE ENDOMORPHISM 9 Proof. By Lemma 2.13, −(f ◦ g)∗D is a movable R-divisor that is no greater than −g∗(f ∗D). Using Lemma 2.8, we have −(f ◦ g)∗D ≤ Pσ(−g∗(f ∗D)). On the other hand, Pσ(−f ∗D) is an X-nef R-Weil b-divisor whose trace on X is D. Hence, EnvX (−D) ≥ Pσ(−f ∗D). The lemma follows. (cid:3) Lemma 2.17. [BdFF12, Corollary 2.13] If EnvX (D) is R-Cartier with a determi- nant on Y , then the trace of EnvX (D) on Y is X-nef. (cid:3) We record the following Negativity Lemma in the context of b-divisors. Lemma 2.18. [BdFF12, Proposition 2.12] Let W be an X-nef R-Weil b-divisor over X and f : Y → X be a birational model. Then WY ≤ −f ∗(−WX ). (cid:3) 2.5. Nef envelope of R-Weil b-divisors and volume. All the definitions below are from [BdFF12], where the reader can find the details. Let W be an R-Weil b-divisor over X. We define the nef envelope EnvX (W) of W to be the largest X-nef R-Weil b-divisor Z such that Z ≤ W, if exists. Let C be an R-Cartier b-divisor over X whose center on X is 0 ∈ X. The self-intersection Cn = Cn Y if Y is a determinant of C. It is obvious that Cn is independent of the choice of Y . If W is X-nef whose center on X is 0 ∈ X, we define W n = inf{Cn} where the infimum is taken for all X-nef R-Carter b-divisors such that C ≥ W. We copy the following properties from [BdFF12]. Proposition 2.19. [BdFF12, Theorem 4.14 and Proposition 4.16] Let W be an X-nef R-Weil b-divisor whose center on X is 0 ∈ X. (1) If W1 ≤ W2 are two X-nef R-Weil b-divisors over X, then W n (2) W n = 0 if and only if W = 0. (3) Let φ : (Y, 0) → (X, 0) be a finite map of degree d between isolated singu- (cid:3) larities. Then (φ∗W)n = d(W n). 1 ≤ W n 2 ≤ 0. For a model f : Y → X, we fix canonical divisors such that f∗KY = KX. For every positive integer m, we define the m-th limiting log discrepancy divisor as Am,Y /X = KY + Ef − f ♮(mKX ), 1 m where Ef is the reduced exceptional divisor. We denote by Am,X /X the correspond- ing Q-Weil b-divisor. Similarly, we define the log discrepancy divisor as AY /X = KY + E − f ∗KX and AX /X to be the corresponding R-Weil b-divisor. It is shown in [BdFF12, yZhang14] that if X has isolated singularity, then EnvX (AX /X ) and EnvX (Am,X /X) exist. We define volm(X) = − EnvX (Am,X /X )n and volBdFF(X) = − EnvX (AX /X )n. It is known that these volumes are finite nonnegative numbers. 10 YUCHEN ZHANG In the case that X is Q-Gorenstein, it is shown in [yZhang14, Theorem 3.2] that volBdFF(X) = volm(X) = −An Y /X for m sufficiently large and divisible, where Y is the log canonical modification of X. When X is non-Q-Gorenstein, [yZhang14, Corollary 4.3] shows that for every m ≥ 2, one can always pick a boundary ∆ on X such that volm(X) is calculated on the log canonical modification of the pair (X, ∆). We do not know whether volBdFF(X) can be calculated by intersection number on some model. The following theorem gives a criteria for log canonical singularity. Theorem 2.20. [yZhang14, Corollary 4.6] If volm(X) = 0 for some (hence any multiple of an) integer m ≥ 1, then there exists a boundary ∆ on X such that (X, ∆) is log canonical. (cid:3) 2.6. Numerically Q-Cartier divisors. The numerically Q-Cartier divisors are defined in [BdFF12] and are further studied in [BdFFU14], which behave like Q- Cartier divisors under birational pullbacks. We give the following definition which generalizes Mumford's numerical pullback. Definition 2.21. Let X be a normal variety. (1) A Weil divisor D on X is numerically Cartier if there exists a resolution of singularities f : Y → X and an f -numerically trivial Cartier divisor D′ on Y such that f∗D′ = D. (2) A Q-Weil divisor is numerically Q-Cartier if some multiple is numerically Cartier. (3) When the canonical divisor KX is numerically Q-Cartier, we say X is nu- merically Q-Gorenstein. Proposition 2.22. Let D be a Weil divisor on X. (1) D is numerically Q-Cartier if and only if EnvX (−D) = − EnvX (D). (2) Suppose X has rational singularities. Then D is numerically Q-Cartier if and only if D is Q-Cartier. (3) Suppose D is numerically Q-Cartier. Let f : Y → X be a birational model such that f ∗D is Q-Cartier. Then f ∗D is f -numerically trivial and − EnvX (−D) = f ∗D. Proof. (1) is [BdFFU14, Proposition 5.9] and (2) is [BdFFU14, Theorem 5.11]. For (3), let g : Z → Y be any Q-factorial birational model over Y and φ = f ◦ g. By Lemma 2.16, we have −φ∗D = Pσ(−g∗(f ∗D)) ≤ −g∗(f ∗D), and −φ∗(−D) = Pσ(−g∗(f ∗(−D))) ≤ −g∗(f ∗(−D)). ON ISOLATED SINGULARITIES WITH A NONINVERTIBLE FINITE ENDOMORPHISM 11 Applying (1), the sum yields 0 = −φ∗D − φ∗(−D) ≤ −g∗(f ∗D) − g∗(f ∗(−D)) = 0. Hence, EnvX (D) = f ∗D and − EnvX (D) = −f ∗D. By Lemma 2.17, both f ∗D and −f ∗D are f -nef. Thus, f ∗D is f -numerically trivial. (cid:3) 3. Partial Resolutions 3.1. Movable modification of non-Q-Gorenstein varieties. The construction of our volume depends on the following theorem which is known to experts. Theorem 3.1. There exists a birational morphism f : Y → X such that (1) Y is Q-factorial, (2) (Y, Ef ) is dlt, and (3) KY + Ef ∈ Mov(Y /X), where Ef is the reduced exceptional divisor. Moreover, AY /X ≤ 0. We call f : Y → X a movable modification of X. Remark 3.2. Theorem 3.1 does not require that X is Q-Gorenstein. It is proved in [OX12] that in the setting of KX + ∆ being Q-Cartier, one may further require that KY + Ef + ∆ is f -semi-ample, where ∆ is the strict transform of ∆ on Y . The general case of semi-ampleness is true if one assume the full minimal model program (MMP) including abundance. Proof. Let g : Z → X be a log resolution and Eg be the reduced exceptional divisor. Let H be a g-ample divisor on Z. We run the (KZ + Eg)-MMP with scaling H over X. We obtain a sequence of flips and divisorial contractions: and a decreasing sequence Z = Z0 99K Z1 99K · · · , λi = inf{s ∈ RKZi + Egi + sHi is nef over X}. We know that for every λi ≥ t ≥ λi+1, KZi + Egi + tHi is relatively semi-ample over X and limi→∞ λi = 0 (see [OX12, Lemma 2.6]). For every divisor E ⊂ B−(KZ + Eg/X) or equivalently every component E of Nσ(KZ + Eg), there is some t > 0 such that E ⊂ B(KZ + Eg + tH/X). We may find an i such that λi ≥ t ≥ λi+1. Since KZi + Egi + tHi is relatively semi-ample over X, E must be contracted on Zi. Since there are finitely many such E, we conclude that there is a model Zj such that B−(KZj + Egj /X) contains no divisor. By Lemma 2.1, we have KZj + Egj ∈ Mov(Zj/X). The theorem follows by setting Y = Zj. For the moreover part, by Lemma 2.13, −f ∗KX ∈ Mov(Y /X), hence so is KY + (cid:3) Ef − f ∗KX . By Lemma 2.3, we have AY /X = KY + Ef − f ∗KX ≤ 0. 12 YUCHEN ZHANG 3.2. Log canonical modification of numerically Q-Gorenstein varieties. Definition 3.3. For a numerically Q-Gorenstein variety X, we say X is numerically log canonical (numerically log terminal, resp.) if for every exceptional divisor E on φ : Z → X, ordE(KZ − f ∗KX) ≥ −1 (> −1, resp.). Remark 3.4. The above definition coincide with Mumford's numerically pullbacks [KM98, Notation 4.1] on surfaces. The following proposition generalizes the well-known property in the surface case [KM98, Notation 4.1][dFH09, Proposition 7.14]. Proposition 3.5. If X is numerically log canonical, then X is Q-Gorenstein. Proof. Let f : Y → X be a movable modification of X. Then KY + Ef − f ∗KX ≤ 0 by Lemma 2.3. Since X is numerically log canonical, there must be KY + Ef = f ∗KX . Thus, (Y, Ef ) is a dlt pair such that KY + Ef is f -numerically trivial. By the abundance theorem [FG14, Theorem 4.9], KY + Ef is f -semi-ample. Let φ : Z = ProjX Lm f∗OY (m(KY + Ef )) → X be the log canonical modification over X. Then KZ + Eφ is φ-ample. On the other hand, KZ + Eφ = φ∗KX is φ- numerically trivial by Proposition 2.22(3). We conclude that φ is an isomorphism. In particular, X is Q-Gorenstein. (cid:3) The existence of log canonical modifications is predicted by full relative minimal model program and is proved for a pair (X, ∆ = P ai∆i) such that KX + ∆ is Q-Cartier and ai ∈ [0, 1] [OX12]. The same idea applies in the case that KX + ∆ is numerically Q-Cartier. For the reader's convenience, we include a sketch below for the case ∆ = 0. Theorem 3.6. Let X be a numerically Q-Gorenstein normal variety. Then there exists a unique birational model f : Y → X, such that (1) KY + Ef is a f -ample Q-Cartier divisor, and (2) (Y, Ef ) is log canonical, where Ef is the reduced exceptional divisor. Proof. Step 1. By Theorem 3.1, there exists a movable modification φ : Z → X such that Z is Q-factorial, KZ + Eφ ∈ Mov(Z/X) and (Z, Eφ) is dlt. By Lemma 2.1, we know that B−(KZ +Eφ/X) contains no divisor. We denote the complement of the support of φ(−⌊AZ/X ⌋) by Xlc. Step 2. We show that if E is an exceptional divisor on Z such that φ(E) ⊆ X\Xlc, then ordE(AZ/X ) < 0. If this is not true, by Koll´ar-Shokurov's Connectedness Lemma [KM98, Corollary 5.49], there exist exceptional divisors E0 and E1 with centers in X\Xlc such that ordE0 (AZ/X ) = 0, ordE1(AZ/X ) < 0 and E0 ∩ E1 6= ∅. Then (AZ/X + ǫH)E0 is not effective for 0 < ǫ ≪ 1 and H ample. Since φ∗X ON ISOLATED SINGULARITIES WITH A NONINVERTIBLE FINITE ENDOMORPHISM 13 is numerically trivial by Proposition 2.22(3), we have E0 ⊆ B−(KZ + Eφ/X), a contradiction. Step 3. Set B = Eφ − ǫAZ/X for some small ǫ such that B ≥ 0. We show that (Z, B) has a good minimal model over X. The idea is to apply [HX13, Theorem 1.1] over the open subset Xlc. A good minimal model exists over Xlc by Proposition 3.5 and [HX13, Lemma 2.11]. One can check that no strata of ⌊B⌋ are contained in φ−1(X\Xlc) using Step 2. Step 4. Since KZ + B = (1 + ǫ)(KZ + Eφ) − ǫφ∗KX ≡f (1 + ǫ)(KZ + Eφ), a good minimal model of KZ + B is also a good minimal model of KZ + Eφ. Hence the canonical ring R = Lm φ∗OZ (m(KZ + Eφ)) is finitely generated over X and we may take Y = ProjX R. (cid:3) 4. Movable Volume We start with a lemma showing the behavior of Zariski decomposition in the sense of [BdFF12] on a movable modification. Lemma 4.1. Let X be a normal variety which is not necessarily Q-Gorenstein. Let f : Y → X be a movable modification. Then the trace of EnvX (AX /X ) on Y is AY /X . Proof. Consider the R-Weil b-divisor P− = Pσ(KY + Ef ) + EnvX (−KX ). Let g : Z → Y be a projective birational morphism and φ = g ◦ f . Since (Y, Ef ) is dlt, we have g∗(KY + Ef ) ≤ KZ + Eφ. Hence, (P−)Z = Pσ(g∗(KY + Ef )) − φ∗KX ≤ g∗(KY + Ef ) − φ∗KX ≤ AZ/X . Thus, P− is an X-nef R-Weil b-divisor that is less than or equal to AX /X. By the definition of nef envelope, we have P− ≤ EnvX (AX /X ). Taking the trace on Y , we obtain AY /X = (P−)Y ≤ (EnvX (AX /X))Y ≤ (AX /X )Y = AY /X . The lemma follows. (cid:3) Inspired by the lemma above, we give the following definition: Definition 4.2. Let X be a normal variety and f : Y → X be a movable modifi- cation. The diminished positive part P− is an R-Weil b-divisor over X: P− = Pσ(KY + Ef ) + EnvX (−KX). Lemma 4.3. Let f : Y → X be a movable modification, g : Z → Y be a Q-factorial birational model and φ = f ◦ g. Then the trace of P− on Z is Pσ(KZ + Eφ)− φ∗KX . In particular, P− is independent of the choice of Y . Proof. We only need to show that Pσ(KZ + Eφ) = Pσ(g∗(KY + Ef )). 14 YUCHEN ZHANG Since (Y, Ef ) is dlt, we have KZ + Eφ ≥ g∗(KY + Ef ). Hence, Pσ(KZ + Eφ) ≥ Pσ(g∗(KY + Ef )). Now we have Pσ(g∗(KY + Ef )) ≤ Pσ(KZ + Eφ) ≤ KZ + Eφ. As we know that KY + Ef ∈ Mov(Y /X), by Proposition 2.8(3), g∗(KY + Ef ) − Pσ(g∗(KY + Ef )) is g-exceptional, and so is KZ + Eφ − Pσ(g∗(KY + Ef )). We obtain that Pσ(KZ + Eφ) − g∗(KY + Ef ) must be g-exceptional. Notice that Pσ(KZ +Eφ)−g∗(KY +Ef ) ∈ Mov(Z/Y ). By Lemma 2.3, we have Pσ(KZ +Eφ) ≤ g∗(KY + Ef ). Proposition 2.8(2) gives the desired inequality. (cid:3) Remark 4.4. If we assume the termination of MMP for dlt pairs, then there exists a movable modification with KY + Ef nef over X, which is called a dlt modification [OX12]. In this case, it is not hard to see that Pσ(KY + Ef ) is a Q-Cartier b-divisor determined by f . Lemma 4.5. Let φ : Z → X be a Q-factorial birational model over X not necessar- ily factoring through a movable modification. Then (P−)Z ≤ Pσ(KZ + Eφ)− φ∗KX . Proof. Let f : Y → X be a movable modification and W be a common resolution of Y and Z as below: W s ~⑥⑥⑥⑥⑥⑥⑥⑥ t ❆❆❆❆❆❆❆❆ Z Y We only need to show that t∗Pσ(s∗(KY + Ef )) ≤ Pσ(KZ + Eφ). It is clear that t∗Pσ(s∗(KY + Ef )) ≤ t∗(s∗(KY + Ef )) ≤ t∗(KW + Ef ◦s) = KZ + Eφ and t∗Pσ(s∗(KY + Ef )) ∈ Mov(Z/X). The lemma now follows from Proposition 2.8(2). (cid:3) Definition 4.6. Suppose that (X, 0) has only isolated singularity at 0 ∈ X. We define the movable volume volmov(X) = −(P−)n. Lemma 4.7. We have the following inequality between the volumes: 0 ≤ volBdFF(X) ≤ volmov(X) < +∞. Proof. According to [Koll´ar13, 1.12.1], there exists a log resolution s : W → X with an effective s-exceptional divisor H on W such that −H is s-ample. We fix an integer m such that KW + Es − mH and −s∗KX − mH are both s-ample. For ~ ON ISOLATED SINGULARITIES WITH A NONINVERTIBLE FINITE ENDOMORPHISM 15 any model φ : Z → X dominating a movable model and factoring through s via t : Z → W , we have (P−)Z = Pσ(KZ + Eφ) − φ∗KX ≥ Pσ(t∗(KW + Es)) + Pσ(−t∗(s∗KX)) ≥ Pσ(t∗(KW + Es − mH)) + Pσ(t∗(−s∗KX − mH)) = t∗(KW + Es − mH) + t∗(−s∗KX − mH) = t∗(AW/X − 2mH). Thus, by Proposition 2.19, volmov(X) = −(P−)n ≤ −(AW/X − 2mH)n < +∞. By the proof of Lemma 4.1, we have P− ≤ EnvX (AX /X ). Hence, volmov(X) = −(P−)n ≥ −(EnvX (AX /X))n = volBdFF(X) ≥ 0. (cid:3) Theorem 4.8. volmov(X) = 0 if and only if X is Q-Gorenstein and log canonical. Proof. Suppose that volmov(X) = −(P−)n = 0. By Proposition 2.19, we have that P− = 0. Let f : Y → X be a movable modification, g : Z → Y be any Q-factorial model over Y and φ = f ◦ g. Then 0 = (P−)Y = AY /X , hence KY + Ef = f ∗KX. In particular, f ∗KX is Q-Cartier. For the trace on Z, we have 0 = (P−)Z = Pσ(g∗(KY + Ef )) − φ∗KX = Pσ(g∗(f ∗KX )) + Pσ(−g∗(f ∗KX)) ≤ g∗(f ∗KX ) − g∗(f ∗KX ) = 0, In particular, we obtain that where the second row follows from Lemma 2.16. EnvX (−KX) = Pσ(−f ∗KX) is a Q-Cartier b-divisor determined by f . Thus, −f ∗KX is f -nef by Lemma 2.17. Similarly, since Pσ(g∗(f ∗KX )) = g∗(f ∗KX ), we have that f ∗KX is also f -nef. Thus, f ∗KX is f -numerically trivial and X is numerically Q-Gorenstein. The inequality 0 = P− ≤ AX /X yields that X is nu- merically log canonical. We may apply Proposition 3.5 to conclude the proof. The other direction is obvious. (cid:3) 5. Finite Endomorphisms We briefly review the pullback of Weil b-divisors. Let φ : Y → X be a generically finite dominant morphism between normal varieties. Every divisorial valuation ν on Y induces a divisorial valuation φ∗ν via the natural inclusion of the function field φ∗ : k(X) ֒→ k(Y ) given by φ∗ν(f ) = ν(f ◦ φ) [BdFF12, Lemma 1.13]. In other words, suppose that F is a prime divisor on X ′ over X. Then there is a birational model Y ′ over Y such that φ lifts to a morphism φ′ : Y ′ → X ′. We may obtain such Y ′ by blowing up the indeterminancy of the 16 YUCHEN ZHANG rational map Y 99K X ′. φ′(E) = F , then φ∗νE is a scalar multiple of νF with the scalar νE(φ′∗F ). If E is an irreducible component of (φ′)−1F such that Let W be an R-Weil b-divisor on X. We define the pullback of W as the R-Weil b-divisor φ∗W such that νE(φ∗W ) = (φ∗νE)(W ). For a normal variety Z, we denote by KZ the Weil b-divisor given by a canonical divisor on each model and EZ the Weil b-divisor given by the reduced exceptional divisor over Z on each model. Lemma 5.1. [BdFF12, Lemma 2.19 and 3.3] Let φ : Y → X be a finite dominant morphism between normal varieties. Then (1) If D is an R-Weil divisor on X, then EnvY (φ∗D) = φ∗ EnvX (D). (2) If W is an R-Weil b-divisor over X such that EnvX (W) is well-defined, then EnvY (φ∗W) = φ∗ EnvX (W). (3) If E is a prime exceptional divisor over Y , then νE(KY + EY ) = νE(φ∗(KX + EX )). (4) If φ is ´etale in codimension 1, then KY + EY = φ∗(KX + EX ). Proof. Part (1) and (2) are the same as [BdFF12, Lemma 2.19]. Let E be an exceptional divisor over Y . Let X ′ be a smooth model over X such that the center of φ∗νE on X ′ is a divisor F . Let Y ′ be a smooth model over Y such that E is a divisor on Y ′ and φ′ : Y ′ 99K X ′ is a morphism. We summarize this in the following diagram: E ⊂ Y ′ g Y φ′ φ X ′ ⊃ F f / X. We have φ∗νE = bνF where b = νE(φ′∗F ). Hence the ramification order of φ′ at the generic point of E is b − 1. We get νE(KY ′ ) = νE(φ′∗KX ′) + b − 1 = bνF (KX ′) + b − 1. Thus νE(KY + EY ) = νE(KY ′ ) + 1 = bνF (KX ′) + b = φ∗νE(KX + EX ). Part (3) follows. If φ is ´etale in codimension 1, then KY = φ∗KX. Hence, KY + EY and φ∗(KX + (cid:3) EX ) also coincide on non-exceptional valuations. Lemma 5.2. Let f : Y → X be a movable modification. Then EnvX (KX + EX ) = Pσ(KY + Ef ). In particular, P− = EnvX (KX + EX ) + EnvX (−KX ). / /     / ON ISOLATED SINGULARITIES WITH A NONINVERTIBLE FINITE ENDOMORPHISM 17 Proof. The proof is similar to Lemma 4.1. Let g : Z → X be any Q-factorial model factoring through f via φ : Z → Y . By Lemma 4.3, the trace on Z satisfies: (Pσ(KY + Ef ))Z = Pσ(KZ + Eg) ≤ KZ + Eg. Hence, (EnvX (KX + EX ))Z ≤ (Pσ(KY + Ef ))Z . On the other hand, Pσ(KY + Ef ) ≤ KX + EX . By the definition of nef envelope, we have EnvX (KX + EX ) ≥ Pσ(KY + Ef ). (cid:3) We are ready to prove the main theorem. Theorem 5.3. Let φ : (Y, 0) → (X, 0) be a finite morphism of degree d between normal isolated singularities such that φ is ´etale in codimension 1. Then (1) P−,Y = φ∗P−,X . (2) volmov(Y ) = d volmov(X). Proof. By Lemma 5.1 and 5.2, P−,Y = EnvY (KY + EY ) + EnvY (−KY ) = EnvY (φ∗(KX + EX )) + EnvY (−φ∗KX) = φ∗ EnvX (KX + EX )) + φ∗ EnvX (−KX) = φ∗P−,X . Part (2) now follows from Lemma 2.19. (cid:3) Theorem 5.4. If φ : (X, 0) → (X, 0) is a finite endomorphism of degree d ≥ 2 such that φ is ´etale in codimension 1, then volmov(X) = 0. Proof. By Theorem 5.3, volmov(X) = d volmov(X). But volmov(X) is a nonnegative finite number. There must be volmov(X) = 0. (cid:3) References [BdFF12] S´ebastien Boucksom, Tommaso de Fernex, and Charles Favre. The volume of an isolated singularity. Duke Math. J., 161:1455 -- 1520, 2012. [BdFFU14] S´ebastien Boucksom, Tommaso de Fernex, Charles Favre, and Stefano Urbinati. Valuation spaces and multiplier ideals on singular varieties. Recent Advances in Algebraic Geometry, 417:29, 2014. [BDPP13] S´ebastien Boucksom, Jean-Pierre Demailly, Mihai Paun, and Thomas Peternell. The pseudo-effective cone of a compact Kahler manifold and varieties of negative Kodaira dimension. J. Algebraic Geom., 22:201 -- 248, 2013. [BH14] Amael Broustet and Andreas Horing. Singularities of varieties admitting an endo- morphism. Math. Ann., 360(1-2):429 -- 456, 2014. [dFH09] Tommaso de Fernex and Christopher Hacon. Singularities on normal varieties. [ELMNP06] Compos. Math., 145(2):393 -- 414, 2009. Lawrence Ein, Robert Lazarsfeld, Mircea Mustat¸a, Michael Nakamaye, and Mi- hnea Popa. Asymptotic invariants of base loci. Annales de l'institute Fourier, 56(6):1701 -- 1734, 2006. 18 YUCHEN ZHANG [Favre10] Charles Favre. Holomorphic endomorphisms of singular rational surfaces. Publ. [FG14] [FN05] [FN07] Mat., 54(2):389 -- 432, 2010. Osamu Fujino and Yoshinori Gongyo. Log pluricanonical representations and the abundance conjecture. Compos. Math., 150(4):593 -- 620, 2014. Yoshio Fujimoto and Noboru Nakayama. Compact complex surfaces admitting non- trivial surjective endomorphisms. Tohoku Math. J.(2), 57(3):395 -- 426, 2005. Yoshio Fujimoto and Noboru Nakayama. Endomorphisms of smooth projective 3- folds with nonnegative Kodaira dimension, II. J. Math. Kyoto Univ., 47(1):79 -- 114, 2007. [Fujimoto02] Yoshio Fujimoto. Endomorphisms of smooth projective 3-folds with nonnegative Kodaira dimension. Publ. RIMS. Kyoto Univ., 38:33 -- 92, 2002. [Fujino11] Osamu Fujino. Semi-stable minimal model program for varieties with trivial canon- [Fulger13] [GKP16] [HX13] [KM98] ical divisor. Proc. Japan Acad. Ser. A Math. Sci., 87(3):25 -- 30, 2011. Mihai Fulger. Local volumes of Cartier divisors over normal algebraic varieties. Ann. Inst. Fourier (Grenoble) 63(5):1793 -- 1847, 2013. Daniel Greb, Stefan Kebekus, and Thomas Peternell. ´Etale fundamental groups of Kawamata log terminal spaces, flat sheaves, and quotients of Abelian varieties. To appear in Duke Math. J.. Christopher Hacon and Chenyang Xu. Existence of log canonical closures. Invent. Math., 192(1):161 -- 195, 2013. J´anos Koll´ar and Shigefumi Mori. Birational Geometry of Algebraic Varieties, vol- ume 134 of Cambridge Tracts in Mathematics. Cambridge University Press, Cam- bridge, 1998. [Koll´ar13] [KR15] J´anos Koll´ar. Singularities of the minimal model program, volume 200 of Cam- bridge Tracts in Mathematics. Cambridge University Press, Cambridge, 2013. Holly Krieger and Paul Reschke. Cohomological conditions on endomorphisms of projective varieties. Preprint, arXiv:1505.07088. [Lesieutre15] John Lesieutre. A pathology of asymptotic multiplicity in the relative setting. preprint, arXiv:1502.03019, 2015. [Nakayama04] Noboru Nakayama. Zariski-decomposition and abundance, volume 14 of MSJ Mem- oirs. Mathematical Society of Japan, Tokyo, 2004. [NZ09] [NZ10] Noboru Nakayama and De-Qi Zhang. Building blocks of ´etale endomorphisms of complex projective manifolds. Proc. Lond. Math. Soc.(3), 99(3):725 -- 756, 2009. Noboru Nakayama and De-Qi Zhang. Polarized endomorphisms of complex normal varieties. Math. Ann., 346(4):991 -- 1018, 2010. [OX12] Yuji Odaka and Chenyang Xu. Log-canonical models of singular pairs and its ap- [Wahl90] [sZhang06] [yZhang14] plications. Math. Res. Lett., 19(2):325 -- 334, 2012. Jonathan Wahl. A characteristic number for links of surface singularities. J. Amer. Math. Soc., 3(3):625 -- 637, 1990. Shou-Wu Zhang. Distributions in algebraic dynamics, volume 10 of Survey in Dif- ferential Geometry, 381 -- 430. Int. Press, 2006. Yuchen Zhang. On the volume of 150(8):1413 -- 1424, 2014. isolated singularities. Compos. Math., Department of Mathematics, University of Michigan, Ann Arbor, MI 48109, USA E-mail address: [email protected]
1105.2197
2
1105
2011-07-27T19:57:57
Functors on triangulated tensor categories
[ "math.AG" ]
We define and study the functorial spectrum for every triangulated tensor category. A reconstruction result for topologically noetherian schemes similar to (and based on) a theorem by Balmer is proved. An alternative proof of the reconstruction theorem by Bondal-Orlov for smooth projective varieties with ample (anti-)canonical bundles is given.
math.AG
math
Functors on triangulated tensor categories Y a i 1 Introduction 1.1 Introduction (1.1.1)  hi ae we de(cid:12)e he f  ia e   jST j f evey iag aed e aegy; hi ea ha T i a iag aed aegy  whi h we have a ida aegy   e  a \e d " a; b 7! a (cid:10) b whi h i exa  i bh vaiabe. The ea idea f ea ig he aegy T wih he f   ST  i eee had aeady aeaed i [8℄. R ghy eakig eee i jST j ae ihi ae f  f   f T i b D Ve   whi h eeve he i b je  ad he e d . Thi  i ay k be viewed a aag   he devee f agebai geey f a f  ia i f view a de i [6℄.  geea jST j e wih e a a   e:  i away a  ay iged a e 2.3 ad a a afai bewee f   give a \ah ageba"   e 2.4. The  i T 7! jST j i avaia i T . (1.1.2)    i i ey eaed  he ie e   Se T  de(cid:12)ed by Bae [2℄. deed hee i away a ai hi Thi hi i  away a ihi 3.1.3. jST j ! Se T : 1 Bae [2℄ hwed ha i he ae whe X i a gi ay eheia  hee ad T := X DX  he iag aed e aegy f efe  exe [12 3.1℄  X ag wih af he deived e d  we have a a a ai ihi X ! Se T : X We wi hw i 3.3.4.1 ha de he ae a i we have ai ihi X ! jST j ! Se T : X X (1.1.3)  he he had he ah ageba   e aw   e ve (cid:12)ie deed  ive f hei eeeai [8℄; ee 3.4.3.1 bew. (1.1.4) F he i f view f he e  i e  3.3.4.1 we ee ha  ihi  hee X; X wih e ivae aegie T T give dii  e d   he ae X X = 0 (cid:24) 0 iag aed aegy. Cveey aig wih a  hee X  i de  dead he  hee X havig e ivae aegy f efe  exe we eed  dead ibe 0 e d   he iag aed aegy T . X Fwig hi hihy we give a aeaive f f he e  i hee 4.2.1.1 d e  Bda v [3℄. The f i [3℄ ay be ieeed a ayig ha a h  je ive vaiey X wih ae ai  ai a b de i he d i a e f \i b je " i T = D X . Thi idea wa ae ed    (cid:13) by Bidgeed i [4℄ whee he X b eaig f \i b je " deed  he hi e f a    e. F   i f view hweve he a e f X i  ed a he f  ia e   jST j (cid:12). The e  i hee by Bda v i he eieeed a he X i ee f e d   he iag aed aegy T . X (1.1.5) Whe we wk i he deived aegy D X d a f   ideed ae deived f  . F exae if f : Y ! X i a hi bewee  hee we dee iy by   f ad f he deived f   f ad Rf  ee ivey.   2 Representable functors on triangulated tensor categories 2 2.1 The set of points (2.1.1) e TT be he aegy f eeiay a iag aed aegie whi h i a a i  ida aegy [9 Chae V e i 1.℄ aifyig he fwig aibiiy dii wih he iag aed aegy   e: The e d  i e ied  be exa  i ea h vaiabe. hi i TT ae exa  f   whi h ae a g ida f   ha i he i b je  ad e d  ae eeved   a a ihi [9 Chae V e i 1.℄. hi i TT wi be iy aed e f  . f g : T ! T i a e f   whi h i a a e ivae e f aegie he by [9 Chae 0 V e i 4℄ ad [10 ea 1.2℄ we kw ha i  ai ivee g : T ! T i exa  ad 0 0 a be hw  be a a g ida f  . F ay T 2 TT dee by ST  he e va ed vaia f   TTT ; = (cid:24) =  TT edig S 2 TT  he a a ihi ae f e f   f T  S . We wi yi ay dee eee i ST S  iy by eeeaive f   F : T ! S . (2.1.2) F ay (cid:12)ed k dee by Ve  k he abeia aegy f (cid:12)ie dieia k ve  a e ad by T he deived aegy D Ve   wih he  a e d  ad i k k b b je  k. We dee ST T  iy by ST k. k The aegy T i e ivae  he  b aegy Ve  [j ℄ by akig hgy V 7! k k   V . e ha akig hgy i a e f  ; we wi fe ideify T wih k j  j Ve  [j ℄. k (2.1.3) F ay (cid:12)ed exei k ! k 0 0 we have a e f   (cid:10) k : T ! T ad he e 0 k k k a a ST k ! ST k . e 0 jST j := i ST k: ! k e e iey jST j i he diji i f ST k d  he e ivae e eai geeaed by F (cid:24) F if F 2 ST k i aed  F 2 ST k  f e (cid:12)ed exei 0 0 0 k ! k . F exae jST j i f e i f evey (cid:12)ed k. We dee by [F ℄ he k 0 eee i jST j eeeed by a f   F : T ! T . k  he fwig e i we id e e a a   e  he e jST j. 3 (2.1.4) i e ha he  i abve ake ee f ay e va ed vaia f   X  he aegy TT givig a e jX j. Thi way we ge a e va ed vaia f    he aegy F TT; Set f f    TT. We ay a  bf   U (cid:26) X i e if f evey (cid:12)ed exei k ! k 0 id ig f : X k !  we have f 1U k 0  = U k; e ha he aie (cid:27) hd f ay  bf   f X k 0 X . Thi ey aw   \fd" he e ivae e eai: f U (cid:26) X i e he a i [ ℄ 2 jX j ie i he iage f he a a a jU j ! jX j if ad y if 2 U k f evey eeeaive f [ ℄.  ai a i hi ae jU j a be a ay idei(cid:12)ed wih a  be f jX j.  i he aighfwad  hw ha (cid:12) (cid:12) (cid:12) (cid:12) [ [ (cid:12) U (cid:12) = jU j; (cid:12) (cid:12) i i (cid:12) (cid:12) i i ad jU \ U j = jU j \ jU j; 1 2 1 2 a  be f jX j f ay e i U f e  bf   f X . i 2.2 Topology (2.2.1) F ay e i  f b je  i T  dee by U  he  bf   f ST  de(cid:12)ed by (cid:10) U  S  := fF : T ! S j 0 2 F  g; (cid:10) whee  dee he he e i f (cid:12)ie e d  f b je  i  . i e ha we have ad [ U = U ;  a a2 U \ U = U ;     1 2 1 2 whee    dee he e i f b je  a  a wih a 2  . 1 2 1 2 i i (2.2.2) Si e he f   (cid:10) b je  he  bf   U 0 0 k : T ! T f evey (cid:12)ed exei k ! k i ije ive  0 k k k i e he ee f 2.1.4.  4 A  be f jST j i aed e if i i f he f U := jU j f e  (cid:26) T ; by 2.1.4   we ee ha hi ideed de(cid:12)e a gy  jST j. i e ha if  i a (cid:12)ie e e a be he e d  f b je  i   he we have e e evey  ai a  e  be f jST j i f he f U = jU a a j f e b je  a 2 T . U = U :  a (2.2.3) A a exae f a 2 T he e e U a i f ae [F ℄ 2 jST j f f   F : T ! T wih F a = 0. De(cid:12)e he   a f a  be he ee f U  he k a he a iai a 7! a give a   daa jST j;   T i he ee f [1 De(cid:12)ii 3.1℄. 2.3 Structure sheaf (2.3.1) F ay e e U (cid:26) jST j e \ U T = keF ; [F ℄2U e ha hi i we de(cid:12)ed i e f ay (cid:12)ed exei k ! k we have ha ke (cid:10) k  k 0 0 i f b je  ihi  ze he e he f iag aed  b aegy keF  i ideede f he hi e f eeeaive i he a [F ℄ 2 U . i e ha T i a hi k U e idea ad he  aizai T =T i agai a iag aed e aegy. U e exi iy wih he ai i 2.2.3 we ee ha T i f he b je  a 2 T U  h ha U i aied i U ;  e ivaey he   a i aied i he ed a  be jST j U . De(cid:12)e whee 1 dee he iage f he i b je  i T =T . U RU  := Ed U 1; T =T i e ha he  i U 7! T evee i i he e U 7! RU  i a eheaf f U  aive ig [2 ea 9.6℄. Dee by  i hea(cid:12)(cid:12) ai he he ai jST j;   T T i a iged a e. 5 (2.3.2) We  aize e ie bevai  be ed feey bew: (2.3.2.1) Lemma e  ;  0 0 0 be  e i f bje  i T  a; a bje  i T  U; U e  be f jST j ad [F ℄ 2 jST j. The f wig aee hd: 1  (cid:26)   U (cid:26) U . 0   0 2 U (cid:26) U  T (cid:27) T . 0 U U 0 3 [F ℄ 2 U  keF  \  6= ;.  4 [F ℄ 2 U  keF  (cid:27) T . U 5 [F ℄ 2 U  a 2 keF   keF  (cid:27) T . a U a 6 a 2 T (cid:26) T . U a U 0 aa 7 U (cid:26) U . 0 aa a (2.3.3) We w  e he ak  T ;[F ℄ a a i [F ℄ 2 jST j. By de(cid:12)ii we have  = i Ed 1: U T ;[F ℄ T =T ! [F ℄2U [ Re a ha evey e e U i f he f U = U f e  (cid:26) T . e e if [F ℄ 2 U  a a2 he [F ℄ 2 U f e a whi h i e ivae  he dii ha a 2 keF .  he a wd we have  = i Ed U 1: T ;[F ℄ T =T a ! a2keF  i e ha f evey a 2 keF  we have a fa izai F T "EEEEEEEEE T =T U a :tttttttttt F (cid:22) / T = keF : T k The hhi Ed U a 1 ! Ed 1 he id e a hhi T =T T = keF  (cid:30) :  ! Ed 1: T ;[F ℄ T = keF  (2.3.3.1) Proposition The hhi (cid:30) abve i a ihi. eve Ed T = keF  1 i a  a ig; i ai a jST j;   i a  a y iged a e. T 6 / / " / : f. We wi hw e geeay ha he a a a (cid:30) : i  T =T U a x; y !  x; y T = keF  ! a2keF  i a ihi f ay x; y 2 T  viewed a b je  i he  aizai. We (cid:12) hw ha i i  je ive: Ay eee f 2  x; y i f he f T = keF   g x w ! y ; whee  ad g ae hi i T  ha F  i a ihi. Thi ea ha F  e = 0; e a := e 2 keF  he we ee ha f i a hi i T =T i e U a U a a 2 T  he e he eee [f ℄ i i  U a x; y eeeed by f 2  U x; y a T =T T =T a  f de (cid:30). ! a2keF  We w hw ha (cid:30) i ije ive: e [f ℄ be i he kee f (cid:30). By de(cid:12)ii [f ℄ i eeeed by e f 2  T =T U a x; y f e a 2 keF . e e f i f he f x w ! y wih  g ; g i T  ad  i a ihi i T =T . Si e  i a a ihi i T = keF  we U a ee ha g i aed  ze i T = keF . Cide diig ihed iage i T : g h w ! y ! eg: We ee ha h ha a ef ivee i T = keF . By he  je iviy abve we a (cid:12)d a hi  2  T =T U b  eg; y ha a  hi ef ivee i T = keF . Si e U U a ab U ab T (cid:26) T  bh f ad  ay be viewed a a hi i T =T . w he ii  AE h AE g i e a  ze i T =T  b   AE h i a ihi i U b T = keF . Thi ea ha := e AE h ie i keF  ad eve  AE h i aeady a ihi i T =T . e d = a  b   he i T =T we have  AE h AE g = 0 b   AE h U U d i a ihi he e g i ze i T =T  ha i [f ℄ = 0 a e ied f he ije iviy U d f (cid:30). Fiay we hw ha Ed 1 i a  a ig. Dee by F he hhi T = keF  1 (cid:22) Ed 1 ! Ed F 1 k; = T = keF  T k (cid:24) e a ha F 1 k a ed a degee ze i T = D Ve  .   AE e  hw ha = k k (cid:24) b evey f 2 Ed 1  yig i ke F  i iveibe b  hi i j  he de(cid:12)ii f T = keF  1 (cid:22)  aizai T = keF : Si e k i a (cid:12)ed f =2 ke F  iie ha F f  i iveibe i T  1 1 k (cid:22) (cid:22) i he wd a ihi. e e f i aeady a ihi i T = keF . 7 2.4 Path algebra (2.4.1) F ay (cid:12)ed k ad F ; G 2 ST k de(cid:12)e F ; G  be he k ve  a e aed by a a afai f F  G. e AT ; k be he die    f a F ; G wih F ; G 2 ST k he by ig a a afai ad k ieaiy AT ; k i a k ageba. (2.4.2) Wih T (cid:12)xed he a iai k 7! AT ; k i f  ia i he ee ha f evey (cid:12)ed exei k ! k we have a ageba hhi 0 : AT ; k (cid:10) k ! AT ; k : 0 k!k k 0 0 Thi way AT ;  i a ageba va ed f    he aegy f (cid:12)ed; we a hi f   he ah ageba  jST j. Re a ha we have a ST k ! jST j f evey k edig F 7! [F ℄. The ageba AT ; k ay be ideed a givig a  ive wih ST k a he e f vei e. 2.5 Functoriality (2.5.1) Ay hi g : T ! T 0 i TT id e by ii a a a afai ST  ! ST  bewee e va ed f  . Thi i   id ed a a 0 (cid:13) : jST j ! jST j: 0  he fwig we hw ha hi id ed a eeve he vai    e id ed abve. (2.5.2) e  (cid:26) T be a e i f b je  ad U  = jU  j he a iaed e  be f jST j. e  = g  he i i aighfwad  he k ha ha U = (cid:13) U  0   0 1 he e (cid:13) i i  . (2.5.3) e U (cid:26) jST j be a e  be ad U 0 1 = (cid:13) U . Re a ha a b je  a 2 T ie i T if ad y if U (cid:26) U . Thi a aie iie U (cid:26) (cid:13) U  = U  he e a a g a U 0 1 g give a f   f T i T . e e we have a id ed f   U 0U 0 (cid:22)g : T =T ! T =T : U U 0 0U 0 8 Re a he eheaf R : U 7! Ed U 1  jST j whe hea(cid:12)(cid:12) ai wa de(cid:12)ed  be T =T 0 0  ; iiay we have a eheaf R  jST j. The U 7! (cid:22)g give a eheaf hi T U whi h i   give R ! (cid:13) R ;  0 # (cid:13) :  ! (cid:13)  ; 0 T  T akig (cid:13) ; (cid:13)  a hi bewee iged a e. # (2.5.4) Give [F 0 0 0 0 ℄ 2 jST j we have (cid:13) [F ℄ = [F AE g ℄ =: [F ℄ 2 jST j. The id ed a # (cid:13) :  !  0 0 0 [F ℄ T ;[F ℄ T ;[F ℄  ak i Ed 1 ! Ed 1; 0 0 T = keF  T = keF  whi h i id ed by he ei i f g f keF  = keF AE g  keF . We kw ha 0 0 (cid:22) (cid:22) 0 # hee ae  a ig wih axia idea ee ivey ke F 1 ad ke F 1. Ceay (cid:13) 0 [F ℄ a he (cid:12) idea i he e d he e (cid:13) ; (cid:13)  i a hi f  ay iged a e. # (2.5.5) f (cid:30) : F ! G i a a a afai wih F ; G 2 ST 0 k he f ay a 2 T  a 7! (cid:30) : F ga ! Gga g a i a a a afai f F AE g  G AE g . Thi exed  a a a afai Ag : AT ;  ! AT ;  0 bewee ah ageba. 2.6 Examples (2.6.1) f g : T ! T 0 i e f   whi h i a a e ivae e f aegie he he id ed hi (cid:13) i a ihi; ee 2.1.1. (2.6.2) Cide he  aizai f   T ! T =T U a wih a 2 T . Re a ha a 2 T by U a 2.3.2.1 he e we have a a a ije i ST =T U a k ! U k a 9 f evey (cid:12)ed k: he e  he ef ide i f f   F : T ! T edig T  0 k U a whie he e  he igh ide i f F edig a  ze. B  if F a = 0 he [F ℄ 2 U  a he e T (cid:26) keF  ad F ie i he ef ide. Theefe hi a i i fa  a bije i. U a Takig ii we he bai: (2.6.2.1) Corollary F ay a 2 T  we have a ihi bewee  a y iged a e jST =T j ! U a U a (cid:24) = id ed by he  aizai f   T ! T =T  whee U U a = jU j i a  a y iged a a  ba e f jST j. (2.6.3) ee we   e iag aed e  b aegie whi h wi be ed ae i he f f 3.3.4.1. F ay e f   F : T ! T dee by F : T = keF  ! T he k k (cid:22) id ed f   ad (cid:22) (cid:24) F : Ed 1 ! Ed 1 k 1 T = keF  T k = he eva ai f F  edhi f 1. (cid:22) We dee by kF  he (cid:12)ed (cid:22) (cid:22) F Ed 1 = Ed 1= ke F : T = keF  T = ke F  1 ad by RF  he  a ig Ed 1. i e ha hey deed y  keF  ad i T = keF  ai a y  he a [F ℄. The (cid:12)ed kF  i aied i evey (cid:12)ed k  ha [F ℄ i 0 eeeed by a f   T ! T . 0 k Cide he f   0 (cid:15) F : a 7!  1; a (cid:10) kF ; T = keF  RF  hee  x; y dee he die    S (cid:15)  j2Z  x; y [j ℄ S f ay iag aed aegy S . i e ha hi i a d e ve Ed x. S The f   F ake va e i T  i e F a i a ay a gaded kF  ve  a e; i kF  0 0 i exa  b  iby  away a e f  . eve eva ai f F a hi give a a a afai (cid:30) f := F (cid:10) k  F : kF  0 (cid:30) : a =  1; a (cid:10) k !  F 1; F a F a: = a RF  T = keF  T k (cid:15) (cid:15) (cid:24) 10 We w have a diaga ha i iby   aive: 0 F T T !CCCCCCCCC F kF  (cid:10) k kF  T : k i e ha (cid:30) : a ! F a i a ihi wheeve he iage f a i T = keF  i a ihi  1. Si e bh ad F ae exa  f   he e i f b je  a 2 T  h ha (cid:30) i a ihi i ed de hifig ad akig e. a Dee by  he i y f iag aed  b aegy f T geeaed by b je  a 2 T F whi h ae ihi  1 i T = keF ; hee by \i y" we ea ha if a 2  ad a b F = (cid:24) i T he b 2  . E ivaey  i he i y f  b aegy f T iig f b je  F F whe iage i T = keF  ie i he iag aed  b aegy geeaed by 1 2 T = keF . i e ha  i a i y f iag aed e  b aegy f T deedig y  he F a [F ℄ ad we have a  aive diaga 0 F T  F "EEEEEEEE F kF  2.6.3. (cid:10) k kF  T : k e e iey he a a afai (cid:30) i a ihi bewee he f   ad F whe ei ed   . F Si e (cid:10) k i ije ive  b je  we ee ha F i a exa  e f   f   kF  F 0 T . Thi give he fwig kF  (2.6.3.1) Lemma e F : T ! T k ad G : T ! T be e f   wih keF  = keG;  i ai a  =  =:  . The [F ℄ ad [G℄ a  he ae i de he a a F G a jST j ! jS j id ed by he i i  ! T . f. deed he dii keF  = keG iie eve ha 0 0 F = G : T ! T = T : kF  kG Diaga 2.6.3. ad he fa  ha F i a e f   hw ha a i i jS j we 0 have [F ℄ = [F ℄ = [G ℄ = [G℄. 0 0 11 / / !   / / "   3 Comparison morphisms 3.1 Prime spectrum (3.1.1) Bae [2℄ de(cid:12)ed f evey iag aed e aegy T a  ay iged a e Se T  whe deyig a e i f ie e idea i T . Thi a e i (cid:12)a ag a gi a a e adiig a   daa whi h i a a f he b je  i T  he e f ed  be f he gi a a e aifyig e a a axi [2 De(cid:12)ii 3.1℄. By 2.2.3 we have a i   a Exi iy f : [F ℄ 7! keF . eve a bai f e e f Se T  ae give by f : jST j ! Se T : ad we ee ha f U a = U . The ivea   daa  Se T  i give by a 1 U a := f 2 Se T  j a 2  g; a 7! Se T  U a: (3.1.2) F ay ed  be Z f Se T  e T := fa 2 T j Z a (cid:26) Z g; Z i he wd T i f b je  a 2 T  h ha U a (cid:27) U := Se T  Z . F hi Z we have \ T =  : Z  2U F ay e  be U f Se T  we he have a a a i i 1 f U  T (cid:27) T ; Z whee Z i he ee f U . Thi id e a f   ad i   T =T ! T =T ; Z 1 f U  Ed 1 ! f RU  T =T  Z wih ai a i 2.3.1. Si e he   e heaf  i de(cid:12)ed  be he hea(cid:12)(cid:12) a Se T  i f he eheaf give by he ig  he ef ide we bai a hi f ; f  bewee # iged a e. 12 (3.1.3) We eak ha he ai hi f : jST j ! Se T  i  a bije i i geea. F a ie exae e ` be a (cid:12)ed ad T be he bi aegy T =[℄ whe ` b je  ae he ae a he f T  b  wih `   a; b =  a; b[j℄: T T k j2Z Cii ae de(cid:12)ed i he \bvi " way; ee [5 De(cid:12)ii 2.3℄.  [7 Se i 4 Thee℄ ad [11 ea 2.3℄ T i ved  be iag aed  ha he a a f   T ! T ideiy  b je  i a exa  f  .  he he had eay ` he e d   T give a bif   T  T ! T  ad i i eay  hw ha hi id ed ` e d  (cid:10)  T i exa  i ea h vaiabe:   AE e  beve ha he e i T f b je  a 2 T  ha a (cid:10) : T ! T i a exa  f   i ed de hifig ad ` T akig e ad ai he i b je . The e jST j i ey:  T evey b je  a i ihi  a[℄ ad i ai a he i b je  ` i ihi  i hif `[℄ he e hee a be ay exa  e f   f T  T a g a  6= 0. eawhie Se T  i away  ey [2 ii 2.3 k d℄. 3.2 Scheme (3.2.1) F ay  hee X dee by T X he iag aed e aegy DX  [12 3.1℄; af i i b je  i  . We de(cid:12)e a hi f  ay iged a e X a fw. (cid:11) : X ! jST j X F ay i x 2 X  dee by x : T ! T he deived bae hage f   via X kx  Se kx ! X  whee kx i he eid e (cid:12)ed f  . The (cid:11) : x 7! [x ℄ de(cid:12)e (cid:11)  he X;x  deyig e. (3.2.2) Ui he ed f hi e i we wi a e ha X i gi ay eheia; ha i i e e aify he a edig hai dii. i e ha hi i e ivae  ha a e e ae  ai a  ad i ai a iie ha X i  ai a  ad  ai eaaed. 13 (3.2.3) T ee ha (cid:11) i i   e U a (cid:26) jSe T j be a bai e  be. The we X have 1   (cid:11) U  = fx 2 X j x a = 0g; a aey he ee  X f he hgy  . By [12 ea 3.4℄ hi i e he e (cid:11) i i  . (3.2.4) w we eed  de(cid:12)e (cid:11) # :  U  !  (cid:11) U  f ay e e U i jST j; T X X 1 U X 1 0 f ii iy we dee (cid:11) U  by U . F hi i  AE e  de(cid:12)e  eheave a h hi RU  = Ed 1 !  U : U T =T X X X 0 i e ha  U  = Ed   i e T i a f  b aegy f he deived aegy 0 0 X T U U 0 U 0 0 0 DU  = D d f he  hee U . Theefe i  AE e  beve ha he a a 0 U ei i f   T ! T ed b je  i T  ze i e he we have a fa izai 0 X U U X U T ! T =T ! T 3.2.4. 0 X X U X f he ei i f   j : T ! T . 0 X U  (3.2.4.1) Lemma  he i ai abve we have T U  (cid:26) kej : T ! T . 0 X X U f. a 2 T iie ha f evey x 2 U we have ha x a i give by a a y i X U 0  ex i T . By [12 ea 3.3 a℄ hi iie ha j a = aj = 0. kx U  0 (3.2.5) e x 2 X  he x  : T ! T fa  h gh 3.2.4. f ay e e U (cid:26) X kx jST j aiig [x ℄ i ai a U ai x. e e if  2  i eeeed   0 by ~ 2 Ed 1 ee 2.3.3 ad aed  ze de x : T = kex  ! T  he T =T U X kx he iage f ~ i Ed 1 a a  ze de x : T ! T . e e (cid:11); (cid:11)  i a 0 T 0 U U kx  # hi f  ay iged a e. T ;[x ℄  (cid:22)  3.3 Comparison theorem for schemes (3.3.1) Cbiig 3.1 ad 3.2 we have hi (cid:11) f X ! jST j ! Se T  X X 14 f ay  hee X . Dee by (cid:12) he ii f AE (cid:11).  [2 Cay 5.6℄ i wa ved ha (cid:12) i a ihi f iged a e whe X i gi ay eheia. We wi ve de he ae a i ha (cid:11) i a a i hi. (3.3.1.1) Lemma f X i a gi a y eheia  hee he f evey [F ℄ 2 jST j X we have f a i e x 2 X  h ha keF  = kex .  f. The a (cid:12) ed x 7! kex  ad f ed [F ℄ 7! keF .  (3.3.2) The dii f b je  i  F a  hee. deed we have i 2.6.3 ii  he i f efe  exe  (3.3.2.1) Proposition e X be a gi a y eheia  hee; T X af = DX  . The f evey [F ℄ 2 jST j we have T =  . X X F f. e a be a ex eeeig a b je  i T  ad [F ℄ 2 jST j. By 3.3.1.1 X X hee i a i x 2 X  h ha kex  = keF . e e  ve ha a ie i  i  AE e  F   hw ha a ie i he i y f iag aed  b aegy geeaed by 1 i T = kex . X S e  (cid:26) T = kex  be a i y f iag aed  b aegy aiig 1; we eed  X  hw ha a 2  . Che a e eighb hd U f x i X  whi h a i  ai ihi  a i  efe  0 ex; i exi by he de(cid:12)ii f efe  exe. By hikig U if e eay we 0 ay a e ha a i  ai ihi  U  a b ded ex f (cid:12)ie die    f 0 1 =  . By [12 ea 3.4℄ we a (cid:12)d a b je  b i T whe    X i e a  0 U X X U ; i ai a we have (cid:11) U  = U . b 0 1 0 Wih ai a i 3.3.2.2 bew we ee ha j a ie i he iag aed  b aegy  (cid:22)  (cid:22) 0 0  j   aiig 1 =  2 T = kex . B  by 3.3.2.2 ad he i  f e f  we U U ee ha  j   j   =   he e a ie i  .  1  (cid:22) (cid:22) (3.3.2.2) Lemma e X be a gi a y eheia  hee; T X af = DX  . F ay e  be U (cid:26) jST j e U = (cid:11) U  (cid:26) X ; e x 2 U . The he a a f   X 0 1 0    (cid:22)  (cid:22) j : T = kex  ! T = kex  i f y faihf .  ai a if j a i ihi  0 X U  (cid:22)  j b he a i ihi  b i T = kex . X 15 f. We (cid:12) ve ha j i faihf . e a; b be b je  i T = kex  ad a ! b  (cid:22)   f be a hi whi h i aed  he ze hi i T = kex . Thi ea ha we U X 0  have a  aive diaga i T 0 U d  }zzzzzzzz 0 !CCCCCCCC   j a j   j  j f / j  b whee x  i a ihi. Thi iie ha x i a a ihi he e by ayig   [12 ea 3.3 a℄  x e  we ee ha hee i e e eighb hd U f x i  00 0  00 0 U  h ha k i aeady a ihi whee k : U ! U i he i i. Thi iie ha k j f = 0 = k j 0 i T . By ayig [13 ii 5.2.4 a℄  he 00     U w g i i U (cid:26) X  we have e 2 T ad hi e ! b wih x w a ihi ad X 00   h ha he fwig diaga i  aive: e AEw  ??????? w f AEw=0 / b: a  f  f Thi ay e iey ha he hi a ! b we bega wih i ze. w we hw ha j i f . e a; b be i T ad j a ! j b be a eee i X  (cid:22)    f  j a; j b; i ai a x  i a ihi. The a abve hee i a     T = kex  0 U e eighb hd U f x i U  h ha k  i aeady a ihi. 00 0  w k f  AE k  : k j a ! k j b i a hi i T . By [13 ii 5.2.3 a℄ 00 U   1     hee i a b je  d 2 T ag wih hi a d ! b  h ha k j  i a ihi X  g   i T ad he fwig diaga i  aive: 00 U   k j d k  j   zvvvvvvvvv j   k $IIIIIIIII g / k    j b: / k   k j a  1 k   k f We wi hw ha he iage f a d ! b de j i e a  j a ! j b.   g f    (cid:22) By [13 ii 5.2.3 a℄ aied  he hi k  AE k j  : k j d ! k abve  1      hee i a b je  e i T ag wih hi j d e !  ha k  i a ihi 0 U    h ad he fwig diaga ae  aive: 16 !   } o o /     o o / z $ / /  k w   k j d  k e k  j   zvvvvvvvvv {xxxxxxxxx h k     k j a k ;  k   k e   k j d  k w / #FFFFFFFFF h k  j   k $IIIIIIIII g  k  k f / k   j b: w by he faihf e f k : T = kex  ! T = kex  ved abve we ee ha hee 0 00 U U    (cid:22) i ahe b je  e i T ag wih hi e ! e e ! j d ad e !  wih he 0 U 0 0 0  0 (cid:12) hi be ig a ihi de x  ad  h ha he fwig diaga i   aive:  j d  j a  j  }{{{{{{{{ bDDDDDDDD  0 e ; j  g !CCCCCCCC =zzzzzzzz f / j  b whee he aw  he ef ad igh ae de(cid:12)ed  be he bvi  ii. Thi diaga ay ha he iage f a d ! b de j i e a  j a ! j b ad  j i  g  f     (cid:22) (cid:22) f . (3.3.3) Re ig  he ai hi (cid:11) f X ! jST j ! Se T ; X X we w have (3.3.3.1) Proposition The a (cid:11) i a hehi whe X i a gi a y eheia  hee. f. By [2 Cay 5.6℄ he ii (cid:12) = f AE (cid:11) i a hehi. e e i  AE e  hw ha he i   a (cid:11) i a bije i ad i y eai  hw ha i i  je ive. e [F ℄ 2 jST j. By 3.3.1.1 hee i a i e x 2 X  h ha keF  = kex . By X  3.3.2.1 we have T =  =   ad by 2.6.3.1 we have [F ℄ = [x ℄.  X F x  17 z o o { o o / # $ / } !   O O o o / = b (3.3.4) w we a (cid:12)ih he f f (3.3.4.1) Theorem f X i a gi a y eheia  hee ad T X = DX   he he af ai hi (cid:11) X ! jST j X i a ihi. f. By 3.3.3.1 we kw ha (cid:11) i a hehi  i y eai  hw ha # (cid:11) a de(cid:12)ed i 3.2.4 i a heaf ihi. F hi i  AE e  hw ha f evey e e U (cid:26) jST j he hhi X Ed U 1 ! Ed 1 T =T X X T 0 U id ed by he ei i f   j : T =T ! T i a ihi; e ha U = 0 X U X  U 0 1 (cid:11) U  i w idei(cid:12)ed wih U . B  by [13 ii 5.2.3 a ii 5.2.4 a℄ he f   j i f y faihf .  3.4 Comparison theorem for quivers (3.4.1) e be a  ive ibe wih eai. We e ie ha he eai aify he ey ha he aegy Re f (cid:12)ie dieia eeeai ve a (cid:12)ed ` i ` ida de he  a veex wie e d . e ha hi e d  i exa  ad T := D  Re  i he a ` iea iag aed e aegy. ;` ` b (3.4.2) e b he e f vei e f . The f ay v 2 we have a e f   0 0 F : T ! T edig v ;` ` V 7! V  ; v  whee he hgy V  f a b je  V 2 T i a ex wih ze di(cid:11)eeia  ;`  wih b je  i Re . The ex f ` ve  a e f V  a he veex v i he ` v a ay a b je  i T . ` Thi de(cid:12)e a a f  ST k f evey (cid:12)ed exei ` ! k ad he e a a 0 ;` f  jST j. 0 ;` The fwig i ved i [8℄: 18 (3.4.2.1) Proposition f i (cid:12)ie ad wih  ieed y e he f evey (cid:12)ed we have hehi ! jST j ! Se T ; 0 ;` ;` whee i give he di ee gy. Thee a ae give by v 7! [F ℄ ad [F ℄ 7! 0 v keF . (3.4.3) A iaed  he  ive we have a ageba va ed f    he aegy f (cid:12)ed edig k 7! A ; k := k ; he ah ageba f . Fixig a (cid:12)ed ` ad ei ig hi  (cid:12)ed exei ` ! k we wi de(cid:12)e a a a afai bewee f   whee he e d f   wa de(cid:12)ed i 2.4.2. (cid:30) : A ;  ! AT ; ; ;` Exi iy if  i a ah f veex v  veex w givig a eee  2 A ; k he f ay (cid:12)ed exei ` ! k (cid:30)  i he eee i F (cid:10) k; F (cid:10) k (cid:26) AT ; k edig k v ` w ` ;` V 7! (cid:30)  := V : F V  ! F V : k V  v w The fwig i ved i [8℄: (3.4.3.1) Theorem f i (cid:12)ie ad wih  ieed y e he f evey (cid:12)ed ` he a a afai (cid:30) abve i a ihi f A ;   AT ; . ;` 4 Rigidifying tensor products 4.1 Identifying tensor products ee we ve a few ie eiiay ea ha wi he  ae e d   iag aed aegie. 19 (4.1.1) We ide (cid:12) he i ai whe hee i a exa  e ivae e g : T ! T 0 f iag aed aegie whe T ha a give e d  (cid:10) . Dee by g a  ai ivee 0 0 0  g  he we a de(cid:12)e a e d  (cid:10)  T by eig 1 a (cid:10) b := g ga (cid:10) gb: 1 0 0 (4.1.1.1) Lemma The f   g : T ; (cid:10) 1  ! T ; (cid:10)  i a e ivae e f iag aed e 0 0 aegie.  ai a we have a ihi jST ; (cid:10) j jST ; (cid:10) j f  a y = 1 0 0 (cid:24) iged a e. f.  y eai  hw ha g i a g ida f  : ga (cid:10) b = gg ga (cid:10) gb ga (cid:10) gb: = 1 0 0 0 (cid:24) (4.1.2) e be a b je  i a iag aed e aegy T ; (cid:10) 1  be  h ha he exa  0 f    : T ! T de(cid:12)ed by a 7! (cid:10) a i a e ivae e wih  ai ivee  . The we 1 a de(cid:12)e a e d  (cid:10)  T by eig 0 a (cid:10) b :=  a (cid:10)  b: 0 1 0 0 (4.1.2.1) Lemma The f    : T ; (cid:10)  ! T ; (cid:10)  i a e ivae e f iag aed 1 0 e aegie.  ai a we have a ihi jST ; (cid:10) j jST ; (cid:10) j f 0 1 = (cid:24)  a y iged a e. e ha i e he i b je  1 wih ee   (cid:10) i ihi  g   he b je  i 1 0 he i b je  wih ee   (cid:10) . 0 f. We  e: a (cid:10) b  a (cid:10)  b = a (cid:10) b: = 1 1 0 (cid:24) 0 0 (4.1.3) e T be a iag aed aegy ad W (cid:26) T a f  b aegy  e eaiy iag aed. We ay ha W geeae T if f evey a 2 T hee ae (cid:12)iey ay b je  w ; w ; : : : ; w i W ad diig ihed iage i T : 1 2  a ! w ! a ! a [1℄; 1 1 1 20 a ! w ! a ! a [1℄; 2 2 1 2 : : : ; w ! w ! a ! w [1℄;  1 1  f e b je  a ; : : : ; a . 1 1 (4.1.3.1) Lemma f T ; (cid:10) i a iag aed e aegy he (cid:10) i deeied by i ei i  W  T ! T f ay f  b aegy W geeaig T . f. e e iey if (cid:10) ad (cid:10) ae e d   T whe ei i  W  T ae 0 ihi  he he ihi exed  a ihi bewee (cid:10) ad (cid:10)  T  T . 0 deed  e (cid:30) : w (cid:10) b ! w (cid:10) b w 2 W  a 2 T  i a a a ihi bewee w ;b 0 (cid:24) = he ei i we eed  exed  a ihi (cid:30) : a (cid:10) b ! a (cid:10) b a; b 2 T . a;b 0 (cid:24) = T hi ed we y eed  (cid:12)d a \e i" f a by b je  i W a abve ad ay (cid:30) : (cid:10) b ! (cid:10) b  evey diig ihed iage. ;b 0 4.2 Example: smooth varieties with ample (anti-)canonical bundles (4.2.1) We wi give a f f he fwig (4.2.1.1) Theorem [3 Thee 2.5℄ e X be a ied ibe h je ive vaiey ve a (cid:12)ed k wih ae ai  ai a b d e. f X i ahe ied ibe h 0 je ive vaiey ve k  h ha D X  ad D X  ae e ivae a iag aed b b 0 aegie he X i ihi  X . 0   f i  eey ideede f ha i [3℄; ee 4.2.2.1 bew. Ude he he a i D X  i e ivae  T = DX   heefe he X af b kwedge f he whe e d  T  T ! T aeady aw   e   X X X X 3.3.4.1. The ai i f he f i ha he aee a i aw   e ve he whe e d  f he ei ed e d  W  T ! T f eai  b aegy W X X geeaig T  whi h i   i deeied by he i e See f  . X 21 (4.2.2) Dee by g : T X ! T a exa  e ivae e wih  ai ivee g ; dee by (cid:10) ad 0 X 0 0 (cid:10) hei ee ive e d . (4.2.2.1) Lemma Wih a i a i 4.2.1.1 g X  i ihi  a ie b d e  X . 0 f. F ay ed i x 2 X  by [3 ii 2.2℄ we have ha g    [i℄ f x x = 0 0 0 (cid:24) 0 e iege i 2 Z ad ed i x 2 X  deedig  x . e ha he eid e (cid:12)ed kx 0 a be e veed a Ed   i ai a we have kx kx ; dee hi (cid:12)ed by k . = T x X (cid:24) 0 0 e e we have  x g ; k [j ℄  g ;  [j ℄   ;  [i  j ℄; = = 0 T 0 k X T X x T X x 0 X X 0 0 (cid:24) (cid:24) whi h i ihi  k whe i  j = 0 ad i e a  ze whe i  j 6= 0. 0  ai a x g  i  ze f evey ed i x 2 X  he e he   f g  X X 0 0 0 i he whe f X . eve eeeig g  by a efe  ex  X we ee ha hi X 0 0 ex ha y e  ze hgy heaf i ai a he iege i i ideede 0 0 f x  whi h   be a ie b de  X i e a f i (cid:12)be have ak e. (4.2.3) w we a give a f f Thee 4.2.1.1. Dee by g : T ! T a exa  e ivae e wih  ai 0 X X ivee g ; dee by (cid:10) ad (cid:10) he ee ive deived e d   hee w iag 0 0 aed aegie. By 4.1.1.1 we have a e d  (cid:10)  T de(cid:12)ed by 1 X a (cid:10) b = g ga (cid:10) gb 1 0 0  ha hee ae ihi f  ay iged a e (cid:24) (cid:24) 0 0 = = X ! jST ; (cid:10) j ! jST ; (cid:10) j: 0 X X 1 Dee by ! ad ! he ai a ie b de  X ad X ee ivey ad by S; S he 0 0 0 hif f he See f    ha S a = a (cid:10) ! ad S a  = a (cid:10) ! ; we wi e he fa  0 0 0 0 0 ha See f    e wih exa  e ivae e [3 ii 1.3℄. e ! := g !  1 0 0 he we have a a ihi a (cid:10) ! g ga (cid:10) !  = g S ga g gS a a (cid:10) ! ; = = = 1 1 (cid:24) (cid:24) (cid:24) 0 0 0 0 0 0 f evey a 2 T .  ai a by akig a =  we have  (cid:10) ! ! . = X X X 1 1 (cid:24) 22 w by 4.2.2.1 we kw ha g  i a ie b de  X  he e i ai a he f   X 0 0 0 0 a 7! g  (cid:10) a i a a e ivae e  T .  he fw ha  : a 7!  (cid:10) a i a 0 X X X 1 a e ivae e  T ; dee by  i  ai ivee. e e by 4.1.2.1 hee i ahe X 0 e d  (cid:10)  T de(cid:12)ed by 0 X a (cid:10) b =  a (cid:10)  b 0 1 0 0  ha we w have ihi (cid:24) (cid:24) 0 = = X ! jST ; (cid:10) j jST ; (cid:10) j: X 1 X 0 w we  e: a (cid:10) !  a (cid:10)   (cid:10) !   a (cid:10) !  S a  S a S a: = = = = = 0 1 X 1 1 1 1 (cid:24) (cid:24) (cid:24) (cid:24) (cid:24) 0 0 0 0 0 e e we have a a ihi a 7! S a a 7! a (cid:10) ! a 7! a (cid:10) !: = = 0 (cid:24) (cid:24) e W (cid:26) T be he f  b aegy iig f b je  f he f ! [ ℄ wih X j  (cid:10)` j ` ;  2 Z ad y (cid:12)iey ay  ad. The i e eihe !  ! i ae W j j j 1 geeae T [13 ii 2.3.1 d℄. B  he ei i f (cid:10) ad (cid:10)  W  T ! T X 0 X X bh give die    f ieai f hif f he See f   ad  by 4.1.3.1 we kw ha (cid:10) ad (cid:10) ae ihi  ad we  de: 0 (cid:24) = 0 (cid:24) (cid:24) = X ! jST ; (cid:10) j jST ; (cid:10)j X: = X 0 X References [1℄ a Bae. eheave f iag aed aegie ad e  i f  hee. ah. A. 3243:557{580 2002. [2℄ a Bae. The e   f ie idea i e iag aed aegie. . Reie Agew. ah. 588:149{168 2005. [3℄ Aexei Bda ad Dii v. Re  i f a vaiey f he deived aegy ad g  f a e ivae e. Cii ah. 1253:327{344 2001. [4℄ T Bidgead. F ad deived aegie. ve. ah. 1473:613{632 2002. 23 [5℄ Ca de Cibi ad Ed ad . a . Skew aegy Gai veig ad ah d  f a k aegy.  . Ae. ah. S . 1341:39{50 ee i  2006. [6℄ i he Deaz e ad ee Gabie. d i  agebai geey ad agebai g  v e 39 f h  ad aheai  S die. h ad  bihig C. Aeda 1980. [7℄ Behad ee.  iag aed bi aegie. D . ah. 10:551{581 2005. [8℄ Y a i ad S a Siea. ie e   f  ive eeeai. ei 2011. [9℄ Sa de a ae. Caegie f he wkig aheai ia v e 5 f Gad ae Tex i aheai . Sige Veag ew Yk e d edii 1998. [10℄ D. . v. E ivae e f deived aegie ad  3  fa e. . ah. S i. ew Yk 845:1361{1381 1997. Agebai geey 7. [11℄ iagag eg ad ie Xia. R aegie ad ie ie ageba. . Ageba 1981:19{56 1997. [12℄ R. W. Tha. The ai(cid:12) ai f iag aed  b aegie. Cii ah. 1051:1{27 1997. [13℄ R. W. Tha ad Tha Tba gh. ighe agebai  hey f  hee ad f deived aegie.  The Ghedie k Fe hif V.  v e 88 f g. ah. age 247{435. Bikh(cid:127)a e B B A 1990. 24
1811.11951
1
1811
2018-11-29T04:04:35
Rank two sheaves with maximal third Chern character in three-dimensional projective space
[ "math.AG" ]
We give a complete classification of semistable rank two sheaves on three-dimensional projective space with maximal third Chern character. This implies an explicit description of their moduli spaces. As an open subset they contain rank two reflexive sheaves with maximal number of singularities. These spaces are irreducible, and apart from a single special case, they are also smooth. This extends a result by Okonek and Spindler to all missing cases and gives a new proof of their result. The key technical ingredient is variation of stability in the derived category.
math.AG
math
RANK TWO SHEAVES WITH MAXIMAL THIRD CHERN CHARACTER IN THREE-DIMENSIONAL PROJECTIVE SPACE BENJAMIN SCHMIDT Abstract. We give a complete classification of semistable rank two sheaves on three-dimensional projective space with maximal third Chern character. This implies an explicit description of their moduli spaces. As an open subset they contain rank two reflexive sheaves with maximal number of singularities. These spaces are irreducible, and apart from a single special case, they are also smooth. This extends a result by Okonek and Spindler to all missing cases and gives a new proof of their result. The key technical ingredient is variation of stability in the derived category. Contents 1. Introduction 2. Preliminaries 3. Classifying rank two reflexive sheaves with maximal third Chern character 4. Geometric structure of the moduli spaces References 1 3 8 15 22 8 1 0 2 v o N 9 2 ] . G A h t a m [ 1 v 1 5 9 1 1 . 1 1 8 1 : v i X r a 1. Introduction Moduli spaces of sheaves are well known to be badly behaved. In [Mum62] Mumford described a generically non-reduced irreducible component of the Hilbert scheme of curves in P3 whose general point parametrizes a smooth curve. The fact that well-behaved geometric objects could have such a disastrous moduli space was a shocking result. In [Vak06] this was vastly generalized. Vakil showed that many classes of moduli spaces satisfy Murphy's law in algebraic geometry. This means every possible singularity can occur on them. The moral of these results is that moduli space are problematic, unless there is a good reason to believe otherwise. In this article, we deal with moduli spaces of rank two sheaves in P3 whose third Chern character is maximal. They defy the general principle. Except for one case they turn out to be smooth and irreducible. We denote the moduli space of Gieseker-semistable sheaves E ∈ Coh(P3) with Chern character ch(E) = v as M (v). Theorem 1.1. Let E ∈ Coh(P3) be a Gieseker-semistable rank two object with ch(E) = (2, c, d, e). (i) If c = −1, then d ≤ − 1 2 . (a) If d = − 1 (b) If d ≤ − 3 2, then e ≤ 5 2 , then e ≤ d2 6 . Moreover, M (2, −1, − 1 2 , 5 6 ) ∼= P3. M (2, −1, d, d2 2 − d + 5 24 ) → P3, where the fiber is the Grassmannian Gr(2, n) for 2 − d + 5 24 . Moreover, there is a locally trivial fibration n =(cid:18) 5 2 − d 2 (cid:19). 2010 Mathematics Subject Classification. 14J60 (Primary); 14D20, 14F05 (Secondary). Key words and phrases. Stable sheaves, Stability conditions, Derived categories. 1 (ii) If c = 0, then d ≤ 0. (a) If d = 0, then e ≤ 0. In case of equality, E ∼= O⊕2. (b) If d = −1, then e ≤ 0. Moreover, M (2, 0, −1, 0) ∼= P5. (c) If d = −2, then e ≤ 2. (d) If d = −3, then e ≤ 4. The moduli space M (2, 0, −3, 4) is the blow up of Gr(3, 10) in a smooth subvariety isomorphic to P3 × P3. (e) If d ≤ −4, then e ≤ d2 2 + d a Pn-bundle over P3 × P3, where n = d(d − 2) − 1. 2 + 1. Moreover, the moduli space M (2, 0, d, d2 2 + d 2 + 1) is It turns out that in the case ch(E) = (2, 0, −2, 2) there are strictly semistable sheaves preventing the moduli space from being smooth. This case had already been deeply analyzed in [MT94]. A special case of a theorem by Hartshorne in [Har88] proves these bounds with the extra assumption that E is reflexive. In [OS85] Okonek and Spindler proved the same bounds for all semistable sheaves of rank two. Moreover, they described the moduli space if either c = 0 and d ≤ −6 or c = −1 and d ≤ − 11 2 . The above theorem fills in all the remaining special cases. Our proof is completely independent of these previous results. Assume that E is reflexive. Then by [Har80, Proposition 2.6] having maximal third Chern character can be interpreted as E having the maximal possible number of singularities, i.e., points where it fails to be a vector bundle. Curiously, making E less close to a vector bundle leads to a nice moduli space. Other descriptions of components of moduli spaces of semistable rank two sheaves in P3 have been found in examples such as [AJT18, AJTT17, Cha83, JMT17a, JMT17b, Man81]. So called instanton bundles satisfy additional cohomology vanishings. A long standing conjecture that these bundles represent smooth points in their moduli spaces was settled in [JV14]. However, all these cases are different, since their closures in the moduli space of all semistable sheaves are well known to contain many singularities. The situation is similar to the Hilbert scheme of plane curves of degree d in P3. Interpreted as the moduli space of its ideal sheaves, they also maximize the third Chern character. The rank one and two examples lead us to make the following slightly adventurous conjecture. Conjecture 1.2. Let r ∈ Z≥0, c ∈ Z, d ∈ 1 Z such that M (r, c, d, e) is non-empty, but 2 M (r, c, d, e′) = ∅ for all e′ > e. If all Gieseker-semistable sheaves with Chern character (r, c, d, e) are Gieseker-stable, then M (r, c, d, e) is smooth and irreducible. Z, e ∈ 1 6 1.1. Ingredients. Our proof is fundamentally different than what was done before in [Har88] and [OS85]. We use the notion of tilt stability (see Section 2) in the derived category due to [Bri08, AB13, BMT14]. It generalizes the notion of slope stability by varying the abelian category from coherent sheaves to certain categories of two-term complexes Cohβ(P3) ⊂ Db(P3) dependent on a real parameter β. A new slope function να,β depends on another positive real number α. Varying α, β varies the set of semistable objects. The key is that for α ≫ 0 and β ≪ 0 all Gieseker- semistable sheaves are να,β-semistable. In the upper half-plane parametrized by α > 0 and β ∈ R, there is a locally-finite wall and chamber structure such that the set of semistable objects is constant within each chamber. In [Mac14b] Macr`ı proves an inequality for the Chern characters of tilt-semistable objects (see Theorem 2.7). This inequality can be used to show that if ch3(E) is larger than or equal to the claimed bound, then E has to be destabilized somewhere in tilt stability. If ch3(E) is strictly larger, then we get a contradiction by showing that there is no wall by mostly numerical arguments. In case of equality, we show that there is a unique wall unless ch1(E) = 0 and ch2(E) = 3. This unique wall leads to a classification of all semistable sheaves, and the description of the moduli spaces follows. The special case uses techniques close to what was done for some Hilbert schemes of curves in [Sch15] and [GHS18]. 2 Acknowledgments. I would like to thank Tom Bridgeland, Izzet Coskun, Sean Keel, Marcos Jardim, Emanuele Macr`ı, Ciaran Meachan, and Benjamin Sung for very useful discussions. This work has been supported by an AMS-Simons Travel Grant. Notation. Db(P3) bounded derived category of coherent sheaves on P3 over C Hi(E) H i(E) ch(E) Chern character of an object E ∈ Db(P3) the i-th cohomology group of a complex E ∈ Db(P3) the i-th sheaf cohomology group of a complex E ∈ Db(P3) ch≤l(E) (ch0(E), . . . , chl(E)) 2. Preliminaries In this section, we will recall several notions of stability in the bounded derived category of coherent sheaves and its basic properties. By abuse of notation, we write chi(E) for E ∈ Db(P3) but mean its intersection with 3 − i powers of the hyperplane class. 2.1. Stability for sheaves. Definition 2.1. (i) The classical slope for a coherent sheaf E ∈ Coh(P3) is defined as µ(E) := ch1(E) ch0(E) , where division by zero is interpreted as +∞. (ii) A coherent sheaf E is called slope-(semi)stable if for any non-trivial proper subsheaf F ֒→ E the inequality µ(F ) < (≤)µ(E/F ) holds. In many cases, this notion is not quite refined enough. Definition 2.2. Let f, g ∈ R[m] be two polynomials. (i) If deg(f ) < deg(g), then f > g. Vice versa, deg(g) < deg(f ) implies g > f . (ii) Assume d = deg(f ) = deg(g), and let a, b be the coefficients of md in f , g. Then we define f < (≤)g if f (m) b Definition 2.3. a < (≤) g(m) (i) For any E ∈ Coh(P3) we can define numbers αi(E) for i ∈ {0, 1, 2, 3} via for all m ≫ 0. the Hilbert polynomial P (E, m) := χ(E(m)) = α3(E)m3 + α2(E)m2 + α1(E)m + α0(E). Moreover, we set P2(E, m) := α3(E)m2 + α2(E)m + α1(E). (ii) A sheaf E ∈ Coh(P3) is called Gieseker-(semi)stable if for any non-trivial proper subsheaf F ֒→ E the inequality P (F, m) < (≤)P (E/F document, m) holds. (iii) A sheaf E ∈ Coh(P3) is called 2-Gieseker-(semi)stable if for any non-trivial proper subsheaf F ֒→ E the inequality P2(F, m) < (≤)P2(E/F, m) holds. Gieseker stability was introduced by Gieseker for torsion-free sheaves and later generalized to torsion sheaves by Simpson. The notion of 2-Gieseker stability is less known, but it is in fact the precise notion that we need to connect it to tilt stability as described in the next subsection. Finally, there are the following relations between these notions. slope-stable 3 2-Gieseker-stable 3 Gieseker-stable slope-semistable 2-Gieseker-semistable Gieseker-semistable 3 + +  k s k s 2.2. Tilt Stability. By using the derived category it is possible to obtain a more flexible form of stability. The key idea due to Bridgeland is to change the category of coherent sheaves for another heart of a bounded t-structure inside the bounded derived category. The following notion of tilt stability was introduced by Bridgeland for K3 surfaces [Bri08], later generalized to all surfaces by Arcara-Bertram [AB13], and finally extended to higher dimensions by Bayer-Macr`ı-Toda in [BMT14]. Let β be an arbitrary real number. Then the twisted Chern character chβ is defined to be e−βH · ch, where H is the hyperplane class. Note that for β ∈ Z one has chβ(E) = ch(E(−β)) for any E ∈ Db(P3). Explicitly: chβ 0 = ch0, chβ 1 = ch1 −β ch0, chβ 2 = ch2 −β ch1 + β2 2 ch0, chβ 3 = ch3 −β ch2 + β2 2 ch1 − β3 6 ch0 . The process of tilting is used to construct a new heart of a bounded t-structure. For more information on the general theory of tilting we refer to [BvdB03, HRS96]. Definition 2.4. (i) A torsion pair is defined by Tβ := {E ∈ Coh(P3) : any quotient E ։ G satisfies µ(G) > β}, Fβ := {E ∈ Coh(P3) : any non-trivial subsheaf F ⊂ E satisfies µ(F ) ≤ β}. We define Cohβ(P3) as the extension closure hFβ[1], Tβi. (ii) Let α > 0 be a positive real number. The tilt-slope is defined as να,β := 2 chβ 0 chβ 2 − α2 chβ 1 . (iii) Similarly as before, an object E ∈ Cohβ(P3) is called tilt-(semi)stable (or να,β-(semi)stable) if for any non-trivial proper subobject F ֒→ E the inequality να,β(F ) < (≤)να,β(E/F ) holds. Note that Cohβ(P3) consists of some two term complex. More precisely, it contains exactly those complexes E ∈ Db(P3) such that H0(E) ∈ Tβ, H−1(E) ∈ Fβ, and Hi(E) = 0 whenever i 6= −1, 0. The following proposition was proved for K3 surfaces in [Bri08, Proposition 14.2], but the proof holds without trouble in our case. Proposition 2.5. If E ∈ Cohβ(P3) for β < µ(E) is να,β-(semi)stable for α ≫ 0, then it is a 2-Gieseker-(semi)stable sheaf. Vice versa, if E is a 2-Gieseker-(semi)stable sheaf, then it is να,β- semistable for any α ≫ 0 and β < µ(E). Note that in case µ(E) = ∞, this proposition holds for arbitrary β. The Chern characters of semistable objects satisfy certain inequalities. The first one is well known as the Bogomolov inequality. It was first proved for slope-semistable sheaves on surfaces ([Rei78, Bog78, Gie79]). Theorem 2.6 (Bogomolov inequality for tilt stability, semistable object E ∈ Cohβ(P3) satisfies [BMT14, Corollary 7.3.2]). Any να,β- ∆(E) := ch1(E)2 − 2 ch0(E) ch2(E) ≥ 0. The following inequality involving the third Chern character is part of a more general conjecture in [BMT14] and was brought into the following form in [BMS16]. The case of P3 was proved in [Mac14b] 4 Theorem 2.7. For any να,β-semistable object E ∈ Cohβ(P3) the inequality Qα,β(E) := α2∆(E) + 4 chβ 3 (E) ≥ 0 2 (E)2 − 6 chβ 1 (E) chβ holds. We will need to understand interactions between the derived dual and tilt stability. Proposition 2.8 ([BMT14, Proposition 5.1.3]). Assume E ∈ Cohβ(P3) is να,β-semistable with να,β(E) 6= ∞. Then there is a να,−β-semistable object E ∈ Coh−β(P3) and a sheaf T supported in dimension zero together with a distinguished triangle E → RHom(E, O)[1] → T [−1] → E[1]. 2.3. Walls. Let Λ = Z ⊕ Z ⊕ 1 Z. Then H · ch≤2 maps to Λ. Varying (α, β) changes the set of 2 semistable objects. A numerical wall in tilt stability with respect to a class v ∈ Λ is a non-trivial proper subset W of the upper half plane given by an equation of the form να,β(v) = να,β(w) for another class w ∈ Λ. We will usually write W = W (v, w). A subset S of a numerical wall W is called an actual wall if the set of semistable objects with class v changes at S. The structure of walls in tilt stability is rather simple. Part (i) - (v) is usually called Bertram's Nested Wall Theorem and appeared in [Mac14a], while part (vi) and (vii) can be found in [BMS16, Appendix A]. Theorem 2.9 (Structure Theorem for Walls in Tilt Stability). Let v ∈ Λ be a fixed class. All numerical walls in the following statements are with respect to v. (i) Numerical walls in tilt stability are either semicircles with center on the β-axis or rays If v0 6= 0, there is exactly one numerical vertical wall given by parallel to the α-axis. β = v1/v0. If v0 = 0, there is no actual vertical wall. (ii) The curve να,β(v) = 0 is given by a hyperbola, which may be degenerate if v0 = 0. Moreover, this hyperbola intersects all semicircular walls at their top point. (iii) If two numerical walls given by classes w, u ∈ Λ intersect, then v, w and u are linearly dependent. In particular, the two walls are completely identical. (iv) If a numerical wall has a single point at which it is an actual wall, then all of it is an actual wall. (v) If v0 6= 0, then there is a largest semicircular wall on both sides of the unique numerical vertical wall. If v0 = 0, then there is a unique largest semicircular wall. (vi) If there is an actual wall numerically defined by an exact sequence of tilt-semistable objects 0 → F → E → G → 0 such that ch≤2(E) = v, then ∆(F ) + ∆(G) ≤ ∆(E). Moreover, equality holds if and only if ch≤2(G) = 0. (vii) If ∆(E) = 0, then E can only be destabilized at the unique numerical vertical wall. In particular, line bundles, respectively their shifts by one, are tilt-semistable everywhere. If W = W (v, w) is a semicircular wall in tilt stability for two numerical classes v, w ∈ Λ, then we denote its radius by ρW = ρ(v, w) and its center on the β-axis by sW = s(v, w). The structure of the locus Qα,β(E) = 0 fits right into into the semicircle wall picture. Indeed, a straightforward computation shows the following. Lemma 2.10. Let E ∈ Db(P3). The equation Qα,β(E) = 0 is equivalent to να,β(E) = να,β(ch1(E), 2 ch2(E), 3 ch3(E)). In particular, Qα,β(E) = 0 describes a numerical wall in tilt stability. 5 We denote the numerical wall Qα,β(E) = 0 by WQ = WQ(E), its radius by ρQ = ρQ(E), and its center by sQ = sQ(E). The following lemma is a highly convenient tool to control the rank of destabilizing subobjects. A very close version appeared first in [CH16, Proposition 8.3]. Lemma 2.11 ([MS18, Lemma 2.4]). Assume that a tilt-semistable object E is destabilized by either a subobject F ֒→ E or a quotient E ։ F in Cohβ(P3) inducing a non-empty semicircular wall W . Assume further that ch0(F ) > ch0(E) ≥ 0. Then the inequality ρ2 W ≤ ∆(E) 4 ch0(F )(ch0(F ) − ch0(E)) holds. If we understand the radius ρQ(E), this lemma will provide a key tool to control the rank of destabilizing subobjects. 2.4. Bridgeland stability. Tilt stability has well-behaved computational properties. However, for dimension greater than or equal to three it does not have well-behaved moduli spaces. This is similar to the issues of slope stability on surfaces, where Gieseker stability turns out to be the better notion. We need to recall the constructions of Bridgeland stability on P3 due to [BMT14, Mac14b]. The idea is to perform another tilt as previously. Let α,β := {E ∈ Cohβ(P3) : any quotient E ։ G satisfies να,β(G) > 0}, T ′ α,β := {E ∈ Cohβ(P3) : any non-trivial subobject F ֒→ E satisfies να,β(F ) ≤ 0} F ′ and set Aα,β(P3) := hF ′ α,β[1], T ′ α,βi. For any s > 0 they define Zα,β,s := − chβ 3 +(s + 1 6 )α2 chβ 1 +i(chβ 2 − α2 2 chβ 0 ), λα,β,s := − ℜ(Zα,β,s) ℑ(Zα,β,s) . Definition 2.12. An object E ∈ Aα,β(P3) is called λα,β,s-(semi)stable if for any non-trivial sub- object F ֒→ E, we have λα,β,s(F ) < (≤)λα,β,s(E). 2.5. Moduli Spaces. For any v ∈ K0(P3), α > 0, β ∈ R, s > 0 we make the following definitions. (i) The moduli space of slope-semistable sheaves with Chern character v is Definition 2.13. denoted by M (v). (ii) The moduli space of να,β-semistable objects with Chern character v is denoted M tilt α,β(v). (iii) The moduli space of λα,β,s-semistable objects with Chern character v is denoted M B α,β,s(v). For some Bridgeland stability conditions it is possible to exchange the heart Aα,β(P3) by a finite length category. Theorem 2.14 ([Mac14b]). Let α < 1/3, β ∈ (−5/3, −1], and 0 < s ≪ 1. For any γ ∈ R, we can define a torsion pair γ := {E ∈ Aα,β(P3) : any quotient E ։ G satisfies λα,β,s(G) > γ}, T ′′ γ := {E ∈ Aα,β(P3) : any non-trivial subobject F ֒→ E satisfies λα,β,s(F ) ≤ γ}. F ′′ There is a choice of γ ∈ R such that hT ′′ γ , F ′′ γ [1]i = hO(−2)[3], T (−3)[2], O(−1)[1], Oi. In particular, moduli spaces of Bridgeland-stable objects for these special choices of stability condi- tions are the same as moduli spaces of representation of finite-dimensional algebras as defined in [Kin94]. This means they have projective moduli spaces. 6 Through most of this article, we will only study tilt stability. The following statement makes a connection to Bridgeland stability. Theorem 2.15 ([Sch15, Theorem 6.1(3)]). Let v ∈ K0(P3), α0 > 0, β0 ∈ R, and s > 0 such that να0,β0(v) = 0, H 2 · vβ0 1 > 0, and ∆(v) ≥ 0. Assume that all να0,β0-semistable objects of class v are να0,β0-stable. Then there is a neighborhood U of (α0, β0) such that M B α,β,s(v) = M tilt α,β(v) for all (α, β) ∈ U with να,β(v) > 0. Moreover, in this case all objects parametrized in M B λα,β,s-stable. α,β,s(v) are The following result by Piyaratne and Toda is a major step towards the construction of well- It applies in particular to the case of P3, since the conjectural BMT- behaved moduli spaces. inequality is known. Theorem 2.16 ([PT15]). Let X be a smooth projective threefold such that the conjectural con- struction of Bridgeland stability from [BMT14] works. Then any moduli space of semistable objects for such a Bridgeland stability condition is a universally closed algebraic stack of finite type over C. If there are no strictly semistable objects, the moduli space becomes a proper algebraic space of finite type over C. 2.6. Some known bounds. We recall further known results about Chern character bounds for tilt stability in P3. Lemma 2.17 ([Sch15, Lemma 5.4]). (i) Let E ∈ Cohβ(P3) be tilt-semistable with ch(E) = (1, 0, −1, 1). Then E ∼= IL for a line L ⊂ P3. (ii) Let E ∈ Cohβ(P3) be tilt-semistable with ch(E) = (0, 1, d, e), then E ∼= IZ/V (d+1/2) where Z is a dimension zero subscheme of length 1 24 + d2 2 − e. Proposition 2.18. Let E ∈ Cohβ(P3) be a να,β-semistable object for some (α, β) with either ch(E) = (1, 0, −d, e) or ch(E) = (−1, 0, d, e). Then In case of equality and ch0(E) = 1, the object E is the ideal sheaf of a plane curve. e ≤ d(d + 1) 2 . Proof. The bounds where shown in [MS18, Proposition 3.2]. If ch0(E) = 1, then the proof also shows that in case of equality, E is destabilized by a morphism O(−1) ֒→ E unless ch0(E) = 1 and d = 1. The special case is solved directly by Lemma 2.17. Let G = E/O(−1) be the quotient. Then ch(G) =(cid:18)0, 1, −d − 1 2 , d(d + 1) 2 + 1 6(cid:19) . By Lemma 2.17, we get G ∼= OV (−d), and the claim follows. (cid:3) Line bundles can be easily identified by their Chern characters as follows. This was shown in [Sch15, Proposition 4.1, 4.5]. Proposition 2.19. Let E be a tilt-semistable or Bridgeland-semistable object. Assume that there are integers n, m with m > 0 such that either (i) v = m ch(O(n)), or (ii) v = −m ch≤2(O(n)). Then E ∼= O(n)⊕m or a shift of it. Moreover, in the case m = 1 the line bundle O(n) is stable. 7 The following statement from [MS18, Theorem 3.4] had a further error term which we will not need. For the convenience of the reader, we will give a proof, since it is shorter and substantially easier without the error term. Theorem 2.20. Let E ∈ Cohβ(P3) be a να,β-semistable object with ch(E) = (0, c, d, e), where c > 0. Then e ≤ + c3 24 d2 2c . Proof. Note that if a subobject of rank zero destabilizes E, then it does so independently of (α, β). Therefore, any wall must be induced by a subobject or quotient of positive rank. By Lemma 2.11 this means that any wall W must have radius satisfying ρW ≤ c 2 . Therefore, E has to be tilt-semistable for some (α, β) inside or on the semicircular wall with radius c 2 . But for such (α, β) the inequality Qα,β(E) ≥ 0 implies the statement. (cid:3) 3. Classifying rank two reflexive sheaves with maximal third Chern character Before stating the theorem we require further notation. Let α > 0, β ∈ R. Any να,β-semistable object E ∈ Cohβ(P3) has a Jordan-Holder filtration 0 = E0 → E1 → . . . → En = E, where all the factors Fi = Ei/Ei+1 are να,β-stable and have slope να,β(Fi) = να,β(E). In Bridgeland stability Jordan-Holder factors are unique up to order. The same is not true in tilt stability. This is a serious issue that we will have to deal with. We say that that E satisfies the JH-property with respect to (α, β) if the Jordan-Holder factors of E are unique up to order. Theorem 3.1. Let E ∈ Cohβ(P3) be a tilt-semistable rank two object with ch(E) = (2, c, d, e). (i) If c = −1, then d ≤ − 1 2 . (a) If d = − 1 2 , then e ≤ 5 6 . In case of equality, E is destabilized by a short exact sequence 0 → O(−1)⊕3 → E → O(−2)[1] → 0. Moreover, E satisfies the JH-property along the wall. (b) If d < − 1 2, then e ≤ d2 2 − d + 5 24 . In case of equality, E is destabilized by a short exact sequence 0 → O(−1)⊕2 → E → OV (cid:18)d − 1 2(cid:19) → 0, where V ⊂ P3 is a plane. Moreover, E satisfies the JH-property the wall. (ii) If c = 0, then d ≤ 0. (a) If d = 0, then e ≤ 0. In case of equality, E ∼= O⊕2. (b) If d = −1, then e ≤ 0. In case of equality, there is a short exact sequence 0 → T (−3) → O(−1)⊕5 → E → 0, where T is the tangent bundle of P3. (c) If d = −2, then e ≤ 2. In case of equality, E is destabilized by a short exact sequence 0 → O(−1)⊕4 → E → O(−2)⊕2[1] → 0. 8 (d) If d = −3, then e ≤ 4. In case of equality, E is destabilized by one of the short exact sequences 0 → O(−1)⊕3 → E → O(−3)[1] → 0, 0 → F → E → OV (−2) → 0, 0 → OV (−2) → E → F → 0, where V ⊂ P3 is a plane and F ∈ M (2, −1, − 1 satisfies E satisfies the JH-property along its destabilizing wall. 2 , 5 6 ). Moreover, any semistable E (e) If d ≤ −4, then In case of equality, E is destabilized via an exact sequence e ≤ d2 2 + d 2 + 1. where V ⊂ P3 is a plane and F ∈ M (2, −1, − 1 property along the wall. 0 → F → E → OV (d + 1) → 0, 2 , 5 6 ). Moreover, E satisfies the JH- As an immediate Corollary, we can determine the maximal third Chern character for rank −2 objects as well. Corollary 3.2. Let E ∈ Cohβ(P3) be a tilt-semistable object with ch(E) = (−2, c, d, e). (i) If c = −1, then d ≥ 1 2 and e ≤ d2 2 + d + 5 24 . (ii) If c = 0, then d ≥ 0. (a) If d = 0 or d = 1, then e ≤ 0. (b) If d ≥ 2, then e ≤ d2 2 − d 2 + 1. Proof. By Proposition 2.8, there is a sheaf T supported in dimension 0 and an object E that is tilt-semistable together with a distinguished triangle T → E → RHom(E, O)[1] → T [1]. We have ch( E) = (2, c, −d, e + ch3(T )), and in particular, e ≤ ch3( E). Applying the bounds in Theorem 3.1 finishes the proof. (cid:3) The strategy to prove Theorem 3.1 is induction on the discriminant. We start with a series of special cases that will either serve as base cases or require special arguments. The bounds in some of these special cases, where already dealt with in [SS18]. Nevertheless, for the convenience of the reader, we include full proofs. Lemma 3.3. Let E ∈ Cohβ(P3) be a tilt-semistable rank two object with ch(E) = (2, 0, 0, e). Then e ≤ 0, and if e = 0, then E ∼= O⊕2. Proof. The fact that E ∈ Cohβ(P3) implies β < 0. The inequality Qα,β(E) ≥ 0 is equivalent to e ≤ 0 for any α > 0. For the fact that e = 0 implies E ∼= O⊕2 see Proposition 2.19. (cid:3) Lemma 3.4. Let E ∈ Cohβ(P3) be a tilt-semistable rank two object with ch(E) = (2, −1, − 1 Then e ≤ 5 6 , then E is destabilized in tilt stability by an exact sequence 6 , and if e = 5 2 , e). 0 → O(−1)⊕3 → E → O(−2)[1] → 0. Moreover, E satisfies the JH-property along the wall. 9 Proof. The point α = 0, β = −1 lies on the numerical wall with center − 3 2 . A straightforward computation shows that for any point inside this numerical wall Qα,β(E) ≥ 0 implies e < 5 6 . Therefore, we only have to deal with objects stable at or outside this wall. 2 and radius 1 We have ch−1(E) = (2, 1, − 1 1 (E) is the minimal positive value, and β = −1 is not the vertical wall, E must be semistable for all α > 0 when β = −1. Finally, Q0,−1 ≥ 0 implies e ≤ 5 (cid:3) 6 . The remaining statement is a special case of [Sch15, Theorem 5.1]. 2 , e − 2 3 ). Since ch−1 Lemma 3.5. Let E ∈ Cohβ(P3) be a tilt-semistable rank two object with ch(E) = (2, 0, −1, e). Then e ≤ 0. If e = 0, then E fits into an exact sequence of the form 0 → T (−3) → O(−1)⊕5 → E → 0. Proof. The fact that E ∈ Cohβ(P3) implies β < 0. The equation ν0,β(E) = 0 holds if and only if β = ±1. If E is destabilized at a semicircular wall, it must intersect the vertical line β = −1. We have ch−1(E) = (2, 2, 0, e − 2 3 ). Assume we have such a wall induced by 0 → F → E → G → 0 that contains a point (α, −1). Since the wall itself is not vertical, ch−1 1 (F ) has to be an integer strictly in between 0 and 2, i.e., ch−1 2 + Z. Then ≤2(F ) = (r, 1, x), where r ∈ Z and x ∈ 1 1 (F ) = 1. We know ch−1 − α2 2 = να,−1(E) = να,−1(F ) = x − α2 2 r. This simplifies to (r − 1)α2 = 2x. If r ≥ 2, then x > 0 and ∆(F ) ≥ 0 implies x ≤ 1 4 . There is no possible value for x with these properties. If r = 1, then x = 0 which is not a valid value for x. If r ≤ 0, then the quotient E/F has positive rank. The same argument with E/F instead of F works. 2r ≤ 1 Overall, there is no wall to the left of the unique vertical wall for E. If H 0(E) 6= 0, we get a non-trivial morphism O → E in contradiction to stability of E. If H 2(E) 6= 0, we can use Serre duality to get a non-trivial morphism E → O(−4)[1]. However, such a morphism would induce a semicircular wall, and therefore, H 2(E) = 0. The Todd class of P3 is given by We get td(TP3) =(cid:18)1, 2, 11 6 , 1(cid:19) . e = χ(E) = −h1(E) − h3(E) ≤ 0. To prove the second statement about the exact sequence, note that ν7/24,−25/24(E) = 0. By Theorem 2.7 and Theorem 2.15 the object E or E[1] is in the finite length category hO(−2)[3], T (−3)[2], O(−1)[1], Oi. The Chern character of E directly implies that it has to be an extension between T (−3)[1] and five copies of O(−1). (cid:3) Lemma 3.6. Let E ∈ Cohβ(P3) be a tilt-semistable rank two object with ch(E) = (2, 0, −2, e). Then e ≤ 2. If e = 2, then E is destabilized in tilt stability by an exact sequence 0 → O(−1)⊕4 → E → O(−2)⊕2[1] → 0. Proof. Assume e ≥ 2. We have ch−1(E) = (2, 2, −1, e − 5 for β = −1. ch−1 F by F/E in the following argument. 3 ). We will show that there is no wall If there is any destabilizing subobject F ⊂ E for some α > 0 and β = −1, then ≤2(F ) = (r, 1, d). We may assume r ≥ 1. If r ≤ 0, then we can simply replace 1 (F ) = 1. Let ch−1 10 If r = 1, we are dealing with the vertical wall, but that is located at β = 0. Thus, r ≥ 2. A straightforward computation shows that the wall occurs for This means d > − 1 2d + 1 r − 1 2 . Furthermore, ∆(F ) ≥ 0 implies α2 = . d ≤ 1 2r < 1 2 . This is a contradiction to the fact that d ∈ 1 [Sch15, Theorem 5.1]. 2 + Z. The remaining statement is a special case of (cid:3) Lemma 3.7. Let E ∈ Cohβ(P3) be a tilt-semistable rank two object with ch(E) = (2, 0, −3, e). Then e ≤ 4. If e = 4, then E is is destabilized in tilt stability by one of the following of the three following sequences 0 → O(−1)⊕3 → E → O(−3)[1] → 0, 0 → F → E → OV (−2) → 0, document 0 → OV (−2) → E → F → 0, where V ⊂ P3 is a plane and F ∈ M (2, −1, − 1 JH-property along these walls. 2 , 5 6 ). Moreover, any semistable E satisfies the Proof. Assume e ≥ 2. A straightforward computation shows that Q0,−1(E) ≥ 0 is equivalent to e ≤ 4. Therefore, we only need to check this inequality for objects that destabilizes along the vertical ray β = −1. We have ch−1(E) = (2, 2, −2, e − 8 3 ). If there is any destabilizing subobject F ⊂ E for some α > 0 and β = −1, then ch−1 ≤2(F ) = (r, 1, d). If r ≤ 0, we replace F by E/F in the following argument. Therefore, we may assume r ≥ 1. If r = 1, we are dealing with the vertical wall, but that is located at β = 0. Thus, r ≥ 2. A straightforward computation shows that the wall occurs for 1 (F ) = 1. Let ch−1 α2 = 2d + 2 r − 1 . This means d > −1. Furthermore, ∆(F ) ≥ 0 implies d ≤ 1 2r < 1 2 . 2 + Z, we have d = − 1 Since d ∈ 1 ch≤2(F ) = (2, −1, − 1 Lemma 2.17 implies ch3(E/F ) ≤ 19 6 , and ch3(E/F ) = 19 have ch3(F ) = 5 2 . Then ∆(E/F ) ≥ 0 implies r = 2. Overall, we have showed 2 ), 6 . Overall, this means e ≤ 4. Moreover, in case of equality we 6 . By Lemma 2.17 we get E/F = OV (−2). 6 . Since ch≤2(E/F ) = (0, 1, − 5 2 ) and by Lemma 3.4 we know ch3(F ) ≤ 5 The first exact sequence in the statement is giving the smallest wall by a special case of [Sch15, (cid:3) Theorem 5.1]. This finishes the special cases. For the rest of the section, we deal with the induction step for the general case. Lemma 3.8. Let E ∈ Coh(P3) be a tilt-semistable object with ch(E) = (2, c, d, e). Assume that either (i) c = −1, d ≤ − 3 (ii) c = 0, d ≤ −4, and e ≥ d2 2 , and e ≥ d2 2 − d + 5 2 + 1. 2 + d 24 , or Then E is destabilized along a semicircular wall by a subobject or quotient of rank at most two. 11 Proof. (i) Assume c = −1, d ≤ − 3 2 , and e ≥ d2 2 − d + 5 24 . Then the radius of Qα,β(E) = 0 is bounded from below by ρ2 Q ≥ 144d4 − 32d3 + 24d2 − 24d + 5 16(4d − 1)2 . We can compute ρ2 Q − ∆(E) 12 ≥ (108d2 − 68d + 11)(2d + 1)2 48(4d − 1)2 > 0. We conclude by Lemma 2.11. (ii) Assume c = 0, d ≤ −4, and e ≥ d2 2 + d 2 + 1. Then the radius of Qα,β(E) = 0 is bounded from below by We can compute ρ2 Q ≥ 9d4 + 34d3 + 45d2 + 36d + 36 16d2 . ρ2 Q − ∆(E) 12 ≥ (27d3 + 37d2 + 24d + 36)(d + 3) 48d2 > 0. We conclude by Lemma 2.11. (cid:3) Lemma 3.9. Let E ∈ Coh(P3) be a tilt-semistable object with ch(E) = (2, c, d, e). (i) Assume c = −1 and d ≤ − 3 2 . If E is destabilized by a subobject F of rank one, then e ≤ d2 2 − d + 5 24 . In case of equality, we have F ∼= O(−1) and E/F is the ideal sheaf of a plane curve. (ii) Assume c = 0 and d ≤ −4. If E is destabilized by a subobject or quotient F of rank one, then e < d2 2 + d 2 + 1. Proof. (i) Assume c = −1 and e ≥ d2 2 − d + 5 24 . We can compute Q0,−2(E) ≤ 4d2 − 20d − 18e + 4 ≤ − (10d − 1)(2d + 1) 4 < 0. Thus, any wall destabilizing E must contain a point (α, −2). In particular, 0 < ch−2 1 (F ) = ch1(F ) + 2 < ch−2 1 (E) = 3. This implies ch1(F ) ∈ {−1, 0}. For proving the bound, we can assume ch1(F ) = −1 and ch(F ) = (1, 0, −y, z) · ch(O(−1)) for some y ≥ 0. If ch1(F ) = 0, the same argument will work when F is replaced by E/F . By Proposition 2.18 we know We can compute z ≤ y(y + 1) 2 . sQ(E) = d + 6e 4d − 1 ≤ 12d2 − 20d + 5 16d − 4 , s(E, F ) = d + 2y − 1. Since s(E, F ) ≤ sQ(E), we have y ≤ 4d2 − 1 8 − 32d 12 < − d 2 − 1 4 . Using Proposition 2.18 on the quotient E/F leads to e ≤ d2 2 + dy + y2 2 − d + z + 5 24 ≤ d2 2 + dy + y2 − d + y 2 + 5 24 . This is a parabola in y with minimum at y = − d at y = 0, where we get 2 − 1 4 . This means the maximum occurs e ≤ d2 2 − d + 5 24 . In case of equality, we must have ch(F ) = ch(O(−1)). By Proposition 2.19, we get F ∼= O(−1). Proposition 2.18 implies that E/F is the ideal sheaf of a plane curve. If instead ch0(F ) = 0, then E/F ∼= O(−1) and F ∼= IC for a plane curve C. Then Ext1(O(−1), IC ) = 0 implies that E is just a direct sum. (ii) Assume c = 0 and e ≥ d2 2 + d 2 + 1. We can compute Q0,−1(E) ≤ 4d2 − 4d − 12e ≤ −2(d + 2)(d + 3) < 0. Thus, any wall destabilizing E must contain a point (α, −1). In particular, 0 < ch−1 1 (F ) = ch1(F ) + 1 < ch−1 1 (E) = 2. This means ch1(F ) = 0, a contradiction to the fact that we are not dealing with the vertical wall. (cid:3) Proof of Theorem 3.1. The proof will be by induction on ∆(E). The start of the induction is done by Lemma 3.3, 3.4, 3.5, 3.6, and 3.7. The induction step will be by contradiction. The strategy is to show that there is no wall outside the semidisk Qα,β(E) < 0 and therefore, there is no wall for such an object unless we have equality in the claimed bound. By Lemma 3.8 we are able to infer that E is destabilized along a semicircular wall W induced by an exact sequence 0 → F → E → G → 0, where F is of rank r ∈ {0, 1, 2}. Note that ∆(F ) < ∆(E), and we intend to use the induction hypothesis on F in case r = 2. (i) Assume c = −1, d ≤ − 3 2 , and e ≥ d2 2 − d + 5 24 . • If r = 1, we can use Lemma 3.9 to get that F = O(−1), and G = IC for a plane curve C. Then there is a map O(−1) → IC. An application of the Snake Lemma shows that there is an injective morphism O(−1)⊕2 ֒→ E. This reduces to the case of rank two walls. • Up to exchanging F and G we can assume r = 2. At the end we will show that F is indeed the subobject and not the quotient. As in the proof of Lemma 3.9, we know that Q0,−2(E) < 0. Moreover, we can compute (E) = 4d2 − 12d − 12e + Q 0,− 3 2 9 4 ≤ −2d2 − 1 4 < 0. 1 1 (F ) < ch−2 (F ) > 0 and ch−2 Both ch−3/2 1 (E) together imply ch1(F ) = −2. Therefore, we may assume that ch(F ) = (2, 0, y, z) · ch(O(−1)) for some y ≤ 0. If y = 0, then z ≤ 0. Applying Lemma 2.17 to the quotient G implies that e ≤ d2 24 with equality if and only if G ∼= OV (d − 1 2 ) for a plane V ⊂ P3. Assume for a contradiction that y ≤ −1. By induction we know that 2 − d + 5 z ≤ y2 2 y 2 + 13 + 1. We can compute sQ(E) = d + 6e 4d − 1 ≤ 12d2 − 20d + 5 16d − 4 , s(E, F ) = d − y − 1. Since s(E, F ) ≤ sQ(E), we have y ≥ 4d2 − 1 16d − 4 > d 2 − 1 4 . If d = − 3 Using Theorem 2.20 on the quotient G leads to 2 , then this means y > −1, a contradiction. Thus, we may assume d ≤ − 5 2 . e ≤ d2 2 − dy + y2 2 − d + z + 5 24 ≤ d2 2 − dy + y2 − d + y 2 + 29 24 . This is a parabola in y with minimum at y = d at y = −1, where we get 2 − 1 4 . This means the maximum occurs e ≤ d2 2 + 41 24 < d2 2 − d + 5 24 . Overall, we showed that the only case in which we can get e = d2 24 is when ch(F ) = 2 ch(O(−1)). We can conclude by Proposition 2.19 that F = O(−1)⊕2. If O(−1)⊕2 is the quotient and not the subobject, then Ext1(O(−1), OV (d − 1 2 )) = 0 shows that E is simply a direct sum. 2 − d + 5 (ii) Assume c = 0, d ≤ −4, and e ≥ d2 2 + 1. By Lemma 3.9, we know that either F or G has to have rank two. Most of the argument is numerical, and for the moment we assume that F has rank two. We will argue at the end that F is indeed the subobject and not the quotient. As in the proof of Lemma 3.9, we get Q0,−1(E) < 0. Thus, 2 + d 0 < ch−1 1 (F ) = ch1(F ) + 2 < ch−1 1 (E) = 2. This means we can assume that ch(F ) = (2, −1, y, z) for some y ≤ − 1 know that 2 . By induction, we We can compute z ≤ y2 2 − y + 5 24 . sQ(E) = 3e 2d ≥ s(E, F ) = d − y. 3d2 + 3d + 6 4d , Since s(E, F ) ≤ sQ(E), we have y ≥ 4d d2 − 3d − 6 > + d 2 Using Theorem 2.20 on the quotient G leads to d2 2 This is a parabola in y with minimum at y = d 2 + 1 y = − 1 − dy + + z + 2 , where we get 1 24 y2 2 d2 2 e ≤ 1 2 . ≤ − dy + y2 − y + 1 4 . 2 . This means the maximum occurs at e ≤ d2 2 d 2 + 14 + 1. 6 ). By Lemma 2.17 we have G ∼= Moreover, equality happens when F ∈ M (2, −1, − 1 OV (d + 1). We are left to show that F is indeed a subobject. As before, the strategy will be to show that Ext1(F, OV (d + 1)) vanishes. By Lemma 3.4 we have a short exact sequence of sheaves. 2 , 5 0 → O(−2) → O(−1)⊕3 → F → 0. The long exact sequence from applying the functor Hom(·, OV (d + 1)) to this sequence immediately concludes the proof. (cid:3) 4. Geometric structure of the moduli spaces From the classification in the last section, we can deduce a geometric description of their moduli spaces. Corollary 4.1. (i) We have M (2, −1, − 1 2 , 5 6 ) ∼= P3 and M (2, 0, −1, 0) ∼= P5. (ii) The moduli space M (2, 0, −3, 4) is the blow up of Gr(3, 10) in a smooth subvariety isomor- phic to P3 × P3. (iii) For d ≤ − 3 2 the moduli space M (2, −1, d, d2 2 − d + 5 24 ) is a Gr(2, n)-bundle over P3, where (iv) For d ≤ −4 the moduli space M (2, 0, d, d2 2 + d 2 + 1) is a Pn-bundle over the product P3 × P3, where n = d(d − 2) − 1. n =(cid:18) 5 2 − d 2 (cid:19). In order to proof this statement we need to recall some notation and known results. Let f : X → Y be a morphism between projective varieties, and let F, G ∈ Coh(X). For any i ∈ Z the relative Ext-sheaf is defined to be Exti f (F, G) := Ri(f∗ Hom(F, ·))(G) = Hi(Rf∗ RHom(F, G)). In [Lan83] Lange constructs universal families of extensions of sheaves using these relative Ext- sheaves. However, the case of the Grassmann bundle requires a few steps beyond what Lange did. Theorem 4.2 ([Lan83, Theorem 1.4]). Let y ∈ Y and assume that the base change morphism τ i(y) : Exti (Fy, Gy) is surjective. f (F, G) ⊗Y C(y) → Exti Xy (i) There is a neighborhood U of y such that the base change morphism τ i(y′) is an isomor- phism for all y′ ∈ U . (ii) The base change morphism τ i−1(y) is surjective if and only if Exti f (F, G) is locally free in a neighborhood of y. Note that Grauert's Theorem [Har77, Corollary III.12.9] shows that if τ i(y) is an isomorphism f (F, G) is locally for all y ∈ Y and the dimension of Exti free. Therefore, Lange's theorem creates opportunities for descending induction on i. (Fy, Gy) is independent of y, then Exti Xy 4.1. The case c = −1, d = − 1 is given by P3, and let M := M (2, −1, − 1 H 0(O(1))∨ ։ U is a three-dimensional quotient, then we get a short exact sequence of sheaves 2 . The moduli space of three-dimensional quotients of H 0(O(1))∨ 6 ). We can define a function ϕ : P3 → M as follows. If 2 , 5 0 → O(−2) → O(−1) ⊗ U → E → 0. We set ϕ(U ) = E. We will have to proof the following lemma. Lemma 4.3. (i) The function ϕ is well-defined, i.e., E is slope-stable. (ii) The function ϕ is bijective 15 (iii) The function ϕ is a morphism of schemes. (iv) The moduli space M is smooth, and therefore, ϕ is an isomorphism. Proof. (i) By Theorem 3.1, we know that there is only one wall for such objects E given by a sequence 0 → O(−1)⊕3 → E → O(−2)[1] → 0. Therefore, showing that E = ϕ(U ) is slope-stable is the same as showing that it is να,β- semistable in a neighborhood above this wall W . If it is not semistable above W , then there is a destabilizing semistable quotient E ։ G. Clearly, E is strictly-semistable along W . Therefore, such a quotient must satisfy να,β(E) = να,β(G) for (α, β) along W . Since E has the JH-property, we know that in any Jordan-Holder filtration of E there are three stable factors O(−1) and one stable factor O(−2)[1]. This means that the stable factors of G have to be a subset of these. A quotient O(−2)[1] does not destabilizes E above W for purely numerical reasons. The vector space Hom(E, O(−1)) is the kernel of the morphism Hom(O(−1) ⊗ U, O(−1)) → Hom(O(−2), O(−1)) which is injective. Thus, G 6∼= O(−1)⊕a for a ∈ {1, 2, 3}. If G is an extension between O(−1) and O(−2)[1], then the kernel is O(−1)⊕2 and this does not destabilize E above the wall. Similarly, if G is an extension between O(−1)⊕2 and O(−2)[1], then the kernel is given by O(−1) which does not destabilize E above the wall. (ii) By Theorem 3.1 we know that any semistable E fits into an exact sequence 0 → O(−1)⊕3 → E → O(−2)[1] → 0. Giving such an extension is the same as giving an element in Ext1(O(−2)[1], O(−1)⊕3) = H 0(O(1))⊕3. In Theorem 3.1 we have already shown that this is the Harder-Narasimhan filtration of E below the wall. The Harder-Narasimhan factors are unique. This means E determines the subobject O(−1)⊕3. However, the group GL(3) acts via automorphisms on O(−1)⊕3 without changing the isomorphism class of E. This means we get a unique subspace of H 0(O(1))⊕3. However, if this subspace is not of dimension three, then there is a destabilizing morphism E ։ O(−1). This proves both surjectivity and injectivity. (iii) We will construct a family on P3 × P3 whose fibers are in bijection with objects in M . The universal property of M then shows that ϕ is a morphism. We have two projections p1, p2 : P3 × P3 → P3. Let O ⊗ H 0(O(1))∨ ։ Q be the universal rank three quotient bundle whose fibers pa- rametrize three-dimensional quotients H 0(O(1))∨ ։ U . We can compose the morphisms p∗ 2O(−1) ⊗ p∗ 2O(−2) → p∗ 1Q. Taking the quotient leads to an object U . By construction this is the desired family. 2O(−1) ⊗ H 0(O(1))∨ → p∗ (iv) In order to show that M is smooth all we have to do is to show that Ext2(E, E) = 0. How- ever, applying the three functors RHom(·, O(−2)[1]), RHom(·, O(−1)), and RHom(E, ·) to 0 → O(−1)⊕3 → E → O(−2)[1] → 0 implies this immediately. (cid:3) 4.2. The case c = 0, d = −1. Note that Hom(T (−3), O(−1)) ∼= C6. Let Q be the generalized Kronecker quiver with two vertices and six arrows between them, all going in the same direction. By Lemma 3.5 and Theorem 2.14 we know that M (2, 0, −1, 0) is isomorphic to the moduli space of quiver representations of Q with dimension vector (1, 5). This space parametrizes six vectors in C5 modulo the action of GL(5). It is not hard to see that this space is P5. 16 4.3. The case c = 0, d = −3. This is the only case in which there is more than one chamber where the moduli space of tilt-semistable objects is non-trivial. By Theorem 3.1 there are exactly two walls in tilt stability for objects with Chern character v = (2, 0, −3, 4). The two walls are W1 = W (cid:18)v,(cid:18)2, −1, − W2 = W (v, O(−1)). 1 2(cid:19)(cid:19) , Note that W2 is located inside W1. By Theorem 3.1, there are no semistable objects inside W2. Let M ′ be the moduli space of tilt-semistable objects in between W1 and W2. Note that M ′ does not parametrize any strictly semistable objects. The first goal is to show that M ′ is isomorphic to Gr(10, 3) the moduli space of three-dimensional quotients of H 0(O(2))∨. We define a function ϕ : Gr(10, 3) → M ′ as follows. If H 0(O(2))∨ → U is a three-dimensional quotient, then we get a short exact sequence of sheaves 0 → O(−3) → O(−1) ⊗ U → E → 0. We set ϕ(U ) = E. Lemma 4.4. (i) The function ϕ is well defined, i.e., E is Gieseker-stable. (ii) The function ϕ is bijective. (iii) The function ϕ is a morphism of schemes. (iv) The moduli space M ′ is smooth, and therefore, ϕ is an isomorphism. Proof. The proof is essentially the same as that of Lemma 4.3 with O(−2) replaced by O(−3). (cid:3) Next, we have to understand crossing the wall W1 to describe M (2, 0, −3, 4). Lemma 4.5. Let F ∈ M (2, 0, − 1 2 , 5 6 ) and V ⊂ P3 be a plane. Then , if i = 0 , if i = 1 , otherwise 1 3 0 exti(F, F ) =  exti(OV (−2), OV (−2)) =  exti(F, OV (−2)) =(1 exti(OV (−2), F ) =(15 1 3 0 0 0 , , , if i = 0 , if i = 1 , otherwise , if i = 1 , otherwise , , if i = 1 , otherwise . Proof. By Theorem 3.1 and Lemma 4.3 there is an exact sequence of sheaves 0 → O(−2) → O(−1)⊕3 → F → 0, where the three linear polynomials defining the first map are linearly independent. The sequence 0 → O(−1) → O → OV → 0 shows that the derived dual of OV is given by OV (1). From here the statement is a straightforward computation involving the appropriate long exact sequences. (cid:3) By Theorem 3.1 any tilt-stable objects E that is destabilized by either 0 → F → E → OV (−2) → 0 or 0 → OV (−2) → E → F → 0 satisfies the JH-property along W1. This means all non-trivial extensions in Ext1(OV (−2), F ) or Ext1(F, OV (−2)) are stable on one side of the wall. 17 Lemma 4.6. The closed subscheme of M ′ parametrizing objects E fitting into a sequence is isomorphic to P3 × P3. 0 → OV (−2) → E → F → 0 Proof. We have showed that the moduli space M (2, −1, − 1 6 ) parametrizes three-dimensional subspaces U ⊂ H 0(O(1)). Moreover, the space M ′ ∼= Gr(3, 10) parametrizes three-dimensional subspaces W ⊂ H 0(O(2)). Any plane in P3 is cut out by a linear equation l. From this we get a closed embedding 2 , 5 P3 × P3 ∼= M (2, −1, − 1 2 , 5 6 ) × Gr(3, 4) ֒→ Gr(3, 10) as follows. If U ⊂ H 0(O(1)) with dim U = 3 and a l is a linear equation cutting out a plane in P3, then we get a three-dimensional subspace l · U ⊂ H 0(O(2)). The goal in this argument is to show that this image is precisely the locus in M ′ that is destabilized at the wall W1. Let W = l · U ⊂ H 0(O(2)) be as above. Then we get a short exact sequence 0 → O(−3) → O(−1) ⊗ U → E → 0. The morphism O(−3) → O(−1) ⊗ W factors through O(−2) → O(−1) ⊗ U whose quotient is an element F ∈ M (2, −1, − 1 6 ). By the Snake Lemma the kernel of E ։ F is given by OV (−2), where V is cut out by l. 2 , 5 Assume vice versa that there is a short exact sequence 0 → OV (−2) → E → F → 0. Then V is cut out by a linear equation l. Since there is also a short exact sequence 0 → O(−3) → O(−1)⊕3 → E → 0, we get a morphism O(−1)⊕3 ։ F whose kernel has to be O(−2). This morphism O(−2) → O(−1)⊕3 gives a three-dimensional subspace U ⊂ H 0(O(1)). By construction the subspace W = l · U ⊂ H 0(O(2)) represents E. (cid:3) We need the following classical result by Moishezon. Recall that the analytification of a smooth proper algebraic spaces of finite type over C of dimension n is a complex manifold with n inde- pendent meromorphic functions. Moishezon's result is originally stated in these terms as his work predated algebraic spaces. Theorem 4.7 ([Moi67]). Any birational morphism f : X → Y between smooth proper algebraic spaces of finite type over C such that the contracted locus E is irreducible and the image f (E) is smooth is the blow up of Y in f (E). Since Ext1(F, OV (−2)) = C independently of F and V , there is a unique stable extension 0 → OV (−2) → E → F → 0 for each F and V . Therefore, we get a morphism M (2, 0, −3, 4) → M ′ which is birational outside of the objects destabilized by this type of sequence. By Lemma 4.6 the exceptional locus maps onto a smooth projective subvariety. The fibers are all irreducible and given by P(Ext1(OV (−2), F )) ∼= P14. Therefore, the exceptional locus is irreducible. Theorem 4.7 concludes the proof together with the following lemma. Lemma 4.8. The moduli space M (2, 0, −3, 4) is smooth. Proof. Applying the three functors RHom(·, F ), RHom(·, OV (−2)), and RHom(E, ·) to leads to 0 → F → E → OV (−2) → 0 Ext2(E, E) = 0. 18 (cid:3) 4.4. The general case with c = −1. Lemma 4.9. Let d ≤ − 3 written as an extension 2 , and let V ⊂ P3 be plane. Giving a slope-stable sheaf E that can be 0 → O(−1)⊕2 → E → OV (cid:18)d − 1 2(cid:19) → 0 is equivalent to giving a subspace of Ext1(OV (d − 1 2 ), O(−1)) of dimension two. Proof. Giving such an extension is the same as giving an element in Ext1(cid:18)OV (cid:18)d − 1 2(cid:19) , O(−1)⊕2(cid:19) . In Theorem 3.1 we have already shown that for semistable objects this short exact sequence is the Harder-Narasimhan filtration of E once E becomes tilt-unstable. The Harder-Narasimhan factors are unique. This means E determines both V and the subobject O(−1)⊕2. However, the group GL(2) acts via automorphisms on O(−1)⊕2 without changing the isomorphism class of E. This means we get a subspace of Ext1(OV (d − 1 2 ), O(−1)). If this subspace is of dimension zero, then E is a direct sum and certainly unstable. Assume the subspace is of dimension one. Then the morphism OV (d − 1 2 ) → O(−1)⊕2[1] factors through O(−1)[1]. The octahedron axiom implies that there is a map E ։ O(−1) in contradiction to stability. Assume vice versa that we have a two dimensional subspace of Ext1(OV (d− 1 2 ), O(−1)). Choosing two arbitrary basis elements leads to an extension 0 → O(−1)⊕2 → E → OV (cid:18)d − 1 2(cid:19) → 0. Its Jordan-Holder factors along This object E is strictly semistable along the induced wall W . the wall are two copies of O(−1) and one copy of OV (d − 1 2 ). The Jordan-Holder factors of any destabilizing subobject must be a subset of these. By construction we have Hom(E, O(−1)) = 0. Therefore, neither OV (d − 1 2 ) and O(−1) can be a subobject of E. (cid:3) 2 ) nor an extension between OV (d − 1 The argument will proceed in three steps. First we construct the Grassmann bundle that we expect to be the moduli space. Then we construct a global family on this space. This family will induce a morphism, and we finish by showing that it is an isomorphism. Let V ⊂ Gr(3, 4) × P3 ∼= P3 × P3 be the universal plane. There are two projections p : Gr(3, 4) × 2 ), O(−1)) = 2 − d)) is independent of the plane V ⊂ P3 and non-zero if and only if i 6= 1. By Theorem P3 → Gr(3, 4) and q : Gr(3, 4) × P3 → P3. The dimension of the group Exti(OV (d − 1 H i−1(OV ( 1 4.2 this implies that A := Ext1 p(OV ⊗ q∗O(d − 1 2 ), q∗O(−1)) ∼= Rp∗ RHom(OV ⊗ q∗O(d − 1 2 ), q∗O(−1))[1] is a vector bundle such that the natural map AV → Ext1(OV (d − 1 every plane V ⊂ P3. 2 ), O(−1)) is an isomorphism for Let Gr(A∨, 2) be the Grassmann bundle parametrizing locally free rank two quotients of A∨. There is a projection π : Gr(A∨, 2) → Gr(3, 4). Let the quotient π∗A∨ ։ Q be the universal quotient bundle. Let E be a stable sheaf as above. Then we get a commutative diagram with exact rows: 19 0 0 / O(−1) ⊗ Ext1(cid:0)OV (cid:0)d − 1 2(cid:1) , O(−1)(cid:1)∨ / O(−1)⊕2 Here the top row is induced by the natural morphism OV (cid:18)d − 1 2(cid:19) → O(−1)[1] ⊗ Hom(cid:18)OV (cid:18)d − / 0 / 0. EV / E OV (cid:0)d − 1 2(cid:1) / OV (cid:0)d − 1 2(cid:1) 2(cid:19) , O(−1)[1](cid:19)∨ 1 . We will globalize this diagram to obtain a family. We can compute Hom(A, A) = Hom(cid:0)p∗A, RHom(OV ⊗ q∗O(d − 1 = Hom(cid:18)p∗A ⊗ OV ⊗ q∗O(cid:18)d − = Hom(cid:18)OV ⊗ q∗O(cid:18)d − 1 1 2 ), q∗O(−1))[1](cid:1) 2(cid:19) , q∗O(−1)[1](cid:19) 2(cid:19) , q∗O(−1) ⊗ p∗A∨[1](cid:19) . Choosing the identity in this group leads to an extension 0 → q∗O(−1) ⊗ p∗A∨ → W → OV ⊗ q∗O(cid:18)d − 1 2(cid:19) → 0, whose restriction to each plane V ⊂ P3 is EV . Let p : Gr(A∨, 2) × P3 → Gr(A∨, 2) be the first projection and let q : Gr(A∨, 2)×P3 → P3 be the second projection. We get a commutative diagram with exact rows: 0 0 / q∗O(−1) ⊗ p∗π∗A∨ W / q∗O(−1) ⊗ p∗Q / U (π × id)∗(cid:0)OV ⊗ q∗O(cid:0)d − 1 2(cid:1)(cid:1) / (π × id)∗(cid:0)OV ⊗ q∗O(cid:0)d − 1 2(cid:1)(cid:1) / 0 / 0. Here U is a family of stable objects with Chern character (2, −1, d, d2 2 − d + 5 that induces a bijective morphism Gr(A∨, 2)) → M (2, −1, d, d2 teristic zero, the following lemma will finish the argument. 2 −d+ 5 24 ) living in Gr(A∨, 2) 24 ). Since we are in charac- Lemma 4.10. The moduli space M (2, −1, d, d2 2 − d + 5 24 ) is smooth. Proof. Applying the three functors RHom(·, O(−1)), RHom(·, OV (d − 1 2 )), and RHom(E, ·) to 0 → O(−1)⊕2 → E → OV (cid:18)d − 1 2(cid:19) → 0 leads to ext1(E, E) = 2 ext1(OV (d − 1 2 ), O(−1)) − 1 = dim Gr(A∨, 2). (cid:3) 4.5. The general case with c = 0. 20 / / /   / /   / / / / / / / /   / /   / / / / / Lemma 4.11. Let F ∈ M (2, −1, − 1 2 , 5 6 ), V ⊂ P3, and d ≤ −4. Then 1 3 0 , if i = 0 , if i = 1 , otherwise exti(F, F ) =  exti(OV (−2), OV (−2)) =  exti(F, OV (d + 1)) =((d + 4)(d + 2) exti(OV (d + 1), F ) =(d(d − 2) , if i = 0 , if i = 1 , otherwise 1 3 0 , , , if i = 2 , otherwise , 0 0 , if i = 1 , otherwise . Proof. By Theorem 3.1 and Lemma 4.3 there is an exact sequence of sheaves 0 → O(−2) → O(−1)⊕3 → F → 0, where the three linear polynomials defining the first map are linearly independent. The sequence shows that the derived dual of OV is given by OV (1). From here the statement is a straightforward computation involving the appropriate long exact sequences. (cid:3) 0 → O(−1) → O → OV → 0 Lemma 4.12. Let d ≤ −4, V ⊂ P3 be a plane, and F ∈ M (2, −1, − 1 stable sheaf E that can be written as an extension 2 , 5 6 ). Giving a 2-Gieseker- 0 → F → E → OV (d + 1) → 0 is equivalent to giving a line in Ext1(OV (d + 1), F ). Proof. Any extension 0 → F → E → OV (d + 1) → 0 corresponds to an element in Ext1(OV (d + 1), F ). In Theorem 3.1 we have already shown that for semistable objects this is the Harder-Narasimhan filtration of E once E becomes tilt-unstable. The Harder-Narasimhan factors are unique. This means E determines both V and F as a subobject of E. Scaling the map F → E does not change the isomorphism class of E. Moreover, if E was a direct sum, it would not be stable. Therefore, we get a line in Ext1(OV (d + 1), F ). Assume vice versa that we have a line in Ext1(OV (d + 1), F ). Choosing an arbitrary non-zero element on this line leads to a non-trivial extension 0 → F → E → OV (d + 1) → 0. This object E is strictly semistable along the induced wall W . By Theorem 3.1 E satisfies the JH-property, and the only relevant destabilizing subobjects of E above the wall could be either F or OV (d + 1). However, F does not destabilize E for purely numerical reasons, and the fact that the exact sequence does not split excludes OV (d + 1). (cid:3) Next we have to construct the variety that we expect to be our moduli space. Let W be the 6 ) ∼= P3, and let V ⊂ Gr(3, 4) × P3 ∼= P3 × P3 be the universal universal family on M (2, −1, − 1 2 , 5 21 plane. We have projections 2 , 5 2 , 5 2 , 5 2 , 5 By Theorem 4.2 and Lemma 4.11 we get that p12 : Gr(3, 4) × M (2, −1, − 1 p13 : Gr(3, 4) × M (2, −1, − 1 p23 : Gr(3, 4) × M (2, −1, − 1 p3 : Gr(3, 4) × M (2, −1, − 1 2 , 5 6 ), 6 ) × P3 → Gr(3, 4) × M (2, −1, − 1 6 ) × P3 → Gr(3, 4) × P3, 6 ) × P3 → M (2, −1, − 1 2 , 5 6 ) × P3 → P3. 6 ) × P3, A := Ext1 p12(p∗ 13OV ⊗ p∗ 3O(d + 1), p∗ 23W) ∼= Rp12∗ RHom(p∗ 13OV ⊗ p∗ 3O(d + 1), p∗ 23W)[1] is a vector bundle. Let π : P(A∨) → Gr(3, 4)×M (2, −1, − 1 6 ) be the projection from the projective bundle of locally free rank one quotients of A∨ to its base. Furthermore, we have a relatively ample line bundle Oπ(1) on this projective bundle and a projection q : P(A∨) × P3 → P(A∨). By [Lan83, Corollary 4.5] there is an extension 2 , 5 0 → (π × id)∗p∗ 23W ⊗ q∗Oπ(1) → U → (π × id)∗(p∗ 13OV ⊗ p∗ 3O(d + 1)) → 0, such that the fibers of U are in bijection with non-trivial extensions 0 → F → E → OV (d + 1) → 0 2 , 5 as previously. This family satisfies a universal property on the category of noetherian (P3)∨ × M (2, −1, − 1 6 )-schemes, but this is not the universal property we need on the category of noe- therian C-schemes. Regardless, the universal property of M (2, 0, −d, d2 2 + 1) implies that there is a bijective morphism P(A∨) → M (2, 0, d, d2 2 + 1). We are done if we can show that M (2, 0, d, d2 2 + d 2 + d 2 + d 2 + 1) is smooth. Lemma 4.13. The moduli space M (2, 0, d, d2 2 + d 2 + 1) is smooth. Proof. Applying the three functors RHom(·, F ), RHom(·, OV (d + 1)), and RHom(E, ·) to 0 → F → E → OV (d + 1) → 0 leads to ext1(E, E) = d(d − 2) + 5 = dim P(A∨). (cid:3) References [AB13] D. Arcara and A. Bertram. Bridgeland-stable moduli spaces for K-trivial surfaces. J. Eur. Math. Soc. (JEMS), 15(1):1 -- 38, 2013. With an appendix by Max Lieblich. [AJT18] C. Almeida, M. Jardim, and A. S. Tikhomirov. Irreducible components of the moduli space of rank two sheaves of odd degree on P3, 2018. In preparation. [AJTT17] C. Almeida, M. Jardim, A. Tikhomirov, and S. Tikhomirov. New moduli components of rank 2 bundles on projective space, 2017. arXiv:1702.06520v1. [BMS16] A. Bayer, E. Macr`ı, and P. Stellari. The space of stability conditions on abelian threefolds, and on some Calabi-Yau threefolds. Invent. Math., 206(3):869 -- 933, 2016. [BMT14] A. Bayer, E. Macr`ı, and Y. Toda. Bridgeland stability conditions on threefolds I: Bogomolov-Gieseker type [Bog78] inequalities. J. Algebraic Geom., 23(1):117 -- 163, 2014. F. A. Bogomolov. Holomorphic tensors and vector bundles on projective manifolds. Izv. Akad. Nauk SSSR Ser. Mat., 42(6):1227 -- 1287, 1439, 1978. T. Bridgeland. Stability conditions on K3 surfaces. Duke Math. J., 141(2):241 -- 291, 2008. [Bri08] [BvdB03] A. Bondal and M. van den Bergh. Generators and representability of functors in commutative and non- [CH16] commutative geometry. Mosc. Math. J., 3(1):1 -- 36, 258, 2003. I. Coskun and J. Huizenga. The ample cone of moduli spaces of sheaves on the plane. Algebr. Geom., 3(1):106 -- 136, 2016. [Cha83] M. Chang. Stable rank 2 bundles on P3 with c1 = 0, c2 = 4, and α = 1. Math. Z., 184(3):407 -- 415, 1983. 22 [GHS18] P. Gallardo, C. Lozano Huerta, and B. Schmidt. Families of elliptic curves in P3 and Bridgeland stability. [Gie79] [Har77] Michigan Math. J., 67(4):787 -- 813, 2018. D. Gieseker. On a theorem of Bogomolov on Chern classes of stable bundles. Amer. J. Math., 101(1):77 -- 85, 1979. R. Hartshorne. Algebraic geometry. Springer-Verlag, New York-Heidelberg, 1977. Graduate Texts in Math- ematics, No. 52. R. Hartshorne. Stable reflexive sheaves. Math. Ann., 254(2):121 -- 176, 1980. R. Hartshorne. Stable reflexive sheaves. III. Math. Ann., 279(3):517 -- 534, 1988. [Har80] [Har88] [HRS96] D. Happel, I. Reiten, and S. O. Smalø. Tilting in abelian categories and quasitilted algebras. Mem. Amer. Math. Soc., 120(575):viii+ 88, 1996. [JMT17a] M. Jardim, M. Maican, and A. S. Tikhomirov. Moduli spaces of rank 2 instanton sheaves on the projective space. Pacific J. Math., 291(2):399 -- 424, 2017. [JMT17b] M. Jardim, D. Markushevich, and A S. Tikhomirov. Two infinite series of moduli spaces of rank 2 sheaves [JV14] [Kin94] on P3. Ann. Mat. Pura Appl. (4), 196(4):1573 -- 1608, 2017. M. Jardim and M. Verbitsky. Trihyperkahler reduction and instanton bundles on CP3. Compos. Math., 150(11):1836 -- 1868, 2014. A. D. King. Moduli of representations of finite-dimensional algebras. Quart. J. Math. Oxford Ser. (2), 45(180):515 -- 530, 1994. H. Lange. Universal families of extensions. J. Algebra, 83(1):101 -- 112, 1983. [Lan83] [Mac14a] A. Maciocia. Computing the walls associated to Bridgeland stability conditions on projective surfaces. Asian J. Math., 18(2):263 -- 279, 2014. [Mac14b] E. Macr`ı. A generalized Bogomolov-Gieseker inequality for the three-dimensional projective space. Algebra Number Theory, 8(1):173 -- 190, 2014. [Man81] N. Manolache. Rank 2 stable vector bundles on P3 with Chern classes c1 = −1, c2 = 2. Rev. Roumaine [Moi67] [MT94] [PT15] Math. Pures Appl., 26(9):1203 -- 1209, 1981. B. Moishezon. On n-dimensional compact complex varieties with n algebraic independent meromorphic functions. Transl., Am. Math. Soc., (63):51 -- 177, 1967. R. M. Mir´o-Roig and G. Trautmann. The moduli scheme M (0, 2, 4) over P3. Math. Z., 216(2):283 -- 315, 1994. E. Macr`ı and B. Schmidt. Derived categories and the genus of space curves, 2018. arXiv:1801.02709v1. [MS18] [Mum62] D. Mumford. Further pathologies in algebraic geometry. Amer. J. Math., 84:642 -- 648, 1962. [OS85] C. Okonek and H. Spindler. Das Spektrum torsionsfreier Garben. II. In Seminar on deformations ( L´od´z/Warsaw, 1982/84), volume 1165 of Lecture Notes in Math., pages 211 -- 234. Springer, Berlin, 1985. D. Piyaratne and Y. Toda. Moduli of Bridgeland semistable objects on 3-folds and Donaldson-Thomas invariants, 2015. arXiv:1504.01177v2. [Rei78] M. Reid. Bogomolov's theorem c2 1 ≤ 4c2. In Proceedings of the International Symposium on Algebraic [Sch15] [SS18] [Vak06] Geometry (Kyoto Univ., Kyoto, 1977), pages 623 -- 642. Kinokuniya Book Store, Tokyo, 1978. B. Schmidt. Bridgeland stability on threefolds - Some wall crossings, 2015. arXiv:1509.04608v1. B. Schmidt and B. Sung. Discriminants of stable rank two sheaves on some general type surfaces, 2018. In preparation. R. Vakil. Murphy's law in algebraic geometry: badly-behaved deformation spaces. Invent. Math., 164(3):569 -- 590, 2006. The University of Texas at Austin, Department of Mathematics, 2515 Speedway, RLM 8.100, Austin, TX 78712, USA E-mail address: [email protected] URL: https://sites.google.com/site/benjaminschmidtmath/ 23
1512.09028
2
1512
2018-02-22T18:55:32
The Classification of Real Singularities Using Singular. Part III: Unimodal Singularities of Corank 2
[ "math.AG", "math.AC" ]
We present a classification algorithm for isolated hypersurface singularities of corank 2 and modality 1 over the real numbers. For a singularity given by a polynomial over the rationals, the algorithm determines its right equivalence class by specifying all representatives in Arnold's list of normal forms (Arnold et al. 1985) belonging to this class, and the corresponding values of the moduli parameter. We discuss how to computationally realize the individual steps of the algorithm for all singularities in consideration, and give explicit examples. The algorithm is implemented in the Singular library realclassify.lib.
math.AG
math
THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR PART III: UNIMODAL SINGULARITIES OF CORANK 2 JANKO BÖHM, MAGDALEEN S. MARAIS, AND ANDREAS STEENPA(cid:223) Abstract. We present a classification algorithm for isolated hypersurface singularities of corank 2 and modality 1 over the real numbers. For a singularity given by a polynomial over the rationals, the algorithm determines its right equivalence class by specifying all representatives in Arnold's list of normal forms (Arnold et al., 1985) belonging to this class, and the corresponding values of the moduli parameter. We discuss how to computationally realize the individual steps of the algorithm for all singularities under consideration, and give explicit examples. The algorithm is imple- mented in the Singular library realclassify.lib. 1. Introduction In the ground breaking work on singularities by Arnold (1974), all isolated hypersurface singularities over the complex numbers up to modality 2 have been classified. Arnold has also given fundamental theorems on the classification of real singularities up to modality 1, which has been made explicit in Arnold et al. (1985). For his classification, Arnold considers stable equivalence of functions. Two function germs are stably equivalent if they are right equivalent after the direct addition of a non-degenerate quadratic form. Over a field K, two function germs f, g ∈ m2 ⊂ K[[x1, . . . , xn]], where m = (cid:104)x1, . . . , xn(cid:105), are right equivalent if there is a K-algebra automorphism φ of K[[x1, . . . , xn]] such that φ(f ) = g. For K = C, modality ≤ 2, and K = R, modality ≤ 1, Arnold presents a finite list of normal forms, each of which is a family of polynomial equations with moduli parameters such that each equivalence class contains at least one, but finitely many, elements of these families. We refer to such elements as normal form equations. Over the complex numbers, Arnold gives an algorithmic determinator which, for certain classes of power series, determines a normal form, in which a representative of a given power series lies. In Boehm et al. (2017a) an algorithm is presented which, in addition, computes an explicit complex normal form equation for power series of modality and corank less or equal to 2. For polynomial input, this algorithm has been implemented in the library (Boehm et al., 2017b) for the computer algebra system Singular (Decker et al., 2017). Considering singularities over the reals, an algorithm for computing the degenerate part and the inertia index is given in Marais and Steenpass (2015a). For singularities of modality 0, an algorithm to determine the corresponding normal form is developed in the same article. In the case of singularities of modality 1, it turns out that an equivalence class may contain several normal form equations, as specified by Arnold. In Marais and Steenpass (2015b), a complete classification of singularities up to modality 1 and corank 2 is given, in Date: 7 November 2017. Key words and phrases. hypersurface singularities, algorithmic classification, real algebraic geometry . This research was supported by grant KIC14081491583 and grant 109327 (Incentive Funding for Rated Researches) of the National Research Foundation (NRF), by grants awarded by the University of Pretoria, and by SPP 1489 and SFB-TRR 195 of the German Research Foundation (DFG). Part of the research was done during a visit of the second author at the Riemann Center for Geometry and Physics at Leibniz Universität Hannover supported by a Riemann fellowship. 1 THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 2 the sense that all complex and real equivalences between complex, respectively real, normal form equations are determined. In this paper, we develop a determinator for real singularities of modality 1 and corank 2, which computes, for a given input polynomial, all normal form equations in Arnold's list (see Table 1). In this setting, the complex types correspond to real main types, which split up into real subtypes by modifying the signs of the terms in the normal form of the real main type, except in the case Yr,r. This complex normal form splits up into the real main types (cid:101)Yr and Yr,r. In fact, we describe an algorithm which computes, for an arbitrary input polynomial f ∈ m3 ⊂ Q[x, y], the following data: all real singularity subtypes of f as well as all normal form equations in the right equivalence class of f with the respective parameter given as the unique real root of a minimal polynomial over Q in a specified interval. The algorithm is implemented in the Singular library realclassify.lib. This paper is structured as follows: In Section 2, we introduce the fundamental definitions and give the required background on the classification of singularities. In Section 3, we develop a general algorithm for the classification of real singularities of modality 1 and corank 2. Although the general algorithm is applicable to all cases under consideration, some steps do not have a straight-forward implementation for certain types of singularities. Therefore, in the subsequent sections, we give explicit computational realizations for all steps of the algorithm. The exceptional cases (W12, W13, Z11, Z12, Z13, E12, E13, E14) follow the general algorithm in a direct way. In Section 4, as an example, we give an explicit algorithm for the case E14. Section 5 handles the parabolic cases (X9, J10). Section 6 deals with the hyperbolic cases (X9+k, J10+k, Yr,s, (cid:101)Yr). The cases X9+k and J10+k follow the general algorithm in a straight-forward manner, whereas the cases Yr,s and (cid:101)Yr require some attention to detail. Acknowledgements. We would like to thank Gert-Martin Greuel, and Gerhard Pfister for many fruitful discussions, and Domnik Bendle and Clara Petroll for implementing a mul- tivariate real root isolation algorithm in the Singular-library rootisolation.lib Bendle et al. (2017). 2. Definitions and Preliminary Results In this section we give some basic definitions and results, as well as some notation that will be used throughout the paper. We only consider classes of germs with respect to right equivalence. Definition 2.1. Let K be a field. Two power series f, g ∈ K[[x1, . . . , xn]] are called right K∼ g, if there exists a K-algebra automorphism φ of K[[x1, . . . , xn]] equivalent, denoted by f such that φ(f ) = g. If K = R, we also write ∼ to denote R∼. Using this equivalence relation, Arnold (1974) gives the following formal definition of a normal form. From now on, let K be either R or C. Definition 2.2. Let K ⊂ K[[x1, . . . , xn]] be a union of equivalence classes with respect to the relation K∼. A normal form for K is given by a smooth map Φ : B −→ K[x1, . . . , xn] ⊂ K[[x1, . . . , xn]] of a finite dimensional K-linear space B into the space of polynomials for which the following three conditions hold: (1) Φ(B) intersects all equivalence classes of K; (2) the inverse image in B of each equivalence class is finite; (3) Φ−1(Φ(B) \ K) is contained in a proper hypersurface in B. THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 3 We call the elements of the image of Φ normal form equations. Remark 2.3. There is a type T associated to each of Arnold's corank 2 real normal forms of modality 0 and 1 (see Table 1), and we denote the corresponding normal form by NF(T ). Depending on whether T is a real or a complex type, we consider K = R or K = C. For b ∈ par(NF(T )) := Φ−1(K) with K as in Definition 2.2, we write NF(T )(b) = Φ(b) for the corresponding normal form equation. We briefly introduce the concepts of weighted jets, filtrations, Newton polygons, and permissible chains. For background regarding the definitions in this section, we refer to Arnold (1974) and de Jong and Pfister (2000). Definition 2.4. Let w = (c1, . . . , cn) ∈ Nn be a weight on the variables (x1, . . . , xn). We If the weight on all variables is 1, we call the weighted degree of a monomial m the standard degree of m and write deg(m) instead of w-deg(m). define the weighted degree on Mon(x1, . . . , xn) by w-deg((cid:81)n i ) = (cid:80)n i=1 cisi. i=1 xsi A polynomial is called quasihomogeneous of degree d with respect to w if w-deg(t) = d for any term t of f. Definition 2.5. Let w = (w1, . . . , ws) ∈ (Nn)s be a finite family of weights on the variables (x1, . . . , xn). For any monomial m ∈ K[x1, . . . , xn] (or term t = c · m, c ∈ K), we define the piecewise weight with respect to w as w-deg(m) := min i=1,...,s wi-deg(m). (1) Let f =(cid:80)∞ Definition 2.6. Let w be a (piecewise) weight on Mon(x1, . . . , xn). i=0 fi be the decomposition of f ∈ K[[x1, . . . , xn]] into weighted homoge- neous parts fi of w-degree i. We denote the weighted j-jet of f by w-jet(f, j) := fi . j(cid:88) i=0 The sum of terms of f of lowest w-degree is the principal part of f with respect to w. (2) A power series in K[[x1, . . . , xn]] has filtration d ∈ N if all its monomials are of weighted degree d or higher. The power series of filtration d form a vector space d ⊂ K[[x1, . . . , xn]] . Ew (3) A power series f ∈ K[[x1, . . . , xn]] is a germ with non-degenerate Newton prin- cipal part if f has filtration d ∈ N with respect to w and if the principal part w-jet(f, d) is non-degenerate, that is, if its Milnor number is finite. If w consists out of a single weight, we call f semi-quasihomogeneous and w-jet(f, d) the quasiho- mogeneous part of f. (4) A power series f ∈ K[[x1, . . . , xn]] is weighted k-determined with respect to the weight w if f ∼ w-jet(f, k) + g for all g ∈ Ew k+1. We define the weighted determinacy of f as the minimum number k such that f is k-determined. Remark 2.7. (1) If for a given type T , w-jet(NF(T )(b), j) is independent of b ∈ par(NF(T )), (2) Note that d < d(cid:48) implies Ew we denote it by w-jet(T, j). d(cid:48) ⊆ Ew filtration d(cid:48) + d, it follows that Ew d(cid:48) · Ew (3) If the weight of each variable is 1, we write Ed and jet(f, j) instead of Ew d . Since elements of the product Ew d is an ideal in the ring of power series. d have d and w-jet(f, j), respectively. THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 4 There are also similar concepts for coordinate transformations: Definition 2.8. Let φ be a K-algebra automorphism of K[[x1, . . . , xn]] and let w be a weight on Mon(x1, . . . , xn). (1) For j > 0 we define the w-jet(φ, j), denoted by φw j (xi) := w-jet(φ(xi), w-deg(xi) + j) φw j , to be the automorphism given by for all i = 1, . . . , n . j . If the weight of each variable is 1, that is, w = (1, . . . , 1), we simply write φj for φw (2) φ has filtration d if, for all λ ∈ N, (φ − id)Ew λ ⊂ Ew λ+d . Remark 2.9. Note that φ0(xi) = jet(φ(xi), 1) for all i = 1, . . . , n. Furthermore note that 0 has filtration ≤ 0 and that φw φw For the next definition we restrict ourselves to the power series ring K[[x, y]] in two j has filtration j if j > 0 and φw j−1 = id. variables. Definition 2.10. Let f =(cid:80) 0 (cid:54)= b ∈ par(NF(T )). We call i,j ai,jxiyj ∈ K[[x, y]], let T be a corank 2 singularity type and supp(f ) supp(T ) := {xiyj ai,j (cid:54)= 0} and := supp(NF(T )(b)), where b ∈ par(NF(T )) is generic, the support of f and NF(T ), respectively. We write supp(T, j) := supp(jet(NF(T )(b), j)) for j ≥ 0. Let (cid:91) Γ+(T ) := xiyj∈supp(T ) ((i, j) + R2 +) and let Γ(T ) be the boundary of the convex hull of Γ+(T ) in R2. Then: (1) Γ(T ) is called the Newton polygon of NF(T ). (2) The compact segments of Γ(T ) are called faces. (3) Let ∆ be a face of Γ(T ). Then ∆ induces a weight w on Mon(x, y) in the following , in lowest terms, and wx, wy > 0, we set w-deg(x) = wx and way: If ∆ has slope − wx w-deg(y) = wy. (4) If w1, . . . , ws are the weights associated to the faces of Γ(T ), ordered by increasing slope, there are unique minimal integers λ1, . . . , λs ≥ 1 such that the piecewise weight associated to w(T ) = (λ1w1, . . . , λsws) by Definition 2.5 is constant on Γ(T ). We denote this constant by d(T ). (5) A monomial m lies strictly underneath, on or above Γ(T ) if, respectively, the w(T )- wy degree of m is less than, equal to or greater than d(T ). Notation 2.11. Given f ∈ K[[x, y]] and m ∈ Mon(x, y), we write coeff(f, m) for the coefficient of m in f. Definition 2.12. Let f ∈ K[x, y] be piecewise quasihomogeneous such that each partial derivative of f is a binomial. (1) The convex hull of the exponent vectors of the monomials of ∂f ∂x is called the funda- (2) A translate of a fundamental segment by an integral vector with non-negative com- mental x-segment, similarly for ∂f ∂y . ponents is called a permissible segment. (3) Two permissible segments are said to be joined if they have a common end. (4) A permissible chain is a sequence of permissible segments such that each consecutive two of them are joined. THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 5 See (Arnold, 1974, Example 9.6) for an example. Remark 2.13. (1) The assumptions of Definition 2.12 are satisfied for the principal parts of all corank 2 singularities of modality 0 and 1. (2) Suppose f is a corank 2 singularity of modality 1 whose Newton polygon has two faces. Then a permissible x-segment is not parallel to any permissible y-segment, that is, they are either disjoint, or intersect in a single point. Definition 2.14. The Jacobian ideal Jac(f ) ⊂ K[[x, y]] of f is generated by the partial derivatives of f. The local algebra Q(f ) is the residue class ring of the Jacobian ideal of f. Proposition 2.15 (Arnold (1974), Proposition 9.7). If the exponent vector of a monomial lies in an infinite permissible chain, then it lies in the Jacobian ideal of f. In (Arnold, 1974, Section 7) the following results are used for the classification of singu- larities of corank 2. Definition 2.16. Suppose f is a germ, e1, . . . , eµ are monomials representing a basis of the local algebra of f, and e1, . . . , es are the monomials in this basis above or on Γ(T ). We then call e1, . . . , es a system of the local algebra of f. Theorem 2.17. Let f be a semi-quasihomogeneous function with quasihomogeneous part f0, and let e1, . . . , eµ be monomials representing a basis of the local algebra of f0. Then e1, . . . , eµ also represent a basis of the local algebra of f. Theorem 2.18. Let f be a semi-quasihomogeneous function with quasihomogeneous part f0 and let e1, . . . , es be a system of the local algebra of f0. Then f is equivalent to a function of the form f0 +(cid:80) ckek. Arnold proved Theorem 2.18 by iteratively applying the following lemma. Lemma 2.19. Let f0 be a quasihomogeneous function of weighted w-degree dw and let e1, . . . , er be the monomials of a given degree d(cid:48) > dw in a system of the local algebra of f0. Then, for every series of the form f0 + f1, where the filtration of f1 is greater than dw, we have f0 + f1 ∼ f0 + f(cid:48) 1, 1 of degree less than d(cid:48) are the same as in f1, and the part of degree d(cid:48) where the terms in f(cid:48) can be written as c1e1 + ··· + crer, c1, . . . , cr ∈ R. Proof. Let g(x) denote the sum of the terms of degree d(cid:48) in f1. There exists a decomposition of g of the form (cid:88) i g(x) = vi(x) + c1e1 + ··· + crer, ∂f0 ∂x since e1, . . . , er represents a monomial vectorspace basis for the local algebra of f0. Applying the transformation defined by xi (cid:55)→ xi − vi(x) to f, we transform f to f0(x) + (f1(x) + (c1e1(x) + ··· + crer(x) − g(x)) + R(x), where the filtration of R is greater than d(cid:48). (cid:3) The following result is proved for a ≥ 4, b ≥ 5 and K = C by Arnold (1974). The same proof is also valid for a = 3, b ≥ 7. Moreover the proof is also valid for K = R. Hence we obtain the following more general result. THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 6 Lemma 2.20. Let K be either R or C. Suppose that f ∈ K[[x, y]] is a germ with non- degenerate Newton principal part f0 = xa + λx2y2 + yb, where 0 (cid:54)= λ ∈ K and a, b ∈ N, such that Γ(f0) has two faces. Then f ∼ f0. Remark 2.21. Note that it follows from Lemma 2.19 and Lemma 2.20 that all germs with a non-degenerate Newton principal part of modality 1, corank 2 are finitely weighted determined. Moreover, we explicitly obtain the weighted determinacy for every such germ. Notation 2.22. A system of a local algebra is in general not unique. For his lists of normal forms of hypersurface singularities with non-degenerate Newton boundary, Arnold has chosen a specific system for the local algebra in each case. In the rest of the paper, we call these systems the Arnold systems. We refer to a set of parameters monomials in a normal form as a parameter system. Again for his lists of normal forms, Arnold has chosen in each case a specific parameter system. We refer to these as the Arnold parameter systems. Note that in the case of a normal form with non-degenerate Newton boundary, corank 2 and modality 1, a system is a parameter system. We proof the following result in addition to (Marais and Steenpass, 2015a, Lemma 9) which has similar results for linear factorizations. Lemma 2.23. Let n ∈ N, n ≥ 2. Suppose f ∈ Q[x, y] is homogeneous and factorizes as 1 (g2) over R, where g1 is a polynomial of degree 1 and g2 is a polynomial of degree 2 that gn does not have a multiple root over C. Then f = ag(cid:48)n 1 g(cid:48) 1 is a polynomial of degree 1 over Q, g2 is a polynomial of degree 2 over Q and a ∈ Q. Proof. Let f = (a1x + a2y)n(a3x2 + a4xy + a5y2) with ai ∈ R. Since the coefficient of xn+2 in f is a rational number, it follows that 2, where g(cid:48) (x + a2 a1 Since Q is a perfect field, (x + a2 Q[x, y] which implies that x + a2 a1 g(cid:48) y2 and a = an 2 = x2 + a4 a3 xy + a5 a3 a1 xy + y)n(x2 + y2) ∈ Q[x, y]. y2) ∈ Q[x, y]. Hence (x + a2 a5 a3 xy + a5 a3 a4 a3 y)(x2 + a4 y ∈ Q[x, y]. Therefore f = ag(cid:48)n a3 1 g(cid:48) 1 a3. 2, where g(cid:48) a1 y)n−1 ∈ y, (cid:3) 1 = x + a2 a1 3. General Classification Algorithm In this section we outline the general structure of an algorithm (see Algorithm 5) to determine, for a given input polynomial f ∈ Q[x, y], the real types as well as, for each type, the corresponding normal form equations to which f is equivalent (see Table 1). Each parameter is given as the unique root of its minimal polynomial over Q in a specified interval. Figures 1 to 4 show in the gray shaded area all monomials which can possibly occur in a polynomial f of the given type T . The Newton polygon Γ(T ) is shown as the blue line with the non-moduli monomials of NF(T ) shown as blue dots. Red dots indicate monomials which are not in Jac(f ). The dot with the thick black circle indicates the moduli monomial in the Arnold system. We start the classification process by classifying the input polynomial according to the complex classification in Arnold et al. (1985). We use the Singular library classify2.lib (Boehm et al., 2017b) for this purpose. As shown by Arnold (1974), the complex types correspond to real main types, which split up into real subtypes by modifying the signs of the terms in the normal form of the real main type, except in the case Yr,r. This complex type splits up into the real main types (cid:101)Yr and Yr,r. The (cid:101)Yr case differs from all the other cases in the sense that these singularities are not right equivalent over the real numbers to a THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 7 Table 1. Normal forms of singularities of modality 1 and corank 2 as given in Arnold et al. (1985) Complex normal form Normal forms of real subtypes 9 9 ) ) ) 9 ) +x4 + ax2y2 + y4 (X ++ −x4 + ax2y2 − y4 (X−− +x4 + ax2y2 − y4 (X +− −x4 + ax2y2 + y4 (X−+ 9 x3 + ax2y2 + xy4 (J + 10) x3 + ax2y2 − xy4 (J− 10) x3 + x2y2 + ay6+k (J + 10+k) x3 − x2y2 + ay6+k (J− 10+k) +x4 + x2y2 + ay4+k (X ++ 9+k) −x4 − x2y2 + ay4+k (X−− 9+k) +x4 − x2y2 + ay4+k (X +− 9+k) −x4 + x2y2 + ay4+k (X−+ 9+k) +x2y2 + xr + ays (Y ++ r,s ) −x2y2 − xr + ays (Y −− r,s ) +x2y2 − xr + ays (Y +− r,s ) −x2y2 + xr + ays (Y −+ r,s ) r ) r ) +(x2 + y2)2 + axr ((cid:101)Y + −(x2 + y2)2 + axr ((cid:101)Y − Restrictions a2 (cid:54)= +4† a2 (cid:54)= +4† a (cid:54)= 0, k > 0 a (cid:54)= 0, k > 0 a (cid:54)= 0, r, s > 4 a (cid:54)= 0, r > 4 x4 + ax2y2 + y4 c X9 i l o b a r a P J10 x3 + ax2y2 + xy4 J10+k x3 + x2y2 + ay6+k X9+k x4 + x2y2 + ay4+k c i l o b r e p y H l a n o i t p e c x E Yr,s x2y2 + xr + ays (cid:101)Yr E12 E13 E14 Z11 Z12 Z13 (x2 + y2)2 + axr x3 + y7 + axy5 x3 + xy5 + ay8 x3 + y8 + axy6 x3y + y5 + axy4 x3y + xy4 + ax2y3 x3y + y6 + axy5 W12 x4 + y5 + ax2y3 W13 x4 + xy4 + ay6 x3 + y7 + axy5 x3 + xy5 + ay8 x3 + y8 + axy6 (E+ 14) x3 − y8 + axy6 (E− 14) x3y + y5 + axy4 x3y + xy4 + ax2y3 x3y + y6 + axy5 (Z + 13) x3y − y6 + axy5 (Z− 13) +x4 + y5 + ax2y3 (W + 12) −x4 + y5 + ax2y3 (W − 12) +x4 + xy4 + ay6 (W + 13) −x4 + xy4 + ay6 (W − 13) - - - - - - - - †Note that the restriction a2 (cid:54)= 4 applies to the normal forms of the real subtypes X ++ well as to the normal forms of the complex types X9 and J10. 9 , X−− 9 , and J + 10 as THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 8 germ with a non-degenerate Newton principal part.1 Therefore we postpone the discussion applicable to this case. Remark 3.1. We can easily distinguish whether an input polynomial f is of real main type of the case (cid:101)Yr to Section 6.3, where we will give a modification of the general algorithm Yr,r or (cid:101)Yr, using (Marais and Steenpass, 2015a, Proposition 8) by considering the number of type (cid:101)Yr does not have any real roots. The Singular library rootsur.lib (Tobis, 2012) real roots (i.e., real homogeneous linear factors) of jet(f, 4). The 4-jet of functions of real type Yr,s has four real roots (counted with multiplicity), while the 4-jet of functions of real can be used to determine the number of roots. In all the other cases, the Newton polygon Γ(T ) of a real subtype T coincides with the Newton polygon of its corresponding complex and main real type. Remark 3.2. Note that, according to Marais and Steenpass (2015b), normal form equations of different complex types, and hence of different real main types, cannot be equivalent. Nor- mal form equations of types Yr,r and (cid:101)Yr can be complex equivalent, but not real equivalent. Suppose NF(T )(b), b ∈ par(NF(T )), is one of the normal form equations to which f is equivalent. Then there exists an R-algebra automorphism φ of R[[x, y]] such that f = φ(NF(T )(b)). We write w = w(T ) for the piecewise weight induced by Γ(T ) and let dw = d(T ) be the degree of the monomials on Γ(T ) with respect to w. Our first goal is to transform f, and thus φ, iteratively by composing φ with a suitable R-algebra automorphism such that, after every step, we have and, after step i, we have if Γ(T ) has one face, and f = φ(NF(T )(b)) supp(jet(f, i)) = supp(T, i) supp(w-jet(jet(f, i), 0) = supp(T, i) if Γ(T ) has two faces. This implies that, after a finite number of steps, we may assume, by an appropriate choice of φ, that φw 0 (x) = cxx and φw 0 (y) = cyy, where w is the weight defined by Γ(T ). After this, in the case where Γ(T ) has two faces, The- orem 2.20 can be used to determine the equivalence class of f by eliminating the monomials above Γ(T ). In the case where Γ(T ) has only one face, the algorithmic proof of Theorem 2.18 can be used to eliminate all monomials above Γ(T ) which are not in supp(T ), by again iteratively transforming f and φ such that we have after each step, and f = φ(NF(T )(b)) supp(w-jet(f, dw + ji)) = supp(w-jet(T, dw + ji)), where 1 ≤ j1 < ··· < ji, after step i. We may thus assume that (y) = cyy, (x) = cxx and φw ji φw ji where cx, cy ∈ R∗. Since the determinacy k of f is finite, terms of degree greater than k may be discarded after each step. Once f is transformed to a germ with non-degenerate Newton principal part and weighted determinacy bound k(cid:48), the elimination process above Γ(T ) can be stopped once dw + ji > k(cid:48). Hence a normal form equation of f can be determined in finitely many steps. To summarize, the general classification algorithm involves the following main steps: 1Thus it does not make sense to include the (cid:101)Yr case in the Figures 1 to 4. THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 9 I. Determine the complex type and the corresponding Newton polygon Γ(T ). II. Eliminate all monomials in supp(f ) \ supp(T ) underneath or on Γ(T ). III. Eliminate all monomials in supp(f ) \ supp(T ) above Γ(T ). IV. Read off the real subtypes and the corresponding normal form equations. Step (I) is straightforward using classify2.lib. We now discuss Steps (II) to (IV) in detail: Step (II): Elimination of monomials underneath and on the Newton Polygon. We first eliminate the monomials in jet(f, d), which are not in jet(T, d), where d is the maximum filtration of f with respect to the standard degree. This can be done by a linear transformation over R, since these terms can only be created from the monomials in φ0(x) and φ0(y). Removing these terms amounts to transforming φ such that φ0(x) = cxx, cx ∈ R, (1) if jet(T, d) = xn for some n ∈ N, or such that φ0(x) = cxx, cx ∈ R, and φ0(y) = cyy, cy ∈ R, (2) if jet(T, d) (cid:54)= xn for any n ∈ N. For the exceptional cases and the parabolic cases J10 and J10+k, this can be done by factorizing jet(f, d) over Q using (Marais and Steenpass, 2015a, Proposition 8) and (Marais and Steenpass, 2015a, Lemma 9). Similarly, can we use Lemma 2.23 for the case X9+k. See Algorithms 1 and 2, respectively, which implement the required transformations. with maximum filtration d Algorithm 1 Input: f ∈ Q[x, y] of real type T , where T is exceptional, or parabolic of type J10 or J10+k, Output: h ∈ Q[x, y] with f ∼ h and supp(jet(h, d)) = supp(T, d) 1: g := jet(f, d) 2: factorize g over Q as cf α 2 , with f1 and f2 linear and non-associated, c ∈ Q, and 1 f β α > β ≥ 0 3: if β > 0 then apply f1 (cid:55)→ x, f2 (cid:55)→ y to f 4: 5: else if f1 (cid:54)= c(cid:48)y, c(cid:48) ∈ Q then apply f1 (cid:55)→ x, y (cid:55)→ y to f 6: 7: else apply f1 (cid:55)→ x, x (cid:55)→ y to f 8: 9: return f In the cases X9 and Yr,s, in general, a real field extension is required for this step. This leads to an implementational problem for the subsequent steps of the algorithm, since if we represent an algebraic number field as Q[z]/(cid:104)m(cid:105), with m ∈ Q[z] irreducible, we have to determine which root of m the generator z corresponds to. We discuss how to handle this problem in the cases X9 and Yr,s in Sections 5 and 6, respectively. Next we eliminate the remaining monomials in supp(f )\ supp(T ) underneath or on Γ(T ). We consider the cases where Γ(T ) has exactly one face and where Γ(T ) has two faces separately. The Newton polygon has only one face. Note that in the case X9, after the linear trans- formation, there are no monomials in supp(f ) \ supp(T ) underneath or on Γ(T ). In the remaining cases, we have jet(T, d) = m0, where m0 is a monomial, and d is the maximum filtration of f with respect to the standard degree as before. Then jet(f, d) = cm0, with THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 10 Algorithm 2 Input: f ∈ Q[x, y] of real type T = X9+k Output: h ∈ Q[x, y] with f ∼ h and supp(jet(h, 4)) = supp(T, 4) 1: g := jet(f, 4) 2: factorize g over Q as cf 2 3: if f1 (cid:54)= c(cid:48)y, c(cid:48) ∈ Q then 4: 5: else 6: 7: write f = a0x4 + a1x3y + a2x2y2 + R, a0, a1 ∈ Q, a2 ∈ Q∗, and R ∈ E5 8: apply y (cid:55)→ y − a1 9: return f apply f1 (cid:55)→ x, y (cid:55)→ y to f apply f1 (cid:55)→ x, x (cid:55)→ y to f 1 f2 with f1 linear, f2 quadratic, and c ∈ Q x, x (cid:55)→ x to f 2a2 0 (cid:54)= c ∈ Q. By considering 1 remove the required terms of 1 c f, we may assume that c = 1. Indeed, the transformations that c f also remove the same terms of f. We first eliminate the monomials in f strictly underneath Γ(T ) and then the monomials on Γ(T ), which are not in supp(T ). Note that, to achieve this goal, also monomials above the Newton polygon will have to be eliminated. We inductively consider jets of f of increasing standard degree, starting at degree d + 1, until all monomials strictly underneath Γ(T ) have been removed. For each degree j, the jet is transformed such that supp(jet(f, j)) = supp(T, j). While jet(T, j) = m0, all terms of degree j in f have been created through a term in homogeneous non-linear polynomials v1 or v2 in φ(x) = linear terms + v1 + terms of higher degree than v1 φ(y) = linear terms + v2 + terms of higher degree than v2. Considering Equations (1) and (2), and taking into account that by c = 1, we may assume cx = 1 and cx = cy = 1, respectively, such a transformation maps m0 = xαyβ in NF(T )(b) to xαyβ + αxα−1yβv1 + βxαyβ−1v2 + terms of higher degree. Since m0 has degree less than j and is the only term in NF(T )(b) with degree less than or equal to j, the terms of degree j in f can be written as v1 ∂m0 ∂x + v2 ∂m0 ∂y , where deg(v1) > 1 and deg(v2) > 1. We can eliminate these terms by the transformation x (cid:55)→ x − v1, y (cid:55)→ y − v2. This transformation possibly creates terms of degree greater than j, but does not change jet(f, j − 1). As soon as the j-jet of NF(T )(b) contains more than one term, the situation becomes more complicated. For example, if j is the degree of a monomial in supp(T ), then some of the monomials of degree j in supp(f )\ supp(T ) may result from the linear terms of φ(x) and φ(y). However, all monomials in f underneath Γ(T ) have already been eliminated before we reach such a case. Indeed, taking Equations (1) and (2) into account, linear terms in φ(x) will not create any additional monomials and φ(y) will only create additional monomials above Γ(T ). Moreover, a case by case analysis shows that terms in f underneath Γ(T ) also cannot result from higher order terms in φ(x) and φ(y): E14, J10 (see Figures 1, 2, and 4). • If m0 = xd, then the real main type of f is one of the following: W12, W13, E12, E13, Considering the normal forms of W12 and W13, we have jet(T, 4) = x4 and jet(T, 5) (cid:54)= x4. Suppose jet(f, 4) = x4. If f is of type W12, then f cannot have any monomials THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 11 underneath Γ(T ). Let f be of type W13 and suppose y5 ∈ supp(f ). Then it follows from Theorem 2.18 that f is of type W12 which leads to a contradiction. Hence y5 (cid:54)∈ supp(f ). If f is of type E12 or E13, then jet(T, 5) = x3 and jet(T, 6) (cid:54)= x3. After applying the algorithm described above, we may assume that jet(f, 5) = x3. If y6 ∈ supp(f ), then f is of type J10. If y7 ∈ supp(f ), then f is not of type E13, but rather of type E12. Hence if f is of type E12 or E13, then f has no monomials underneath Γ(T ). Similarly, it can be shown that if f is of type E14, then f has no monomials underneath Γ(T ) after f has been transformed such that jet(f, 6) = x3. If f has real main type J10 and jet(f, 3) = x3, the only monomials that may occur underneath Γ(T ) are xy3, y4 and y5. Suppose that xy3 ∈ supp(f ). Since, by Equation (1), this term cannot result from linear terms of φ(x) or φ(y), and since jet(f, 3) = x3, it follows that xy3 ∈ supp(v1 ∂x ) = supp(3v1x2). Since x2 (cid:45) xy3, this leads to a contradiction. If y4 ∈ supp(f ), then f is of real main type E6 by Theorem 2.18, which again gives a contradiction. Similarly, if y5 ∈ supp(f ), then f is of real main type E8. • If m0 (cid:54)= xd, then the real main type of f is one of the following: Z11, Z12, Z13 (see ∂x3 Figure 3). We show that, in these cases, there are no monomials underneath Γ(T ) after the linear transformation that transforms f such that jet(f, d) = m0. Suppose f is of one of the main types under consideration and that jet(f, d) = m0. If f is of type Z11, there are no monomials of degree greater than d underneath Γ(T ). Suppose f is of real main type Z12 or Z13. If y5 or xy4 are elements of supp(f ), then, by Theorem 2.18, f is of real main type W12 or Z12, respectively, which leads to a contradiction. Therefore, if m0 (cid:54)= xd, then after the linear transformation given in Algorithm 1, f has no monomials underneath Γ(T ). If m0 = xd and j is such that jet(T, j) = m0, then one can write jet(f, j) = m0 + v1 ∂m0 ∂x + 0 ∂m0 ∂y . Hence, iterative application of Algorithm 4 with t a term of f of standard degree j removes all monomials underneath Γ(T ). Note that in the algorithm we have mx = m0, so we make only use of the case in lines 8 - 9. The other case will be used later to eliminate monomials above the diagonal in a similar manner. Moreover, note that Algorithm 4, as formulated, also works for the polynomial t = jet(f, j) − jet(f, j − 1). that φw−1(x) = 0 and φw−1(y) = 0. After eliminating all the monomials in supp(f ) strictly underneath Γ(T ), we may assume Next we eliminate the remaining monomials in supp(f ) \ supp(T ) on Γ(T ), see line 10 in Algorithm 5. The only case where such monomials can occur is the J10 case. In this case, these monomials can theoretically be eliminated by a weighted homogeneous transformation. However, similar to the linear transformation discussed above, this transformation creates implementational difficulties, since it may require real field extensions. We will discuss how to handle this problem in Section 6. The Newton polygon has two faces. In this case, all the normal forms we have to consider are of the form xn + x2y2 + aym with m, n ∈ N, see Figure 5. After a linear transformation (see Algorithm 1 for the case J10+k, Algorithm 2 for the case X9+k, and Section 6 for the case Yr,s), we may assume that supp(jet(f, d)) = supp(T, d), where d is the maximum filtration of f with respect to the standard grading. We first remove all monomials m with max{wi-deg(m) i} ≤ dw which are not in supp(T ), see line 12 in Algorithm 5. The only cases in which such monomials may occur are the J10+k cases. The monomials y4, y5 and xy3 do not occur in f if it is of type J10+k, k > 0, since presence of either of them leads to a different type. Using Algorithm 3, we transform f by a weighted linear transformation such that THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 12 supp(w-jet(f, 6)) = supp(w-jet(T, 6)) where w = (2, 1), which means that f is in the desired form. Algorithm 3 Input: f of type J10+k with supp((2, 1)-jet(f, 5)) = ∅ Output: h ∈ Q[x, y] with f ∼ h and with supp((2, 1)-jet(h, 6)) = supp((2, 1)-jet(T, 6)) 1: write f as f = x3 + cx2y2 + dxy4 + ey6 + R with c, d, e ∈ Q and R ∈ E(2,1) 2: let q be the double root of x3 + cx2 + dx + e 3: apply x (cid:55)→ x + qy2, y (cid:55)→ y to f 4: return f 7 Similar to the case where Γ(T ) has one face, we may assume that coeff(f, x2y2) = 1 by considering 1 c f, where c = coeff(f, x2y2). We now, again, consider jets of f of increasing standard degree, starting at degree d + 1. However, in this case, we only eliminate those monomials not lying in supp(T ) that are strictly underneath or on Γ(T ). Suppose we are considering the standard weighted j-jet. We first eliminate terms of the form c1xyj−1 and c2yxj−1 which lie underneath or on Γ(T ), using permissible x- and y-segments, that is, by applying the transformation or x (cid:55)→ x, x (cid:55)→ x + yj−3, c1 2 y (cid:55)→ y + y (cid:55)→ y xj−3, c2 2 respectively. This process is described in Algorithm 4 with t = c1xyj−1 or t = c2yxj−1, respectively. Now, if n, m (cid:54)= j, the monomials xj and yj do not occur in supp(f ): Otherwise, by Lemma 2.20 and Theorem 2.18, f would be of a different type than assumed. No other monomials can occur in supp(f ) \ supp(T ) underneath or on Γ(T ). Step (III): Elimination of monomials above the Newton polygon. In the case where Γ(T ) has two faces, Theorem 2.20 allows us to delete all monomials above Γ(T ). We now consider the case where Γ(T ) has one face. Here we use the algorithmic method that was use in the proof of Theorem 2.18 to eliminate the monomials above Γ(T ) in Jac(f ). Let {m1} be the Arnold system for the singularity under consideration. Let d(cid:48) w be the degree of m1. Again we iteratively consider the w-degree j part of f in Jac(f ), for increasing j with dw < j ≤ d(cid:48) w. This polynomial can be written as ∂f0 ∂x v1 + ∂f0 ∂y v2, where f0 is the quasihomogeneous part of f and v1, v2 ∈ R[x, y]. After applying the trans- formation x (cid:55)→ x − v1, y (cid:55)→ y − v2 to f, the w-degree j part has no terms of degree j in Jac(f ). Note that for dw < j < d(cid:48) w, all terms of w-degree j are in Jac(f ). In all the cases we consider, each weighted diagonal above Γ(T ) of w-degree less than or equal to d(cid:48) w will only contain one monomial in Jac(f ). Hence, either v1 or v2 are zero. We therefore can apply Algorithm 4 to remove this monomial. The remaining w-degree d(cid:48) w part of f can be written as ∂f0 ∂x ∂f0 ∂y v1 + v2 + cm1, where v1, v2 ∈ R[x, y] and c ∈ R. Applying the transformation x (cid:55)→ x − v1, y (cid:55)→ y − v2 results in transforming the part of f of weighted degree d(cid:48) w to cm1. A case by case analysis shows that we can always choose v2 = 0. See line 23 in Algorithm 5. Since all terms of w-degree greater than d(cid:48) w are above the weighted determinacy, they can be deleted. THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 13 Algorithm 4 Input: f ∈ K[x, y] of corank 2, modality 1 and type T , and a term t /∈ supp(T ) of f such that either (a) t ∈ Jac(f ), supp(T, j) = 1, supp(jet(f, j − 1)) = supp(T, j − 1) where j = deg(t), (b) t is in a permissible chain, and supp(jet(f, d)) = supp(T, d) where d is the maximum filtration of f, or supp(w(T )-jet(f, j − 1)) = (c) t ∈ Jac(f ), supp(w(T )-jet(T, j − 1)), and j is smaller than the w(T )-degree of the unique ele- ment in Arnold's system. := w(T )-deg(t) > d(T ), j (u1, u2) := ((1, 1), (1, 1)) Output: g ∈ K[x, y] such that f ∼ g and supp(g) ∩ supp(t) = ∅ 1: let f0 be piecewise quasihomogeneous part of f of degree d(T ) w.r.t. w(T ) 2: if t ∈ Jac(f ) then 3: 4: else 5: 6: let mx be the term of ∂f0 7: let my be the term of ∂f0 8: if mxt then 9: ∂x of lowest u2-degree ∂y of lowest u1-degree (u1, u2) := w(T ) α : K[x, y] → K[x, y] x (cid:55)→ x − t/mx y (cid:55)→ y 10: else 11: 12: g := α(f ) 13: return g α : K[x, y] → K[x, y] x (cid:55)→ x y (cid:55)→ y − t/my Remark 3.3. The question may be asked why we change the strategy, in the sense that we iteratively consider the terms of increasing standard degree in the process of removing all terms underneath Γ(T ) and the terms of increasing w-degree above Γ(T ). The reason is that canceling terms underneath Γ(T ) can only be done using transformations of negative filtration with respect to w, since the terms underneath Γ(T ) have smaller w-degree than the terms on Γ(T ). This means that in the process of deleting monomials of a given w- degree, monomials of lower and higher w-degree may be created. Since the filtration of a transformation with respect to the standard degree cannot be negative, this is not the case when canceling terms by increasing standard degree via the above methods. However, terms above Γ(T ) are canceled using transformations with non-negative filtration with respect to w via the above method and therefore we consider terms by increasing w-degree in this case, which leads to a much simpler process. Step (IV): Scaling and reading off the desired information. Lastly we scale f such that the coefficients of the non-moduli terms coincide with those in one of the real subtypes of the real main type of f, read off the needed information and determine all normal form equations in the equivalence class of f, using Marais and Steenpass (2015b). THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 14 Algorithm 5 Classification and Determination of Parameter: General Structure Input: f ∈ m3 ⊂ Q[x, y], a germ of modality 1 and corank 2 Output: the real singularity types and normal forms of f, as well as the corresponding normal form equations in the right equivalence class of f with the respective parameter given as the unique root of its minimal polynomial over Q in a specified interval I. Determine the complex type: 1: compute the complex type T of f using classify() from (Boehm et al., 2017a) II. Eliminate the monomials in supp(f ) underneath or on Γ(T ): 2: let d be the maximum filtration of f w.r.t. the standard grading 3: apply a linear coordinate change to f such that supp(jet(f, d)) = supp(T, d) 4: w := (w1, . . . , wn) := w(T ) 5: dw := d(T ) 6: if n = 1 then 7: 8: 9: 10: for j = d + 1, d + 2, . . . while supp(f ) ∩ (R2 \Γ+(T )) (cid:54)= ∅ do apply a weighted homogeneous transformation to f such that replace f by the output of Algorithm 4 applied to f and t while f has a term t of standard degree j do supp(w-jet(f, dw)) = supp(w-jet(T, dw)) 11: else 12: 13: 14: 15: apply a w2-weighted homogeneous transformation to f such that supp(w2-jet(f, dw)) = supp(w2-jet(T, dw)) for j = d + 1, d + 2, . . . while supp(f ) ∩ (R2 \Γ+(T )) (cid:54)= ∅ do while f has a term t of the form bxyj−1 or byxj−1 underneath or on Γ(T ) do replace f by the output of Algorithm 4 applied to f and t III. Eliminate the monomials above Γ(T ) which are not in supp(T ): 16: if n = 1 then 17: 18: 19: 20: 21: 22: 23: 24: 25: 26: else 27: else f := w-jet(f, dw) if w-jet(Γ(T ), dw) is non-degenerate then let d(cid:48) for j = dw + 1, . . . , d(cid:48) w be the w-degree of the unique element in Arnold's system while f has a term t ∈ Jac(g) of w-degree j do w do replace f by the output of Algorithm 4 applied to f and t f := w-jet(f, d(cid:48) w) modulo Jac(f ), write the sum of the terms of f above Γ(T ) in Arnold's system replace f by the output of Algorithm 14 applied to f IV. Scale and read off the desired information: 28: apply a transformation of the form x (cid:55)→ ±x, y (cid:55)→ ±y to f such that the signs of the coefficients of the non-moduli monomials match the signs of one possible real normal form F in Table 1 29: read off the corresponding type of F 30: apply a transformation of the form x (cid:55)→ λ1x, y (cid:55)→ λ2y with λ1, λ2 > 0 in a real algebraic 31: determine the minimal polynomial over Q for the parameter in F 32: using Marais and Steenpass (2015b), determine all possible types and corresponding extension of Q to transform the non-moduli terms to the terms of F parameters 33: return all types, normal forms, and parameters THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 15 4. Implementing the Exceptional Cases For the exceptional singularities (see Figures 1, 2, and 3), the general algorithm can be implemented in a straightforward manner. As an example, we consider singularities of real main type E14, and give all details in Algorithm 6. In this case, we have w = (8, 3), d = 3, dw = 24, d(cid:48) w = 26, m0 = x3, m1 = xy6 and f0 = x3 + y8 in the general algorithm. Algorithm 6 corresponds directly to our implementation of the respective case in Singular. Note that, although the scaling step in line 30 of Algorithm 5 has not been implemented explicitly, it has been taken into account in Algorithm 6 when determining the minimal polynomial. We use the following notation. Notation 4.1. Suppose p ∈ Q[z] is a univariate polynomial over the rational numbers, and I ⊂ R is an interval such that there is exactly one a ∈ I with p(a) = 0. We then denote the monic irreducible divisor of p with root a by mI (p). When denoting real subcases, we identify + with +1 and − with −1. For example, we use the notation Esign(b) 14 for E+ 14 if b > 0, and E− 14 if b < 0. W12 W13 Figure 1. Exceptional Singularities of type W E12 E13 E14 Figure 2. Exceptional Singularities of type E 01234567123456701234567123456702468246802468123450246812345 THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 16 Algorithm 6 Algorithm for the case E14 Input: f ∈ m3 ⊂ Q[x, y] of complex singularity type T = E14 Output: all normal form equations in the right equivalence class of f, each specified as a tuple of the real singularity type, the normal form, the minimal polynomial of the parameter, and an interval such that the parameter is the unique root of the minimal polynomial in this interval II. Eliminate the monomials in supp(f ) underneath or on Γ(T ): Apply Algorithm 1: 1, where h1 is homogeneous of degree one and c ∈ Q∗ 1: h := jet(f, 3) 2: factorize h as ch3 3: if h (cid:54)= c(cid:48)y, c(cid:48) ∈ Q then apply h1 (cid:55)→ x, y (cid:55)→ y to f 4: 5: else apply h1 (cid:55)→ x, x (cid:55)→ y to f 6: 7: g := 1 c f Consider the terms of g of standard degree j = 4 and apply Algorithm 4: 8: h1 := jet(g,4)−x3 9: apply x (cid:55)→ x − h1, y (cid:55)→ y to g 3x2 Consider the terms of g of standard degree j = 5 and apply Algorithm 4: 10: h2 := jet(g,5)−x3 11: apply x (cid:55)→ x − h2, y (cid:55)→ y to g 3x2 III. Eliminate the monomials above Γ(T ) which are not in supp(T ): 12: w := (8, 3) Consider the terms of g of w-degree j = 25 and apply Algorithm 4: 1 := w-jet(g,25)−w-jet(g,24) 13: h(cid:48) 14: x (cid:55)→ x − h(cid:48) 15: f := cg 3x2 1, y (cid:55)→ y IV. Read off the desired information: apply x (cid:55)→ −x, y (cid:55)→ y to f 16: if c < 0 then 17: 18: write f as f = cx3 + by8 + R with c ∈ Q 19: t := coeff(f, xy6) 20: p := z12 − c−4b−9t12 ∈ Q[z] 21: if t > 0 then 22: 23: else 24: return (Esign(b) 14 return (Esign(b) 14 >0, b ∈ Q∗, and R ∈ E(8,3) 25 , x3 + sign(b) · y8 + axy6, m(0,∞)(p), (0,∞)) , x3 + sign(b) · y8 + axy6, m(−∞,0](p), (−∞, 0]) THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 17 Z11 Z12 Z13 Figure 3. Exceptional Singularities of type Z 5. Parabolic Singularities In this section, we give details on the application of the general algorithm (Algorithm 5) for the parabolic singularities. See Figure 4 for these cases. 5.1. The J10 case. In this case we have w = (6, 3) and dw = d(cid:48) w = 18. As discussed in Section 3, lines 7 - 9 in the general algorithm are redundant. Moreover, by dw = d(cid:48) w, Step (III) is redundant. Unfortunately we may need a real field extension in line 10 of the general algorithm. This leads to an implementational problem for the subsequent steps of the algorithm, since if we represent an algebraic number field as Q[z]/m, with m irreducible, we have to determine which root of m the generator z corresponds to. In this section, we will work around this problem by presenting a method to read off the required information without explicitly doing the required weighted linear transformation, see Algorithm 7. In Marais and Steenpass (2015b) it has been shown that in the case of real main type J10, 10 or 10 and one is of the equivalence class of f either contains exactly one normal form equation of type J + it contains exactly three normal form equations, two of which are of type J + X9 J10 Figure 4. Parabolic Singularities 012345612345601234567123456701234567123456702468123450246812345 THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 18 10. This shows, in particular, that the real subtype J− type J− algorithmic approach to the classification also reconfirms these findings. Let f ∈ m3 be a polynomial of complex type J10 such that f ∈ E(6,3) R ∈ E(6,3) f = cx3 + bx2y2 + dxy4 + ey6 + R, form 18 . 19 Since f is weighted 18-determined, f ∼ f − R. Hence, we may assume that 10 is redundant. The following , that is, it is of the f = cx3 + bx2y2 + dxy4 + ey6. If c < 0, then we apply the transformation x (cid:55)→ −x, y (cid:55)→ y. Since f is weighted homogeneous, a rescaling of the coordinates achieves that c = 1. By applying x (cid:55)→ x − b 3 f is transformed to a polynomial of the form y2, y (cid:55)→ y, Since, on the other hand, for a(cid:48), d(cid:48) ∈ R we have f = x3 + dxy4 + ey6. x3 + a(cid:48)x2y2 ± d(cid:48)xy4 ∼ x3 + ax2y2 ± xy4, a = a(cid:48)(cid:112)d(cid:48) , via the invertible transformation α(cid:48) : x (cid:55)→ x, y (cid:55)→ 1 4(cid:112)d(cid:48) y (3) (4) (as applied in Step (IV) of the general algorithm), the problem is reduced to finding a transformation α such that f = x3 + dxy4 + ey6 is mapped to α(f ) = x3 + a(cid:48)x2y2 ± d(cid:48)xy4, (5) The composition of α with the transformation in (3) and the scaling of the coordinates, as described above, is then a suitable transformation to be applied in line 10 of the general algorithm. a(cid:48), d(cid:48) ∈ R . In the following, we describe a method to determine all polynomials as in (5) in the equivalence class of f. Since α has to be weighted homogeneous, it is of the form (6) with appropriate t, c ∈ R, t (cid:54)= 0. We now determine all possible values for t and c. Applying a transformation as in (6) to f, we obtain α : x (cid:55)→ x + cy2, y (cid:55)→ ty α(f ) = x3 + 3cx2y2 + (3c2 + t4d)xy4 + (c3 + t4dc + et6)y6. (7) Taking c(cid:48) = c t2 , we can rewrite (7) as α(f ) = x3 + 3t2c(cid:48)x2y2 + t4(3c(cid:48)2 + d)xy4 + t6(c(cid:48)3 + dc(cid:48) + e)y6. (8) Clearly, for a fixed value t (cid:54)= 0, c(cid:48) is any real root of k := s3 + ds + e ∈ Q[s]. Due to the scaling of y by α(cid:48), we may assume that t = 1 and c = c(cid:48). To determine all possible normal form equations in the equivalence class of f, we have to consider the following cases: (i) k(s) has one real root; (ii) k(s) has three real roots. THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 19 Algorithm 7 Algorithm for the case J10 Input: f ∈ m3 ⊂ Q[x, y] of complex singularity type J10 Output: all normal form equations in the right equivalence class of f, each specified as a tuple of the real singularity type, the normal form, the minimal polynomial of the parameter, and an interval such that the parameter is the unique root of the minimal polynomial in this interval 1 apply x (cid:55)→ y, y (cid:55)→ x to f 1 with b ∈ Q, b > 0, and g1 homogeneous of degree 1 1: f := jet(f, 6) 2: if coeff(f, x3) = 0 then 3: 4: h := jet(f, 3) 5: factorize h as bg3 6: apply x (cid:55)→ g1, y (cid:55)→ y to f 7: f := coeff(f,x3) f 8: write f as f = x3 + ax2y2 + dxy4 + ey6 with a, d, e ∈ Q 9: apply x (cid:55)→ x − a 10: write f as f = x3 + dxy4 + ey6 with d, e ∈ Q 11: p(σ) := σ(4d3 + 27e2)z6 + (−36d3 − 243e2)z4 + σ(81d3 + 729e2)z2 − 729e2 ∈ Q[z] 12: k := s3 + ds + e ∈ Q[s] 13: if k has exactly one real root then 14: 15: 3 y2, y (cid:55)→ y to f if e < 0 then return (J + 10, x3 + ax2y2 + xy4, m(0,∞)(p+), (0,∞)) 10, x3 + ax2y2 + xy4, m(−∞,0)(p+), (−∞, 0)) if e > 0 then return (J + if e = 0 then 16: 17: 18: 19: 20: else 21: 22: 23: 24: 25: 26: 27: 28: 29: 30: 31: 32: 33: 34: 35: 36: 37: 38: 39: 40: return (J + 10, x3 + ax2y2 + xy4, z, [0, 0]) if k(0) = 0 then C1 := (J + C2 := (J− C3 := (J + 10, x3 + ax2y2 + xy4, z2 − 9 10, x3 + ax2y2 − xy4, z, [0, 0]) 10, x3 + ax2y2 + xy4, z2 − 9 2 , (−∞, 0)) 2 , (0,∞)) else ε := 1 do replace ε by a rational number in the interval (0, ε 2 ) let (z1, . . . , z4) ∈ Q4 be approximations of the four distinct real roots of p+ in increasing order with error smaller than ε (I1, . . . , I4) := ((z1 − ε, z1 + ε), . . . , (z4 − ε, z4 + ε)) while p+ has more than one real root in any of the intervals I1, . . . , I4 if e < 0 then C1 := (J− C2 := (J + C3 := (J + C1 := (J− C2 := (J + C3 := (J + 10, x3 + ax2y2 − xy4, m(−∞,0)(p−), (−∞, 0)) 10, x3 + ax2y2 + xy4, mI1(p+), I1) 10, x3 + ax2y2 + xy4, mI3(p+), I3) 10, x3 + ax2y2 − xy4, m(0,∞)(p−), (0,∞)) 10, x3 + ax2y2 + xy4, mI2(p+), I2) 10, x3 + ax2y2 + xy4, mI4(p+), I4) else return C1, C2, C3 hj(w) = w3 + 2dw2(9 − 3a2 that is, ±aj, j = 1, 2, 3, are the roots of j ) + d2w(9 − 3a2 j )2 − e2(9 − 3a2 j )3 ∈ Q[w], h(z) := (4d3 + 27e2)z6 + (−36d3 − 243e2)z4 + (81d3 + 729e2)z2 − 729e2 ∈ Q[z]. By e = −c1c2c3 = 0 and c2, c3 (cid:54)= 0, it follows that e = 0 ⇔ c1 = 0 ⇔ a = 0. THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 20 Denote by c1, c2 and c3 the complex roots of k(s). Then and − c1c2c3 = e. c1 + c2 + c3 = 0, c1c2 + c1c3 + c2c3 = d, (9) (i) Let c = c1 be the real root and let c2 and c3 be the complex conjugate roots of k(s). 1. Since the product of two non-zero complex conjugates It follows from (9) that c2c3 = d + c2 is positive, d + c2 1 > 0. Hence, considering (8), f is of type 10. Since the real root c1 is uniquely determined, the transformation α as considered above J + and the real subtype are also uniquely determined. In particular, a = 3c1√ is uniquely 3c2 1+d determined. Define a1 := a, a2 := 3c2√ j (9 − 3a2 . Note that da2 j ). j = c2 Since c1, c2, and c3 are the roots of k(s), they are also roots of 1 > 0, which implies d + 3c2 , and a3 := 3c3√ 3+d 2+d 3c2 3c2 k(s) := −k(s) · k(−s) = s6 + 2ds4 + d2s2 − e2. Multiplying k(s) with (9 − 3a2 j )3, we see that da2 j is a root of 2 2 = 9c2 We show that this is the case if and only if a2 ∈ R or a3 ∈ R. Assume that, without loss 2 for some of generality, a2 is real. Then a2 λ ∈ R. Therefore d = λ(cid:48)c2 2 ∈ R. Because c2 ∈ C\ R, it follows that c2 = γi for some γ ∈ R, hence c3 = −γi. This implies that a3 is real, since d = c2c3 = γ2. By (9) we obtain c1 = 0. For the converse, note that if c1 = 0, then c2 = γi and c3 = −γi, hence a2 and a3 are real. We now consider the case where e (cid:54)= 0. Using the fact that c2c3 is positive and e = −c1c2c3, it follows that 2 for some λ(cid:48) ∈ R. Since d ∈ R, it follows that c2 2+d is real, which implies that 3c2 2 + d = λc2 3c2 − sign(e) = sign(c1) = sign(a). Hence, a minimal polynomial for a is obtained as the monic irreducible factor of h(x) with a root in (0,∞) in case e < 0, or in (−∞, 0) in case e > 0. Moreover, a is the unique root of this factor in this interval. aj := j = 1, 2, 3. In case c2 = 0, we have c3 = −c1 = (ii) In the case where k(s) has three real roots, they must be pairwise different, otherwise k(s) and its derivative k(cid:48)(s) have a common root, which implies that f is degenerate, since the coefficient of xy4 in (8) vanishes for c a double root of k(s). Without loss of generality, we can assume that c1 < c2 < c3. Therefore a can attain the three different values 3cj(cid:113)d + 3c2 j , √−d with d < 0, which implies that a1 = − 3√ , a2 = 0, and a3 = 3√ i + d, we obtain that c1 and c3 correspond to type J + Next we consider the case c2 (cid:54)= 0. Since c1 + c2 + c3 = 0, not all three roots have the same sign. Suppose − sign(c1) = sign(c2) = sign(c3). Note that c1 = c2 + c3 = c2 + c3, that 1 > c1c3 is, c1 > c2 and c1 > c3. Then 3c2 1 + d = 3c2 1 > c1c2, and, moreover, c2 and c2 1 > 0, c1c2 < 0, c1c3 < 0, and c2c3 > 0. Hence, c1 corresponds to type J + 10. Furthermore j + c1(c2 + c3) + c2c3 = 3c2 3c2 1 + c1c2 + c1c3 + c2c3 > 0, since c2 j − (c2 + c3)(c2 + c3) + c2c3 = 3c2 10 and c2 corresponds to J− 10. . Considering the sign of 3c2 2 + c2c3 + c2 j + d = 3c2 j − (c2 3), 2 2 THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 21 for j = 2, 3. Since c2 < c3, we obtain that c2 corresponds to type J− 10 and c3 to J + 10. The remaining case − sign(c1) = − sign(c2) = sign(c3) can be treated in a similar manner, leading again to types J + the case J− We now determine the corresponding values of the moduli parameter. We first consider 10 corresponding to c1, c2 and c3. 10 and J + 10, J− 2(9 + 3(a2)2) and − d(iaj)2 = c2 j (9 + 3(iaj)2), j = 1, 3. 10. Note that −d(a2)2 = c2 By multiplying k(s) with 9 + 3(iaj)2, for j = 1, 3, and with 9 + 3a2 the roots of j, for j = 2, it follows that h−(z) := (−4d3 − 27e2)z6 + (−36d3 − 243e2)z4 + (−81d3 − 729e2)z2 − 729e2 ∈ Q[z] are ±ia1, ±a2, and ±ia3. Since sign(c1) = − sign(c3), we have sign(a2) = sign(c2) = sign(e), which determines an appropriate interval containing a2. For the parameters which correspond to J + 10, we can construct, in a similar manner, the polynomial h+(z) := (4d3 + 27e2)z6 + (−36d3 − 243e2)z4 + (81d3 + 729e2)z2 − 729e2 ∈ Q[z] with roots ±a1, ±ia2, and ±a3. Since, in these cases, either c2 < c1 or c2 < c3, it follows that d < 0. An easy calculation shows that c1 < c3 if and only if a3 < a1. If e < 0, that is, c2 < 0, we have that c1 + c2 = c1 + c2 = c3 and, hence, that a3 < a1. Therefore, a1 is the smallest negative real root and a3 is the smallest positive real root of h+(z). Similarly, if e > 0, then a1 < a3, hence a1 is the largest negative real root, and a3 the largest positive real root of h+(z). Remark 5.1. With regard to the implementation, we use the Singular library solve.lib (Wenk and Pohl, 1999) and Sturm chains, as implemented in the library rootsur.lib (Tobis, 2012), to determine intervals with rational boundaries containing the roots. 5.2. The X9 case. According to Theorem 29 in Marais and Steenpass (2015b), the real right equivalence class of a singularity of real main type X9 always contains exactly two normal form equations from Arnold's list, of possibly different real subtypes. There are four different possible cases for a given singularity of real main type X9: rameter a with a > −2. (A) The singularity is right equivalent to NF(cid:0)X ++ (B) The singularity is right equivalent to NF(cid:0)X−− (C) The singularity is right equivalent to both NF(cid:0)X +− (D) The singularity is right equivalent to NF(cid:0)X ++ (cid:1)(a) for some a > 2. parameter and to NF(cid:0)X−− rameter a with a < 2. unique value a ∈ R of the parameter. (cid:1)(a) for two different values of the pa- (cid:1)(a) for two different values of the pa- (cid:1)(a) for some (cid:1)(a) and NF(cid:0)X−+ (cid:1)(a) for some value a < −2 of the 9 9 9 9 9 9 In the following, we discuss how Algorithm 8 determines in which of these cases a given singularity falls, and which are the corresponding values of the parameter. We do not strictly follow Algorithm 5 because line 3 would, in general, require working over algebraic extensions. Due to this difficulty, especially Step (IV) needs a specific approach. First of all, the determinacy of a given polynomial f ∈ Q[x, y] of complex singularity type X9 is 4, thus it suffices to consider jet(f, 4). Some calculations in linear algebra show that we can assume the coefficient of x4 to be non-zero, by applying a linear coordinate transformation as in line 2 of Algorithm 8 if necessary. For convenience, we can then get rid of the term x3y by the transformation given in line 3 such that f is of the form f = bx4 + cx2y2 + dxy3 + ey4 with b, c, d, e ∈ Q and b (cid:54)= 0. THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 22 Algorithm 8 Algorithm for the case X9 Input: f ∈ m3 ⊂ Q[x, y] of complex singularity type T = X9 Output: all normal form equations in the right equivalence class of f, each specified as a tuple of the real singularity type, the normal form, the minimal polynomial of the parameter, and an interval such that the parameter is the unique root of the minimal polynomial in this interval 1: f := jet(f, 4) 2: make sure that the coefficient of x4 in f is non-zero by applying either (x, y) (cid:55)→ (y, x), (x, y) (cid:55)→ (x, x + y), (x, y) (cid:55)→ (x, 2x + y), or (x, y) (cid:55)→ (x, 3x + y) to f if necessary coeff(f,x4) y, y (cid:55)→ y to f 3: apply x (cid:55)→ x − coeff(f,x3y) 4: write f as f = bx4 + cx2y2 + dxy3 + ey4 with b, c, d, e ∈ Q and b (cid:54)= 0 5: p(σ) := (cid:0)−256b3e3 + 128b2c2e2 − 144b2cd2e + 27b2d4 − 16bc4e + 4bc3d2(cid:1) · z6 + σ(cid:0)18432b3e3 + 11520b2c2e2 − 5184b2cd2e + 972b2d4 + 144bc3d2 + 16c6(cid:1) · z4 +(cid:0)−331776b3e3 − 62208b2cd2e + 11664b2d4 − 11520bc4e + 1728bc3d2 − 128c6(cid:1) · z2 + σ(cid:0)331776b2c2e2 − 248832b2cd2e + 46656b2d4 − 18432bc4e + 6912bc3d2 + 256c6(cid:1) ∈ Q[z] else if b > 0 then else C1 :=(cid:0)X ++ C2 :=(cid:0)X ++ 9 9 else C1 :=(cid:0)X−− C2 :=(cid:0)X−− 9 , +x4 + ax2y2 + y4, mI1(p+), I1 , +x4 + ax2y2 + y4, mI2(p+), I2 if f ∼ +x4 + ax2y2 + y4 for some a ∈ [0, 2) then if f ∼ −x4 + ax2y2 − y4 for some a ∈ (−2, 0] then I1 := [0, 2), I2 := (2, 6] I1 := (−2, 0), I2 := (6,∞) I1 := [−6,−2), I2 := (−2, 0] I1 := (−∞,−6), I2 := (0, 2) 6: let r be the number of real roots of f (x, 1) = bx4 + cx2 + dx + e 7: if r = 0 then 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20: 21: 22: if r = 2 then 23: 24: 25: 26: 27: 28: 9 29: if r = 4 then 30: 31: 32: 33: 34: 35: 9 36: return C1, C2 , +x4 + ax2y2 − y4, mI (p−), I(cid:1) , −x4 + ax2y2 + y4, mI (p−), I(cid:1) I1 := (−∞,−6], I2 = (2, 6] I1 := (−6,−2), I2 = (6,∞) , −x4 + ax2y2 − y4, mI1(p+), I1 , −x4 + ax2y2 − y4, mI2(p+), I2 if f ∼ +x4 + ax2y2 + y4 for some a ∈ (−∞,−6] then if f ∼ +x4 + ax2y2 − y4 for some a ∈ (−∞, 0] then else I := (−∞, 0] I := (0,∞) C1 :=(cid:0)X +− C2 :=(cid:0)X−+ , +x4 + ax2y2 + y4, mI1 (p+), I1 , −x4 + ax2y2 − y4, mI2(p+), I2 else C1 :=(cid:0)X ++ C2 :=(cid:0)X−− 9 9 9 (cid:1) (cid:1) (cid:1) (cid:1) (cid:1) (cid:1) THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 23 9 9 9 . 9 The number of real roots of f (x, 1) = bx4 + cx2 + dx + e, which geometrically correspond to the points of the strict transform on the exceptional divisor of the blow-up at the origin, is invariant under right equivalence and can thus be used as a first step to distinguish the four cases mentioned above. In fact, in the cases (A) and (B) the polynomial f (x, 1) has no real roots, in the case (C) it has two real roots, and in the case (D) it has four real roots. Furthermore, the two cases (A) and (B) can be distinguished by the sign of the coefficient of x4 because it is invariant under right equivalence if f is of either one of these cases. It remains to determine the possible values of the parameter a. Using the techniques from Section 5.1 in Marais and Steenpass (2015b), one can determine, for each real subtype, a polynomial whose coefficients depend on those of f and whose roots are precisely the possible values of the parameter in the normal form equations to which f is complex right equivalent. In Algorithm 8, this is the polynomial p(σ) defined in line 5, with σ = +1 for the subtypes X ++ , and with σ = −1 for X +− and X−− and X−+ However, it turns out that p(σ) does not always factorize into linear factors over Q and that the number of its real roots within the intervals specified in the cases (A) to (D) above is larger than the number of admissible values for the parameter. This is due to the fact that p(σ) also takes into account complex transformations. In other words, for some of the real roots of p(σ), there is no real transformation which takes f to the respective normal form equation where the value of the parameter is that root. We use the following method to solve this problem. Considering, again, Theorem 29 in Marais and Steenpass (2015b), one may observe that the real roots of p(σ) lie in fixed disjoint intervals. In the case (A), for example, the polynomial p+ has exactly one real root in each of the intervals (−2, 0), (0, 2), (2, 6), and (6,∞) (and two more in (−∞,−6) and (−6,−2), which are not admissible values for the parameter in this case), or it has two double roots at 0 and 6 (and one more double root at −6, which we do not consider either). According to that theorem, the two roots which are admissible values for the parameter are either those in [0, 2) and (2, 6] or those in (−2, 0) and (6,∞). Using the techniques from Section 5.3 in Marais and Steenpass (2015b), we set up, for a generic parameter a, the ideal of transformations which take f to the normal form equation with this parameter. Note that a real point in the vanishing set of this ideal corresponds to a real transformation of f. We can thus determine whether there exists a real transformation which maps f to a normal form equation where the parameter lies in a specific interval. To determine this, we can use an algorithm for real root isolation implemented in the Singu- lar library rootisolation.lib (Bendle et al., 2017). Applying ideas from Sommese and Wampler (2005, Ch. 6), it is based on a subdivision method using interval arithmetic to do an exclusion test and an interval Newton step for an inclusion test. Alternatively one could also use methods based on quantor ellimination as described in Algorithm 12.8 from Basu et al. (2008), however, these methods are less efficient. Continuing with case (A), if there exists a real transformation for the real root of p+ in [0, 2), then the two admissible values of the parameter are given by this root and the one in (2, 6], otherwise they are given by the two roots of p+ in (−2, 0) and (6,∞), see lines 9 - 14 in Algorithm 8. The cases (B), (C), and (D) are treated in an analogous way. Remark 5.2. Instead of determining the applicable case and then use a prior known in- tervals, one can also use an ansatz and real root isolation to find appropriate intervals for the possible parameter values along with the computation of the case. To test whether f = bx4 +cx2y2 +dxy3 +ey4 with b, c, d, e ∈ Q is right equivalent to f0 = σ1x4 +ax2y2 +σ2y4 with fixed signs σi and undetermined a, we make an ansatz for the right equivalence trans- formation φ(x) = αx + βy, φ(y) = γx + δy and isolate the roots (α, β, γ, δ, a) ∈ R5 for the polynomial system arising from the condition φ(f ) = f0. This system is represented by an ideal I ⊂ Q[α, β, γ, δ, a]. If this system has a real solution, then f is of type X σ1σ2 . 9 THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 24 Root isolation applied to I yields Cartesian products of intervals, where each product contains a unique solution. Let p be a generator of the elimination ideal I ∩ Q[a]. We can design the root isolation algorithm in a way that each of the a-intervals of the solutions of I contains precisely one solution of p. Then the minimal non-empty intersections of a-intervals of solutions of I lead to intervals isolating the possible parameter values as solutions of p. A minimal polynomial for the parameter value is obtained as an irreducible factor of p. Recall that it is not sufficient to do root isolation on p, since the transformations φ corresponding to these roots may not be defined over the reals. For performance reasons, we can consider a projection R5 → R2, (α, β, γ, δ, a) (cid:55)→ (c1α + c2β+c3γ+c4δ, a) with ci ∈ Q. In every case of Algorithm 8 the number values for a occurring for real solutions is known. If the number of solutions computed from the projection coincides with the expected number, the projection was general enough, otherwise we choose a different projection. 6. Hyperbolic Singularities In this section, we give details on the application of the general algorithm (Algorithm 5) for the hyperbolic singularities. See Figure 5 for these cases. X9+k J10+k Yr,s and Yr Figure 5. Hyperbolic Singularities 6.1. The X9+k case. In Algorithm 9 we demonstrate the general algorithm in the cases X9+k. Note that the Milnor number µ of the input polynomial f can be easily computed, using Gröbner basis techniques, and that k = µ− 9, see Arnold et al. (1985). Also note that E5 ∩ E(µ−7,2) is the vector space generated by all monomials above the Newton polygon. In 2µ−9 line 12 we use that if the exponent of the yk+4 = yµ−5 term is odd, there are two possible normal form equations, which can be obtained from each other by the transformation x (cid:55)→ x, y (cid:55)→ −y. In the algorithm we use the following notation. Notation 6.1. If I ⊂ R and λ ∈ R, then we define λI as the set λI := {λx x ∈ I}. The cases J10+k can be handled in a similar manner. 02468123450246812345024682468 THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 25 Algorithm 9 Algorithm for the case X9+k Input: f ∈ m3 ⊂ Q[x, y] of complex singularity type T = X9+k Output: all normal form equations in the right equivalence class of f, each specified as a tuple of the real singularity type, the normal form, the minimal polynomial of the parameter, and an interval such that the parameter is the unique root of the minimal polynomial in this interval II. Eliminate the monomials in supp(f ) underneath or on Γ(T ) 1: apply Algorithm 2 to f Use Algorithm 4 iteratively: 2: µ := 9 + k 3: c := coeff(f, x2y2) 4: for i = 4 . . . ,(cid:4) µ−3 (cid:5) do 2 5: 6: t := coeff(f, xyi) apply x (cid:55)→ x − t 2c yi−2, y (cid:55)→ y to f III. and IV. Discard the monomials above Γ(T ) and read off the desired information: 9+k (cid:16)b0 (cid:17)µ−5 7: write f as f = b0x4 + b1x2y2 + b2yµ−5 with b0, b1, b2 ∈ Q∗ and R ∈ E5 ∩ E(µ−7,2) 2µ−9 8: TR := X sign(b0), sign(b1) 9: F := sign(b0) · x4 + sign(b1) · x2y2 + ayµ−5 10: p := z4 − b4 11: I1 := sign(b2) · (0,∞), I2 := sign(b2) · (−∞, 0) 12: if µ is odd then 13: 14: else 15: return (TR, F, mI1(p), I1), (TR, F, mI2(p), I2) return (TR, F, mI1(p), I1) 2 b2 1 6.2. The Yr,s case. By the following lemma, if the given polynomial f ∈ Q[x, y] is of real main type Yr,s with r (cid:54)= s, then the 4-jet of f always factors into linear factors over Q. Hence we can follow the general algorithm without any additions. The same may happen if f is of real main type Yr,r. However, if f is of real main type Yr,r and the 4-jet of f does not factor into linear factors over Q, line 3 of Algorithm 5 requires a real algebraic field extension. Lemma 6.2. Let f ∈ Q[x, y] be of real main type Yr,s. If the 4-jet of f does not factorize into linear factors over Q, then r = s, that is, f is of real main type Yr,r. Proof. Let f be of real main type Yr,s for some r, s > 4. Then f is of the form f = (a0x + a1y)2(b0x + b1y)2 + higher terms in x and y where (a0, a1), (b0, b1) ∈ R2 are linearly independent. Hence, the strict transform of the blow-up of f at the origin intersects the exceptional divisor in exactly two points. Without loss of generality, we consider the chart where x = 0 is the local equation of the exceptional divisor. Then the strict transform is given by f = (a0 + a1y)2(b0 + b1y)2 + terms that are divisible by x. Therefore, the intersection points of the strict transform with the exceptional divisor {x = 0} correspond to the zeros of the rational polynomial p2 = (a0 + a1y)2(b0 + b1y)2. Since Q is a perfect field and p2 ∈ Q[y], it follows that p = (a0 + a1y)(b0 + b1y) ∈ Q[y]. THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 26 a1 b1 Now consider the standard charts where x = 0 and y = 0, respectively, define the excep- tional divisor. We show that if the strict transform has an irrational point on the exceptional divisor E in any of the two charts, then in both charts it has two irrational points on E. Indeed, if a1 = 0 or b1 = 0, then p has only rational roots. Hence, by the assumption that the strict transform has an irrational point on E, we have a1 (cid:54)= 0 and b1 (cid:54)= 0, and at least one of the two roots − a0 is irrational. Since p is the minimal polynomial for these roots, we conclude that both − a0 are irrational. and − b0 , − b0 a1 b1 b1 a1 ) and (0,− b0 Note that these roots correspond to the irrational points (0,− a0 ) of the strict transform on E. Hence, the points of the strict transform on the exceptional divisor are irrational if and only if p does not have any rational root. This, in turn, is the case if and only if jet(f, 4) = (a0x + a1y)2(b0x + b1y)2 does not factorize into linear factors over Q. Thus, under the assumptions of the lemma, both roots of p and both points of the strict transform on the exceptional divisor are irrational. Consider the blow-up of the normal form equation of f at the origin. The germs at the two intersection points q1 and q2 of the strict transform with the exceptional divisor are right equivalent to f1 = ±x2 ± yr−4 and f2 = ±x2 ± ys−4. Since any right equivalence of a singularity induces right equivalences for each germ of the strict transform, the singularities of the strict transform of f are also right equivalent to f1 and f2. The Galois group of the quadratic extension Q ⊂ Q[y]/p =: K is isomorphic to Z/2Z. Let κ be the non-trivial element in this group, and let ϕ1 and ϕ2 be the K-algebra automorphisms which transform the germs at q1 and q2 to f1 and f2, respectively. Then q1 and q2 are conjugate via κ, and we have κ−1 ◦ ϕ1 ◦ κ = κ(ϕ1) = ϕ2 as K-algebra automorphisms identified with each other by the group action of Gal(K Q) on Aut(K). Also note that the map κ−1 ◦ ϕ1 ◦ κ can be extended to an R-algebra automorphism because its restriction to K is just the identity, and note that the germ at q2 is right equivalent to f1 via this map. Hence f1 ∼ f2 and thus r = s. (cid:3) because both (cid:101)f and f1 are invariant under κ = κ−1. In other words, ϕ1 and ϕ2 can be We now discuss an explicit version of Algorithm 5 which works for all rational input polynomials f ∈ Q[x, y] of complex type Yr,s. The framework is given in Algorithm 10. We realize Step (II) of the general algorithm in Algorithm 11 and Step (IV) in Algorithm 12. Algorithm 10 Algorithm for the cases Yr,s Input: f ∈ m3 ⊂ Q[x, y] of complex singularity type T = Yr,s Output: all normal form equations in the right equivalence class of f, each specified as a tuple of the real singularity type, the normal form, the minimal polynomial of the parameter, and an interval such that the parameter is the unique root of the minimal polynomial in this interval 1: f0 := f II. Eliminate the monomials in supp(f ) underneath or on Γ(T ) 2: replace f by the output of Algorithm 11 applied to f III. Eliminate the monomials above Γ(T ) 3: f := w-jet(f, dw) with w := w(T ) and dw := d(T ) IV. Read off the desired information 4: Output of Algorithm 12 applied to f0 and f Suppose f is of real main type Yr,s. Algorithm 11 removes all monomials below or on the Newton polygon, with the result defined either over Q or over a simple real algebraic extension of Q. From this, Algorithm 12 computes a linear or quadratic minimal polynomial THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 27 Algorithm 11 Step (II) in the general algorithm for the cases Yr,s Input: f ∈ m3 ⊂ Q[x, y] of complex singularity type T = Yr,s Output: f(cid:48) ∈ m3 ⊂ K[x, y], where K is Q or a quadratic extension field of Q, such that f K∼ f(cid:48) and f(cid:48) has no terms underneath Γ(T ) 1: h := jet(f, 4) 2: b := coeff(f, x2y2) 2 over Q, where g1, g2 are homogeneous of degree 1 1g2 Implement line 3 from Algorithm 5: 3: if h has a linear factor over Q then 4: 5: 6: else 7: 8: 9: 10: factorize h as bg2 apply g1 (cid:55)→ x, g2 (cid:55)→ y to f factorize h as bg2 over Q, where g is homogeneous of degree 2 K := Q[t]/g(1, t) over K, factorize h as bg2 1g2 apply g1 (cid:55)→ x, g2 (cid:55)→ y to f 2, where g1, g2 are homogeneous of degree 1 Implement lines 13 - 15 from Algorithm 5: 11: n1 := 3, n2 := 3 12: while coeff(f, xn1) = 0 or coeff(f, yn2 ) = 0 do 13: 14: 15: n1 := n1 + 1 apply x (cid:55)→ x , y (cid:55)→ y − coeff(f,xn1 y) if coeff(f, xn1) = 0 then xn1−2 to f 2b if coeff(f, yn2 ) = 0 then 16: 17: 18: 19: return f(cid:48) := f n2 := n2 + 1 apply x (cid:55)→ x − coeff(f,xyn2 ) 2b yn2−2, y (cid:55)→ y to f for the moduli parameter. The algorithm then determines which of the roots of the minimal polynomial are valid moduli parameters and computes the corresponding real subtypes. We now give a detailed exposition of Algorithms 11 and 12. Before entering the algorithm we remember f0 = f. For Algorithm 11 we set h := jet(f, 4). The algorithm then considers the following two cases: 1. h(1, y) has a linear factor over Q: This case is handled in lines 3 - 5 and 11 - 18 of Algorithm 11. It includes all singularities of real main type Yr,s with r (cid:54)= s and those of type Yr,r for which the 4-jet factorizes into linear factors over Q. We exactly follow the general algorithm. 2. h(1, y) does not have a linear factor over Q: This case is handled in lines 6 - 18 of Algorithm 11. It includes all singularities of type Yr,r for which the 4-jet does not factorize into linear factors over Q. We first discuss 2, with g1 and g2 non-associate lines 6 - 10. Over the reals, jet(f, 4) factorizes as g2 and homogeneous of degree 1. Since Q is a perfect field, we have g1g2 ∈ Q[x, y]. Since g1, g2 (cid:54)∈ Q[x, y], the polynomial g1g2(1, y) is irreducible. Passing to the extension field K = Q[y]/g1g2(1, y), we apply the automorphism g1 (cid:55)→ x, g2 (cid:55)→ y to f. Lines 10 - 18 of Algorithm 11 follow the general algorithm, using transformations over K to find a polynomial which is right equivalent to f and of the form bx2y2 + dxr + eys + R with b ∈ Q, d, e ∈ K, and R ∈ E(n1−2,2) ∩ E(2,n1−2) 1g2 . n1+1 n1+1 We now pass to Algorithm 12. Here we set h := jet(f0, 4). The algorithm considers the following two cases: THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 28 Algorithm 12 Step (IV) in the general algorithm for the real main types Yr,s Input: f0 ∈ m3 ⊂ Q[x, y] of real singularity type Yr,s and f ∈ m3 ⊂ K[x, y], where K = Q or K is a real quadratic extension of Q, such that f is of complex singularity type T = Yr,s, supp(f ) = supp(T ), and f K∼ f0 Output: all normal form equations in the right equivalence class of f0, each specified as a tuple of the real singularity type, the normal form, the minimal polynomial of the parameter, and an interval such that the parameter is the unique root of the minimal polynomial in this interval write f as f = bx2y2 + dxr + eys with b, d, e ∈ Q∗ if r is odd and s is even then apply x (cid:55)→ y, y (cid:55)→ x to f write f as f = bx2y2 + dxr + eys with b, d, e ∈ Q∗ p := z2r − b−rsd2se2r, p := z2s − b−rsd2se2r ∈ Q[z] σ1 := sign(b), σ2 := sign(d), σ3 := sign(e), I := (0,∞) if r and s are even then 1: h := jet(f0, 4) 2: if h has a linear factor over Q then 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: C1 := (Y σ1σ2 C2 := (Y σ1σ3 if r = s and σ2 = σ3 then if r is even and s is odd then , σ1x2y2 + σ2xr + ays, mσ3·I (p), σ3 · I) , σ1x2y2 + σ3xs + ayr, mσ2·I (p), σ2 · I) return C1, C2 return C1 else r,s s,r for i = 1, 2 do Ci := (Y σ1σ2 Ci+2 := (Y σ1 (−1)i return C1, C2, C3, C4 s,r r,s , σ1x2y2 + σ2xr + ays, m(−1)i·I (p), (−1)i · I) , σ1x2y2 + (−1)ixs + ayr, mσ2·I (p), σ2 · I) if r and s are odd then for i = 1, 2 do for j = 1, 2 do C2i+j−2 := (Y σ1 (−1)i C2i+j+2 := (Y σ1 (−1)i s,r r,s , σ1x2y2 + (−1)ixr + ays, m(−1)j·I (p), (−1)j · I) , σ1x2y2 + (−1)ixs + ayr, m(−1)j·I (p), (−1)j · I) 19: 20: 21: 22: 23: 24: 25: 26: 27: 28: 29: 30: else 31: 32: 33: 34: 35: 36: 37: if r = s then return C1, C2, C3, C4 else return C1, C2, C3, C4, C5, C6, C7, C8 write f as f = bx2y2 + dxr + eyr with b ∈ Q∗ and d, e ∈ K∗ σ := sign(b), I := (0,∞) if r is odd then p := z2 − b−r(de)2 ∈ Q[z] for i = 1, 2 do Ci := (Y σ+ Ci+2 := (Y σ− r,r , σx2y2 + xr + ayr, m(−1)i·I (p), (−1)i · I) r,r , σx2y2 − xr + ayr, m(−1)i·I (p), (−1)i · I) 38: return C1, C2, C3, C4 THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 29 else if de < 0 then for i = 1, 2 do 39: 40: 41: 42: 43: 44: 45: 46: 47: 48: 49: Ci := (Y σ (−1)i r,r return C1, C2 else , σx2y2 + (−1)ixr + ayr, z − (−1)ib −r 2 de, (−∞,∞)) write g(1, t) as defined in Algorithm 11 as g(1, t) = αt2+βt+γ and d = λ1+λ2t if sign(λ1 − λ2 · β return (Y σ − 2α ) = −1 then r,r , σx2y2 − xr + ayr, z + b −r r,r , σx2y2 + xr + ayr, z − b −r 2 de, (−∞,∞)) 2 de, (−∞,∞)) return (Y σ + else 1. h(1, y) has a linear factor over Q: This case is handled in lines 2 - 29 of Algorithm 12, which follow the general algorithm. We use Theorem 33 in Marais and Steenpass (2015b) to determine the possible normal form equations, depending on whether r and s are even or odd, respectively. 2. h(1, y) does not have a linear factor over Q: This case is handled in lines 30 - 49 of Algorithm 12. Note that the input polynomial f is an element of K[x, y], where K = Q[t] is a real quadratic number field. After normalizing and completing the square in g(1, t) as defined in Algorithm 11, we may D for some positive discriminant D ∈ Q. Furthermore, note that assume that t = f is of the form √ f = bx2y2 + dxr + eyr with b ∈ Q and d, e ∈ K, and let σ = sign(b). The case where r is odd is handled in lines 33 - 37. In this case, f is right equivalent r,r , taking into account the transformation x (cid:55)→ −x and y (cid:55)→ y. to both the types Y σ± By Theorem 33 from Marais and Steenpass (2015b) there are only two possible values for the parameter, which are both roots of the polynomial z2 − b−r(de)2. If a is a possible value for the parameter, then so is −a by the transformation x (cid:55)→ x, y (cid:55)→ −y. On the other hand, suppose ϕ is an automorphism which transforms f0 to a normal form equation with parameter λ1 + λ2t, and let κ be the conjugation on K. Then κ ◦ ϕ ◦ κ−1 is also an automorphism and transforms f0 to the same normal form equation with the parameter replaced by κ(λ1 + λ2t) = λ1 − λ2t. Hence, λ1 + λ2t = −(λ1− λ2t) or λ1 + λ2t = λ1− λ2t. Thus, λ1 = 0 or λ2 = 0, and the square of the parameter is a rational number. Therefore, z2−b−r(de)2 is a polynomial with rational coefficients. −b r determined. Hence, by a similar argument as above, b− r under conjugation, and we conclude that de ∈ Q. transformation x (cid:55)→ y, y (cid:55)→ x into account, f is both of type Y σ+ and of type Y σ− Consider now the case where r is even. We either have a = +b r 2 de or a = 2 de. By Theorem 33 of Marais and Steenpass (2015b) the parameter a is uniquely 2 a and thus de is invariant The case where r is even and de < 0 is handled in lines 39 - 43. Taking the 2 de, r,r with a = b r r,r with a = −b r The case where r is even and de > 0 is handled in lines 44 - 49. Since de > 0, the coefficient of xr and the parameter a in the normal form equation have the same sign. For any polynomial dxr + bx2y2 + eyr in the right equivalence class of f, the sign of d is the same. Since the extension Q ⊂ K = Q[t] is quadratic, d is of the form 2 de. THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III d = λ1 + λ2t, with λi ∈ Q. Hence, for (cid:101)d(T ) = λ1 + λ2T ∈ Q[T ] and t1,2 = ±√ have sign((cid:101)d(t1)) = sign((cid:101)d(t2)). Thus sign(d) = sign((cid:101)d(0)) = sign(λ1). 6.3. The (cid:101)Yr case. In this case the general algorithm can be followed, except for a slight for T = (cid:101)Yr. We start by removing monomials in f on Γ(T ) that are not in supp(T ). To do modification in Step (III). Note that there are no monomials in supp(f ) underneath Γ(T ) this, we apply a linear transformation to f such that supp(jet(f, d)) = supp(T, d) for d = 4. This transformation is implemented in Algorithm 13. 30 D, we Algorithm 13 Input: f ∈ Q[x, y] of real type (cid:101)Yr Output: h ∈ Q[x, y] with f ∼ h and supp(jet(h, 4)) = supp(T, 4) 1: g := jet(f, 4) 2: factorize g over Q as cf 2 3: apply x (cid:55)→ x + a 4: return f 1 with f1 = x2 + axy + by2, and a, b, c ∈ Q 2 y, y (cid:55)→ y to f Next we remove the monomials in f above Γ(T ) not in supp(T ). Note that the Arnold parameter system of germs of type (cid:101)Yr contains only the monomial xr. Hence, similarly to the case of germs that are equivalent to a germ with a non-degenerate Newton boundary, we iteratively consider the degree j part of f, for increasing j with dw < j < r. Since the coefficients of the elements of a basis of R[x, y]/(cid:104) ∂f ∂y(cid:105) in degree j are zero in the Arnold normal form, we can write this polynomial as ∂x , ∂f ∂f0 ∂x v1 + ∂f0 ∂y v2, be discarted. We summarize this procedure in Algorithm 14. with v1, v2 ∈ R[x, y]. Applying the transformation x (cid:55)→ x − v1, y (cid:55)→ y − v2 results in transforming the part of f of degree j to 0. Similarly, we then can transform the degree r part of f to cxr, c ∈ R. Since a germ of type (cid:101)Yr is r-determined, terms of higher degree can Algorithm 14 Step (III) in the general algorithm for the real main types (cid:101)Yr Input: f ∈ Q[x, y] of real type (cid:101)Yr such that supp(jet(f, 4)) = supp(T, 4) q = x2 + by2 ∈ R[x, y], a, b ∈ R Output: h ∈ Q[x, y] with f ∼ h and supp(jet(h, r)) = supp(T, r) 1: g0 = jet(f, 4) = aq2, 2: i := 5 3: p5 = (jet(f, 5) − jet(f, 4))/(4aq) 4: while i (cid:54)= r + 1 do pxi = pi(x, 0); 5: pyi = pi − pxi; 6: x , y (cid:55)→ y − pyi apply x (cid:55)→ x − pxi 7: 8: i = i + 1 pi = (jet(f, i) − jet(f, 4))/(4aq) 9: 10: return f by to f; THE CLASSIFICATION OF REAL SINGULARITIES USING SINGULAR, PART III 31 References Arnold, V.I., 1974. Normal forms of functions in neighbourhoods of degenerate critical points. Russ. Math. Surv. 29(2), 10-50. Arnold, V.I., Gusein-Zade, S.M., Varchenko, A.N., 1985. Singularities of Differential Maps, Basu, S., Pollack, R., Roy, M.-F., 2008. Algorithms in Real Algebraic Geometry. Springer, Vol. I. Birkhäuser, Boston. Berlin. Böhm, J., Marais, M.S., Pfister G., 2017. A classification algorithm for complex singularities of corank and modality up to two. Singularities and Computer Algebra. Festschrift for Gert-Martin Greuel on the Occasion of his 70th Birthday. Editors: Decker, Wolfram, Pfister, Gerhard, Schulze, Mathias. 2017. Böhm, J., Marais, M.S., Pfister G., 2017. classify2.lib. A Singular 4-1-0 library for classifying isolated hypersurface singularities of modality and corank up to two w.r.t. right equivalence, based on the algorithm by J. Böhm, M.S. Marais and G. Pfister. Singular distribution. Bendle, D., Böhm, J., Petroll, C., 2017. rootisolation.lib. A Singular 4-1-0 library for exact real root isolation. Singular distribution. Decker, W., Greuel, G.-M., Pfister, G., Schönemann, H., 2017. Singular 4-1-0 – A computer algebra system for polynomial computations. http://www.singular.uni-kl.de de Jong, T., Pfister, G., 2000. Local Analytic Geometry. Vieweg, Braunschweig. Greuel, G.-M., Lossen, C., Shustin, E., 2007. Introduction to Singularities and Deformations. Springer, Berlin. ed. Springer, Berlin. Greuel, G.-M., Pfister, G., 2008. A Singular Introduction to Commutative Algebra, second Marais, M., Steenpass, A., 2013. realclassify.lib. A Singular 4-0-2 library for classi- fying isolated hypersurface singularities over the reals w.r.t. right equivalence. Singular distribution. Marais, M., Steenpass, A., 2015. The classification of real singularities using SINGULAR Part I: Splitting Lemma and Simple Singularities. J. Symb. Comput. 68 (2015), 61-71. Marais, M., Steenpass, A., 2015. The classification of real singularities using SINGULAR Part II: The Structure of the Equivalence Classes of the Unimodal Singularities. Symb. Comput. (2015). Sommese, A. J., Wampler, C. W., 2005. The Numerical Solution of Systems of Polynomials - Arising in Engineering and Science, World Scientific Publishing Co. Pte. Ltd., 2005 Tobis, E., 2012. rootsur.lib. A Singular 4-1-0 library for counting the number of real roots of a univariate polynomial. Wenk, M., Winfried, P., 1999. solve.lib. A Singular 4-1-0 library for complex solving of polynomial systems. Department of Mathematics, University of Kaiserslautern, Erwin-Schrödinger-Str., 67663 Kaiserslautern, Germany E-mail address: [email protected] University of Pretoria and African Institute for Mathematical Sciences, Department of Mathematics and Applied Mathematics, Private bag X20, Hatfield 0028, South Africa E-mail address: [email protected] Department of Mathematics, University of Kaiserslautern, Erwin-Schrödinger-Str., 67663 Kaiserslautern, Germany E-mail address: [email protected]
1701.01807
2
1701
2017-01-16T14:46:17
Matrix divisors on Riemann surfaces and Lax operator algebras
[ "math.AG" ]
Matrix divisors are introduced in the work by A.Weil (1938) which is considered as a starting point of the theory of holomorphic vector bundles on Riemann surfaces. In this theory matrix divisors play the role similar to the role of usual divisors in the theory of line bundles. Moreover, they provide explicit coordinates (Tyurin parameters) in an open subset of the moduli space of stable vector bundles. These coordinates turned out to be helpful in integration of soliton equations. We would like to gain attention to one more relationship between matrix divisors of vector G-bundles (where G is a complex semi-simple Lie group) and the theory of integrable systems, namely to the relationship with Lax operator algebras. The result we obtain can be briefly formulated as follows: the moduli space of matrix divisors with certain discrete invariants and fixed support is a homogeneous space. Its tangent space at the unit is naturally isomorphic to the quotient space of M-operators by L-operators, both spaces essentially defined by the same invariants (the result goes back to Krichever, 2001). We give one more description of the same space in terms of root systems.
math.AG
math
MATRIX DIVISORS ON RIEMANN SURFACES AND LAX OPERATOR ALGEBRAS O.K. SHEINMAN Contents Introduction 1. 2. Matrix divisors and flag configurations 3. Canonical form of a matrix divisor. The moduli space 4. Moduli of matrix divisors and Lax operator algebras References 1 4 7 13 15 1. Introduction Matrix divisors are introduced in the work by A.Weil [29] which is considered as a starting point of the theory of holomorphic vector bundles on Riemann surfaces. The classification of the holomorphic vector bundles on Riemann surfaces by A.N.Tyurin [23, 24, 25] based on matrix divisors, the well-known Narasimhan -- Seshadri description of stable vector bundles [13], and subsequent description of the moduli space of vector bundles with the parabolic structure [14, 12] originate in [29]. In the theory of holomorphic vector bundles the matrix divisors play the role similar to the role of usual divisors in the theory of line bundles. The matrix divisor approach to classification of holomorphic vector bundles provides invariants not only of stable bundles but also of fam- ilies of smaller dimensions. Moreover, it provides explicit coordinates, invented in [24], in an open subset of the moduli space of stable vec- tor bundles. In [8], these coordinates were given the name of Tyurin parameters and applied in integration of soliton equations. To be more specific, assume that a holomorphic rank n vector bundle has the n-dimensional space of holomorphic sections. Then any base of Partial support by the Internal Research Project GEOMQ11, University of Lux- embourg, and by the OPEN scheme of the Fonds National de la Recherche (FNR), Luxembourg, project QUANTMOD O13/570706 is gratefully acknowledged. 1 MATRIX DIVISORS 2 the space of the holomorphic sections is called framing, and the bun- dle with a given framing is called a framed bundle. The classification of the framed holomorphic vector bundles is one of the main results of [23, 24, 25]. In particular, it follows from [24, 25] that the moduli space of stable framed rank n holomorphic vector bundles of degree 0 is a quasiprojective variety of dimension (n2 − 1)(g − 1) where g ≥ 2 is the genus of the Riemann surface, and if in the same set-up we con- sider the bundles of degree ng then the dimension of the corresponding quasiprojective variety is equal to n2(g − 1) + 1. It has been also shown by Tyurin that the bundles which do not possess any natural framing depend on a smaller number of parameters. In the present paper, we address the problem of classifying the ma- trix divisors. It is a straight forward generalization of the problem of classifying the framed vector bundles. Indeed, let ψU n be the elements of a framing represented in local coordinates (i.e. the local meromorphic vector-functions defined at the local coordinate set U). Then the collection of matrices ΨU formed by them at every U form a matrix divisor. 1 , . . . , ψU We would like to gain attention to one more relationship between matrix divisors and the theory of integrable systems, namely to the relationship with Lax operator algebras. Those came to existence due to the theory by Krichever [9] of integrable systems with the spectral parameter on a Riemann surface. Originally, this theory has been moti- vated in part by the Tyurin parametrization of framed vector bundles. In [18, 22] it has been developed in the different, and more general set-up related to Z-gradings of the semisimple Lie algebras. The main purpose of the present work is to develop the corresponding set-up in the theory of matrix divisors. The result we obtain on this way can be briefly formulated as follows. Theorem. The moduli space M of matrix divisors with certain discrete invariants and fixed support is a homogeneous space. For its tangent space at the unit we have (1.1) TeM ∼= ML/L where L is the Lax operator algebra essentially defined by the same invariants, ML is the corresponding space of M-operators. This result goes back to [9]. We refer to Section 4, in particular to Theorem 4.1, for the details, notation, and a more precise statement. We were not able to find out any reference for the matrix divisors of G-bundles for G a complex semisimple group. It is one of the purposes of the present work, closely related to the main purpose, to propose a MATRIX DIVISORS 3 treatment of such matrix divisors. To do that, we use the Chevalley groups over the field (resp. ring) of Laurent (resp. Tailor) series. It is a very adequate set-up for matrix divisors by our opinion, because a Chevalley group is defined by a (complex, semisimple) Lie algebra and its faithful representation given by a highest weight. Such data con- tain information both on the group structure and on the fibre of the bundle. Moreover, the Cartan decomposition of Chevalley groups in its general form provides a convenient description of the canonical form of a matrix divisor (Theorems 3.1,3.2 below). The Cartan decomposition, in particular, states that for an arbitrary Chevalley group G over the field of Laurent series the following holds: G = KA+K where K is the same group considered over the ring of Tailor series, and A+ is a cham- ber in the maximal torus. In [24] the same role is played by Lemma 1.2.1. However, the last claims a stronger statement, namely it speci- fies the form of the K-component of the decomposition in the following quite beautiful way: let k be the K-component at a certain point of the divisor support, diag(zd1, . . . , zdn) ∈ A+ be the toric component, d1 ≤ . . . ≤ dn, Eij are the matrix units, then k = E +Xi<j aij(z)Eij, aij(z) ∈ C(z)/zdj −di C(z) where C(z) is the ring of Tailor series. Since we are not able to follow all the arguments by A.N.Tyurin in course of deriving that expression, we reinterpret it, generalize it to the case of an arbitrary reduced root system R, and thus obtain the following description of the tangent space to the moduli space of matrix divisors (see Theorem 3.9 below for the more precise statement). Theorem. The tangent space to M at the unit consists of elements of the form Mγ∈Γ Xα∈R+ aγ α(z)xα, aγ α(z) ∈ C(z)/zα(hγ )C(z) where Γ is the divisor support, hγ comes from the maximal torus com- ponent of the Cartan decomposition at γ. Finally it turns out to be an important argument for establishing the above relationship (given by (1.1)) between matrix divisors and Lax operator algebras. In the present paper we assume G to be semi-simple which corre- sponds to the case of topologically trivial holomorphic vector bundles. To include the topologically non-trivial bundles we would need to con- sider the conformal extensions of semi-simple groups [11] instead. Let G be a complex semi-simple group with the finite center Z equal to a direct sum of r cyclic components. By conformal extension of G we MATRIX DIVISORS 4 call Gc = G ×Z (C∗)r. For example, GL(n, C) is a conformal extension of SL(n, C). We do not focus on this easy modification here. The plan of the present paper is as follows. In Section 2 we give the preliminaries on matrix divisors, our treatment of this notion related to Chevalley groups, and a description of the space of sections of a matrix divisor in terms of certain flag configurations. In Section 3 we define the moduli space of matrix divisors as a certain coset, and prove Theorem 3.9 giving a description of its tangent space at the unit in terms of the root system of the group, and the weight lattice of the underlying module. In Section 4 we give preliminaries on Lax operator algebras (see [22] for the details) and then complete the interpretation of the moduli space from the point of view of integrable systems iden- tifying the tangent space at the unit with the coset of the space of M-operators by the space of L-operators in spirit of [9], relying on the results of [22]. I would like to express my gratitude to I.M.Krichever whom I am indebted with my interest to the subject, and who is a pioneer of many ideas relating integrable systems and holomorphic vector bundles on Riemann surfaces. I am grateful to M.Schlichenmaier with whom we started to discuss the subject many years ago, and to E.M.Chirka for his help in complex analysis. I would like to acknowledge a special role of discussions with E.B.Vinberg on Lax operator algebras and algebraic groups. 2. Matrix divisors and flag configurations Let G denote a Chevalley group given by a semi-simple complex Lie algebra g, a faithful g-module V with dominant highest weight, and a field k. We recall that G is the group of automorphisms of the k-space V k = V ⊗Q k generated by the 1-parameter subgroups of ∞ Pn=0 automorphisms of the form exp tXα = tnX n α/n! (the sums being actually finite on V k) where Xα is the root vector of the root α. Let Σ be a Riemann surface. The following system of definitions reproduces the corresponding def- initions in [24]. Definition 2.1. Assume each point of Σ to be assigned with a germ of meromorphic G-valued functions holomorphic except at a finite set Γ ⊂ Σ. Such a correspondence is called distribution with the support Γ. Definition 2.2. Two distributions Ax and Bx, x ∈ Σ are equivalent if there is a third distribution Cx holomorphic for every x ∈ Σ and such MATRIX DIVISORS 5 that CxAx = Bx. Any class of equivalent distributions is called matrix divisor. That Cx is holomorphic and G-valued implies in particular that it is holomorphically invertible. We delay the discussion of equivalent matrix divisors until Section 3 (Definition 3.5 and below). Definition 2.3. Given a matrix divisor Ψ, by its local section (or just section) we mean a meromorphic V -valued function f on an open sub- set U ⊂ Σ such that f is holomorphic on U\Γ and Ψγf is holomorphic in the neighborhood of any γ ∈ Γ ∩ U. We denote the sheaf of sections by ΓV (Ψ). It has a simple description in terms of flag configurations related to the divisor. Given a matrix divisor Ψ we assign a flag in V to every point in its support. Thus the divisor turns out to be assigned with the system of flags which we call a flag configuration (this term assumes Γ to be fixed). Let f be a meromorphic V -valued function on U ⊂ Σ. In order f be a section it is required that (2.1) s = Ψf is holomorphic at every γ ∈ Γ ∩ U where Γ = support Ψ. Assume Ψ to have an expansion of the form Ψ = Ψizi at γ, and f to have an expansion of the form ∞ Xi=−m ∞ Xi=−k f = fizi there. We take −k = ordγ Ψ−1 the following system of m + k linear equations γ which follows from (2.1). We then have (2.2) Ψ−mf−k = 0, Ψ−mf−k+1 + Ψ−m+1f−k = 0, . . . , Ψ−mfm−1 + Ψ−m+1fm−2 + . . . + Ψk−1f−k = 0 . expressing the fact that the terms containing z−m−k, . . . , z−1 of Ψf vanish, i.e. s in (2.1) is holomorphic. This system of equations is ho- mogeneous, hence all the components of its solutions constitute linear spaces. Let Fi be the subspace in V constituted by fi's for all solutions f−k, f−k+1, . . . , fm−1 to (2.2). MATRIX DIVISORS 6 The following lemma claims that the subspaces Fi, i = −k, . . . , m−1 constitute a flag; we were not able to find any reference for it. Similar arguments are used for the flag interpretation of opers [4]. Lemma 2.4. F−k ⊆ F−k+1 ⊆ . . . ⊆ Fm−1 ⊆ V . Proof. Solutions to (2.2) can be found out by an inductive procedure. The first equation is independent and homogeneous. The space of its solutions is exactly what we have denoted by F−k. Solutions to the second equation can be found out by plugging an arbitrary f−k ∈ F−k, and resolving the obtained system of equations. In particular, we can plug f−k = 0. Then the second equation coincides with the first one, hence F−k ⊆ F−k+1. The ith equation of (2.2) (at z−m−k+i) is as follows: Ψ−mf−k+i + Ψ−m+1f−k+i−1 + . . . + Ψ−m+if−k = 0. The next equation has the form Ψ−mf−k+i+1 + Ψ−m+1f−k+i + . . . + Ψ−m+if−k+1 + Ψ−m+i+1f−k = 0. If f−k = 0, the (i + 1)th equation degenerates to the ith one. Hence every solution of the ith equation is a particular solution of the (i+1)th equation, i.e. F−m−k+i ⊆ F−m−k+i+1. (cid:3) The description of the sheaf ΓV (Ψ) is now as follows. Lemma 2.5. ΓV (Ψ) is the sheaf of local meromorphic V -valued func- tions on Σ satisfying the following requirement for every γ ∈ Γ. Let f i zi be its Laurent expansion at a i ⊆ be such a function, and f (z) = P f γ γ ∈ Γ. Then it is required that f γ i ∈ F γ . . . ⊆ V is the flag corresponding to γ. i where Fγ : {0} ⊆ . . . ⊆ F γ The proof immediately follows from Definition 2.3 and the definition of F . In the Lemma 2.5 we consider the flag to be semi-infinite to the right, where Fi stabilize to V since a certain moment. Definition 2.6. Given a matrix divisor Ψ we call the Lie algebra of meromorphic g-valued functions on Σ leaving invariant ΓV (Ψ) by en- domorphism algebra of Ψ, and denote it by End(Ψ). Next we give a description of the Lie algebra End(Ψ) in terms of the flag configuration related to Ψ. Let g = Lie(G). Given a flag F consider the following filtration of g. Remind that V is a g-module. For every i consider a subspace gi ⊆ g such that giFj ⊆ Fj+i for every j. Then gi ⊆ gi+1 because giFj ⊆ Fj+i ⊆ Fj+i+1. MATRIX DIVISORS 7 Lemma 2.7. End(Ψ) is the subspace of the space of all g-valued merom- orphic functions on Σ satisfying the following requirement for every γ ∈ Γ. Let L be such a function, and L(z) = P Lizi be its Laurent expansion at a γ ∈ Γ. Then Li ∈ gi, ∀i. It is instructive to keep in mind the homological interpretation of matrix divisors [26, 27]. In this approach a matrix divisor is defined as a 0-cochain with coefficients in the sheaf G(R) of rational G-valued functions whose boundary is a 1-cocycle with coefficients in the sheaf G(O) of regular G-valued functions. Thus an open covering {Ui} of Σ is assigned with the system of local rational G-valued functions fi such that its boundary fij = fif −1 is regular and regularly invertible on Ui ∩ Uj, and fijfjkfki = 1 on Ui ∩ Uj ∩ Uk for every triple (i, j, k). It is clear that the cocycle fij (the system of gluing functions in other terminology) is invariant with respect to the right action of any global G-valued function. j 3. Canonical form of a matrix divisor. The moduli space Let o be a principal ideal ring (commutative, with the unit), k be its quotient field, G be a rank l Chevalley group over k, K = Go be the same group over o, H be a maximal torus in G, i.e. the subgroup generated by the 1-parameter subgroups h′ i(t) acts on ev- ery vector of a weight µ in V as a multiplication by tµ(hi), hi ∈ h (i = 1, . . . , l) form a base of the lattice L∗ V dual to LV , the last being generated by the weights of the module V [15, Lemma 35, p.58]. i(t) where h′ For an obvious reason h′ i(t) is denoted also by thi, and H consists of the elements of the form th1 l where ti ∈ k \ {0}, i = 1, . . . , l. 1 The following example shows how the module V affects the torus of the corresponding Chevalley group. · . . . · thl Example 1. Consider g = sl(2, C). Up to the end of the example let h denote the canonical generator of the Cartan subalgebra. The standard g-module has the weights ±1, and the adjoint module the weights {±2, 0}. Hence the dual lattice to the weight lattice LV is generated by h in the first case, and by (1/2)h in the second case. The corresponding tori consist of elements of the form tρ(h)n in the first case, and tρ(h)n/2 in the second case where n ∈ Z and ρ(h) denotes the corresponding representation operator (i.e. ρ(h) = diag(1, −1) in the first case, and ρ(h) = diag(2, 0, −2) in the second case). By definition of a Chevalley group it is assumed that the representa- tion of g in V is a faithful representation. For the reason that the root lattice is always a sublattice of LV we have α(hi) ∈ Z for every α ∈ R, MATRIX DIVISORS 8 i = 1, . . . , l. Let A+ denote the chamber in H given by the tuples ∈ o for every positive i=1 tα(hi) i {t1, . . . , tl} satisfying the condition Ql root α. Theorem 3.1 ([15], Theorem 21). (a) G = KA+K (the Cartan decomposition); (b) The A+-component in (a) is defined uniquely modulo H ∩ K. 1 · . . . · thl th1 h ∈ L∗ as ti = zdi where di ∈ Z (i = 1, . . . , l). Let h = Pl weight µ in V ). The condition Ql We need a particular case of Theorem 3.1 when o = C(z) is the ring of Tailor expansions in z, k = C(z−1, z) is the field of the Laurent expansions. The elements ti up to units of the ring C(z) can be taken i=1 dihi. Then l = zh (once again, zh is defined on a module V such that V , and operates as multiplication by zµ(h) on the subspace of ∈ o means that zα(h) is holomorphic, i.e. α(h) ≥ 0, for every positive α. We conclude that H = {zhh ∈ L∗ In this case we obtain the following Cartan decomposition for current groups out of Theorem 3.1. In certain particular cases it is also known as factorization theorem in the theory of holomorphic vector bundles. V } and A+ = {zhα(h) ≥ 0, ∀α ∈ R+}. i=1 tα(hi) i Theorem 3.2. Let G = G(C(z−1, z)) be the Chevalley group given by a simple Lie algebra g over C and its module V , K = G(C(z)) be the same group over the ring of Taylor series, H and A+ be as introduced above. Then G = KA+K, and the A+-component in the decomposition is determined uniquely up to HC. By virtue of Theorem 3.2 the support of a divisor can be character- ized as the set of those points in Σ for which h 6= 0. Up to equivalence given by left multiplication by a distribution taking values in K we can assume that Ψ = zhk(z), k(z) ∈ K where h and k depend on the point of supp Ψ. We call it the reduced form of the divisor. Given a g-module V of highest weight χ, and an h ∈ (L∗ V )+ we introduce the following flag F in V (below m = χ(h), m ∈ Z+). First we define the grading of the module V : m V = (3.1) Mi=−m Obviously Vi = Lµ(h)=−i Vi where Vi = {v ∈ V hv = −iv}. Vµ, Vµ being the weight subspace of V of weight µ. MATRIX DIVISORS 9 Next we define the flag F : {0} ⊆ F−m ⊆ . . . ⊆ Fm = V by setting (3.2) Fj = Vs. j Ms=−m In particular, F−m is generated by the highest weight vector. Let Q = Z(R) be the root lattice of g. Lemma 3.3. Let h ∈ L∗ to the divisor Ψ = zh by virtue of Lemma 2.4. V . Then F is nothing but the flag corresponding Proof. We resolve the equation (2.1) for Ψ = zh: f = z−hs where s(z) sjzj is holomorphic in the neighborhood of z = 0. Take s(z) = ∞ where sj ∈ V for every j ≥ 0. Let sj = s(i) j be an expansion of sj Pj=0 m Pi=−m s(i) j zi, according to the grading of V , i.e. s(i) j ∈ Vi. Then m hence f (z) = z−hsj = Xi=−m j zi! zj = s(i) ∞ Xj=0 m Xi=−m ∞ Xp=−m p Xi=−m s(i) p−i! zp. Since i ≤ p in the internal sum, the last belongs to the subspace Fp. (cid:3) Remark. The flags of the form (3.2) already occured in [3] in the con- text of infinitesimal parabolic structures. In contrast to any parabolic structure our flag distributions appear as an intrinsic structure for a matrix divisor and the corresponding holomorphic vector bundle. Let D be a nonnegative divisor, Π = support D, Π ∩ Γ = ∅, and gl(Ψ) = {f ∈ ΓV (Ψ) (f ) + D + mΓ ≥ 0, f is global}. ΓD Corollary 3.4. dim ΓD ΓD gl(Ψ) is trivial unless deg D ≥ g. gl(Ψ) = dim V (deg D − g + 1), in particular Proof. Let F D+mΓ denote the space of global sections f satisfying the condition (f ) + D + mΓ ≥ 0, and lD+mΓ = dim F D+mΓ. By the Riemann -- Roch theorem lD+mΓ = (dim V )(deg D + mΓ − g + 1). How- ever, the space of sections has a codimension in F D+mΓ coming from the conditions fs ∈ F γ s , s = −m, . . . , m are the flag subspaces at γ. The contribution of every γ ∈ Γ to the codimension is equal to s . By symmetry of the grading (3.1) the last is equal s=−m codimV F γ s where F γ Pm MATRIX DIVISORS 10 to m dim V . The total codimension is equal to (m dim V )Γ. Hence the rest of dimension is equal to (dim V )(deg D − g + 1), and it should be deg D − g + 1 > 0 for non-triviality of the space of sections. (cid:3) The highest weight χ, and the tuple h = {hγ ∈ L∗ V γ ∈ Γ} are discrete invariants of a divisor. The matrix divisors also have moduli coming from the K-components of their canonical forms at the points in Γ, and from elements γ ∈ Γ themselves. Below we introduce two types of equivalence of matrix divisors, and the corresponding moduli spaces. Definition 3.5. Two matrix divisors are equivalent if they have the same sheaf of sections (up to the common left shift by a constant (in z and γ) element of G). Remark. This equivalence is different from that given in [24, 25] for the purpose of classificaton of the holomorphic vector bundles. Following the last, two matrix divisors are equivalent if one of them can be taken to another by right multiplication by a (global) meromorphic G-valued function. In particular, the divisors equivalent in this sense may have different support. Lemma 3.6. Two matrix divisors are equivalent in sense of Defini- tion 3.5 if, and only if, they have the same flag configuration (up to the common left shift by a constant (in z and γ) element of G), in particular, the same Γ. Proof. Immediately follows from Lemma 2.5. (cid:3) We define a flag set in the same way as a flag configuration except that we relax the requirement that Γ is fixed. Then our second equiv- alence relation is as follows. Definition 3.7. Two matrix divisors are equivalent if they have the same flag set (up to the common left shift by a constant (in z and γ) element of G). Denote the moduli space of matrix divisors with given invariants χ, h = {hγ ∈ Q∗ ∩ LV γ ∈ Γ} by Mχ h (it corresponds to the equivalence given by Definition 3.7), and with additionally fixed Γ by Mχ,Γ (it corresponds to the equivalence given by Definition 3.5). Denote the part of the moduli space over a point γ ∈ Γ by Mχ h hγ . It is our next step to represent Mχ,Γ h as a homogeneous space and describe the stationary group of a point of this space. Given a matrix divisor, consider the corresponding h = {hγ ∈ L∗ Γ}. Observe that for a faithful g-module V one has L∗ V γ ∈ V ⊆ Q∗ [15, MATRIX DIVISORS 11 Lemma 27]. Hence α(hγ) ∈ Z for every α ∈ R, γ ∈ Γ. For every γ the hγ defines a grading g = gγ k, and the corresponding increasing filtration gγ −k ⊕ . . . ⊕ gγ k = g where −k ⊂ . . . ⊂ gγ gγ i = Mα(hγ )≤i gα, gγ i =Ms≤i gγ s . gγ gα, i = Mα(hγ )=i Pj=−k ∞ In the next section we give equivalent definitions to these objects. Let gγ = {L(z) = Ljzj Lj ∈ gγ j , j = −k, . . . , ∞}. Let Gγ ⊂ G(C(z, z−1)) be the subgroup with the Lie algebra gγ, Kγ = K ∩ Gγ, Kγ. K0 = Qγ∈Γ Proposition 3.8. Mχ,Γ h = K × . . . × K /K0. Proof. By Lemma 2.7 Qγ∈Γ Γtimes {z } ∩ Qγ∈Γ Γtimes {z } Gγ is exactly the stationary subgroup of the given flag distribution in the group of all G-distributions. The propo- sition follows by K0 = K × . . . × K (cid:3) Gγ. Theorem 3.9. The tangent space to Mχ,Γ ments of the form h at the unit consists of ele- Mγ∈Γ Xα∈R+ aγ α(z)xα, aγ α(z) ∈ C(z)/zα(hγ )C(z). Proof. Since Mχ is the Lie algebra of the group K considered over C. hγ = K/Kγ, we have TeMχ hγ = k(z)/k(z) ∩ gγ where k Observe that gγ can be characterized as the subalgebra in g(z, z−1) consisting of elements of the form L(z) = Xα∈R, i≥α(hγ ) xαzi. gα. Hence the Indeed, Li = Ps≤i Ls i where Ls i ∈ gs, and gs = Lα(hγ )=s terms xαzi, i ≥ α(hγ) are absent in the quotient k/k(z) ∩ gγ. In par- ticular, the whole lower triangle subalgebra is filtered out because it is generated by xα, α ∈ R−, hence only xαzi with i < 0 could be in the remainder (i < α(hγ) ≤ 0), but there is no negative degrees in k(z) (moreover, the whole lower parabolic subalgebra gγ 0 is filtered out for the same reason). Thus we are left with only upper nilpotents, and their exponents exactly give elements in the form claimed in the statement of the theorem. (cid:3) MATRIX DIVISORS 12 Corollary 3.10. 1◦. dim Mχ hγ = s dim gγ s . 2◦. dim Mχ,Γ s dim gγ s (Γ is fixed). k Ps=1 3◦. dim Mχ s dim gγ s ) (Γ is not fixed). where k is the depth of the above defined grading. k Ps=1 (1 + h = Pγ∈Γ h = Pγ∈Γ k Ps=1 hγ = Pα∈R+ Proof. Indeed, for every α ∈ R+ the number of moduli is equal to α(hγ), hence dim Mχ s = α(hγ). Next, gγ gα, i.e. dim gγ ♯{α α(hγ) = s}. Since dim gα = 1 we obtain the first and second claims. 3◦ follows from 2◦ taking account of elements γ ∈ Γ which are also counted as moduli. (cid:3) s = Lα(hγ )=s Examples. 1) For G = GL(n), V = Cn, hγ = diag(d1, . . . , dn) Theorem 3.9 gives aij ∈ k(z)/zdi−dj k(z) for the entry aij of an element of the tangent space. These are the conditions claimed in [24, 25] for the canonical form of any divisor. Here they have a different meaning. Assume hγ = diag(1, 0, . . . , 0) for every γ. Then the corresponding grading is as follows: g = g−1 ⊕ g0 ⊕ g1 where dim g±1 = n − 1 [22], i.e. according to Corollary 3.10.3◦ every point contributes (n − 1) + 1 = n parameters into the dimension of the moduli space. If Γ = ng then the total number of parameters is equal to n2g. Observe that Ad G is a subgroup of the group of equivalencies of any matrix divisor because the right multiplication by g−1 (g ∈ G) is a multiplication by a global (constant) function while the left multiplication by g is a multiplication by a K-valued distribution. Taking quotient by Ad G kills n2 − 1 parameters, and we conclude that dim M = n2(g − 1) + 1 which coincides with the known dimension of the moduli space of holomorphic rank n vector bundles on Σ. 2) Consider G = SO(2n), V = C2n, hγ = diag(1, 0, . . . , 0) for ev- ery γ. Then again g = g−1 ⊕g0 ⊕g1 where dim g±1 = 2n−2 [22]. Hence every γ ∈ Γ contributes 2n − 1 into dimension of the moduli space. If Γ = ng then dim M = (2n − 1)ng − dim G = (2n − 1)n(g − 1) (the subtracting of dim G = (2n−1)n is due to the action of Ad G as above). 3) Let G = Sp(2n), V = C2n, hγ = diag(1, 0, . . . , 0) for every γ. Then g = g−2 ⊕ g−1 ⊕ g0 ⊕ g1 ⊕ g2 where dim g±1 = 2n − 2, dim g±2 = 1 [22]. By Corollary 3.10.2◦ every point γ ∈ Γ contributes 2 · 1 + (2n − 2) + 1 = 2n + 1 into the dimension of the moduli space. If Γ = ng then dim M = (2n + 1)ng − dim G = (2n + 1)n(g − 1) (again the MATRIX DIVISORS 13 subtracting of dim G is due to the action of Ad G as above but this time dim G = (2n + 1)n). In all considered cases the dimension of the space of matrix divi- sors coincides with the dimension of the corresponding moduli space of semi-stable G-bundles. To obtain the same result for SL(n) and G = SO(2n + 1) we would need consider the matrix divisors with val- ues in the conformal extensions of the corresponding groups (see the Introduction). 4. Moduli of matrix divisors and Lax operator algebras Here we will establish an isomorphism of the tangent space at the unit to the moduli space of matrix divisors (with given discrete in- variants) with the quotient of the space of M-operators by the Lax operator algebra (basically defined by the same invariants). The result goes back to [9]. For brevity, we consider only Chevalley groups with a trivial center here. We start with the definition of a Lax operator algebra. Let g be a semi-simple Lie algebra over C, h be its Cartan subalgebra, and h ∈ h be such element that pi = αi(h) ∈ Z+ for every simple root αi of g. If we denote the root lattice of g by Q then h belongs to the positive chamber of the dual lattice Q∗. Let gp = {X ∈ g (ad h)X = pX}, and k = max{p gp 6= 0}. Then the decomposition g = gp gives a Z-grading on g. For the theory k Li=−k and classification results on such kind of gradings we refer to [28]. Call gα where R is the root k a depth of the grading. Obviously, gp = Lα∈R α(h)=p system of g. Define also the following filtration on g: gp = gq. Then gp ⊂ gp+1 (p ≥ −k), g−k = g−k, . . . , gk = g, gp = g, p > k. As above, let Σ be a complex compact Riemann surface with two given non-intersecting finite sets of marked points: Π and Γ. Assume every γ ∈ Γ to be assigned with an hγ ∈ Q∗ +, and the corresponding grading and filtration. We equip the notation gp, gp with the upper γ indicating that the grading (resp. filtration) subspace corresponds to γ. We stress that gγ p are the same as defined in the previous section. Let L be a meromorphic mapping Σ → g holomorphic outside the marked points which may have poles of an arbitrary order at the points p, gγ p Lq=−k MATRIX DIVISORS 14 in Π, and has the decomposition of the following form at every γ ∈ Γ: (4.1) L(z) = Lpzp, Lp ∈ gγ p ∞ Xp=−k where z is a local coordinate in the neighborhood of γ. For simplicity, we assume that the depth of grading k is the same all over Γ, though it would be no difference otherwise. We denote by L a linear space of all such mappings. Since the relation (4.1) is preserved under commutator, L is a Lie algebra called Lax operator algebra. Lax operator algebras have emerged in [10] due to the observation by I.Krichever [9] that the Lax operators of integrable systems with the spectral parameter on a Riemann surface have a very special Laurent expansions related to the Tyurin parameters of holomorphic vector bundles. In [17, 18] they were generalized to the form described here. For the current state of the theory of Lax operator algebras and their applications to integrable systems we refer to [22, 16, 17, 18, 19, 20, 21] and references therein. To give the Lax operator algebra description of the moduli space of matrix divisors introduce the space of M-operators, the counterparts of Lax operators in the Lax pairs of integrable systems. A meromorphic mapping M : Σ → g, holomorphic outside Π and Γ, is called an M-operator if at any γ ∈ Γ it has a Laurent expansion (4.2) M(z) = νγhγ z + Mizi ∞ Xi=−k where Mi ∈ gγ i for i < 0, Mi ∈ g for i ≥ 0, and νγ ∈ C. We denote by MΠ,Γ,h the space of M-operators corresponding to given data Π, Γ, h (as above, h = {hγ γ ∈ Γ}). Obviously, LΠ,Γ,h ⊂ MΠ,Γ,h. Let (L)Π and (M)Π be the restrictions of the corresponding divisors to Π. For an non-negative divisor D introduce LD MD Π,Γ,h = {L ∈ LΠ,Γ,h (L)Π + D ≥ 0}, Π,Γ,h = {M ∈ MΠ,Γ,h (M)Π + D ≥ 0}. For a Lax operator algebra LΠ,Γ,h, the mapping LΠ,Γ,h → Lγ∈Γ taking any L ∈ LΠ,Γ,h to the set of its Laurent expansions at the points in Γ is called localization. gγ Theorem 4.1. For the tangent space at the unit to the moduli space of matrix divisors we have (4.3) TeMχ,Γ h ∼= MD Π,Γ,h/LD Π,Γ,h, MATRIX DIVISORS 15 independently of Π, χ and D, provided deg D < Γ. The isomorphism is given by the localization map. Proof. For deg D < Γ the localization map is injective for the rea- son that an element in the kernel would have Γ zeroes and no more than deg D < Γ poles, hence it is trivial. At any point γ ∈ Γ the main parts of the Laurent expansions for L and M operators satisfy the same conditions, hence vanish in the quotient MΠ,Γ,h/LΠ,Γ,h. As for the Tailor parts, for i = 0, . . . , k − 1 the coefficients at zi in the quotient run over g/gi = gi+1 ⊕ . . . ⊕ gk (and vanish for i ≥ k). Hence MD the quotient of localizations on the right hand side of the relation. It is gα. Hence so where it is exactly i=1 Lizi Li ∈Li<s gγ similar to the proof of Corollary 3.10 thatLi<s gγ s = Li<α(hγ ) i=1 Lizi Li ∈Li<α(hγ ) gαo. The last space Π,Γ,h = Lγ∈ΓnPk Π,Γ,h =Lγ∈ΓnPk Π,Γ,h/LD Π,Γ,h/LD MD coincides with TeMχ,Γ h by Theorem 3.9. (cid:3) Remark. The independence of χ in the theorem is related to the trivi- ality of the centers of the Chevalley groups in question (see remarks in the beinning of the section, and in Introduction). Remark. In general, the right hand side of (4.3) depends on D. Theo- rem 4.1 may be reformulated using the space Mχ,Γ,D corresponding to modified equivalence of matrix divisors. Namely, two matrix divisors will be equivalent if their spaces ΓD gl(Ψ) (cf. Corollary 3.4) are non- trivial and equal up to the shift by a constant element of the group. h We note the similarity with [24, Lemma 5] which reduces the clas- sification problem for degree 0 framed bundles to the corresponding problem for effective framed bundles by tensoring the first with two linear bundles of degrees g − 1 and 1, respectively. It is the same as to assume the existence of an external (with respect to Γ) degree g divisor D. To stress the similarity we note that for a semisimple group G all G-bundles have degree 0, and that by Corollary 3.4 ΓD gl(Ψ) is non-trivial if deg D ≥ g. References [1] M.F.Atiyah. Vector bundles over an elliptic curve, Proc. London Math. Soc., V. 7 (1957), 414-452. [2] A.Grothendieck. Sur la classification des fibres holomorphes sur la sph`ere de Riemann, Amer. J. Math. 79 (1957), 121-138. [3] G.Faltings. Stable G-bundles and projective connections, J. Algebraic Geometry, V. 2 (1993), 507-568. MATRIX DIVISORS 16 [4] Feigin, B.L. Integrable systems, shaffle algebras and the Bethe equations. Trans. Moscow Math. Soc. 2016, 203-246. [5] Hitchin, N. Stable bundles and integrable systems, Duke Math. J. 54:1 (1987), 91-114. [6] Hitchin, N. The self-duality equations on a Riemann surface, Proc. Lond. Math. Soc., 55(3):1, 1987, 59-126. [7] Krichever I.M., Novikov S.P. Holomorphic bundles over Riemann surfaces and the Kadomtsev -- Petviashvili equation. I. Funct. Anal. and Appl., 4, 12 (1978), 276 -- 286. [8] Krichever I.M., Novikov S.P. Holomorphic bundles over algebraic curves and non-linear equations. Russ. Math. Surv., 6, 35 (1980), 53 -- 79. [9] Krichever, I.M. Vector bundles and Lax equations on algebraic curves. Comm. Math. Phys. 229, 229 -- 269 (2002). hep-th/0108110. [10] Krichever, I.M., Sheinman, O.K. Lax operator algebras. Funct. Anal. i Prilozhen., 41 (2007), no. 4, p. 46-59. math.RT/0701648. [11] Levin A., Olshanetsky M., Smirnov A., Zotov A. Characteristic Classes and Integrable Systems. General Construction, arXiv 1006.0702. [12] Mehta, V.B., Seshadri, C.S. Moduli of vector bundles on urves with parabolic structures. Math.Ann., V. 248, p. 205 -- 239 (1980). [13] Narasimhan, M.S., Seshadri, C.S. Stable and unitary vector bundles on a com- pact Riemann surface. Ann.Math., V. 82, 540-564, 1965. [14] Seshadri, C.S. Moduli of vector bundles with parabolic structures, Bull. Amer. Math. Soc. 83:1 (1977), 124-126. [15] Steinberg, R. Lectures on Chevalley groups. Yale University, 1967. [16] Sheinman, O.K. Current algebras on Riemann surfaces, De Gruyter Exposi- tions in Mathematics, 58, Walter de Gruyter GmbH & Co, Berlin-Boston, 2012, ISBN: 978-3-11-026452-4, 150 pp. [17] Sheinman, O.K. Lax operator algebras and gradings on semi-simple Lie alge- bras. Doklady Mathematics, V. 461, no. 2, 2015, 143 -- 145. [18] Sheinman, O.K. Lax operator algebras and gradings on semi-simple Lie algebras. Transformation Groups, Vol. 21, No. 1, 2016, 181 -- 196. DOI 10.1007/s00031-015-9340-y. [19] Sheinman, O.K. Hierarchies of finite-dimensional Lax equations with a spectral parameter on a Riemann surface, and semi-simple Lie algebras. Theoret. and Math. Phys., 185:3 (2015), 1816 -- 1831. [20] Sheinman, O.K. Semi-simple Lie algebras and Hamiltonian theory of finite- dimensional Lax equations with the spectral parameter on a Riemann surface. Proc. Steklov Inst. Math., 290 (2015), 178 -- 188. [21] Sheinman,O.K. Global current algebras and localization on Riemann surfaces, Moscow Math. Journ., V. 15, n. 4, (2015), p. 833-846. [22] Sheinman,O.K. Lax operator algebras and integrable systems. Russian Math. Surveys, 71:1 (2016), 109 -- 156. [23] Tyurin, A.N. On the classification of rank 2 vector bundles over an alge- braic curve of arbitrary genus, Izv. AN SSSR. Ser. Math. (1964) 28:1 p. 2152. Also published in: Tyurin, A.N., Vector Bundles, Collected Works, V.1, F.Bogomolov, A.Gorodentsev, V.Pidstrigach, M.Reid, and N.Tyurin eds., Gottingen 2008 MATRIX DIVISORS 17 [24] Tyurin, A.N. Classification of vector bundles over an arbitrary genus algebraic curve. Amer. Math. Soc. Transl. Ser. 2, 63 (1967), 245-279. [25] Tyurin, A.N. On classification of rank n vector bundles over an arbitrary genus algebraic curve. Amer. Math. Soc. Transl. Ser. 2, 73 (1968), 196 -- 211. [26] Tyurin, A.N. On the classifications of rank 2 vector bundles on an algebraic curve of an arbitrary genus., Izv. AN SSSR, ser. math. 28:1 (1964), 21-52 (see also Collected papers). [27] Tyurin, A.N. Geometry of modules of vector bundles, Russ Math. Surv., 29:6(180) (1974), 59-88 (see also Collected papers). [28] E.B.Vinberg, V.V.Gorbatsevich, A.L.Onischik. Structure of Lie groups and Lie algebras. In: Itogi nauki i techniki, vol. 41, pp. 5-258. [29] A.Weil. Generalisation des fonctions abeliennes, J.Math. Pures et Appl., v.17 (1938), 47-87.
1410.3280
15
1410
2016-06-04T13:32:19
Some results of algebraic geometry over Henselian rank one valued fields
[ "math.AG" ]
We develop geometry of affine algebraic varieties in $K^{n}$ over Henselian rank one valued fields $K$ of equicharacteristic zero. Several results are provided including: the projection $K^{n} \times \mathbb{P}^{m}(K) \to K^{n}$ and blow-ups of the $K$-rational points of smooth $K$-varieties are definably closed maps, a descent property for blow-ups, curve selection for definable sets, a general version of the \L{}ojasiewicz inequality for continuous definable functions on subsets locally closed in the $K$-topology and extending continuous hereditarily rational functions, established for the real and $p$-adic varieties in our joint paper with J. Koll\'ar. The descent property enables application of resolution of singularities and transformation to a normal crossing by blowing up in much the same way as over the locally compact ground field. Our approach relies on quantifier elimination due to Pas and a concept of fiber shrinking for definable sets, which is a relaxed version of curve selection. The last three sections are devoted to the theory of regulous functions and sets over such valued fields. Regulous geometry over the real ground field $\mathbb{R}$ was developed by Fichou--Huisman--Mangolte--Monnier. The main results here are regulous versions of Nullstellensatz and Cartan's Theorems A and B.
math.AG
math
ALGEBRAIC GEOMETRY OVER HENSELIAN RANK ONE VALUED FIELDS KRZYSZTOF JAN NOWAK Abstract. This paper develops algebraic geometry over Henselian rank one valued fields K of equicharacteristic zero. Several re- sults are provided including: the projection K n × Pm(K) → K n and blow-ups of the K-points of smooth K-varieties are definably closed maps; a descent property for blow-ups; a general version of the Lojasiewicz inequality for continuous definable functions on subsets locally closed in the K-topology; curve selection for defin- able sets and extending continuous hereditarily rational functions, established for the real and p-adic varieties in our joint paper with J. Koll´ar. The descent property enables application of desingular- ization and transformation to a normal crossing by blowing up in much the same way as over the locally compact ground field. Our approach relies on quantifier elimination due to Pas and a concept of fiber shrinking for definable sets, which is a relaxed version of curve selection. The last three sections are devoted to the theory of regulous functions and sets over such valued fields. Regulous geometry over the real ground field R was developed by Fichou– Huisman–Mangolte–Monnier. The main results here are regulous versions of Nullstellensatz and Cartan’s Theorems A and B. 1. Introduction We develop algebraic geometry over Henselian rank one valued fields K of equicharacteristic zero with valuation v, valuation ring R and residue field k. We shall always assume that the value group Γ ⊂ R contains 1. Every rank one valued field has a metric topology induced by its absolute value. Examples of such fields are the quotient fields of the rings of formal power series and of Puiseux series with coefficients 2000 Mathematics Subject Classification. Primary: 14G27, 03C10; Secondary: 12J25, 14P10. Key words and phrases. Henselian rank one valued fields, topological fields with density property, quantifier elimination, cell decomposition, restricted for- mal power series, Puiseux’s theorem, fiber shrinking, descent property for blow- ups, Lojasiewicz inequality, curve selection, hereditarily rational functions, regulous functions and sets, Nullstellensatz, Cartan’s Theorems A and B.. 1 2 KRZYSZTOF JAN NOWAK from a field k of characteristic zero and the field of generalized formal power series k((tΓ)) :=(f (t) =Xγ∈Γ aγtγ : aγ ∈ k, supp f (t) is well ordered) . The K-points X(K) of any K-algebraic variety X inherit from K a topology, called the K-topology. Several results concerning algebraic geometry over such ground fields K are established. Let L be the 3-sorted language of Denef–Pas. We prove that the projection K n × Pm(K) → K n is an L-definably closed map (Theorem 3.1). Further, we shall draw several conclusions, including the theorem that blow-ups of the K- points of smooth K-varieties are definably closed maps, a descent prop- erty for such blow-ups, a general version of the Lojasiewicz inequal- ity for continuous L-definable functions on subsets locally closed in the K-topology (Theorem 9.2) and curve selection for L-definable sets (Theorem 8.2) and for (valuative) semialgebraic sets (Theorem 8.1). Also given is the theorem on extending continuous hereditarily ratio- nal functions over such ground fields (Theorem 10.2), established for the real and p-adic varieties in our joint paper [24] with J. Koll´ar. The proof makes use of the descent property and of the Lojasiewicz inequality. The descent property enables application of desingulariza- tion and transformation to a normal crossing by blowing up in much the same way as over the locally compact ground field. Our approach relies on quantifier elimination due to Pas and on a certain concept of fiber shrinking for definable sets, which is a relaxed version of curve selection. Let us mention that this paper comprises our two earlier preprints [36, 37]. We should emphasize that our approach to the subject of this paper is possible just because the language L in which we investigate valued fields is not too rich; in particular, it does not contain the inclusion language on the auxiliary sorts and the only symbols of L connecting the sorts are two functions from the main K-sort to the auxiliary Γ-sort and k-sort. Hence and by elimination of K-quantifiers, the L-definable subsets of the products of the two auxiliary sorts are precisely finite unions of the Cartesian products of sets definable in those two sorts. This allows us to reduce our reasonings to an analysis of ordinary cells (i.e. fibers of a cell in the sense of Pas). The organization of the paper is as follows. In Section 2, we set up notation and terminology including, in particular, the language L ALGEBRAIC GEOMETRY OVER VALUED FIELDS 3 of Denef–Pas and the concept of a cell. We recall the theorems on quantifier elimination and on preparation cell decomposition, due to Pas [38]. When the value group Γ is divisible, we prove a theorem on cell decomposition for L-definable sets (Theorem 2.4). When Γ is arbitrary, what is crucial for our proof of the closedness theorem is a weak version of cell decomposition to the effect that every L-definable subset of K n+1 is a finite union of sets each of which is a subset of a cell determined by finitely many congruences (Theorem 2.6). In Section 3, we state the closedness theorem and several direct corollaries, including the descent property. In the case where the value group Γ is discrete, Section 4 gives a proof of this theorem which is based on an algorithm. In Section 5, we study L-definable functions of one variable. A result playing an important role in the sequel is the theorem on existence of the limit (Theorem 5.3). Its proof makes use of Puiseux’s theorem for the local ring of convergent power series. In Section 6, we introduce a certain concept of fiber shrinking for L- definable sets (Theorem 6.1). Section 7 provides a proof of the closed- ness theorem (Theorem 3.1) for the general case. This proof makes use of fiber shrinking and existence of the limit for functions of one variable (Theorem 5.3). In the subsequent three sections, some further conclusions from the closedness theorems are drawn. Section 8 provides some versions of curve selection, a general one for arbitrary L-definable sets and one for (valuative) semialgebraic sets. The next section is devoted to a general version of the Lojasiewicz inequality for continuous L-definable functions on subsets locally closed in the K-topology (Theorem 9.2). In Section 10, the theorem on extending continuous hereditarily rational functions, established for the real and p-adic varieties in [24], is carried over to the case where the ground field K is a Henseliam rank one valued field of equicharacteristic zero. Finally, let us mention that in real algebraic geometry applications of hereditarily rational functions and the extension theorem, in particular, are given in papers [25, 26, 27] and [28], which discuss rational maps into spheres and stratified- algebraic vector bundles on real algebraic varieties. The last three sections are devoted to the theory of regulous functions and sets over Henselian rank one valued fields of equicharacteristic zero. Regulous geometry over the real ground field R was developed by Fichou–Huisman–Mangolte–Monnier [15]. In Section 11 we set up notation and terminology as well as provide basic results about regulous functions and sets, including the noetherianity of the constructible and 4 KRZYSZTOF JAN NOWAK regulous topologies. Those results are valid over arbitrary fields with the density property. The next section establishes a regulous version of Nullstellensatz (Theorem 12.4), valid over Henselian rank one valued fields of equicharacteristic zero. The proof relies on the Lojasiewicz inequality (Theorem 9.2). Also drawn are several conclusions, including the existence of a one-to-one correspondence between the radical ideals of the ring of regulous functions and the closed regulous subsets, or one-to-one correspondences between the prime ideals of that ring, the irreducible regulous subsets and the irreducible Zariski closed subsets (Corollaries 12.5 and 12.10). Section 13 provides an exposition of the theory of quasi-coherent reg- ulous sheaves, which generally follows the approach given in the real algebraic case by Fichou–Huisman–Mangolte–Monnier [15]. It is based on the equivalence of categories between the category of eRk-modules and the category of Rk-modules which, in turn, is a consequence of the one-to-one correspondences mentioned above. The main results here are the regulous versions of Cartan’s Theorems A and B. We also establish a criterion for a continuous function on an affine regulous sub- variety to be regulous (Theorem 13.10), which relies on our theorem on extending continuous hereditarily rational functions (Theorem 10.2). Finally, note that the metric topology of a non-archimedean field K with a rank one valuation v is totally disconnected. Rigid analytic geometry (see e.g. [6] for its comprehensive foundations), developed by Tate, compensates for this defect by introducing sheaves of ana- lytic functions in a Grothendieck topology. Another approach is due to Berkovich [3], who filled in the gaps between the points of K n, pro- ducing a locally compact Hausdorff space (the so-called analytification of K n), which contains the metric space K n as a dense subspace if the ground field K is algebraically closed. His construction consists in re- placing each point x of a given K-variety with the space of all rank one valuations on the residue field κ(x) that extend v. The theory of stably dominated types, developed in paper [21], deals with non-archimedean fields with valuation of arbitrary rank and generalizes that of tame topology for Berkovich spaces. Currently, various analytic structures over Henselian rank one valued fields are intensively investigated (see e.g. [11, 12] for more information, and [31] for the case of algebraically closed valued fields). ALGEBRAIC GEOMETRY OVER VALUED FIELDS 5 2. Quantifier elimination and cell decomposition We begin with quantifier elimination due to Pas in the language L of Denef–Pas with three sorts: the valued field K-sort, the value group Γ-sort and the residue field k-sort. The language of the K-sort is the language of rings; that of the Γ-sort is any augmentation of the language of ordered abelian groups (and ∞); finally, that of the k-sort is any augmentation of the language of rings. We denote K- sort variables by x, y, z, . . ., k-sort variables by ξ, ζ, η, . . ., and Γ-sort variables by k, q, r, . . .. In the case of non-algebraically closed fields, passing to the three sorts with additional two maps: the valuation v and the residue map, is not sufficient. Quantifier elimination due to Pas holds for Henselian valued fields of equicharacteristic zero in the above 3-sorted language with additional two maps: the valuation map v from the field sort to the value group, and a map ac from the field sort to the residue field (angular component map) which is multiplicative, sends 0 to 0 and coincides with the residue map on units of the valuation ring R of K. Not all valued fields have an angular component map, but it exists whenever the valued field has a cross section or the residue field is ℵ1- saturated (cf. [39]). In general, unlike for p-adic fields and their finite extensions, adding an angular component map does strengthen the fam- ily of definable sets. For both p-adic fields (Denef [13]) and Henselian equicharacteristic zero valued fields (Pas [38]), quantifier elimination was established by means of cell decomposition and a certain prepa- ration theorem (for polynomials in one variable with definable coeffi- cients) combined with each other. In the latter case, however, cells are no longer finite in number, but parametrized by residue field variables. In the proof of the closedness theorem (which is a fundamental tool for many other results of this paper), we may use an angular component map, because we may always assume — via the transfer principle of Ax–Kochen–Ershov (see e.g. [9]) — that the residue field k is suffi- ciently saturated. Finally, let us mention that quantifier elimination based on the sort RV := K ∗/(1 + m) ∪ {0} (where K ∗ := K \ {0} and m is the maximal ideal of the valuation ring R) was introduced by Besarab [4]. This new sort binds together the value group and residue field into one structure. In paper [20, Section 12], quantifier elimination for Henselian valued fields of equicharacteristic zero, based on this sort, was derived directly from that by Robinson for algebraically closed valued fields. Yet another, more general result, including Henselian valued fields of 6 KRZYSZTOF JAN NOWAK mixed characteristic, was achieved by Cluckers–Loeser [10] for so-called b-minimal structures (from ”ball minimal”); in the case of valued fields, however, countably many sorts RVn := K ∗/(1 + nm) ∪ {0}, n ∈ N, are needed. Below we state the theorem on quantifier elimination due to Pas [38, Theorem 4.1]. Theorem 2.1. Let (K, Γ, k) be a structure for the 3-sorted language L of Denef–Pas. Assume that the valued field K is Henselian and of equicharacteristic zero. Then (K, Γ, k) admits elimination of K- quantifiers in the language L. ✷ We immediately obtain the following Corollary 2.2. The 3-sorted structure (K, Γ, k) admits full elimination of quantifiers whenever the theories of the value group Γ and the residue field k admit quantifier elimination in the languages of their sorts. ✷ We now recall the necessary notation and terminology concerning cell decomposition. Consider an L-definable subset D of K n × km, three L-definable functions a(x, ξ), b(x, ξ), c(x, ξ) : D → K and a positive integer ν. For each ξ ∈ km set C(ξ) := {(x, y) ∈ K n x × Ky : (x, ξ) ∈ D, v(a(x, ξ)) ✁1 v((y − c(x, ξ))ν) ✁2 v(b(x, ξ)), ac(y − c(x, ξ)) = ξ1}, where ✁1, ✁2 stand for <, ≤ or no condition in any occurrence. If the sets C(ξ), ξ ∈ km, are pairwise disjoint, the union C := [ξ∈km C(ξ) is called a cell in K n × K with parameters ξ and center c(x, ξ); C(ξ) is called a fiber of the cell C. Theorem 2.3. (Preparation Cell Decomposition, [38, Theorem 3.2]) Let f1(x, y), . . . , fr(x, y) be polynomials in one variable y with coefficients being L-definable func- tions on K n x . Then K n × K admits a finite partition into cells such that on each cell C with parameters ξ and center c(x, ξ) and for all i = 1, . . . , r we have: v(fi(x, y)) = v(cid:16) fi(x, ξ)(y − c(x, ξ))νi(cid:17) , ALGEBRAIC GEOMETRY OVER VALUED FIELDS 7 ac fi(x, y) = ξµ(i), where fi(x, ξ) are L-definable functions, νi ∈ N for all i = 1, . . . , r, and the map µ : {1, . . . , r} → {1, . . . , m} does not depend on x, y, ξ. We then say that the the functions f1(x, y), . . . , fr(x, y) are prepared with respect to the variable y. ✷ From now on, we shall assume, unless otherwise stated, that the ground field K is a Henselian rank one valued field of equicharacteristic zero. An arbitrary subgroup Γ ⊂ R, Γ 6= {0}, of the additive group of real numbers admits quantifier elimination in the language indicated below. • When Γ is divisible, it admits quantifier elimination in the language of ordered groups. • When Γ is discrete, we may assume that Γ = Z. We consider the Presburger language (<, +, −, 0, 1) with binary relation symbols ≡n for congruences modulo n, n ∈ N, n > 1. • When Γ is not discrete, we consider the language (<, +, −, 0) with binary relation symbols ≡n for congruences modulo n, n ∈ N, n > 1, defined as follows. For any k, r ∈ Γ, k ≡n r ⇔ k − r = nq for some q ∈ Γ. Note that if the value group Γ is divisible, the each congruence ≡n is satisfied by every pair of its elements. Combining theorems 2.1 and 2.3 with quantifier elimination in the Γ-sort, we obtain the following Theorem 2.4. (Cell decomposition) If, in addition, the value group Γ is divisible, then every L-definable subset B of K n × K is a finite union of cells. Proof. By elimination of K-quantifiers, B is a finite union of sets de- fined by conditions of the form (v(f1(x, y)), . . . , v(fr(x, y))) ∈ P, (ac g1(x, y), . . . , ac gs(x, y)) ∈ Q, where fi, gj ∈ K[x, y] are polynomials, and P and Q are definable subsets of Γr and ks, respectively (since a = 0 iff ac a = 0). Thus we may assume that B is such a set. The sort Γ admits quantifier elimination in the language of ordered groups. Therefore we can assume that the set P is defined by finitely 8 KRZYSZTOF JAN NOWAK many inequalities of the form {α ∈ Γr : k1α1 + · · · + krαr + β ✁ 0}, where k1, . . . , kr ∈ Z, β ∈ Γ and ✁ stands for < or =. But there exists a finite partition of K n×K into cells, which prepares the polynomials f1(x, y), . . . , fr(x, y) and g1(x, y), . . . , gs(x, y). On each cell C of this partition (of the form considered before), we thus have ac fi(x, y) = ξµ(i), v(fi(x, y)) = v(cid:16) fi(x, ξ)(y − c(x, ξ))νi(cid:17) , v(gi(x, y)) = v(cid:0)gi(x, ξ)(y − c(x, ξ))ϑi(cid:1) , v rYi=1 fi(x, ξ)ki(y − c(x, ξ))kiνi! + β ✁ 0 ac gi(x, y) = ξη(i). and and Hence the intersection B ∩ C(ξ) is defined by finitely many conditions of the form (ξη(1), . . . , ξη(s)) ∈ Q. After renumbering the integers ki, we may assume that k1, . . . , kp, with p ≤ r, are non-negative integers and kp+1, . . . , kr are negative integers. Take b ∈ K such that v(b) = β. Then the conditions of the first form are equivalent to v b fi(x, ξ)ki(y − c(x, ξ))kiνi!✁ v rYi=p+1 pYi=1 fi(x, ξ)−ki(y − c(x, ξ))−kiνi! . We can therefore partition the cell C into a finite number of finer cells with the same center c(x, ξ), so that the intersection B ∩ C(ξ) is the union of some of them. This finishes the proof. (cid:3) Remark 2.5. A more careful analysis of cells in the above proof leads to the sharpening of Theorem 2.4 stated below: Every L-definable subset B of K n ×K can be partitioned into a finite union of cells. When the value group Γ is arbitrary, the method of the proof of Theorem 2.4 yields immediately the following weaker version of cell decomposition. of a cell F := [ξ∈km C := [ξ∈km F (ξ) C(ξ) ALGEBRAIC GEOMETRY OVER VALUED FIELDS 9 Theorem 2.6. Every L-definable subset B of K n × K is a finite union of sets each of which is a subset determined by finitely many congruences: F (ξ) :=(cid:8)(x, y) ∈ C(ξ) : v(cid:0)fi(x, ξ)(y − c(x, ξ))ki(cid:1) ≡n 0, i = 1, . . . , s(cid:9) , where fi are L-definable functions, ki ∈ N for i = 1, . . . , s, and n ∈ N, n > 1. ✷ Remark 2.7. The above theorem will be applied to establish the closed- ness theorem (Theorem 3.1) in Section 7. 3. Closedness theorem One of the most fundamental theorems, on which many other results of this paper rely, is the following Theorem 3.1. (Closedness theorem) Let D be an L-definable subset of K n. Then the canonical projection π : D × Rm −→ D is definably closed in the K-topology, i.e. if B ⊂ D × Rm is an L- definable closed subset, so is its image π(B) ⊂ D. We shall provide two different proofs for this theorem. The first, given in Section 4, is valid whenever the value group Γ is discrete, and is based on an algorithm. The other, given in Section 7, is valid for the general case, and makes use of fiber shrinking from Section 6 which, in turn, relies on some results on L-definable functions of one variable from Section 5. When the ground field K is locally compact, the closedness theorem holds by a routine topological argument. We immediately obtain five corollaries stated below. Corollary 3.2. Let D be an L-definable subset of K n and Pm(K) stand for the projective space of dimension m over K. Then the canonical projection π : D × Pm(K) −→ D is definably closed. ✷ 10 KRZYSZTOF JAN NOWAK Corollary 3.3. Let A be a closed L-definable subset of Rm or of Pm(K). Then every continuous L-definable map f : A → K n is defin- ably closed in the K-topology. Corollary 3.4. Let φi, i = 0, . . . , m, be regular functions on K n, D be an L-definable subset of K n and σ : Y −→ KAn the blow-up of the affine space KAn with respect to the ideal (φ0, . . . , φm). Then the restriction σ : Y (K) ∩ σ−1(D) −→ D is a definably closed quotient map. Proof. Indeed, Y (K) can be regarded as a closed algebraic subvariety of K n × Pm(K) and σ as the canonical projection. (cid:3) Since the problem is local with respect to the target space, the above corollary immediately generalizes to the case where the K-variety Y is the blow-up of a smooth K-variety X. Corollary 3.5. Let X be a smooth K-variety, φi, i = 0, . . . , m, regular functions on X, D be an L-definable subset of X(K) and σ : Y −→ X the blow-up of the ideal (φ0, . . . , φm). Then the restriction σ : Y (K) ∩ σ−1(D) −→ D is a definably closed quotient map. ✷ Corollary 3.6. (Descent property) Under the assumptions of the above corollary, every continuous L-definable function g : Y (K) ∩ σ−1(D) −→ K that is constant on the fibers of the blow-up σ descends to a (unique) continuous L-definable function f : D −→ K. ✷ 4. Proof of Theorem 3.1 when the valuation is discrete The proof given in this section is based on an algorithm. Through the transfer principle of Ax–Kochen–Ershov (see e.g. [9]), it suffices to prove Theorem 3.1 for the case where the ground field K is a complete, discretely valued field of equicharacteristic zero. Such fields are, by virtue of Cohen’s structure theorem, the quotient fields K = k((t)) of formal power series rings k[[t]] in one variable t with coefficients from a field k of characteristic zero. The valuation v and the angular component ac of a formal power series are the degree and the coefficient of its initial monomial, respectively. ALGEBRAIC GEOMETRY OVER VALUED FIELDS 11 The additive group Z is an example of ordered Z-group, i.e. an or- dered abelian group with a (unique) smallest positive element (denoted by 1) subject to the following additional axioms: ∀ k k > 0 ⇒ k ≥ 1 and ∀ k ∃ q k = nq + r n−1_r=0 for all integers n > 1. The language of the value group sort will be the Presburger language of ordered Z-groups, i.e. the language of ordered groups {<, +, −, 0} augmented by 1 and binary relation symbols ≡n for congruence modulo n subject to the axioms: ∀ k, r k ≡n r ⇔ ∃ q k − r = nq for all integers n > 1. This theory of ordered Z-groups in the Pres- burger language has quantifier elimination and definable Skolem (choice) functions. The above two countable axiom schemas can be replaced by universal ones when we augment the language by adding the function n(cid:3) (of one variable k) for division with remainder, which fulfil symbols(cid:2) k the following postulates: (cid:20) k n(cid:21) = q ⇔ n−1_r=0 k = nq + r for all integers n > 1. The theory of ordered Z-groups admits there- fore both quantifier elimination and universal axioms in the Presburger language augmented by division with remainder. Thus every definable function is piecewise given by finitely many terms and, consequently, is piecewise linear. In the residue field sort, we can add new relation symbols for all definable sets and impose suitable postulates. This enables quantifier elimination for the residue field in the augmented language. In this fashion, we have full quantifier elimination in the 3-sorted structure (K, Z, k) with K = k((t)). Now we can readily pass to the proof of Theorem 3.1 which, of course, reduces easily to the case m = 1. So let B be an L-definable closed (in the K-topology) subset of D × Ry ⊂ K n x × Ry. It suffices to prove that if a lies in the closure of the projection A := π(B), then there is a point b ∈ B such that π(b) = a. Without loss of generality, we may assume that a = 0. Put Λ := {(v(x1), . . . , v(xn)) ∈ Zn : x = (x1, . . . , xn) ∈ A}. 12 KRZYSZTOF JAN NOWAK The set Λ contains points all coordinates of which are arbitrarily large, because the point a = 0 lies in the closure of A. Hence and by definable choice, Λ contains a set Λ0 of the form Λ0 = {(k, α2(k), . . . , αn(k)) ∈ Nn : k ∈ ∆} ⊂ Λ, where ∆ ⊂ N is an unbounded definable subset and α2, . . . , αn : ∆ −→ N are increasing unbounded functions given by a term (because a function in one variable given by a term is either increasing or decreasing). We are going to recursively construct a point b = (0, w) ∈ B with w ∈ R by performing the following algorithm. Step 1. Let Ξ1 := {(v(x1), . . . , v(xn), v(y)) ∈ Λ0 × N : (x, y) ∈ B}, and β1(k) := sup {l ∈ N : (k, α2(k), . . . , αn(k), l) ∈ Ξ1} ∈ N∪{∞}, k ∈ Λ0. If lim supk→∞ β1(k) = ∞, there is a sequence (x(ν), y(ν)) ∈ B, ν ∈ N, such that v(x(ν) 1 ), . . . , v(x(ν) n ), v(y(ν)) → ∞ when ν → ∞. Since the set B is a closed subset of D × Ry, we get (x(ν), y(ν)) → 0 ∈ B when ν → ∞, and thus w = 0 is the point we are looking for. Here the algorithm stops. Otherwise Λ1 × {l1} ⊂ Ξ1 for some infinite definable subset Λ1 of Λ0 and l1 ∈ N. The set {(v(x1), . . . , v(xn); ac(y)) ∈ Λ1 × k : (x, y) ∈ B, v(y) = l1} is definable in the language L. By elimination of K-quantifiers, it is given by a quantifier-free formula with variables only from the value group Γ-sort and the residue field k-sort. Therefore there is a finite partitioning of Λ1 into definable subsets over each of which the fibres of the above set are constant, because quantifier-free L-definable subsets of the product Zn × k of the two sorts are finite unions of the Cartesian products of definable subsets in Zn and in k, respectively. One of those definable subsets, say Λ′ 1, must be infinite. Consequently, for some ξ1 ∈ k, the set Ξ2 := {(v(x1), . . . , v(xn), v(y − ξ1tl1)) ∈ Λ′ contains points of the form (k, l) ∈ Nn+1, where k ∈ Λ′ 1 × N : (x, y) ∈ B} 1 and l > l1. ALGEBRAIC GEOMETRY OVER VALUED FIELDS 13 Step 2. Let β2(k) := sup {l ∈ N : (k, α2(k), . . . , αn(k), l) ∈ Ξ2} ∈ N∪{∞}, k ∈ Λ′ 1. If lim supk→∞ β2(k) = ∞, there is a sequence (x(ν), y(ν)) ∈ B, ν ∈ N, such that v(x(ν) 1 ), . . . , v(x(ν) n ), v(y(ν) − ξ1tl1) → ∞ when ν → ∞. Since the set B is a closed subset of D × Ry, we get (x(ν), y(ν)) → (0, ξ1tl1) ∈ B when ν → ∞, and thus w = ξ1tl1 is the point we are looking for. Here the algorithm stops. Otherwise Λ2 × {l2} ⊂ Ξ2 for some infinite definable subset Λ2 of Λ′ ξ2 ∈ k, the set 1 and l2 > l1. Again, for some Ξ3 := {(v(x1), . . . , v(xn), v(y − ξ1tl1 − ξ2tl2)) ∈ Λ′ 2 × N : (x, y) ∈ B} contains points of the form (k, l) ∈ Nn+1, where k ∈ Λ′ definable subset of Λ2 and l > l2. 2, Λ′ 2 is an infinite Step 3 is carried out in the same way as the previous ones; and so on. In this fashion, the algorithm either stops after a finite number of steps and then yields the desired point w ∈ R (actually, w ∈ k[t]) such that (0, w) ∈ B, or it does not stop and then yields a formal power series w := ξ1tl1 + ξ2tl2 + ξ3tl3 + . . . , 0 ≤ l1 < l2 < l3 < . . . such that for each ν ∈ N there exists an element (x(ν), y(ν)) ∈ B for which v(y(ν) − ξ1tl1 − ξ2tl2 − . . . − ξνtlν ) ≥ lν + 1 ≥ ν, v(x(ν) 1 ), . . . , v(x(ν) 1 ) ≥ ν. Hence v(y(ν) − w) ≥ ν, and thus the sequence (x(ν), y(ν)) tends to the point b := (0, w) when ν tends to ∞. Since the set B is a closed subset of D × R, the point b belongs to B, which completes the proof. ✷ 5. Definable functions of one variable In this section, we shall again consider the general case where K is an arbitrary Henselian rank one valued field of equicharacteristic zero. We begin with the following 14 KRZYSZTOF JAN NOWAK Proposition 5.1. Let f : A → K be an L-definable function on a subset A of K n. Then there is a finite partition of A into L-definable sets Ai and irreducible polynomials Pi(x, y), i = 1, . . . , k, such that for each a ∈ Ai the polynomial Pi(a, y) in y does not vanish and Pi(a, f (a)) = 0 for all a ∈ Ai, i = 1, . . . , k. Proof. By elimination of K-quantifiers, as in the proof of Theorem 2.4, the graph of f is a finite union of sets Bi, i = 1, . . . , k, defined by conditions of the form (v(f1(x, y)), . . . , v(fr(x, y))) ∈ P, (ac g1(x, y), . . . , ac gs(x, y)) ∈ Q, where fi, gj ∈ K[x, y] are polynomials, and P and Q are L-definable subsets of Γr and ks, respectively. Each set Bi is the graph of the restriction of f to an L-definable subset Ai. Since, for each point a ∈ Ai, the fibre of Bi over a consists of one point, the above condition imposed on angular components includes one of the form ac gj(x, y) = 0 or, equivalently, gj(x, y) = 0, for some j = 1, . . . , s, which may depend on a, where the polynomial gj(a, y) in y does not vanish. This means that the set {(ac g1(x, y), . . . , ac gs(x, y)) : (x, y) ∈ Bi} is contained in the union of hyperplanes Ss j=1{ξj = 0} and, further- more, that, for each point a ∈ Ai, there is an index j = 1, . . . , s such that the polynomial gj(a, y) in y does not vanish and gj(a, f (a)) = 0. Clearly, for any j = 1, . . . , s, this property of points a ∈ Ai is L- definable. Therefore we can partition the set Ai into subsets each of which fulfils the condition required in the conclusion with some irre- ducible factors of the polynomial gj(x, y). (cid:3) Consider a complete rank one valued field L. For every non-negative integer r, let L{x}r be the local ring of all formal power series φ(x) = akxk ∈ L[[x]] in one variable x such that v(ak) + kr tends to ∞ when k → ∞; L{x}0 coincides with the ring of restricted formal power series. Then the local ring ∞Xk=0 L{x} := L{x}r ∞[r=0 ALGEBRAIC GEOMETRY OVER VALUED FIELDS 15 is Henselian, which can be directly deduced by means of the implicit function theorem for restricted power series in several variables (see [7, Chap. III, § 4], [16] and also [17, Chap. I, § 5]). Now, let L be the completion of the algebraic closure K of the ground field K. Clearly, the Henselian local ring L{x} is closed under division by the coordinate and power substitution. Therefore it follows from our paper [35, Section 2] that Puiseux’s theorem holds for L{x}. We still need an auxiliary lemma. Lemma 5.2. The field K is a closed subspace of its algebraic closure K. Proof. Indeed, Denote by cl (E, F ) the closure of a subset E in F , and But, through the transfer principle of Ax-Kochen–Ershov (see e.g. [9]), cl (K, K) = cl (K, L) ∩ K = bK ∩ K. let bK be the completion of K. We have K is an elementary substructure of bK and, a fortiori, is algebraically closed in bK. Hence cl (K, K) = bK ∩ K = K, as asserted. Consider an irreducible polynomial (cid:3) in two variables x, y of y-degree d ≥ 1. Let Z be the Zariski closure of its zero locus in K × P1(K). Performing a linear fractional transformation over the ground field K of the variable y, we can assume that the fiber {w1, . . . , ws}, s ≤ d, of Z over x = 0 does not contain the point at infinity, i.e. w1, . . . , ws ∈ K. Then pd(0) 6= 0 and pd(x) is a unit in L{x}. Via Hensel’s lemma, we get the Hensel decomposition P (x, y) = pd(x) · Pj(x, y) sYj=1 of P (x, y) into polynomials Pj(x, y) = (y − wj)dj + pj1(x)(y − wj)dj−1 + · · · + pjdj (x) ∈ L{x}[y − wj] which are Weierstrass with respect to y − wj, j = 1, . . . , s, respectively. By Puiseux’s theorem, there is a neighbourhood U of 0 ∈ K such that the trace of Z on U × K is a finite union of sets of the form Zφj = {(xq, φj(x)) : x ∈ U} with some φj ∈ L{x}, q ∈ N, j = 1, . . . , s. Obviously, for j = 1, . . . , s, the fiber of Zφj over x ∈ U tends to the point φj(0) = wj when x → 0. P (x, y) = pi(x)yi ∈ K[x, y] dXi=0 16 KRZYSZTOF JAN NOWAK If φj(0) ∈ K \ K, it follows from Lemma 5.2 that Zφj ∩ ((U ∩ K) × K) = ∅, after perhaps shrinking the neighbourhood U. Let us mention that if then φj(0) ∈ K and φj ∈ L{x} \ bK{x}, Zφj ∩ ((U ∩ K) × K) = {(0, φ(0))} after perhaps shrinking the neighbourhood U. Indeed, let φj(x) = akxk ∈ L[[x]] ∞Xk=0 and p be the smallest positive integer with ap ∈ L \ bK. Since bK is a closed subspace of L, we get akxk = xp ap + x · ∞Xk=p ∞Xk=p+1 akxk−(p+1)! 6∈ bK for x close enough to 0, and thus the assertion follows. Suppose now that an L-definable function f : A → K satisfies the equation P (x, f (x)) = 0 for x ∈ A and 0 is an accumulation point of the set A. It follows immediately from the foregoing discussion that the set A can be partitioned into a finite number of L-definable sets Aj, j = 1, . . . , r with r ≤ s, such that, after perhaps renumbering of the fiber {w1, . . . , ws} of the set {P (x, f (x)) = 0} over x = 0, we have lim x→0 f Aj (x) = wj for each j = 1, . . . , r. Hence and by Proposition 5.1, we immediately obtain the following Theorem 5.3. (Existence of the limit) Let f : A → K be an L- definable function on a subset A of K and suppose 0 is an accumulation point of A. Then there is a finite partition of A into L-definable sets A1, . . . , Ar and points w1 . . . , wr ∈ P1(K) such that lim x→0 f Aj (x) = wj for j = 1, . . . , r. Moreover, there is a neighbourhood U of 0 such that each definable set {(v(x), v(f (x))) : x ∈ (Aj ∩ U) \ {0}} ⊂ Γ × (Γ ∪ {∞}), j = 1, . . . , r, ALGEBRAIC GEOMETRY OVER VALUED FIELDS 17 is contained in an affine line with non-negative rational slope l = pj q · k + βj, j = 1, . . . , r, with pj, q ∈ N, q > 0, βj ∈ Γ, or in Γ × {∞}. ✷ Remark 5.4. In the above theorem, the existence of the limits could also be established through the lemma on the continuity of roots of a monic polynomial, which can be found in e.g. [6, Chap. 3, § 3]). The second conclusion, in turn, follows from Puiseux’s parametrization. 6. Fiber shrinking for definable sets Let A be an L-definable subset of K n with accumulation point a = (a1, . . . , an) ∈ K n and E an L-definable subset of K with accumulation point a1. We call an L-definable family of sets Φ =[t∈E {t} × Φt ⊂ A an L-definable x1-fiber shrinking for the set A at a if lim t→a1 Φt = (a2, . . . , an), i.e. for any neighbourhood U of (a2, . . . , an) ∈ K n−1, there is a neigh- bourhood V of a1 ∈ K such that ∅ 6= Φt ⊂ U for every t ∈ V ∩ E, t 6= a1. When n = 1, A is itself a fiber shrinking for the subset A of K at an accumulation point a ∈ K. This concept is a relaxed version of curve selection, available in the non-archimedean geometry treated in this paper. Theorem 6.1. (Fiber shrinking) Every L-definable subset A of K n with accumulation point a ∈ K n has, after a permutation of the coor- dinates, an L-definable x1-fiber shrinking at a. Proof. We proceed with induction with respect to the dimension of the ambient affine space n. The case n = 1 is trivial. So assuming the assertion to hold for n, we shall prove it for n + 1. We may, of course, assume that a = 0. Let x = (x1, . . . , xn+1) be coordinates in K n x . If 0 is an accumulation point of the intersections A ∩ {xi = 0}, i = 1, . . . , n + 1, 18 KRZYSZTOF JAN NOWAK we are done by the induction hypothesis. Thus we may assume that the intersection A ∩ n+1[i=1 {xi = 0} = ∅ is empty. Then the definable (in the Γ-sort) set P := {(v(x1), . . . , v(xn+1)) ∈ Γn+1 : x ∈ A} has an accumulation point (∞, . . . , ∞). Since the Γ-sort admits quantifier elimination in the language of or- dered groups augmented by binary relation symbols ≡n for congruence modulo n, every definable subset of Γn+1 is a finite union of subsets of semi-linear sets contained in Γn+1 that are determined by a finite number of congruences (6.1) n+1Xj=1 rij · kj ≡N αi, i = 1, . . . , s; here N ∈ N, N > 1, rij ∈ Z, αi ∈ Γ for i = 1, . . . , s, j = 1, . . . , n + 1. Consequently, there exists a semi-linear subset P0 of Rn+1 given by finitely many linear equations and inequalities with integer coefficients and with constant terms from Γ such that the subset P1 of P0 ∩ Γn+1 determined by congruences of the form 6.1 is contained in P and has an accumulation point (∞, . . . , ∞). Therefore there exists an affine semi-line L := {(r1 · k + γ1, . . . , rn+1 · k + γn+1) : k ∈ Γ, k ≥ 0} , where r1, . . . , rn+1 are positive integers, passing through a point (γ1, . . . , γn+1) ∈ P1 ⊂ Γn+1 and contained in P0. It is easy to check that the set L0 := {(γ1 + rr1N, . . . , γn+1 + rrn+1N) : r ∈ N} ⊂ P1 is contained in P1. Then Φ := {x ∈ A : (v(x1), . . . , v(xn+1)) ∈ L0} is an L-definable x1-fiber shrinking for the set A at 0. This finishes the proof. (cid:3) ALGEBRAIC GEOMETRY OVER VALUED FIELDS 19 7. Proof of Theorem 3.1 for the general case The proof reduces easily to the case m = 1. We must show that if B is an L-definable subset of D × R and a point a lies in the closure of A := π(B), then there is a point b in the closure of B such that π(b) = a. We may obviously assume that a = 0 6∈ A. By Theorem 6.1, there exists, after a permutation of the coordinates, an L-definable x1-fiber shrinking Φ for A at a: Φ =[t∈E {t} × Φt ⊂ A, lim t→0 Φt = 0; here E is the canonical projection of A onto the x1-axis. Put B ∗ := {(t, y) ∈ K × R : ∃ u ∈ Φt (t, u, y) ∈ B}; it easy to check that if a point (0, w) ∈ K 2 lies in the closure of B ∗, then the point (0, w) ∈ K n+1 lies in the closure of B. The problem is thus reduced to the case n = 1 and a = 0 ∈ K. By Theorem 2.6, we can assume that B is a subset F of a cell C F ⊂ C ⊂ Kx × R ⊂ Kx × Ky of the form F (ξ) := {(x, y) ∈ Kx × Ky : (x, ξ) ∈ D, v(a(x, ξ)) ✁1 v((y − c(x, ξ))ν) ✁2 v(b(x, ξ)), ac(y − c(x, ξ)) = ξ1, v(cid:0)fi(x, ξ)(y − c(x, ξ))ki(cid:1) ≡n 0, i = 1, . . . , s}. But the set {(v(x), ξ) ∈ Γ × km : ∃ y ∈ R (x, y) ∈ F (ξ)} is an L-definable subset of the product Γ×km of the two sorts, which is, by elimination of K-quantifiers, a finite union of the Cartesian products of definable subsets in Γ and in km, respectively. It follows that 0 is an accumulation point of the projection π(F (ξ ′)) of the fiber F (ξ ′) for a parameter ξ ′ ∈ km. We are thus reduced to the case where B is the fiber F (ξ ′) of the set F for a parameter ξ ′. For simplicity, we abbreviate c(x, ξ ′), a(x, ξ ′), b(x, ξ ′) and fi(x, ξ ′) to c(x), a(x), b(x) and fi(x), i = 1, . . . , s. Denote by E ⊂ K the domain of these functions; then 0 is an accumulation point of E. In the statement of Theorem 3.1, we may equivalently replace R with the projective line P1(K), because the latter is the union of two open and closed charts biregular to R. By Theorem 5.3, we can thus assume that the limits, say c(0), a(0), b(0),fi(0) of c(x), a(x), b(x), fi(x) 20 KRZYSZTOF JAN NOWAK (i = 1, . . . , s) when x → 0 exist in P1(K) and, moreover, there is a neighbourhood U of 0 such that, each definable set {(v(x), v(fi(x))) : x ∈ (E ∩ U) \ {0}} ⊂ Γ × (Γ ∪ {∞}), i = 1, . . . , s, is contained in an affine line with non-negative rational slope (7.1) l = pi q · k + βi, i = 1, . . . , s, with pi, q ∈ N, q > 0, βi ∈ Γ, or in Γ × {∞}. Performing a linear fractional transformation of the coordinate y, we get c(0), a(0), b(0) ∈ K. The role of the center c(x) is immaterial. We can assume, without loss of generality, that it vanishes, c(x) ≡ 0, for if a point b = (0, w) ∈ K 2 lies in the closure of the cell with zero center, the point (0, w + c(0)) lies in the closure of the cell with center c(x). When ✁1 occurs and a(0) = 0, the set F (ξ ′) is itself an x-fiber shrinking at (0, 0) and the point b = (0, 0) is an accumulation point of B lying over a = 0, as desired. So suppose that either only ✁2 occurs or ✁1 occurs and a(0) 6= 0. By elimination of K-quantifiers, the set v(E) is a definable subset of Γ. The value group Γ admits quantifier elimination in the language of ordered groups augmented by symbols ≡n for congruences modulo n, n ∈ N, n > 1 (cf. Section 2). Therefore the set v(E) is of the form (7.2) v(E) = {k ∈ (α, ∞) ∩ Γ : mjk ≡N γj, j = 1, . . . , t}, where α, γj ∈ Γ, mj ∈ N for i = 1, . . . , s. Now, take an element (u, w) ∈ F (ξ ′) with u ∈ (E ∩ U) \ {0}. By equality 7.2, there is a point xr ∈ E, r ∈ N, with By equality 7.1, we get v(ur) = v(u) + rqnN. v(fi(ur)) = v(fi(u)) + rpinN, i = 1, . . . , s. Hence (7.3) v(cid:0)fi(ur)wki(cid:1) = v(fi(ur)) + kiv(w) = = v(fi(u)) + rpinN + kiv(w) = v(cid:0)fi(u)wki(cid:1) + rpinN ≡n 0. Of course, after shrinking the neighbourhood U, we may assume that v(a(x)) = v(a(0)) < ∞ for all x ∈ (E ∩ U) \ {0}. Consequently, v(a(ur)) ✁1 v(wν) ✁2 v(b(ur)). ALGEBRAIC GEOMETRY OVER VALUED FIELDS 21 Hence and by 7.3, we get (ur, w) ∈ F (ξ ′). Since ur tends to 0 ∈ K when r → ∞, the point (0, w) is an accumulation point of F (ξ ′) lying over 0 ∈ K, which completes the proof of the closedness theorem. ✷ 8. Curve selection We now pass to curve selection over non-locally compact ground fields under study. While the real version of curve selection goes back to papers [8, 42] (see also [32, 33, 5]), the p-adic one was achieved in papers [41, 14]. Before proving a general version for L-definable sets, we give a version for semialgebraic sets. By a (valuative) semialgebraic subset of K n we mean a (finite) Boolean combination of elementary (valuative) semialgebraic subsets, i.e. sets of the form {x ∈ K n : v(f (x)) ≤ v(g(x))}, where f and g are regular functions on K n. We call a map ϕ semial- gebraic if its graph is a semialgebraic set. Theorem 8.1. Let A be a semialgebraic subset of K n. If a point a ∈ K n lies in the closure (in the K-topology) of A \ {a}, then there is a semialgebraic map ϕ : R −→ K n given by restricted power series such that ϕ(0) = a and ϕ(R \ {0}) ⊂ A \ {a}. Proof. It is easy to check that every semialgebraic set is a finite union of basic semialgebraic sets, i.e. sets of the form {x ∈ K n : v(f1(x))✁1v(g1(x))} ∩ . . . ∩ {x ∈ K n : v(fr(x))✁rv(gr(x))}, where f1, . . . , fr, g1, . . . , gr are regular functions and ✁1, . . . , ✁r stand for ≤ or <. We may assume, of course, that A is a set of this form and a = 0. Take a finite composite σ : Y −→ KAn of blow-ups along smooth centers such that the pull-backs of the coor- dinates x1, . . . , xn and the pull-backs f σ 1 := f1 ◦ σ, . . . , f σ r := fr ◦ σ and gσ 1 := g1 ◦ σ, . . . , gσ r := gr ◦ σ are normal crossing divisors ordered with respect to divisibility relation, unless they vanish. Since the restriction σ : Y (K) −→ K n is definably closed (Corollary 3.5), there is a point b ∈ Y (K) ∩ σ−1(a) which lies in the closure of the set B := Y (K) ∩ σ−1(A \ {a}). 22 KRZYSZTOF JAN NOWAK 1 (y))✁1v(gσ 1 (y))} ∩ . . . ∩ {v(f σ Further, we get Y (K)∩σ−1(A) = {v(f σ r (y))}, and thus σ−1(A) is in suitable local coordinates y = (y1, . . . , yn) near b = 0 a finite intersection of sets of the form {v(yα) ≤ v(u(y))}, {v(u(y)) ≤ v(yα)}, {v(yβ) < ∞} or {∞ = v(yγ)}, where α, β, γ ∈ Nn and u(y) is a regular function which vanishes nowhere. r (y))✁rv(gσ The first case cannot occur because b = 0 lies in the closure of B; the second case holds in a neighbourhood of b; the third and fourth cases are equivalent to yβ 6= 0 and yγ = 0, respectively. Consequently, since the pull-backs of the coordinates x1, . . . , xn are monomial divisors too, B contains the set (R \ {0}) · c when c ∈ B is a point sufficiently close to b = 0. Then the map ϕ : R −→ K n, ϕ(z) = σ(z · c) has the desired properties. (cid:3) We now pass to the general case of curve selection for L-definable sets. Theorem 8.2. Let A be an L-definable set subset of K n. If a point a ∈ K n lies in the closure (in the K-topology) of A \ {a}, then there exist a semialgebraic map ϕ : R −→ K n given by restricted power series and an L-definable subset E of R with accumulation point 0 such that ϕ(0) = a and ϕ(E \ {0}) ⊂ A \ {a}. Proof. We proceed with induction with respect to the dimension of the ambient space n. The case n = 1 being evident, suppose n > 1. By elimination of K-quantifiers, similarly as in Section 2, the set A \ {a} is a finite union of sets defined by conditions of the form (v(f1(x)), . . . , v(fr(x))) ∈ P, (ac g1(x), . . . , ac gs(x)) ∈ Q, where fi, gj ∈ K[x] are polynomials, and P and Q are definable subsets of Γr and ks, respectively (since x = 0 iff ac x = 0). Thus we may assume that A is such a set and, of course, that a = 0. Again, take a finite composite σ : Y −→ KAn of blow-ups along smooth centers such that the pull-backs f σ 1 , . . . , f σ r and gσ 1 , . . . , gσ r ALGEBRAIC GEOMETRY OVER VALUED FIELDS 23 are normal crossing divisors unless they vanish. Since the restriction σ : Y (K) −→ K n is definably closed (Corollary 3.5), there is a point b ∈ Y (K) ∩ σ−1(a) which lies in the closure of the set B := Y (K) ∩ σ−1(A \ {a}). Take local coordinates y1. . . . , yn near b in which b = 0 and every pull-back above is a normal crossing. We shall first select a semial- gebraic map ψ : R −→ Y (K) given by restricted power series and an L-definable subset E of R with accumulation point 0 such that ψ(0) = b and ψ(E \ {0}) ⊂ B. Since the valuation map and the angular component map composed with a continuous function are locally constant near any point at which this function does not vanish, the conditions which describe the set B near b are of the form where eP and eQ are definable subsets of Γn and kn, respectively. The set B0 determined by the conditions (v(y1), . . . , v(yn)) ∈ eP , (ac y1, . . . , ac yn) ∈ eQ, (v(y1), . . . , v(yn)) ∈ eP , n[i=1 (ac y1, . . . , ac yn) ∈ eQ ∩ {ξi = 0}, is contained near b in the union of hyperplanes {yi = 0}, i = 1, . . . , n. If b is an accumulation point of the set B0, then the desired map ψ exists by the induction hypothesis. Otherwise b is an accumulation point of the set B1 := B \ B0. Analysis from the proof of Theorem 6.1 (fiber shrinking) shows that the congruences describing the definable subset eP of Γn are not an essential obstacle to finding the desired map ψ, but affect only the definable subset E of R. Neither are the conditions {ξi = 0} n[i=1 eQ \ imposed on the angular components of the coordinates y1, . . . , yn, be- cause then none of them vanishes. Therefore, in order to select the map ψ, we must first of all analyze the linear conditions (equalities and inequalities) describing the set eP . 24 KRZYSZTOF JAN NOWAK cumulation point of B. We see, similarly as in the proof of Theorem 6.1 The set eP has an accumulation point (∞, . . . , ∞) as b = 0 is an ac- (fiber shrinking), that eP contains a definable subset of a semi-line L := {(r1 · k + γ1, . . . , rn · k + γn) : k ∈ Γ, k ≥ 0} , where r1, . . . , rn are positive integers, passing through a point clearly, (∞, . . . , ∞) is an accumulation point of that definable subset of L. γ1, . . . , γn ∈ eP ⊂ Γn; Now, take some elements (ξ1, . . . , ξn) ∈ eQ \ n[i=1 {ξi = 0} and next some elements w1, . . . , wn ∈ K for which v(w1) = γ1, . . . , v(wn) = γn and ac w1 = ξ1, . . . , ac wn = ξn. There exists an L-definable subset E of R which is determined by some congruences imposed on v(t) (as in the proof of Theorem 6.1) and the conditions ac t = 1 such that the subset F := {(w1 · tr1, . . . , wn · trn) : t ∈ E} of the arc ψ : R → Y, ψ(t) = (w1 · tr1, . . . , wn · trn) is contained in B1. Then ϕ := σ ◦ ψ is the map we are looking for. This completes the proof. (cid:3) 9. Lojasiewicz inequality In this section we provide certain general versions of the Lojasiewicz inequality. For the classical version over the real ground field, we refer the reader to [5, Thm.2.6.6]. Theorem 9.1. Let f, g : A → K be two continuous L-definable func- tions on a closed (in the K-topology) L-definable subset A of Rm. If {x ∈ A : g(x) = 0} ⊂ {x ∈ A : f (x) = 0}, then there exist a positive integer s and a continuous rational function h on A such that f s(x) = h(x) · g(x) for all x ∈ A. ALGEBRAIC GEOMETRY OVER VALUED FIELDS 25 Proof. It is easy to check that the set Aγ := {x ∈ A : v(f (x)) = γ} is a closed L-definable subset of A for every γ ∈ Γ. Hence and by Corollary 3.3 to the closedness theorem, the set g(Aγ) is a closed L- definable subset of K \ {0}, γ ∈ Γ. The set v(g(Aγ)) is thus bounded from above, i.e. v(g(Aγ)) < α(γ) for some α(γ) ∈ Γ. By elimination of K-quantifiers, the set Λ := {(v(f (x)), v(g(x))) ∈ Γ2 : x ∈ A} ⊂ {(γ, δ) ∈ Γ2 : δ < α(γ)} is a definable subset of Γ2, and thus it is described by a finite number of linear inequalities and congruences. Hence Λ ∩ {(γ, δ) ∈ Γ2 : γ > γ0} ⊂ {(γ, δ) ∈ Γ2 : δ < s · γ} for a positive integer s and some γ0 ∈ Γ. We thus get v(g(x)) < s · v(f (x)) if x ∈ A, v(f (x)) > γ0, whence v(g(x)) < v(f s(x)) if x ∈ A, v(f (x)) > γ0. Consequently, the quotient f s/g extends by zero through the zero set of the denominator to a (unique) continuous L-definable function on A. This finishes the proof. (cid:3) The above theorem can be generalized as follows. Theorem 9.2. Let U and F be two L-definable subsets of K m, suppose U is open and F closed in the K-topology and consider two continuous L-definable functions f, g : A → K on the locally closed subset A := U ∩ F of K m. If {x ∈ A : g(x) = 0} ⊂ {x ∈ A : f (x) = 0}, then there exist a positive integer s and a continuous rational function h on A such that f s(x) = h(x) · g(x) for all x ∈ A. Proof. We shall adapt the foregoing arguments. Since the set U is open, its complement V := K m \ U is closed in K m and A is the following union of open and closed subsets of K m and of Pm(K): Xβ := {x ∈ K m : v(x1), . . . , v(xm) ≥ −β, v(x − y) ≤ β for all y ∈ V }, where β ∈ Γ, β ≥ 0. As before, we see that the sets Aβ,γ := {x ∈ Xβ : v(f (x)) = γ} with β, γ ∈ Γ 26 KRZYSZTOF JAN NOWAK are closed L-definable subsets of Pm(K), and next that the sets g(Aβ,γ) are closed L-definable subsets of K \ {0} for all β, γ ∈ Γ. Likewise, we get Λ := {(β, v(f (x)), v(g(x))) ∈ Γ3 : x ∈ Xβ} ⊂ ⊂ {(β, γ, δ) ∈ Γ3 : δ < α(β, γ)} for some α(β, γ) ∈ Γ. Consequently, since Λ is a definable subset of Γ3, there exist a positive integer s and elements γ0(β) ∈ Γ such that Λ ∩ {(β, γ, δ) ∈ Γ3 : γ > γ0(β)} ⊂ {(β, γ, δ) ∈ Γ3 : δ < s · γ}. Since A is the union of the sets Xβ, it is not difficult to check that the quotient f s/g extends by zero through the zero set of the denominator to a (unique) continuous L-definable function on A, which is the desired result. (cid:3) 10. Extending continuous hereditarily rational functions We first recall an elementary lemma from [24]. Lemma 10.1. If the ground field K is not algebraically closed, then there are polynomials Gr(x1, . . . , xr) in any number of variables whose only zero on K r is (0, . . . , 0). Proof. Indeed, take a polynomial g(t) = td + a1td−1 + · · · + ad ∈ K[t], d > 1, which has no roots. Then its homogenization G2(x1, x2) = xd 1 + a1xd−1 1 x2 + · · · + adxd 2 is a polynomial in two variables we are looking for. Further, we can recursively define polynomials Gr by putting Gr+1(x1, . . . , xr, xr+1) := G2(Gr(x1, . . . , xr), xr+1). (cid:3) We keep further the assumption that K is a Henselian rank one valued field of equicharacteristic zero and, additionally, that it is not algebraically closed. We have at our disposal the descent property (Corollary 3.6) and the Lojasiewicz inequality (Theorem 9.2). There- fore we are able, by adapting mutatis mutandis its proof, to carry over Proposition 11 from [24] to the case of such non-archimedean fields. Theorem 10.2. (Extending continuous hereditarily rational functions) Let X be a smooth K-variety and W ⊂ Z ⊂ X closed subvarieties. Let f be a continuous hereditarily rational function on Z(K) that is regular at all K-points of Z(K) \ W (K). Then f extends to a continuous ALGEBRAIC GEOMETRY OVER VALUED FIELDS 27 hereditarily rational function F on X(K) that is regular at all K-points of X(K) \ W (K). Sketch of the Proof. We shall keep the notation from paper [24]. The main modification of the proof in comparison with that paper is the definition of the functions G and F2n which improve the rational function P/Q. Now we need the following corrections: G := P Q · Qd G2(Q, H) and Fdn := G · Qdn G2(Qn, H) = P Q · Qd G2(Q, H) · Qdn G2(Qn, H) , where the positive integer d and the polynomial G2 are taken from Lemma 10.1 and its proof. It is clear that the restriction of Fdn to Z \ W equals f2, and thus Theorem 10.2 will be proven once we show that the rational function Fdn restricts to a continuous function Φdn on X(K) for n ≫ 1. We work on the variety π : X1(K) −→ X(K) obtained by blowing up the ideal (P Qd−1, G2(Q, H)); let E := π−1(W ) be its exceptional divisor. Equivalently, X1(K) is the Zariski closure of the graph of G in X(K) × P1(K). Two open charts are considered: • a Zariski open neighbourhood U ∗ of the closure (in the K-topology) Z ∗ of π−1(Z(K) \ W (K)); • an open (in the K-topology) set V ∗ := X1(K) \ Z ∗, which is an L-definable subset of X1(K). Via the descent property (Corollary 3.6), it suffices to show that the rational function Fdn ◦ π (with n ≫ 1) extends to a continuous function on X1(K) that vanishes on E(K). The subtlest analysis is on the latter chart, on which Fdn ◦ π can be written in the form Fdn ◦ π = (P ◦ π) ·(cid:18) Qdn−1 H d ◦ π(cid:19) ·(cid:18) Qd ◦ π(cid:19) ·(cid:18) H d G2(Qn, H) ◦ π(cid:19) . G2(Q, H) Note that on V ∗ the function H ◦ π vanishes only along E(K) and Q ◦ π vanishes along E(K) too. Therefore we can apply Theorem 9.2 ( Lojasiewicz inequality) to the numerator and denominator of the first factor, which are regular functions on the chart V ∗, to immediately deduce that the first factor extends to a continuous rational function on V ∗ for n ≫ 1. 28 KRZYSZTOF JAN NOWAK What still remains to prove (cf. paper [24]) is that the factors Qdn G2(Qn, H) , Qd G2(Q, H) and H d G2(Qn, H) are regular functions off W (K) whose valuations are bounded from below. But this follows immediately from an auxiliary lemma: Lemma 10.3. Let g be the polynomial from the proof of Lemma 10.1. Then the set of values v(cid:18) td g(t)(cid:19) ∈ Z, t ∈ K, is bounded from below. In order to prove this lemma, observe that g(t)(cid:19) = v(cid:16)1 + v(cid:18) td a1 t + · · · + ad td(cid:17) for t ∈ K, t 6= 0. Hence the values under study are zero when iv(t) < v(ai) for all i = 1, . . . , d. Therefore, we are reduced to analysing the case where v(t) ≥ k := min(cid:26)v(ai) i : i = 1, . . . , d(cid:27) . Denote by Γ the value group of v. Thus we must show that the set of values v(g(t)) ∈ Γ when v(t) ≥ k is bounded from above. Take elements a, b ∈ R, a, b 6= 0, such that aai ∈ R for all i = 1, . . . , d, and bt ∈ R whenever v(t) ≥ k. Then (ab)dg(t) = (abt)d+aba1(abt)d−1+(ab)2a2(abt)d−2+· · ·+(ab)dad =: h(abt), where h is a monic polynomial with coefficients from R which has no roots in K. Clearly, it is sufficient to show that the set of values v(h(t)) ∈ Γ when t ∈ R is bounded from above. Consider a splitting field eK = K(u1, . . . , ud) of the polynomial h, where u1, . . . , ud are the roots of h. Let ev be a (unique) extension to eK of the valuation v, eR be its valuation ring and eΓ, Γ ⊂ eΓ, its value group (see e.g. [43, Chap. VI, § 11] for valuations of algebraic field extensions). Then u1, . . . , ud ∈ eR \ R and h(t) = (t − ui). dYi=1 ALGEBRAIC GEOMETRY OVER VALUED FIELDS 29 Since R is a closed subring of eR by Lemma 5.2, there exists an l ∈eΓ such thatev(t−ui) ≤ l for all i = 1, . . . , d and t ∈ R. Hence v(h(t)) ≤ dl for all t ∈ R, and thus the lemma follows. ✷ In this fashion, we have demonstrated how to adapt the proof of Proposition 11 from paper [24] to the case of Henselian rank one valued fields of equicharacteristic zero. Note that the proofs of all remaining results from that paper work over general topological fields with the density property. We say that a topological field K enjoys density property (cf. [24]) if the following equivalent conditions hold: (1) If X is a smooth, irreducible K-variety and ∅ 6= U ⊂ X is a Zariski open subset, then U(K) is dense in X(K) in the K- topology. (2) If C is a smooth, irreducible K-curve and ∅ 6= C 0 ⊂ C is a Zariski open subset, then C 0(K) is dense in C(K) in the K- topology. (3) If C is a smooth, irreducible K-curve, then C(K) has no isolated points. The examples of such fields are, in particular, all Henselian rank one valued fields. 11. Regulous functions and sets In these last three sections, we shall carry the theory of regulous functions over the real ground field R, developed by Fichou–Huisman– Mangolte–Monnier [15], over to non-archimedean algebraic geometry over Henselian rank one valued fields K of equicharacteristic zero. We assume that the ground field K is not algebraically closed. (Otherwise, the notion of a regulous function on a normal variety coincides with that of a regular function and, in general, the study of continuous rational functions leads to the concept of seminormality and seminormalization; cf. [1, 2] or [23, Sec.10.2] for a recent treatment.) Every such field enjoys the density property. The K-points X(K) of any algebraic K-variety X inherit from K a topology, called the K-topology. In this section, we deal with the ground fields K with the density property. Observe first that if f is a rational function on an affine K- variety X which is regular on a Zariski open subset U, then there exist two regular functions p, q on X such that f = p q and q(x) 6= 0 for all x ∈ U(K). 30 KRZYSZTOF JAN NOWAK When X ⊂ KAn, then p, q can be polynomial functions. Every rational function f is regular on the largest Zariski open subset of X, called its regular locus and denoted by dom (f ). Further, assume that Z is a closed subvariety of a K-variety X. Then every rational function f on Z that is regular on Z(K) extends to a rational function F on X that is regular on X(K). Both the results can be deduced via Lemma 10.1 (cf. [24], the proof of Lemma 15 on extending regular functions). Suppose now that X is a smooth affine K-variety or, at least, an affine K-variety that is smooth at all K-points X(K). We say that a function f on X(K) is k-regulous, k ∈ N ∪ {∞}, if it is of class Ck and there is a Zariski dense open subset U of X such that the restriction of f to U(K) is a regular function. A function f on X(K) is called regulous if it is 0-regulous. Denote by Rk(X(K)) the ring of k-regulous functions on X(K). The ring R∞(X(K)) of ∞-regulous functions on X(K) coincides with the ring O(X(K)) of regular functions on X(K). This follows easily from the faithful flatness of the formal power series ring K[[x1, . . . , xn]] over the local ring of regular function germs at 0 ∈ K n. Therefore we shall most often restrict ourselves to the case k ∈ N. When K is a Henselian rank one valued field of equicharacteristic zero, transformation to a normal crossing by blowing up along smooth centers and the descent property (Corollary 3.6) enable the following characterization: Given a smooth algebraic K-variety X, a function f : X(K) → K with smooth centers such that the pull-back f σ := f ◦ σ is a regular is regulous iff there exists a finite composite σ : eX → X of blow-ups function on eX(K). We say that a subset V of K n is k-regulous closed if it is the zero set of a family E ⊂ Rk(Kn) of k-regulous functions: V = Z(E) := {x ∈ K n : f (x) = 0 for all f ∈ E}. A subset U of K n is called k-regulous open if its complement Kn \ U is k-regulous closed. The family of k-regulous open subsets of K n is a topology on K n, called the k-regulous topology on K n. If f 6= 0 is a k-regulous function on K n with regular locus U = dom (f ), then p q f = where p, q ∈ K[x1, . . . , xn], Z(q) = K n \ U(K), ALGEBRAIC GEOMETRY OVER VALUED FIELDS 31 and p, q are coprime polynomials. Clearly, Z(q) ⊂ Z(p) and it follows, by passage to the algebraic closure of K, that the zero set Z(q) is of codimension ≥ 2 in K n. Thus the complement K n \ dom (f ) is of codimension ≥ 2 in K n. Consequently, every k-regulous function on K is regular and every k-regulous function on K 2 is regular at all but finitely many points. We now recall some results about algebraic sets over arbitrary fields F . Let V be an affine F -variety. Then the regular locus Reg (V ) of V is a non-empty, Zariski open subset of V and, moreover, its trace on the set V (F ) of F -rational points of V is non-empty; cf. [29], Chap. VI, Corollary 1.17 to the Jacobian criterion for regular local rings and the remark preceding it. Therefore, for every affine F -subvariety V of F An, the set V (F ) of its F -rational points is a finite (disjoint) union of smooth, Zariski locally closed subsets of pure dimension. We call a subset E of K n constructible if it is a (finite) Boolean combination of Zariski closed subsets of K n. Every such set E is, of course, a finite union of Zariski locally closed subsets. Moreover, by the above observation, E is a finite union of smooth, Zariski locally closed subsets of pure dimension with irreducible Zariski closure. Thus it follows immediately from the density property that every closed (in the K-topology) constructible subset E is a finite union of a (unique) family Σ(E) of subsets each of which is the regular locus Reg (V ) ∩ K n of an irreducible affine K-subvariety V of KAn; obviously, Reg (V )∩K n is a smooth, Zariski locally closed subset of pure dimension dim V . Below we introduce the constructible topology on K n. We shall see in the next section that the k-regulous topology coincides with the constructible topology for all k ∈ N. Theorem 11.1. If K is a topological field with the density property, then the family of all closed (in the K-topology) constructible subsets of K n is the family of closed sets for a topology, called the constructible topology on K n. Furthermore, this topology is noetherian, i.e. every descending sequence of closed constructible subsets of K n stabilizes. Proof. Clearly, it suffices to prove only the last assertion. We shall follow the reasoning from our paper [34] which showed that the quasi- analytic topology is noetherian. For any closed (in the K-topology) constructible subset E of K n, let µi(E) be the number of elements from the family Σ(E), constructed above, of dimension i, i = 0, 1, . . . , n, and put µ(E) = (µn(E), µn−1(E), . . . , µ0(E)) ∈ Nn+1. 32 KRZYSZTOF JAN NOWAK Consider now a descending sequence of closed constructible subsets K n ⊃ E1 ⊃ E2 ⊃ E3 ⊃ . . . . It is easy to check that for any two closed (in the K-topology) con- structible subsets D ⊂ E, we have µ(D) ≤ µ(E) and, furthermore, D = E iff µ(D) = µ(E). Hence we get the decreasing (in the lexico- graphic order) sequence of multi-indices µ(E1) ≥ µ(E2) ≥ µ(E3) ≥ . . . , which must stabilize for some N ∈ N: µ(EN ) = µ(EN +1) = µ(EN +2) = . . . Then as desired. EN = EN +1 = EN +2 = . . . , (cid:3) Proposition 11.2. Suppose K is a field with the density property. Then there is a one-to-one correspondence between the irreducible closed constructible subsets E of K n and the irreducible Zariski closed subsets V of K n: α : E 7−→ E Z and β : V 7−→ Reg (V ) c , Z c stands for the Zariski closure of E and A where E A in the constructible topology. for the closure of Proof. We have α ◦ β = Id, because Reg (V ) = V for every irreducible Zariski closed subset V of K n. In view of the foregoing discussion, the assignment β is surjective. Therefore every irreducible closed con- structible subset E of K n is of the form E = β(V ) = Reg (V ) for an irreducible Zariski closed subset V of K n. Hence c Z (β ◦ α)(E) = (β ◦ α)(β(V )) = (β ◦ α ◦ β)(V ) = (β ◦ Id)(V ) = β(V ) = E, and thus β ◦ α = Id, which finishes the proof. (cid:3) Below we recall Proposition 8 from paper [24] which holds over any topological fields with the density property. Proposition 11.3. Let X be an algebraic K-variety and f a rational function on X that is regular on X 0 ⊂ X. Assume that f X 0(K) has a continuous extension f c : X(K) → K. Let Z ⊂ X be an irreducible subvariety that is not contained in the singular locus of X. Then there is a Zariski dense open subset Z 0 ⊂ Z such that f cZ 0(K) is a regular function. ✷ We immediately obtain two corollaries: ALGEBRAIC GEOMETRY OVER VALUED FIELDS 33 Corollary 11.4. Let X be an algebraic K-variety that is smooth at all K-rational points X(K) and f a rational function on X that is regular on X 0 ⊂ X. Assume that f X 0(K) has a continuous extension f c : X(K) → K. Then there is a sequence of closed subvarieties ∅ = X−1 ⊂ X0 ⊂ · · · ⊂ Xn = X such that for i = 0, . . . , n the restriction of f to Xi(K) \ Xi−1(K) is regular. Moreover, we can require that each set Xi \ Xi−1 be smooth of pure dimension i. ✷ Corollary 11.5. If f is a regulous function on K n, then there is a sequence of Zariski closed subsets ∅ = E−1 ⊂ E0 ⊂ · · · ⊂ En = K n such that for i = 0, . . . , n the restriction of f to Ei \ Ei−1 is regular. Moreover, we can require that each set Ei \ Ei−1 be smooth of pure dimension i. ✷ Remark 11.6. Given a finite number of regulous functions f1, . . . , fp, there is a filtration ∅ = E−1 ⊂ E0 ⊂ · · · ⊂ En = K n as in the above corollary such that for i = 0, . . . , n the restriction of each function fj, j = 1, . . . , p, to Ei \ Ei−1 is regular. Now, three further consequences of the above corollary will be drawn. We say that a map f = (f1, . . . , fp) : K n → K p is k-regulous if all its components f1, . . . , fp are regulous functions on K n. Corollary 11.7. If two maps g : K m → K n and f : K n → K p are k-regulous, so is its composition f ◦ g. Proof. Indeed, let U be the common regular locus of the components g1, . . . , gn of the map g: U := dom (g1) ∩ . . . ∩ dom (gn) ⊂ Rm. Take a filtration ∅ = E−1 ⊂ E0 ⊂ · · · ⊂ En = K n 34 KRZYSZTOF JAN NOWAK for the functions f1, . . . , fp described in Remark 11.6. Then U is the following union of Zariski locally closed subsets U = n[i=0(cid:0)U ∩ g−1(Ei \ Ei−1)(cid:1) . Clearly, one of these sets, say U ∩ g−1(Ei0 \ Ei0−1) must be a Zariski dense open subset of U and of K m too. Hence f ◦g is a regular function on U ∩ g−1(Ei0 \ Ei0−1), which is the required result. (cid:3) Corollary 11.8. The zero set Z(f ) of a regulous function f on K n is a closed (in the K-topology) constructible subset of K n. ✷ Corollary 11.9. The zero set Z(f1, . . . , fp) of a finite number of reg- ulous functions f1, . . . , fp on K n is a closed (in the K-topology) con- structible subset of K n. Proof. This follows directly from Corollary 11.8 and Lemma 10.1. (cid:3) Hence and by Theorem 11.1, we immediately obtain Proposition 11.10. The k-regulous topology on K n is noetherian. ✷ Corollary 11.11. Every k-regulous closed subset of K n is the zero set Z (f ) of a k-regulous function f on K n, and thus is a closed (in the K-topology) constructible subset of K n. Hence every k-regulous open subset of K n is of the form D (f ) := K n \ Z (f ) = {x ∈ K n : f (x) 6= 0} for a k-regulous function f on K n. ✷ Corollaries 11.11 and 11.7 yield the following Corollary 11.12. Every k-regulous map f : K n → K m is continuous in the k-regulous topology. ✷ 12. Regulous Nullstellensatz We keep further the assumption that the ground field K is a Henselian rank one valued field of equicharacteristic zero and that it is not al- gebraically closed. Throughout this section, k will be a non-negative integer. We begin with a consequence of the Lojasiewicz inequality (Theorem 9.2). Proposition 12.1. Let f, g be rational functions on KAn such that f extends to a continuous function on K n and g extends to a continuous function on the set D (f ). Then the function f sg extends, for s ≫ 0, by zero through the set Z (f ) to a continuous rational function on K n. ALGEBRAIC GEOMETRY OVER VALUED FIELDS 35 Proof. We can find a finite composite σ : M → KAn of blow-ups along smooth centers such that the pull-backs f σ := f ◦ σ and gσ := g ◦ σ are regular functions at all K-rational points on M(K) and M(K) \ σ−1(Z (f )), respectively. Then there are regular functions p, q on M such that gσ = p q and Z (q) := {y ∈ M(K) : q(y) = 0} ⊂ Z (f σ). It follows immediately from Theorem 9.2 that the rational function (f σ)s q , for s ≫ 0, extends by zero through the set Z (f σ) to a continuous function on M(K), whence so does the rational function (f σ)s · gσ. By the descent property (Corollary 3.6), the continuous function (f σ)s · gσ descends to a continuous function on K n that vanishes on Z (f ). This is the required result. (cid:3) The two corollaries stated below are verbatim counterparts of Lem- mata 5.1 and 5.2 from paper [15], established over the real ground field R. Nevertheless, the proofs given here seem to be simpler. Corollary 12.2. Let f be a k-regulous function on K n and g a k- regulous function on the open subset D (f ). Then the function f sg extends, for s ≫ 0, by zero through the zero set Z (f ) to a k-regulous function on K n. Hence the ring of k-regulous functions on D (f ) is the localization Rk(K n)f . Proof. The case k = 0 is just Proposition 12.1. If s is an exponent large enough so that f sg extends to a continuous function on K n vanishing on Z (f ), then f s+kg is k flat on Z (f ), and thus is k-regulous on K n, as desired. (cid:3) Corollary 12.3. Let U be a k-regulous open subset of K n, f a k- regulous function on U and g a k-regulous function on the open subset D (f ) ⊂ U. Then the function f sg extends, for s ≫ 0, by zero through the zero set Z (f ) ⊂ U to a k-regulous function on U. Proof. By Corollary 11.11, U = D(h) for a k-regulous function on K n. From the above corollary we get f hs ∈ Rk(K n) for s ≫ 0. 36 Hence KRZYSZTOF JAN NOWAK g ∈ Rk(K n)f hs for integers s large enough, and thus the conclusion follows. (cid:3) Now we can readily pass to a regulous version of Nullstellensatz, which follows directly from Corollary 12.2 and the fact that the k- regulous topology is noetherian. Theorem 12.4. If I is an ideal in the ring Rk(K n) of k-regulous functions on K n, then where Rad (I) = I (Z (I)), I (E) := {f ∈ Rk(K n) : f (x) = 0 for all x ∈ E} for a subset E of K n. Proof. The inclusion Rad (I) = I (Z (I)) is obvious. For the converse inclusion, observe that, since the k-regulous topology is noetherian (Proposition 11.10), there is a function g ∈ I such that Z (I) = Z (g). Then Z (g) ⊂ Z (f ) for any f ∈ I (Z (I)), and thus the function 1/g is k-regulous on the set D (f ). By Corollary 12.2, we get f s g ∈ Rk(K n) for s ≫ 0 large enough. Hence f s ∈ g · Rk(K n) ⊂ I, concluding the proof. (cid:3) Corollary 12.5. There is a one-to-one correspondence between the radical ideals of the ring Rk(K n) and the k-regulous closed subsets of K n. Consequently, the prime ideals of Rk(K n) correspond to the irre- ducible k-regulous closed subsets of K n, and the maximal ideals m of Rk(K n) correspond to the points x of K n so that we get the bijection K n ∋ x −→ mx := {f ∈ Rk(K n) : f (x) = 0} ∈ Max(cid:0)Rk(K n)(cid:1) . The resulting embedding ι : K n ∋ x −→ mx ∈ Spec(cid:0)Rk(K n)(cid:1) is continuous in the k-regulous and Zariski topologies, and induces a one-to-one correspondence between the k-regulous closed subsets of K n and the subsets of Spec(cid:0)Rk(K n)(cid:1) closed in the Zariski topology. precisely of the form yield immediately ALGEBRAIC GEOMETRY OVER VALUED FIELDS 37 Proof. The embedding ι is continuous by the very definition of the Zariski topology. The last assertion follows immediately from the Null- The above corollary along with Proposition 11.10 and Corollary 11.11 stellensatz and the fact that the closed subsets of Spec(cid:0)Rk(K n)(cid:1) are {p ∈ Spec(cid:0)Rk(K n)(cid:1) : p ⊃ I}, where I runs over all radical ideals of Spec(cid:0)Rk(K n)(cid:1). Corollary 12.6. With the above notation, the space Spec(cid:0)Rk(K n)(cid:1) and the subsets of Spec(cid:0)Rk(K n)(cid:1) open in the Zariski topology. particular, every open subset of Spec(cid:0)Rk(K n)(cid:1) is of the form with the Zariski topology is noetherian, and the embedding ι induces a one-to-one correspondence between the k-regulous open subsets of K n In U(f ) := {p ∈ Spec(cid:0)Rk(K n)(cid:1) : f 6∈ p}, corresponding to the subset D(f ) of K n. (cid:3) ✷ f ∈ Rk(K n), Remark 12.7. As demonstrated in paper [15] (see also [30, Ex. 6.11]), the ring Rk(K n) is not noetherian for all k, n ∈ N, n ≥ 2. From Theorem 12.4 and Corollary 11.11, we immediately obtain Corollary 12.8. Every radical ideal of Rk(K n) is the radical of a principal ideal of Rk(K n). ✷ Finally, we return to the the comparison of the regulous and con- structible topologies. Below we state the non-archimedean version of [15, Theorem 6.4] by Fichou–Huisman–Mangolte–Monnier, which says that those topologies coincide in the real algebraic geometry. The proof relies on their Lemmata 5.1 and 5.2, and can be repeated verbatim in the case of the ground fields K studied in our paper. Theorem 12.9. The k-regulous closed subsets of K n are precisely the closed (in the K-topology) constructible subsets of K n. ✷ The above theorem along with Proposition 11.2 and Corollary 12.5 yield the following Corollary 12.10. There are one-to-one correspondences between the prime ideals of the ring Rk(K n), the irreducible closed constructible subsets of K n and the irreducible Zariski closed subsets of K n. ✷ Corollary 12.11. The dimension of the topological space K n with the regulous topology and the Krull dimension of the ring Rk(K n) is n. ✷ 38 KRZYSZTOF JAN NOWAK 13. Quasi-coherent regulous sheaves The concepts of quasi-coherent k-regulous sheaves on K n and k- regulous affine varieties, k ∈ N ∪ {∞}, can be introduced over valued fields studied in this paper, similarly as by Fichou–Huisman–Mangolte– Monnier [15] over the real ground field R. Also, the majority of their results concerning these concepts carry over to the non-archimedean geometry with similar proofs. For the sake of completeness, we provide an exposition of the theory of quasi-coherent regulous sheaves. Here we shall deal only with k-regulous functions with a non-negative integer k, because for k = ∞ we encounter the classical case of regular functions and quasi-coherent algebraic sheaves. Consider an affine scheme Y = Spec (A) with structure sheaf OY . Any A-module M determines a quasi-coherent sheaf fM on Y (cf. [19, Chap. II]). The functor M 7−→ fM gives an equivalence of categories between the category of A-modules and the category of quasi-coherent OY -modules. Its inverse is the global sections functor F 7−→ H 0(Y, F ) (op.cit., Chap. II, Corollary 5.5). This abuse of notation does not lead to confusion. We thus obtain the following version of Cartan’s Theorem A: egories between the category of Rk(K n)-modules and the category of inverse is the direct image functor ι∗. We say that F is a quasi-coherent k-regulous sheaf on K n if it is By a k-regulous sheaf F we mean a sheaf of Rk-modules. Through Corollaries 12.5 and 12.6, the functor ι−1 of restriction to K n gives and by Rk the sheaf of k-regulous function germs (in the k-regulous topology equal to the constructible topology) on K n. It follows directly Denote by eRk the structure sheaf of the affine scheme Spec(cid:0)Rk(K n)(cid:1) from Corollaries 12.5 and 12.2 that the restriction ι−1eRk of eRk to K n coincides with the sheaf Rk; conversely, ι∗ Rk = eRk. an equivalence of categories between eRk-modules and Rk-modules. Its the restriction to K n of a quasi-coherent eRk-module. Thus the functor ι−1 induces an equivalence of categories between quasi-coherent eRk- image functor ι∗. For any Rk(K n)-module M, we shall denote by fM both the associated sheaf on Spec(cid:0)Rk(K n)(cid:1) and its restriction to K n. Theorem 13.1. The functor M 7−→ fM gives an equivalence of cat- modules and quasi-coherent Rk-modules, whose inverse is the direct ALGEBRAIC GEOMETRY OVER VALUED FIELDS 39 quasi-coherent Rk-modules. Its inverse is the global sections functor F 7−→ H 0(K n, F ). In particular, every quasi-coherent sheaf F is generated by its global sections H 0(K n, F ). ✷ The regulous version of Cartan’s Theorem B, stated below, follows directly from the version for affine (not necessarily noetherian in view of Remark 12.7) schemes (cf. [18, Theorem 1.3.1]) via the discussed equivalence of categories (being the functor ι−1 of restriction to K n). Theorem 13.2. If F is a quasi-coherent k-regulous sheaf on K n, then H i(K n, F ) = 0 for all i > 0. ✷ Corollary 13.3. The global sections functor F 7−→ H 0(K n, F ) on the category of quasi-coherent k-regulous sheaves on K n is exact. ✷ Let V be a k-regulous closed subset of K n and I(V ) the sheaf of those k-regulous function germs on K n that vanish on V . It is a quasi- coherent sheaf of ideals of Rk, the sheaf Rk/I(V ) has support V and is generated by its global sections (Theorem A); moreover H 0(cid:0)K n, Rk/I(V )(cid:1) = H 0(cid:0)K n, Rk(cid:1) /H 0 (K n, I(V )) (Theorem B). The subset V inherits the k-regulous topology from K n and constitutes, together with the restriction Rk V of the sheaf Rk/I(V ) to V , a locally ringed space of K-algebras, called an affine k-regulous subvariety of K n. More generally, by an affine k-regulous variety we mean any locally ringed space of K-algebras that is isomorphic to an affine k-regulous subvariety of K n for some n ∈ N. We can define in the ordinary fashion the category of quasi-coherent Rk V -modules. Each such module extends trivially by zero to a quasi- coherent Rk-module on K n. The sections Rk V (V ) of the structure sheaf Rk It follows from Cartan’s V are called k-regulous functions on V . Theorem B that each k-regulous function on V is the restriction to V of a k-regulous function on K n. Hence we immediately obtain the following two results. Proposition 13.4. Let W and V be two affine k-regulous subvarieties of K m and K n, respectively. Then the following three conditions are equivalent: 40 KRZYSZTOF JAN NOWAK 1) f : W → V is a morphism of locally ringed spaces; 2) f = (f1, . . . , fn) : W → K n where f1, . . . , fn are k-regulous func- tions on W such that f (W ) ⊂ V ; 3) f extends to a k-regulous map K m → K n. We then call f : W → V a k-regulous map. ✷ Corollary 13.5. Let W , V and X be affine k-regulous subvarieties of K m, K n and K p, respectively. If two maps g : W → V and f : V → X are k-regulous, so is its composition f ◦ g. ✷ It is clear that Cartan’s theorems remain valid for quasi-coherent k-regulous sheaves on affine k-regulous varieties V . Theorem 13.6. The functor M 7−→ fM gives an equivalence of cat- egories between the category of Rk quasi-coherent Rk V -modules. Its inverse is the global sections functor V (V )-modules and the category of F 7−→ H 0(V, F ). In particular, every quasi-coherent sheaf F is generated by its global sections H 0(V, F ). ✷ Theorem 13.7. If F is a quasi-coherent k-regulous sheaf on V , then H i(V, F ) = 0 for all i > 0. ✷ Corollary 13.8. The global sections functor F 7−→ H 0(V, F ) on the category of quasi-coherent k-regulous sheaves on V is exact. ✷ Note that every non-empty k-regulous open subset U of K n is an affine k-regulous variety. Indeed, if U = D (f ) for a k-regulous function f on K n (Corollary 11.11), then U is isomorphic to the affine k-regulous subvariety V := Z (yf (x) − 1) ⊂ K n x × Ky. Remark 13.9. Consider a smooth algebraic subvariety X of the affine space KAn. We may look at the set V := X(K) of its K-points both as an algebraic variety X(K) and as a k-regulous subvariety V of K n. Every function f : V → K that is k-regulous on V in the second sense remains, of course, k-regulous on X(K) in the sense of the definition from the beginning of Section 11. ALGEBRAIC GEOMETRY OVER VALUED FIELDS 41 Open problem. The problem whether the converse implication is true for k > 0 is unsolved as yet. For k = 0 the answer is in the affirmative, which follows immediately from Theorem 10.2: given an arbitrary algebraic subvariety X of KAn, every continuous hereditarily rational function f on X(K) extends to a continuous rational function on K n, whence f is regulous on V . This theorem was proved for real and p-adic varieties in paper [24]. Finally, we wish to give a criterion for a continuous function to be regulous. It relies on Theorem 10.2 on extending continuous hereditar- ily rational functions on algebraic K-varieties. Theorem 13.10. Let V be an affine regulous subvariety of K n and f : V → K a function continuous in the K-topology. Then a necessary and sufficient condition for f to be a regulous function is the following: (*) For every Zariski closed subset Z of K n there exist an open Zariski dense subset U of the Zariski closure of V ∩ Z in K n and a regular function g on U such that f (x) = g(x) for all x ∈ V ∩ Z ∩ U. Proof. By Corollary 11.5, the necessary condition is clear, because f is the restriction to V of a regulous function on K n (Corollary 13.3 to Cartan’s Theorem B). In order to prove the sufficient condition, we proceed with induction with respect to the dimension d of the set V which is a closed (in the K-topology) constructible subset of K n. The case d = 0 is trivial. Assuming the assertion to hold for dimensions less than d, we shall prove it for d. So suppose V is of dimension d. By Proposition 11.2, the Zariski closure W of V in K n is of dimension d and we have W 0 := {x ∈ W : W is smooth of dimension d at x} ⊂ V. Obviously, W ∗ := W \ W 0 is a Zariski closed subset of K n of dimension less than d. Therefore Y := V ∩ W ∗ is a regulous closed subset of V of dimension less that d and W = W 0 ∪ W ∗ ⊂ V ∪ W ∗ ⊂ W. Since the restriction f Y satisfies condition (*), it is a regulous function on Y by the induction hypothesis. It is thus the restriction to Y of a regulous function F on K n (Corollary 13.3 to Cartan’s Theorem B). Further, the function f and the restriction F W ∗ can be glued to a function g : W = V ∪ W ∗ → K, g(x) =(cid:26) f (x) : x ∈ V F (x) : x ∈ W ∗ 42 KRZYSZTOF JAN NOWAK which satisfies condition (*) as well. Now, it follows from Theorem 10.2 that g extends to a regulous function G on K n. Since f is the restriction to V of the function G which is regulous, so is f , as required. (cid:3) Acknowledgements. The author wishes to express his gratitude to Wojciech Kucharz for several stimulating discussions on the topics of this paper, especially in the context of real algebraic geometry. References [1] A. Andreotti, E. Bombieri, Sugli omeomorfismi delle variet`a algebriche, Ann. Scuola Norm. Sup Pisa 23 (3) (1969), 431–450. [2] —, F. Norguet, La convexit´e holomorphe dans l’espace analytique des cycles d’une vari´et´e alg´ebrique, Ann. Scuola Norm. Sup. Pisa 21 (3) (1967), 31–82. [3] V. Berkovich, Spectral Theory and Analytic Geometry over Non-Archimedean Fields, Math. Surveys and Monographs 33, AMS, Providence, RI, 1990. [4] S. Besarab, Relative elimination of quantifiers for Henselian valued fields, Ann. Pure Appl. Logic 53 (1991), 51–74. [5] J. Bochnak, M. Coste, M.-F. Roy, Real Algebraic Geometry, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 36, Springer-Verlag, Berlin, 1998. [6] S. Bosch, U. Guntzer, R. Remmert, Non-Archimedian Analysis: a system- atic approach to rigid analytic geometry, Grundlehren der math. Wiss. 261, Springer-Verlag, Berlin, 1984. [7] N. Bourbaki, Alg`ebre Commutative, Hermann, Paris, 1962. [8] F. Bruhat, H. Cartan, Sur la structure des sous-ensembles analytiques r´eels, C. R. Acad. Sci. 244 (1957), 988–990. [9] G. Cherlin, Model Theoretic Algebra, Selected Topics, Lect. Notes Math. 521, Springer-Verlag, Berlin, 1976. [10] R. Cluckers, F. Loeser, b-minimality, J. Math. Logic 7 (2) (2007), 195–227. [11] —, L. Lipshitz, Fields with analytic structure, J. Eur. Math. Soc. 13 (2011), 1147–1223. [12] —, —, Strictly convergent analytic structures, arXiv:1312.5932 [math.LO] v.1 (2013). [13] J. Denef, p-adic semi-algebraic sets and cell decomposition, J. Reine Angew. Math. 369 (1986), 154–166. [14] —, L. van den Dries, p-adic and real subanalytic sets, Ann. Math. 128 (1988), 79–138. [15] G. Fichou, J. Huisman, F. Mangolte, J.-P. Monnier, Fonctions r´egulues, J. Reine Angew. Math., to appear. [16] B. Fisher, A note on Hensel’s lemma in several variables, Proc. Amer. Math. Soc. 125 (11) (1997), 3185–3189. [17] H. Grauert, R. Remmert, Analytische Stellenalgebren, Grundlehren der math. Wiss. 176, Springer-Verlag, Berlin, 1971. [18] A. Grothendieck, ´El´ements de G´eom´etrie Alg´ebrique. III. ´Etude cohomologique des faisceaux coh´erents, Publ. Math. IHES 11 (1961) and 17 (1963). [19] R. Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics 52, Springer-Verlag, 1977. ALGEBRAIC GEOMETRY OVER VALUED FIELDS 43 [20] E. Hrushovski, D. Kazhdan, Integration in valued fields; In: Algebraic Ge- ometry and Number Theory, Progr. Math. 253, Birkhauser Boston, Boston, MMA, 2006, 261—405. [21] —, F. Loeser, Non-Archimedean tame topology and stably dominated types, arXiv:1009.0252 [math.AG], v.3 (2012). [22] J. Koll´ar, Lectures on resolution of singularities, Ann, Math. Studies, Vol. 166, Princeton Univ. Press, Princeton, New Jersey, 2007. [23] —, Singularities of the minimal model program, (With a collaboration of S. Kov´acs) Cambridge Tracts in Mathematics, vol. 200. Cambridge Univer- sity Press, Cambridge, 2013. [24] —, K. Nowak, Continuous rational functions on real and p-adic varieties, Math. Zeitschrift 279 (2015), 85–97. [25] W. Kucharz, Continuous rational maps into the unit 2-sphere, Arch. Math. 102 (2014), 257–261. [26] —, Approximation by continuous rational maps into spheres, J. Eur. Math. Soc. (JEMS) 16 (2014), 1555–1569. [27] —, Continuous rational maps into spheres, arXiv: 1403.5127 [math.AG] (2014). [28] —, K. Kurdyka, Stratified-algebraic vector bundles, arXiv 1308.4376 [math.AG] (2013). [29] E. Kunz, Introduction to Commutative Algebra and Algebraic Geometry, Birkhauser, Boston, 1985. [30] K. Kurdyka, Ensemble semi-alg´ebriques sym´etriques par arcs, Math. Ann. 282 (1988), 445–462. [31] L. Lipshitz, Z. Robinson, Uniform properties of rigid subanalytic sets, Trans. Amer. Math. Soc. 357 (11) (2005), 4349–4377. [32] S. Lojasiewicz, Ensembles Semi-analytiques, I.H.E.S., Bures-sur-Yvette, 1965. [33] J. Milnor, Singular points of complex hypersurfaces, Princeton Univ. Press, Princeton, New Jersey, 1968. [34] K.J. Nowak, On the Euler characteristic of the links of a set determined by smooth definable functions, Ann. Polon. Math. 93 (2008), 231–246. [35] —, Supplement to the paper ”Quasianalytic perturbation of multiparameter hy- perbolic polynomials and symmetric matrices” (Ann. Polon. Math. 101 (2011), 275–291), Ann. Polon. Math. 103 (2012), 101-107. [36] —, Some results from algebraic geometry over complete discretely valued fields, arXiv:1311.2051 [math.AG] (2013). [37] —, Some results from algebraic geometry over Henselian real valued fields, arXiv:1312.2935 [math.AG] (2013). [38] J. Pas, Uniform p-adic cell decomposition and local zeta functions, J. Reine Angew. Math. 399 (1989), 137–172. [39] —, On the angular component map modulo p, J. Symb. Logic 55 (1990), 1125– 1129. [40] A. Robinson, Complete Theories, North-Holland, Amsterdam, 1956. [41] P. Scowcroft, L. van den Dries, On the structure of semi-algebraic sets over p-adic fields, J. Symbolic Logic 53 (4) (1988), 1138–1164. [42] A.H. Wallace, Algebraic approximation of curves, Canadian J. Math. 10 (1958), 242–278. 44 KRZYSZTOF JAN NOWAK [43] O. Zariski, P. Samuel, Commutative Algebra, Vol. II, Van Nostrand, Princeton, 1960. Institute of Mathematics Faculty of Mathematics and Computer Science Jagiellonian University ul. Profesora Lojasiewicza 6 30-348 Krak´ow, Poland e-mail address: [email protected]
0904.4204
2
0904
2010-02-28T22:39:51
Birational modifications of surfaces via unprojections
[ "math.AG", "math.AC" ]
We describe elementary transformations between minimal models of rational surfaces in terms of unprojections. These do not fit into the framework of Kustin-Miller unprojections as introduced by Papadakis and Reid, since we have to leave the world of projectively Gorenstein varieties. Also, our unprojections do not depend on the choice of the unprojection locus only, but need extra data corresponding to the choice of a divisor on this unprojection locus.
math.AG
math
BIRATIONAL MODIFICATIONS OF SURFACES VIA UNPROJECTIONS CHRISTIAN LIEDTKE AND STAVROS ARGYRIOS PAPADAKIS March 1, 2010 Abstract. We describe elementary transformations between minimal mod- els of rational surfaces in terms of unprojections. These do not fit into the framework of Kustin–Miller unprojections as introduced by Papadakis and Reid, since we have to leave the world of projectively Gorenstein varieties. Also, our unprojections do not depend on the choice of the unprojection locus only, but need extra data corresponding to the choice of a divisor on this unprojection locus. 0 1 0 2 b e F 8 2 ] . G A h t a m [ 2 v 4 0 2 4 . 4 0 9 0 : v i X r a Introduction Unprojection, introduced by Miles Reid in [R2], is a technique to describe birational transformations in higher dimensional geometry explicitly in terms of commutative algebra. As explained in [PR, Section 2.3], the prototype and easiest example of an unprojection is the Castelnuovo blow-down of a rational (−1)-curve lying on a smooth cubic surface in P3 as the inverse of a projec- tion from a del Pezzo surface of degree 4 in P4, which also explains the name unprojection. Since then, unprojections have been applied to many explicit constructions in birational geometry of surfaces and 3-folds, compare [ABM], [CPR], [RS],... Moreover, these techniques have also proved to be useful for the construction of key varieties, see, for example, [R2] or [NP]. Finally, the general theory of unprojection has been developed further by the second author and a general framework has been proposed in [P]. In this note we return to the two-dimensional case. Here, every birational map between two smooth projective surfaces can be factored into a sequence of blow-ups of points and Castelnuovo blow-downs of rational (−1)-curves. A prominent class of birational transformations between smooth surfaces are the elementary transformations, which relate minimal models of surfaces of Kodaira dimension κ = −∞ to each other, compare [B, Chapter 12]. Their higher- dimensional generalisations are Sarkisov links, compare [CPR]. Question. Can one describe elementary transformations of surfaces in terms of unprojections? For reasons of simplicity and since all relevant problems already occur within this class of surfaces, we will restrict ourselves to minimal rational surfaces. Already here we encounter new types of unprojections and new phenomena 2000 Mathematics Subject Classification. 14M05, 14E05, 14J26, 13H10. 1 2 CHRISTIAN LIEDTKE AND STAVROS ARGYRIOS PAPADAKIS show up. This is related to the fact that an elementary transformation depends not only on the choice of a curve but needs also the choice of a point on it. Answer. Yes, this can be done for minimal rational surfaces. However, the unprojection is no longer determined by the unprojection locus alone. Moreover, this cannot be done within the framework of projectively Gorenstein varieties, as in the classical case of Kustin–Miller unprojections. Minimal rational surfaces consist of P2 and Hirzebruch surfaces. By defi- nition, the Hirzebruch surface, sometimes also called Segre surface, Fd is the P1-bundle P(OP1 ⊕ OP1(d)) → P1. An elementary transformation of Fd is the following: we choose a point lying on a fibre of this projection and blow it up. The strict transform of this fibre on the blow-up is a rational (−1)-curve and blowing it down we obtain the desired elementary transformation of Fd. Depending on the position of the point we blew up to start with, the resulting surface is isomorphic to Fd+1 or Fd−1. As embedding for the Hirzebruch surfaces we choose their realisations as surfaces of minimal degree, i.e., as scrolls. As unprojection locus Γ we choose a line lying on this scroll, which corresponds to the fibre of the projection of this Hirzebruch surface onto P1. In terms of rings and ideals we have V (I) = Γ ⊂ Proj S ⊂ PN . The point of departure for unprojections is the S-module HomS(I, S), which in our situation turns out to be generated by two elements in degree zero. This implies that the associated unprojection ring is ”not geometric“, although somewhat similar rings have been considered in [R2] to describe 3-fold flips. Geometrically, this is related to the fact that unprojections correspond to con- tractions and that Γ has not negative self-intersection, which would be necessary in order to contract it. Instead, we will use natural submodules of HomS(I, S) to construct our un- projections. Geometrically, these submodules correspond to choosing a divisor D on Γ = V (I). In case D is a divisor of degree k ≥ 1, whose support consists of k distinct points, our results specialise to the following Theorem. The unprojection of X = Proj S ⊂ PN with respect to D ⊂ Γ ⊂ X is a normal and projectively Cohen–Macaulay surface inside P(1N +1, k), which arises from X by first blowing up D and then contracting the strict transform of Γ. In this special case the unprojection is smooth outside a toric singularity k (1, 1), which is induced from the singularity of the ambient weighted of type 1 projective space. We refer to Section 2 for the general case. The case k = 1 corresponds to elementary transformations of Hirzebruch surfaces. Moreover, if Proj S ∼= Fd and we vary the divisor D, which is just one point in this case, we obtain a 1-parameter family of unprojections, all of which are isomorphic to Fd−1 except one surface which is isomorphic to Fd+1. This fits nicely into the deformation and degeneration theory of Hirzebruch surfaces, confer [BHPV, Theorem VI.8.1]. BIRATIONAL MODIFICATIONS OF SURFACES 3 Finally, we give an application to odd Horikawa surfaces, i.e., to minimal sur- faces of general type with K 2 = 2pg − 3. More precisely, for an odd Horikawa surface with pg ≥ 7 the canonical and the bicanonical image are rational sur- faces. Moreover, the canonical image is a Hirzebruch surface realised as surface of minimal degree. Then the birational transformation relating canonical and bicanonical image corresponds to an unprojection of the type considered in this article with k = 2. Acknowledgement. We thank Francesco Zucconi and the referee for remarks, comments and pointing out a couple of inaccuracies. Stavros Papadakis is a participant of the Project PTDC/MAT/099275/2008, a member of CAMSGD (IST/UTL), and gratefully acknowledges funding from the Portuguese Funda¸cao para a Ciencia e a Tecnologia (FCT) under research grant SFRH/BPD/22846/2005 of POCI2010/FEDER. Much of this work was done while he was visiting the university of Dusseldorf. We thank the Mathematisches Institut and the DFG- Forschergruppe Classification of Algebraic Surfaces and Compact Complex Mani- folds for the kind hospitality and financial support. 1. Hirzebruch surfaces and scrolls We fix once and for all an arbitrary field k over which all our schemes will be defined. Let d ≥ 0 be a non-negative integer. Then the Hirzebruch surface, or, Segre surface, Fd is defined to be the P1-bundle P(OP1 ⊕ OP1(d)) → P1. We denote by Γ the class of a fibre of this projection. Moreover, there exists a section ∆0 with self-intersection −d, which is unique if d 6= 0. We remark that F0 is isomorphic to P1 × P1, and that F1 is isomorphic to P2 blown-up in one point. Moreover, F1 is the only Hirzebruch that is not minimal, and among Hirzebruch surfaces, the only del Pezzo surfaces are F0 and F1. Now, a projectively Cohen–Macaulay scheme X is projectively Gorenstein if and only if there exists a k ∈ Z such that ωX ∼= OX (k), see [Eis, Section 21.11]. Thus, apart from F0 and F1, Hirzebruch surfaces do not possess embeddings into projective space that are projectively Gorenstein. In particular, elementary transformations of minimal rational surfaces in terms of unprojections cannot be described within the framework of Kustin–Miller unprojections as in [PR]. However, Hirzebruch surfaces do possess nice embeddings into projective space. Namely, for integers m, n satisfying n ≥ m ≥ 1 we define F(m, n) to be the surface scroll in Pm+n+1 defined by the vanishing of the 2 × 2-minors of (1) (cid:18) x00 x01 ... x0m−1 x10 x11 ... x0m ... x1n−1 ... x1n (cid:19) . Abstractly, this scroll is isomorphic to Fn−m. Moreover, F(m, n) corresponds to embedding Fd with d = n − m via the complete linear system ∆0 + nΓ into projective space. Under this embedding, the projection onto P1 is given by the ratios of the columns of (1) F(m, n) [x00 : ... : x0m : x10 : ... : x1,n] → 7→ [x00 : x01] = ... = [x1n−1 : x1n] . P1 4 CHRISTIAN LIEDTKE AND STAVROS ARGYRIOS PAPADAKIS Let us fix the fibre Γ over [0 : 1], which is given by the vanishing of the first row in (1): (2) Γ = {x00 = ... = x0m−1 = x10 = ... = x1n−1 = 0} ∩ F(m, n) . Apart from the nice determinantal description there is another reason why these embeddings of the Hirzebruch surfaces are distinguished. Namely, a non- degenerate and integral surface in PN has degree at least N − 1. If such a surface has degree equal to N − 1, then a theorem of del Pezzo states that it is precisely one of the F(m, n)’s above, P2, the Veronese surface in P5, or the cone over a rational normal curve, confer [EH] for a modern account. The case of the cone over a rational normal curve corresponds to having only one block in the matrix (1) above. We set R = k[x0i, x1j ] with i = 0, ..., m and j = 0, ..., n, which is the ho- mogeneous coordinate ring of Pm+n+1. Let Q be the ideal of R corresponding to F(m, n), i.e., the homogeneous ideal generated by the 2 × 2 minors of (1). We set S = R/Q and define I to be the ideal of S defining Γ, i.e., the ideal corresponding to (2) I = (x00, ..., x0m−1, x10, ..., x1n−1) ⊂ S = R/Q . The ring S is Cohen–Macaulay, which is related to the fact that the embedding of F(m, n) into Pm+n+1 is given by a complete linear system, confer [Eis, Ex- ercise 18.16]. Alternatively, it also follows from the determinantal description of S by Eagon’s theorem, see [Eis, Theorem 18.18]. Since we want to unproject from Γ = V (I), we need an analysis of the graded S-module HomS(I, S), compare [P, Section 4]. From loc.cit. or [PR, Section 1] we recall the short exact sequence (3) 0 −→ S −→ HomS(I, S) res −→ Ext1 S(S/I, S) −→ 0 , where res stands for Poincar´e residue map. Obviously, the inclusion ı of I into S lies in HomS(I, S). Moreover, the map φ : x0i x1j 7→ x0i+1 7→ x1j+1 i = 0, ..., m − 1 j = 0, ..., n − 1 sending the first row of (1) to the second row defines a homomorphism of degree zero of graded S-modules from I to S. This is best seen by considering the element s = x01/x00 = ... = x1n/x1n−1 in the field of fractions k(S) of the domain S. Then multiplication by s induces a homomorphism of S-modules from I to k(S) which yields φ. Proposition 1.1. The S-module HomS(I, S) is generated by the two elements ı and φ, both of which are of degree zero. Since this module is crucial for the construction and analysis of unprojections, we decided to give two proofs – one more geometric and one purely algebraic: First Proof. Let X = Proj S together with its very ample invertible sheaf OX (1). From the determinantal description we infer that X is projectively normal, which implies S = Ln∈Z H 0(X, OX (n)). Moreover, the sheafification BIRATIONAL MODIFICATIONS OF SURFACES 5 of HomS(I, S) is OX (Γ), which implies that there is a natural injection of graded S-modules α : HomS(I, S) → Mn∈Z H 0(X, OX (Γ)(n)) =: M. Thus, there are no elements of negative degree in HomS(I, S). In degree zero, we have ı and φ in HomS(I, S) and h0(X, OX (Γ)) = 2, which implies that α is an isomorphism in degree zero. Using the explicit description of global sections of invertible sheaves on scrolls in terms of bihomogeneous polynomials as in [R, Chapter 2], it follows easily that M is generated as an S-module in degree zero. It follows that α is an isomorphism of graded S-modules and hence that HomS(I, S) is generated by ı and φ. (cid:3) Second Proof. Denote by B the following subset of R 0m · xb B = {1} ∪ {x0i · xa 1n 0m · xb 1n 10 · xb 1n 10 · x1i · xb 1n ∪ {xa ∪ {xa ∪ {xa 2 ≤ i ≤ m − 1, a, b ≥ 0} a, b ≥ 0, (a, b) 6= (0, 0)} a, b ≥ 0, (a, b) 6= (0, 0)} 1 ≤ i ≤ n − 1, a, b ≥ 0} . First claim: The set B is a basis of the k-vector space R/(Q + (x00, x01)). Set Q1 = Q + (x00, x01). Using the relations x0ix0j = x0i−1x0j+1, x1ix1j = x1i−1x1j+1 and x0ix1j = x0i−1x1j+1 of Q1 it is not difficult to see that given a monomial w ∈ R there exists another monomial w′ ∈ R with w − w′ ∈ Q1 such that w′ ∈ B. This shows that B spans R/Q1 as k-vector space. To prove linear independence, we consider the k-algebra homomorphism g : R → k[z, s, t] defined by g(x0i) = ztm−isi, 0 ≤ i ≤ m and g(x1j ) = tn−jsj, 0 ≤ j ≤ n . Now, assume that we have an element a = Xb∈B ab · b ∈ Q1, where almost all ab = 0 . Using Q ⊆ ker g, we see that g(a) = Pb abg(b) lies inside the ideal of k[z, s, t] generated by ztm−1. On the other hand, none of the monomials g(b), b ∈ B is divisible by ztm−1. Hence if g(a) 6= 0 we get a contradiction, so g(a) = 0. Since it is clear that the set {g(b) b ∈ B} is linearly independent, we get ab = 0 for all b ∈ B and we conclude linear independence of B. This proves the claim. Second claim: If u ∈ R fulfills ux1n−1 ∈ (x00) + Q then u ∈ Q1. Changing by elements of Q1 and using the first claim, we may assume that u is of the form u = Pb∈B abb with almost all ab = 0. By assumption we have g(u)tsn−1 = g(ux1n−1) ∈ (ztm), hence g(u) ∈ (ztm−1). However, we have seen in the proof of the first claim that this implies ab = 0 for all b ∈ B and proves the second claim. Finally, we prove our assertion about HomS(I, S): Since S = R/Q is a domain, the element x00 is S-regular. Moreover, since I is an ideal of S and x00 is S-regular, it follows that the S-module homomorphism HomS(I, S) → S given by f 7→ f (x00), is injective. Now, let f ∈ HomS(I, S) and set u = 6 CHRISTIAN LIEDTKE AND STAVROS ARGYRIOS PAPADAKIS f (x00) ∈ S. We are done if we show that u lies inside the ideal generated by x00 and x01 of S. However, this follows from the computation ux1n−1 = f (x00)x1n−1 = f (x00x1n−1) = x00f (x1n−1) ∈ (x00) ⊆ S. together with the second claim above. (cid:3) Remark 1.2. In fact, ı and φ are defined over the integers. Since they generate HomS(I, S) over any field, in particular over all prime fields, it follows that Proposition 1.1 holds in fact over the integers. Since HomS(I, S) has two generators in degree zero, the Kustin–Miller unpro- jection with respect to the whole S-module HomS(I, S) yields a graded ring, whose component of degree zero is a vector space of dimension at least two, i.e., the unprojection ring is ”not geometric“. Although even negatively graded rings occur in the description of 3-fold flips [R2, Section 11], we will use natural submodules of HomS(I, S) instead. A geometric interpretation why the unprojection ring associated to the whole S-module HomS(I, S) does not give the ”right“ object is the following obser- vation: the unprojection locus Γ = V (I) is a curve with self-intersection zero, whereas for the existence of a morphism contracting Γ we would need that Γ has negative self-intersection. 2. Generalised unprojections We keep the notations introduced so far. As already noted before, taking the unprojection ring with respect to the whole of HomS(I, S) yields a graded ring, which is not ”geometric“, which is why we consider suitable submodules. In view of the natural short exact sequence (3) we will consider submodules of HomS(I, S) of the form res−1(N ), where N is a submodule of Ext1 S(S/I, S). Recall that HomS(I, S) is generated as S-module by two elements ı, φ in our setup by Proposition 1.1. Then, in case N is a cyclic S-module we are led to considering submodules of HomS(I, S) that are generated by ı and another element f φ, where f ∈ S is a homogeneous element. Motivated by [PR] and [P] we define Definition 2.1. Let S be the homogeneous coordinate ring of F(m, n) inside Pm+n+1 and let f ∈ S be homogeneous of degree k ≥ 1. The generalised unprojection ring of S with respect to the unprojection ideal I and to f is defined as Sun(f ) = S[T ] (T u − f φ(u), u ∈ I) , where T is a variable of degree k. Lemma 2.2. Let f1, f2 be homogeneous elements of S of the same degree with f1 − f2 ∈ I. Then there exists an isomorphism of graded rings In particular, if f1 ∈ I then Sun(f1) is not a domain. Sun(f1) ∼= Sun(f2) . BIRATIONAL MODIFICATIONS OF SURFACES 7 Proof. Let i = f2 − f1, which is an element of I by assumption. Then for all u ∈ I we calculate T u − f2φ(u) = T u − (f1 + i)φ(u) = (T − φ(i)) u − f1φ(u). Thus a change of variables from T to T − φ(i) (note that both elements are of the same degree) yields the desired isomorphism of graded rings. (cid:3) Remark 2.3. That Sun(f ) depends only on the submodule of HomS(I, S) generated by ı and f φ and not on the particular choice of generators also follows from the intrinsic setup of [P, Section 2]. The previous lemma thus tells us that there is no loss of generality choosing f to be a homogeneous polynomial in x0m and x1n. Since x0m and x1n are homogeneous coordinates on Γ = V (I) we remark the following. Remark 2.4. If the ground field is algebraically closed then choosing a homo- geneous element f of degree k is equivalent to choosing k points (counted with multiplicities) on Γ. Theorem 2.5. Let f 6= 0 be homogeneous of degree k ≥ 1 in x0m and x1n. Then Proj Sun(f ) is an integral, normal, and projectively Cohen–Macaulay sur- face inside weighted projective space P(1m+n+2, k). Its homogeneous ideal is generated by the 2 × 2-minors of the matrix (4) (cid:18) x00 x01 ... x0m−1 x10 x11 ... x0m ... x1n−1 x1n ... f T (cid:19) , where the xij are of degree one and T is of degree k. Proof. The description of the homogeneous ideal follows directly from the pre- sentation of S as the vanishing of the 2 × 2 minors of (1) and the definition of Sun(f ). Let Q2 be the ideal of R2 = k[xij, T ] generated by the 2×2 minors of (4). We want to show that if P is a minimal prime ideal over Q2 then it has codimension equal to m + n, which is the maximum possible by a result of Eagon, compare [Eis, Exercise 10.9]. Denote by I e the ideal of R2 generated by the subset I + Q2. First, as- sume that I e ⊆ P , which implies codim(P ) ≥ codim(I e). However, I e contains x00, ..., x0m−1, x10, ..., x1n−1, f x0m, which form a regular sequence, which im- plies codim(I e) ≥ m + n + 1. Hence this case does not exist and we have I e 6⊆ P , i.e., V (P ) ∩ (V (Q2) − V (I e)) 6= ∅. By [NP2, Remark 2.5], the inclu- sion of rings induces an isomorphism Spec Sun(f ) − V (I e) ∼= Spec S − V (I). Since Spec S − V (I) is irreducible of dimension three, we see that P has codi- mension m + n. Thus, every minimal prime over Q2 has codimension m + n, which means that Sun(f ) is a determinantal ring and such rings are known to be Cohen–Macaulay by a result of Eagon, compare [Eis, Theorem 18.18]. By the above arguments, the irreducible open subset Spec Sun(f ) − V (I e) of Spec Sun(f ) meets every irreducible component of Spec Sun(f ). From this we conclude that Spec Sun(f ) is irreducible. From the isomorphism Spec Sun(f ) − V (I e) ∼= Spec S − V (I) it follows that Sun(f ) is generically reduced. In particular, being Cohen–Macaulay there are 8 CHRISTIAN LIEDTKE AND STAVROS ARGYRIOS PAPADAKIS no embedded components and it follows that Sun(f ) is reduced. Together with the irreducibility it follows that Sun(f ) is a domain. Using the isomorphism Spec Sun(f ) − V (I e) ∼= Spec S − V (I) once more, we obtain normality outside V (I e). A straightforward calculation using the Jacobian criterion shows normality along V (I e). Thus, Sun(f ) is normal. (cid:3) Theorem 2.6. Let : X → X = Proj S be the blow-up of the ideal (I, f ). Then there exists a factorisation ysssssssssss X 'OOOOOOOOOOOOO cont Γ X = Proj S /_______ Xun = Proj Sun(f ) , where contΓ is contraction of the strict transform of Γ = V (I) on X. Proof. Considered as an ideal of R = k[xij], the ideal (I, f ) is generated by the regular sequence x00, ..., x0m−1, x10, ..., x1n−1, f . By [EH, Exercise IV-26], the Rees algebra R of R with respect to (I, f ) is isomorphic to R[T00, ..., T0m−1, T10, ..., T1n−1, Tf ]/B, where the Tij and Tf are indeterminants and where B is the ideal generated by the 2 × 2 minors of (cid:18) x00 T00 ... x0m−1 x10 ... T0m−1 T10 ... x1n−1 ... T1n−1 Tf (cid:19) . f Taking Proj , we obtain the blow-up : X → X. We denote by Γ the strict transform of Γ, which is cut out by x00 = ... = x0m−1 = 0, x10 = ... = x1n−1 = 0, T00 = ... = T0m−1 = 0, and T10 = ... = T1n−1 = 0. On the level of commutative algebra, corresponds to eliminating the T0i’s, the T1j’s and Tf , whereas eliminating only the T0i’s and the T1j’s induces a map cont Γ from X onto Xun. A straightforward calculation shows that contΓ is in fact a morphism, that it is an isomorphism outside Γ and that it contracts Γ to the vertex of the weighted projective space in which Xun lies. Since Xun is normal by Theorem 2.5, Zariski’s main theorem shows that cont Γ is in fact the contraction of Γ. (cid:3) Let us assume that D = Pi kiPi where the ki are positive integers with k = Pi ki and where the Pi are distinct points that are rational over the ground field. Note that this assumption on D can always be fulfilled if the ground field is algebraically closed. Then calculations similar to those in [EH, Chapter IV.2.3] show that we obtain Xun = Proj Sun(f ) as follows: (1) For each i, blow up Proj S at Pi. Then blow up the intersection point of the strict transform of Γ with the resulting (−1)-curve of the blow-up etc. until, for every i we get a chain Ci of (ki − 1) rational (−2)-curves and a (−1)-curve. It is understood that Ci is empty if ki = 1. (2) The strict transform Γ is a rational curve with self-intersection −k. Contracting Γ and all the Ci’s we obtain Xun. y ' / BIRATIONAL MODIFICATIONS OF SURFACES 9 From this description we can read off the singularities of Xun: it has a toric sin- k (1, 1) coming from the contraction of Γ, and every contracted gularity of type 1 chain Ci contributes a cyclic quotient singularity of type 1 (1, ki − 1), i.e., a ki Du Val singularity of type Aki−1. 3. Elementary transformations As in the previous sections, let S be the homogeneous coordinate ring of F(m, n) inside Pm+n+1 given by (1) and recall that we assumed n ≥ m ≥ 1. As unprojection divisor we take the line Γ = V (I) lying on F(m, n) as in (2). We note that x0m and x1n can be viewed as coordinates on Γ. Moreover, in our unprojection setting we choose 0 6= f ∈ S of degree one, which for our unprojection purposes we may assume to be of the form fa,b = a x0m + b x1n with [a : b] ∈ P1, cf. Lemma 2.2. As already noted in Remark 2.4, our unprojection data consists of an unprojection locus, which is a line, and a point on this line. Proposition 3.1. There exists an isomorphism Proj Sun(fa,b) ∼= (cid:26) F(m, n + 1) F(m + 1, n) if [a : b] = [0 : 1] else, which is induced by a projective linear transformation of the ambient Pm+n+2. Proof. If a = 0 or b = 0 this follows directly from comparing (4) with (1). We may thus assume a 6= 0. For all i = 0, ..., n − 1 we add b times the (n − i).th column of the middle block of (4) to the (m − i).th column of the left block of (4). This is possible since we assumed n ≥ m and a linear change of variables yields the desired isomorphism. (cid:3) Remark 3.2. By Theorem 2.6, we have realised all elementary transformations of Hirzebruch surfaces in our setting. Moreover, we obtain a family of unprojections parametrised by P1. One member of this family is isomorphic to Fn−m+1, whereas all the others are isomorphic to Fn−m−1. This fits nicely into the deformation and degeneration theory of Hirzebruch surfaces as explained in [BHPV, Theorem VI.8.1]. The inverse of the unprojection Proj S 99K Proj Sun(fa,b), corresponds to eliminating the new variable T of Sun(fa,b), and is induced by a projection from Pm+n+2 onto Pm+n+1, confer [Ha, Proposition 8.20]. 4. (Bi)canonical images of Horikawa surfaces In this final section we work over the complex numbers. If S is a minimal surface of general type then Noether’s inequality K 2 ≥ 2pg − 4 holds true, confer [BHPV, Theorem VII.3.1]. In case of equality K 2 = 2pg − 4, i.e., if S is a so-called even Horikawa surface, then the canonical map is a generically finite morphism of degree 2 onto a surface of minimal degree in Ppg−1, which is the key to the classification of these surfaces, compare [BHPV, Chapter VII.9]. Also, it is not difficult to show that the bicanonical map is a morphism that coincides with the canonical map followed by the second Veronese embedding. 10 CHRISTIAN LIEDTKE AND STAVROS ARGYRIOS PAPADAKIS Now, let S be an odd Horikawa surface, i.e., a minimal surface of general type with K 2 = 2pg − 3. These have been classified in [Ho], and we will assume that we are in case A in Horikawa’s terminology with smooth canonical image. This is the generic case, and if pg ≥ 7 it is even automatically fulfilled, compare [Ho, Theorem 1.3]. Then the image of the canonical map is a smooth surface X of minimal degree in Ppg −1. Hence there exist integers n ≥ m ≥ 1 with pg = m + n + 2 such that the canonical image is F(m, n), which is abstractly isomorphic to Fn−m. The canonical system KS has a unique base point, whose indeterminacy is resolved by a single blow-up S → S. The resulting (−1)-curve on S maps to a line Γ ⊂ F(m, n) ⊂ Ppg −1 and after an appropriate choice of coordinates we may assume that Γ and F(m, n) are as in Section 1. Then S determines two points x, y on Γ, possibly infinitely near, cf. [Ho, Theorem 1.3]. We denote by π : X → X their blow-up, by Ex, Ey the corresponding exceptional divisors, and by Γ the strict transform of Γ on X. Moreover, we obtain a factorisation S → S ∗ → X → X, where S → S ∗ is birational, S ∗ has at worst Du Val singularities, and where S ∗ → X is finite and flat of degree 2. On X we consider the line bundle L = O X (π∗∆0 + (n − 4)π∗Γ − 2Ex − 2Ey) . Then the canonical map of S (and hence S) factors over the complete linear system L on X and we already noted that we can identify its image with F(m, n). Moreover, the (−1)-curve on S maps to Γ on X and thus maps to Γ under the canonical map. From [Ho, Theorem 1.3] it follows easily that the bicanonical map factors over L⊗2 on X. This map contracts Γ to an A1-singularity. A straightforward computation counting the quadratic relations coming the 2×2 minors of (1), we see that the map H 0(L)⊗2 → H 0(L⊗2) has 1-dimensional coimage. This implies that the canonical ring of S has pg generators in degree one and precisely one new generator in degree two. Let us denote by X1 ⊂ P(1pg ) and by by X2 ⊂ P(1pg , 2) the projection from the canonical model of S onto the weighted projective space correspond- ing to generators in degree 1 (canonical image) and generators in degree ≤ 2 (weighted bicanonical image). Then, putting all observations above together we can interpret Horikawa’s results [Ho, Section 1] in our setting as follows: Proposition 4.1. The odd Horikawa surface S determines a line Γ on X1 and two points {x, y} on this line, which are possibly infinitely near. Then the inverse of the natural projection X2 → X1, X1 99K X2 is a generalised unprojection with unprojection data Γ and divisor D = x + y on Γ. Finally, we remark that we cannot realise the other birational modifications of Theorem 2.6 by images of pluricanonical maps of surfaces of general type: If the canonical map has two-dimensional image then pg ≥ 3 and in order to BIRATIONAL MODIFICATIONS OF SURFACES 11 get a scroll as canonical image we even need pg ≥ 4. For such surfaces we have K 2 ≥ 4 by Noether’s inequality. However, for minimal surfaces of general type X with K 2 X ≥ 3 all pluricanonical images im(ϕi(X)) with i ≥ 3 are birational to X by Bombieri’s theorem (confer [BHPV, Theorem VII.5.1]) and thus no rational surfaces. References [ABM] S. Altınok, G. Brown, M. Reid, Fano 3-folds, K3 surfaces and graded rings, Topol- ogy and geometry: commemorating SISTAG, 25-53, Contemp. Math. 314, Amer. Math. Soc., Providence, RI, (2002). L. Badescu, Algebraic Surfaces, Universitext, Springer (2001). [B] [BHPV] W. P. Barth, K. Hulek, C. Peters, A. van de Ven, Compact Complex Surfaces, 2nd [CPR] [Eis] [EH] [EH2] [Ha] [Ho] [NP] [NP2] [PR] [P] [R] [R2] [RS] edition, Erg. d. Math., 3. Folge Volume 4, Springer (2004). A. Corti, A. Pukhlikov, M. Reid, Fano 3-fold hypersurfaces, Explicit birational geometry of 3-folds, 175-258, London Math. Soc. Lecture Note Ser. 281, Cambridge University Press (2000). D. Eisenbud, Commutative Algebra With a View Toward Algebraic Geometry, Grad- uate Texts in Mathematics 150, Springer (1995). D. Eisenbud, J. Harris, On Varieties of Minimal Degree (A Centennial Account), Algebraic Geometry, Bowdoin 1985, Proc. Symp. Pure Math. 46, Part 1, 3-13 (1987). D. Eisenbud, J. Harris, The Geometry of Schemes, Graduate Texts in Mathematics 197, Springer (2000). J. Harris, Algebraic Geometry. A First Course, Graduate Texts in Mathematics 133, Springer (1992). E. Horikawa, Algebraic surfaces of general type with small c2 121-155 (1976). J. Neves, S. Papadakis, A construction of numerical Campedelli surfaces with tor- sion Z/6Z, arXiv:0707.0244, to appear in Trans. AMS. J. Neves, S. Papadakis, Parallel Kustin–Miller unprojections with an application to Calabi–Yau geometry, preprint, arXiv:0903.1335v1 (2009). S. Papadakis, M. Reid, Kustin–Miller unprojection without complexes, J. Algebraic Geom. 13, 563-577 (2004). S. Papadakis, Towards a general theory of unprojection, J. Math. Kyoto Univ. 47, 579-598 (2007). M. Reid, Chapters on Algebraic Surfaces, IAS/Park City Math. Series, Volume 3, Amer. Math. Soc., Providence, RI (1997). M. Reid, Graded Rings and Birational Geometry in Proc. of algebraic Symposium (Kinosaki Oct. 2000), K. Ohno (ed.) 1-72, from http://www.maths.warwick.ac.uk/∼miles/3folds M. Reid, K. Suzuki, Cascades of projections from log del Pezzo surfaces, Number theory and algebraic geometry, 227–249, London Math. Soc. Lecture Note Ser. 303, Cambridge University Press (2003). 1, II, Invent. Math. 37, available Christian Liedtke Department of Mathematics Stanford University 450 Serra Mall Stanford, CA 94305, USA e-mail: [email protected] 12 CHRISTIAN LIEDTKE AND STAVROS ARGYRIOS PAPADAKIS Stavros Papadakis, Center for Mathematical Analysis, Geometry, and Dynamical Systems Departamento de Matem´atica, Instituto Superior T´ecnico Universidade T´ecnica de Lisboa 1049-001 Lisboa, Portugal e-mail: [email protected]
1612.03861
4
1612
2018-01-25T16:41:01
Newton-Okounkov bodies and Toric Degenerations of Mori dream spaces via Tropical compactifications
[ "math.AG" ]
Given a smooth Mori dream space $X$ we construct a model dominating all the small $\mathbb{Q}$-factorial modifications via tropicalization. This construction allows us to recover a Minkowski basis for the Newton-Okounkov bodies of divisors on $X$ and hence the movable cone of $X$. The existence of such basis allows us to prove the polyhedrality of the global Newton-Okounkov body and the existence of toric degenerations.
math.AG
math
NEWTON-OKOUNKOV BODIES AND TORIC DEGENERATIONS OF MORI DREAM SPACES VIA TROPICAL COMPACTIFICATIONS ELISA POSTINGHEL AND STEFANO URBINATI Abstract. Given a smooth Mori dream space X we construct a model dom- inating all the small Q-factorial modifications via tropicalization. This con- struction allows us to recover a Minkowski basis for the Newton-Okounkov bodies of divisors on X and hence the movable cone of X. The existence of such basis allows us to prove the polyhedrality of the global Newton-Okounkov body and the existence of toric degenerations. 1. Introduction In this work we make a connection between the theory of Newton-Okounkov bodies and tropical geometry, sharing the aim of the recent preprints [13, 15]. The results of the present parer yield a different point of view on how tropical geometry can be extremely helpful in birational geometry. Our principal aim is to study Mori dream spaces (MDS) via tropicalization. Via this connection we will describe a simple and computable way of reconstructing the movable cone of such varieties. The results obtained complete the picture intro- duced in [20, 22], where Minkowski bases for Newton-Okounkov bodies were given respectively for surfaces and for toric varieties, with respect to certain admissible flags. Mori dream spaces are special projective varieties for which the Minimal Model Program is particularly simple, since they only admit a finte number of small Q- factorial modifications (SQM's). The key property of these varieties is that they always admit a particularly nice embedding into toric varieties, see [8]. Given a MDS X and such an embedding X ⊆ Z into a toric variety, let T ⊆ Z be the maximal torus of the given toric variety. The main result of this paper, that is Theorem 3.1, can be summarized as follows: Theorem 1.1. Let X ⊂ Z be a MDS with the embedding of [8] into a toric variety Z. Then the tropicalization of the variety restricted to T , Trop(XT ), induces a model h : X → X, embedded in a toric variety j : X ⊂ Z, that dominates all the SQM's induced by the Minimal Model Program. Moreover the fan of Z is supported on Trop(XT ). Via this construction we are able to prove several consequences. First of all, the map h is given as an embedded map into toric varieties and this allows us to obtain a Minkowski basis for the Newton-Okounkov bodies on X. In particular we can 2010 Mathematics Subject Classification. Primary: 14C20. Secondary: 14E25, 14M25, 14E30. The second author was partially supported by the European Commission, Seventh Framework Programme, Grant Agreement n◦ 600376. 1 2 ELISA POSTINGHEL AND STEFANO URBINATI reconstruct the movable cone of X. The main result is the following, more details are given in Lemma 4.2 and Theorem 4.5. Theorem 1.2. In the notation of Theorem 1.1, certain admissible flags on Z induce admissible flags on X, such that if {DZ i }i∈I is a set of generators of the nef cone of Z, then {Di := h∗j∗DZ i }i∈I is a Minkowski basis for X with respect to induced flag. Moreover, another main result of this paper is that the Newton-Okounkov bodies of the Minkowski basis elements are rational and polyhedral. Moreover the value semigroup is finitely generated. This implies the following result: Theorem 1.3. If X is a smooth Mori dream space, the global Newton-Okounkov body of X with respect to the flags obtained in Theorem 1.2 is a rational polyhedron. Moreover if A is an ample line bundle on X, (X, A) admits a flat embedded degeneration to a not necessarily normal toric variety whose normalisation is the toric variety defined the Newton-Okounkov body of A. Details are given in Corollary 4.7 and Theorem 4.9. The first statement of Theorem 1.3 answers affirmatively a question posed by Lazarsfeld and Mustat¸a in [18]. The second statement is based on work of Anderson [1]. The article is organized as follows. In section 2 we recall the notations and basic properties of MDS's and Cox rings. We review the constructions via GIT of the embedding of a MDS into a toric variety and we deduce the connection with tropical geometry. In section 3 we construct the universal model of MDS's via tropical geometry. In section 4, after recalling the constructions of Newton- Okounkov bodies and Minkowski bases, we prove our main result, i.e. we show the existence of a Minkowski basis for a MDS, we show how to construct it and we prove Theorem 1.3. Finally, in Section 5, we give some examples of the construction, explicitly computed via the free software Gfan [9]. 2. General setting In this section we collect several results from [6, 8, 11, 24] providing the main connections between tropicalization and Mori dream spaces. 2.1. Newton-Okounkov bodies and toric degenerations. Let X be a smooth complex projective variety of dimension n over an algebraically closed field k and let D be a Cartier divisor on X. Okounkov's construction associates to D a convex body ∆Y• (D) ⊆ Rn, which we call the Newton-Okounkov body. It depends on the choice of an admissible flag Y•, i.e. a chain Y0 = X ⊃ Y0 ⊃ . . . ⊃ Yn−1 ⊃ Yn, where Yi is a subvariety of codimension i in X which is smooth at the point Yn. Using the flag, one defines a rank n valuation ν = νY• which, in turn, defines a graded semigroup ΓY• ⊆ N × Nn. The convex body ∆Y• (D) is the intersection of {1} × Rn with the closure of the convex hull of ΓY• in R × Rn. One can also define the notion of global Newton-Okounkov body of X which is the closed convex cone in Rn × N 1(X)R whose fibers over any big divisor D on X coincides with the Newton-Okounkov body ∆Y• (D) of such divisor. We refer to [14, 18] for details on this construction. NEWTON-OKOUNKOV BODIES & TORIC DEGENERATIONS OF MORI DREAM SPACES 3 Newton-Okounkov bodies are quite difficult to compute in general. They often are not polyhedral and, when polyhedral, they may be not rational. Even when the body is polyhedral, the semigroup ΓY• need not be finitely generated. For toric varieties Newton-Okounkov bodies turn out to be nice. In fact Lazars- feld and Mustat¸a [18] proved that if X is a smooth toric variety, D is a T -invariant ample divisor on X and the Yi's are T -invariant subvarieties, then ∆Y• (D) is the lattice polytope associated with D in the usual toric construction. Moreover the global Newton-Okounkov bodies are rational polyhedral. It is natural to investigate whether this construction behaves well for special classes of varieties. A consequence of the work contained in [3, 8] is that divisors on Mori dream spaces often have toric-like behaviour, so it makes sense to pose the following question. Question 2.1 ([18, Problem 7.1]). Let X be a smooth Mori dream space. Does there exist an admissible flag with respect to which the global Newton-Okounkov body of X is rational polyhedral? In [1] Anderson extended the connection between Okounkov bodies and toric varieties, by introducing a geometric criterion for ∆Y• (A) to be a lattice polytope for A ample and, in this situation, by constructing an embedded toric degeneration of (X, A). Theorem 2.2 ([1, Theorem 5.8]). Let A to be an ample divisor on X and assume the value semigroup associated to A with respect to the valuation induced by a complete flag is finitely generated. Then X admits a flat degeneration to a toric variety whose normalization is X∆Y• (A). Another natural question is the following. Question 2.3. For which varieties is it possible to find a flag such that the value semigroup of an ample divisor is finitely generated? Notice that the latter is a very strong condition, and it is much stronger than the finite generation of the divisorial ring. We will give an affirmative answer to both Questions 2.1 and 2.3 for Mori dream spaces in Section 4. 2.2. Mori dream spaces. Let X be a Q-factorial projective variety of dimension n over the complex numbers. Let r be the rank of PicQ(X) = N1(X) and let {L1, . . . , Lr} be a set of line bundles on X whose numerical classes form a basis of N1(X) and whose affine hull contains NE Definition 2.4. For any vector of integers v = (n1, . . . , nr), we set Lv := L⊗n1 · · · ⊗ L⊗nr . The Cox ring of X is given by 1 ⊗ 1 (X). r Cox(X) := R(X; L1, . . . , Lr) = Mv∈Nr H 0(X, Lv). Definition 2.5 ([8]). Let X be a Q-factorial projective variety such that PicQ(X) = N 1(X). Then we say that X is a Mori Dream Space (MDS) if and only if Cox(X) is a finitely generated C-algebra. Notation 2.6. We denote by N the dimension of Cox(X) over C and we set where r as above is the rank N 1(X). m := N − r, 4 ELISA POSTINGHEL AND STEFANO URBINATI Notation 2.7. A toric variety Z is quasi-smooth, i.e. it is a finite abelian quotient of a smooth toric variety, if the fan is given by rational simplicial cones. Notation 2.8. We will denote by Nef(X) and Mov(X) the cones of nef and mov- able divisors of X (see [17] for precise definitions). Here we just recall that nef divisors are the ones intersecting positively all the effective curves and movable divisors are the ones with stable base locus of codimension at least 2. It is well known that that Nef(X) ⊆ Mov(X); when X is a MDS we know much more. Proposition 2.9 ([8, Proposition 2.11]). Let X be a Mori dream space. Then there is an embedding X ⊆ Z into a quasi-smooth projective toric variety (with torus T ) such that: (1) the restriction Pic(Z)Q → Pic(X)Q is an isomorphism, 1 (2) the previous isomorphism induces an isomorphism NE (3) every Mori chamber of X is a union of finitely many Mori chambers of Z, (4) for every rational contraction ϕ : X → X0 there is toric rational contraction ϕT : Z → Z0, that is regular at the generic point of X, such that ϕ = ϕT X . (Z) → NE 1 (X), Combining the previous proposition and the definition of MDS's, Hu and Keel [8] proved that both the nef cone and the movable cone of a MDS are rational polyhedral. Moreover, the movable cone is reconstructed as a finite union of cones, the so called Mori chamber decomposition as follows. Let us denote by F := {fi : X 99K Xi} the set of finitely many SQM's of X, then we have Mov(X) = [fi∈F f ∗ i (Nef(Xi)). 2.3. GIT and Chow quotients for toric varieties. The main references for this section are [11, 24]. Let us consider a projective toric variety W ⊆ PN with torus T . Let us assume that the T -action on W extends to PN . Moreover let G be a subtorus of its defining torus T . In [11] Kapranov, Sturmfels and Zelevisky constructed a quotient of such variety called the Chow quotient, denoted by W//G. Proposition 2.10 ([11, Section 3]). The toric variety (T //G, W//G) dominates each GIT quotient of W with respect to G and it is the minimal normal variety with such property. Notation 2.11. The variety (T //G, W//G) will appear as Z in the rest of the paper. Let us now consider Y a connected closed subvariety of the algebraic torus T . Let Y be a compactification of Y in a toric variety Z of T . Definition 2.12 ([24, Definition 1.1]). We say that Y is a tropical compactification if the multiplication map is faithfully flat and Y is proper. Φ : T × Y → Z, (t, y) 7→ ty Notation 2.13. The following notation is commonly used: the pair (Y , Z), with Y and Z as in Definition 2.12, is a tropical compactification of Y ⊆ T . NEWTON-OKOUNKOV BODIES & TORIC DEGENERATIONS OF MORI DREAM SPACES 5 Theorem 2.14 ([24, Theorem 1.2]). Any subvariety Y of a torus has a tropical compactification Y such that the corresponding toric variety Z is smooth. In this case the boundary Y \Y is divisorial and has combinatorial normal crossings: for any collection B1, . . . , Br ∈ Y \Y of irreducible divisors, T Bi has codimension r. Proposition 2.15 ([24, Proposition 2.5]). Assume that (Y, Z) is a tropical com- pactification. Then the fan of Z is supported on the tropicalization of Y , Trop(Y ). Remark 2.16. Note that the variety Z in Notation 2.11 is projective but might be singular, whereas Z in Definition 2.12 is never projective, since the associated fan is never complete, but it is always smooth. The key observation of Tevelev in [24, Section 5] is that the tropical compact- ification is a refinement of the Chow compactification, i.e. every compactification induced by GIT is dominated by the tropical compactification. Tevelev also ob- serves that the previous statement might fail if the torus sits trivially or in special position in the variety, but we will see that with a good choice of an auxiliary compactification of the torus, i.e. the one given by [8], the statement is always true. We are now ready to introduce the key object linking tropical geometry and MDS's. Notation 2.17. Let X be a normal algebraic variety and W a normal algebraic toric variety with X ⊆ W . Let ΣW be the fan defining the toric variety W . We will consider the following subfan of ΣW ΣW,X := {σ ∈ ΣV(σ) ∩ X 6= ∅}. This subfan has already been considered in closely related works (cf. [6, 24]). In the next section we will construct a birational model of the toric variety W viewed as the toric variety associated to a subfan of the Chow quotient. The idea was first introduced by Lafforgue in [16] in the case of Grassmannians. 3. Universal model via tropicalization Let X be a Mori dream space and let X ⊆ Z be an embedding of X into a toric variety Z, as in Proposition 2.9. Via ambient toric modifications it is possible to give a sequence of blow-ups in order to obtain an inclusion X ⊆ Z such that for every small Q-factorial modification of X, fi : X 99K Xi, we have the commutative diagram of Figure 1, where the Zi corresponds to the toric GIT quotient for the given SQM. Note that the diagram exists because MDS's admit finitely many SQM's. A similar construction is studied in [7]. Z ⊆ X g gi h hi Z ⊇ X fi Xi ⊆ Zi Figure 1. Universal commutative diagram 6 ELISA POSTINGHEL AND STEFANO URBINATI Theorem 3.1. Let X be a smooth Mori dream space with rank(N 1(X)) = r and let X ⊆ Z be the embedding of Proposition 2.9, where Z is a toric variety with maximal torus T . Set XT := XT , the restriction of X to the maximal torus of Z. Then the following holds. (1) There exists a quasi-smooth toric variety Z and an embedded variety X ֒→ Z resolving all the small Q-factorial modifications of X induced by the Minimal Model Program. (2) Let ΣZ be the fan associated to Z and let and ΣZ,X be the subfan of ΣZ whose T -invariant orbits intersect X, as in Notation 2.17. Then ΣZ,X is supported on Trop(XT ). Proof. Since MDS's admit only a finite number of SQM's, it is always possible to construct a variety resolving all such morphisms. It is well known that the Mori chambers of X are refined by the Mori chambers of Z, cfr. Proposition 2.9 (3). This implies that all of the SQM's are induced by a sequence of SQM's at the level of the ambient toric variety. Therefore Z is obtained as a finite sequence of star subdivisions corresponding to the resolution of all the SQM's. This proves (1). By the previous argument we know that there exists a normal toric variety Z that dominates all possible GIT quotients associated to the Cox ring of X. Then Z must be the Chow quotient of Proposition 2.10. Let TZ,X denote the toric variety associated to ΣZ,X . We claim that (X, TZ,X ) is a tropical compactification of XT ⊆ T . Since the inclusion is obviously proper, as it is constructed from an embedding via a sequence of blow-ups, the pair is a tropical compactification if and only if the boundary divisors have combinatorial normal crossings (cfr. [24, Proof of Theorem 1.2]). Note that proposition 2.9 implies that the intersection has the correct dimension for divisors. That these restrictions, when non-empty, are with combinatorial nor- mal crossings it is not true a priori. However the desired property can be achieved via a sequence of star subdivisions of the defining fan of the ambient toric variety; these star subdivisions giving rise to Z. This concludes the proof of (2). (cid:3) Remark 3.2. Note that in the proof of Theorem 3.1 we implicitly showed that the tropicalization itself detects what are the star subdivisions to be performed in order to solve such issue: these correspond to the blow-up of the loci with the "incorrect" intersection dimension. In particular we confirmed that the Chow quotient is dominated by the tropicalization as stated in [24, Section 5]. Corollary 3.3. There exists a smooth toric resolution eZ of Z such that X embeds in eZ and such that (1) and (2) of Theorem 3.1 hold for the inclusion X ֒→ eZ. ΣZ of Z, π : Σ eZ → ΣZ such that the resulting morphism eX → X is an isomorphism Proof. Since TZ,X is smooth, it is possible to choose a smooth refinement of the fan and the induced morphism of fans Σ eZ,X → ΣZ,X is the identity. (cid:3) Using this construction we can recover the movable cone of X. Proposition 3.4. With the above notation, we have that h∗Nef(X) = Mov(X). Proof. First of all we have that for every SQM of X, fi : X 99K Xi, then h∗ i Nef(Xi) ⊆ Nef(X). In particular, since X is a MDS and X dominates all the SMQ's, this implies that Mov(X) ⊆ h∗Nef(X). NEWTON-OKOUNKOV BODIES & TORIC DEGENERATIONS OF MORI DREAM SPACES 7 For the converse inclusion, recall that the push-forward of the movable cone via any birational morphism is contained in the movable cone of the image, see for instance [5, Corollary 3.12]. Therefore we have the following inclusions of cones: h∗Nef(X) ⊆ h∗Mov(X) ⊆ Mov(X). (cid:3) 4. Movable cones of Mori Dream Spaces The idea behind the notion of Newton-Okounkov bodies is that of associating polyhedral objects to line bundles on a given variety. It is then natural to try to understand the correlation between properties of polytopes and their algebro- geometric counterpart. Let us recall the definition of Minkowski basis. Definition 4.1 ([22]). Let X be a smooth projective variety of dimension n and Y• : X = Y0 ⊇ Y1 ⊇ · · · ⊇ Yn−1 ⊇ Yn = {pt} an admissible flag on X. A collection {D1, . . . , Dr} of pseudo-effective divisors on X is called Minkowski basis of X with respect to Y• if (1) for any pseudo-effective divisor D on X there exist non-negative numbers {a1, . . . , ar} and a translation φ : Rn → Rn such that φ(∆Y• (D)) =X ai∆Y• (Di), (2) the Newton-Okounkov bodies ∆Y• (Di) are indecomposable, if ∆Y• (Di) = P1 + P2 for convex bodies P1, P2, then Pj = kj · ∆Y• (Di) for non-negative numbers k1, k2 with k1 + k2 = 1. i.e., The following is the key result linking tropical geometry and Newton-Okounkov bodies. • : Z = Y Z Lemma 4.2. Let Y Z torus-invariant subvarieties of Z of dimension dim(Y Z Then Y Z • such element to X and the choice of an irreducible element of the intersection: m be an admissible flag of i ) = m − i, for i = 0, . . . , m. induces a flag on X via truncation to the first n elements, restriction of 1 ⊃ · · · ⊃ Y Z 0 ⊃ Y Z m−1 ⊃ Y Z Y• : X = Y0 ⊃ Y1 ⊃ · · · ⊃ Yn, where Yi ⊆ Y Z m−n+iX of dimension dim(Yi) = n − i, for i = 0, . . . , n, is an admissible flag on X. Proof. By Theorem 3.1, X ⊆ TZ,X is a tropical compactification, hence the in- tersection with T -invariant divisors is combinatorially normal crossings, i.e. the intersection with the torus-invariant flag elements has always the correct codimen- sion. If the flag obtained in this way is not given by irreducible elements, we can derive an admissible one by choosing irreducible components of the restrictions Y Z m−n+iX . This proves the statement. (cid:3) Remark 4.3. Notice that if Z is not smooth, an admissible flag might not exist. In we have the identity Σ eZ,X → ΣZ,X , it is always possible, modulo reordering the this case, let us consider the resolution eZ → Z constructed in Corollary 3.3. Since terms, to obtain the desired flag starting from an admissible flag on eX. Remark 4.4. The flags constructed in Lemma 4.2 are supported on X. Those are infinitesimal flags for X, and all the properties of Newton-Okounkov bodies are preserved. 8 ELISA POSTINGHEL AND STEFANO URBINATI Theorem 4.5. In the notation of Section 3, set j : X ֒→ Z. Let {DZ Minkowski basis for the nef cone of Z. Then i }i∈I be a D := {Di := h∗j∗DZ i }i∈I is a Minkowski basis for X with respect to the flag constructed in Lemma 4.2. Proof. Note that the inclusion j : X ֒→ Z allows us to reconstruct the Nef cone of X, in fact [6, Proposition 5.1] proves that j∗Mov(Z) = Nef(X). Then the movable cone on X can be reconstructed via push-forward and Proposition 3.4. We claim that any valuation vector induced by the flag constructed in Lemma 4.2 is obtained by the truncation of a valuation vector of the ambient toric variety. In order to prove the claim, let us denote by DZ a divisor on Z and by D its restriction to X. Let π be the projection map π : Rm → Rn, obtained via the truncation. We first need to prove that: π(∆Y Z • (DZ )) = ∆Y• (D). Since the valuation vectors inducing ∆Y Z • (DZ ) are given by T -invariant sections, it is clear that π(∆Y Z • (DZ )) ⊆ ∆Y• (D). In the general case it is quite difficult to prove the reverse inclusion, sometimes it might even be false. In this case, since X is a MDS, the movable cone is covered by pull-back's of nef cones of small modifications. For every movable divisor D on X, there exists an SQM fi : X 99K Xi such that Di := f∗D is nef. We can then prove the reverse inclusion via the diagram (1) using the following property and the fact that every nef divisor is a limit of ample divisors. For any ample divisor A on Z we have that volZX (A) = volX (AX ) = An · X, i.e. the restricted volume coincides with the volume of the induced linear series via restriction, see [4] for precise definitions and properties. Now, every movable divisor is a limit of ample divisors in some given small birational model. Hence we can conclude that the two Newton-Okounkov bodies coincide. Finally, the additivity of bodies of X is simply induced by the additivity of the corresponding bodies of Z (that holds for the results contained in [22]), since this property is preserved under projection. (cid:3) Remark 4.6. Lemma 4.2 and Theorem 4.5 give a direct connection between the Minkowski basis for the MDS X and the toric variety where it embeds to. The explicit description also allows us to study restrictions of sections from the toric variety to the MDS. 4.1. Application 1. As an immediate consequence of Theorem 4.5 we can then state the following result, that answers positively the Question 2.1 Corollary 4.7. Let X be a Mori dream space. The global Newton-Okounkov body if X with respect to an admissible flag obtained as in Theorem 4.5 is a rational polyhedral polytope in Rn+r, where n = dim(X) and r = rank(N 1(X)). Remark 4.8. Work done recently led to partial answers to the question posed by Lazarsfeld and Mustat¸a, Question 2.1. In [21] the author gave an answer to the question for surfaces and, in higher dimension, for Mori dream spaces with certain NEWTON-OKOUNKOV BODIES & TORIC DEGENERATIONS OF MORI DREAM SPACES 9 particular flags provided that such flags exist. In [23] the authors gave a criterion for the polyhedrality of the global Newton-Okounkov body depending on the exis- tence of a Minkowski basis and they proved polyhedrality for certain homogeneous threefolds. 4.2. Application 2. Via our construction we are able to extend Theorem 2.2 to the case of arbitrary ample divisors on smooth Mori dream spaces. Theorem 4.9. Let X be a smooth Mori dream space and let D be the Minkowski basis for X constructed in Theorem 4.5. For any ample divisor A on X, let ∆Y• (A) be the Newton-Okounkov body constructed as a Minkowski sum of elements of D. Then X admits a flat degeneration to the toric variety X∆Y• (A). Proof. The main condition for the existence of the flat degeneration is the finite generation of the value semigroup. This condition yields that the extremal points of the Newton-Okounkov body shall correspond to the valuation of some section (i.e. the point is not a limit but an actual valuation vector). By the construction, the restriction of torus-invariant sections to X will give the exact valuation vector. Even more, the inclusion map will ensure that the section corresponds to a section of chosen ample line bundle A. (cid:3) 5. Examples 5.1. Example 1. Let X = Bl4(P2) the del Pezzo surface of degree 5, namely the blow-up of P2 in 4 points in linearly general position. This is isomorphic to M 0,5,. Let H denote the pull-back of the class of a line of P2 and let E1, . . . , E4 denote the classes of the exceptional divisors. We have Pic(X) = hH, E1, . . . , E4i. After a change of basis, we can take {H − E3 − E4, E1, . . . , E4} to be a set of generators of the Picard group. Moreover Cox(X) is generated by the classes {Eij := H − Ei − Ej : 1 ≤ i < j ≤ 4} ∪ {Ei : 1 ≤ i ≤ 4}. We now rewrite the generators of Cox in the new basis of Pic(X): H − Ei − Ej = (H − E3 − E4) − Ei − Ej + E3 + E4, ∀1 ≤ i < j ≤ 4 Ei = Ei, ∀1 ≤ i ≤ 4. Using this, we can describe the map with the following 5 × 10 matrix deg : Cox(X) → Pic(X) E34 E1 E2 E3 E4 1   E12 E13 E14 E23 E24 E34 E1 E2 E3 E4 1 0 1 −1 −1 −1 −1 1 1 1 0 0 −1 −1 1 1 0 0 1 0 0 1 0 1 1 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1   . Let A be the 5 × 5 matrix such that the above matrix is the concatenated matrix The matrix representing (cid:0) A I5 (cid:1) . ker(deg) → Cox(X) 10 ELISA POSTINGHEL AND STEFANO URBINATI is defined by the Gale transform −A (cid:19) = (cid:18) I5   0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 1 −1 −1 −1 −1 −1 0 1 0 1 1 1 −1 0 −1 −1 −1 0 1 1 0 0 0 −1 0 −1   . The rows of the above matrix define the fan Z5 of a blow-up of P5. Precisely, the first six rays are the rays of the fan of P5 in the standard basis of Z5. Let x1, . . . , x5, x0 be corresponding homogeneous coordinates. The four last rays define exceptional divisors of the blow-up of T -invariant 2-planes of P5, with T = (C∗)5, defined by Π1 ={x1 = x2 = x3 = 0}, Π2 ={x1 = x4 = x5 = 0}, Π3 ={x0 = x2 = x4 = 0}, Π4 ={x0 = x3 = x5 = 0}. Notice that {Πi ∩ Πj : i 6= j} is the set of the six coordinate points of P5. The map i : X → Z is the embedding described in Proposition 2.9. Remark 5.1. The embedding might not be unique since it depends on the lin- earization by an ample line bundle. However the tropical compactification will be the same. Levitt in [19] computed a set of equations for such an embedding that do not respect intersection conditions and cannot be the image of the embedding constructed by [8]. Let P be the 2−plane of P5 described by the following equations: x2 + x4 + x0 = 0, x1 + x4 + x5 = 0, x2 + x3 + x4 + x5 = 0, with the following intersections p1 :=P ∩ Π1 = [0, 0, 0, 1, −1, −1], p2 :=P ∩ Π2 = [0, 1, −1, 0, 0, −1], p3 :=P ∩ Π3 = [−1, 0, −1, 0, 1, 0], p4 :=P ∩ Π4 = [−1, −1, 0, 1, 0, 0]. The image of X inside Z is the blow-up of P at p1, p2, p3, p4. Using Gfan, we computed the tripicalization of P ⊂ P5 obtaining the following description. Q[ x1 , x2 , x3 , x4 , x5 ] { x2+x4 +1 , x1+x4+x5 , x2+x3+x4+x5 NEWTON-OKOUNKOV BODIES & TORIC DEGENERATIONS OF MORI DREAM SPACES 11 } LP a l g o r i t h m b e i n g used : "cddgmp " . a p p l i c a t i o n f a n v e r s i o n 2 . 2 t y p e SymmetricFan AMBIENT DIM 5 DIM 2 LINEALITY DIM 0 RAYS −1 −1 −1 0 0 # 0 −1 0 0 −1 −1 # 1 −1 0 0 0 0 # 2 0 −1 0 0 0 # 3 0 0 −1 0 0 # 4 0 0 0 −1 0 # 5 0 0 0 0 −1 # 6 1 # 7 0 1 1 1 0 # 8 1 # 9 1 1 1 0 1 0 1 1 N RAYS 10 F VECTOR 1 10 15 SIMPLICIAL 1 PURE 1 CONES Dimension 0 {} Dimension 1 12 ELISA POSTINGHEL AND STEFANO URBINATI { 0 } { 1 } { 2 } { 3 } { 4} {5 }{ 6 }{ 7} {8 }{ 9} Dimension 2 {0 2}{1 2}{0 3}{0 4}{1 5}{1 6}{3 6}{4 5}{2 9}{3 7}{5 7}{4 8} {6 8}{7 9}{8 9} Let Ei denote the exceptional divisor of Πi in Z and let H denote the hyperplane class in Z. Let us use the notation Eij := H − Ei − Ej for the sake of simplicity. The 1− and 2−dimensional cones of the fan correspond to the orbits in Z that intersect X. The rays correpond to the divisors Ei and Eij The 2-cone {0, 2} corresponds to E1.E12, the class of a hyperplane on the exceptional divisor E1 of Π1 in Z; the 2−cone {3, 6} corresponds to the intersection E13.E24, that is the strict transform on Z of the 3−plane spanned by the points Πi ∩ Πj, with (i, j) ∈ {1, 3} × {2, 4} Whilst, for example, let us consider the cone {0, 1} that corresponds to the intersection E1.E2, that is the strict transform in Z of the point of P5 of coordinates [0, 0, 0, 0, 0, 1]: this orbit does not intersect X and it does not appear in the tropicalization. Note that the rays exactly coincide with the ones appearing in the fan of Z. This is explained by the fact that a surface does not admit flips and hence we have a unique chamber of the Movable cone given by the Nef cone for X. The combinatorial normal crossing property of the boundary allows us to induce a flag from the ambient toric variety Z to the MDS X. for example, Let us consider, the maximal cone in ΣZ generated by {E13, E24, E14, E23, E3}. This gives a T -invariant flag by complete intersection. The tropicalization selects all the subcones that induce a flag on X: in this case one could choose the following {E13, E24}, {E14, E23}, {E13, E3} or {E23, E3}. It is then a simple computation to show that the Minkowski basis is obtained by projection to the corresponding selected components. 5.1.1. A toric degeneration of the del Pezzo surface of degree 5. We consider X embedded in P5 with the linear system of cubics through four points, given by the (ample) anticanonical divisor A = 3H − E1 − E2 − E3 − E4. Let us consider on Z the admissible flag Y Z • : Z ⊃ E13 ⊃ (E13 ∩ E24) ⊃ · · · that induces the following admissible flag on X Y• : X ⊃ E13 ⊃ {(E13 ∩ E24)}, where Eij denotes the strict transform on X of the line in P2 passing through two of the four blown-up points, using Lemma 4.2. Following Theorem 4.5, we find a Minkowski basis for X which is given by the following divisor: {H, H − Ei, H − Ei − Ej , 2H − E1 − E2 − E3 − E4 : i, j, i 6= j}. One can compute the corresponding Newton-Okounkov bodies. For instance the Newton-Okounkov body of H is a triangle with vertices (0, 0), (0, 1), (1, 0) in R2, while the Newton-Okounkov body of 2H − E1 − E2 − E3 − E4 is a segment with vertices (0, 0), (1, 1). The ample divisor A admits the Minkowski decomposition A = H + (2H − E1 − E2 − E3 − E4) and in particular its Newton-Okounkov body is obtained as the Minkowski sum of the two polytopes of H and 2H−E1−E2−E3−E4, see Figure 2. This gives rise to a toric degeneration of the del Pezzo surface (X, A) NEWTON-OKOUNKOV BODIES & TORIC DEGENERATIONS OF MORI DREAM SPACES 13 to the singular toric surface X∆Y• (A) obtained as the blow-up of P2 in two pairs of infinitely near points. • ❅ ❅• • • + • • ❅ • ❅ • • • = • Figure 2. Newton-Okounkov body of 3H −P4 i=1 Ei 5.2. Example 2. Let now X := Bl5(P3) be the blow-up of P3 in 5 points in linearly general position. Let us denote by H the hyperplane class and by E1, . . . , E5 the exceptional divisors. The following divisors form a basis for the Picard group of X: Generators of the Cox ring of X are: H − E3 − E4 − E5, E1, . . . , E5. {H − Ei1 − Ei2 − Ei3 : 1 ≤ i1 < i2 < i3 ≤ 5} ∪ {Ei : 1 ≤ i ≤ 5}. Let Y• be the flag on X obtained starting from the flag {E13, E13 ∩ E24, . . . } on Z. As in the previous example, the maps deg : Cox(X) → Pic(X) and ker(deg) → Cox(X) are defined, respectively, by the following fundamental matrices:   (cid:0) A I6 (cid:1) := and 1 1 1 1 1 1 −1 −1 −1 −1 −1 −1 −1 −1 −1 1 1 0 1 0 0 −1 −1 −1 1 1 0 0 0 0 0 0 0 1 0 0 1 0 1 0 1 0 1 1 0 1 0 0 0 1 1 0 1 0 1 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 1   , −A (cid:19) = (cid:18) I9   0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 −1 −1 −1 −1 −1 −1 −1 −1 −1 0 1 1 0 1 1 1 1 0 −1 −1 0 −1 −1 0 −1 0 −1 −1 0 1 1 0 0 0 −1 0 0 −1 1 0 0 0 −1 0 0 1 0 0 −1 0 0 −1 1 1   . The rows of the above matrix define a simplicial fan. The associated quasi-smooth projective variety is the variety Z that appears in Proposition 2.9, that we can describe as follows. The first 9 rows give rise to the fan of P9, while the five last 14 ELISA POSTINGHEL AND STEFANO URBINATI rows represent the exceptional divisors of the blow-up of P9 along the torus invariant 3−planes defined by Π1 ={x1 = x2 = x3 = x4 = x5 = x6 = 0}, Π2 ={x1 = x2 = x3 = x7 = x8 = x9 = 0}, Π3 ={x0 = x1 = x4 = x5 = x7 = x8 = 0}, Π4 ={x0 = x2 = x4 = x6 = x7 = x9 = 0}, Π5 ={x0 = x3 = x5 = x6 = x8 = x9 = 0}. Therefore Z is the blow-up of P9 along Π1, . . . , Π5. We now describe the embedding X ⊂ Z of Proposition 2.9. Consider the 3−plane P of P9, that intersects each Πi in a point, defined by x1 − x2 + x3 = 0, x5 − x8 + x0 = 0, x1 − x7 + x8 = 0, x2 − x7 + x9 = 0, x6 − x9 + x0 = 0 x1 − x4 + x5 = 0. The intersection points are given by p1 :=P ∩ Π1 = [0, 0, 0, 0, 0, 0, 1, 1, 1, 1], p2 :=P ∩ Π2 = [0, 0, 0, 1, 1, 1, 0, 0, 0, −1], p3 :=P ∩ Π3 = [0, 1, 1, 0, 0, −1, 0, 0, −1, 0], p4 :=P ∩ Π4 = [1, 0, −1, 0, −1, 0, 0, −1, 0, 0], p5 :=P ∩ Π5 = [1, 1, 0, 1, 0, 0, 1, 0, 0, 0]. It is easy to check that P ∩ Πi ⊂ P9 is a point, say pi. The image of X in Z is the blow-up of P in p1, . . . , p5. Remark 5.2. Obviously the choice of a subspace P satisfying the same properties is not unique. For instance the following set of equations describes another such a 3−plane. x1 + x7 + x8 = 0, x6 + x9 + x0 = 0, x2 + x7 + x9 = 0, x4 + x5 + x7 + x8 = 0, x5 + x8 + x0 = 0, x3 + x6 + x8 + x0 = 0. Using Gfan, we computed the tripicalization of P ⊂ P9 obtaining for both sets of equations the following description. Q[ x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 , x9 ] { x1+x7+x8 , x6+x9 +1 , x5+x8 +1 , x2+x7+x9 , x4+x5+x7+x8 , x3+x6+x8+1 } LP a l g o r i t h m b e i n g used : "cddgmp " . a p p l i c a t i o n f a n v e r s i o n 2 . 2 t y p e SymmetricFan NEWTON-OKOUNKOV BODIES & TORIC DEGENERATIONS OF MORI DREAM SPACES 15 AMBIENT DIM 9 DIM 3 LINEALITY DIM 0 # 0 ( ∗ ) # 1 ( ∗ ) # 2 # 3 # 4 # 5 ( ∗ ) # 6 # 7 # 8 ( ∗ ) # 9 # 10 # 11 ( ∗ ) # 12 ( ∗ ) # 13 ( ∗ ) # 14 ( ∗ ) # 15 ( ∗ ) # 16 ( ∗ ) # 17 ( ∗ ) # 18 ( ∗ ) # 19 ( ∗ ) # 20 ( ∗ ) # 21 # 22 # 23 # 24 ( ∗ ) RAYS −1 −1 −1 −1 −1 −1 0 0 0 −1 −1 −1 0 0 0 −1 −1 −1 −1 −1 −1 0 0 0 0 0 0 −1 0 0 −1 −1 0 0 0 0 −1 0 0 0 0 0 −1 −1 0 −1 0 0 0 0 0 0 0 0 0 −1 0 −1 0 −1 0 0 0 0 −1 0 0 0 0 −1 0 −1 0 −1 0 0 0 0 0 0 0 0 0 −1 0 −1 −1 0 0 0 0 0 −1 0 0 0 0 −1 −1 0 0 −1 0 0 0 0 0 0 0 0 0 −1 0 0 0 0 0 0 0 0 0 −1 0 0 0 0 0 0 0 0 0 −1 0 0 0 0 0 0 0 0 0 −1 0 0 0 0 0 0 0 0 0 −1 0 0 0 0 0 0 0 0 0 −1 0 1 1 0 0 1 0 0 1 1 0 1 0 1 0 0 1 0 1 1 0 1 0 0 1 0 0 1 1 1 0 1 1 0 1 1 1 1 1 1 0 1 1 0 1 1 1 1 1 1 0 1 1 0 1 1 1 1 1 1 1 1 1 N RAYS 25 F VECTOR 1 25 105 105 SIMPLICIAL 1 ELISA POSTINGHEL AND STEFANO URBINATI 16 PURE 1 The rays marked with (∗) are the rows of the matrix defining Z (modulo a sign change). There are 10 more rays, that we are now going to interpret. Notice first that the span of each pair of 3−planes hΠi, Πji is a 6−plane in P9. Moreover the 3−plane P intersects hΠi, Πji in the line hpi, pji. There are exactly 10 of these 6−planes in P9 and, consequently, 10 of these lines in P . Each of the remaining 10 rays above is the exceptional divisor of the blow-up of a (strict transform in Z of the) 6-plane hΠi, Πji. For instance the ray #2 (which is the sum of the rays #5, #8, #11 ) gives the exceptional divisors of the blow-up of the 6−plane of P9 of equations x1 = x2 = x3 = 0, namely the one spanned by Π1 and Π2. Now, in the notation of Section 3, the toric variety Z is the blow-up of Z along all these 10 6−planes hΠi, Πji. Moreover X ⊂ Z is the blow-up of X ⊂ Z along the 10 corresponding lines hpi, pji. THerefore X ∼= M 0,7, cfr. Section ??. The blow-up X resolves all the flips of (−1)−curves (given by the strict transforms of the lines hpi, pji in X). The subfan ΣZ,X is supported on the tropicalization of P and it is described in the rest of the Gfan scrip below. Notice that all rays and 2−dimensional cones of ΣZ appear in ΣZ,X because X intersects all T -invariant subvarieties of coidimension 1 and 2, just by a simple dimension count. CONES Dimension 0 {} Dimension 1 { 0 } { 1 } { 2 } { 3 } { 4 } { 5} { 6} { 7} { 8} {9 } {1 0 }{ 1 1} {1 2 }{ 1 3} { 1 4 } { 1 5 } { 1 6 } { 1 7 }{ 1 8} {1 9 }{ 20 } {2 1 }{ 22 } {2 3} {24} Dimension 2 {0 2}{1 2}{0 3}{0 5}{1 4}{1 5}{2 5}{3 5}{4 5} {0 6}{0 8}{1 7}{1 8}{2 8}{0 9}{0 11}{1 10}{1 11}{2 11} {0 12}{0 13}{0 14}{1 15}{1 16}{1 17}{3 12}{3 13}{3 17} {4 14}{5 14}{4 15}{4 16}{5 17}{3 18}{4 18}{5 18}{6 8}{7 8} {6 12}{6 14}{6 16}{7 13}{8 13}{7 15}{7 17}{8 16}{9 11} {10 11}{10 12}{11 12}{9 13}{9 14}{9 15}{11 15}{10 16} {10 17}{12 16}{12 17}{13 15}{13 17}{14 15}{14 16}{2 24} {12 18}{13 18}{15 18}{16 18}{5 23}{5 24}{6 19}{7 19}{8 19} {12 19}{14 19}{15 19}{17 19}{8 22}{8 24}{9 20}{10 20}{11 20} {11 21}{13 20}{14 20}{16 20}{17 20}{11 24}{12 21}{15 21} {13 22}{16 22}{14 23}{17 23}{18 21}{18 22}{18 24}{19 21} {19 23}{19 24}{20 22}{20 23}{20 24}{21 24}{22 24}{23 24} NEWTON-OKOUNKOV BODIES & TORIC DEGENERATIONS OF MORI DREAM SPACES 17 MAXIMAL CONES Dimension 3 {0 2 5} {1 2 5}{0 3 5}{1 4 5} {0 2 8}{1 2 8}{0 2 11}{1 2 11}{0 3 12} {0 3 13}{0 5 14}{1 4 15}{1 4 16}{1 5 17}{3 5 17} {4 5 14}{3 5 18}{4 5 18}{0 6 8}{1 7 8}{0 6 12} {0 6 14}{0 8 13}{1 7 15}{1 7 17}{1 8 16}{0 9 11} {1 10 11}{0 11 12}{0 9 13}{0 9 14}{1 11 15} {1 10 16}{1 10 17}{3 12 17}{3 13 17}{4 14 15} {4 14 16}{2 5 24}{3 12 18}{3 13 18}{4 15 18}{4 16 18} {6 8 16}{7 8 13}{6 12 16}{6 14 16}{7 13 15}{7 13 17} {2 8 24}{10 11 12}{9 11 15}{10 12 16}{10 12 17} {9 13 15}{9 14 15}{2 11 24}{12 16 18}{13 15 18} {5 14 23}{5 17 23}{5 18 24}{6 8 19}{7 8 19}{6 12 19} {6 14 19}{7 15 19}{7 17 19}{12 17 19}{14 15 19} {8 13 22}{8 16 22}{9 11 20}{10 11 20}{9 13 20} {9 14 20}{10 16 20}{10 17 20}{11 12 21}{11 15 21} {13 17 20}{14 16 20}{12 18 21}{15 18 21}{13 18 22} {16 18 22}{5 23 24}{8 19 24}{12 19 21}{15 19 21} {14 19 23} {17 19 23} {8 22 24} {11 20 24} {11 21 24} {13 20 22} {16 20 22} {14 20 23} {17 20 23} {18 21 24} {18 22 24} {19 21 24} {19 23 24} {20 22 24} {20 23 24} References [1] D. Anderson, Okounkov bodies and toric degenerations. Math. Ann. (2013), vol. 356, 1183 -- 1202. [2] M. Artebani and A. Laface, Hypersurfaces in Mori dream spaces. Journal of Algebra 371 (2012): 26-37. [3] C. Birkar, P. Cascini, C. D. Hacon, and J. McKernan, Existence of minimal models for varieties of log general type, Journal of the American Mathematical Society, vol. 23 (2010), no. 2, 405 -- 468. [4] L. Ein, R. Lazarsfeld, M. Mustat¸a, M. Nakamaye and M. Popa, Restricted volumes and base loci of linear series, Amer. J. Math., vol.131 (2009), no. 3, 607 -- 651. [5] M. Fulger and B. Lehmann, Zariski decompositions of numerical cycle classes, ArXiv e-prints- 1310.0538, (2013). [6] A. Gibney and D. Maclagan, Lower and upper bounds for nef cones, International Mathe- matics Research Notices. IMRN (2012), n. 14, 3224 -- 3255. [7] J. Hausen, Cox rings and combinatorics II. Mosc. Math. J 8.4 (2008): 711-757. [8] Y. Hu and S. Keel, Mori dream spaces and GIT, Michigan Mathematical Journal, vol. 48 (2000), 331 -- 348. [9] A. N. Jensen, Gfan, a software system for Grobner fans and tropical varieties, Available at http://home.imf.au.dk/jensen/software/gfan/gfan.html [10] M. Kapranov, V eronese curves and Grothendieck-Knudsen moduli space M0,n, J. Algebraic Geom. 2 (1993), no. 2, pp. 239-262. [11] M. M. Kapranov, B. Sturmfels A. V. and Zelevinsky, Quotients of toric varieties, Mathema- tische Annalen, vol. 290 (1991), n. 4, 643 -- 655. [12] A.M. Castravet, The Cox ring of M 0,6, Trans. Amer. Math. Soc. 361 (2009), no. 7, 3851-3878 [13] E. Katz, S. Urbinati, Newton-Okounkov Bodies over Discrete Valuation Rings and Linear Systems on Graphs, arXiv:1609.06036, 2016. [14] K. Kaveh, A. Khovanskii, Newton-Okounkov bodies, semigroups of integral points, graded algebras and intersection theory, Annals of Mathematics, 176 (2): 925?978, 2012 18 ELISA POSTINGHEL AND STEFANO URBINATI [15] K. Kaveh, C. Manon, Khovanskii bases, Newton-Okounkov bodies and tropical geometry of projective varieties, arXiv:1610.00298, (2016). [16] L. Lafforgue, Chirurgie des grassmanniennes, CRM Monograph Series, vol. 19 (2003). [17] R. Lazarsfeld, Positivity in Algebraic Geometry, I, Springer (2004). [18] Lazarsfeld, R., Mustat¸a, M., Convex bodies associated to linear series, Ann. Sci. ´Ec. Norm. Sup´er. (4) 42 (2009), no. 5, 783-835. [19] J. Levitt, On embeddings of Mori Dream Spaces, Geom Dedicata 170, 281 -- 288 (2014) [20] Luszcz-´Swidecka, P., Schmitz, D., Minkowski decomposition of Okounkov bodies on surfaces, Journal of Algebra, Vol. 414, (2014), Pages 159174. [21] S. Okawa, On global Okounkov bodies of Mori dream spaces. Preprint, proceedings of the Miyako-no-Seihoku Algebraic Geometry Symposium (2010) [22] P. Pokora, D. Schmitz and S. Urbinati, Minkowski decomposition and generators of the moving cone for toric varieties, The Quarterly Journal of Mathematics, vol. 66 (2015), no. 3, 925 -- 939. [23] D. Schmitz and H. Seppanen, On the polyhedrality of global Okounkov bodies. Advances in Geometry 16.1 (2016): 83-91. [24] J. Tevelev, Compactifications of subvarieties of tori, American Journal of Mathematics, vol. 129 (2007), n. 4, 1087 -- 1104. Department of Mathematical Sciences, Loughborough University, Leicestershire LE11 3TU, United Kingdom E-mail address: [email protected] Dipartimento di Matematica, Politecnico di Milano, via Bonardi 9, Milano, Italy E-mail address: [email protected]
1307.6527
3
1307
2016-04-20T13:16:21
Alpha invariants and K-stability for general polarisations of Fano varieties
[ "math.AG", "math.DG" ]
We provide a sufficient condition for polarisations of Fano varieties to be K-stable in terms of Tian's alpha invariant, which uses the log canonical threshold to measure singularities of divisors in the linear system associated to the polarisation. This generalises a result of Odaka-Sano in the anti-canonically polarised case, which is the algebraic counterpart of Tian's analytic criterion implying the existence of a K\"{a}hler-Einstein metric. As an application, we give new K-stable polarisations of a general degree one del Pezzo surface. We also prove a corresponding result for log K-stability.
math.AG
math
ALPHA INVARIANTS AND K-STABILITY FOR GENERAL POLARISATIONS OF FANO VARIETIES RUADHA´I DERVAN Abstract. We provide a sufficient condition for polarisations of Fano vari- eties to be K-stable in terms of Tian's alpha invariant, which uses the log canonical threshold to measure singularities of divisors in the linear system associated to the polarisation. This generalises a result of Odaka-Sano in the anti-canonically polarised case, which is the algebraic counterpart of Tian's analytic criterion implying the existence of a Kahler-Einstein metric. As an application, we give new K-stable polarisations of a general degree one del Pezzo surface. We also prove a corresponding result for log K-stability. Contents 1. Introduction 2. Prerequisites 2.1. K-stability 2.2. Odaka's Blowing-up Formalism 2.3. Log Canonical Thresholds 3. Alpha Invariants and K-stability 4. Examples 5. Log Alpha Invariants and Log K-stability References 1 3 3 4 6 8 12 16 20 1. Introduction A central problem in complex geometry is to find necessary and sufficient con- ditions for the existence of a constant scalar curvature Kahler (cscK) metric in a given Kahler class. One of the first sufficient conditions is due to Tian, who in- troduced the alpha invariant. The alpha invariant α(X, L) of a polarised variety (X, L) is defined as the infimum of the log canonical thresholds of Q-divisors in the linear system associated to L, measuring singularities of these divisors. Tian [31] proved that if X is a Fano variety of dimension n with canonical divisor KX , the lower bound α(X,−KX ) > n n+1 implies that X admits a Kahler-Einstein metric in c1(X) = c1(−KX). The Yau-Tian-Donaldson conjecture states that the existence of a cscK metric in c1(L) for a polarised manifold (X, L) is equivalent to the algebro-geometric notion of K-stability, related to geometric invariant theory. This conjecture has recently been proven in the case that L = −KX [8, 6, 7, 32]. By work of Donaldson [9] and Stoppa [28], it is known that the existence of a cscK metric in c1(L) implies 1 2 RUADHA´I DERVAN that (X, L) is K-stable, provided the automorphism group of X is discrete. Odaka- Sano [22] have given a direct algebraic proof that α(X,−KX ) > n n+1 implies that (X,−KX) is K-stable. This provides the first algebraic proof of K-stability of varieties of dimension greater than one. On the other hand, few sufficient criteria are known for K-stability in the general case. We give a sufficient condition for general polarisations of Fano varieties to be K-stable. A fundamental quantity will be the slope of a polarised variety (X, L), defined as µ(X, L) = −KX.Ln−1 Ln = RX c1(X).c1(L)n−1 RX c1(L)n . (1) The slope is therefore a topological quantity which, after rescaling L, can be as- sumed equal to 1. Our main result is then as follows. Theorem 1.1. Let (X, L) be a polarised Q-Gorenstein log canonical variety with canonical divisor KX . Suppose that n+1 µ(X, L) and (i) α(X, L) > n (ii) −KX ≥ n n+1 µ(X, L)L. Then (X, L) is K-stable. Here, for divisors H, H ′, we write H ≥ H ′ to mean H − H ′ is nef. Note that when L = −KX, the slope of (X, L) is equal to 1, the second condition is vacuous and this theorem is then due to Odaka-Sano. The condition that X is log canonical ensures that α(X, L) ≥ 0, while the condition that X is Q-Gorenstein ensures that −KX exists as a Q-Cartier divisor. The second condition implies that X is either Fano or numerically Calabi-Yau, see Remark 3.5. By proving a continuity result for the alpha invariant, we also show in Corollary 4.3 that provided the inequality in the second condition is strict, the conditions to apply Theorem 1.1 are open when varying the polarisation. Theorem 1.1 gives the first non-toric criterion for K-stability of general polarisa- tions of Fano varieties. On the analytic side, a result of LeBrun-Simanca [14] states that the condition that a polarised variety (X, L) admits a cscK metric is an open condition when varying L, provided the automorphism group of X is discrete. As the existence of a cscK metric in c1(L) implies K-stability, this gives an analytic proof that in the situation of 1.1, K-stability is an open condition again in the case that the automorphism group of X is discrete. On the other hand, our result can also be used to give explicit K-stable polarisations, see for example Theorem 1.2. Many computations [4, 3] of alpha invariants have been done for anti-canonically polarised Fano varieties. Cheltsov [3], building on work of Park [25], has calculated alpha invariants of del Pezzo surfaces. As a corollary, Cheltsov's results imply that general anti-canonically polarised del Pezzo surfaces of degrees one, two and three are K-stable. Following the method of proof of Cheltsov, we give new examples of K-stable polarisations of a general del Pezzo surface X of degree one. Noting that X is isomorphic to a blow-up of P2 at 8 points in general position, we denote by H i=1 Ei− λE8 the hyperplane divisor, Ei the 8 exceptional divisors and Lλ = 3H −P7 arising from this isomorphism. Theorem 1.2. (X, Lλ) is K-stable for 19 25 ≈ 1 9 (10 − √10) < λ < √10 − 2 ≈ 29 25 . (2) ALPHA INVARIANTS AND K-STABILITY 3 We note that Theorem 1.1 is merely sufficient to prove K-stability. It would be interesting to know exactly which polarisations of a general degree one del Pezzo surface are K-stable. Analytically, a result of Arezzo-Pacard [1] implies that (X, Lλ) admits a cscK metric for λ sufficiently small. In particular, by work of Donaldson [9] and Stoppa [28], this implies (X, Lλ) is K-stable for λ sufficiently small. However, using the technique of slope stability, Ross-Thomas [27, Example 5.30] have shown that there are polarisations of such an X which are K-unstable. The recent proof of the Yau-Tian-Donaldson conjecture [8, 6, 7, 32] in the case L = −KX has emphasised the importance of log K-stability. This concept extends K-stability to pairs (X, D) and conjecturally corresponds to cscK metrics with cone singularities along D. With this in mind, we extend Theorem 1.1 to the log setting as follows. Theorem 1.3. Let ((X, D); L) consist of a Q-Gorenstein log canonical pair (X, D) with canonical divisor KX, such that D is an effective integral reduced Cartier divisor on a polarised variety (X, L). Denote µβ((X, D); L) = −(KX +(1−β)D).Ln−1 . Suppose that Ln (i) α((X, D); L) > n (ii) −(KX + (1 − β)D) ≥ n n+1 µβ((X, D); L) and Then ((X, D); L) is log K-stable with cone angle β along D. n+1 µβ((X, D); L)L. Notation and conventions: By a polarised variety (X, L) we mean a normal complex projective variety X together with an ample line bundle L. We often use the same letter to denote a divisor and the associated line bundle, and mix multiplicative and additive notation for line bundles. 2. Prerequisites 2.1. K-stability. K-stability of a polarised variety (X, L) is an algebraic notion conjecturally equivalent to the existence of a constant scalar curvature Kahler met- ric in c1(L), which requires the so-called Donaldson-Futaki invariant to be positive for all non-trivial test configurations. Definition 2.1. A test configuration for a normal polarised variety (X, L) is a normal polarised variety (X ,L) together with • a proper flat morphism π : X → C, • a C∗-action on X covering the natural action on C, • and an equivariant very ample line bundle L on X such that the fibre (Xt,Lt) over t is isomorphic to (X, L) for one, and hence all, t ∈ C∗. if X is C∗- Definition 2.2. We say that a test configuration is almost trivial isomorphic to the product configuration away from a closed subscheme of codimen- sion at least 2. Definition 2.3. We will later be interested in a slightly modified version of test configurations. In particular, we will be interested in the case where we have a proper flat morphism π : X → P1 with target P1 rather than C such that L is just relatively semi-ample over P1, that is, a multiple of the restriction to each fibre over P1 is basepoint free. We call this a semi-test configuration. 4 RUADHA´I DERVAN As the C∗-action on (X ,L) fixes the central fibre (X0,L0), there is an induced action on H 0(X0,Lk 0) for all k. Denote by w(k) the total weight of this action, which is a polynomial in k of degree n + 1 for k ≫ 0, where n is the dimension of X. Denote the Hilbert polynomial of (X, L) as P(k) = χ(X, Lk) = a0kn + a1kn−1 + O(kn−2) and denote also the total weight of the C∗-action on H 0(X0,Lk 0) as w(k) = b0kn+1 + b1kn + O(kn−1). (3) (4) Definition 2.4. We define the Donaldson-Futaki invariant of a test configuration (X ,L) to be DF(X ,L) = b0a1 − b1a0 a2 0 . (5) We say (X, L) is K-stable if DF(X ,L) > 0 for all test configurations which are not almost trivial. Remark 2.5. For more information on the following remarks, or for a more detailed discussion of K-stability, see [27]. • The definition of K-stability is independent of scaling L → Lr. In particu- lar, it makes sense for pairs (X, L) where X is a variety and L is a Q-line bundle. is given by f1. kP(k) = f0+f1k−1+O(k−2), the Donaldson-Futaki invariant • If one expands w(k) • One should think of test configuration as geometrisations of the one-parameter subgroups that are considered when applying the Hilbert-Mumford cri- terion to GIT stability. In fact, asymptotic Hilbert stability implies K- semistability, since the Donaldson-Futaki invariant appears as the leading coefficient in a polynomial associated with asymptotic Hilbert stability. • The notion of almost trivial test configurations was introduced by Stoppa [29] to resolve a pathology noted by Li-Xu [15, Section 2.2]. Conjecture 2.1. (Yau-Tian-Donaldson) A smooth polarised variety (X, L) admits a constant scalar curvature metric in c1(L) if and only if (X, L) is K-stable. Remark 2.6. This conjecture as stated has recently been proven by Chen-Donaldson- Sun [8, 6, 7] and separately Tian [32] in the case L = −KX (so X is Fano). It is expected to hold in the general case, with possibly some slight modifications to the definition of K-stability, see [30]. 2.2. Odaka's Blowing-up Formalism. In [20], Odaka shows that to check K- stability, it suffices to check the positivity of the Donaldson-Futaki invariant on semi-test configurations arising from flag ideals. Definition 2.7. A flag ideal on X is a coherent ideal sheaf I on X × A1 of the form I = I0 + (t)I1 + . . . + (tN ) with I0 ⊆ I1 ⊆ . . . ⊆ IN −1 ⊆ OX a sequence of coherent ideal sheaves. The ideal sheaves Ij thus correspond to subschemes Z0 ⊇ Z1 ⊇ . . . ⊇ ZN −1 of X. Flag ideals can be equivalently characterised by being C∗-invariant with support on X × {0}. Remark 2.8. The flag ideal I naturally induces a coherent ideal sheaf on X × P1, which we also denote by I. Blowing-up I on X × P1, we get a map π : B = BlI(X × P1) → X × P1. (6) ALPHA INVARIANTS AND K-STABILITY 5 Denote by E the exceptional divisor of the blow-up π : B → X × P1, that is, O(−E) = π−1I. Abusing notation, write L − E to denote (p1 ◦ π)∗L ⊗ O(−E), where p1 : X × P1 → X is the natural projection. Note that the induced map from B → P1 is flat by [27, Remark 5.2]. There is a natural C∗ action on X × P1, acting trivially on X, which lifts to an action on B. With this action, provided L − E is relatively semi-ample over P1 and B is normal, we have that (B,L − E) is a semi-test configuration. Theorem 2.9. [20, Corollary 3.11] Assume that (X, L) is a normal polarised va- riety. Then (X, L) is K-stable if and only if DF(B,Lr − E) > 0 for all r > 0 and for all flag ideals I 6= (tN ) with B normal and Gorenstein in codimension one and with Lr − E relatively semi-ample over P1. Remark 2.10. That B can be assumed normal was noted by Odaka-Sano [22, Proposition 2.1]. The condition that I 6= (tN ) is to ensure B is not almost trivial, see Definition 2.2. Remark 2.11. As a general test configuration (X ,L) is C∗-isomorphic to (X × A1, Lr) away from the central fibre, it is C∗-birational to (X × A1, L). In particular, it is dominated by a blow-up of X × A1 along a flag ideal. Odaka shows that one can choose a flag ideal such that the Donaldson-Futaki invariant of the two test configurations are equal. In order to use the machinery of intersection theory, one must also compactify X × A1 to X × P1. Remark 2.12. In the case the flag is of the form I = I0 + (t), blowing-up I on X × A1 leads to deformation to the normal cone. In [26], Ross-Thomas study test configurations arising from this process. Stability with respect to test configurations arising from blow-ups of the form I = I0 + (t) is called slope stability. Note that Panov-Ross [24, Example 7.8] have shown that the blow-up of P2 at 2 points is slope stable but is not K-stable. One must therefore consider more general flag ideals to check K-stability. One benefit of this formalism is that, for test configurations arising from flag ideals, there is an explicit intersection-theoretic formula for the Donaldson-Futaki invariant. Theorem 2.13. [20, Theorem 3.2] For a semi-test configuration of the form (B = BlIX × P1,L − E) arising from a flag ideal I with B normal and Gorenstein in codimension one, the Donaldson-Futaki invariant is given by (up to multiplication by a positive constant) DF = −n(Ln−1.KX )(L − E)n+1 + (n + 1)(Ln)(L − E)n.(KX + KB/X×P1). (7) Here we have denoted by KX the pull back of KX to B. The intersection numbers Ln−1.KX and Ln are computed on X, while the remaining intersection numbers are computed on B. Replacing L and L by Lr and Lr respectively in formula 7 gives the formula for the Donaldson-Futaki invariant of a test configuration of the form (B,Lr − E). Note that KX + KB/X×P1 = KB/P1. The benefit of splitting this into two terms is that positivity of the contribution from the second term, the relative canonical divisor over X × P1, can be controlled under assumptions on the singularities of X. 6 RUADHA´I DERVAN 2.3. Log Canonical Thresholds. The log canonical threshold of a pair (X, D) is a measure of singularity, related to the complex singularity exponent. It takes into consideration both the singularities of X and D. See [12] for more information on log canonical thresholds. Definition 2.14. Let X be a normal variety and let D =P diDi be a divisor on X such that KX + D is Q-Cartier, where Di are prime divisors. Let π : Y → X be a log resolution of singularities, so that Y is smooth and π−1D ∪ E has simple normal crossing support where E is the exceptional divisor. We can then write KY − π∗(KX + D) ≡X a(Ei, (X, D))Ei where Ei is either an exceptional divisor or Ei = π−1 ∗ Di for some i. That is, either Ei is exceptional or the proper transform of a component of D. We usually abbreviate a(Ei, (X, D)) to ai. We say the pair (X, D) is (8) • log canonical if a(Ei, (X, D)) ≥ −1 for all Ei, • Kawamata log terminal if a(Ei, (X, D)) > −1 for all Ei. By [12, Lemma 3.10] these notions are independent of log resolution. We will later need a form of inversion of adjunction for log canonicity. Theorem 2.15. [11] Let D = D′ + D′′ be a Q-divisor with D′ an effective reduced normal Cartier divisor and D′′ an effective Q-divisor which has no common com- ponents with D′. Then (X, D) is log canonical on some open neighbourhood of D′ if and only if (D′, D′′D′ ) is log canonical. Definition 2.16. We say a variety X is log canonical if (X, 0) is log canonical. Note in particular that log canonical varieties are normal by assumption. For a not necessarily log canonical pair (X, D) we can still use the idea of log canonicity to measure singularities. Definition 2.17. Let X be a normal variety, and let D be a Q-divisor. The log canonical threshold of a pair (X, D) is lct(X, D) = sup{λ ∈ Q>0 (X, λD) is log canonical}. (9) One can generalise the log canonical threshold of a divisor to general coherent ideal sheaves as follows. Definition 2.18. Let I ⊂ OX be a coherent ideal sheaf, and let D be an effective Q-divisor on X. We say that π : Y → X is a log resolution of I and D if Y is smooth and there is an effective divisor F on Y with π−1I = OY (−F ) such that F ∪ E ∪ D has simple normal crossing support, where D is the proper transform of D. Let π : Y → X be such a log resolution and assume the pair (X, D) is log canonical. For a real number c ∈ R, we define the discrepancy of ((X, D); cI) to be (10) a(Ei, ((X, D); cI)) = a(Ei, (X, D)) − c valEi(I). Here by valEi(I) we mean the valuation of the ideal I on Ei, while the a(Ei, (X, D)) are as in Definition 2.14. We say ((X, D); cI) is log canonical if a(Ei, ((X, D); cI)) ≥ −1 for all Ei appearing in a log resolution of I and D. The log canonical threshold of ((X, D); I) is then defined as lct((X, D); I) = sup{λ ∈ Q>0 ((X, D); λI) is log canonical}. (11) ALPHA INVARIANTS AND K-STABILITY 7 Remark 2.19. [12, Proposition 8.5] For a proper birational map f : X ′ → X, we have that lct((X, D); cI) ≤ min Ei⊂X ′(cid:26) 1 + valEi KX ′/X − valEi D c valEi(I) (cid:27) , (12) where our convention for the appearance of the Ei is as in Definition 2.14. Equality is achieved on a log resolution, where this is essentially a rephrasing of the definition of the log canonical threshold. Definition 2.20. Let (X, L) be a log canonical polarised variety. We define the alpha invariant of (X, L) to be α(X, L) = inf m∈Z>0 inf D∈mL lct(X, 1 m D). In particular, for c > 0 the alpha invariant satisfies the scaling property α(X, cL) = 1 c α(X, L). (13) (14) This definition of the alpha invariant is the algebraic counterpart of Tian's orig- inal definition. For further details on the following analytic definition, see [4, Ap- pendix A]. Definition 2.21. Let h be a singular hermitian metric on L, written locally as h = e−2φ. We define the complex singularity exponent c(h) to be c(h) = sup{λ ∈ R>0 for all z ∈ X, hλ = e−2λφ is L1 in a neighbourhood of z}. (15) We then define Tian's alpha invariant αan(X, L) of (X, L) to be αan(X, L) = inf h with ΘL,h≥0 c(h) (16) where the infimum is over all singular hermitian metrics h with curvature ΘL,h ≥ 0. For a compact subgroup G of Aut(X, L), one defines αan similarly, however considering only G-invariant metrics. Theorem 2.22. [4, Appendix A] The alpha invariant α(X, L) defined algebraically equals Tian's alpha invariant αan(X, L). That is, α(X, L) = αan(X, L). (17) Remark 2.23. As every divisor D ∈ L gives rise to a singular hermitian metric, one sees that α(X, L) ≥ αan(X, L). Equality follows from approximation techniques for plurisubharmonic functions. The main consequence of the definition of the alpha invariant is the following theorem of Tian, which states that certain lower bounds on the alpha invariant imply the existence of a Kahler-Einstein metric. Theorem 2.24. [31, Theorem 2.1] Let G be a compact subgroup of Aut(X) and suppose X is a smooth Fano variety with αG(X,−KX) > n n+1 . Then X admits a Kahler-Einstein metric. 8 RUADHA´I DERVAN 3. Alpha Invariants and K-stability In this section we provide a sufficient condition for polarised varieties (X, L) of dimension n to be K-stable. A fundamental quantity will be the slope of a polarised variety. Definition 3.1. We define the slope of (X, L) to be µ(X, L) = −KX.Ln−1 Ln = RX c1(X).c1(L)n−1 RX c1(L)n . (18) The slope of a polarised variety is thus a topological quantity which, after rescaling L, may be assumed equal to 1. Note in particular that µ(X,−KX ) = 1. Remark 3.2. In [27], Ross-Thomas defined a similar quantity, which they also call the slope, defined to be n 2 times our definition. Theorem 3.3. Let (X, L) be a polarised Q-Gorenstein log canonical variety with canonical divisor KX . Suppose that n+1 µ(X, L) and (i) α(X, L) > n (ii) −KX ≥ n n+1 µ(X, L)L. Then (X, L) is K-stable. Remark 3.4. Here, for divisors H, H ′, we write H ≥ H ′ to mean H − H ′ is nef. Note that both conditions are independent of positively scaling L. For L = −KX, the second condition is vacuous, so in this case, this is the algebraic counterpart of a theorem of Tian (Theorem 2.24) and in this case is due to Odaka-Sano [22, Theorem 1.4]. The condition that X is log canonical ensures that α(X, L) ≥ 0, while the condition that X is Q-Gorenstein ensures that −KX exists as a Q-Cartier divisor. Remark 3.5. If µ(X, L) = 0, i.e. Ln−1.KX = 0, the second condition requires −KX to be nef. Suppose −KX is nef but not numerically equivalent to zero, and suppose Ln−1.KX = 0. Then, by the Hodge Index Theorem [16, Theorem 1] we would have Ln−2.K 2 X < 0, contradicting the fact that −KX is nef. In particular, for the second condition of the theorem to hold, X must either be numerically Calabi- Yau or Fano. In the Calabi-Yau case, this theorem also follows from a theorem due to Odaka [21, Theorem 1.1]. A Corollary of Theorem 3.3 is that the automorphism group of (X, L) is dis- crete. Indeed, if Aut(X, L) were to admit a one parameter subgroup, this would give two test configurations with Donaldson-Futaki invariants of opposite sign. But K-stability requires strict positivity of the Donaldson-Futaki invariant, a contradic- tion. Corollary 3.6. If the criteria of Theorem 3.3 are satisfied, then Aut(X, L) is discrete. To prove Theorem 3.3, we first establish an upper bound on the alpha invariant. Proposition 3.7. (c.f. [22, Proposition 3.1]) Let B be the blow-up of X × P1 along a flag ideal, with B normal and Gorenstein in codimension one, L − E relatively semi-ample over P1 and notation as in Remark 2.8. Denote the natural map arising from the composition of the blow-up map and the projection map by Π : B → P1. Denote also ALPHA INVARIANTS AND K-STABILITY 9 • the discrepancies as: KB/X×P1 =P aiEi, • the multiplicities of X × {0} as: Π∗(X × {0}) = Π−1 • the exceptional divisor as: Π−1I = OB(−P ciEi) = OB(−E). ∗ (X × {0}) +P biEi, Then α(X, L) ≤ min i (cid:26) ai − bi + 1 ci (cid:27) . (19) Proof. We are seeking an upper bound on the alpha invariant, where this upper bound is related to the flag ideal I = I0 + (t)I1 + . . . + (tN ) on X × P1. As the divisors considered in the definition of the alpha invariant are divisors on X, we pass from I to its first component I0. The choice of I0 is because I0 is a subsheaf of the full flag ideal I. Let π0 : BlI0 X → X be the blow-up of I0 with exceptional divisor E0. We claim π∗ 0 (mL) being base-point free for some m. However, as L − E is semi-ample restricted each fibre, we know that Imπ∗(mL) is base-point free on each fibre of X × P1. As I0 is a subsheaf of I, the result follows. 0L − E0)) = 0 (mL)) has a section, which exists since multiples of semi-ample line bun- 0 L − E0 is semi-ample. This is equivalent to I m Choose m sufficiently large and divisible such that H 0(BlI0 X, m(π∗ H 0(X, I m dles have sections. Let D be in the linear series H 0(X, I m 0 (mL)). We show that α(X, L) ≤ m lct(X, D) ≤ min i (cid:26) ai − bi + 1 ci (cid:27) . (20) For general ideal sheaves I, J, [19, Property 1.12] states that I ⊂ J implies lct(X, I) ≤ lct(X, J). We therefore see that lct(X, D) = lct(X, ID) ≤ 1 m lct(X, I0). (21) Note that X × {0} is a divisor on X × A1. By a basic form of inversion of adjunction of log canonicity, we have lct(X, I0) = lct((X × P1, X × {0}); I0). (22) One can see this by taking a log resolution of ((X × P1, X × {0}); I0) of the form X × P1 → X × P1, where X → X is a log resolution of (X, I0). Note that for all divisors Ei over X, we have In particular, we see that valEi(I0) ≥ valEi(I). lct((X × P1, X × {0}); I0) ≤ lct((X × P1, X × {0});I) ≤ min i (cid:26) ai − bi + 1 ci (cid:27) . The last inequality is by Remark 2.19. (23) (24) (25) (cid:3) The final ingredients of the proof of Theorem 3.3 are the following lemmas on computing the positivity of terms in Odaka's formula for the Donaldson-Futaki invariant, which are due to Odaka-Sano. We repeat their proof for the reader's convenience. 10 RUADHA´I DERVAN Lemma 3.8. [22, Lemma 4.2] Let L and R be ample and nef divisors respectively on X, with p∗L = L and p∗R = R where p : B → X is the natural map arising from the composition of the blow-up map B → X × P1 and the projection X × P1 → P1. Suppose that L− E is semi-ample on the blow-up BlI X × P1 for some flag ideal I. Then (26) Proof. Firstly note that, because R and L are the pull back of ample and nef divisors respectively from X, which has dimension n, we have Ln.R = 0. Now note that we have the equality (L − E)n.R ≤ 0. −(L − E)n.R = Ln.R − (L − E)n.(R − E) − (L − E)n.E (27) = E.R.(Ln−1 + Ln−2.(L − E) + . . . + (L − E)n−1). (28) As L is nef and the restriction of L − E to the central fibre of the map B → P1 is semi-ample, hence nef, and as E.R is a non-zero effective cycle with support in the central fibre, the result follows. Lemma 3.9. [22, Lemma 4.7] Let E = P ciEi be the exceptional divisor of the blow-up B → X × P1. Then (cid:3) (29) Moreover, strict positivity holds for some Ei, that is, (L − E)n.Ei ≥ 0. (L − E)n.E > 0. (30) Proof. Since each Ei has support in the central fibre, and L − E restricted to the central fibre is semi-ample, hence nef, we have that (L − E)n.Ei ≥ 0. To show (L − E)n.E > 0, we first show (L − E)n.(L + nE) > 0. Note that we have the equality of polynomials n (x − y)n(x + ny) = xn+1 − Xi=1 (n + 1 − i)(x − y)n−ixi−1y2. (31) In fact, the polynomials (x − y)n−ixi−1y2 for 1 ≤ i ≤ n are linearly independent over Q, and for all 0 < s < n, the monomial xsyn+1−s can be written as a linear combination of these polynomials with coefficients in Z. Note that Ln+1 = 0, as L is the pull back of an ample line bundle from X, which has dimension n. In particular, we can write (L − E)n.(L + nE) = −E2. n (n + 1 − i)(L − E)n−i.Li−1! . Xi=1 (32) Let s = dim(Supp(O/I)), where I = I0 + (t)I1 + . . . + (tN ). By dividing I by a power of t if necessary, which does not change the resulting blow-up BlIX × P1 and hence does not change the Donaldson-Futaki invariant of the associated semi-test configuration, we can assume s < n. Perturbing the coefficients in equation (32), we get (L−E)n.(L+nE) = −E2. n Xi=1 where 0 < ǫi ≪ 1 and 0 < ǫ′ ≪ 1. (n + 1 − i + ǫi)(L − E)n−iLi−1!−ǫ′(Ls.(−E)n+1−s) (33) ALPHA INVARIANTS AND K-STABILITY 11 The following lemma then shows that (L − E)n.(L + nE) > 0. Lemma 3.10. [22, Lemma 4.7] (i) −E2.(L − E)n−i.Li−1 ≥ 0 for all 0 < i < n. (ii) Ls.(−E)n+1−s < 0. Proof. (i) Cutting B by general elements of rL and r(L − E) for r ≫ 0, we can assume dim X = 2. In this case, the required equation becomes −E2.(L − E) ≥ 0. Note that L − E is semi-ample restricted to fibres of B → P1 and E has support in the central fibre. In particular, E.(L − E) is an effective cycle with support in fibres of the blow-up map B → X × P1. Since −E is relatively ample over fibres of B → X × P1, we have −E.(E.(L − E)) ≥ 0 and the result follows. (ii) Again cutting B by general elements of Lr for r ≫ 0, we can assume s = 0. The required result then follows by relative ampleness of −E over fibres of B → X × P1. (cid:3) Finally, since (L − E)n.(L + nE) > 0 and (L − E)n.L ≤ 0 by Lemma 3.8, we have (L − E)n.E > 0 as required. (cid:3) Proof. (of Theorem 3.3) We show that the Donaldson-Futaki invariant is positive for all semi-test configurations of the form (B,Lr − E) arising from flag ideals. We assume r = 1 for notational simplicity, the proof in the general case is essentially the same. The idea is to first split formula 2.13 for the Donaldson-Futaki invariant into two terms, which we consider separately. We split the Donaldson-Futaki invariant as DF(B,L − E) = DFnum + DFdisc, DFnum = (L − E)n.(−n(Ln−1.KX)L + (n + 1)(Ln)KX ), DFdisc = (L − E)n.((n + 1)(Ln)KB/X×P1 + n(Ln−1.KX)E). Our second hypothesis in Theorem 3.3 is − KX ≥ n n + 1 µ(X, L)L. (34) (35) (36) (37) In particular, n(Ln−1.KX )L−(n+1)(Ln)KX is nef. So, by Lemma 3.8, DFnum ≥ 0. As (L − E)n.E > 0 by Lemma 3.9, it suffices to show that there exists an ǫ > 0 such that (38) Here we mean that each coefficient of Ei is non-negative in the difference of the divisors. As Ln is positive, this is equivalent to showing (n + 1)(Ln)KB/X×P1 + n(Ln−1.KX )E ≥ ǫE. µ(X, L)E ≥ ǫE. By the first assumption in 3.3, namely that α(X, L) > n KB/X×P1 − n n + 1 (39) n+1 µ(X, L), we see that KB/X×P1 − n n + 1 µ(X, L)E > KB/X×P1 − α(X, L)E. But by the upper bound on the alpha invariant, Proposition 3.7, we see that KB/X×P1 − α(X, L)E ≥ KB/X×P1 − min i { ai − bi + 1 ci }E. (40) (41) 12 RUADHA´I DERVAN Here we have used notation as in Proposition 3.7. Finally, we see that KB/X×P1 − n n+1 µ(X, L)E > KB/X×P1 − mini{ ai−bi+1 = P aiEi − mini{ ai−bi+1 − mini{ ai−bi+1 = P( ai−bi+1 ≥ 0. }P ciEi }P ciEi } + bi−1 ci ci ci ci ci The result follows as (L − E)n.E > 0, by Lemma 3.9. )ciEi (cid:3) Remark 3.11. One can marginally strengthen Theorem 3.3 as follows. The posi- tivity of the alpha invariant is used in the proof of Theorem 3.3 as it appears as a coefficient of the exceptional divisor E. In particular, one has a term with positive contribution of the form (α(X, L) − n n+1 µ(X, L))E. By the proof of Lemma 3.9, we have that (L − E)n.(L + nE) > 0. Using this, one can use the positivity of the contribution of the term (α(X, L) − n n+1 µ(X, L))E to slightly weaken the require- ment that −KX ≥ n n+1 µ(X, L)L. However, the resulting hypothesis still implies that −KX is either ample or numerically trivial. We therefore omit the details. Remark 3.12. For any compact subgroup G ⊂ Aut(X, L), Odaka-Sano [22, Sec- tion 2.2] have defined a form of stability, which they call G-equivariant K-stability and conjecture to be equivalent to K-stability. Definition 3.13. Let G ⊂ Aut(X, L) be compact, and define a G-test configuration (X ,L) be a test configuration equipped with an extension of the natural G-action on (X ,L)π−1(A1−{0}) to (X ,L) which commutes with the C∗-action on (X ,L). We say (X, L) is G-equivariantly K-stable if the Donaldson-Futaki invariant of all G-test configuration is strictly positive for all non-trivial test configurations. As G-test configurations give rise to G-invariant flag ideals, Theorem 3.3 can be adapted to G-equivariant K-stability as follows. Corollary 3.14. Let (X, L) be a polarised Q-Gorenstein log canonical variety with canonical divisor KX , and let G ⊂ Aut(X, L) be a compact subgroup. Suppose that (i) αG(X, L) > n (ii) −KX ≥ n n+1 µ(X, L)L. n+1 µ(X, L) and Then (X, L) is G-equivariantly K-stable. 4. Examples By showing that the alpha invariant of a line bundle is a continuous function of the line bundle, we first show that the conditions of Theorem 3.3 are open when varying the polarisation. To prove this, we need a lemma regarding adding ample divisors and alpha invariants. Lemma 4.1. Let (X, L) be a log canonical polarised variety, and let D be an ample Q-divisor on X. Then α(X, L + D) ≤ α(X, L). Proof. Take any divisor D′ ∈ m′L. We find a divisor F ∈ p(L + D) such that lct(X, 1 m′ D′). Suppose that mD is a Z-divisor. Let F = mD′ + mm′D ∈ mm′(L + D). As F − mD′ = mm′D is ample, hence effective, the discrepancies satisfy (42) p F ) ≤ lct(X, 1 a(Ei, (X, F )) ≤ a(Ei, (X, mD′)) ALPHA INVARIANTS AND K-STABILITY 13 for all divisors Ei over X [13, Lemma 2.27], so we have 1 m′ D′). 1 mm′ F ) ≤ lct(X, lct(X, Therefore α(X, L + D) ≤ α(X, L). (43) (cid:3) Using this we can show that the alpha invariant of a polarised variety (X, L) is a continuous function of L. Proposition 4.2. Let (X, L) be a polarised klt variety and D be a Q-divisor on X. Then for all ǫ > 0 there exists a δ > 0 depending on D such that α(X, L) − α(X, L + δD) < ǫ. Proof. Firstly, suppose both γL + D and γL − D are ample for some 0 < γ < 1. By the inverse linearity property of the alpha invariant noted in Definition 2.20, we then have (44) Lemma 4.1 implies that subtracting ample divisors raises the alpha invariant. Ap- plying Lemma 4.1 by subtracting γL − D from (1 + γ)L, we see that α(X, L) = (1 + γ)α(X, (1 + γ)L). (1 + γ)α(X, (1 + γ)L) ≤ (1 + γ)α(X, L + D). This in particular implies α(X, L) − α(X, L + D) ≤ γα(X, L + D). On the other hand, since γL + D is ample, applying Lemma 4.1 by adding γL + D to (1 − γ)L gives α(X, L) = (1 − γ)α(X, (1 − γ)L) ≥ (1 − γ)α(X, L + D). (45) (46) (47) (48) (49) (50) Therefore Note that equation (48) implies that α(X, L + D) ≤ 1 1−γ α(X, L), so we have α(X, L) − α(X, L + D) ≤ γα(X, L + D). α(X, L) − α(X, L + D) ≤ α(X, L). γ 1 − γ Since ampleness is an open condition, there exists a c > 0 such that both L + cD cǫ and L − cD are ample. We now show continuity of the alpha invariant at L. Given ǫ > 0 let δ = 2α(X,L)+ǫ . Then both (δc−1)L + δD and (δc−1)L − δD are ample. In our situation δc−1 γ = δc−1 = 1−δc−1 = 2α(X,L) , we therefore have 2α(X,L)+ǫ < 1, so we can apply equation (50). Noting that ǫ ǫ α(X, L) − α(X, L + δD) ≤ ǫ 2 < ǫ. (51) (cid:3) Corollary 4.3. Suppose (X, L) is a klt Q-Gorenstein polarised variety such that (i) α(X, L) > n (ii) −KX > n n+1 µ(X, L) and n+1 µ(X, L)L. Note that both inequalities are strict. Then for all Q-divisors D, there exists an ǫ > 0 such that L + ǫD is K-stable. 14 RUADHA´I DERVAN Proof. This follows by Proposition 4.2 and continuity of intersections of divisors. (cid:3) We now apply Theorem 3.3 to a general degree one del Pezzo surface X. Here the genericity condition means that −KX contains no cuspidal curves, so α(X,−KX) = 1 ([3], Theorem 1.7). Note that X is isomorphic to the blow-up of P2 at a configu- ration of 8 points in general position. Denote by H the hyperplane divisor pulled back from P2, let Ei be the 8 exceptional divisors arising from an isomorphism X ∼= Blp1,...,p8P2 and let Lλ = 3H −P7 Theorem 4.4. (X, Lλ) is K-stable for i=1 Ei − λE8. 19 25 ≈ 1 9 (10 − √10) < λ < √10 − 2 ≈ 29 25 . (52) To prove this result, we first obtain a lower bound for α(X, Lλ) using the follow- ing two lemmas. Lemma 4.5. [33, Corollary 6] Let S be a smooth variety, let p ∈ S, and let D, B be effective Q-divisors on S with p ∈ S, p ∈ B such that (S, D) is not log canonical at p but (S, B) is log canonical at p. Then, for all c ∈ [0, 1) ∩ Q, is not log canonical at p. (S, 1 1 − c (D − cB)) (53) Lemma 4.6. [18, Lemma 2.4 (i)] Let (S, D) be pair consitisting of a smooth surface S and an effective Q-divisor D such that (S, D) is not log canonical at p. Then multp D > 1. Proposition 4.7. For X be a general del Pezzo surface of degree one as above, and Lλ = 3H −P7 i=1 Ei − λE8 with λ ≥ 0, we have , 1(cid:27) . α(X, Lλ) ≥ min(cid:26) 1 2 − λ Proof. Suppose for contradiction ω < min{ 1 2−λ , 1}, and there exists an effective Q- divisor D with mD ∈ mLλ for some m such that (S, ωD) is not log canonical at some point p ∈ X. Write D = aC + Ω, where C ∈ − KX is a Z-divisor with p ∈ C, and C 6⊆ Supp(Ω). Note that since X is a general degree one del Pezzo surface, we have that ωC is log canonical by ([25], Proposition 3.2). Since Ω = D − aC, we see that (54) Ω.H = (D − aC).H = (1 − a)(−KX ).H + (1 − λ)E8.H = 3(1 − a). (57) But since H is ample and Ω is effective, Ω.H ≥ 0. Thus a ≤ 1, and in particular, ωa < 1. 1−ωa (ωD − ωaC)) is not log canonical at p. Note 1−ωa Ω) > 1 by Lemma 4.6. But since By Lemma 4.5, we see that (S, that ωD − ωaC = ωΩ. Therefore multp( ω C 6⊆ Supp(Ω), we have that 1 ωC.Ω ≥ ω multp Ω > 1 − ωa. (58) (55) (56) ALPHA INVARIANTS AND K-STABILITY 15 Thus ω(2 − λ) = ωD.C = ω(aC.C + Ω.C) > ωa + 1 − ωa = 1. But this implies ω > 1 2−λ , a contradiction. (59) (60) (61) (62) (cid:3) Using this lower bound we can prove Theorem 4.4. Proof. (of Theorem 4.4) For Theorem 3.3 to apply, the two equations that must be satisfied are (i) α(X, Lλ) > 2 (ii) −KX − 2 In our case µ(X, Lλ) = 2−λ 3 µ(X, Lλ) and 3 µ(X, Lλ)Lλ is nef. 1 2−λ2 . By Proposition 4.7, for λ ≤ 1, we have α(X, Lλ) ≥ 2−λ and the first condition is always satisfied. When λ ≥ 1, we have α(X, Lλ) ≥ 1 3 (1 + √7). and the first condition requires 2 − 3λ2 + 2λ > 0, which is true for λ < 1 For the second condition to apply, we require 3H − 7 Xi=1 Ei − 6 − 4λ − λ2 2 + 2λ − 3λ2 E8 (63) to be nef. Note for λ = 1 this holds. By the cone theorem [13, Theorem 3.7] applied to a del Pezzo surface, to check when a line bundle on a del Pezzo surface is nef, it suffices to check it has non- negative intersection with all curves of negative self-intersection. However, by the adjunction formula, all curves C on a del Pezzo surface of negative self-intersection are exceptional, that is, C.C = −1. Therefore, to check when a line bundle is nef on a del Pezzo surface, it suffices to check it has non-negative intersection with all exceptional curves. For the blow-up of P2 at 8 points, from [17, Theorem 26.2] we know that the exceptional curves are the proper transforms of: • points which are blown up, with class Ei, • lines through pairs of points, with class H − Ei − Ej, • conics through 5 points, with class 2H −P5 Ei, • cubics through 7 points, vanishing doubly at Ej for some j, with class 3H − Ej −P7 Ei, • quartics through 8 points, vanishing doubly at Ej , Ek, El, with class 4H − Ei − Ej − Ek −P8 El, • quintics through 8 points, vanishing doubly at 6 points, with class 5H − Ej − Ek − 2P6 Ei, • sextics through 8 points, vanishing doubly at 7 points and triply at another, with class 6H − 3Ej − 2P7 Ei. For a line bundle of the form W = 3H −P7 i=1 Ei − δE8, the first condition requires δ ≥ 0, the second and third conditions require δ ≤ 2, the fourth, fifth and sixth require δ ≤ 3 3 . In particular, δ = 4 2 , while the seventh condition requires δ ≤ 4 3 is the maximal value of δ with W nef. 16 RUADHA´I DERVAN The equation that therefore must be satisfied for 3H −P7 to be nef is i=1 Ei − 6−4λ−λ2 2+2λ−3λ2 E8 . (64) As λ > 0, the condition that 6− 4λ− λ2 ≥ 0 requires λ < √10− 2 ≈ 29 bound is equivalent to 25 . The upper 0 ≤ 6 − 4λ − λ2 2 + 2λ − 3λ2 ≤ 4 3 9λ2 − 20λ + 10 ≤ 0, which is true for 1 which both conditions required to apply Theorem 3.3 are satisfied is 1 9 (10 − √10) ≤ λ ≤ 1 9 (10 + √10) ≈ 29 20 . Therefore, the range for 9 (10−√10) < λ < √10 − 2. (65) (cid:3) Remark 4.8. Note that the lower bound for α(X, Lλ) may not be sharp. A more delicate analysis of the Q-divisors linearly equivalent to Lλ may provide a sharper lower bound. However, both the upper and lower bounds obtained in Theorem 4.4 were given by the requirement that −KX ≥ 2 3 µ(X, Lλ)Lλ. Since we calculated exactly for which λ that this condition holds, we have calculated precisely the range of λ for which Theorem 3.3 applies. Remark 4.9. Analytically, Arezzo-Pacard [1, Theorem 1.1] have shown that if a general (X, L) admits a constant scalar curvature Kahler metric in c1(L), and π : Y → X is the blow-up of X at a point p, then (Y, π∗L − ǫE) admits a constant scalar curvature Kahler metric in c1(π∗L − ǫE) for sufficiently small ǫ, provided Aut(X, L) is discrete. As the existence of a cscK metric in c1(L) implies K-stability by work of Stoppa [28, Theorem 1.2] and Donaldson [9], this in particular implies that (X, Lλ) as in Theorem 4.4 is K-stable for λ sufficiently small. Theorem 4.4 shows that (X, Lλ) is K-stable for 1 using the techniques of slope stability, Ross-Thomas [26, Example 5.30] have shown that there are polarisations of a general degree one del Pezzo surface X which are K-unstable. It would be interesting to know exactly which polarisations of a general degree one del Pezzo surface are K-stable. 9 (10−√10) < λ < √10− 2. On the other hand, 5. Log Alpha Invariants and Log K-stability In this section we extend Theorem 3.3 to K-stability with cone singularities along an anti-canonical divisor, which conjecturally corresponds to the existence of cscK metrics with cone singularities along a divisor. For a general introduction to log K-stability, see [23]. Definition 5.1. Let (X, Lr) be a normal polarised variety, and let D be an effective integral reduced divisor on X. We define a log test configuration for ((X, D); Lr) to be a pair of test configurations (X ,L) for (X, Lr) and (Y,LY ) for (D, LrD) with a compatible C∗ action. We denote by ((X ,Y);L) the data of a log test configuration. Denote the Hilbert polynomials of (X, L) and (D, LD) respectively as P(k) = χ(X, Lk) = a0kn + a1kn−1 + O(kn−2), P(k) = χ(D, Lk D) = a0kn−1 + a1kn−2 + O(kn−3). (66) (67) ALPHA INVARIANTS AND K-STABILITY 17 Denote also the total weights of the C∗-actions on H 0(X0,Lk respectively as 0) and H 0(Y0,Lk0Y0 ) w(k) = b0kn + b1kn−1 + O(kn−2), w(k) = b0kn+1 + b1kn + O(kn−1). (68) (69) We define the log Donaldson-Futaki invariant of ((X ,Y);L) with cone angle 2πβ for 0 ≤ β ≤ 1 to be DFβ((X ,Y);L) = 2(b0a1 − b1a0) + (1 − β)(a0b0 − b0a0). (70) We say that ((X, D); L) is log K-stable with cone angle 2πβ if DFβ((X ,Y);L) > 0 for all log test configurations ((X ,Y);L) with X ,Y normal, Gorenstein in codimension one and such that ((X ,Y);L) is not almost trivial. Note that the usual Donaldson-Futaki invariant for the test configuration (X ,L) is b0a1−b1a0 0 for ease of notation. Since K-stability is independent of positively scaling L, this makes no difference to the definition of K-stability. , we have multiplied by 2a2 a2 0 Odaka-Sun [23] have extended the blowing-up formalism of Odaka to the log case. Recall that to certain flag ideals I on X × A1 one can associate a semi-test configuration by the following method. Blowing-up I on X × P1, we get a map π : B = BlI(X × P1) → X × P1. (71) (D × P1) and let E be the exceptional divisor of the Denote B(D×P1) = BlI(D×P1 ) blow-up π : B → X × P1, that is, O(−E) = π−1I. Abusing notation, write L − E to denote (p1 ◦ π)∗L ⊗ O(−E), where p1 : X × P1 → X is the natural projection. Theorem 5.2. [23, Corollary 3.6] A normal polarised variety (X, L) is log K-stable with cone angle 2πβ if and only if DFβ((B,B(D×P1));Lr − E) > 0 for all r > 0 and for all flag ideals I such that B,B(D×P1) are normal and Gorenstein in codimension one, Lr − E is relatively semi-ample over P1 and I 6= (tN ). Moreover, there is an explicit formula for the log Donaldson-Futaki invariant for log test configurations arising from flag ideals. Theorem 5.3. [23, Theorem 3.7] With all notation as above, we have DFβ((B,B(D×P1));L − E) = −n(Ln−1.(KX + (1 − β)D))(L − E)n+1+ + (n + 1)(Ln)(L − E)n.(KX + (1 − β)D + (KB/((X,(1−β)D)×P1)exc). (73) Here we have denoted by KB/((X,(1−β)D)×P1)exc the exceptional terms of KB − π∗(KX×P1 +(1−β)D), and KX the pull back of KX to B. The intersection numbers Ln−1.KX and Ln are computed on X, while the remaining intersection numbers are computed on B. Replacing L and L by Lr and Lr respectively in formula 72 gives the formula for the Donaldson-Futaki invariant of a test configuration of the form (B,Lr − E). (72) We can extend the definition of the alpha invariant to the log setting as follows. Definition 5.4. Let ((X, D); L) consist of a log canonical pair (X, D) with L an ample Q-line bundle. We define the log alpha invariant of ((X, D); L) to be α((X, D); L) = inf m∈Z>0 inf F ∈mL lct((X, D); 1 m F ). (74) 18 RUADHA´I DERVAN Berman [2, Section 6] has provided an analytic counterpart to the log alpha invariant as follows. Definition 5.5. Let ((X, D); L) consist of a Kawamata log terminal pair (X, D) with X smooth, L an ample Q-line bundle and D = P diDi a simple normal crossing divisor with Di = {fi = 0}. Let h be a singular hermitian metric on L, written locally as h = e−2φ. We define the complex singularity exponent cD(h) to be c(h) = supnλ ∈ R>0 for all z ∈ X, hλ = e−2λφYfi−λdi is L1 near zo . We then define Tian's log alpha invariant αan((X, D); L) of ((X, D); L) to be αan((X, D); L) = inf h with ΘL,h≥0 c(h) (75) (76) where the infimum is over all singular hermitian metrics h with curvature ΘL,h ≥ 0. For a compact subgroup G of Aut(X, L), one defines αan similarly, however considering only G-invariant metrics. Theorem 5.6. [2, Section 6] The log alpha invariant α((X, D); L) defined alge- braically equals Tian's log alpha invariant αan((X, D); L). That is, α((X, D); L) = αan((X, D); L) (77) We can now extend Theorem 3.3 to the log setting. Theorem 5.7. Let ((X, D); L) consist of a Q-Gorenstein log canonical pair (X, D) with canonical divisor KX, such that D is an effective integral reduced Cartier divisor on a polarised variety (X, L). Denote µβ((X, D); L) = −(KX +(1−β)D).Ln−1 Suppose that Ln (i) α((X, D); L) > n (ii) −(KX + (1 − β)D) ≥ n n+1 µβ((X, D); L) and n+1 µβ((X, D); L)L. Then ((X, D); L) is log K-stable with cone angle β along D. Remark 5.8. In the case L = −KX, and D ∈ − KX, this result is due to Odaka- Sun [23, Theorem 5.6]. Again in the case L = −KX, this is the analytic counterpart of a result of Berman ([2], Theorem 3.11) and Jeffres-Mazzeo-Rubinstein ([10], Lemma 6.9). Explicit examples are given in [5]. To prove this theorem, we extend Proposition 3.7 to the log setting. Proposition 5.9. Let B be the blow-up of X × P1 along a flag ideal, with B nor- mal and Gorenstein in codimension one, L − E relatively semi-ample over P1 and notation as in Remark 2.8. Denote the natural map arising from the composition of the blow-up map and the projection map by Π : B → P1. Denote also • the discrepancies as: KB/X×P1 =P aiEi, • the multiplicities of X × {0} as: Π∗(X × {0}) = Π−1 • the multiplicities of D as: Π∗D = Π−1 ∗ D +P diEi, • the exceptional divisor as: Π−1I = OB(−P ciEi) = OB(−E). (cid:27) . i (cid:26) ai − bi + 1 − (1 − β)di α((X, (1 − β)D); L) ≤ min ci Then ∗ (X × {0}) +P biEi, (78) ALPHA INVARIANTS AND K-STABILITY 19 Proof. Let π0 : BlI0X → X be the blow-up of I0 with exceptional divisor E0. As in Proposition 3.7, we have that π∗ 0 L − E0 is semi-ample. Choose m sufficiently large and divisible such that H 0(BlI0 X, m(π∗ 0 (mL)) has a section, and let F be such a section. We show that 0 L − E0)) = H 0(X, I m α((X, (1 − β)D); L) ≤ m lct((X, D); F ) ≤ min i (cid:26) ai − bi + 1 − (1 − β)di ci (cid:27) . (79) Since for general ideal sheaves I, J, we have I ⊂ J implies lct(X, I) ≤ lct(X, J), we see that lct((X, D); F ) = lct((X, D); IF ) ≤ 1 m lct((X, D); I0). (80) Since (X, D) is log canonical, using inversion of adjunction of log canonicity (Theorem 2.15), we have lct((X, (1 − β)D); I0) = lct((X × P1, X × {0} + (1 − β)D × P1); I0) ≤ lct((X × P1, X × {0} + (1 − β)D × P1);I) ≤ min i (cid:26) ai − bi + 1 − (1 − β)di (cid:27) . ci The last inequality is by Remark 2.19. Using this we can prove Theorem 5.7. (81) (82) (83) (cid:3) Proof. (of Theorem 5.7) We treat the case r = 1 for notational simplicity, the general case is similar. The log Donaldson-Futaki invariant is given by DFβ((B,B(D×P1));L − E) = −n(Ln−1.(KX + (1 − β)D))(L − E)n+1+ + (n + 1)(Ln)(L − E)n.(KX + (1 − β)D + (KB/((X,(1−β)D)×P1)exc). (84) (85) For ease of notation, we let K ′ split the log Donaldson-Futaki invariant into two terms as X = KX + (1 − β)D and K′ = KX + (1 − β)D. We DFβ((B,B(D×P1));L − E) = DFβ,num + DFβ,disc, DFβ,num = (L − E)n.(−n(Ln−1.K ′ DFβ,disc = (L − E)n.((n + 1)(Ln)(KB/((X,(1−β)D)×P1)exc) + n(Ln−1.K ′ X )L + (n + 1)(Ln)K′ X ), X)E). (86) (87) (88) Since −(KX +(1−β)D) ≥ n 0. n+1 µβ((X, D); L)L, Lemma 3.8 implies that DFβ,num ≥ To prove DFβ,disc > 0, since (L − E)n.E > 0 by Lemma 3.9 it suffices to prove the existence of an ǫ > 0 such that KB/((X,(1−β)D)×P1)exc − n n + 1 µβ((X, D); L)E ≥ ǫE. (89) By Proposition 5.9 and the first hypothesis of the theorem, we have that 20 RUADHA´I DERVAN n µβ((X, D); L)E > n + 1 (KB/((X,(1−β)D)×P1)exc − (KB/((X,(1−β)D)×P1)exc − α((X, D); L)E ≥X(ai − (1 − β)di)Ei − min =X(cid:18) ai − bi − (1 − β)di + 1 ci i (cid:26) ai − bi + 1 − (1 − β)di i (cid:26) ai − bi − (1 − β)di − min (cid:27)X ciEi + 1(cid:27) + ci ci ≥ 0. The result follows. (90) (91) (92) ci (cid:19) ciEi bi − 1 (93) (94) (cid:3) Acknowledgements: I would like to thank my supervisor Julius Ross for his support, advice and for many useful discussions. I would also like to thank Je- sus Martinez-Garcia, Ivan Cheltsov, Costya Shramov and John Ottem for helpful conversations. Finally I would like to thank the referee for useful comments, in particular for pointing out Corollary 3.6. References [1] C. Arezzo and F. Pacard. Blowing up and desingularizing constant scalar curvature Kahler manifolds. Acta Math., 196(2):179–228, 2006. [2] R. J. Berman. A thermodynamical formalism for Monge-Ampere equations, Moser-Trudinger inequalities and Kahler-Einstein metrics. arXiv:1011.3976, Nov. 2010. [3] I. Cheltsov. Log canonical thresholds of del Pezzo surfaces. Geom. Funct. Anal., 18(4):1118– 1144, 2008. [4] I. Cheltsov., C. Shramov. With an appendix by J.-P. Demailly. Log-canonical thresholds for nonsingular Fano threefolds. Uspekhi Mat. Nauk, 63(5):73–180, 2009. [5] I. Cheltsov., Y. A. Rubinstein. Asymptotically log Fano varieties. arXiv:1308.2503, Aug. 2013. [6] X.-X. Chen, S. Donaldson, and S. Sun. Kahler-Einstein metrics on Fano manifolds, II: limits with cone angle less than 2π. arXiv:1212.4714, Dec. 2012. [7] X.-X. Chen, S. Donaldson, and S. Sun. Kahler-Einstein metrics on Fano manifolds, III: limits as cone angle approaches 2π and completion of the main proof. arXiv:1302.0282, Feb. 2013. [8] X.-X. Chen, S. Donaldson, and S. Sun. Kahler-Einstein metrics on Fano manifolds, I: ap- proximation of metrics with cone singularities. arXiv:1211.4566, Nov. 2012. [9] S. K. Donaldson. Lower bounds on the Calabi functional. J. Differential Geom., 70(3):453– 472, 2005. [10] T. Jeffres, R. Mazzeo, Y. A. Rubinstein. Kahler-Einstein metrics with edge singularities. arXiv:1105.5216, May 2011. [11] M. Kawakita. Inversion of adjunction on log canonicity. Invent. Math., 167(1):129–133, 2007. [12] J. Koll´ar. Singularities of pairs. In Algebraic geometry-Santa Cruz 1995, volume 62 of Proc. Sympos. Pure Math., pages 221–287. Amer. Math. Soc., Providence, RI, 1997. [13] J. Koll´ar and S. Mori. Birational geometry of algebraic varieties, volume 134 of Cambridge Tracts in Mathematics. Cambridge University Press, Cambridge, 1998. [14] C. LeBrun and S. R. Simanca. Extremal Kahler metrics and complex deformation theory. Geom. Funct. Anal., 4(3):298–336, 1994. [15] C. Li and C. Xu. Special test configurations and K-stability of Fano varieties. arXiv:1111.5398, Nov. 2011. [16] T. Luo. A note on the Hodge index theorem. Manuscripta Math., 67(1):17–20, 1990. [17] Y. I. Manin. Cubic forms, volume 4 of North-Holland Mathematical Library. North-Holland Publishing Co., Amsterdam, second edition, 1986. Algebra, geometry, arithmetic, Translated from the Russian by M. Hazewinkel. ALPHA INVARIANTS AND K-STABILITY 21 [18] J. Martinez-Garcia. Log canonical thresholds of Del Pezzo Surfaces in characteristic p. arXiv:1203.0995, Mar. 2012. [19] M. Mustat¸a. IMPANGA lecture notes on log canonical thresholds. In Contributions to alge- braic geometry, EMS Ser. Congr. Rep., pages 407–442. Eur. Math. Soc., Zurich, 2012. Notes by Tomasz Szemberg. [20] Y. Odaka. A generalization of the Ross-Thomas slope theory. Osaka J. Math., 50:171–185, 2013. [21] Y. Odaka. The Calabi conjecture and K-stability. Int. Math. Res. Not. IMRN, (10):2272– 2288, 2012. [22] Y. Odaka and Y. Sano. Alpha invariant and K-stability of Q-Fano varieties. Adv. Math., 229(5):2818–2834, 2012. [23] Y. Odaka and S. Sun. Testing log K-stability by blowing up formalism. arXiv:1112.1353, Dec. 2011. [24] D. Panov and J. Ross. Slope stability and exceptional divisors of high genus. Math. Ann., 343(1):79–101, 2009. [25] J. Park. Birational maps of del Pezzo fibrations. J. Reine Angew. Math., 538:213–221, 2001. [26] J. Ross and R. Thomas. An obstruction to the existence of constant scalar curvature Kahler metrics. J. Differential Geom., 72(3):429–466, 2006. [27] J. Ross and R. Thomas. A study of the Hilbert-Mumford criterion for the stability of projec- tive varieties. J. Algebraic Geom., 16(2):201–255, 2007. [28] J. Stoppa. K-stability of constant scalar curvature Kahler manifolds. Adv. Math., 221(4):1397– 1408, 2009. [29] J. Stoppa. A note on the definition of K-stability. arXiv:1111.5826, Nov. 2011. [30] G. Sz´ekelyhidi. Filtrations and test configurations. arXiv:1111.4986, Nov. 2011. [31] G. Tian. On Kahler-Einstein metrics on certain Kahler manifolds with C1(M ) > 0. Invent. Math., 89(2):225–246, 1987. [32] G. Tian. K-stability and Kahler-Einstein metrics. arXiv:1211.4669, Nov. 2012. [33] A. Wilson. Smooth exceptional del Pezzo surfaces. PhD Thesis, University of Edinburgh, 2010. Ruadha´ı Dervan, University of Cambridge, UK [email protected]
1007.5314
1
1007
2010-07-29T20:03:16
On the Smoothness of Functors
[ "math.AG" ]
In this paper we try to introduce a good smoothness notion for a functor. We consider properties and conditions from geometry and algebraic geometry which we expect a smooth functor should to have.
math.AG
math
On the Smoothness of Functors A. Bajravani Department of Pure Mathematics, Faculty of Mathematical Sciences, Tarbiat Modares University, Tehran, Iran P. O. Box 14115-134 A. Rastegar∗ Faculty of Mathematics, Sharif University, Tehran, Iran P. O. Box 11155 0 1 0 2 l u J 9 2 ] . G A h t a m [ 1 v 4 1 3 5 . 7 0 0 1 : v i X r a Abstract In this paper we will try to introduce a good smoothness notion for a functor. We consider properties and conditions from geometry and algebraic geometry which we expect a smooth functor should have. Keywords: Abelian Category, First Order Deformations, Multicategory, Tangent Category, Topologizing Subcategory. Mathematics Subject Classification: 14A20, 14A15, 14A22. 1 Introduction Nowadays noncommutative algebraic geometry is in the focus of many basic topics in mathematics and mathematical physics. In these fields, any under consideration space is an abelian category and a morphism between noncommutative spaces is a functor between abelian categories. So one may ask to generalize some aspects of morphisms between commutative spaces to morphisms between noncommutative ones. One of the important aspects in commutative case is the notion of smoothness of a morphism which is stated in some languages, for example: by lifting property E-mail Addresses: [email protected] [email protected] (*)Corresponding Author 1 as a universal language, by projectivity of relative cotangent sheaves as an algebraic language and by inducing a surjective morphism on tangent spaces as a geometric language. In this paper, in order to generalize the notion of smooth morphism to a functor we propose three different approaches. A glance description for the first one is as follows: linear approxima- tions of a space are important and powerful tools. They have geometric meaning and algebraic structures such as the vector space of the first order deformations of a space. So it is legitimate to consider functors which preserve linear approximations. On the other hand first order deformations are good candidates for linear approximations in categorical settings. These observations make it reasonable to consider functors which preserve first order deformations. The second one is motivated from both Schlessinger's approach and simultaneous deformations. Briefly speaking, a simultaneous deformation is a deformation which deforms some ingredients of an object simultaneously. Deformations of morphims with nonconstant target, deformations of a couple (X, L), in which X is a scheme and L is a line bundle on X, are examples of such deforma- tions. Also we see that by this approach one can get a morphism of moduli spaces of some moduli families. We get this, by fixing a universal ring for objects which correspond to each other by a smooth functor. Theorem 7 connects this notion to the universal ring of an object. In 3.1 and 3.2 we describe geometrical setting and usage of this approach respectively. The third notion of smoothness comes from a basic reconstruction theorem of A. Rosenberg, in- fluenced by ideas of A. Grothendieck. We think that this approach can be a source to translate other notions from commutative case to noncommutative one. In remarks 3 and 4 we notice that these three smoothness notions are independent of each other. Throughout this paper Art will denote the category of Artinian local k-algebras with quotient field k. By Sets, we denote the category of sets which its morphisms are maps between sets. Let F and G be functors from Art to Sets. For two functors F, G : Art → Sets the following is the notion of smoothness between morphisms of F and G which has been introduced in [8]: A morphism D : F → G between covariant functors F and G is said to be a smooth morphism of functors if for any surjective morphism α : B → A, with α ∈ Mor(Art), the morphism F (B) → F (A) × G(B) G(A) is a surjective map in Sets. Note that this notion of smoothness is a notion for morphisms between special functors, i.e. func- tors from the category Art to the category Sets, while the concepts for smoothness which we introduce in this paper are notions for functors, but not for morphisms between them. A functor F : Art → Sets is said to be a deformation functor if it satisfies in definition 2.1. of [5]. For a fixed field k the schemes in this paper are schemes over the scheme Spec(k) otherwise 2 it will be stated. 2 First Smoothness notion and some examples 1.1 Definition: Let M and C be two categories. We say that the category C is a multicategory over M if there exists a functor T : C → M , in which for any object A of M , T −1(A) is a full subcategory of C. Let C and C be two multicategories over M and M respectively. A morphism of multicategories C and C is a couple (u, ν) of functors, with u : C → C and ν : M → M such that the following diagram is commutative: f → M C u ↓ ↓ ν C → M The category of modules over the category of rings and the category of sheaves of modules over the category of schemes are examples of multicategories. 1.2 Definition: For a S-scheme X and A ∈ Art, we say that X is a S-deformation of X over A if there is a commutative diagram: X → X ↓ ↓ S → S × k A in which X is a closed subscheme of X , the scheme X is flat over S × k A and one has X ∼= S × A S× k X . Note that in the case S = Spec(k), we would have the usual deformation notion and as in the usual case the set of isomorphism classes of first order S-deformations of X is a k-vector space. OX2). The addition of two deformations (X1, OX1) and (X2, OX2) is denoted by (X1S X2, OX1 × OX X 1.3 Definition: i) Let C be a category. We say C is a category with enough deformations, if for any object c of C, one can associate a deformation functor. We will denote the associated deformation functor of c, by Dc. Moreover for any c ∈ Obj(C) let Dc(k[ǫ]) be the tangent space of c, where k[ǫ] is the ring of dual numbers. ii) Let C1 and C2 be two multicategories with enough deformations over Sch /k, and (F, id) be a morphism between them. We say F is a smooth functor if it has the following properties: 1 : For any object M of C1, if M1 is a deformation of M in C1 then F (M1) is a deformation of F (M ) on A in C2. 3 2 : The map is a morphism of tangent spaces. DM (k[ε]) → DF (M)(k[ε]) X 7→ F (X ) The following are examples of categories with enough deformations: 1) Category of schemes over a field k. 2) Category of coherent sheaves on a scheme X. 3) Category of line bundles over a scheme. 4) Category of algebras over a field k. We will need the following lemma to present an example of smooth functors: Lemma 1. Let X, X1, X2 and X be schemes over a fixed scheme S. Assume that the following diagram of morphisms between schemes is a commutative diagram. X ❄ X2 i1 ✲ X1 g ❄ ✲ X i2 If i1 is homeomorphic on its image, then so is i2. Proof. See Lemma (2.5) of [9]. ✷ Example 1. Let Y be a flat scheme over S. Then the fibered product by Y over S is smooth. More precisely, the functor: is smooth. F : Sch /S → Sch /Y Y F (X) = X × S Let X be a closed subscheme of X . Then X × S Y is a closed subscheme of X × S Y . To get the A, it suffices to has flatness of Y over S. It can also be verified easily flatness of X × S Y over S × k that the isomorphism: (X × S Y ) × A S× k S ∼= X × S Y 4 is valid. Therefore X × S Y is a S-deformation of X × S Y if X is such a deformation of X. This verifies the first condition of item (ii) of definition 1.3. To prove the second condition we need the following: Lemma 2. Let Y , X1 and X2 be S-schemes. Assume that X is a closed subscheme of X1 and X2. Then we have the following isomorphism: (X1S X X2) × S Y ∼= (X1 × S Y ) S Y X× S (X2 × S Y ). Proof. For simplicity we set: X1 ∪ X X2 = X , (X1 × S Y ) [ Y X× S (X2 × S Y ) = Z By universal property of Z we have a morphism θ : Z → X × S Y . We prove that θ is an isomorphism. Let i1 : X1 → X , i2 : X2 → X , j1 : X1 × S Y → Z and j2 : X2 × S Y → Z be the inclusion morphisms. Set theoretically we have: j1(X1 × S Y ) S j2(X2 × i1(X1) S i2(X2) S Y ) = Z = X (I) (II) Now consider the following commutative diagrams: X X1 ✒ f g ❘ X2 Y X1 × S ✒ i1 ❘ X ✒ i2 j1 j2 e ❘ X2 × S Y h 5 ✲ ✒ Z θ ❄❘ ✲ Y X × S X × S Y g1 g2 Let z ∈ X × S Y , α = PX (z) ∈ X and β = PY (z) ∈ Y in which PX and PY are the first and second projections from X × S Y to X and Y respectively. Then by relation (II) one has α ∈ i1(X1) or α ∈ i2(X2). If α = i1(α1) ∈ i1(X1), then α1 and β go to the same element in S by ηX1 and ηY in which ηX1 : X1 → S and ηY : Y → S are the maps which make X1 and Y schemes over S. Y such that P X1 (γ) = α1 and P Y (γ) = β in which P X1 Therefore there exists an element γ in X1 × S and P Y are the first and second projections from X × S Y to X1 and Y respectively. By universal property of fibered products γ belongs to X × S Y and θ(γ) = z. The proof for the case α ∈ i2(X) is similar. This implies that θ is surjective. For injectivity of θ assume that θ(z1) = θ(z2). The relation (I) implies that z1 and z2 belong to im(j1) S im(j2). Set z1 = j1(c1) and z2 = j2(c2). There are two cases: if z1, z2 ∈ im(j1) ∩ im(j2), then the lemma 1 implies e(c1) 6= e(c2) when c1 6= c2. Now by commutativity of the subdiagram: X1 × S Y ✲ Y X × S ✻ θ s j1 we have θ(z1) 6= θ(z2) when z1 6= z2. Otherwise assume that z1 ∈ im(j1) and z2 ∈ im(j2) − im(j1). In this case one can see easily that i1P X1 (c1) = i2q2(c2) in which q2 is the first projection from X2 × Y to X2. Since X is the fibered S Z sum of X1 and X2, there exists an element x ∈ X such that i1f (x) = i2g(x), f (x) = P X1 (c1) and g(x) = q2(c2). Set y = p2e(c1) in which p2 is the second projection from X × S Y to Y . By a diagram chasing we see that x and y go to the same element in S. This implies that there exists an element ǫ in X × S which is mapped to x and y by first and second projections, respectively. Also it is easy to see that the equalities g1(x, y) = c1 and g2(x, y) = c2 are valid. Since Z is the fibered sum of X1 × Y S Y Y , we have z1 = z2 which means that θ is injective. This together with the and X2 × S Y on X × S surjectivity of θ implies that θ is bijective. Continuity of θ and its inverse, follow by a diagram chasing. Finally we should prove that OX × Y ∼= OZ . Since the claim is local, it is sufficient to prove it for affine schemes. Let X be an affine scheme, so X1, X2 and X are affine schemes, since they are closed subschemes of X each one defined by a nilpotent sheaf of ideals. Set X = Spec(A), X1 = Spec(A1), Y ∼= OZ X2 = Spec(A2), X = Spec(A0), Y = Spec(B) and S = Spec(C). The isomorphism OX × reduces to the following isomorphism: S S (A1 × A0 A2) ⊗ C B ∼= (A1 ⊗ C B) × A0⊗ C B (A2 ⊗ C B). 6 Define a morphism as follows: d : (A1 × A0 A2) ⊗ C B → (A1 ⊗ C B) × A0⊗ C B (A2 ⊗ C B) d((a1, a2) ⊗ b) = (a1 ⊗ b, a2 ⊗ b). By a simple commutative algebra argument it can be shown that this is in fact an isomorphism. This completes the proof of lemma. ✷ This lemma shows that the fibered product functor, induces an additive homomorphism on tangent spaces. To check linearity with respect to scalar multiplication, take an element a in the field k. Multiplication by a is a ring homomorphism on D. This homomorphism induces a morphism from D and scalar multiplication on tDX , comes from composition of this map with π. S × k D to S × k Y . These together give the linearity of In other words this gives a map from X × S Y into X × S homomorphism induced from F with respect to scalar multiplication. This observation together with the lemma 2, give the smoothness of the fibered product functor. Lemma 3. Let X and Y be arbitrary schemes and assume that there exist morphisms h and g from η to η1 and η2, where η, η1, η2 are sheaves of OX -modules on the scheme X. Then for any morphism f : X → Y we have the following isomorphisms: f∗(η1 × η f ∗(ρ1 × ρ η2) ∼= f∗(η1) × f∗(η) ρ2) ∼= f ∗(ρ1) × f ∗(ρ) f∗(η2) f ∗(ρ2). Proof. For the first isomorphism, it is enough to consider the definition of direct image of sheaves. To prove the second one, assume that (Mi)i∈I , (Ni)i∈I and (Pi)i∈I are direct systems of modules over a directed set I. We have to prove that lim i∈I (Mi × Pi Ni) ∼= (lim i∈I (Mi)) × (limi∈I (Pi)) (lim i∈I (Ni)). The above isomorphism can be proved by elementary calculations and using elementary properties of direct limits. ✷ Example 2. Let f : X → Y be a flat morphism of schemes. Then f∗ and f ∗ are smooth functors. In fact let η be a coherent sheaf on X and η1 ∈ Coh(X × k assumptions we would have: D) be a deformation of η. By these (f∗(η)) ⊗ D k = f∗(η1 ⊗ D k) = f∗(η). 7 Moreover f∗(η1) is flat on D, because η is flat on D. This implies that f∗ satisfies in the first condition of smoothness. The second one is the first isomorphism of lemma 3. Therefore f∗ is smooth. Smoothness of f ∗ is similar to that of f∗. Assuming this notion of smoothness we can generalize another aspect of geometry to categories. 1.9 Definition: Let C be a category with enough deformations. We define the tangent category of C, denoted by T C, as follows: Obj(T C) := S c∈Obj(C) TcC MorT C (υ, ω) := Mor(V, W ) which by TcC, we mean the tangent space of Dc. Moreover υ and ω are first order deformations of V and W . Remark 1. (i) It is easy to see that a smooth functor induces a covariant functor on the tangent categories. (ii) Let C be an abelian category. Then its tangent category is also abelian. The following is a well known suggestion of A. Grothendieck: Instead of working with a space, it is enough to work on the category of quasi coherent sheaves on this space. This suggestion was formalized and proved by P. Gabriel for noetherian schemes and in its general form by A. Rosenberg. To do this, Rosenberg associates a locally ringed space to an abelian category A. In a special case he gets the following: Theorem 4. Let (X, OX ) be a locally ringed space and let A = QCoh(X). Then (Spec(A), OSpec(A)) = (X, OX ) where Spec(A) is the ringed space which is constructed from an abelian category by A. Rosenberg. Proof. See Theorem (A.2) of [7]. ✷ The definition of tangent category and theorem 4 motivates the following questions which the authors could not find any positive or negative answer to them until yet. Question 1: For a fixed scheme X consider T QCoh(X) and T X, the tangent category of category of quasi coherent sheaves on X and the tangent bundle of X respectively. Can T X be recovered from T QCoh(X) by Rosenberg construction? Question 2: Let M be a moduli family with moduli space M . Consider M as a category and consider its tangent category T M. Is there a reconstruction from T M to T M ? 8 3 Second Smoothness Notion Definition 3.1 : Let F : Sch /k → Sch /k be a functor with the following property: For any scheme X and an algebra A ∈ Obj(Art), F (X ) is a deformation of F (X) over A if X is a deformation of X over A. We say F is smooth at X, if the morphism of functors ΘX : DX → DF (X) is a smooth morphism of functors in the sense of Schlessinger (See [8]). F is said to be smooth if for any object X of Sch /k, the morphism of functors ΘX is smooth. The following lemma describes more properties of smooth functors. Lemma 5. (a) Assume that C1, C2 and C3 are multicategories over the category Sch /k. Let F1 : C1 → C2 and F2 : C2 → C3 be smooth functors with the first notion. Then so is their composition. (b) Let F1 : Sch /k → Sch /k and F2 : Sch /k → Sch /k be smooth functors with second notion. Then so is their composition. (c) Let F : Sch /k → Sch /k and G : Sch /k → Sch /k be functors to which F and GoF are smooth with second notion. Then G is a smooth functor. (d) Let F, G, H : Sch /k → Sch /k be smooth functors in the sense of second notion with morphisms H is smooth functor with the of functors F → G and H → G between them. Then the functor F × G second one. Proof. Part (a) of lemma is trivial. (b) Let X ∈ Sch /k and B → A be a surjective morphism in Art. By smoothness of F1, F2 and by remark 2.4 of [8], there exists a surjective map ΘF2(X),F2oF1(X) : DF2oF1(X)(B) × DF2 oF1 (X)(A) DX(A) → DF1(X)(B) × DF1 (X)(A) DX (A) such that we have ΘX,F2oF1(X) = ΘF2(X),F2oF1(X)oΘX,F2(X) in which ΘX,F2(X) is the surjective map induced by smoothness of F2. From this equality it follows the map ΘX,F2oF1(X) is surjective immediately. (c) For a scheme X in the category Sch /k consider a surjective morphism B → A in Art. By smoothness of F , the morphism DX → DF (X) is a surjective morphism of functors. Now apply proposition (2.5) of [8] to finish the proof. (d) Let X ∈ Sch /k and B → A be a surjective morphism in Art. Consider the following commu- tative diagram: 9 DX ✲ DF (X) ✻ s DG(X) Since the morphisms of functors DX → DF (X) and DX → DG(X) are smooth morphisms of functors, proposition 2.5(iii) of [8] implies that DF (X) → DG(X) is a smooth morphism of functors. Similarly DH(X) → DG(X) is a smooth morphism of functors. Again by 2.5(iv) of [8], the morphism of functors: is a smooth morphism of functors. Since in the diagram: DH(X) × DG(X) DF (X) → DH(X) DX ✲ DF (X) DH(X) × DG(X) ✻ s DH(X) the morphisms DX → DH(X) and DH(X) × DG(X) DF (X) are smooth morphisms of functors, part (c) of this lemma implies that DH(X) × DG(X) DF (X) is smooth. This completes the proof. ✷ Remark 2. (i) The same proof works to generalize part (c) of lemma 5 as follows: (´c) Let F : Sch /k → Sch /k and G : Sch /k → Sch /k be functors with GoF smooth and F surjective in the level of deformations in the sense that for any X ∈ Sch /k and any A ∈ Obj(Art) the morphism DX (A) → DF (X)(A) is surjective in Art. Then G is smooth. (ii) One may ask to find a criterion to determine smoothness of a functor. We could not get a complete answer to this question. But by the following fact, one may answer the question at least partially: A functor F : Sch /k → Sch /k is not smooth at X if there exists an algebra A ∈ Art such that the map DX (A) → DF (X)(A) is not surjective in Art, (See [8]). Theorem 7 relates the second smoothness notion to the hull of deformation functors. Recall the hull of a functor is defined in [8]. We need the following: Lemma 6. Let F : Art → Sets be a functor. Then its hulls are non-canonically isomorphic if there exist. 10 Proof. See Proposition 2.9 of [8]. ✷ Theorem 7. Let F : Sch /k → Sch /k be a functor and for a scheme X the functor F has the following properties: (a) F (X ) is a deformation of F (X) if X is a deformation of X. (b) The functor F induces isomorphism on tangent spaces. Then F is smooth at X if and only if (R, F (ξ)) is a hull of DF (X) whenever (R, ξ) is a hull of DX . Proof. To prove the Theorem it is enough to apply (b), (c) of lemma 5, and lemma 6 to the functors ΘX : DX → DF (X) , hR,X : hR → DX , hR,F (X) : hR → DF (X). ✷ For a scheme X let: {pairs (X , ΩX /k) which X is an infinitesimal deformation of X over A } be the isomorphism classes of fibered deformations of X. In the following example we use this notion of deformations of schemes. Example 3. The functor defined by: F : Sch /k → QCoh F (X) = ΩX/k is a smooth functor. Note that if one considers deformations of ΩX/k as usual case, the above functor will not be smooth. The usual deformation of ΩX/k can be described as simultaneous deformation of an object, and differential forms on that object. Also this observation is valid for T X and ωX instead of ΩX . Remark 3. The first and second smoothness notions are in general different. Note that a functor which is smooth with the second notion induces surjective maps on tangent spaces. Since the morphism induced on tangent spaces with first notion of smoothness is not necessarily surjective, a functor which is smooth in the sense of first notion is not necessarily smooth with the sense of second notion. Also a functor which is smooth in the sense of second notion can not be necessarily smooth with the first notion in general. In fact the map induced on tangent spaces by second notion is not necessarily a linear map. It is easy to see that the example 3 is smooth with both of the notions, but examples 1 and 2 are smooth just in the sense of first one. 11 3.1 A Geometric interpretation Let F be a smooth functor at X. By theorem 7, X and F (X) have the same universal rings and this can be interpreted as we are deforming X and F (X) simultaneously. Therefore we have an algebraic language for simultaneous deformations. The example 3 can be interpreted as follows: we are deforming a geometric space and an ingredient of that space, e.g. the structure sheaf of the space or its sheaf of relative differential forms, and these operations are smooth. 3.2 Relation with smoothness of a morphism Let M be a moduli family of algebro - geometric objects with a variety M as its fine moduli space and suppose Y (m) → M is the fiber on m ∈ M . With this assumptions we would have the following bijections: Tm,M ∼= Hom(Spec(k[ǫ]), M ) ∼= {classes of first order deformations of X over A} In fact these bijections states that why deformations are important in geometric usages. Now suppose we have two moduli families M1 and M2 with varieties M1 and M2 as their fine moduli spaces. Also describe M1 and M2 as categories in which there exists a smooth functor F between them. In this setting, if we have a morphism between them, induced from F , then it is a smooth morphism. 4 Third Smoothness Notion This notion of smoothness is completely motivated from Rosenberg's reconstruction theorem, The- orem (A.2) of [7]. For this notion of smoothness we do not use deformation theory. 3.1 Definition: Let F : C1 → C2 be a functor between abelian categories such that there exists a morphism f : Spec(C1) → Spec(C2) induced by the functor F . We say F is a smooth functor if f is a smooth morphism of schemes. Remark 4. (a) Since this smoothness notion uses a language completely different from the two previous ones, it does not imply non of them and vice versa. We did not verified this claim with details but it is not so legitimate to expect that this smoothness implies the previous ones, because deformation theory is not consistent with the Rosenberg construction. This observation together with the remark 3 show that these three notions are independent of each other, having nice geometric and algebraic meaning in their own rights separately. 12 (b) It seems that a functor of abelian categories induces a morphism of schemes in rarely cases. But the cases in which this happens are the cases of enough importance to consider them. Here we mention some cases which this happens. (i) Let f : X → Spec(k) be a morphism of finite type between schemes. Then it can be shown f is induced by f∗ : QCoh(X) → QCoh(Spec(k)) by Rosenberg's construction. This example is important because it can be a source of motivation, to translate notions from commutative case to noncommutative one. (ii) Also the following result of Rosenberg is worth to note: Proposition 8. Let A be an abelian category. (a) For any topologizing subcategory T of A, the inclusion functor T → A induces an embedding Spec(T ) → Spec(A). (b) For any exact localization Q : A → A/S and for any P ∈ Spec(A), either P ∈ Obj(S) or Q(P ) ∈ Spec(A/S); hence Q induces an injective map from Spec(A) − Spec(S) to Spec(A/S). Proof. See Proposition (A.0.3) of [7]. ✷ Acknowledgements: The authors are grateful for referee/s carefully reading of the paper, notable remarks and valuable suggestions about it. References [1] J. Harris, I. Marrison, Moduli of Curves, Graduate Texts in Mathematics, Springer-Verlag, 1994. [2] R. Hartshorne, Deformation Theory, Springer-Verlag, 2010. [3] R. Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics, Springer-Verlag, 1977. [4] W. Lowen , M. V. Bergh, Deformation theory of Abelian categories, Trans. AMS, v.358, n.12, p.5441-5483, 2006. [5] M. Manetti, Extended deformation functors, arxiv:math.AG/9910071 v2 16Mar2001. [6] H. Matsumura, Commutative Ring Theory, Cambridge University Press, 1986. [7] A. L. Rosenberg, Noncommutative schemes, Compositio Mathematica 112: 93-125, 1998. [8] M. Schlessinger, Functors of Artin rings, Trans. AMS 130, 1968, 208-222. 13 [9] K. Schwede, Gluing schemes and a scheme without closed points, unpublished, K.Schwede, math.stanford.edu [10] E. Sernesi, An Overview of Classical Deformation Theory, Notes from seminars Algebraic geometry 2000/2001, Univ. La Sapienza. [11] E. Sernesi, Deformations of schemes, Series: Grundlehren der Mathematicien Wissenchaften, Vol.334, Springer-Verlag, 2006. 14
1105.3156
4
1105
2012-02-10T23:25:11
The conjugacy classes of finite nonsolvable subgroups in the plane Cremona group
[ "math.AG" ]
The aim of this paper is to give a finer geometric description of the algebraic varieties parametrizing conjugacy classes of nonsolvable subgroups in the plane Cremona group.
math.AG
math
THE CONJUGACY CLASSES OF FINITE NONSOLVABLE SUBGROUPS IN THE PLANE CREMONA GROUP. VLADIMIR IGOREVICH TSYGANKOV Abstract. The aim of this paper is to give a finer geometric description of the algebraic varieties parametrizing conjugacy classes of nonsolvable subgroups in the plane Cremona group. Keywords: Cremona group, del Pezzo surface, conic bundle, automorphisms group. MSC: 14E07; 14J26 1. Introduction. The classification of finite subgroups in the plane Cremona group over the field C denoted by Cr2(C) is a classical problem. The history of this problem begins with the work of E. Bertini [2], where are classified the conjugacy classes of subgroups of order 2 in Cr2(C). There were obtained three families of conjugacy classes now called as involution de Jonqui`eres, Geiser and Bertini. However, the classification was incomplete, and the proof was not rigorous. Only recently in [1] was obtained complete and short proof. In 1895 Kantor [11] and Wiman [14] gave a description of finite subgroups in Cr2(C). The list was fairly comprehensive, but was not full in the following aspects. Firstly, for a given finite subgroup on this list could not be defined, whether it is contained in the Cremona group or not. Secondly, the question of conjugacy between the subgroups was not considered. Modern approach to the problem was initiated by the work of Manin [12] and continued in works of Iskovskikh [7], [8], [9]. In the paper [12] is established a clear link between the classification of conjugacy classes of finite subgroups of the Cremona group and the classification of G-minimal rational surfaces (S, G) and G- equivariant birational maps between them. The consideration is divided into two cases: when S is a del Pezzo surface, and when S is a conic bundle. Definition 1.1. Let G be a finite group. A G-surface is a triple (S, G, ρ), where S is a nonsingular projective surface, and ρ is a monomorphism of the group G to the automorphisms group of the surface S. For brevity, G-surface will be denoted by (S, G). Let G be a finite subgroup in Cr2(C) with an embedding θ : G ֒→ Cr2(C). It turns out that the action of G on P2 can be regularized, i.e there exists a smooth rational surface S and a birational map µ : S 99K P2 such that µ−1 ◦ θ(G) ◦ µ is a subgroup of automorphisms of S. The research leading to these results has received partial funding from the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement 248826 and from the grant NSh-4731.2010.1. 1 Certainly, any regularization is not unique. For example, if we blow up any G-orbit of points on S. Two distinct G-surfaces (S, G) and (S ′, G) define two conjugate embeddings θ : G → Cr2(C) and θ′ : G → Cr2(C) respectively, iff there exist a G-equivariant birational map ζ : S 99K S ′. Definition 1.2. A G-surface (S, G) is called G-minimal, if any G-equivariant bi- rational morphism S → Y onto a smooth G-surface Y is a G-isomorphism. Theorem 1.3 (([12])). There are two types of the rational G-minimal surfaces (S, G): • S is a del Pezzo surface, and Pic(S)G ≃ Z; • S has a G-equivariant structure of conic bundle φ : S → P1, and Pic(S)G ≃ Z2. Notation 1.4. The classes of G-minimal rational surfaces from the first and the second cases of the Theorem 1.3 will be denoted respectively as D and CB. More recently, I.V. Dolgachev and V.A. Iskovskikh [4] improved the list of Kantor and Wiman. The answer was obtained in terms of action of the groups G on the del Pezzo surfaces and on the conic bundles. It was considered question about conjugacy of the finite subgroups in Cr2(C), using the theory of elementary links of V.A. Iskovskikh (see [10]). For general case this paper is currently the most precise classification of conjugacy classes of finite subgroups in Cr2(C). I note that J. Blanc in [3] obtained a more precise classification in case of finite cyclic subgroups. However in [4] explicit equations of G-minimal surfaces (S, G) in weighted pro- jective spaces and explicit descriptions of actions of G on surfaces S were obtained only in case of Del Pezzo surfaces. Also description of groups G, acting on G- minimal conic bundles (S, G, φ), was given only in terms of groups extensions. In other words, for a given abstract finite group it is still impossible, using [4], to say whether the group is isomorphic to a subgroup of Cr2(C). Also classification of conjugacy classes of finite subgroups in Cr2(C) has some gaps. If G ⊂ Cr2(C) is regularized as a subgroup of automorphisms of a conic bundle φ : S → P1 with S = 1, or 2, and (S, G, φ) ∈ CB. I will show it consistently for K 2 K 2 Let K 2 S = 1. Consider [4, Section 8.1, Pages 534-535]. There is stated non- existence of triples (S, G, φ) ∈ CB with K 2 S = 1 and ample divisor −KS. However, it's wrong. An example of such triples is presented in [13, Section 6.2.3, Theorem 6.8]. In this case the authors of [4] applied an old incorrect version of [4, Theorem 5.7]. This version existed until J. Blanc reported about a mistake to I. Dolgachev. I note that in published version of [4] Theorem 5.7 is presented in correct form. Un- fortunately, for large volume of work, the authors forgot to update some conclusions from the theorem. S = 1 and 2. Let K 2 S = 2. In [4, Section 8.1, Pages 535] is stated: if (S, G, φ) ∈ CB with K 2 S = 2 and non-ample divisor −KS then the surface S is exceptional conic bundle (see Definition 3.3). In other words the surface S contains two smooth non-intersecting rational (−3)-curves. This is also wrong. In [13, Section 5.1.1, Theorem 5.4] is presented an example of triple (S, G, φ) ∈ CB with K 2 S = 2 and nef, non-ample divisor −KS. In the paper [13] I continue classification of G-minimal conic bundles, which was begun in [4]. For given arbitrary value of K 2 S, it was constructed a method of classification by means of explicit equations of G-minimal conic bundles (S, G) in weighted projective spaces and explicit descriptions of the actions of G on the Picard 2 S > 0. If K 2 group Pic(S) and on the surface S. The classification is carried on completely for K 2 S ≤ 0 then the G-minimal conic bundle (S, G) is birationally rigid. So there is no question about conjugacy (see [4, Section 8]). The aim of this paper is to give a finer geometric description of the algebraic varieties parameterizing conjugacy classes of finite nonsolvable subgroups in Cr2(C), applying methods of papers [4] and [13]. It is obtained explicit equations of G- minimal surfaces (S, G) in weighted projective spaces and explicit descriptions of actions of G on surfaces S. Also all possibilities for the groups G are fully described. In [4, Section 9] were stated the following problems for Cr2(C): • Find the finer classification of the conjugacy classes of de Jonqui`eres groups. • Give a finer geometric description of the algebraic varieties parameterizing conjugacy classes. This article gives a solution of these problems for the nonsolvable finite subgroups in Cr2(C). It's important to note that investigation method described in the paper can be employed to solve these problems for all finite subgroups in Cr2(C), i.e. not necessary nonsolvable. However due to large amount of routine work investigation was conducted only for nonsolvable subgroups. The paper has the following structure. In Section 2 I study surfaces from the class D, i.e. the G-minimal del Pezzo surfaces (S, G), where Pic(S)G ≃ Z and G is a finite nonsolvable group. I will apply here results of [4]. There are no my results in this section. In Section 3 I study surfaces from the class CB, i.e. the G-minimal surfaces (S, G, φ), where a morphism φ : S → P1 defines a G-equivariant conic bundle structure, Pic(S)G ≃ Z2, and G is a finite nonsolvable group. The main my results are described in this section. In Section 4 I study conjugacy classes of embeddings G → Cr2(C) defined by G-minimal surfaces (S, G), for all finite nonsolvable subgroups G ⊂ Cr2(C). Here I reprove results in [4, Section 7] for the sake of completeness. In Section 5 I present a list of the finite nonsolvable subgroups in Cr2(C), ob- tained on the basis of results of sections 2 and 3. This work is dedicated to my supervisor Vasily Alekseevich Iskovskikh, who initiated my study of the Cremona group. I am very grateful to Yuri Gennadievich Prokhorov and Ilya Alexandrovich Tyomkin for useful tips and remarks. The base field is assumed everywhere to be C. Throughout this paper we will use the following notations. • εn denotes a primitive n-th root of unity. • Sn denotes the permutation group of degree n. • An denotes the alternating group of degree n. • Consider a subgroup A5 ⊂ P GL(2, C), which is isomorphic to the icosa- hedral automorphisms group, and the standard projection ψ : SL(2, C) → P GL(2, C). Then ¯A5 denotes the group ψ−1(A5), which is isomorphic to the binary icosahedral group. • A.B, where A and B are some abstract groups, is one of the possible ex- tensions with help of the exact sequence: 0 → A → G → B → 0. • Let H be an abstract group. Then H≀Sn will denote the semidirect product H n ⋊ Sn, where Sn is the symmetric group, acting on H n by permuting the factors. 3 • A△DB is a diagonal product of abstract groups A and B over their common homomorphic image D (i.e. the subgroup of A× B of pairs (a, b), such that α(a) = β(b) for some epimorphisms α : A → D, β : B → D). • P(a1, . . . , an), where ai ∈ Z, i = 1, . . . , n, is the weighted projective space, with the set of weights (a1, . . . , an). 2. Case of del Pezzo surfaces. In this section we study the surfaces (S, G) ∈ D, i.e. S is a G-minimal del Pezzo surface, and Pic(S)G ≃ Z. The groups G are supposed to be finite nonsolvable. We will apply here results of [4]. There are no author's results in this section. Recall that a surface S is called a del Pezzo surface, if S is smooth, and −KS S ≤ 9. We will carry our investigation, It's well known that 1 ≤ K 2 is ample. considering different values of K 2 S. In the next theorem we study the case K 2 S = 9. In this case S ≃ P2. Theorem 2.1. Let (S, G) ∈ D, K 2 S = 9, and G be a finite nonsolvable group. Then S ≃ P2 with the coordinates (x0 : x1 : x2), and G is any finite nonsolvable subgroup of Aut(P3) ≃ P GL(3, C). The subgroup G ⊂ P GL(3, C) can be conjugated to one of the following subgroups. (1) H is a group, consisting of maps (x0 : x1 : x2) 7→ (ax0 + bx1 : cx0 + dx1 : x2). The image of matrices (cid:18)a b d(cid:19) ∈ GL(2, C) c in P GL(2, C) under the natural projection GL(2, C) → P GL(2, C) is iso- morphic to A5. The group H is isomorphic to Zn × ¯A5, n ≥ 1. (2) The icosahedral group A5 isomorphic to L2(5). It leaves invariant a non- singular conic C ⊂ P2. phism group of the Klein's quartic x3 (3) The Klein group isomorphic to L2(7). This group is realized as automor- 0x1 + x3 1x2 + x3 2x0 = 0. (4) The Valentiner group isomorphic to A6. It can be realized as automorphism group of the nonsingular plane sextic 1x2 1 + 9x2x5 0 + x6 10x3 0x2 0x3 1 − 45x2 2 − 135x0x1x4 2 + 27x6 2 = 0. Proof. The statement follows directly from [4, Corollary 4.6, Theorems 4.7, 4.8]. We need only to check the isomorphism H ≃ Zn × ¯A5, n ≥ 1 in the first case of theorem. It follows from [4, Lemma 4.5, case (i)]. (cid:3) In the next theorem we consider the case K 2 S = 8. Theorem 2.2. Let (S, G) ∈ D, K 2 S = 8, and G be a finite nonsolvable group. Then S ≃ F0 ≃ P1 × P1 with the coordinates (x0 : x1, t0 : t1). We will employ definition of the group St(A5) (see Notation 3.5), and define involution τ : (x0 : x1, t0 : t1) 7→ (t0 : t1, x0 : x1). We have the following possibilities for G. (1) The subgroup G ⊂ Aut(P1 × P1) is conjugate to the subgroup St(A5) ≀ hτi. 4 (2) The subgroup G ⊂ Aut(P1 × P1) is conjugate to the subgroup H × hτi, where H is the image of the diagonal embedding of St(A5) in P GL(2, C)× P GL(2, C). Proof. One knows that if S is a del Pezzo surface with K 2 S = 8 then S ≃ F0 or F1. However in the second case the exceptional section of ruled surface F1 is G-invariant. Therefore the pair (F1, G) is not G-minimal. Hence S ≃ F0 ≃ P1× P1. It's well known that Aut(P1 × P1) ≃ P GL(2, C) ≀ hτi. Whence G is generated by a nonsolvable subgroup H ⊂ P GL(2, C) × P GL(2, C) and by an element η = µ ◦ τ , where µ ∈ P GL(2, C) × P GL(2, C). We write µ = (B, B′), where B, B′ ∈ P GL(2, C). For any ς = (A, A′) ∈ P GL(2, C) × P GL(2, C) we have (2.1) η ◦ ς ◦ η−1 = (BA′B−1, B′AB′−1). Let's study the structure of group H, applying Goursat's Lemma (see [4, Lemma 4.1]). Consider projections πi : P GL(2, C) × P GL(2, C) → P GL(2, C), i = 1, 2 on the first and the second factor respectively. We get H ≃ π1(H)△Dπ2(H), where D ≃ Im(π1H)/ Ker(π2H). Obviously, either Im(π1H) ≃ A5 or Im(π2H) ≃ A5 (see Klein's classification of finite nonsolvable subgroups in P GL(2, C) in [4, Section 5.5]). The group A5 is simple. Therefore D ≃ 1 or A5. Suppose that D ≃ 1. From (2.1) we get Im(π1H) ≃ Im(π2H) ≃ A5. Hence H can be conjugated to St(A5)×St(A5). We will prove that B, B′ ∈ St(A5). Suppose that it doesn't holds. Then from (2.1) we get St(A5)⋊B, St(A5)⋊B′ ⊂ P GL(2, C). But St(A5) is a maximal finite subgroup of P GL(2, C). Contradiction. We get the first case of the theorem. Suppose that D ≃ A5. Then H is conjugated to the image of diagonal embedding of St(A5) in P GL(2, C) × P GL(2, C). Arguing as above, we get B, B′ ∈ St(A5). By (2.1) these elements define the same inner automorphism of A5. Hence B = B′. We get the second case of the theorem. (cid:3) S = 7, K 2 In the next theorem we consider cases: K 2 S = 6, K 2 Theorem 2.3. There are no surfaces (S, G) ∈ D, such that K 2 7, or 6, or 4 , or 1, and G is a finite nonsolvable group. Proof. Let's consider the case K 2 S = 7. The surface S is presented as a blowing up of two different points in P2. However the strict transform of line, containing this two points, is a G-invariant rational (−1)-curve. Hence the surface S is not G-minimal. S = 6 follows directly from [4, Theorem 6.3, Corollary 4.6, Theorem S = 4, and K 2 S = 1. S is equal to either The case K 2 4.7]. The case K 2 The case K 2 S = 4 follows directly from [4, Theorem 6.9]. S = 1 follows directly from [4, Table 8]. In the next theorem we consider the case K 2 S = 5. (cid:3) Theorem 2.4. Let (S, G) ∈ D, K 2 S = 5, and G be a finite nonsolvable group. Introduce on P2 the coordinates (T0 : T1 : T2). Then the surface S is isomorphic to the blowing up of P2 at points: (0 : 0 : 1), (0 : 1 : 0), (1 : 0 : 0) and (1 : 1 : 1). The group Aut(S) is isomorphic to S5, and is generated by the maps: (2.2) (T0 : T1 : T2) 7→ (T1 : T2 : T0), (T0 : T1 : T2) 7→ (T2 : −T0 + T2 : −T1 + T2), (T0 : T1 : T2) 7→ (T0(T2 − T1) : T2(T0 − T1) : T0T2). 5 The subgroup G ⊂ Aut(S) is isomorphic to A5 or S5. Proof. This follows directly from [4, Theorem 6.4] and arguments of [5, Theorem 8.4.15]. (cid:3) In the next theorem we consider the case K 2 S = 3. Theorem 2.5. Let (S, G) ∈ D, K 2 S = 3, and G be a finite nonsolvable group. Then the surface S can be represented by the following equations in P3 with the coordinates (T0 : T1 : T2 : T3): T 2 0 T1 + T 2 3 T0 = 0. 1 T2 + T 2 2 T3 + T 2 The group G is isomorphic to S5 and is generated by the following maps: (2.3) (T0 : T1 : T2 : T3) 7→ (T0 : ε4 (T0 : T1 : T2 : T3) 7→ (T1 : T2 : T3 : T0). 5T1 : ε5T2 : ε2 5T3), Proof. This follows directly from [4, Theorem 6.14]. (cid:3) In the next theorem we consider the case K 2 S = 2. Theorem 2.6. Let (S, G) ∈ D, K 2 S = 2, and G be a finite nonsolvable group. Then the surface S can be represented by the following equation in P(2, 1, 1, 1) with the coordinates (T0 : T1 : T2 : T3): T 2 3 + T 3 (2.4) The group Aut(S) is isomorphic to Z2 × L2(7). The subgroup G ⊂ Aut(S) is isomorphic to either L2(7), or Z2 × L2(7). Proof. This follows directly from [4, Theorem 6.17]. 1 T2 + T 3 0 T1 + T 3 2 T0 = 0. (cid:3) 3. Case of conic bundles In this section we study the surfaces (S, G, φ) in the class CB, i.e. G-minimal surfaces S with Pic(S)G ≃ Z2, having a G-equivariant conic bundle structure φ : S → P1. The groups G are supposed to be finite nonsolvable. Recall (see [4, Item 3.7]) that a rational G-surface (S, G) has a conic bundle structure, if there exist a G-equivariant morphism φ : S → P1, whose each fiber Ft = φ−1(t), t ∈ P1 is either a nondegenerate plane conic (isomorphic to P1) or a reducible reduced conic, i.e. a pair of intersecting lines. A conic bundle (S, G, φ) is said to be relatively G-minimal, if the fibres of φ do not contain G-orbits, consisting of nonintersecting rational (−1)-curves (i.e. components of reducible fibres -- equivalently to Pic(S)G = φ∗ Pic(P1) ⊕ Z ≃ Z2). Recall that a G-surface (S, G) is said to be G-minimal, if any G-equivariant birational morphism S → Y onto a smooth G-surface Y is a G-isomorphism. It is clear that a G-minimal surface, having a conic bundle structure, is relatively minimal. The inverse statement is not always valid. Denote by r the number of the reducible fibers of a conic bundle (S, G, φ). By S = 8, then S is isomorphic Noether formula we have r = 8 − K 2 to Hirzebruch's surface Fn, n ≥ 0. Theorem 3.1. Let (S, G, φ) ∈ CB with K 2 S = 8, and G be a finite nonsolvable group. Then the surface S is isomorphic to Hirzebruch's surface Fn, n ≥ 0. The morphism φ : S → P1 coincides with the standard projection Fn → P1. S ≤ 8. If K 2 S, so K 2 6 (1) Let n = 0. Then F0 ≃ P1 × P1. The group Aut(F0) is isomorphic to P GL(2, C) ≀ S2. The subgroup G ⊂ P GL(2, C) × P GL(2, C) ⊂ Aut(F0) is isomorphic to one of the following: A5 × B, B × A5, A5△A5A5, where B is any finite subgroup of P GL(2, C). (2) Let n > 0. Then n > 1. Consider Fn → P(n, 1, 1) the blowdown of excep- tional section of Fn. We have Aut(Fn) ≃ Cn+1 ⋊ (GL(2, C)/µn), where GL(2, C)/µn acts on Cn+1 by means of its natural linear representa- tion in the space of binary forms with degree n. The subgroup G ⊂ Aut(Fn) is isomorphic to one of the following groups: (3.1) G ≃( Zm × A5, m ≥ 1, Zm × ¯A5, m ≥ 1, if n is even; if n is odd. Proof. If S ≃ F0 ≃ P1 × P1, then G is a nonsolvable subgroup in Aut(F0) ≃ P GL(2, C) ≀ S2. Note that G ⊂ P GL(2, C) × P GL(2, C), so as Pic(S)G ≃ Z2. We apply Goursat's lemma (see [4, Lemma 4.1]) and Klein's classification of the finite subgroups in P GL(2, C) (see [4, Section 5.5]). We get that G ≃ B△DC, where one of groups B and C is isomorphic to A5. Since the group A5 is simple, the group D is isomorphic to either 1 or A5. Therefore the group G is isomorphic to one of the following groups: A5 × B, B × A5, A5△A5A5, where B is any finite subgroup of P GL(2, C). Remark that a group A5△A5A5 is conjugate to image of a diagonal embedding of group A5 to P GL(2, C) × P GL(2, C). Consider the case S ≃ Fn, n > 0. Let Fn → P(n, 1, 1) be the blowdown of the exceptional section of Fn. We note that if n = 1 then P(1, 1, 1) is a smooth surface. Hence the triple Introduce the coordinates (x : t0 : t1) on P(n, 1, 1). (S, G, φ) is not minimal. Therefore n 6= 1. It's well known (see [4, Theorem 4.10]) that Aut(Fn) ≃ Cn+1 ⋊ (GL(2, C)/µn). The group Cn+1 is gen- erated by maps (x : t0 : t1) 7→ (x + fn(t0, t1) : t0 : t1), where fn is a binary form with degree n. The group GL(2, C)/µn is generated by invertible maps (x : t0 : t1) 7→ (x : at0 + bt1 : ct0 + dt1). Moreover, we have GL(2, C)/µn ≃( C∗ ⋊ SL(2, C), C∗ ⋊ P GL(2, C), if n is odd; if n is even. h2−→ P GL(2, C), Consider the sequence of homomorphisms G where h2 is natural projection. Obviously that the homomorphism h1 is injective, and Im(h2) ≃ A5 (see [4, Section 5.5]). Zm. ¯A5, if n is odd. Applying [4, Lemma 4.4], we get (3.1). Thus the group G is isomorphic to a central extension Zm.A5, if n is even, or (cid:3) h1−→ GL(2, C)/µn There are no relatively G-minimal conic bundles (S, G, φ), if K 2 S = 7. So there are no G-minimal conic bundles too. S = 3, 5 or 6, then there exists a G- equivariant morphism (S, G) → (S ′, G), where (S ′, G) ∈ D (see [8, Proposition 2.1, Theorem 4.1] and, for example, [13, Section 2]). Thus study of this cases reduces to study of G-minimal del Pezzo surfaces. For other values K 2 S = 4, 2, 1, . . . relatively G-minimal conic bundles are always G-minimal. If K 2 7 The morphism φ : S → P1 induces a homomorphism φ∗ : G → Aut(P1). We have the following exact sequence 1 → GK → G → GB → 1, (3.2) where GK ≃ Ker(φ∗), and GB ≃ Im(φ∗). Also consider the natural representation of group G in the automorphisms group of lattice Pic(S). By G0 we denote the kernel of this representation. The group G0 fixes the divisor classes of the sections with negative self-intersection. Such sections obviously exist. Hence G0 fixes it pointwisely. Considering one of these sections as a point on a general fibre, we conclude that G0 is a cyclic group. Theorem 3.2 ([4, Proposition 5.5]). Let (S, G, φ) ∈ CB with K 2 S 6= 3. Suppose that G0 6= {1}. Then the surface S has an exceptional conic bundle structure (see below). S ≤ 4, K 2 Definition 3.3. Define the exceptional conic bundles. This is the minimal resolu- tion of singularities of surface, given by the equation in weighted projective space P(1, 1, g + 1, g + 1), where g ≥ 1: Yg : F2g+2(t0, t1) + t2t3 = 0, where F2g+2 is a binary form without multiple factors with degree 2g + 2. After the resolution of indeterminacy points, the projection onto P1 with coor- dinates (t0, t1) will induce a conic bundle structure φ :fYg → P1. This conic bundle has reducible fibres over the points from P1, where F2g+2(t0, t1) = 0. The surfacefYg and t3 = 0. Automorphisms of the surface fYg are induced by automorphisms of contains 2 nonintersecting rational (−g − 1)-curves defined by the equations t2 = 0 P(1, 1, g + 1, g + 1). The case of exceptional conic bundles will be considered in Section 3.1. There is a theorem about the structure of minimal finite groups, acting on the non-exceptional conic bundles. Theorem 3.4 ([4, Theorem 5.7]). Let (S, G, φ) ∈ CB with K 2 S 6= 3, and Σ be the set of reducible fibres of φ. Suppose that G0 ≃ 1. Then one of the following cases occurs. S ≤ 4, K 2 (1) Case GK ≃ Z2. The central involution ι, generating the group GK, fixes pointwise a smooth bisection C of the fibration φ and switches the com- ponents of fibres in a subset Σ′ ⊂ Σ. The morphism φ defines the linear system g1 2 on the curve C having branch points in the singular points of the fibres in the set Σ′. Genus of the curve C is equal to g = (m − 2)/2, where m = Σ′. The group GB is isomorphic to a subgroup of the automorphism group of curve C modulo the involution defined by g1 2. 2. Each nontrivial element ιi, i = 0, 1, 2 of the group GK fixes pointwise an irreducible smooth bisection Ci. The set Σ is partitioned into three subsets Σ0, Σ1, Σ2, such that Σi = (Σj ∪ Σk) \ (Σj ∩ Σk), i 6= j 6= k 6= i. The morphisms φCi , i = 0, 1, 2 are branched over the singular points of fibres in subsets Σi. The group GB leaves invariant the set of points φ(Σ) ∈ P1 and its partition into three subsets φ(Σi), i = 0, 1, 2. (2) Case GK ≃ Z2 Consider cases of Theorem 3.4 separately. We note that the subgroup GB ⊂ Aut(P1) ≃ P GL(2, C) (see (3.2)) is nonsolvable, since GK is solvable by Theorem 8 3.4. Hence GB ≃ A5 (see Klein's classification of the finite subgroups in P GL(2, C) in [4, Section 5.5]). We will use the fact in sections 3.2, 3.3, 3.4 without mentioning. (1) Case GK ≃ Z2, and Σ′ = Σ. This case will be investigated in Section 3.2. (2) Case GK ≃ Z2, and Σ′ 6= Σ. This case will be investigated in Section 3.3. (3) Case GK ≃ Z2 We will often use the following facts about finite nonsolvable subgroups ¯P ⊂ SL(2, C) (see [4, Section 5.5]). Obviously that ¯P ≃ ¯A5. Any group of this type is conjugated to a group with the following generators: 5(cid:19) . 5 − ε3 ε2 5 −ε5 + ε4 2. This case will be investigated in Section 3.4. √5(cid:18)ε5 − ε4 ε2 5 − ε3 g1 =(cid:18)ε10 g2 =(cid:18)0 0(cid:19) , 0 ε−1 10(cid:19) , g3 = (3.3) 0 1 5 5 i i Notation 3.5. We will denote a group generated by (3.3) as St( ¯A5). It's image in P GL(2, C) we will denote as St(A5). Consider the natural representation of St( ¯A5) in space of polynomials C[t0, t1]. Space of relative invariants of the representation is generated by the following Grundformens: (3.4) 0 + T 30 Φ1 = T 30 Φ2 = −(T 20 Φ3 = T0T1(T 10 1 + 522(T 25 0 T 5 1 − T 5 0 T 25 1 ) + 228(T 15 0 T 5 1 − T 5 1 − T 10 1 ). 0 T 5 0 + 11T 5 0 + T 20 1 ) − 10005(T 20 0 T 15 1 ) − 494T 10 0 T 10 1 + T 10 0 T 10 1 , 0 T 20 1 ), Since ¯A5/(±1) ∼= A5 is a simple group and all Grundformens have even degree, we easily see that g(Φi) = Φi, i = 1, 2, 3, for any g ∈ St( ¯A5). In other words, the characters are trivial. Notation 3.6. We will denote space of invariants of group ¯A5 generated by this Grundformens as I St( ¯A5). 3.1. Case of exceptional conic bundles. In this section we will prove the fol- lowing theorem. Theorem 3.7. Let (S, G, φ) ∈ CB be an exceptional conic bundle, and G be a finite nonsolvable group. Then the surface S can be represented as the minimal resolution of singularities of surface given by the equation in the weighted projective space P(1, 1, g + 1, g + 1), g ≥ 1 with coordinates (t0 : t1 : t2 : t3): Yg : F2g+2(t0, t1) + t2t3 = 0, where F2g+2 ∈ I St( ¯A5) is a binary form without multiple factors with degree 2g + 2. The morphism φ : S → P1 is induced by the map φ′ : Yg 99K P1 given by φ′ : (t0 : t1 : t2 : t3) 7→ (t0 : t1). The group G is isomorphic to G ≃( Dn × ¯A5, n ≥ 1, if g is even; Dn × A5, n ≥ 1, if g is odd. 9 10 t1 : t2 : t3), All possibilities for G occur. The group G is generated by the maps: (3.5) (t0 : t1 : t2 : t3) 7→ (ε10t0 : ε−1 (t0 : t1 : t2 : t3) 7→ (it1 : it0 : t2 : t3), (t0 : t1 : t2 : t3) 7→ ((ε5 − ε4 5)t0 + (ε2 (t0 : t1 : t2 : t3) 7→ (t0 : t1 : εmt2 : ε−1 (t0 : t1 : t2 : t3) 7→ (t0 : t1 : t3 : t2), where m = n, if g is odd, and m = 2n, otherwise. Proof. By [4, Proposition 5.3] we have Aut Yg ≃ N.P , where N ≃ C∗ ⋊ Z2 is a group generated by the maps: 5)t0 + (−ε5 + ε4 5 − ε3 m t3), 5)t1 : t2 : t3), 5)t1 : (ε2 5 − ε3 (3.6) (t0 : t1 : t2 : t3) 7→ (t0 : t1 : t3 : t2), (t0 : t1 : t2 : t3) 7→ (t0 : t1 : ct2 : c−1t3), c ∈ C, c 6= 0. And P is the subgroup of P GL(2, C), leaving the form F2g+2(t0, t1) semi-invariant. Obviously that P ≃ A5. We conjugate the subgroup P ⊂ P GL(2, C) to the sub- group St(A5). Then F2g+2 ∈ I St( ¯A5). Whence we get that the group Aut(Yg) is generated by maps (3.5) and (3.6). Hence Aut(Yg) ≃( (N/µ2) × ¯A5, if g is even; N × A5, if g is odd, where the group µ2 acts by (t0 : t1 : t2 : t3) 7→ (t0 : t1 : −t2 : −t3). It follows from the description of exceptional conic bundles (see [4, Section 5.2]) that the triple (S, G, φ) is minimal, iff the group G permutes points: (0 : 0 : 1 : 0) and (0 : 0 : 0 : 1). Therefore G ∩ N ≃ Dn. Further arguments are obvious. (cid:3) 3.2. Case, when G0 ≃ 1, GK ≃ Z2, and Σ′ = Σ. Here we will apply arguments of [13, Section 3.1]. The group GK is generated by involution ι. By [13, Theorem 3.2] we get S/ι ≃ Fe. The morphism π : S → S/ι ≃ Fe is G-equivariant, and a faithful action of the group GB is defined on Fe(see exact sequence (3.2)). Recall (see Theorem 3.4) that the morphism π : S → Fe is branched over a nonsingular hyperelliptic curve C. Let ¯C = π(C). We consider cases e = 0 and e > 0 in Theorems 3.8 and 3.10 respectively. We make some preparations before statement of Theorem 3.8. Introduce the coordinates (x0 : x1, t0 : t1) on F0 ≃ P1 × P1. The morphism φ : S → P1 induces projection (x0 : x1, t0 : t1) 7→ (t0 : t1). The curve ¯C is represented by the equation: (3.7) 0 + 2p1(t0, t1)x0x1 + p2(t0, t1)x2 Equat( ¯C) = p0(t0, t1)x2 1 = 0, where pi, i = 0, 1, 2 are binary forms with degree 2d. Note that the degree is even, so as the divisor class ¯C ∈ 2 Pic(F0) (where 2 Pic(F0) ⊂ Pic(F0) ≃ Z2 is the even sublattice). The form Disc( ¯C) = p0p2−p2 1 has no multiple factors, since ¯C is nonsin- gular. We will apply the Segre embedding ν : P1× P1 → P3 to represent the surface S by equations. Introduce the coordinates (x : y : z : w) on P3. This embedding is given by ν : (x0 : x1, t0 : t1) 7→ (x0t0 : x0t1 : x1t0 : x1t1). We choose some poly- Equat( ¯C) = ν ∗(Fi). nomials Fi(x, y, z, w), i = 0, . . . , 2d − 2, such that xi 0x2d−2−i 1 10 The surface S is represented by the equations in P(dd, 14) with the coordinates ui, x, y, z, w, i = 0, . . . , d − 1: (3.8) uiuj = Fi+j , 0 ≤ i ≤ j ≤ d − 1, xj−iui = ujzj−i, yj−iui = ujwj−i, 0 ≤ i < j ≤ d − 1, xw = yz. Theorem 3.8. Let (S, G, φ) ∈ CB , and G be a finite nonsolvable group. Suppose that G0 ≃ 1, GK ≃ hιi ≃ Z2, Σ′ = Σ, and S/ι ≃ F0. Then the surface S is represented by equations (3.8). The morphism φ : S → P1 is given by φ : (u0 : . . . : ud−1 : x : y : z : w) 7→( (x : y), if (x : y) 6= (0 : 0); (z : w), if (z : w) 6= (0 : 0). There is defined a faithful action of GB (see exact sequence (3.2)) on F0, and GB ⊂ P GL(2, C) × P GL(2, C). One of the following cases occurs. (1) The subgroup GB ⊂ P GL(2, C) × P GL(2, C) is conjugated to the subgroup (2) The subgroup GB ⊂ P GL(2, C) × P GL(2, C) is conjugated to the diagonal 1 × St(A5), and G ≃ Z2 × A5. embedding St(A5) ֒→ P GL(2, C) × P GL(2, C), and G ≃( ¯A5, if d is even in (3.7), Z2 × A5, otherwise. All cases exist. And all possibilities for G occur. In all cases G acts on S by the following way. Embedding GB ≃ St(A5) to P GL(2, C)×P GL(2, C) defines a unique embedding St( ¯A5) ֒→ SL(2, C)×SL(2, C). This defines an action of St( ¯A5) on the surface S given by equation (3.8). The action of St( ¯A5) on coordinates ui, i = 0, . . . , d − 1 coincides with the action on monomials xi , i = 0, . . . , d − 1, respectively. An action of G is generated by the action of St( ¯A5) and by the map 0xd−1−i 1 (u0 : . . . : ud−1 : x : y : z : w) → (−u0 : . . . : −ud−1 : x : y : z : w). Proof. Recall that GB ≃ A5. The subgroup GB ⊂ Aut(F0) acts trivially on Pic(F0). Hence GB ⊂ P GL(2, C) × P GL(2, C). We apply Goursat's Lemma (see [4, Lemma 4.1]) to study the subgroups A5 ⊂ P GL(2, C) × P GL(2, C). Con- sider projections πi : P GL(2, C) × P GL(2, C) → P GL(2, C), i = 1, 2 on the first and the second factor respectively. We get GB ≃ π1(GB)△Dπ2(GB), where D ≃ Im(π1GB)/ Ker(π2GB). The group A5 is simple. Therefore D ≃ 1 or A5. These cases corresponds respectively to cases 1 and 2 of the theorem. It's need to check existence of the nonsingular curve ¯C ⊂ F0 for each of these cases. In the first case this curve obviously exists. Because we can choose binary forms pi ∈ I St( ¯A5), i = 0, 1, 2 in (3.7) without multiple and common factors (see the generators of I St( ¯A5) in (3.4)). It remains to verify existence of the nonsingular curve ¯C in the second case. Also we need to show that the parameter d in (3.7) can be odd and even. This follows from the next lemma. Lemma 3.9. There exist nonsingular curves ¯C ∈ F0 with odd and even parameter d given by equation (3.7) and invariant under the diagonal action of group St(A5) on F0 ≃ P1 × P1. 11 and RSt( ¯A5) 2d−2 2d 0 + 2p1t0t1 + p2t2 ⊕ RSt( ¯A5) 2d−2 . 2d+2 ⊕ RSt( ¯A5) 2d 2RSt( ¯A5) 60(k−1) ⊂ RSt( ¯A5) 60k−20 (see (3.4)). The space RSt( ¯A5) Proof. Consider the linear space of polynomials C[x, y]. The space has the natural structure of SL(2, C)-module. Denote by Rn ⊂ C[x, y] the subspace of polynomials with degree n. We have Equat( ¯C) ∈ R2 ⊗ R2d. It's known (see [6, Exercise 11.11]) that R2 ⊗ R2d ≃ R2d+2 ⊕ R2d ⊕ R2d−2 as SL(2, C)-module. Consider a linear system J of St( ¯A5)-invariant curves with bidegree (2, 2d) in F0. Obviously, we have J ≃ (R2 ⊗ R2d)St( ¯A5) ≃ RSt( ¯A5) First, we prove existence of a nonsingular curve ¯C with odd parameter d. We It's easy to check (see (3.4)) that each set take d = 30k + 15, k ∈ N, k ≥ 2. RSt( ¯A5) 2d+2 , RSt( ¯A5) is not empty. To apply Bertini theorem, we need to study the base points of system J . We have (x0t1 − x1t0)2R2d−2(t0, t1)St( ¯A5) ∈ J . It's easy to check that RSt( ¯A5) 2d−2 = 3RSt( ¯A5) Φ4 60k−20, and Φ2 60(k−1) has not a common factor, since k ≥ 2. Hence the base points of J lie in the union of sets: x0t1 − x1t0 = 0, Φ2(t0, t1) = 0 and Φ3(t0, t1) = 0. Consider the projection ξ : R2 ⊗ R2d ≃ R2d+2 ⊕ R2d ⊕ R2d−2 → R2d+2. It is given by the polynomial p0t2 1. This polynomial defines intersection of the curve ¯C and of diagonal x0t1 − x1t0 = 0. We have Φ2Φ3RSt( ¯A5) ⊂ RSt( ¯A5) 2d+2 . The space RSt( ¯A5) has not a common factor, since k ≥ 2. Hence we can take a polynomial Equat( ¯C) ∈ R2 ⊗ R2d, such that ξ(Equat( ¯C)) = Φ2Φ3h(t0, t1), where the forms Φ2, Φ3, h(t0, t1) have not pairwise common factors. Therefore the base points of J lie in the union of sets: Φ2(t0, t1) = 0 and Φ3(t0, t1) = 0. However by choose of Equat( ¯C) we get that the curve ¯C is nonsingular at these sets. It remains to prove existence of a nonsingular curve ¯C with even parameter d. It is easy to check (see (3.4)) that each set is not empty. To apply Bertini theorem, we need to We take d = 30k, k ∈ N, k ≥ 3. RSt( ¯A5) 2d+2 , RSt( ¯A5) study the base points of system J . We have Φ1J ′ ⊂ J , where J ′ is a linear system of St( ¯A5)-invariant curves with bidegree (2, 2d − 30) in F0. By previous arguments, we know that the base points of J ′ lie in the union of sets: Φ2(t0, t1) = 0 and Φ3(t0, t1) = 0. Hence the base points of J lie in the union of sets: Φ1(t0, t1) = 0, Φ2(t0, t1) = 0 and Φ3(t0, t1) = 0. Again consider projection ξ. We have Φ1Φ2Φ3RSt( ¯A5) 2d+2 . The space RSt( ¯A5) 60(k−1) has not a common factor, since k ≥ 3. Hence we can take a polynomial Equat( ¯C) ∈ R2 ⊗ R2d, such that ξ(Equat( ¯C)) = Φ1Φ2Φ3h(t0, t1), where the forms Φ1, Φ2, Φ3, h(t0, t1) have not pairwise common factors. Again the conditions of Bertini theorem are satisfied. (cid:3) 60(k−1) ⊂ RSt( ¯A5) and RSt( ¯A5) 2d−2 60k 60k 2d It remains to describe the action of group G. The embedding of the group GB ≃ St(A5) to P GL(2, C) × P GL(2, C) defines a unique embedding St( ¯A5) ֒→ SL(2, C)×SL(2, C). We note that Equat( ¯C) is invariant under the action of St( ¯A5). Hence there is defined an action of St( ¯A5) on the surface S given by equation (3.8). The action of St( ¯A5) on coordinates ui, i = 0, . . . , d − 1 coincides with the action on monomials xi , i = 0, . . . , d − 1, respectively. The remaining arguments are obvious. 0xd−1−i (cid:3) 1 The case e > 0 will be considered in the next theorem. 12 Theorem 3.10. Let (S, G, φ) ∈ CB , and G be a finite nonsolvable group. Suppose that G0 ≃ 1, GK ≃ hιi ≃ Z2, Σ′ = Σ, and S/ι ≃ Fn, n > 0. Then there is a G-invariant curve E, which is the preimage of exceptional section Fe. Consider the contraction of this curve (S, G, φ) → (S ′, G, φ′), where a map φ′ : S ′ 99K P1 is defined by φ.The surface S ′ is given by the following equation in P(d + e, e, 1, 1) with the coordinates (u : x : t0 : t1): (3.9) u2 + p0(t0, t1)x2 + 2p1(t0, t1)x + p2(t0, t1) = 0, where pi, i = 0, 1, 2 are binary forms with degree 2d, 2d + e, 2d + 2e, respectively. The binary form p0p2 − p2 1 has no multiple factors. The map φ′ is given by φ′ : (u : x : t0 : t1) 7→ (t0 : t1). Moreover, e is even. The group G is generated by the maps: u 7→ −u, (u : x : t0 : t1) 7→ (u : x + Fe(t0, t1) : at0 + bt1 : ct0 + d′t1), where (3.10) (cid:18)a c b d′(cid:19) ∈ St( ¯A5), and Fe(t0, t1) is a some binary form with degree e, unique for each matrix (3.10). The group G is isomorphic to (3.11) G ≃( ¯A5, if d is odd, Z2 × A5, if d is even. All possibilities for G occur. Proof. We will use the following construction to represent the surface S by equa- tions. Consider the morphism Fe → P(e, 1, 1), which is the blowing down of excep- tional section Fe. Introduce the coordinates (x : t0 : t1) on P(e, 1, 1). The morphism φ induces projection (x : t0 : t1) 7→ (t0 : t1). The curve ¯C will be represented by the following equation in P(e, 1, 1): p0(t0, t1)x2 + 2p1(t0, t1)x + p2(t0, t1) = 0. 1 has no multiple factors, since ¯C is nonsingular. We The form Disc( ¯C) = p0p2 − p2 construct a double cover of P(e, 1, 1), branched along ¯C. We get the surface S ′ given by the equations (3.9). Note that deg(p0) is even, since ¯C ∈ 2 Pic(Fe). Denote the degree as 2d. The automorphism group of P(e, 1, 1) consists of the maps (x : t0 : t1) 7→ (a′x + Pe(t0, t1) : b′t0 + c′t1 : d′t0 + v′t1). where Pe is a binary form with degree e. We can choose coefficients b′, c′, d′, v′, so that (cid:18)b′ d′ c′ v′(cid:19) ∈ SL(2, C). We have GB ≃ A5 6⊂ ¯A5 (see (3.2)). Therefore e is even. We conjugate GB to a group consisting of the following maps (x : t0 : t1) 7→ (vx + Fe(t0, t1) : at0 + bt1 : ct0 + d′t1), 13 where the coefficients a, b, c, d′ and the binary form Fe satisfy conditions of the theorem. But v = 1, since A5 is a simple group, and all it's characters A5 → C∗ are trivial. Obviously, we get (3.11). Finally, we need to prove that all possibilities for G in (3.11) occur. It's sufficient to construct nonsingular curves ¯C invariant under an action of GB with odd and even parameter d. We assume that GB is a group consisting of maps (x : t0 : t1) 7→ (x : at0 + bt1 : ct0 + d′t1), with condition (3.10). Let d is even. Then the curve ¯C is represented by the following equation in P(4, 1, 1): Φ3(t0, t1)x2 + Φ2(t0, t1) = 0, where Φi, i = 2, 3 are binary forms in (3.4). Let's construct the curve ¯C with odd d. Consider the equation of ¯C in P(30, 1, 1): Φ1(t0, t1)x2 + 2h(t0, t1)x + h′(t0, t1) = 0, where Φ1 is a binary form in (3.4), and h, h′ are some binary forms in I St( ¯A5). It's easy to check by counting of parameters that h and h′ can be chosen, such that Disc( ¯C) = Φ1h′ − h2 has no multiple factors. Then ¯C is nonsingular. (cid:3) 3.3. Case, when G0 ≃ 1, GK ≃ Z2, and Σ′ 6= Σ. Here we will apply arguments of [13, Section 3.2]. Let r = Σ, and m = Σ′. Let g1 : eS → S be blowing up of the singular points of reducible fibres Σ \ Σ′, and g2 : eS → S ′ be the contraction of proper transform of Σ \ Σ′. The surface S ′ has 2(r − m) singular points of type A1. Obviously that maps g1 and g2 are G-equivariant. We have the G-equivariant commutative diagram. (3.12) S h eS g2 "❊❊❊❊❊❊❊❊❊ g1 }④④④④④④④④④ eh S ′ g′ eS/ι ~⑤⑤⑤⑤⑤⑤⑤⑤ 1 h′ 2 g′ !❈❈❈❈❈❈❈❈ S ′/ι S/ι In the diagram the vertical arrows correspond to the quotient map by the invo- lution ι, and maps g′ 2 are induced by maps g1 and g2. The triple (S, G, φ) defines a triple (S ′, G, φ′), where the morphism φ′ : S ′ → P1 is induced by the morphism φ. 1 and g′ By [13, Lemma 3.4] we get that surfaces eS/ι and S ′/ι are nonsingular. Moreover, S ′/ι ≃ Fe. C ′ ∪ g2∗(g∗ image of the curve g2∗(g∗ fibres. Denote these fibres as Si, i = 1, . . . , r − m. Also let C = h′(C ′). The morphism h′ : S ′ → Fe is a double cover branched over the union of curves 1(Σ \ Σ′)), where the curve C ′ is the proper transform of curve C. The 1(Σ \ Σ′)) on the ruled surface Fe is the union of r − m 14 }   "     ~ ! For each fiber Si, i = 1, . . . , r − m denote by xi1 and xi2 two distinct points of the intersection Si∩ C. Obviously, there is defined a faithful action of GB (see exact sequence (3.2)) on Fe. By [13, Lemma 3.5] the triple (S, G, φ) is minimal, iff points xi1 and xi2 lie in the same orbit under the action of GB for each i = 1, . . . , r − m. We consider cases e = 0 and e > 0 in Theorems 3.11 and 3.12 respectively. We make some preparations before statement of Theorem 3.11. Introduce the coordinates (x0 : x1, t0 : t1) on F0 ≃ P1 × P1. The morphism φ′ : S ′ → P1 induces projection σ : (x0 : x1, t0 : t1) 7→ (t0 : t1). The fibres Si ⊂ F0, i = 1, . . . , r − m are represented by the equations: Si : ait0 + bit1 = 0. Consider the form Qr−m =Qr−m we get that(cid:12)(cid:12)StabSt(A5)(σ(Si)) is either 1, or 2, or 3, or 5. Applying [13, Lemma i=1 (ait0 + bit1). We conjugate the subgroup GB ≃ A5 ⊂ P GL(2, C) to the group St(A5). Let StabSt(A5)(σ(Si)), i = 1, . . . , r − m be a stabilizer of point σ(Si) ⊂ P1 in the group St(A5). Considering equations (3.4), 3.5], we get StabSt(A5)(σ(Si)) ≃ Z2, i = 1, . . . , r − m. Hence r − m = 30, and Qr−m = Φ1. The curve C is represented by the equation: (3.13) Equat( C) = p0(t0, t1)x2 0 + 2p1(t0, t1)x0x1 + p2(t0, t1)x2 1 = 0, where pi, i = 0, 1, 2 are binary forms with degree 2d. Note that the degree is even, so as the divisor class C +Pi Si ∈ 2 Pic F0 (see [13, Lemma 3.6]), and r− m = 30 is 1 has no multiple factors, since C is nonsingular. even. The form Disc( C) = p0p2− p2 We will apply the Segre embedding ν : P1 × P1 → P3 to represent the surface S by equations. Introduce the coordinates (x : y : z : w) on P3. This embedding is given by ν : (x0 : x1, t0 : t1) 7→ (x0t0 : x0t1 : x1t0 : x1t1). We choose some Φ1 Equat( C) = polynomials Fi(x, y, z, w), i = 0, . . . , 2d + 28, such that xi ν ∗(Fi). The surface S is represented by the equations in P(dd+15, 14) with the coordinates ui, x, y, z, w, i = 0, . . . , d + 14: 0x2d+28−i 1 (3.14) uiuj = Fi+j , 0 ≤ i ≤ j ≤ d + 14, xj−iui = ujzj−i, yj−iui = ujwj−i, 0 ≤ i < j ≤ d + 14, xw = yz. Theorem 3.11. Let (S, G, φ) ∈ CB , and G be a finite nonsolvable group. Suppose that G0 ≃ 1, GK ≃ hιi ≃ Z2, Σ′ 6= Σ and S ′/ι ≃ F0. Then there is defined a faithful action of GB (see (3.2)) on F0, and GB ⊂ P GL(2, C) × P GL(2, C). There is a G-invariant birational map (S, G, φ) 99K (S ′, G, φ′) described in the diagram (3.12). The surface S ′ can be represented by equations (3.14). The parameter d in (3.13) is odd. The morphism φ′ : S ′ → P1 is given by φ′ : (u0 : . . . : ud+14 : x : y : z : w) 7→( (x : y), if (x : y) 6= (0 : 0); (z : w), if (z : w) 6= (0 : 0). We have (3.15) Φ1(t0, t1) 6 (p0(t0, t1)t2 0 + 2p1(t0, t1)t0t1 + p2(t0, t1)t2 1), where pi, i = 0, 1, 2 are binary forms in (3.13), and Φ1 is the binary form from (3.4). 15 The group GB ⊂ P GL(2, C) × P GL(2, C) is the image of diagonal embedding St(A5) ֒→ P GL(2, C) × P GL(2, C). The group G is isomorphic to ¯A5. This possi- bility for G occur. The group G acts on S ′ by the following way. Embedding of GB ≃ St(A5) to P GL(2, C)×P GL(2, C) defines a unique embedding St( ¯A5) ֒→ SL(2, C)×SL(2, C). This defines an action of St( ¯A5) on the surface S ′ given by the equation (3.14). The action of St( ¯A5) on coordinates ui, i = 0, . . . , d + 14 coincides with the action on monomials xi , i = 0, . . . , d + 14. An action of G is generated by the action of St( ¯A5) and by the map 0xd+14−i 1 (u0 : . . . : ud+14 : x : y : z : w) → (−u0 : . . . : −ud+14 : x : y : z : w). Proof. The subgroup GB ⊂ Aut(F0) acts trivially on Pic(F0). Hence GB ⊂ P GL(2, C) × P GL(2, C). Consider projections πi : P GL(2, C) × P GL(2, C) → P GL(2, C), i = 1, 2 on the first and the second factor respectively. By Gour- sat's Lemma (see [4, Lemma 4.1]) we get that GB ≃ π1(GB)△Dπ2(GB), where D ≃ Im(π1GB)/ Ker(π2GB). We have D ≃ 1 or A5. By [13, Lemma 3.5] we need to find conditions, when points xi1 and xi2 for each i = 1, . . . , 30 lie in the same orbit under an action of GB. Obviously, D ≃ A5. We conjugate GB ⊂ P GL(2, C) × P GL(2, C) to the image of diagonal embedding St(A5) ֒→ P GL(2, C) × P GL(2, C). It's easy to check that none of points xi1 and xi2 for each i = 1, . . . , 30 lies on the diagonal x0t1 − x1t0 = 0. This is a sufficient condition, and it's equivalent to (3.15). Therefore we need to prove existence of curves C ⊂ F0 with odd parameter d, such that condition (3.15) holds. Also we will prove that d cannot be even. We will use notations and arguments of Lemma 3.9. Consider a linear system J of St( ¯A5)-invariant curves with bidegree (2, 2d) in F0. Consider the projection ξ : R2 ⊗ R2d → R2d+2. It's is given by the polynomial p0t2 Suppose that d is even. It's easy to check that any polynomial f (t0, t1) ∈ RSt( ¯A5) Let d is odd. In Lemma 3.9 we proved that a general member of J is nonsingular, if d = 30k + 15, k ∈ N, k ≥ 2. But it's obvious that there exist polynomial f (t0, t1) ∈ RSt( ¯A5) 2d+2 with degree 2d + 2 = 60k + 32, k ≥ 2, which is not divided by The remaining arguments follow from the construction of equations (3.14) and (cid:3) is divided by Φ1. Therefore it's impossible. 0 + 2p1t0t1 + p2t2 1. 2d+2 Φ1. are obvious. In the next theorem we consider case e > 0. Theorem 3.12. Let (S, G, φ) ∈ CB, and G be a finite nonsolvable group. Suppose that G0 ≃ 1, GK ≃ hιi ≃ Z2, Σ′ 6= Σ, and S ′/ι ≃ Fe, e > 0. Then there is a G- invariant birational map (S, G, φ) 99K (S ′, G, φ′) described in diagram (3.12). The surface S ′ contains a G-invariant curve E, which is the preimage of exceptional section Fe. Consider the contraction of this curve (S ′, G, φ′) → (S ′′, G, φ′′), where a map φ′′ : S ′′ 99K P1 is defined by φ′.The surface S ′′ is given by the following equation in P(d + e + 15, e, 1, 1) with the coordinates (u : x : t0 : t1): (3.16) u2 + Φ1(t0, t1)(p0(t0, t1)x2 + 2p1(t0, t1)x + p2(t0, t1)) = 0, where pi, i = 0, 1, 2 are binary forms with degree 2d, 2d + e, 2d + 2e, respectively, and Φ1 is the binary form from (3.4). Also Φ1 6 (p0p2 − p2 1). 16 The map φ′′ is given by φ′′ : (u : x : t0 : t1) 7→ (t0 : t1). Moreover, e ≡ 2 (mod 4). The group G is generated by the maps: u 7→ −u, (u : x : t0 : t1) 7→ (u : x + Fe(t0, t1) : at0 + bt1 : ct0 + d′t1), where (3.17) (3.18) (cid:18)a c b d′(cid:19) ∈ St( ¯A5), G ≃( ¯A5, if d is even, Z2 × A5, if d is odd. and Fe(t0, t1) is a some binary form with degree e, unique for each matrix (3.17). The group G is isomorphic to All possibilities for G occur. Proof. We will use the following construction to represent the surface S ′′ by equa- tions. Consider the morphism Fe → P(e, 1, 1), which is the blowing down of ex- ceptional section Fe. Introduce on P(e, 1, 1) the coordinates (x : t0 : t1). The map φ′ : S ′ → P1 induces the projection σ : (x : t0 : t1) 7→ (t0 : t1). The fibres Si ⊂ Fe, i = 1, . . . , r − m are represented by the equations: Si : ait0 + bit1 = 0. i=1 (ait0 + bit1). We conjugate the group GB ≃ A5 ⊂ P GL(2, C) to the group St(A5). Let StabSt(A5)(σ(Si)), i = 1, . . . , r − m be a stabilizer of point σ(Si) ⊂ P1 in the group St(A5). Considering equations (3.4), 3.5], we get StabSt(A5)(σ(Si)) ≃ Z2, i = 1, . . . , r − m. Hence r − m = 30, and Qr−m = Φ1 (see (3.4)). The curve C is represented by the following equation in P(e, 1, 1) with the coordinates (x : t0 : t1): Consider the form Qr−m = Qr−m we get that(cid:12)(cid:12)StabSt(A5)(σ(Si)) is either 1, or 2, or 3, or 5. Applying [13, Lemma p0(t0, t1)x2 + 2p1(t0, t1)x + p2(t0, t1) = 0. Each fibre Si, i = 1, . . . , 30 intersects the curve C in two distinct points: xi1 and xi2. Hence Φ1 6 (p0p2 − p2 1). We construct a double cover of P(e, 1, 1) branched along C and Φ1(t0, t1) = 0. We get the surface S ′′ given by the equation (3.16). Note that deg(p0) is even, since C +Pi Si ∈ 2 Pic(Fe) (see [13, Lemma 3.6]), and r − m = 30 is even. Denote the degree as 2d. conjugate GB to a group consisting of the following maps Then we apply the arguments as in Theorem 3.10. We prove that e is even, and (x : t0 : t1) 7→ (x + Fe(t0, t1) : at0 + bt1 : ct0 + d′t1), where the coefficients a, b, c, d′ and the binary form Fe satisfy conditions of the Theorem. By [13, Lemma 3.5] we need to find conditions, when the points xi1 and xi2 for each i = 1, . . . , 30 lie in the same orbit under an action of GB. Obviously, it's sufficient to check, that each element g ∈ GB with ord(g) = 2 doesn't have fixed points on the curve C. The element g can be conjugated to the map: (x : t0 : t1) 7→ (x : it0 : −it1). 17 It's easy to see that g doesn't have fixed points on C, iff e ≡ 2 (mod 4). We easily get (3.18). We need to prove existence of curves C ⊂ Fe with odd and even parameters d, such that listed above conditions holds. We assume that GB is a group consisting of maps with condition (3.17). (x : t0 : t1) 7→ (x : at0 + bt1 : ct0 + d′t1), Let d is even. We take e = 34, and the curve C is represented by the following equation in P(34, 1, 1) with the coordinates (x : t0 : t1) Φ3(t0, t1)x2 + Q60(t0, t1)Φ2(t0, t1) = 0, where Φ2 and Φ3 are binary forms from (3.4), and Q60 ∈ I St( ¯A5) (see Notation 3.6) is a some binary form with degree 60 and without multiple factors. Let d is odd. We can employ here example constructed in proof of Theorem 3.10. (cid:3) 3.4. Case, when G0 ≃ 1, GK ≃ Z2 2. In this section we prove the next theorem. Theorem 3.13. Let (S, G, φ) ∈ CB, and G be a finite nonsolvable group. Suppose that G0 ≃ 1, GK ≃ Z2 2. Then there exists an embedding S ֒→ P(E), where E is a line bundle on P1. We have E = E0 ⊕ E1 ⊕ E2, and isomorphisms fi : Ei → O(ai), i = 0, 1, 2, a0 = 0, 0 ≤ a1 ≤ a2. The surface S can be represented by the equation in P(E) ≃ P(O ⊕ O(a1) ⊕ O(a2)): (3.19) 2Xi=0 aiXj,k=0 pj,k i (t0, t1)ξj i ξk i = 0, i are binary forms with degree d, and ξj where pj,k ), i = 0, 1, 2, 0 ≤ j ≤ ai. The morphism φ : S → P1 is induced by the natural projection P(E) → P1. The following conditions holds. i = f −1 0tai−j (tj 1 i H0 = p0,0 0 (t0, t1) ∈ I St( ¯A5), a1Xj,k=0 1 (t0, t1)tj+k pj,k a2Xj,k=0 2 (t0, t1)tj+k pj,k 0 0 (3.20) H1 = H2 = t2a1−j−k 1 t2a2−j−k 1 ∈ I St( ¯A5), ∈ I St( ¯A5). Also the binary forms H0, H1 and H2 do not have multiple and pairwise common factors. The group GK acts by the following way (see Theorem 3.4). ι0(ξ) = ∓ξ0 ± ξ1 ± ξ2, ι1(ξ) = ±ξ0 ∓ ξ1 ± ξ2, ι2(ξ) = ±ξ0 ± ξ1 ∓ ξ2, for any ξ = ξ0 + ξ1 + ξ2, ξi ∈ Ei, i = 0, 1, 2. action of St( ¯A5). The action of St( ¯A5) on sections ξj The action of G on the surface S is generated by the action of GK and an ), i = 0, 1, 2, (tj i = f −1 0tai−j 1 i 18 0 ≤ j ≤ ai is induced by the action on C[t0, t1]. We have (3.21) G ≃( Z2 × ¯A5, if either a1, or a2 is odd, Z2 2 × A5, otherwise. All possibilities for G occur. 1) ⊕ O(a′ 2). Obviously, we can take a′ 0) ⊕ O(a′ An action of G on O(−KS) defines an action on E. In the next lemma we show Proof. Denote as f the fibre's divisor class of the conic bundle (S, G, φ). We have −KS · f = 2. It's well known that a line bundle O(−KS) is locally free of rank 3. Hence the line bundle O(−KS) is relatively very ample and defines an embedding S ֒→ P(E ′), where E ′ = φ∗(O(−KS)). By Grothendieck theorem we have E ′ = O(a′ 2. Hence we can take the bundle E in the statement of theorem to be equal E = E ′ ⊗ O(−a′ 0). that the action of GK on E is diagonalizable. Lemma 3.14. We can choose a decomposition E = E0⊕E1⊕E2, where Ei ≃ O(ai), i = 0, 1, 2, such that GK acts by the following way. Denote three different nontrivial elements in GK ≃ Z2 2 as ι0, ι1, ι2. Then 0 ≤ a′ 1 ≤ a′ (3.22) ι0(ξ) = ∓ξ0 ± ξ1 ± ξ2, ι1(ξ) = ±ξ0 ∓ ξ1 ± ξ2, ι2(ξ) = ±ξ0 ± ξ1 ∓ ξ2, the action of GK. Hence the statement is obvious. for any ξ = ξ0 + ξ1 + ξ2, ξi ∈ Ei, i = 0, 1, 2. Proof. Recall that a0 = 0, 0 ≤ a1 ≤ a2. Also remind that GK acts trivially on the base of fibration φ. Suppose that 0 < a1 < a2. Then each bundle Ei, i = 0, 1, 2 is invariant under Suppose that ai = aj, ai 6= ak for some i 6= j 6= k 6= i. Without loss of generality we can take 0 = a1 < a2. Then the action of GK on E0 ⊕ E1 defines an embedding GK ֒→ GL(2, C). But, obviously, any subgroup Z2 2 ⊂ GL(2, C) is diagonalizable. Suppose that 0 = a1 = a2. Then the statement follows from the fact that any subgroup Z2 (cid:3) 2 ⊂ GL(3, C) is diagonalizable. We apply Lemma 3.14. Fix isomorphisms fi : Ei → O(ai), i = 0, 1, 2. Let ξj i = f −1 ), i = 0, 1, 2, 0 ≤ j ≤ ai be generators of global section spaces of bundles Ei. Then the surface S is presented by the following equation in P(E) 0tai−j (tj 1 i X0≤i≤j≤2 Xk≤ai,l≤aj pk,l i,j (t0, t1)ξk i ξl j = 0, where pk,l (3.22) that pk,l (3.19). i,j (t0, t1) are binary forms with degree d. But it easily follows from equations i,j = 0, if i 6= j. Hence the surface S can be represented by equation Let's find relations on the forms pj i , 0 ≤ j ≤ ai, i = 0, 1, 2. By Theorem 3.4 each nontrivial element ιi, i = 0, 1, 2 of the subgroup GK fixes pointwise an irreducible smooth bisection Ci. Hence, there is defined an action of G on the set of these curves, since GK ⊳ G. This action defines a homomorphism σ : G → S3. But GB ≃ A5 is simple. Therefore σ is trivial. The curves Ci, i = 0, 1, 2 on the surface S are cut out by the hypersurfaces: ξj i = 0, 0 ≤ j ≤ ai. 19 We conjugate GB ⊂ P GL(2, C) to St(A5). We employ now notations (3.20). From 2 ∈ I St( ¯A5). triviality of σ we get: H0H1, H0H2, H1H2 ∈ I St( ¯A5). Hence H 2 One knows that all characters of St( ¯A5) are trivial (see (3.4)). Therefore we get (3.20). 1 , H 2 0 , H 2 It's easy to check that the surface S is nonsingular, iff the binary forms H0, H1 and H2 do not have multiple and pairwise common factors. Now we can describe an action of group G on the surface S. The action of St( ¯A5) on C[t0, t1] induces an action on ξj ), i = 0, 1, 2, 0 ≤ j ≤ ai. The action of G is generated by the action of GK and the action of St( ¯A5). Therefore we easily get (3.21). It's easy to check that all possibilities for G in (3.21) occur. (cid:3) i = f −1 0tai−j (tj i 1 4. Conjugacy question. The main result of this section is Theorem 4.2. Here we reprove results of [4, Section 8] for the sake of completeness. As the main tool we will use the next theorem. Theorem 4.1 ([10, Theorem 1.6]). (1) Let (S, G) be a surface in the class D with degree K 2 S = 1, and let χ : S 99K S ′ be a birational G-invariant map onto an arbitrary surface (S ′, G) ∈ D∪ CB. Then S ′, like S, is a del Pezzo surface of degree 1 and χ is an isomorphism. (2) Let χ : S 99K S ′ be a birational map, where (S, G) ∈ D and (S ′, G) ∈ D∪CB. S, where OrbG(x) is Suppose that S has no points x with OrbG(x) < K 2 an orbit of point x under action of G. Then χ is an isomorphism. (3) Let χ : S 99K S ′ be a birational G-invariant map, where (S, G, φ) ∈ CB and (S ′, G) ∈ D ∪ CB. Suppose that K 2 S = K 2 S ′, and χ takes a pencil of conics on S to a pencil of conics on S ′, that is, the diagram S ≤ 0; then (S ′, G, φ′) ∈ CB, K 2 χ /❴❴❴ S ′ S φ φ′ P1 π / P1 is commutative, where π is an isomorphism over C. Theorem 4.2. (1) Let (S, G, φ) be a surface in the class CB, K 2 S ≤ 0, and G be a finite nonsolvable group. Let χ : S 99K S ′ be a birational G-invariant map, where (S ′, G) ∈ D ∪ CB. Then (S ′, G, φ′) ∈ CB, and K 2 S = K 2 S ′. The map χ is a composition of elementary transformations elmx1 ◦ . . . ◦ elmxn , where (x1, . . . , xn) is a G-invariant set of points not lying on a singular fibre with no two points lying on the same fibre. (2) Let (S, G, φ) be a surface in the class CB, K 2 S > 0, and G be a finite nonsolvable group. Let χ : S 99K S ′ be a birational G-invariant map, where (S ′, G) ∈ D ∪ CB. Then K 2 (a) Let n is odd. Then one the following cases occurs S = 8, S ≃ Fn, n 6= 1. (4.1) S ′ ≃ P2, and G ≃ Zn′ × ¯A5; S ′ ≃ Fm, where m is odd and m 6= 1. (b) Let n is even. Then S ′ can be isomorphic to Fm, where m is even. If n = 0, and m 6= 0, then G ≃ Zm × A5. 20 /     / (3) Let (S, G) be a surface in the class D, and G be a finite nonsolvable group. Let χ : S 99K S ′ be a birational G-invariant map, where (S ′, G) ∈ D ∪ CB. Then we have the following. (a) Let S ≃ P2, and G ≃ Zm× ¯A5. Then the surface S ′ may be isomorphic to either P2, or Fn, where n is odd. (b) Otherwise we have S ′ ≃ S. If K 2 S < 9, then χ is an automorphism of S. Remark 4.3. I.Cheltsov proved the following. If (Fn, G, φ) ∈ CB, n 6= 1, and G be a finite nonsolvable group, then there exist a birational G-invariant map χ : Fn 99K S ′, where S ′ ≃ P2, if n is odd; S ′ ≃ F0, if n is even. Proof. The first case of theorem follows from Theorem 4.1 and [4, Theorems 7.7, Proposition 7.14]. Let's prove the second case of theorem. First we prove that K 2 S = 8. Denote by r the number of the reducible fibres of conic bundle (S, G, φ). Suppose that r 6= 0. Obviously, we have GB ≃ A5 (see exact sequence (3.2) and Klein's classification of the finite subgroups in P GL(2, C) in [4, Section 5.5]). Then by (3.4) we have r ≥ 12. Hence by Noether formula K 2 We will use results of Theorem 3.1 and theory of elementary links (see [4, Section 7.2]). By [4, Theorem 7.7] the map χ is equal to a composition of elementary links χ1◦ . . .◦ χk. It's easy to check that χk is an elementary link of type II (see Theorem 3.1). We have χk(Fn) ≃ Fl. Then we apply Theorem 3.1. For any point x ∈ P1 we have OrbSt(A5)(x) is even (see Notation 3.5). Hence we easily get that l − n is even. Therefore if n is even, then χi, i = 1, . . . , k are elementary links of type II. And S = 8 − r ≤ −4. Therefore r = 0, K 2 S = 8. we easily get the case 2b of theorem. Consider case, when n is odd. Then G ≃ Zn′ × ¯A5 by Theorem 3.1. Suppose that χ is not a composition of elementary links of type II. Then one of elementary links χi, 1 ≤ i ≤ k− 1 must be a link of type III. We may suppose that the links χj, i < j ≤ k are of type II. In this case χi ◦ . . . ◦ χk(S) ≃ P2. Below we will see that S ′ 6≃ X, where (X, G) ∈ D, and K 2 X < 9. Therefore we get the case 2a of theorem. Let's prove the third case of theorem. To apply Theorem 4.1, we need to show, that S has no points x with OrbG(x) < K 2 S. We will argue, considering different values of K 2 S. If K 2 S is equal to either 7, or 6, or 4 , or 1, then by Theorem 2.3 there is no such pairs (S, G). Let K 2 S = 2. We apply Theorem 2.6. Consider the following two automorphisms of surface S, given by equation (2.4): α : (T0 : T1 : T2 : T3) 7→ (T1 : T2 : T0 : T3), β : (T0 : T1 : T2 : T3) 7→ (ε7T0 : ε4 7T1 : ε2 7T2 : T3). Let K 2 It's easy to check by calculations that α, β ∈ G, and there is no point x ∈ S fixed under an action of subgroup H ⊂ G generated by α and β. S = 3. We apply Theorem 2.5. Consider the maps (2.3). Again by easy calculations we can check that there is no point x ∈ S, such that OrbG(x) < 3. S = 5. We apply Theorem 2.4. It's sufficient to consider the case G ≃ A5. Suppose that there is a point x ∈ S, such that OrbG(x) < K 2 S = 5. Denote by Let K 2 21 StabG(x) the stabilizer of point x in the group G. Then StabG(x) is a subgroup of G with order either 15, or 20, or 30 or 60. It's well known that there are no subgroups of A5 with order either 15, or 20, or 30. Hence StabG(x) = 60 = G. But it's easy to see from (2.2) that the action of group G on the surface S has no fixed points. Contradiction. Let K 2 The case K 2 S = 8. We apply Theorem 2.2. Consider the action of St(A5) on P1. It's known (see (3.4)) that OrbSt(A5)(x) ≥ 12 for any point x ∈ P1. Therefore OrbG(x) ≥ 12 for any point x ∈ S. S = 9 easily follows from the above investigation and Theorem 2.1. However in this case the condition OrbSt(A5)(x) ≥ 9 for any point x ∈ S ≃ P2 not always holds. For example in the second case of Theorem 2.1. Therefore we cannot apply Theorem 4.1 to prove that χ is an isomorphism. But in fact we need only to know an isomorphism class of S ′ to describe conjugacy classes. (cid:3) Corollary 4.4. Suppose that the conditions of case 1 of Theorem 4.2 are satisfied. After suppose that GK 6≃ Z2 (see (3.2)). Then χ is an isomorphism. Proof. By Theorems 3.4 and 3.7 we have GK ≃ Z2 2 or Dn, n ≥ 2. Therefore it's easy to see that any G-orbit of points (x1, . . . , xn) has two points lying on the same fibre of conic bundle φ. Contradiction to case 1 of Theorem 4.2. (cid:3) 5. The list of finite nonsolvable subgroups in Cr2(C). Summarizing results obtained in sections 2 and 3 we get the following list of finite nonsolvable subgroups in Cr2(C). • L2(7) This group is presented in Cr2(C) by pairs (S, L2(7)) ∈ D, which were described in Theorems 2.1 and 2.6. • Z2 × L2(7) This group is presented in Cr2(C) by pairs (S, Z2 × L2(7)) ∈ D, which were described in Theorem 2.6. • A6 • S5 This group is presented in Cr2(C) by pairs (S, A6) ∈ D, which were described in Theorem 2.1. This group is presented in Cr2(C) by pairs (S, S5) ∈ D, which were described in Theorems 2.4, 2.5. • St(A5) ≀ hτi This group is presented in Cr2(C) by pairs (S, St(A5) ≀ hτi) ∈ D, which were described in Theorem 2.2. • A5 × A5, A5 × S4, A5 × A4. These groups are presented in Cr2(C) by triples (S, G, φ) ∈ CB (G is one of our groups), which were described in Theorem 3.1. • Dn × ¯A5, n ≥ 3 This group is presented in Cr2(C) by triples (S, Dn× ¯A5, φ) ∈ CB, which were described in Theorem 3.7. • Dn × A5, n ≥ 3 This group is presented in Cr2(C) by triples (S, Dn× A5, φ) ∈ CB, which were described in Theorems 3.1 and 3.7. 22 • Zn × ¯A5, n ≥ 3 • Zn × A5, n ≥ 3 This group is presented in Cr2(C) by pairs (S, Zn × ¯A5) ∈ D and triples (S, Zn × ¯A5, φ), which were described in Theorems 2.1, 3.1. This group is presented in Cr2(C) by triples (S, Zn × A5, φ) ∈ CB, which were described in Theorem 3.1. 2 × ¯A5 This group is presented in Cr2(C) by triples (S, Z2 were described in Theorem 3.7. 2 × A5 This group is presented in Cr2(C) by triples (S, Z2 were described in Theorems 3.1, 3.7 and 3.13. 2 × ¯A5, φ) ∈ CB, which 2 × A5, φ) ∈ CB, which • Z2 • Z2 • Z2 × ¯A5 • Z2 × A5 This group is presented in Cr2(C) by pairs (S, Z2 × ¯A5) ∈ D and triples (S, Z2 × ¯A5, φ) ∈ CB, which were described in Theorems 2.1, 3.1, and 3.13. This group is presented in Cr2(C) by pairs (S, Z2 × A5) ∈ D and triples (S, Z2 × A5, φ) ∈ CB, which were described in Theorems 2.2, 3.1, 3.8, 3.10, 3.12. • ¯A5 This group is presented in Cr2(C) by pairs (S, ¯A5) ∈ D and triples (S, ¯A5, φ) ∈ CB, which were described in Theorems 2.1, 3.1, 3.8, 3.10, 3.11, 3.12. • A5 This group is presented in Cr2(C) by pairs (S, A5) ∈ D and triples (S, A5, φ) ∈ CB, which were described in Theorems 2.1, 2.4, 3.1. References [1] L. Bayle and A. Beauville, Birational involutions of P2. Asian J. Math. 4 (2000), no. 1, 11-17. [2] E. Bertini, Richerche sulle transformazioni univoche involutorie nel piano. Annali di Mat. 8 (1877), 244-286. [3] J. Blanc, Elements and cyclic subgroups of finite order of the Cremona group. Comment. Math. Helv. 86 (2011), 2, 469-497. [4] I. Dolgachev and V. Iskovskikh, Finite subgroups of the plane Cremona group. In: Algebra, arithmetic, and geometry: in honor of Yury I. Manin, Vol. I, pp. 443-548, Boston 2009. [5] I. Dolgachev, Topics in classical algebraic geometry. Preprint (2010). http://www.math.lsa.umich.edu/idolga/lecturenotes.html [6] W. Fulton and J. Harris, Representation theory: A first course. Graduate Texts in Mathe- matics, Springer-Verlag New York Inc 1991. [7] V. A. Iskovskikh, Rational surfaces with a pencil of rational curves. Sbornik Mathematics (N.S.). 74(116):4 (1967), 608-638. [8] V. A. Iskovskikh, Rational surfaces with a pencil of rational curves and with positive square of the canonical class. Sbornik Mathematics (N.S.). 83(125):1(9) (1970), 90-119. [9] V. A. Iskovskikh, Minimal models of rational surfaces over arbitrary fields. Izv. Akad. Nauk USSR Ser. Mat.. 43:1 (1979), 19-43. [10] V. A. Iskovskikh, Factorization of birational maps of rational surfaces from the viewpoint of Mori theory. Russ. Math. Survey. 51(4) (1996), 585-652. [11] S. Kantor, Theorie der endlichen Gruppen von eindeutigen Transformationen in der Ebene. Berlin (1895). [12] Yury I. Manin, Rational surfaces over perfect fields, II. Sbornik Mathematics (N.S.). 72(114):2 (1967), 161-192. 23 [13] V. I. Tsygankov, Equations of G-minimal conic bundles. Sbornik: Mathematics (N. S.). 202:11 (2011), 1667-1721. [14] A. Wiman, Zur Theorie der endlichen Gruppen von birationalen Transformationen in der Ebene. Math. Ann.. 48 (1896), 195 -- 241. Johannes Gutenberg-Universitat, Institut fur Mathematik, Staudingerweg 9, 55099 Mainz, Germany E-mail address: [email protected] 24
1702.08631
2
1702
2018-12-11T05:30:03
Topological recursion with hard edges
[ "math.AG", "math-ph", "math-ph" ]
We prove a Givental type decomposition for partition functions that arise out of topological recursion applied to spectral curves. Copies of the Konstevich-Witten KdV tau function arise out of regular spectral curves and copies of the Brezin-Gross-Witten KdV tau function arise out of irregular spectral curves. We present the example of this decomposition for the matrix model with two hard edges and spectral curve $(x^2-4)y^2=1$
math.AG
math
TOPOLOGICAL RECURSION WITH HARD EDGES LEONID CHEKHOV AND PAUL NORBURY ABSTRACT. We prove a Givental type decomposition for partition functions that arise out of topological recursion applied to spectral curves. Copies of the Konstevich-Witten KdV tau function arise out of regular spectral curves and copies of the Brezin-Gross-Witten KdV tau function arise out of irregular spectral curves. We present the example of this decomposition for the matrix model with two hard edges and spectral curve (x2 − 4)y2 = 1. CONTENTS Introduction 1. 1.1. Acknowledgements 2. Kontsevich-Witten and Brezin-Gross-Witten tau functions 3. Partition function for topological recursion 3.1. Topological field theory and asymptotic behaviour of ωg,n. 4. Givental decomposition 4.1. Translations 4.2. Graphical expansion 5. Legendre Ensemble 5.1. Chekhov-Givental decomposition Appendix A Quantisation References 1. INTRODUCTION 1 4 4 6 6 8 8 10 15 16 20 21 The partition functions for Gromov-Witten invariants of P1, the Gaussian Hermitian matrix model, the Legendre ensemble, and enumeration of dessins d'enfant, which are formal series in an infinite sequence of variables {¯h, vk,α k ∈ N, α ∈ {1, 2}}, have in common a decomposition given by a differential operator R acting on the product of two species of the Kontsevich-Witten KdV tau function ZKW or of the Brezin-Gross- Witten KdV tau function ZBGW -- defined in Section 2. ZGW(¯h, {vk,α}) = R · T1 · ZKW(2¯h, { ZGUE(¯h, {vk,α}) = R · T2 · ZKW(2¯h, { ZLeg(¯h, {vk,α}) = R · T3 · ZBGW(2¯h, { 1 ZDes(¯h, {vk,α}) = R · T4 · ZBGW(− 2 √2vk,1})ZKW(−2¯h, {i√2vk,2}) √2vk,1})ZKW(−2¯h, {i√2vk,2}) √2vk,1})ZBGW(−2¯h, {i√2vk,2}) vk,1})ZKW(32¯h, {4√2vk,2}). i √2 ¯h, { (1.1) The operator R, which is the exponential of a quadratic differential operator, is common to all four models, whereas Ti are operators of translations vk,α 7→ vk,α + ck,α i described in Section 4.1. The partition func- tion ZGW stores ancestor Gromov-Witten invariants of P1. Its decomposition in (1.1) is a particular case of Date: December 12, 2018. 2010 Mathematics Subject Classification. 14N10; 05A15; 32G15. 1 2 LEONID CHEKHOV AND PAUL NORBURY Givental's decomposition of partition functions of Gromov-Witten invariants of targets X with semi-simple quantum cohomology [21] which applies more generally to partition functions of semi-simple cohomolog- ical field theories. It is usually expressed as a function of variables corresponding to cohomology classes in H∗(P1) denoted {tk,β} which are related to the variables in the decomposition by vk,1 = 1√2 (tk,1 + tk,2), vk,2 = i√2 (−tk,1 + tk,2). The partition functions ZGUE and ZLeg store moments of the probability measure RHN exp (NV(M))DM as asymptotic expansions in ¯h = 1/N2 for N → ∞ where HN consists of N × N Hermitian matrices. For ZGUE, we use V(M) = −tr (M2). For ZDes we use V(M) = −tr M · χ[0,∞)(M) where the indicator function is defined by χ[0,∞)(M) := ∏N i=1 χ[0,∞)(λi) and for A ⊂ R, χA(x) = 1 for x ∈ A and 0 otherwise. This produces a potential with infinite wall, or hard edge, at zero eigenvalue. For ZLeg we use V(M) = χ[−2,2](M) (again χ[−2,2](M) := ∏N i=1 χ[−2,2](λi)) i.e. one restricts eigenvalues to lie in the interval [−2, 2] and sets V(M) = 0. This produces a potential with infinite walls, or hard edges, at eigenvalues −2 and 2. Note that for the integration over the compact domain [−2, 2]N we do not need a Gaussian term for convergence. The decomposition of ZGUE is a decomposition of the partition function for a Hermitian matrix model with Gaussian potential proven by the first author in [6] and in fact gives an example of Givental's decomposition via an associated cohomological field theory [3]. The decomposition of ZLeg into two copies of ZBGW is described in detail in this paper as a particular example of the more gen- eral result involving copies of both ZKW and ZBGW. An example of mixed ZBGW and ZKW factors is given by ZDes the partition function for enumeration of dessins d'enfant [10]. The conclusion is that the partition functions ZKW and ZBGW are fundamental to a large class of partition functions arising from many areas. For the two choices of V(M) above, the limit y(x) = lim N→∞ N−1ZHN(cid:28)tr 1 x − M(cid:29) exp (NV(M))DM is a holomorphic function near x = ∞ which analytically continues to define a Riemann surface, known as a spectral curve, given as a double cover of the x-plane x = z + 1/z on which y(x)dx extends to a well-defined differential r(z)dz for r(z) a rational function. One can also associate a Riemann surface to Gromov-Witten invariants of P1 via an associated Landau-Ginzburg model. Each of the examples in (1.1) can be formulated in terms of a recursive construction of meromorphic differentials defined on the associated Riemann surface known as topological recursion. In this paper we prove a decomposition theorem for partition functions arising out of topological recursion that generalises the examples in (1.1). Topological recursion developed by Eynard, Orantin and the first author [7, 8, 18] produces invariants ωg,n for integers g ≥ 0 and n ≥ 1, which we will refer to as correlators, of a Riemann surface Σ equipped with two meromorphic functions x, y : Σ → C and a bidifferential B(p1, p2) for p1, p2 ∈ Σ. The zeros Pα of dx must be simple and disjoint from the zeros of dy. We refer to the data S = (Σ, B, x, y) as a spectral curve. We allow Σ to be (a possibly disconnected) open subset of a compact Riemann surface, in which case S is known as a local spectral curve. For integers g ≥ 0 and n ≥ 1, the correlator ωg,n is a multidifferential on Σ or, in other words, a tensor product of meromorphic differentials on Σn. It is defined recursively via ω0,1(p) = −y(p)dx(p), ω0,2(p1, p2) = B(p1, p2) which are used to define the kernel in a neighbourhood of p2 = Pα for dx(Pα) = 0 K(p1, p2) = 1 2 R p2 σα(p2) B(p, p1) (y(p2) − y(σα(p2)))dx(p2) . The point σα(p) ∈ Σ is defined to be the unique point σα(p) 6= p close to α such that x(σα(p)) = x(p) which is well-defined since each zero Pα of dx is assumed to be simple. For L = {2, ..., n} define (1.2) ωg,n(p1, pL) = D ∑ α=1 Res p=Pα K(p1, p)(cid:20)ωg−1,n+1(p, σα(p), pL) + ωg1,I+1(p, pI) ωg2,J+1(σα(p), pJ)(cid:21) ◦ ∑ g1+g2=g I⊔J=L TOPOLOGICAL RECURSION WITH HARD EDGES 3 where the outer summation is over the zeros Pα of dx and the ◦ over the inner summation means that we ex- clude terms that involve ω0 1. The recursive definition of ωg,n(p1, . . . , pn) uses only local information around zeros of dx so a local spectral curve containing the zeros of dx is sufficient. A zero of dx is regular if y is analytic there. A spectral curve is regular if y is analytic at all zeros of dx. In this paper we consider irregular spectral curves where y may have a simple pole at any zero of dx. The correlators ωg,n are polynomial in a basis of differentials vk,m = Vk,m(pi) on Σ depending only x and B -- defined in (3.1) in Section 3. Define the topological recursion partition function of the spectral curve S = (Σ, B, x, y) by ZS(¯h, {vk,i}) = exp ∑ g,n ¯hg−1 n! ωg,n({vk,i})! . The topological recursion partition function of the curve x = 1 2 y2 (equipped with the Cauchy kernel B = dy1dy2/(y1 − y2)2) is the Kontsevich-Witten KdV tau function Z = ZKW and we write ωAiry g,n for the corre- lators of this curve known as the Airy curve due to its relation to the differential equation satisfied by the Airy function. Similarly, the topological recursion partition function of the curve xy2 = 1 2 yields the Brezin- Gross-Witten KdV tau function Z = ZBGW, defined in Section 2, and we write ωBes g,n due to a relation of the curve with the Bessel equation [11]. The BGW model was first identified with the KdV τ-function in [28]. For a general spectral curve S, it is straightforward to prove that the asymptotic behaviour, or largest order principal part, of ωg,n near each zero Pα of dx is given by the correlators for the local model of the curve 2 y2 and xy2 = 1 x = 1 for some constant c ∈ C near any regular zero of dx and ωg,n ∼ c2g−2+nωBes g,n near any irregular zero of dx where y has a simple pole. What is much deeper is that ωg,n can be constructed completely from copies of ω g,n . This is described in terms of the partition functions in Theorem 1 below. 2 , i.e. ωg,n ∼ c2g−2+nωAiry g,n Airy g,n and ωBes In [13] Dunin-Barkowski, Orantin, Shadrin and Spitz proved that the partition function ZS of a regular spectral curve satisfying a finiteness assumption possesses a decomposition in terms of products of ZKW acted on by differential operators built out of spectral curve data. Furthermore, they showed that this decomposition coincides with a decomposition of Givental [21] for partition functions Z arising out of semi- simple cohomological field theories. An immediate consequence is that, under some assumptions on the spectral curve, topological recursion produces partition functions Z for semi-simple cohomological field theories. The results of [13] require the spectral curve to have regular singularities, i.e. dy must be analytic at the zeros of dx. The main result of this paper is a generalisation of the decomposition theorem to allow irregular singularities. Theorem 1. Consider a spectral curve S = (Σ, B, x, y) with m irregular zeros of dx at which y has simple poles, and D − m regular zeros. There exist operators R, T and ∆ defined in Definition 4.7 determined explicitly by (Σ, B, x, y) such that (1.3) ZS = R T ∆ZBGW(¯h, {vk,1}) · · · ZBGW(¯h, {vk,m})ZKW(¯h, {vk,m+1}) · · · ZKW(¯h, {vk,D}). Moreover, R depends only on (Σ, B, x), T is a translation operator, and ∆ acts by rescaling ¯h and vk,i. Remark 1.1. Theorem 1 generalises the result of [13] even when applied to a regular spectral curve since it contains a translation term T that does not appear in the decomposition theorem of [13]. More precisely, [13] considers a restricted class of spectral curves on which the function y is determined by (Σ, B, x) to- gether with the m numbers dy(Pα) ∈ C and corresponds to cohomological field theories with flat identity. In particular [13] does not apply to Weil-Petersson volumes studied by Mirzakhani [29] which gives a fun- damental example of a cohomological field theory without flat identity. This is despite the proof in [16, 20] 4 LEONID CHEKHOV AND PAUL NORBURY that Weil-Petersson volumes do indeed arise out of topological recursion applied to a spectral curve. Theo- rem 1 remedies this situation. The translation term in Theorem 1 (for regular spectral curves) corresponds to a translation of cohomological field theories -- see [31]. Remark 1.2. In the examples (1.1) the different operators ∆ are visible via the differential rescalings of ¯h. The operator R is built out of the data (Σ, B, x) -- see Section 4. It is the same for all four examples of (1.1) reflecting the fact that the associated spectral curves differ only in the definition of y, i.e. (Σ, B, x) is the same in these four cases. Remark 1.3. The partition functions arising from Gromov-Witten invariants, cohomological field theories, matrix models and topological recursion possess two sets of natural coordinates -- flat coordinates and canon- ical coordinates. Theorem 1 is expressed with respect to canonical coordinates. The change of coordinates between canonical and flat is given by an m × m matrix, i.e. it is linear and independent of the first param- eter d. This change of coordinates appears in Givental's decomposition as a further operator acting on the left of (1.3). The proof of the formula (1.3) is built up progressively via special cases proven throughout the text. The reader may find some of these simpler versions more digestable. The basic cases of ( R, T, ∆) = (Id, Id, Id) and D = 1 appear in (2.2) and (2.3). The case of ( R, T, ∆) = (Id, Id, ∆) which gives rise to a topological field theory appears in (3.8). Translations can be best understood via the case ( R, T, ∆) = (Id, T, ∆) and D = 1 given in (4.4). In Section 2 we recall the definitions of the two KdV tau functions which form the fundamental pieces of the decomposition. In Section 3 we introduce the decomposition without the differential operator R via the elementary topological part of the correlators. In Section 4 we prove the decomposition (1.3). We apply the decomposition to the example of the Legendre ensemble in Section 5 and demonstrate an application of the decomposition (1.3) pictorially. In this paper we use the convention N = {0, 1, 2, ...}. 1.1. Acknowledgements. The authors would like to thank Maxim Kazarian, Nicolas Orantin, Sergey Shad- rin and Peter Zograf for useful conversations and the referee for many useful comments. PN was partially supported by the Australian Research Council grant DP170102028. The work of L.Ch. was partially sup- ported by the Russian Foundation for Basic Research (Grant No. 17-01-00477) and by the ERC Advanced Grant 291092 "Exploring the Quantum Universe" (EQU). 2. KONTSEVICH-WITTEN AND BREZIN-GROSS-WITTEN TAU FUNCTIONS The fundamental components ZKW(¯h, t0, t1, ...) and ZBGW(¯h, t0, t1, ...) of the factorisation (1.3) are tau func- tions of the KdV hierarchy. The Kontsevich-Witten tau function ZKW was introduced in [32], and the Brezin- Gross-Witten tau function ZBGW arises out of a unitary matrix mode studied in [5, 23]. A tau function Z(t0, t1, ...) of the KdV hierarchy (equivalently the KP hierarchy in odd times p2k+1 = tk/(2k + 1)!!) gives rise to a solution of the KdV hierarchy via Z = exp F, U = ¯h ∂2F ∂t2 0 (2.1) Ut1 = UUt0 + ¯h 12 Ut0t0t0, U(t0, 0, 0, ...) = f (t0). The first equation in the hierarchy is the KdV equation (2.1), and later equations Utk = Pk(U, Ut0, Ut0t0, ...) for k > 1 determine U uniquely from U(t0, 0, 0, ...). The Kontsevich-Witten tau function ZKW is defined by U(t0, 0, 0, ...) = t0 (and is in fact determined uniquely by (2.1) and the string equation Ft0 = 1 the higher equations giving Utk for k > 1 are automatically satisfied). It is famously a generating function for intersection numbers over the moduli space of stable curves equipped with tautological line bundles Li → Mg,n, i = 1, ..., n and ψi = c1(Li). 0 + ∑ ti+1Fti i.e. 2 t2 TOPOLOGICAL RECURSION WITH HARD EDGES 5 Theorem 2 (Witten-Kontsevich 1992 [24, 32]). FKW(¯h, t0, t1, ...) = ∑ g,n ¯hg−1 1 n! ∑ ~k∈NnZMg,n n ∏ i=1 ψki i tki Its first few terms are given by FKW(¯h, t0, t1, ...) = ¯h−1( t3 0 3! t3 0t1 3! + t4 0t2 4! + ...) + t1 24 + ... Its dispersionless limit is nontrivial: lim ¯h→0 U = It arises via topological recursion applied to the Airy curve [19]. For + + t0 1 − t1 t2 0t2 2(1 − t1)3 + .... SAiry = {x = 1 2 z2, y = z, B = and ω Airy g,n defined by (1.2), we have dzdz′ (z − z′)2} ω Airy g,n = ∑ ~k∈Zn +ZMg,n n ∏ i=1 ψki i (2ki + 1)!! dzi z2ki+2 i . Out of the correlators ω from Definition 3.1 gives Airy g,n , (3.2) builds a partition function which in this case using Vk(z) = (2k + 1)!! dz z2k+2 (2.2) This is the first case of (1.3) where D = 1 and R = Id = T = ∆. ZSAiry = ZKW. The Brezin-Gross-Witten solution of the KdV hierarchy is defined by the initial condition The first few terms of its tau function are given by U(t0, 0, 0, ...) = ¯h 8(1 − t0)2 . log ZBGW = FBGW(¯h, t0, t1, ...) = 1 8 t0 + 1 16 t2 0 + 1 24 t3 0 + ... + ¯h( 3 128 t1 + 9 128 t0t1 + ...) + .. and we see that its dispersionless limit is trivial: It arises via topological recursion applied to the Bessel curve [11] U = 0. lim ¯h→0 as follows. Write SBes = {x = z2, y = 1 2 1 z , B = ωBes g,n = ∑ µ∈Zn + bg,n(µ1, . . . , µn) dzdz′ (z − z′)2 } n ∏ i=1 dzi zµi+1 i for the correlators of topological recursion applied to the curve SBes. As above Vk(z) = (2k + 1)!! dz it is proven in [11] that z2k+2 and FSBes g = FBGW g = ∑ n 1 n! ∑ k∈Nn bg,n(2k1 + 1, . . . , 2kn + 1) ∏n i=1(2ki + 1)!! tk1 . . . tkn hence ZSBes = ZBGW (2.3) which is again a case of (1.3) for D = 1 and R = Id = T = ∆. 6 LEONID CHEKHOV AND PAUL NORBURY Remark 2.1. Kontsevich and Soibelman [25] have studied topological recursion via an algebraic structure which they call an Airy structure referring to the fact that a regular spectral curve locally resembles the Airy curve. The more general setup of irregular spectral curves that locally resemble the Bessel curve also determines an Airy structure, using the technique of abstract topological recursion [2]. 3. PARTITION FUNCTION FOR TOPOLOGICAL RECURSION In this section we define the partition function ZS built out of the correlators ωg,n of the spectral curve S. It is a generating function for all ωg,n with the substitution of variables vk,i for differentials on the curve. We then give a leisurely introduction to the formula (1.3) by considering only part of this formula, obtained by arranging R = Id = T. Any correlator ωg,n(p1, ..., pn) has the property that its principal part in any pi at any zero Pi of dx is skew- invariant under the local involution defined by dx around Pi. Eynard [15] defined a collection of auxiliary differentials (defined below) on the curve which span all those meromorphic differentials with principal part at any zero P of dx skew-invariant under the local involution defined by dx around P. Hence ωg,n is a polynomial in these auxiliary differentials. Definition 3.1. For a Riemann surface Σ equipped with a meromorphic function x : Σ → C define the auxiliary differentials on Σ as follows: (3.1) Vα 0 (p) = B(Pα, p), Vα k+1(p) = d(cid:18) Vα dx(p)(cid:19) , α = 1, ..., D, k (p) k = 0, 1, 2, ... where B(Pα, p) is evaluation at Pα, a zero of dx . Evaluation of any meromorphic differential ω at a simple zero P of dx is defined by ω(P ) := Res p=P ω(p) p2(x(p) − x(P )) where we choose a branch ofpx(p) − x(P ) once and for all at each P to remove the ±1 ambiguity. The meromorphic differentials Vα k (p) defined above constitute local Krichever -- Whitham systems [26] of 1-differentials skew-invariant with respect to local involutions. In [15] Eynard defines dξα,k(p), using local coordinates, which give the principal part of Vα For each locally defined involution σα defined in a neighbourhood of a zero Pα of dx, we have Vα k (σα(p)) is analytic at Pα. The Vα Vα skew-invariant under each involution σα and ωg,n is a polynomial in them: k (p) + k (p) form a basis for meromorphic differentials which have principal part k (p) and serve a similar role. ωg,n(p1, ..., pn) = ∑ ~α,~k cg,~α,~k n ∏ i=1 Vαi ki (pi). The partition function of a spectral curve S = (Σ, B, x, y) is defined by: (3.2) ZS(¯h, {vk,α}) = exp ∑ g,n ¯hg−1 1 n! ωg,n(p1, ..., pn){V αi ki (pi)=vki,αi}! . 3.1. Topological field theory and asymptotic behaviour of ωg,n. Given a spectral curve S = (Σ, B, x, y), around any regular zero Pi of dx the pair (Σ, x) resembles the Airy curve x = 1 2 s2. A consequence of this proven in [18] is that near a branch point, the asymptotic behaviour of ωg,n(p1, ..., pn) is described by ωAiry g,n (s1, ..., sn). More precisely, consider a local variable s in a neighbourhood of ai chosen so that x = TOPOLOGICAL RECURSION WITH HARD EDGES x(Pi) + 1 where 2 ǫ2s2, for ǫ > 0 a small real constant. With respect to this local coordinate y = y(Pi) + η1/2 i 7 ǫs + ... (3.3) ηi = dy(Pi)2 = Res z=Pi (dy)2 dx . ωg,n(p1, ..., pn) = ǫ6−6g−3ndy(Pi)2−2g−nωAiry The dominant asymptotic term as ǫ → 0 is (3.4) where si = s(pi) is the local coordinate s evaluated at the point pi ∈ Σ. The analogous behaviour near an irregular zero Pi of dx where the spectral curve resembles the Bessel curve xy2 = 1 2 also holds. With respect to the local variable s defined above we now have y = η1/2 g,n (s1, ..., sn) + O(ǫ7−6g−3n) s−1 + ... for i (3.5) ηi = (ydx)(Pi)2 = Res z=Pi y2dx. The dominant asymptotic term as ǫ → 0 is (3.6) ωg,n(p1, ..., pn) = ǫ2−2g−n(ydx)(Pi)2−2g−nωBessel g,n (s1, ..., sn) + O(ǫ3−2g−n). Collect the top order pole parts of ωg,n(p1, ..., pn) at each Pi, which have order 6g − 6 + 4n by (3.4), respec- tively 2g − 2 + 2n by (3.6), into a single correlator ωtop g,n (p1, ..., pn) using a local spectral curve S0 built out of S as follows. Given a spectral curve S = (Σ, B, x, y), define the local spectral curve S0 = (Σ, B0, x, y) with B0 the trivial Bergman kernel -- it is given in a local coordinate s around 2 s2 by B0(p, p′) = ds(p)ds(p′) top any zero Pi of dx defined by x = x(Pi) + 1 g,n (p1, ..., pn) of S0 consist of the top order pole parts of ωg,n(p1, ..., pn) at each zero of dx. g,n (p1, ..., pn). Alternatively define ωtop (s(p)−s(p′))2 . The correlators ω In preparation for studying the full formula (1.3), we restate (3.4) and (3.6) by (3.8) below in terms of the partition functions ZKW and ZBGW by replacing the R operator in (1.3) with the identity operator. Recall that a two-dimensional topological field theory (2D TFT) is a vector space H and a sequence of sym- metric linear maps Ig,n : H⊗n → C for integers g ≥ 0 and n > 0 satisfying the following conditions. The map I0,2 is a non-degenerate bilinear form on H i.e. it defines a metric h, with dual bivector ∆ = hαβeα ⊗ eβ (defined with respect to a basis {eα} of H). The map I0,3 together with h = I0,2 defines a product· on H via h(v1·v2, v3) = I0,3(v1, v2, v3) with identity 11 given by the dual of I0,1 = 11∗ = h(11, ·). It satisfies the natural insertion of 11 condition Ig,n+1(11 ⊗ v1 ⊗ ... ⊗ vn) = Ig,n(v1 ⊗ ... ⊗ vn) and gluing conditions Ig,n(v1 ⊗ ... ⊗ vn) = Ig−1,n+2(v1 ⊗ ... ⊗ vn ⊗ ∆) = Ig1,I+1 ⊗ Ig2,J+1(cid:0)Oi∈I vi ⊗ ∆ ⊗Oj∈J vj(cid:1) for g = g1 + g2 and I ⊔ J = {1, ..., n}. Via the natural isomorphism H0(Mg,n) ∼= C we consider Ig,n : H⊗n → H0(Mg,n) and integrate these classes next to Chern classes of the tautological line bundles Li on the moduli space of stable curves Mg,n. With respect to a basis {eνi} of H, we define the partition function: (3.7) ¯hg−1 1 Ig,n(eν1, ..., eνn) · c1(Lj)kjvkj,νj. Z(¯h, {vk,j}) = exp ∑ g,n n!ZMg,n n ∏ j=1 The partition function (3.7) for the trivial dimension 1 TFT, where Ig,n(11⊗n) = 1, stores intersection numbers of ψ classes on Mg,n and hence Z(¯h, {vk,1}) = ZKW(¯h, {vk,1}), the Kontsevich-Witten partition function, as described in Section 2. 8 LEONID CHEKHOV AND PAUL NORBURY A 2D TFT is known as semisimple if its associated Frobenius algebra (H, η,·) is semisimple, i.e. H ∼= C ⊕ C ⊕ ... ⊕ C, hei, eji = δijηi, ei · ej = δijei (j) i = δij is the standard basis. For a semisimple 2D TFT, the for some ηi ∈ C \ {0}, i = 1, ..., D where e Ig,n decompose into a sum of 1-dimensional TFTs and hence are extremely simple. A 1-dimensional TFT depends on a single complex number I0,1(11) = η ∈ C which determines Ig,n(11⊗n) = η1−g. Its partition function (3.7) is simply ZKW(η−1 ¯h, {uk,1}) since the η1−g is naturally absorbed by the ¯hg−1. The coordinate 1 2 uk,1 corresponding to an orthonormal uk,1 corresponds to the unit vector 11, and we instead use vk,1 = η basis, so ZKW(η−1 ¯h, {uk,1}) = ZKW(η−1 ¯h, {η− 1 2 vk,1}). Hence the partition function of a semisimple 2D TFT, where Ig,n vanishes on mixed monomials in the ei and Ig,n(e⊗n , is given by a product ) = η1−g i i Z(¯h, {vk,j}) = ZKW(η−1 1 ¯h, {η− 1 2 1 vk,1}) · · · ZKW(η−1 D ¯h, {η− 1 D vk,D}). 2 For storing the numbers Ig,n(eν1, ..., eνn) this may seem more complicated than necessary, however it brings useful insight to the formula (1.3), and is precisely necessary to store the highest order parts ωtop g,n (p1, ..., pn) of the correlators ωg,n of a spectral curve S = (Σ, B, x, y). A generalisation of this idea couples a TFT to ZBGW(¯h, {vk,1}) via ZBGW(η−1¯h, {η− 1 2 uk,1}). Thus ZS0 (¯h, {vk,i}) = ∆ZBGW(¯h, {vk,1}) · · · ZBGW(¯h, {vk,m})ZKW(¯h, {vk,m+1}) · · · ZKW(¯h, {vk,D}) (3.8) where S, hence S0, has k irregular zeros of dx at which y has simple poles, and D − m regular zeros, ∆(¯h) = η−1 i ¯h in the ith factor and ∆(vk,i) = η− 1 vk,i. 2 i The expressions (3.4) and (3.6), hence also (3.8), depend only on the local behaviour of x and y in a neigh- bourhood of Pi, and are independent of B. The formula (1.3) strengthens this result to show that all lower asymptotic terms of ωg,n, hence ωg,n itself, can also be obtained from ωAiry by also using B. g,n and ωBessel g,n 4. GIVENTAL DECOMPOSITION 4.1. Translations. The translation term T in (1.3) translates the arguments vk,α in the tau functions: ZKW(¯h, {vk,α}) 7→ ZKW(¯h, {vk,α + ck,α}), ZBGW(¯h, {vk,β}) 7→ ZBGW(¯h, {vk,β + ck,β}). Translations arise from local expansions of y near each zero Pα of dx. To study these translations, we restrict ykzk. This is a deformation of the Bessel curve, to the case where dx has a single zero. Let x = 1 or a deformation of the Airy curve when y−1 = 0 and y1 6= 0. General deformations of the Airy curve were studied in [13, 16]. Propositions 4.1 and 4.2 below show that the correlators of deformations of the Airy and Bessel curves depend linearly on the coefficients ag,n and bg,n of the correlators ω ∞ ∑ k=−1 Airy g,n and ωBes 2 z2, y = g,n defined by (4.1) ωAiry g,n = ∑ µ∈Zn + ag,n(µ) n ∏ i=1 dzi zµi+1 i , ωBes g,n = ∑ µ∈Zn + bg,n(µ) n ∏ i=1 dzi zµi+1 i and homogeneously in the coefficients yk of y. We will begin with the statement of the known result for deformations of the Airy curve. Proposition 4.1 ([13, 15, 16]). For x = 1 2 z2, y = ∞ ∑ k=1 ykzk, B = dzdz′ (4.2) ωg,n(z1, ..., zn) = y2−2g−n 1 ∞ ∑ m=0 n ∏ i=1 ∑ ~d,~α dzi zdi+1 i (z−z′)2 and 2g − 2 + n > 0, k=1(cid:18) yαk−2 ag,n+m(~d,~α) m ∏ y1 (−1)m m! 1 αk(cid:19) TOPOLOGICAL RECURSION WITH HARD EDGES 9 The statements of Proposition 4.1 in the literature, such as Lemma 3.5 in [13], use intersection numbers on where ~d = (d1, ..., dn), ~α = (α1, ..., αm) with αk > 3, ag,n is defined in (4.1) and the product Note that ag,n+m(~d,~α) = 0 when any di or αi is even or when ~d + ~α 6= 6g − 6 + 3n so the sum is finite. the moduli space of stable curves related via ag,n(2m1 + 1, ..., 2mn + 1) =RMg,n (z−z′)2 and 2g − 2 + n > 0, Proposition 4.2. For x = 1 ykzk, B = dzdz′ 2 z2, y = ∏n k=1(2mk + 1)!!ψmk k . (·) is 1 when m = 0. ∞ ∑ k=−1 m ∏ k=1 (4.3) ωg,n(z1, ..., zn) = y2−2g−n −1 g−1 ∑ m=0 n ∏ i=1 ∑ ~d,~α dzi zdi+1 i (−1)m m! bg,n+m(~d,~α) m ∏ k=1(cid:18) yαk−2 y−1 1 αk(cid:19) where ~d = (d1, ..., dn), ~α = (α1, ..., αm) with αk Note that bg,n+m(~d,~α) = 0 when any di or αi is even or when ~d + ~α 6= 2g − 2 + n so the sum is finite. > 1, bg,n is defined in (4.1) and the product (·) is 1 when m = 0. m ∏ k=1 Proof. The dependence of the kernel K(z1, z) on y2k−1 in the recursion (1.2) is given by K(z1, z) = 1 2 R z −z B(z′, z1) (y(z) − y(−z))dx(z) = 1 y−1 2(z2 1 − z2)(1 + zdz1 ∞ ∑ k=1 = 2(z2 1 − z2) dz1 ∞ y2k−1z2k−1dz ∑ k=0 = 1 y−1 zdz1 1 − z2)dz 2(z2 z2k)dz y2k−1 y−1 (−1)m ∞ ∑ k=1 ∞ ∑ m=0 z2k!m . y2k−1 y−1 Thus each summand contributes a term of homogeneous degree m in the y2k−1 for k ≥ 1 to K(z1, z). Thus we can write ωg,n as ωg,n(z1, z2, .., zn) = Res z=0 K(z1, z)(cid:20)ωg−1,n+1(z, −z, z2, .., zn) + ◦ ∑ ωg1,I+1(z, zI) ωg2,J+1(−z, zJ)(cid:21) g1+g2=g I⊔J={2,..,n} = y2−2g−n −1 g−1 ∑ m=0 n ∏ i=1 ∑ ~d,~α dzi zdi+1 i (−1)m m ∏ k=1 yαk y−1 Cg,n+m(~d,~α) for some constants Cg,n+m(~d,~α) which we will show coincide with coefficients 1 g,n (for ~α + 2 = (α1 + 2, ..., αk + 2)). The dependence on only odd αk is realised by the same property of bg,n+m(~d,~α). It is easy to prove by induction that ωg,n is a polynomial in the yk for odd positive k. m! bg,n+m(~d,~α + 2) of ωBes A variational formula for a family of spectral curves depending on a parameter was proven in [9, 18]. When the variation of ydx is given by integration of B(p1, p2) around a generalised cycle ∂ ∂yk (−ydx) = −zk+1dz = − 1 k + 2 d(zk+2) = − Res z0=z 1 k + 2 zk+2 0 B(z0, z) = Res z0=∞ 1 k + 2 zk+2 0 B(z0, z) it determines the variation of ωg,n to be integration of ωg,n+1 around the same generalised cycle ∂ ∂yk ωg,n(z1, ..., zn) = Res z0=∞ 1 k + 2 zk+2 0 ωg,n+1(z0, z1, ..., zn). 10 LEONID CHEKHOV AND PAUL NORBURY Write [yiyj...yk] f for the coefficient of yiyj...yk in a polynomial f of the variables yi. " m ∏ k=1 yαk# ωg,n = 1 Aut(α) ∂yα1 ∂m ...∂yαm ωg,n(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)yi=0 m ∏ k=1 m ∏ k=1 1 1 Aut(α) 1 = = Aut(α) dzi zdi+1 i n ∏ i=1 = ∑ ~d Res zn+1=∞ ... Res zn+m=∞ Res zn+1=∞ ... Res zn+m=∞ (−1)m Aut(α) m ∏ k=1 αk + 2 zαk+2 n+k ωg,n+m(z1, ..., zn+m)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)yi=0 g,n+m(z1, ..., zn+m) zαk+2 n+k ωBes 1 αk + 2 1 αk + 2 bg,n(d1, ..., dn, α1 + 2, ..., αm + 2) . Here Aut(α) = d1!...dj! is the order of the automorphism group of since ωBes g,n = ∑ µ∈Zn + bg,n(µ) n ∏ i=1 d1 z } { dzi zµi+1 i dj z } { α1, .., α1, ..., α = (α1, ..., αm) = ( we replace 1/Aut(α) with 1/m!, and we shift each αk to αk − 2 to get (4.3) as required. It may be helpful to read (4.3) via small genus examples: αj, .., αj) which contains j ≤ m distinct αk. After summing over all m-tuples, (cid:3) ω1,n = y−n −1 n ∏ i=1 dzi z2 i ω2,n = y−n−2 −1 = y−n−2 −1 ω3,n = y−n−4 −1 −"∑ j = y−n−4 −1 −"∑ j n ∏ i=1 dzi z2 i · b1,n(1, 1, ..., 1) 1 8 dzi z2 · 1 z2 j ∑ j=1 9 256 (n − 1)! = y−n −1 i n i n i (cid:16) n ∑ j=1 ∑ j=1 1 z4 j 1 z2 j 8192 75 dzi z2 dzi z2 n ∏ i=1 n ∏ i=1 n ∏ i=1 (n + 1)! − b2,n(3, 1, ..., 1)− (n + 3)! + ∑ i,j (n + 2)!! b2,n+1(1, ..., 1, 3)! 189 20480 (n + 3)! 3 256 y1 y−1 y1 y−1 1 z2 i z2 j 1 3 1 z2 j n ∏ i=1 y1 y−1 dzi z2 i (cid:16) n ∑ j=1 63 20480 (n + 4)! + 15 8192 y3 y−1 (n + 4)!# + y2 1 y2 −1 21 40960 (n + 5)!(cid:17) 1 z4 j b3,n(5, 1, ..., 1) + ∑ i,j 1 z2 i z2 j b3,n(3, 3, 1, ..., 1) b3,n+1(1, 1, ..., 1, 5)# + 1 z2 j y1 y−1 1 3 b3,n+1(3, 1, ..., 1, 3) + 1 5 y3 y−1 (−1)2 2! y2 1 y2 −1 1 32 b3,n+2(1, ..., 1, 3, 3)(cid:17) Note that we place the arguments of the symmetric function bg,n at different ends to emphasise their origin. 4.2. Graphical expansion. In this section we generalise the weighted graphical expansion of the correlators ωg,n for regular spectral curves proven in [13, 15] to allow for irregular spectral curves. Given a set {1, ..., .D} which will correspond to the zeros of dx on a spectral curve consider the following set of decorated graphs. Definition 4.3. For a graph γ denote by V(γ), E(γ), H(γ), L(γ) = L∗(γ) ⊔ L•(γ) its set of vertices, edges, half-edges and leaves. The disjoint splitting of L(γ) into ordinary leaves, L∗, and dilaton leaves, L•, is part of the structure on γ. The set of half-edges consists of leaves and oriented TOPOLOGICAL RECURSION WITH HARD EDGES 11 edges so there is an injective map L(γ) → H(γ) and a multiply-defined map E(γ) → H(γ) denoted by E(γ) ∋ e 7→ {e+, e−} ⊂ H(γ). The map sending a half-edge to its vertex is given by v : H(γ) → V(γ). Decorate γ by functions: g : V(γ) → N α : V(γ) → {1, ..., D} p : L∗(γ) ∼=→ {p1, p2, ..., pn} ⊂ Σ k : H(γ) → N such that kL•(γ) The genus of γ is g(γ) = b1(γ) + ∑ > 1 and n = L∗(γ). We write gv = g(v), αv = α(v), αℓ = α(v(ℓ)), pℓ = p(ℓ), kℓ = k(ℓ). g(v) and γ is stable if any vertex labeled by g = 0 is of valency ≥ 3. We write nv for the valency of the vertex v. For 2g − 2 + n, define Γg,n to be the set of all stable connected genus g decorated graphs with n ordinary leaves. v∈V(γ) g,n or ωBes We now express the correlators of a spectral curve as a sum over decorated graphs, with vertices weighted by coefficients of ωAiry g,n , edges weighted by coefficients of local expansions of B, ordinary leaves weighted by differentials determined by (Σ, B, x), and dilaton leaves weighted by coefficients of local ex- pansions of y. We follow the exposition in [13] for the regular case and generalise it to the irregular case. A spectral curve S = (Σ, B, x, y) defines a disjoint splitting V(γ) = Vreg(γ) ⊔ Virreg(γ) of V(γ) into regular and irregular vertices. Label the zeros of dx by P1, ..., PD and define a vertex v of γ to be irregular if y has a pole at Pαv (v). The spectral curve S also defines weights on any decorated graph γ ∈ Γg,n which will be used to produce correlators ωg,n of S as weighted sums over all decorations on graphs of type (g, n). The weights are defined as follows. Definition 4.4. Vertex weights. W(v) =( ¯hg−1y2−2gv−nv ¯hg−1y2−2gv−nv 1,αv −1,αv agv,nv ({kh h ∈ H(v)}), bgv,nv ({kh h ∈ H(v)}), v ∈ Vreg v ∈ Virreg where y1,αv = dy(Pαv), y−1,αv = (ydx)(Pαv) and ag,n(k1, ..., kn) and bg,n(k1, ..., kn) -- defined in (4.1) -- are symmetric functions of ki so it makes sense to take in a set of cardinality n. Edge weights. where with respect to the local coordinates z, z′ defined by x = 1 W(e) = ¯h(2ke+ − 1)!!(2ke− − 1)!!B αe+ ,αe− 2ke+,2ke− Bα,α′ (z, z′) = δα,α′ dzdz′ (z − z′)2 + ∑ Bα,α′ 2 z2 + x(Pα) and x = 1 m,m′ zmz′m′ dzdz′. 2 z′2 + x(Pα′ ) Here {e+, e−} are different orientations of the edge e, and by the symmetry Bα,α′ αe+ ,αe− B 2ke+ ,2ke− depends only on the (unoriented) edge e. m,m′ = Bα′,α m′,m the weight Ordinary leaf weights W(ℓ) = Vαℓ kℓ (pℓ) The dependence of ωg,n on pi ∈ Σ occurs via Vα Dilaton leaf weights k (p) which is defined in (3.1). W(λ) = 1 y1,αλ 1 y−1,αλ (2kλ + 1)!!y2kλ+1,αλ, (2kλ + 1)!!y2kλ+1,αλ, vλ ∈ Vreg vλ ∈ Virreg 12 LEONID CHEKHOV AND PAUL NORBURY where y2k+1,α are the odd coefficients of the local expansion y = ∑ yk,αzk with respect to the local coordinate z defined by x = 1 2 z2 + x(Pα). The recursive structure (1.2) can be encoded in graphs [18] with edges decorated by ω0,1(p), ω0,2(p, p′) and K(p, p′) obtained by applying (1.2) repeatedly which corresponds to the construction of a surface of type (g, n) via recursively attaching 2g − 2 + n pairs of pants. This enables one to express ωg,n as a weighted sum over graphs with 2g − 2 + n vertices. Theorem 3. For a spectral curve (Σ, B, x, y) and 2g − 2 + n > 0, ωg,n(p1, ..., pn) = 1 n! ∑ γ∈Γg,n ∏ v∈V(γ) W(v) ∏ e∈E(γ) W(e) ∏ W(ℓ) ∏ W(λ) ℓ∈L∗(γ) λ∈L•(γ) with weights W defined in Definition 4.4. Proof. The regular case of this theorem is proven in [13] and [15]. Graphs arise in the expression for ωg,n, as originally described in [18], as a means of encoding occurrences of the kernels K(z1, z2) and B(z1, z2) in formulae for the correlators obtained by iterating the recursion 1.2. Importantly the contributions to graphs by B(p, p′) are independent of the contributions to graphs by y. Hence the proof of Theorem 3 in [13] adapts immediately to allow irregular vertices, with changes only to vertex weights, and corresponding dilaton leaf weights, for vertices corresponding to irregular zeros of dx. The irregular vertex weights are determined from Proposition 4.2. (cid:3) Remark 4.5. Although the main theorem in [13] requires a finiteness assumption on the spectral curve, Theorem 3 is proven there in full generality for any regular spectral curve. The theorem is also proven for any regular spectral curve in [15] Proposition 4.1, where the graphical sum is replaced by a moduli space of D-coloured stable Riemann surfaces. The graphical expansion in Theorem 3 of the correlators ωg,n of S can be restated in terms of differential operators acting on the partition function ZS leading to a proof of the main result Theorem 1. The function R(z) and operator R below are obtained from [13]. Definition 4.6. Given a spectral curve S = (Σ, B, x, y) define R(z) = ∑ Rkzk ∈ End(V)[[z]] for a dimension D(= number of zeros of dx) vector space V by hR−1(z)iα β √z √2πZΓβ = − (x(Pβ)−x(p)) z B(Pα, p) · e where Γβ is a path of steepest descent of −x(p)/z containing β, and define the sequence rk ∈ End(V) by R(z) = exp( ∑ ℓ>0 rℓzℓ). Definition 4.7. Given a spectral curve S = (Σ, B, x, y) define ∂ ∑ k=0 ∑ ∑ ℓ=1 vk,β(rk)α β α,β ∞ R = exp( ∞ T = exp − ∂ ∂vk,α − ∆ acts on the αth factor ZBGW by ¯h 7→= y−2 y2k−1,α y−1,α (2k − 1)!! m ∑ α=1 ∂vk+ℓ,α + ¯h 2 ℓ−1 ∑ m=0 ∑ k>0 ∂2 ∂vm,α∂vℓ−m−1,β!) ∂vk,α! ∂ (−1)m+1(rℓ)α β D ∑ α=m+1 ∑ k>1 y2k−1,α (2k − 1)!! y1,α 1,α ¯h, for ZKW) −1,α¯h (or y−2 We have assumed that y is irregular at Pα for α = 1, ..., m and regular otherwise. Here rk is defined in Definition 4.6 and y2k−1,α are the odd coefficients of the local expansion y = ∑ yj,αzj with respect to the local coordinate z defined by x = 1 2 z2 + x(Pα). TOPOLOGICAL RECURSION WITH HARD EDGES 13 The following restatement of Propositions 4.1 and 4.2 is a special case of Theorem 1 and in fact is used in the proof of Theorem 1. Proposition 4.8. The partition function of the spectral curve S = (Σ, B, x, y) = (P1, dzdz′ (z − z′)2 , 1 2 z2, is obtained via translation of the appropriate tau function: ykzk) ∞ ∑ k=−1 or equivalently ZS = T = exp − ∑ (4.4) for k>ǫ ZBGW(y−2 ZKW(y−2 −1¯h, {y−1 1 ¯h, {y−1 −1v0,1, y−1 −1v0,1, y−1 −1vk,1 − (2k − 1)!! y2k−1 y−1 −1v1,1, y−1 1 vk,1 − (2k − 1)!! y2k−1 , k > 0}), y1 , k > 1}), y−1 6= 0 y−1 = 0, y1 6= 0. or ZS = T ∆ZBGW(¯h, {vk,1}) ∂vk,1! , ∆(¯h, {vk,1}) = (y−2 y2k−1 ymin (2k − 1)!! T ∆ZKW(¯h, {vk,1}) min¯h, {y−1 minvk,1}), (ymin, ǫ) = (y−1, 0) or (y1, 1). weight of a dilaton leaf with label i is (2i − 1)!! y2i−1 Proof. Propositions 4.1 and 4.2 can be expressed graphically by adding dilaton leaves to each vertex. The at an irregular vertex. But this is also the graphical realisation of translation given by the exponential of a constant vector field. (cid:3) at a regular vertex and (2i − 1)!! y2i−1 y−1 y1 Remark 4.9. The regular case of Proposition 4.8 is a consequence of work of Manin and Zograf [27] and Eynard [16] as follows. Consider FKW(¯h, t0, t1, ...) = ∑ g,n ¯hg−1 1 n! ∑ ~k∈NnZMg,n n ∏ i=1 ψki i tki and define a generating function for higher Weil-Petersson volumes Fκ(¯h,~t,~s) = ∑ g,n ¯hg−1 1 n! ∑ ~k∈NnZMg,n n ∏ i=1 ψki i tki ∞ ∏ j=1 mj j κ s mj j mj! . Manin and Zograf [27] proved that Fκ is a translation of FKW Fκ(¯h,~t,~s) = FKW(¯h, t0, t1, t2 + p1(~s), ..., tk + pk−1(~s), ...) where the pj are (weighted) homogeneous polynomials of degree j defined by 1 − exp − sizi! = ∞ ∑ i=1 ∞ ∑ j=1 pj(s1, ..., sj)zj. Eynard [16] proved the same relation between higher Weil-Petersson volumes and topological recursion. Associate to y(z) = ∞ ∑ k=1 ykzk (its Laplace transform) L( f )(z) = 1 y1 ∞ ∑ k=1 (2k + 1)!!y2k+1zk = − ∞ ∑ j=1 pj(s1, ..., sj)zj which defines si as a function of y2k+1. Then for S = ( 1 words, which is the regular case of Proposition 4.8. FS = FKW(¯h, t0, t1, t2 − 3!! y3 y1 , ..., tk − (2k − 1)!! y2k−1 y1 , ...) 2 z2, y(z), dzdz′ (z−z′)2 ) he proves FS = Fκ(¯h,~t,~s). In other 14 LEONID CHEKHOV AND PAUL NORBURY Proof of Theorem 1. The graphical expansion in Theorem 3 corresponds to a Feynman expansion of the ac- tion of the exponential of a differential operator with quadratic and linear terms on a product of D functions of single variables. This viewpoint, i.e. the determination of the differential operators corresponding to the weighted graphical expressions is dealt with thoroughly in [14]. The weights give coefficients of dif- ferential operators. The factor ∏v∈V(γ) W(v) corresponds to the product ZBGW(¯h, {vk,1}) · · · ZKW(¯h, {vk,D}) of tau functions, the factor ∏e∈E(γ) W(e) corresponds to quadratric terms in the differential operator, the in the differential operator, and the factor ∏λ∈L•(γ) W(λ) factor ∏ℓ∈L∗(γ) W(ℓ) corresponds to terms tj corresponds to translations. ∂ ∂tk It is crucial here that the definition of R depends only on (Σ, B, x) which is independent of y and hence also of the property of the spectral curve being regular or irregular. This allows us to partly follow the proof of the regular case in [13] since the same operator R is being used. In [14] the equality exp( ∞ ∑ k=0 ∑ ∑ ℓ=1 α,β ∞ = exp( ∞ ∑ ℓ=1 vk,β(rk)α β ∂ ∂vk+ℓ,α + α,β ∞ ∑ k=0 ∑ vk,β(rk)α β ¯h 2 ℓ−1 ∑ m=0 β (−1)m+1(rℓ)α ∂vk+ℓ,α!) exp( ¯h 2 ∂ ∑ k,ℓ≥0 ∂2 ∂vm,α∂vℓ−m−1,β!) (2k − 1)!!(2ℓ − 1)!!Bα,β 2k,2ℓ ∂2 ∂vk,α∂vℓ,β) is proven via the Campbell -- Baker -- Hausdorff formula. The general form of the Campbell -- Baker -- Hausdorff identity that we need pertains to three differential operators with constant coefficient matrices A, A′, and B (the matrices A and A′ are symmetric): A = (cid:0)~∂A~∂T(cid:1), A′ = (cid:0)~∂A′~∂T(cid:1), and B = (cid:0)~tB~∂T(cid:1) with derivatives acting to the right. In particular, [A, [A, B]] = 0. Then eB+A = eB eA′ provided BTA′ + A′B = A − e−BT A e−B, or equivalently A′ =Z 1 0 dx e−xBT A e−xB. Via the action of (constant coefficient) quadratic differential operators expressed graphically (also described well in [14]) this gives precisely the correct edge weights yielding the operator R in (1.3). The translation term T in (1.3) follows immediately from Propositions 4.1, 4.2 and 4.8. This is a modification to the proof in [13] in two ways. Firstly, the proof in [13] uses an operator R0 = R T0 where T0 is the translation arising out of a very special choice of y -- see below for a discussion of this point. Secondly, here we needed Proposition 4.2 which resulted in terms given by translations of the tau function ZBGW in addition to translations of the tau function ZKW. (cid:3) Note that related decompositions for matrix models appears in [1]. Example 4.10. The four decompositions of partition functions 1.1 follow from Theorem 1 applied to the following spectral curves. Each of the spectral curves is rational with common x and B given by the Cauchy kernel: x = z + , B = 1 z dzdz′ (z − z′)2 . In particular, the operator R is the same in all four examples. The examples differ by their choice of y: ZGW uses y = ln z -- [13, 30]. ZGUE uses y = z -- [17]. ZLeg uses y = z ZDes uses y = z z2−1 -- Section 5. z+1 -- [10]. Theorem 1 generalises a result for regular spectral curves in [13] and its proof is modeled on that result. We end this section with a comparison of the two decomposition formulae applied to regular spectral curves. TOPOLOGICAL RECURSION WITH HARD EDGES 15 Definition 4.11. Given a regular spectral curve S = (Σ, B, x, y) define Yα locally in a neighbourhood of a zero Pα of dx by the property that it satisfies √ζ √2πZΓα dYα(p)e(x(p)−x(Pα))ζ = D ∑ β=1 dy(Pβ) · −1 √2πζ ZΓα B(p, Pα)e(x(p)−x(Pα))ζ. The function Yα is well-defined only up to addition of a function of x, because the Laplace transform annihi- lates functions of x, but this ambiguity disappears in the odd coefficients of the local expansion of Yα which is all that is needed in the sequel. The construction of [13] begins with Givental's decomposition of the partition function of (the correlators of) a cohomological field theory (with flat identity) and produces a spectral curve satisfying the restriction that y = Yα for each α = 1, ..., D (where y = Yα up to local functions of x). Theorem 4 ([13]). Given a regular spectral curve S = (Σ, B, x, y) satisfying y = Yα for each α = 1, ..., D the partition function ZS satisfies the decomposition (1.3). The statement of this theorem is the reverse of the result in [13], but it is easily seen to be reversible on spectral curves satisfying the condition on y, [12]. The translation due to local expansions of y is implicit in the statement in [13] and appears inside an operator R0 = exp(−(rm)α 11 ∂ ∂vm+1,α + ∞ ∑ k=0 vk,α(rm)β α ∂ ∂vm+k,β + ¯h 2 m−1 ∑ i=0 (−1)i+1(rm)α,β ∂vi,α∂vm−i−1,β) ∂2 (which they denote by R) where the vector 11 = {dy(P (α)}. Raising of the indices of rm is required for endo- morphisms with respect to a general basis, but in this paper we express R in a basis (known as normalised canonical coordinates [22]) with respect to which the metric is the identity, so upper and lower indices are the same. It is related to the operator here by R0 = R T for a translation T. This viewpoint allows us to interpret the decomposition (1.3) in the regular case as Givental's decomposition of the partition function of a cohomological field theory, now without the restriction of flat identity. Corollary 4.12. For any regular spectral curve S = (Σ, B, x, y), ZS is a partition function for a cohomological field theory. 5. LEGENDRE ENSEMBLE One application of Theorem 1 applied to the curve (x2 − 4)y2 = 1 is a Givental type decomposition for the partition function of the Legendre ensemble. Proposition 5.1. Consider the Legendre ensemble with trivial potential (5.1) The resolvents Wg,n(x1, ..., xn), defined via the large N asymptotic expansion of the cumulant by DM = ZHN [−2,2] x1 − M(cid:19) · · · tr(cid:18) 1 −2 1 −2 ...Z 2 −2Z 2 N!Z 2 xn − M(cid:19)(cid:29)c 1 N→∞ ∼ ∑ g≥0 (cid:28)tr(cid:18) ∏ i<j (λi − λj)2dλ1...dλN. N2−2g−nWg,n(x1, ..., xn), satisfy topological recursion for the spectral curve (5.2) S = (P1, B = dzdz′ (z − z′)2 , x = z + 1 z , y = z ). z2 − 1 In other words Wg,n(x1, ..., xn)dx1...dxn gives an expansion of ωg,n at xi = ∞. 16 LEONID CHEKHOV AND PAUL NORBURY Note that the existence of the asymptotic expansion Wn = ∑g≥0 N2−2g−nWg,n is not trivial and cannot be obtained from the loop equations only, but it is a consequence of the results of Borot and Guionnet [4]. We omit the proof of Proposition 5.1 which is rather standard, via the usual loop equations. A corollary of Theorem 1 is: ZLeg({vk,1, vk,2}) = R T ∆ZBGW({vk,1}) × ZBGW({vk,2}). We now consider this example closely and demonstrate how the decomposition (1) produces calculations via pictures realising the graphical treatment of Theorem 3. We need only consider connected graphs to produce terms F = log Z. In the calculations via pictures we instead use a related decomposition with only edge and vertex weights leading to simpler pictures. It is analogous to the decomposition proven by the first author in [6] and described next. it contains only terms of type A = (cid:0)~∂A~∂T(cid:1) and no terms of type B = (cid:0)~tB~∂T(cid:1), whose absence was 5.1. Chekhov-Givental decomposition. In [6] the first author obtained a decomposition for the Gaussian model that differs from (1.3). As in (1.3), it consists of the exponential of a quadratic differential operator acting on two copies of ZKW, with the added translation term present in (1.3) but without a rotational term, i.e. ensured by choosing global coordinates properly matching the local ones -- see below. It is equivalent to (1.3) under a linear change of coordinates between vk,α and τ±k described in (??). For the Legendre model which exhibits hard-edges, we present here an analogous decomposition almost identical to the Gaussian model decomposition in [6]. As in the proof of Theorem 1, two ingredients of the decomposition arise from B(p, q) -- the Bergman kernel -- and from the 1-form ydx. (i) ydx. We need ydx in the global and local models. (a) Global model: x = eλ + e−λ, y = 1/(eλ − e−λ), ydx = dλ, λ ∈ C, −π/2 ≤ Im λ < 3π/2 (b) Local models: (x1, y1) and (x2, y2) correspond to two branch points b1 : λ = 0 and b2 : λ = iπ. 1x1 = 1/4, x1 = λ2, y1dx1 = dλ for λ ∈ D1(0) a disk of radius 1 around 0 ∈ C; y2 2x2 = 1/4, x2 = (λ − iπ)2, y2dx2 = dλ for λ ∈ D1(iπ) a disk of radius 1 around iπ ∈ C. y2 So, the global and local 1-forms coincide, ydxglobal = ydxlocal = dλ, which means that the linear differential part is absent in the Givental decomposition in this case. (ii) B(p(λ), q(µ)). We again have global and local B's: B(p, q)global = B(p, q)local = deλdeµ (eλ − eµ)2 , dλdµ (λ − µ)2 , defined in the corresponding domains −π/2 ≤ Im {λ, µ} < 3π/2 and λ, µ ∈ D1(0) for the first local model and λ, µ ∈ D1(iπ) for the second local model, and their corresponding primitives E(p(λ), q(µ))global = log(eλ − eµ), E(p(λ), q(µ))local = log(λ − µ) with the same domains of definition as for B's. We then have that, for {λ, µ} ⊂ D1(0) ∪ D1(iπ), so that these points are in the vicinities of the corresponding branching points bµ and bλ, which can be either b1 or b2, 1 2πiZCp(λ) 1 2πiZCq(µ) 2πiZCp(λ) 1 1 2πiZCq(µ) B(p(λ), q(µ))global = B(p(λ), η)localE(η, ρ)globalB(ρ, q(µ))local = δbλ,bµ B(p(λ), q(µ))local + B(p(λ), η)local(E(η, µ)global − δbλ,bµ E(η, µ)local)B(q(µ), η)local. In this expression, the integration contours Cp(λ) and Cq(µ) are inside the corresponding discs D1(bλ) and D1(bµ) and, if they are in the same disc, they must separate the points p(λ) and q(µ), the first term in the right hand side describes the propagator of the local model, and the second term corresponds to the quadratic differential operator, which we consider in details below. TOPOLOGICAL RECURSION WITH HARD EDGES 17 The local times are t−k = 1 times are λ2k+1 for λ ∈ D1(0) and t+ k = 1 (λ−iπ)2k+1 for λ ∈ D1(iπ); the corresponding global (5.3) τ∓k = 1 (2k)! ∂2k ∂λ2k 1 . ±eλ − 1 The times τ±k are related by a linear change of coordinates to the times vα k used in Theorem 1 as follows. The global meromorphic differentials dτ±k (λ), defined on C, have poles only at eλ = ±1 and are skew-invariant under the involution λ 7→ −λ. Hence they lie in the vector space spanned by the differentials Vα j (p(λ)) defined in (3.1) and dτ±k (λ) = ∑j,α c±,j j (p(λ)). This defines the linear relation between the times: α,k Vα (5.4) τ±k = ∑ j,α c±,j α,k vα j . The correlation function expansion is W(. . . , xj, . . . ) ∼ ∑∞ h. . . tr (Hr) . . . i = r ∂ ∂tr h. . .i r=1 tr Hrx−r−1 j and recalling that we obtain that the expansion of the local correlation function in the local coordinates has the form W(. . . , xj, . . . ) = ∞ ∑ r=1 rx−r−1 j ∂ ∂tr . When bλ = bµ, the coefficient expansion of the quadratic differential operator stems from that of the function E(η, µ)global − E(η, µ)local = log eη − eµ η − µ = η + µ 2 + ∞ ∑ m=2 Bm (η − µ)m m · m! , where Bm are the Bernoulli numbers (recall that B2s+1 = 0 for s ≥ 1), and with the accounting of the skew- symmetrisation under the transformations η ↔ −η and µ ↔ −µ, we obtain the expansion (5.5) ∞ ∑ m=0 − B2m+2 (2m + 2) m ∑ k=0 η2k+1 (2k + 1)! µ2(m−k)+1 (2(m − k) + 1)! with only odd powers of µ and η, so the corresponding differential operator, after evaluating the residues in η and µ, will contain derivatives only in odd times, as expected. This operator has the following form (where we recall that in the final expression we have to substitute local times by the global ones): (5.6) A±,± = − 1 2 ∞ ∑ m=0 B2m+2 (2m + 2) m ∑ k=0 (2k + 1)∂/∂τ±k (2k + 1)! (2(m − k) + 1)∂/∂τ±m−k (2(m − k) + 1)! . For different branch points, E(η, µ)global = log(eη − eµ−iπ) = log(eη + eµ) and the corresponding expansion is log(eη + eµ) = η + µ 2 + ∞ ∑ m=2 Bm m (1 − 2m) (η − µ)m m! so after the skew-symmetrisation and substitution of differential operators in times, we obtain (5.7) A+,− = − ∞ ∑ m=0 B2m+2 (2m + 2) (1 − 22m+2) m ∑ k=0 (2k + 1)∂/∂τ+ k (2k + 1)! (2(m − k) + 1)∂/∂τ−m−k (2(m − k) + 1)! . The differential operators A±,± and A+,− exactly coincide with the corresponding operators from [6] where the analogous transformation was first derived in application to the Gaussian model. Theorem 5. Consider the free energy of the Legendre ensemble on the interval [−2, 2] with the potential V(H) = ∑∞ k Hk where the times tk are expressed using the Miwa-type transform tk = ∑i(eλi + e−λi )−k as λ → +∞. k=1 tk 18 LEONID CHEKHOV AND PAUL NORBURY g where the term FLeg The free energy of this model has the 1/N expansion of the form FLeg = ∑∞ quadratic polynomial in tk, and this term is to be interpreted as the normalisation. We then have the exact relation g=0 N2−2gFLeg 0 is a Leg eFLeg−N2 F 0 = eA+,++A−,−+A+,− ZBGW(τ+(λ))ZBGW(τ−(λ)), (5.8) where ZBGW(τ±(λ)) are partition functions of the (local) BGW model (exponentials of tau-functions of the KdV hierarchy) expressed in terms of times (5.3) and A·,· are quadratic differential operators (5.6) and (5.7). Remark 5.2. Expressing the free energy FLeg in terms of times tk always results in infinite-term expan- sions for any model even if we restrict ourselves to finite-order monomials corresponding to finite-order correlation functions Wg(x1, . . . , xn in the corresponding free theory. It is however well-known (see, e.g., ACNP1) that these expressions are comprised of a finite numbers of terms if we express them in variables k := (eλ − e−λ)−2k−1 and t1 t0 k , α = 0, 1, are finite linear combinations of τ±k . k := (eλ + e−λ)(eλ − e−λ)−2k−1, k = 0, 1, . . . . It is easy to see that all tα We demonstrate Theorem 5 by calculating low genus examples. Unlike the Gaussian case, we see that the Leg free energy F is in some sense finite. It is a finite sum of rational functions apparent in the exact formulae g below. Example 5.3. In genus 1, the only contributing graphs have no edges. FLeg 1 = ∑ ± 1± = 1 8 (1 − log(1 − τ+ 0 )) + 1 8 (1 − log(1 − τ−0 )) Example 5.4. We now consider the decomposition for the F2 term of the free energy. Leg 2 = ∑ F ± 2± + ∑ ± 1± A0,0 ±,± 1± + 1+ A0,0 +,− 1− + ∑ ± 1± A0,0 ±,± Here 2± = 3τ±1 256(1 − τ±0 )3 , 1± = Leg ∂F 1 ∂τ±0 = 1 8(1 − τ±0 ) , 1± = Leg ∂2F 1 ∂(τ±0 )2 = 1 8(1 − τ±0 )2 , Ak,l ±,± = − 1 2 · B2(k+l+1) 2(k + l + 1)(2k)!(2l)! Recalling that B2 = 1/6, we have that A0,0 Leg and fourth terms in the graphical expansion for F 2 (1 − τ±0 )2 and the third term of the same graphical expansion yields (cid:16)− 1 82 · 24 − B2(k+l+1) (22(k+l+1) − 1). 2(k + l + 1)(2k)!(2l)! +,− = −1/4, so the combination of the second , Ak,l +,− = − ±,± = −1/24 and A0,0 8 · 24(cid:17) yield 1 1 = − 3 512 1 (1 − τ±0 )2 − 1 82 · 4 1 0 )(1 − τ−0 ) (1 − τ+ = − 1 256 1 0 )(1 − τ−0 ) (1 − τ+ . TOPOLOGICAL RECURSION WITH HARD EDGES 19 Hence Leg 2 = F 3τ±1 256(1 − τ±0 )3 − 3 512 1 (1 − τ±0 )2 − 1 256 1 0 )(1 − τ−0 ) (1 − τ+ . Example 5.5. We now consider the decomposition for F3 term of the free energy. Leg 3 = ∑ F ± 3± 2 + ∑ ± 2± A0,0 ±,± 1± +1 2± A0,0 +,− 1∓ 1 + ∑ ± 2± A0,0 ±,± + ∑ ± 2 × 2 2± 1 A1,0 ±,± 1± +1 × 2 2± 1 A1,0 ±,∓ 1∓ + ∑ ± 2 × 2 1 2± A1,0 ±,± A0,0 ±,± A0,0 ±,± A0,0 ±,± 1/2 + ∑ ± 1± 2 + ∑ ± A0,0 ±,± 1± 1± 1 + ∑ ± A0,0 ±,± A0,0 ±,∓ 1∓ 1± 1 + ∑ ± 1± A0,0 ±,± A0,0 ±,± 1± +1/2 1+ A0,0 +,− A0,0 +,− 1− 2 + ∑ ± 1± A0,0 ±,± A0,0 ±,± 1± 1± 2 + ∑ ± 1± A0,0 ±,± 1± A0,0 ±,∓ 1∓ 1/2 + ∑ ± 1± A0,0 ±,∓ A0,0 ∓,± 1± 1∓ Here 2± 1 = 3 256(1 − τ±0 )3 , 1 2± = 3 · 3 256(1 − τ±0 )4 , 2± = 3 · 3 · 4τ±1 256(1 − τ±0 )5 , and 1± k = Leg ∂k F 1 ∂(τ±0 )k = (k − 1)! 8(1 − τ±0 )k , Ak,l ±,± = − 1 2 · B2(k+l+1) 2(k + l + 1)(2k)!(2l)! , Ak,l +,− = − B2(k+l+1) 2(k + l + 1)(2k)!(2l)! (22(k+l+1) − 1). = C1τ2/(1 − τ0)6 + C2(τ1)2/(1 − τ0)7. Here τ±2 is obtained by a Laplace transform from a "pure" FBGW 3 state corresponding to the monomial b4(∓1)b. Recalling that B2 = 1/6 and B4 = −1/30, we have that ±,± = −1/24, A0,0 A0,0 ±,∓ = 1/16. so we fix everything indicating in gray the symmetry factors of diagrams. All derivatives in the times τ±1 are to be multiplied by factors of two in order to obtain F3. ±,± = 1/(30 · 16), A1,0 +,− = −1/4, A1,0 20 LEONID CHEKHOV AND PAUL NORBURY The terms in the first line are FBGW 3 (τ+) + FBGW 3 (τ−) − 2 · 3 · 3 · τ±1 24 · 256 · 8(1 − τ±0 )5 − 3 · 3 · τ±1 4 · 256 · 8(1 − τ±0 )4(1 − τ∓0 ) − 3 · 3 · 4 · τ±1 24 · 256(1 − τ±0 )5 and all other terms depend only on zeroth times and are a linear combination of (1 − τ+ (1 − τ+ cients. 0 )−4 + (1 − τ−0 )−4, 0 )−2(1 − τ−0 )−2 with necessarily positive coeffi- 0 )−3(1 − τ−0 )−1 + (1 − τ−0 )−3(1 − τ+ 0 )−1, and (1 − τ+ APPENDIX A QUANTISATION In this section we give a brief background for the construction of R from R following Givental [21]. We first consider quantisation in finite dimensions which easily generalises to infinite dimensions. Consider the standard holomorphic form on C2N = T∗CN given by ω = ∑ dpα ∧ dqα with Darboux coor- dinates {qα, pα}. A transformation A : C2N → C2N that preserves the symplectic form is symplectic. We will consider only linear symplectic transformations which correspond to matrices A ∈ Sp(2N, C). So- called infinitesimal symplectic transformations, corresponding to elements of the Lie algebra sp(2N, C), give rise to vector fields that preserve ω, known as Hamiltonian vector fields. The 2N2 + N-dimensional space sp(2N, C) is isomorphic to the 2N2 + N-dimensional vector space of Hamiltonians spanned by pα pβ, pαqβ, qαqβ. H(p, q) = p1q2 is an example of a Hamiltonian. Quantisation of these coordinates, i.e. promotion to opera- tors pα 7→ pα and qα 7→ qα, gives an algebra defined by [ pα, qβ] = δαβ¯h, [ pα, pβ] = 0 = [ qα, qβ] so we naturally choose pα = ¯h ∂ and qα acts by multiplication by qα. Quantisation of a function, or ob- ∂qα servable, in {pα, qβ} is not unique. We can consistently define quantisation of the quadratic Hamiltonians by [pα pβ = ¯h2 ∂2 ∂qα∂qβ , [pαqβ = ¯hqβ ∂ ∂qα , dqαqβ = qαqβ. Linear combinations of these give the quantisation of infinitesimal symplectic transformations. Hence we can define the quantisation of a linear symplectic transformation A = exp a for a ∈ sp(2N, C) to be A = exp a. This construction generalises to the infinite dimensional symplectic manifold (H, Ω) = (H[z−1][[z]], Ω) for H ∼= CD, where the symplectic form Ω is defined by Ω( f (z), g(z)) = Res z=0 f (−z)g(z)dz. H+ = H[[z]] is Lagrangian with respect to Ω and (H, Ω) ∼= (T∗H+, ωcanonical). Darboux coordinates for Ω are qk,α, pk,α defined by H ∋ f (z) = ∑ k≥0 qk,αzk + ∑ k<0 pk,αz−k. Given a dimension D vector space V and a sequence of operators rm : V → V, m = 1, 2, ... such that rm(−v) = (−1)m+1rm(v) define [rmzm := ∞ ∑ k=0 vk,α(rm)β α ∂ ∂vm+k,β + ¯h 2 m−1 ∑ i=0 (−1)i+1(rm)α,β ∂2 ∂vi,α∂vm−i−1,β . Then the quantisation of R(z) = exp r(z) is given by R(z) = exp r(z). TOPOLOGICAL RECURSION WITH HARD EDGES 21 REFERENCES [1] Alexandrov, A; Moronov, A and Morozov, A. BGWM as second constitutent of complex matrix model. J. High Energy Phys. 2009 (12), (2009) 053. [2] Andersen, J., Borot, G., Chekhov, L., and Orantin, N. ABCD of topological recursion, (2017) arXiv:1703.03307 [3] Andersen, J., Chekhov, L., Norbury, P., and Penner, R. Models of discretized moduli spaces, cohomological field theories, and Gaussian means. J. Geom. Phys. 98 (2015) 312 -- 339. [4] Borot, G. and Guionnet, A. Asymptotic Expansion of β Matrix Models in the One-cut Regime. Commun. Math. Phys. 317 (2013) 447-483. [5] Br´ezin, E and Gross, D. The external field problem in the large N limit of QCD. Phys. Lett. B97 (1980) 120. [6] Chekhov, L. Matrix models tools and geometry of moduli spaces, ActaAppl.Mathematicae 48 (1997) 33 -- 90. [7] Chekhov, Leonid and Eynard, Bertrand. Hermitian matrix model free energy: Feynman graph technique for all genera. J. High Energy Phys. 2006 (3), (2006), 014. [8] Chekhov, L., Eynard, B., and Orantin, N. Free energy topological expansion for the 2-matrix model. J. High-Energy Phys. (2006) 12 053. [9] Chekhov, L., Marshakov, A., Mironov, A., and Vasiliev, D. Complex geometry of matrix models, Proc. Steklov Math. Inst. 251 (2005) 254 -- 292. [10] Do, N. and Norbury, P. Topological recursion for irregular spectral curves. Journal of the London Mathematical Society 97 (2018), 398-426. [11] Do, N. and Norbury, P. Topological recursion on the Bessel curve. Comm. Number Theory and Physics 12 (2018), 53-73. [12] Dunin-Barkowski, P; Norbury, P; Orantin, N; Popolitov, A; Shadrin, S. Dubrovin's superpotential as a global spectral curve. To appear in J. Inst. Math. Jussieu. arXiv:1509.06954 [13] Dunin-Barkowski, P.; Orantin, N.; Shadrin, S. and Spitz, L. Identification of the Givental formula with the spectral curve topological recursion procedure. Comm. Math. Phys. 328, (2014), 669 -- 700. [14] Dunin-Barkowski, P; Shadrin, S and Spitz, L. Givental graphs and inversion symmetry. Lett. Math. Phys. 103 (2013), 533 -- 557. [15] Eynard, B. Invariants of spectral curves and intersection theory of moduli spaces of complex curves. Commun. Number Theory Phys. 8 (2014), 541 -- 588. [16] Eynard, B. Recursion between Mumford volumes of moduli spaces. Annales Henri Poincar´e 12 (2011), 1431 -- 1447. [17] Eynard, B. Topological expansion for the 1-hermitian matrix model correlation functions. J. High Energy Phys. 2004, 11 (2005), 31. [18] Eynard, B. and Orantin, N. Invariants of algebraic curves and topological expansion. Commun. Number Theory Phys. 1 (2), (2007), 347 -- 452. [19] Eynard, B and Orantin, N. Topological recursion in enumerative geometry and random matrices. J. Phys. A 42, no. 29, 293001, 117 pp., (2009). [20] Eynard, B. and Orantin, N. Weil-Petersson volume of moduli spaces, Mirzakhani's recursion and matrix models. arXiv:0705.3600. [21] Givental, A. Gromov-Witten invariants and quantization of quadratic hamiltonians.Moscow Math. Journal 4, (2001), 551 -- 568. [22] Givental, A. Semisimple Frobenius structures at higher genus. Int. Math. Res. Not. (2001), 1265-1286. [23] Gross, D. and Witten, E. Possible third-order phase transition in the large-N lattice gauge theory. Phys. Rev. D21 (1980) 446. [24] Kontsevich, M. Intersection theory on the moduli space of curves and the matrix Airy function. Comm. Math. Phys. 147, (1992), 1 -- 23. [25] Kontsevich, M. and Soibelman, Y. Airy structures and symplectic geometry of topological recursion. arXiv:1701.09137 [26] Krichever, I. The τ-function of the universal Whitham hierarchy, matrix models and topological field theories. Commun. Pure Appl. Math. 47 (1992) 437 -- 457. [27] Manin, Y. and Zograf, P. Invertible cohomological field theories and Weil-Petersson volumes. Ann. Inst. Fourier (Grenoble) 50 (2000), 519 -- 535. [28] Mironov, A., Morozov, A., and Semenoff, G. Unitary matrix integrals in the framework of Generalized Kontsevich Model. I. Brezin -- Gross -- Witten model. Intl. J. Mod. Phys. A10 (1995) 2015 -- 2050. [29] Mirzakhani, M. Weil-Petersson volumes and intersection theory on the moduli space of curves. J. Amer. Math. Soc., 20 (2007), 1 -- 23. [30] Norbury, P. and Scott, N. Gromov -- Witten invariants of P1 and Eynard -- Orantin invariants. Geometry & Topology 18 (2014), 1865 -- 1910. [31] Pandharipande, R; Pixton, A. and Zvonkine, D. Relations on Mg,n via 3-spin structures. J. Amer. Math. Soc. 28 (2015), no. 1, 279 -- 309. [32] Witten, E. Two-dimensional gravity and intersection theory on moduli space. Surveys in differential geometry (Cambridge, MA, 1990), 243 -- 310, Lehigh Univ., Bethlehem, PA, 1991. STEKLOV MATHEMATICAL INSTITUTE AND LAB. PONCELET, MOSCOW, RUSSIA; NIELS BOHR INSTITUTE, COPENHAGEN, DENMARK SCHOOL OF MATHEMATICS AND STATISTICS, UNIVERSITY OF MELBOURNE, VIC 3010 AUSTRALIA E-mail address: [email protected], [email protected]
1007.2737
1
1007
2010-07-16T10:49:59
A curvature formula for the complexified index cone of a cubic form
[ "math.AG", "math.DG" ]
We study the Kaehler metric given by the logarithm of a cubic form on its complexified index cone. Under mirror symmetry, this metric should asymptotically correspond to the Weil-Petersson metric. Using the theory of special Kaehler manifolds, a proof of a curvature formula for this metric is given.
math.AG
math
A CURVATURE FORMULA FOR THE COMPLEXIFIED INDEX CONE OF A CUBIC FORM THOMAS TRENNER Abstract. We study the Kahler metric given by the logarithm of a cubic form on its complexified index cone. Under mirror symmetry, this metric should asymptotically correspond to the Weil -- Petersson metric. Using the theory of special Kahler manifolds, a proof of a curvature formula for this metric is given. Contents Introduction 1. 2. Affine special geometry 3. Projective special geometry 4. Complex moduli space as a Special Kahler Manifold 5. Complexified index cone of a cubic form as a projective Special Kahler manifold References 1 2 7 12 14 16 1. Introduction In this paper, we will prove a curvature formula for the Asymptotic Mir- ror Weil -- Petersson metric -- or AMWP metric -- on the complexified index cone of a cubic polynomial, as studied in [7]. This is motivated by classical mirror symmetry, where on the complexified Kahler cone of a Calabi-Yau threefold we can define a metric corresponding, under the mirror map, to the Weil -- Petersson metric on the complex moduli space of a mirror manifold. To obtain the AMWP metric, we neglect the contributions by instanton corrections. A curvature formula for the Weil -- Petersson metric was given in [6], where it was also claimed that a similar result should hold on the complexified Kahler cone. Further justification has been given in [7], where the precise conjecture proved in here was made ([7, p.10]) and proved for the special case of ternary cubic polynomials. As the setting in which Stro- minger proves the curvature formula for the Weil -- Petersson metric in [6] is special geometry, we will make use of the same setup here. Date: November 11, 2018. 1 2 THOMAS TRENNER In the first two sections we give the basic definitions of special geometry and state the main results that are needed to prove the curvature formula. In section 4, we quickly recall how special geometry is applied for the complex moduli space of a Calabi -- Yau threefold. The proof that the same curvature formula applies to the complexified index cone of a cubic polynomial will be given in section 5. This will include the examples of complexified Kahler cones of the Calabi -- Yau threefolds investigated in [7]. 2. Affine special geometry In this section, we give the basic definitions of affine special Kahler man- ifolds and derive a formula for the curvature tensor in terms of the pre- potential. The main reference for this section is Freed's paper on special Kahler manifolds [4], although we sometimes use slightly different notation and supply proofs where Freed omits them. Definition 2.1. Let (M, ω) be a Kahler manifold. A special Kahler struc- ture on M is a flat, torsion-free connection ∇ on T M such that (2.1) d∇I = 0 (2.2) Here d∇ : Ωp(M ; T M ) → Ωp+1(M ; T M ) denotes the associated exterior covariant derivative. ∇ω = 0 ∇ and d∇ give the de Rham complex with coefficients in T M : 0 −→ Ω0(T M ) (2.3) Flatness of ∇ is then expressed by requiring d2 Poincar´e lemma is valid for (2.3). d∇−−→ Ω1(T M ) d∇−−→ Ω2(T M ) d∇−−→ ... ∇ = 0. It implies that the Lemma 2.2. We can express the torsion-free condition of ∇ by d∇(idT M ) = 0, where we identify Γ(End(T M )) and Ω1(T M ). Proof. Recall that the torsion tensor is defined as T (X, Y ) = ∇XY − ∇Y X − [X, Y ]. For the exterior covariant derivative we compute d∇(idT M )(X, Y ) = d∇(id(Y ))X − d∇(id(X))Y − id([X, Y ]) = d∇Y (X) − d∇X(Y ) − [X, Y ] = ∇XY − ∇Y X − [X, Y ] Now choose a flat local framing {ei}, i = 1, . . . , n, for T M with dual (cid:3) framing {θi}, i = 1, . . . , n, for T ∗M . Lemma 2.3. dθi = 0. CURVATURE FOR THE COMPLEXIFIED INDEX CONE OF A CUBIC FORM 3 Proof. In local coordinates, idT M is given by Pi ei ⊗ θi. Then d∇(idT M ) = Pi ei ⊗ dθi + ∇ei ⊗ θi. As the ei are a flat T M -framing, i.e. ∇ei = 0, the result follows. (cid:3) Definition 2.4. We will call a coordinate system {xi, yj}, i, j = 1, . . . , n Darboux if dxk ∧ dyk. ω =Xk As our local coframing is exact, we have θi = dti for local coordinate functions ti. Because ∇ω = 0, we can choose the coordinate functions so that they form a Darboux coordinate system. If we cover M by local Darboux coordinate systems, the transition functions take the form (cid:18)x y(cid:19) = S(cid:18)x y(cid:19) +(cid:18)a b(cid:19) , S ∈ Sp(2n; R), a, b ∈ Rn. As giving a complex structure is equivalent to giving the decomposition of the complexified tangent bundle in holomorphic and anti-holomorphic parts, instead of equation (2.1) we can also require (2.4) d∇π(1,0) = 0, where we identify π(1,0) ∈ Hom(T M C, T M 1,0) with its dual and use Hom(T M 1,0, T M C) ∼= Ω1,0(T M C). By the Poincar´e lemma for (2.3) and d∇π(1,0) = 0 there exists a complex vector field ξ such that (2.5) ∇ξ = π(1,0). In local Darboux coordinates, ξ = 1 2 (zi ∂ ∂xi − wi ∂ ∂yi ). n Xi=1 Xi=1 n 1 2 and thus ∇ξ = π(1,0) = (dzi ⊗ ∂ ∂xi − dwi ⊗ ∂ ∂yi ). This implies Re(dzi) = dxi, Re(dwi) = −dyi, which can be seen by taking the real part of the last equation and using that the identity on T M is the sum of the two projections: id = π1,0 + π0,1. Thus Re(π(1,0)) = 1 2 id = 1 2 n ( Xi=1 (dxi ⊗ ∂ ∂xi + dyi ⊗ ∂ ∂yi ). Definition 2.5. Let (M, ω, ∇) be a special Kahler manifold. A holomorphic coordinate system {zi} is called special if ∇Re(dzi) = 0. Two special coor- dinate systems {zi}, {wj } are called conjugate if there exists a flat Darboux system {xi, yj} such that Re(dzi) = dxi, Re(dwi) = −dyi. 4 THOMAS TRENNER By [1, p.92],this extension of the Darboux coordinate system to a conju- gate pair of special holomorphic coordinate systems is unique up to purely imaginary translations. The conjugate coordinate systems are related by a change of coordinates via holomorphic functions: (2.6) dwi =Xj τijdzj everywhere vanishing. Lemma 2.6. [2, p. 757] The holomorphic 2-form Θ = Pi dwi ∧ dzi is This implies that θ =Pi widzi is closed, thus exact. Definition 2.7. The holomorphic function F such that θ = dF is called the prepotential. Let us choose the prepotential such that the following equations hold: (2.7) (2.8) wi = τij = 4∂F ∂zi 4∂2F ∂zi∂zj Then define K = 1 ¯zi). We compute i∂ ¯∂K = 2i∂ ¯∂Im( 2 Im(wi ¯zi) = 2Im( ∂F ∂zi ∂F ∂zi ∂2F ∂zi∂zj ¯zi) = 2iIm( )dzi ∧ d¯zj = − i 2 Im(τij)dzi ∧ d¯zj Note that this differs from the conventions in [4], where the Kahler po- tential is 1 2 Im( ∂F ∂zi ¯zi). On our Kahler manifold M we have two affine connections: the Levi -- Civita connection D and the flat, torsion-free connection ∇. Define AR = ∇ − D; AR ∈ Ω1(M, EndRT M ) Extending D and ∇ to the complexified tangent bundle, we obtain A and ¯A. Theorem 2.8. [2, p. 759, Thm. 3.9] A ∈ Ω1,0(Hom(T 1,0, T 0,1)) ¯A ∈ Ω0,1(Hom(T 0,1, T 1,0)) For the curvature of the Kahler metric on M we obtain the following formula: CURVATURE FOR THE COMPLEXIFIED INDEX CONE OF A CUBIC FORM 5 Theorem 2.9. RD = −(A ∧ ¯A + ¯A ∧ A) Proof. By definition, ∇ = D + AR is flat, so: (2.9) (2.10) (2.11) 0 = ∇2 = (D + AR)2 = D2 + D(AR) + AR(D) + A2 R = RD + dD(A) + A ∧ ¯A + ¯A ∧ A + A ∧ A + ¯A ∧ ¯A = RD + dD(A) + A ∧ ¯A + ¯A ∧ A, as A ∧ A and ¯A ∧ ¯A vanish. We split the covariant exterior derivative of D: dD = ∂D + ¯∂D. As D is the Levi -- Civita connection corresponding to ω, we conclude that ¯∂D = ¯∂. By decomposing the equation above into its (r, s)-components, we arrive at: (2.12) 0 = RD + A ∧ ¯A + ¯A ∧ A + ∂D( ¯A) + ¯∂(A) We can further split this up into complex linear and anti-linear pieces, which both have to vanish: (2.13) (2.14) 0 = RD + A ∧ ¯A + ¯A ∧ A 0 = ∂D( ¯A) + ¯∂(A) The complex linear part gives the equation we wanted. Also, in the last equation the two summands take values in different bundles, so we can conclude that A is holomorphic and (2.15) 0 = ∂D( ¯A) Thus we arrive at the formula for R we wanted. (cid:3) Definition 2.10. We can also measure the failure of ∇ to preserve the complex structure by the cubic form (2.16) Ξ = −ω(π(1,0), ∇π(1,0)) = −ω(∇ξ, ∇2ξ), where we use (2.5) for the last equality, which of course only holds locally. In local coordinates we can then compute that the coefficients of Ξ are given by the third partial derivatives of F: Lemma 2.11. Ξijk = 1 2 ∂ 3F ∂zi∂zj∂zk Proof. We may write π(1,0) in local coordinates as π(1,0) =Xj ∂ ∂zj ⊗ dzj . 6 THOMAS TRENNER Thus, using (2.7) and (2.8) we obtain Ξ = −ω Xi = − Xi,j,k,l,m 8Xi,j,k 2Xi,j,k = = 1 1 ∂τij ∂zk dzj ⊗ ∂ ∂zj ∂ ∂ym ), − 1 2 )  ∂τjl ∂zk dzi ⊗ ∂ ∂zi ω(cid:18) 1 2 ( ∂ ∂xi , ∇(Xj − τim dzi ⊗ dzj ⊗ dzk ∂3F ∂zi∂zj∂zk dzi ⊗ dzj ⊗ dzk. dzk ⊗ ∂ ∂yl(cid:19) ⊗ dzi ⊗ dzj To compute the components of the curvature tensor in a local coordinate system {zi}i=1,...,n, we use the formula 0 = RD + A ∧ ¯A + ¯A ∧ A . We have: (cid:3) (2.17) (2.18) (2.19) (2.20) i 2 ω =Xk,j RD = Xi,j,k,l (RD)i¯jk¯l =Xm Ξ =Xi,j,k gk¯jdzk ∧ d ¯zj (Rj ik¯l)i,jdzk ∧ d ¯zl gm¯jRm ik¯l Ξijkdzi ⊗ dzj ⊗ dzk As A ∈ Ω1,0(Hom(T 1,0, T 0,1)), we can expand it in a local framing for the cotangent bundle: (2.21) (2.22) Aidzi, Ai ∈ Hom(T 1,0, T 0,1) A =Xi We now show how, given a smooth cubic form Ξ ∈ C ∞(M, Sym3T ∗M ), we recover the corresponding ∇, or rather A. In local coordinates ω is given by equation (2.17). Also, the curvature tensor A is locally given by ¯k A ijdzi ⊗ dzj ⊗ ∂ ∂z¯k . A =Xi,j,k CURVATURE FOR THE COMPLEXIFIED INDEX CONE OF A CUBIC FORM 7 Then Ξijk = = = = = i i 2Xp,q 2Xp,q 2Xp,q 2Xp,q 2Xq i i i gp¯qdzp ∧ d¯zq(cid:18) ∂ gp¯qdzp ∧ d¯zq  gp¯qdzp ∧ d¯zq  gp¯qdzp ∧ d¯zq ∂ ∂ ∂zi ∂ ∂zi ∂zi gi¯qA¯q kj. , (A ∂ ∂zj π(1,0)) ∂ ∂zk(cid:19) ∂zi ⊗ dzj) ∂ ∂zk  ∂ ∂zj )(Xj ∂zk  ∂ ) A ¯m jl dzl ⊗ A ¯m jl dzl ⊗ ∂ ∂ ¯zm ∂ ∂ ¯zm A ¯m jk ∂ ∂ ¯zm! , (Xl,m , (Xl,m ,Xm Or equivalently A¯q jk = 2iXi g ¯qiΞijk. Using (2.13), we obtain the following formula for the curvature of (M, ω) in terms of the cubic form Ξ: Lemma 2.12. (2.23) (2.24) Ri¯jk¯l = 4Xp,q =Xp,q gp¯qΞikp ¯Ξjlq gp¯qFikp ¯Fjlq 3. Projective special geometry Projective special Kahler manifolds are basically affine special Kahler manifolds with a C∗-action. In Freed's setup this is achieved by taking a line bundle L over the Kahler manifold M . Then by deleting the zero section one obtains a principal C∗-bundle, which naturally equips the total space with a C∗-action. Definition 3.1. Let (M, ω) be a Kahler manifold, L → M a holomorphic line bundle. (M, L, ω) is called a Hodge manifold if c1(L) = [ω]. This implies that the Kahler class is integral. As (L, ωL) is in particular a vector bundle, it has a global zero section. π Removing this section, we obtain a principal bundle M −→ M with structure group C∗. Every holomorphic line bundle admits a hermitian metric, which 8 THOMAS TRENNER implies the existence of the Chern connection ∇L on L, which preserves the hermitian structure of the line bundle and has the property that ∇L induces a connection on the principal bundle M . ∇0,1 L = ¯∂. Definition 3.2. Given a vector bundle π : V → M and a continuous map f : M ′ → M , we define the pullback bundle f ∗V := {(p, v) ∈ M ′ × V f (p) = π(v)} ⊆ M ′ × V together with the natural projection to the first factor: π′ : f ∗V → M ′. For a section s ∈ Γ(M, V ) we define a pullback section f ∗s ∈ Γ(M ′, V ) by f ∗s = s ◦ f . The pullback bundle π∗L → M has a canonical nonzero holomorphic section s. We define a pseudo-Kahler metric on M : Definition 3.3. Let ω ∈ Ω1,1( M ) denote the form which is given by: (1) ω(U, X) = 0 (2) ω(X, Y ) = s2π∗ω(X, Y ) (3) ω(U, V ) = − i 2π ∂ ¯∂s2(U, V ), where X, Y are horizontal vector fields and U, V are vertical vector fields. In the following we will establish a curvature formula for ω in terms of ω and the cubic form Ξ on M . Lemma 3.4. ( M , ω) is a pseudo-Kahler manifold. Furthermore and ω = i 2π ∂ ¯∂s2 π∗ω = i 2π ∂ ¯∂ log s2 Proof. The horizontal lift of a vector field X ∈ T U is X − λh−1∂h(X) ∂ ∂λ . Also, as by this formula ω is exact and real of type (1, 1), it defines a pseudo-Kahler metric of signature (dim(M ), 1). on M . It will not define a Kahler metric as it will be negative on vertical vector fields. The second formula is just the standard curvature formula for the Chern connection of a holomorphic line bundle, up to the constant factor of (cid:3) i 2π . Using this characterisation of the metrics, we can compute the Kahler metric g on M in terms of the Kahler metric g on M . We choose local coordinates zi on U ⊂ M and a coordinate λ on LU . Choosing a nonzero local section t of LU , we get local coordinates on π∗(U ) ⊂ M by (z, λ) 7→ λt(z), λ ∈ C∗. CURVATURE FOR THE COMPLEXIFIED INDEX CONE OF A CUBIC FORM 9 Lemma 3.5. Let g, g be the Kahler metric on M and M . Let Ki := ∂i log K, K the Kahler potential of ω. Also, we use the index 0 in the metric for the fibre coordinate. Then (3.1) (3.2) (3.3) (3.4) (3.5) (3.6) gi¯j = K(−gi¯j + KiK¯j) g0¯i = Kλ−1Ki g0¯0 = (λ¯λ)−1K gi¯j = −K −1gi¯j n g0¯i = ¯λK −1 gk¯iK¯k Xk=1 g0¯0 = K −1λ¯λ(1 − KiK¯jg ¯ji) n Xi,j=1 The entries of the inverse metric have been found by fixing the gi¯j for i, j = 1, . . . , n to give the right answer when computing the Strominger for- mula in the projective case. This determines the other entries and indeed gives an inverse matrix to g. A tedious but easy calculation leads to expressions for the connection coefficients of the Levi -- Civita connection of g in terms of the coefficients of the Levi -- Civita connection for g. As both metrics are Kahler , we only have to compute the ones which might be non-vanishing: Lemma 3.6. With notation as in the previous lemma, we compute the con- nection coefficients Γi (3.7) (3.8) (3.9) (3.10) jk = gi¯l∂jg¯lk: k + Kkδi j jk + Kjδi jk, using Γi Γi jk = Γi Γi j0 = λ−1δi j Γ0 ij = λΓk 00 = Γ0 Γ0 ijKk + 2λKiKj − K −1λ∂i∂jK i0 = Γi 00 = 0 It immediately follows that (3.11) jk = Γi Γi jk − λΓi j0Kk − λΓi 0kKi We can now compute the (3, 1) curvature tensor Rm ik¯l of the Levi -- Civita connection on M , using the following formula: Rm ik¯l = −∂¯lΓm ik Expressed in terms of the (3, 1) curvature tensor Rm potential K, we obtain: ik¯l = −( Rm Rm ik¯l − ∂¯l(δm i Kk + δm k Ki)), ik¯l on M and the Kahler 10 THOMAS TRENNER where i, j, k, m ∈ {1, . . . , n}. Expressing Rm obtain: ik¯l in terms of the cubic form, we Rm ik¯l = −4 Xp,q,m,n gp¯q gn ¯m Ξikp ¯Ξmlq + ∂¯l(δm i Kk + δm k Ki) Plugging in (2.23), we obtain for the (4, 0) curvature tensor: gm¯j Rm ik¯l Ri¯jk¯l =Xm =Xm = −Xm gm¯j (− Rm ik¯l + ∂¯l(δm i Kk + δm k Ki)) 1 K gp¯q 1 K gn ¯m Ξikp gm¯j4Xp,q,n ¯Ξmlq +Xm gm¯j ∂¯lδm i Kk +Xm gm¯j∂¯lδm k Ki = − = − = − 4 4 K 2 Xp,q K 2 Xp,q K 2 Xp,q 1 gp¯qΞikp ¯Ξjlq + gi¯j∂¯lKk + gk¯j∂¯lKi gp¯qΞikp ¯Ξjlq + gi¯jgk¯l + gk¯jgi¯l gp¯qFikp ¯Fjlq + gi¯jgk¯l + gk¯jgi¯l For the last equality we take the above description of ω. Using this in definition 2.10 and remembering that ∇ is just the pullback of ∇ by the C∗-action, we see that Ξ and Ξ do not differ for the vector fields coming from M . It should be noted that for a vector field X on M we have to use the horizontal lift to M to make this identification. The final step is to rewrite the Kahler potential in terms of the cubic form: Proposition 3.7. Denote the coordinates on M by λ, z1, . . . , zn with zj = xj + iyj. In special coordinates tj = zj/λ, j = 1, . . . , n, writing ∂j for ∂ ∂tj , the Kahler potential for the Weil -- Petersson metric is given by K = − log i(Xj ((tj − ¯tj)(∂jF ′ + ¯∂j ¯F ′)) + 2 ¯F ′ − 2F ′) = 8F(y1, . . . , yn) Proof. Let the tj = zj λ form a local coordinate system and define F ′(t) = λ−2F(z). Then observe that λ∂jF ′ = ∂F . Going to inhomogeneous coordi- ∂zj nates, we have to account for the following term that has to be corrected in the sum: λ ∂F ∂λ = λ2(2F ′ −Xj tj∂jF ′) Plugging this into the formula given in the Lemma above we obtain the result. (cid:3) CURVATURE FOR THE COMPLEXIFIED INDEX CONE OF A CUBIC FORM 11 Definition 3.8. [4, p.46] Let (M, L, ω) be an n-dimensional Hodge mani- fold. A projective special Kahler structure is a triple (V, ∇, Q), where (1) V → M is a rank n+1 holomorphic vector bundle with a holomorphic inclusion L ֒→ V . (2) ∇ is a flat connection on VR, the underlying real vector bundle, such that the extension of ∇ to the complexification (VR)C satisfies ∇(L) ⊂ V and the section M → P[(VR)C] m 7→ Lm is an immersion with respect to ∇, or equivalently, L, seen as a section of P[(VR)C], is transverse to the horizontal distribution of the connection. (3) Q is a non-degenerate skew-symmetric form on VR of type (1, 1) with respect to the complex structure. Moreover, it should be flat with respect to ∇. Finally, taking the extension to (VR)C, we assume QL× ¯L = i 2π ωL. Remark 3.9. By ∇(L) ⊂ V we mean that if we take a section s ∈ Γ(L), where L is embedded in V , we get ∇(s) ∈ Γ(V ) ⊗ ΩM . A projective special Kahler structure defines a certain kind of variation of Hodge structures. If we furthermore require the existence of a lattice in VR such that Q restricted to the lattice takes integer values, the corresponding variation of Hodge structures will be polarised. The connection to affine special geometry is established by the following proposition: Proposition 3.10. [4, p.46] Let (M, L, ω) be a Hodge manifold with asso- ciated pseudo-Kahler manifold ( M , ω) and canonical holomorphic non-zero section s. Then a projective special Kahler structure on (M, L, ω) is equiv- alent to a C∗-invariant special pseudo-Kahler structure ∇ on ( M , ω) with ∇s = π(1,0). The condition ∇s = π(1,0) is needed to guarantee that the map in part (2) of the definition of a projective Kahler structure is an immersion: consider s as a holomorphic vertical vector field on M and denote by ∇ the pullback of the connection to π∗V . Then the map (3.12) ∇s : T M → π∗V ∂/∂zi 7→ ∇∂/∂zis is an isomorphism if and only if s is transverse to the horizontal distribu- tion of the connection. The underlying real isomorphism induces a real flat connection on T M . 12 THOMAS TRENNER If we denote the sections of V by s, l1, . . . , ln, the immersion condition says that the effect of the connection on s is given by: ∇s = ∂ ∂zj n Xj=0 ⊗ dzj This implies that under (3.12), s corresponds to a complex vector ζ field that satisfies ∇ζ = π(1,0) because of π(1,0) =Pj ∂ ∂zj ⊗ dzj. Another useful fact is that Q, the pullback of the skew-symmetric form via π, pulls back to −ω under the isomorphism, which can be seen by dif- ferentiating the equation we assume in point (iii) of the projective special Kahler structure. Using this together with definition 2.10 and the fact that ω(ζ, ∇ζ) = 0 as ζ is a holomorphic (1, 0) vector field, we conclude that the cubic form is given by (3.13) Ξ = −ω(ζ, ∇3ζ) = Q( ∇3s, s). In the following sections we will describe two instances of projective spe- cial Kahler manifolds that arise in Mirror Symmetry, the complex moduli space and the (complexified) Kahler moduli space of a Calabi -- Yau three- fold. A curvature formula for the former can be found in [6], although it is derived in a slightly different fashion. We want to compute the curvature of the complexified Kahler moduli space of a Calabi -- Yau threefold with the line bundle H0,0. We do that by proving the formula for the more general case of a complexified index cone of a cubic form. 4. Complex moduli space as a Special Kahler Manifold Let X be a Calabi -- Yau threefold, M the moduli space of complex struc- tures. Over M, consider the Hodge bundle H with fibre H 3(X, C) and Hodge filtration F •. This gives a 2n + 2-dimensional vector bundle over M. Using integral cohomology, we can equip it with a flat connection, called the Gauss -- Manin connection. There exists a symplectic basis for H 3(X, C) and this provides local flat sections for M with respect to the Gauss -- Manin connection on H. These bases are unique up to Sp(2n + 2, R) transforma- tions. Thus H is a flat, holomorphic Sp(2n + 2, R) bundle. As all the information about the Hodge structure is already contained in H 3,0 ⊕ H 2,1, we get an n + 1-dimensional sub-bundle V ∼= F 2. On H, we have two metrics: One is a hermitian metric given by the polarisation of Hodge structure Q(C·, ·) (in most accounts called Hodge metric), where C is the Weil operator, i.e. acting by multiplication by ip−(n−p) on H p,q. The other one is given just by taking the polarisation with a leading factor of −in, which in general is not positive definite. We will call it the indefinite metric. These two metrics differ by (−1)p on H p,q. While the indefinite metric is CURVATURE FOR THE COMPLEXIFIED INDEX CONE OF A CUBIC FORM 13 by definition preserved by the Gauss -- Manin-connection, the Hodge metric is not. This implies that ∇GM is not the Chern connection of the hermitian line bundle H3,0. On any Calabi -- Yau manifold of the family described by the complex moduli space, the nowhere-vanishing holomorphic 3-form Ω is only defined up to multiplication by a non-zero complex number. So in this family, we can multiply Ω by a non-vanishing holomorphic function f (z). As we want no-where vanishing 3-forms, we take the following transformation for Ω: Ω′ = ef (z)Ω, where f is a continuous function of the moduli space coordinates. Now at different points in the complex moduli space, we have different com- plex structures and thus a different decomposition of the tangent bundle of our Calabi -- Yau threefold in holomorphic and anti-holomorphic parts. This means that (3, 0)-forms that are holomorphic in one complex structure may not stay holomorphic in another complex structure. As Ω is by definition a holomorphic form, it defines us a section of H 3. The multiples of Ω define the line bundle H3,0 which is a sub-bundle of both H and V. We also note for later that because of the freedom in choice for Ω we obtain a section of the projectivization of V. We equip M with a projective special Kahler structure (V, ∇, Q) given by: (1) The vector bundle V := H3,0 ⊕ H2,1 with the holomorphic inclusion L := H3,0 ֒→ H3,0 ⊕ H2,1. (2) The Gauss -- Manin connection ∇GM on the underlying real bundle VR. For the rest of this chapter, we will denote the Gauss -- Manin connection simply by ∇. (3) The non-degenerate form Q is the polarisation of the variation of Hodge structures on the complex moduli space, divided by 2π. Then Q is flat with respect to ∇GM . Lemma 4.1. (V, ∇, Q) defines a projective special Kahler structure. Here F p = P3 Proof. The holomorphicity of the inclusion L ֒→ V follows from a theorem of Griffiths [5, p.24] stating that the bundles in the Hodge filtration F p cor- responding to H are holomorphic sub-bundles of the full Hodge bundle H). i=p Hp,3−p. The requirement that ∇(L) ⊂ V after complexifi- cation of the underlying real bundles is just Griffiths transversality for the Hodge bundles. For the immersion condition in part (2) of the definition of a projective special Kahler structure, we note that V is generated by Ω and Ω, i = 1, . . . , n. But L is generated by the non-vanishing (3, 0)-form Ω ∇ ∂ ∂zi and ∇ ∂ ∂zi Ω is a non-zero section of V, so L, seen as a section of P[(VR)C], is transverse to the horizontal distribution of the connection. The remaining required properties for Q are immediate from the definition. (cid:3) 14 THOMAS TRENNER With this setup and the calculations of the previous section, we obtain the following curvature formula for the complex moduli space appearing in Strominger's paper [6]: Ri¯jk¯l = gi¯jgk¯l + gi¯lgk¯j − e2KXp,q gp¯qYikpYjlq, where g is the Weil -- Petersson metric, given by taking as Kahler potential log Q(Ω, ¯Ω), and Yijk denotes the Yukawa couplings (see [3, p.102] for de- tails). We use (3.13) to obtain as the cubic form Ξ the well-known formula for the Yukawa couplings. The cubic form does not fix the prepotential com- pletely, but as the curvature only depends on Ξ, we obtain the well-known original formula for the complex moduli space. Proofs of this formula for the Weil -- Petersson metric on the complex mod- uli space of a Calabi -- Yau threefold have appeared in the literature before. A proof relying on the earlier extrinsic definition of special geometry via special coordinates and their behaviour under coordinate changes can be found in [6]. Two more proofs, using different methods -- most importantly a reformulation of the problem in terms of the period mapping and a theo- rem by Griffiths on the curvature of Hodge bundles -- have been found by Wang [8]. 5. Complexified index cone of a cubic form as a projective Special Kahler manifold Take any cubic polynomial with real coefficients f (y1, . . . , yn) defined on Rn. Definition 5.1. The index cone of f is the open cone W ⊂ Rn where f is positive and the Hessian matrix (∂2f /∂yi∂yj) has index (1, n − 1). Definition 5.2. The complexified index cone M is given by (Rn+iW )/im(Zn), where Zn is mapped into the first summand. Let ti = xi + iyi. As was shown in [7, p.7ff.], starting with the com- plexification of a cubic polynomial f -- e.g. the cubic intersection form of a Calabi -- Yau threefold -- we get a Kahler potential from it by using the formula K = − log i(Xj (tj − ¯tj)(∂j f + ¯∂j ¯f ) + 2 ¯f − 2f ). This potential function is independent of the xi, and is just a multiple of the original polynomial f . We consider the Hessian metric (− 1 4 ∂2 log f /∂yi∂yj) on the complexified index cone, obtained by taking K as the Kahler potential. By [7, Lemma 2.1], this defines a Kahler metric on the whole complexified index cone. CURVATURE FOR THE COMPLEXIFIED INDEX CONE OF A CUBIC FORM 15 It was conjectured in [7, p.10] that the following equivalent of the Stro- minger formula holds for the Hessian metric on the complexified index cone: (5.1) Ri¯jk¯l = gi¯jgk¯l + gi¯lgk¯j −Xp,q gp¯q fikpfjlq 64f 2 The factor of 64 is due to the fact that K = 8F(y) = 8f , so 1 K 2 = 1 64f 2 . In the following we will show that the complexified index cone together with the Hessian metric given by the cubic polynomial f and the trivial complex line bundle L = C × M form a projective special Kahler manifold. Denote the total space obtained by deleting the zero section from L again by M . We choose local coordinates z1, . . . , zn, λ, where λ denotes the fibre coordinate. Then a projective special Kahler structure (V, ∇, Q) on M given by: (1) The vector bundle V := Cn+1 with the obvious holomorphic inclusion of L = C. The underlying real bundle is just Rn+1. (2) Choose rational non-vanishing sections s, s1, . . . , sn for V , with s the section for L. Our aim is to identify these sections with the basic vector fields ∂/∂zi, ∂/∂λ, where the former should be identified with s1 up to sn and the last one should be identified with s. We use the shorthand notation ∂z for the vector field ∂/∂z. On the real bundle define a connection by setting ∇∂zis = si n ∇∂zisj = Xk=1 ∇∂zi ¯sj = δi j ¯s ∇∂zi ¯s = 0, ∂3f ∂zi∂zj∂zk ¯sk where ¯sk are rational sections of (Cn)∨, the dual to V \L. (3) The non-degenerate form Q is given by Q(α, β) = 1 We compute 2 αβ. Ξijk = Q( ∇∂zi = Q( ∇∂zi = Q( ∇∂zi ∇∂zk s, s) ∇∂zj ∇∂zj sk, s) n ∂3f ∂zi∂zj∂zk ¯si, s) = Q( ¯s, s) Xi=1 ∂3f ∂zi∂zj∂zk ∂3f ∂zi∂zj ∂zk = 1 2 which shows that (5.1) holds. 16 THOMAS TRENNER References 1. Alekseevsky, D.V., Cortes, V., Devchand, C.: Special complex manifolds, J. Geom. Phys. 42, 85 -- 105 (2002) 2. Bartocci, C., Mencattini, I.: Some remarks on special Kahler geometry, J. Geom. Phys. 59, 755 -- 763 (2009) 3. Cox, D.; Katz, S.: Mirror Symmetry and Algebraic Geometry. Mathematical Surveys and Monographs, vol. 68. AMS, Providence (1999) 4. Freed, D. S.: Special Kahler manifolds, Commun. Math. Phys. 203, no. 1, 31 -- 52 (1999) 5. Griffiths, P., ed.: Topics in Transcendental Algebraic Geometry. Princeton Univ. Press (1984) 6. Strominger, A.: Special Geometry, Commun. Math. Phys. 133, 163 -- 180 (1990) 7. Trenner, T., Wilson, P. M. H.: Asymptotic curvature of moduli spaces for Calabi -- Yau threefolds, arXiv:0902.4611v2, to appear in J. Geom. Anal. (2010) 8. Wang, C.-L.: Curvature Properties of The Calabi -- Yau Moduli, Doc. Math. 8, 577 -- 590 (2003) Centre for Mathematical Sciences, Wilberforce Road, Cambridge CB3 0WA E-mail address: [email protected]
1411.7363
3
1411
2015-07-08T13:39:59
Higher convexity for complements of tropical varieties
[ "math.AG", "math.CO" ]
We consider Gromov's homological higher convexity for complements of tropical varieties, establishing it for complements of tropical hypersurfaces and curves, and for nonarchimedean amoebas of varieties that are complete intersections over the field of complex Puiseaux series. Based on these results, we conjecture that the complement of a tropical variety has this higher convexity, and we prove a weak form of our conjecture for the nonarchimedean amoeba of a variety over the complex Puiseaux field. One of our main tools is Jonsson's limit theorem for tropical varieties.
math.AG
math
HIGHER CONVEXITY FOR COMPLEMENTS OF TROPICAL VARIETIES MOUNIR NISSE AND FRANK SOTTILE Abstract. We consider Gromov's homological higher convexity for complements of tropical varieties, establishing it for complements of tropical hypersurfaces and curves, and for nonarchimedean amoebas of varieties that are complete intersections over the field of complex Puiseux series. Based on these results, we conjecture that the complement of a tropical variety has this higher convexity, and prove a weak form of this conjecture for the nonarchimedean amoeba of any variety over the complex Puiseux field. One of our main tools is Jonsson's limit theorem for tropical varieties. A tropical hypersurface is a polyhedral complex in Rn of pure dimension n−1 that is dual to a regular subdivision of a finite set of integer vectors. This implies that every connected component of its complement is convex. A classical (archimedean) amoeba of a complex hypersurface also has the property that every connected component of its complement is convex [2, Ch. 6, Cor. 1.6]. Gromov [3, § 1 2] introduced higher convexity. A subset X ⊂ Rn is k-convex if for all affine planes L of dimension k+1, the natural map on kth reduced homology ιk : eHk(X ∩ L) −→ eHk(X) (1) is an injection. Connected and 0-convex is ordinary convexity. Henriques [4] rediscovered this notion and conjectured that the complement of an amoeba of a variety of codimension k+1 in (C×)n is k-convex, and established a weak form of this conjecture: the map ιk sends no positive class to zero [4]. Bushueva and Tsikh [1] used complex analysis to prove Henriques' conjecture when the variety is a complete intersection. Other than these cases, Henriques' conjecture remains open. An amoeba is the image in Rn of a subvariety V of the torus (C×)n under the coor- dinatewise map z 7→ log z. Similarly, the coamoeba is the image in (S1)n of V under the coordinatewise argument map. Lifting to the universal cover and taking closure gives the lifted coamoeba in Rn. There is also a nonarchimedean coamoeba and a lifted nonar- chimedean coamoeba [11]. The complement of either type of lifted coamoeba of a variety of codimension k+1 is k-convex [10], which was proven using tropical geometry. We investigate Gromov's higher convexity for complements of tropical varieties. We show that the complement of a tropical curve in Rn is (n−2)-convex. Both this and the convexity of tropical hypersurface complements rely only on some properties of tropical 2010 Mathematics Subject Classification. 14T05. Research of Sottile is supported in part by NSF grant DMS-1001615. This is based upon work done at the NIMS, Daejeon, Korea, This work was supported by National Institute for Mathematical Sciences 2014 Thematic Program. 1 2 MOUNIR NISSE AND FRANK SOTTILE varieties and hold for polyhedral complexes having these properties. This leads us to conjecture that the complement of a tropical variety of pure codimension k+1 is k-convex. Suppose that K is an algebraically closed field having a nonarchimedean valuation with value group dense in R and let V ⊂ (K×)n be a subvariety. The closure in Rn of the image of V under the coordinatewise valuation map is its nonarchimedean amoeba, which is a tropical variety. When K is the field of complex Puiseux series, methods from analysis, both archimedean and nonarchimedean, may be used to study nonarchimedean amoebas. Jonsson showed that such a nonarchimedean amoeba is a limit of archimedean amoebas [5, Thm. B], generalizing work of Rullgard [12, Thm. 9] and Mikhalkin [8, Cor. 6.4] for hypersurfaces. This, together with a technical lemma, allows us to use the result of Bushueva and Tsikh [1] to establish our conjecture for the nonarchimedean amoeba of a complete intersection in (K×)n. The weak form (that the map ιk (1) sends no positive cycle to zero) for any subvariety of (K×)n also follows by Henriques's result for amoebas. In Section 2, we define combinatorial tropical hypersurfaces and curves in Rn, and prove that their complements are 0-convex and (n−2)-convex, respectively. In Section 3, we prove our results about nonarchimedean amoebas stated above. In Section 1 we give background material. This paper is organized as follows. 1. Background We provide some background on tropical varieties, state Jonsson's limit theorem, and discuss higher convexity. 1.1. Tropical varieties. The map z 7→ log z is a homomorphism from the non-zero complex numbers C× to the real numbers R. This induces the map Log : (C×)n → Rn. The image of subvariety V of (C×)n under Log is its amoeba, A (V ). Let K be an algebraically closed valued field whose value group G is a non-zero divisible additive subgroup of R. Its valuation is a surjective homomorphism ν : K× → G, which induces a map ν : (K×)n → Gn. The closure in Rn of the image of a variety V ⊂ (K×)n under the map ν is its nonarchimedean amoeba, T (V ). There is an equivalent definition of T (V ). An integer vector a ∈ Zn forms the expo- n . A Laurent polynomial f is a K-linear 1 · · · xan nents of a Laurent monomial, xa := xa1 combination of Laurent monomials, f = X caxa a∈A where ca ∈ K× . The finite subset A ⊂ Zn is the support of f . The coordinate ring of (K×)n is the ring of Laurent polynomials K[x1, x−1 n ]. Given a vector w ∈ Rn and a Laurent polynomial f with support A ⊂ Zn, we have a piecewise linear map 1 , . . . , xn, x−1 Rn ∋ z 7−→ min{ν(ca) + w · a a ∈ A} . (2) The tropical hypersurface trop(f ) is the set where this minimum occurs at least twice. This is also the set where the piecewise linear map (2) is not differentiable. Given a variety HIGHER CONVEXITY OF TROPICAL VARIETIES 3 V ⊂ (K×)n, let I be its ideal in the ring of Laurent polynomials. Its tropical variety is \ f ∈I trop(f ) . By the fundamental theorem of tropical geometry [7], this tropical variety equals its nonar- chimedean amoeba, T (V ). If V has dimension r, then T (V ) admits (non-canonically) the structure of a polyhedral complex of pure dimension r. There are positive integral weights ασ associated to each polyhedron σ of maximal dimension r so that the weighted complex is balanced. We explain this. Let τ be an (r−1)-dimensional polyhedron in T (V ). Modulo the affine span hτ i of τ , each r-dimensional polyhedron σ incident on τ (σ ∈ star(τ )) determines a primitive vector vσ. The balancing condition is that X σ∈star(τ ) ασvσ = 0 mod hτ i . We are primarily concerned with nonarchimedean amoebas when the field K is the com- plex Puiseux field. This algebraically closed field contains the field of rational functions C(s), which is the quotient field of the ring of univariate polynomials C[s]. An element of the Puiseux field is a fractional power series of the form c = X m≥0 ams p+m q , where the coefficients am are complex numbers and p, q are integers with q > 0. The valuation ν(c) is the minimum exponent appearing in the power series c with a non-zero coefficient. This is a homomorphism ν : K× → Q. 1.2. Jonsson's limit theorem. Let d(x, y) be the Euclidean distance in Rn between x and y. If ∅ 6= A ⊂ Rn is closed and x ∈ Rn, then the distance between x and A is d(x, A) := inf{d(x, a) a ∈ A} , which is attained as A is closed. If B is closed, then d(A, B) = inf{d(a, b) a ∈ A, b ∈ B}. A family {At t > 0} of closed subsets of Rn has Kuratowski limit T ⊂ Rn if we have the following equality, T = {x ∈ Rn ∀ǫ > 0 ∃δ > 0 such that 0 < t < δ ⇒ d(x, At) < ǫ} = {x ∈ Rn ∀ǫ > 0 ∀δ > 0 ∃0 < t < δ with d(x, At) < ǫ} . (3) When this occurs, we write lim t→0 At = T . This is distinct from the more familiar notion of Hausdorff limit: Consider the family of lines in the plane R2 through the origin with slope t > 0. As t → 0, these lines have Kuratowski limit the x-axis, but they do not have a Hausdorff limit. Observe that the Kuratowski limit T is closed. More interesting is if Z is a compact set disjoint from T . 4 MOUNIR NISSE AND FRANK SOTTILE Lemma 1.1. Suppose that {At t > 0} is a family of closed subsets of Rn with Kuratowski limit T . If Z is a compact subset of Rn that is disjoint from T , then there is a δ > 0 such that if 0 < t < δ, then Z ∩ At = ∅. Proof. By (3), if z ∈ Z, so that z 6∈ T , then there are ǫ, δ > 0 such that for every 0 < t < δ, we have d(z, At) > ǫ. Since Z is compact, there are ǫ, δ > 0 such that if z ∈ Z and 0 < t < δ, then d(z, At) > ǫ. In particular, 0 < t < δ implies that Z ∩ At = ∅. (cid:3) Let V ⊂ C× × (C×)n be a subvariety whose every component maps dominantly onto the first factor, C×, which has coordinate s. Then V is a family of varieties over an open subset U of C×, with fiber Vs over s ∈ U. Equivalently, V is a variety in the torus (C(s)×)n over C(s). Extending scalars to the Puiseux field K gives a variety V ⊂ (K×)n with tropicalization T (V ). In this context, Jonsson [5] proved the following. Proposition 1.2 (Jonsson). We have lim s→0 −1 log s A (Vs) = T (V ). For this, ν(s) = 1. More generally, T (V ) should be scaled by ν(s). To summarize Jonsson's Theorem, the nonarchimedean amoeba of V is the limit of (appropriately scaled) amoebas of fibers of the family V. Stated in this way, Jonsson's Theorem holds in the broader context of tropicalizations of varieties in (K×)n. Proposition 1.3. Let W ⊂ (K×)n be any variety. Then there is a smooth curve C, a point o ∈ C, a local parameter u at o, and a family of varieties V ⊂ (C r {o}) × (C×)n over C r {o} with fiber Va over a ∈ C r {o} such that lim a→o −1 log u(a) A (Va) = T (W ) . If W is a complete intersection, then we may choose the family V so that every fiber Va is a complete intersection. The proof we give uses the following result of Katz. Proposition 1.4 (Thm. 1.5 [6]). Let W be a variety in (K×)n. Then there is a finite extension L of C(s) and a variety W ′ in (L×)n with the same tropicalization and the same Hilbert polynomial as W . This is discussed in the paragraph following the statement of Theorem 1.5 in [6]. Proof of Proposition 1.3. By Proposition 1.4, we may assume that W is defined over a finite extension of C(s), which is the function field, C(C), of a smooth complex curve C. The inclusion C(s) ⊂ C(C) induces a dominant rational map π : C − → C. If we let o ∈ π−1(0) and u be a local parameter at o, then u generates C(C) over C(s). Replacing C by an affine neighborhood of o if necessary, there is a family of varieties W ⊂ (C r {o}) × (C×)n over C r {o} such that, when scalars are extended to K, gives the variety W ⊂ (K×)n. We claim that lim a→o −1 log u(a) A (Wa) = ν(u) · T (W ) . This follows from nearly the same arguments as Proposition 1.2, which are given in Section 4 of [5]. To obtain the statement of the proposition, set V := e−ν(u)W. HIGHER CONVEXITY OF TROPICAL VARIETIES 5 Finally, if the original variety W ⊂ (K×)n was a complete intersection then so is the variety we replace it by in (C(C)×)n as they have the same Hilbert polynomial. Thus there is an open subset U of C such that W is flat over U, and in fact so that Wa is a complete intersection for a ∈ U. (cid:3) 1.3. Higher convexity. Let Y ⊂ Rk+1 be a closed set with finitely many connected components such that the pair (Rk+1, Y ) is triangulated. Set Y c := Rk+1 r Y . Let Y1, . . . , Ym be the bounded connected components of Y . Then there exist open subsets β1, . . . , βm of Rk+1 with disjoint closures, where βi a neighborhood of Yi and γi := ∂βi, the boundary of βi, is a k-cycle in Y c. Lemma 1.5. The classes of the cycles γ1, . . . , γm form a basis for eHk(Y c). We remark that we take coefficients in Z. Fixing an orientation for Rk+1 orients each βi and gives each γi = ∂βi the outward orientation. A nonzero homology class ζ ∈ eHk(Y c) A cycle class [γ] ∈ eHk(Y c) vanishes in eHk(Rk+1), and so it is the boundary of a (k+1)- chain Z in Rk+1. Then [γ] ∈ eHk(Y c) is non-zero if and only if for every (k+1)-chain Z Gromov [3, § 1 2] gave a homological generalization of convexity. An open set X ⊂ Rn is is positive if it is a nonnegative integer combination of the classes [γi]. with ∂Z = γ, we have Z ∩ Y 6= ∅. k-convex if for any affine linear space L of dimension k+1, the natural map ιk : eHk(L ∩ X) −→ eHk(X) is an injection. Note that 0-convex and connected is the ordinary notion of convex. Henriques [4] considered bu did not define a weak version of this notion: The set X is weakly k-convex if the map ιk does not send any positive cycle to zero. This is independent of choices, changing the orientation of L replaces each positive cycle γ by −γ. Let Y be a polyhedral complex of codimension k+1 in Rn consisting of finitely many polyhedra. An affine subspace L in Rn of dimension k+1 is transverse to Y if L meets each polyhedron σ of Y transversally. That is, if L ∩ σ 6= ∅, then σ has dimension n−k−1 and L meets it transversally, necessarily in a single point in the relative interior of σ. Henriques proved a moving lemma [4, Lemma 3.6], which implies that it suffices to take the affine space L in the definition of k-convexity to be a translate of some rational affine space -- one of the form M ⊗Q R, for M an affine subspace of Qn of dimension k+1. The same proof shows that it suffices to take L to lie in a dense subset of such subspaces. 2. Combinatorial tropical varieties A "combinatorial tropical variety" of dimension r is a polyhedral complex in Rn of pure dimension r in Rn that has some of the properties of a nonarchimedean amoeba. Mikhalkin and Rau [9] call these tropical cycles. Consequently, a result about some type of combinatorial tropical variety implies the same result for nonarchimedean amoebas. Let A ⊂ Rn be a finite set of points and c ∈ RA a vector whose components ca are indexed by elements a of A. These define a piecewise linear function on Rn, T (A, c) := x 7−→ min{ca + w · a a ∈ A} . (4) 6 MOUNIR NISSE AND FRANK SOTTILE Its graph Γ(A, c) is a polyhedral complex of dimension n whose facets lie over the domains of linearity of T (A, c). Its ridge set is the union of faces of dimension at most n−1, which lies over the set T (A, c) where T (A, c) is not differentiable. This set T (A, c) is a combinatorial tropical hypersurface and it consists of the points x ∈ Rn where the minimum in (4) is attained at least twice. The following is elementary and not original. Proposition 2.1. The set T (A, c) is a polyhedral complex of pure dimension n−1 whose complement is 0-convex. Proof. The projection of Γ(A, c) to Rn is a piecewise linear homeomorphism. The first statement follows as its ridge set has pure dimension n−1 and the second as the compo- nents of the complement of the ridge set consists of the interiors of the facets of Γ(A, c). (cid:3) When A ⊂ Zn, the set T (A, c) is the tropicalization of the hypersurface {x ∈ (K×)n 0 = X scaxa } , where K is a valued field with value group R and s ∈ K has ν(s) = 1. a∈A By the fundamental theorem of tropical geometry [13], the nonarchimedean amoeba C of a curve C in (K×)n admits the structure of a finite balanced rational polyhedral complex in Rn of pure dimension one. Putting such a structure on C , we have that C consists of finitely many vertices and edges, with each edge an interval (possibly unbounded) of a line. Furthermore, each edge e is equipped with a positive integral wight αe and is parallel to a vector ve ∈ Zn. We assume that ve is primitive in that its components are relatively prime. There are exactly two primitive vectors, ve and −ve, that are parallel to e, one for each direction along e. Finally, balanced means that for every point p ∈ C , we have 0 = X αeve , (5) e the sum over all edges e incident on p where ve points away from p. This sum (5) is nonempty as C is pure and therefore has no isolated points. Lemma 2.2. Suppose that w ∈ Rn and that p ∈ C . Then either we have that w · ve = 0 for all edges e incident on p, or else there are two edges e, f incident on p with w · ve < 0 < w · vf . Proof. This follows from (5) as each weight αe is positive. (cid:3) Definition 2.3. A locally finite polyhedral complex C in Rn of pure dimension one is weakly balanced if Lemma 2.2 holds for C . That is, if for all w ∈ Rn and p ∈ C , either w · ve = 0 for all edges e incident on p or there are two edges e, f incident on p with w · ve < 0 < w · vf . Here, ve and vf are any vectors pointing away from p that are parallel to e and f , respec- tively. Weakly balanced graphs admit unbounded paths in nearly every direction. HIGHER CONVEXITY OF TROPICAL VARIETIES 7 Lemma 2.4. Let C be a weakly balanced graph and w ∈ Rn be nonzero. Then, for every point p of C which has an incident edge e with w · ve 6= 0, there is a continuous path γ : [0, ∞) → C with γ(0) = p where w · γ(t) is unbounded and strictly increasing for t ∈ [0, ∞), and we may assume that γ contains an edge f incident to p with w · vf > 0. In particular, every component of a weakly balanced graph is unbounded. Proof. Let us assume that if w · q > w · p and q is a vertex of C having an incident edge e with w · ve 6= 0, then there is a continuous function γ1 : [0, ∞) → C with γ(0) = q such that w · γ1(t) is an increasing unbounded function. The base case of the induction are those vertices p of C with w · p maximal, which is covered later in this paragraph as edges e emanating from p with w · ve > 0 are unbounded. Let p ∈ C be a point with an incident edge f such that w · vf 6= 0. Since C is weakly balanced, there is an edge e incident to p with w · ve > 0. If e is unbounded in the direction of ve, set γ(t) := p + tve, which gives the desired path. Otherwise, let q be the vertex incident to e in the direction of ve. Then w · q > w · p and the direction of e at q is −ve. Since w · (−ve) 6= 0, the induction hypothesis holds. Let γ1 : [0, ∞) → C be a path with γ1(0) = q and w · γ1(t) increasing and unbounded. Suppose that p + t0ve = q. Then t0 > 0 and we define γ(t) by γ(t) := p + tve for 0 ≤ t ≤ t0 and γ(t) = γ1(t − t0) for t ≥ t0. Then γ is the desired increasing path. (cid:3) Theorem 2.5. The complement of a weakly balanced graph in Rn is (n−2)-convex. Proof. We must show that for any hyperplane L that meets C , the map is injective. ιn−2 : eHn−2(L ∩ C c) −→ eHn−2(C c) connected component of L ∩ C . We assume that C 6⊂ L, for otherwise there are no The reduced homology group eHn−2(L ∩ C c) is free with one generator for each bounded bounded connected components of L ∩ C and eHn−2(L ∩ C c) vanishes. By Lemma 1.5, if C1, . . . , Cm are the bounded connected components of L ∩ C , then there are open subsets β1, . . . , βm with disjoint closures and βi ⊃ Ci such that if γi := ∂βi for i = 1, . . . , m, then the cycle classes [γi] are a basis for eHn−2(L ∩ C c). Suppose that ζ ∈ eHn−2(L ∩ C c) with ιn−2(ζ) = 0 in eHn−2(C c). There are unique integers b1, . . . , bm such that ζ = [γ] where γ is the cycle γ = mX i=1 bi γi . Since ιn−2(ζ) = 0, there is a (n−1)-chain Z in C c with ∂Z = γ. If we set β := − mX i=1 biβi , then Z + β is closed in Rn, and hence bounds an n-chain D in C c. Suppose that ζ 6= 0, so that some coefficient bi is nonzero. Then −biγi forms part of the boundary of D, and we conclude that for every point p of the interior of βi, D contains a neighborhood of p in one of the halfspaces defined by the hyperplane L. 8 MOUNIR NISSE AND FRANK SOTTILE Since βi ∩ C is nonempty and contained in the interior of βi, and C 6⊂ L, there is a point p of βi ∩ C lying in the closure of C r (L ∩ C ). If w is a vector normal to L, then there is an edge e of C incident to p with w · ve 6= 0. Possibly replacing e by another edge incident to p and w by −w, we may assume that w · ve > 0 and that e meets the interior of D. By Lemma 2.4, there is a path γ : [0, ∞) → C with γ(0) = p, w · γ(t) an unbounded and increasing function of t, and whose image contains the edge e. Observe that L is the set of points x with w · x = w · p. By our choice of e and w, there is some t0 > 0 such that γ(t0) lies in the interior of the n-chain D. Since w · γ(t) is unbounded, but D is bounded, there is a point t1 > t0 with γ(t1) lying on the boundary Z + β of D. As w · γ(t1) > w · γ(0) and γ(0) = p, we see that γ(t1) 6∈ L, and therefore is a point of Z. But this implies that C ∩ Z 6= ∅, contradicting that Z is a chain in C c. We conclude that ζ = 0, which implies that ιn−2 is injective. (cid:3) 3. Complex nonarchimedean amoebas Let K be the field of complex Puiseux series in the variable s with ν(s) = 1. We use Jonsson's Theorem to study the complement of the nonarchimedean amoeba T of a variety V in (K×)n. By Proposition 1.3, there is a smooth curve C, a point o ∈ C, a local parameter u at o, and a family V ⊂ (C r {o}) × (C×)n with lim a→o −1 log u(a) A (Va) = T . If V has codimension r, then V and its fibers have codimension r, and if in addition V is a complete intersection, then so are the fibers Va of V. Let a(t) for t positive and near 0 be the analytic arc defined by u(a(t)) = t. If we set then we have At := −1 log t A (Va(t)) , lim t→0 At = T . (6) In particular, for every point x ∈ T and every t sufficiently small, there is a point of At close to x. We need more, that these points of At lie in a given affine (k+1)-plane through x. The following technical lemma, whose proof we defer, guarantees this. It uses the weak higher convexity of the scaled amoebas At. Lemma 3.1. Suppose that L is an affine (k+1)-plane that is transverse to T . For every ǫ > 0, there is a δ > 0 such that for every x ∈ L ∩ T and 0 < t < δ we have We deduce our main result. d(x , L ∩ At) < ǫ . Theorem 3.2. The complement T c of the nonarchimedean amoeba T of a variety V ⊂ (K×)n of codimension k+1 is weakly k-convex. If V is a complete intersection, then the complement is k-convex. HIGHER CONVEXITY OF TROPICAL VARIETIES 9 Remark 3.3. Both Lemma 3.1 and Theorem 3.2 use only that T is a polyhedral complex of pure codimension k+1 that is the Kuratowski limit of a family {At t > 0} whose com- plements are (weakly) k-convex. Thus we have proven a stronger result about polyhedral complexes that are limits of sets whose complements are k-convex. Proof of Theorem 3.2. It suffices to use affine (k+1)-planes L that meet T transversally to test the k-convexity of T c. Let L be an oriented affine (k+1)-plane that meets T transversally and let ζ ∈ eHk(L ∩ T c) be a cycle. Let {xp p ∈ Π} be the finite set L ∩ T . Let ǫ > 0 be such that if βp is the closed ball of radius ǫ in L centered at xp for p ∈ Π, then these balls are disjoint. If γp := ∂βp for its cone of positive cycles. Thus there is an integer combination γ of the γp with [γ] = ζ. p ∈ Π, then the cycle classes of {γp p ∈ Π} are a basis for eHk(L ∩ T c) and they span Suppose that ιk(ζ) = 0 in eHk(T c). Then there is a (k+1)-chain Z in T c with ∂Z = γ. Let At for t positive and near 0 be a family of scaled amoebas (6) of varieties of codimension k+1 that converges to T . As Z is compact, Lemma 1.1 implies that there is some δ > 0 such that Z is disjoint from At for all 0 < t < δ. As [∂Z] = ζ, we conclude that ιk(ζ) = 0 in eHk(A c t ) for any 0 < t < δ. By Lemma 3.1, after possibly shrinking δ, if p ∈ Π and 0 < t < δ, then d(xp, L∩At) < ǫ. Thus for all 0 < t < δ, the scaled amoeba At meets each ball βp and each sphere γp is disjoint from At, as γp ⊂ Z. In particular, the cycle classes [γp] for p ∈ Π are linearly t ) and they span a subset of its positive cone. t of the scaled amoeba is t ) and so it is not a positive integer combination of the [γp]. Consequently, ζ is not a positive class of independent in eHk(L ∩ A c By Henriques' Theorem [4, Thm. 4.1], the complement A c weakly k-convex. Since ιk(ζ) = 0, ζ is not in the positive cone of eHk(L ∩ A c eHk(L ∩ T c). This shows that T c is weakly k-convex. Now suppose that V is a complete intersection. By Proposition 1.3, we may assume that At is the amoeba of a complete intersection, and so by the Theorem of Bushueva and t ), we conclude that ζ = 0 in eHk(L∩ A c Tsikh [1], A c t ). t Since the cycle classes [γp] are linearly independent in both eHk(L ∩ A c t ) and eHk(L ∩ T ), we conclude that ζ = 0 in eHk(L ∩ T ), and so T c is k-convex. is k-convex. As ιk(ζ) = 0 in eHk(A c Proof of Lemma 3.1. The tropical variety T admits the structure of a finite polyhedral complex of dimension n−k−1. Since L is has dimension (k+1) and is transverse to T , it meets T only in polyhedra of maximal dimension. Let {σp p ∈ Π} be the set of maximal polyhedra of T meeting L. If we set xp := L ∩ σp, then {xp p ∈ Π} = L ∩ T . Since T ∩ L is finite, it suffices to prove the lemma for a single point xp ∈ L ∩ T . (cid:3) Shrinking ǫ if necessary, we may assume that if σ is a polyhedron of T , then (1) If σ does not meet L, then d(L, σ) > 2ǫ, (2) If σ 6= σp, then d(xp, σ) > 2ǫ. A plane L′ is parallel to L if L′ = L + v for some v ∈ Rn. By Assumption (1), if L′ is parallel to L with d(L′, L) ≤ 2ǫ, then L′ meets σp transversally in a point x′ p in the relative interior of σp. Let θ be the minimum angle between σ and L. Then π/2 ≥ θ > 0, and if L′ is parallel to L with d(L′, L) ≤ ρ with ρ ≤ 2ǫ, then d(x′ p, xp) < ρ/ sin θ. Let β ⊂ L be 10 MOUNIR NISSE AND FRANK SOTTILE the ball of radius ǫ centered at xp and γ = ∂β the corresponding sphere. Observe that d(γ, σp) = ǫ sin θ. Let Z be the union of translates γ + (z − xp) where z ∈ σp with d(z, xp) ≤ ǫ. Then Z is compact and it lies in the closed ball centered at xp of radius 2ǫ. By construction, Z ∩ σp = ∅, and by Assumption (2), Z meets no other polyhedron of T , and is therefore disjoint from T . By the definition of Kuratowski limit and Lemma 1.1, there is a δ > 0 such that if 0 < t < δ then d(x, At) < 1 2ǫ sin θ and At is disjoint from Z. Fix a positive t < δ. Let y′ ∈ At be a point with d(xp, y′) < 1 2 ǫ sin θ. Set x′ L + (y′ − xp), a plane parallel to L with d(L′, L) < 1 d(x′ p, xp) < ǫ/2. Let Γ ⊂ Z be the cylinder 2 ǫ sin θ. Set L′ := p := L′ ∩ σp, then [{γ + t(x′ p − xp) t ∈ [0, 1]} . Observe that p of radius ǫ. Since d(x′ p, y′) ≤ d(x′ Thus y′ ∈ β′, where β′ = β + (x′ γ′ := ∂β′ ⊂ Z, it is disjoint from At, and so we have that 0 6= [γ′] ∈ eHk(L′ ∩ A c t ) p, xp) + d(xp, y′) < 1 p − xp) is the ball in L′ centered at x′ 2ǫ + 1 2 ǫ sin θ ≤ ǫ . is a positive cycle (we fix an orientation of L′). By Henriques' Theorem [4, Thm. 4.1], A c t is weakly k-convex, and so ιk[γ′] ∈ eHk(A c t ) is also non-zero. As γ′ is the boundary of the cycle Γ ∪ β, we canot have that Γ ∪ β lies in the complement A c t , we conclude that β meets At. As the boundary γ of β is disjoint from At, there is a point y ∈ At in the interior of β, which proves the lemma. (cid:3) t , and so (Γ ∪ β) ∩ At 6= ∅. Since Γ ⊂ A c We thank Mattias Jonsson and Eric Katz for their help in understanding their work. Acknowledgements References [1] N. A. Bushueva and A. K. Tsikh, On amoebas of algebraic sets of higher codimension, Proc. Steklov Inst. Math. 279 (2012), 52 -- 63. [2] I. M. Gel′fand, M. M. Kapranov, and A. V. Zelevinsky, Discriminants, resultants, and multidimen- sional determinants, Birkhauser Boston Inc., Boston, MA, 1994. [3] M. Gromov, Sign and geometric meaning of curvature, Rend. Sem. Mat. Fis. Milano 61 (1991), 9 -- 123 (1994). [4] A. Henriques, An analogue of convexity for complements of amoebas of varieties of higher codimen- sion, Advances in Geometry 4 (2004), no. 1, 61 -- 73. [5] Mattias Jonsson, Degenerations of amoebae and Berkovich spaces, Mathematische Annalen (2015), 1 -- 19. [6] Eric Katz, Tropical realization spaces for polyhedral complexes, Algebraic and combinatorial aspects of tropical geometry, Contemp. Math., vol. 589, Amer. Math. Soc., Providence, RI, 2013, pp. 235 -- 251. [7] Diane Maclagan and Bernd Sturmfels, Introduction to Tropical Geometry, Graduate Studies in Math- ematics, vol. 161, American Mathematical Society, Providence, RI, 2015. HIGHER CONVEXITY OF TROPICAL VARIETIES 11 [8] Grigory Mikhalkin, Decomposition into pairs-of-pants for complex algebraic hypersurfaces, Topology 43 (2004), no. 5, 1035 -- 1065. [9] Grigory Mikhalkin and Johannes Rau, Tropical geometry, 2014, Draft of a book. [10] M. Nisse and F. Sottile, Higher convexity of coamoeba complements, 2014, arXiv.org/1405.4900. [11] , Non-archimedean coamoebae, Tropical and Non-Archimedean Geometry, Contemporary Mathematics, vol. 605, American Mathematical Society, 2014, pp. 73 -- 91. [12] H. Rullgard, Polynomial amoebas and convexity, Tech. report, Stockholm University, 2001, Research Reports in Mathematics, No. 8. [13] David Speyer and Bernd Sturmfels, The tropical Grassmannian, Adv. Geom. 4 (2004), no. 3, 389 -- 411. School of Mathematics KIAS, 87 Hoegiro Dongdaemun-gu, Seoul 130-722, South Korea. E-mail address: [email protected] Department of Mathematics, Texas A&M University, College Station, Texas, USA E-mail address: [email protected] URL: www.math.tamu.edu/~sottile
1901.00839
2
1901
2019-01-07T11:35:05
Elliptic Gromov-Witten Invariants of Del-Pezzo Surfaces
[ "math.AG" ]
We obtain a formula for the number of genus one curves with a variable complex structure of a given degree on a del-Pezzo surface that pass through an appropriate number of generic points of the surface. This is done using Getzler's relationship among cohomology classes of certain codimension 2 cycles in $\overline{M}_{1,4}$ and recursively computing the genus-one Gromov-Witten invariants of del Pezzo surfaces. Using completely different methods, this problem has been solved earlier by Bertram and Abramovich, Ravi Vakil, Dubrovin and Zhang and more recently using Tropical geometric methods by M. Shoval and E. Shustin. We also subject our formula to several low degree checks and compare them to the numbers obtained by the earlier authors.
math.AG
math
ELLIPTIC GROMOV-WITTEN INVARIANTS OF DEL-PEZZO SURFACES CHITRABHANU CHAUDHURI AND NILKANTHA DAS Abstract. We obtain a formula for the number of genus one curves with a variable complex structure of a given degree on a del-Pezzo surface that pass through an appropriate number of generic points of the surface. This is done using Getzler's relationship among cohomology classes of certain codimension 2 cycles in M 1,4 and recursively computing the genus one Gromov-Witten invariants of del-Pezzo surfaces. Using completely different methods, this problem has been solved earlier by Bertram and Abramovich ([3]), Ravi Vakil ([25]), Dubrovin and Zhang ([8]) and more recently using Tropical geometric methods by M. Shoval and E. Shustin ([24]). We also subject our formula to several low degree checks and compare them to the numbers obtained by the earlier authors. Contents Introduction 1. 2. Main Result 3. Recursive formula 4. Del-Pezzo surfaces 5. Basic Strategy 6. Axioms for Gromov-Witten Invariants 7. 8. Low degree checks Acknowledgements References Intersection of cycles 1 3 3 4 4 6 6 9 10 10 One of the most fundamental problems in enumerative algebraic geometry is: 1. Introduction d , the number of genus g degree d curves in CP2 (with a variable complex Question 1.1. What is E(g) structure) that pass through 3d − 1 + g generic points? Although the computation of E(g) (even for genus zero) was unknown until the early 90 Manin ([17]) obtained a formula for E(0) d . d is a classical question, a complete solution to the above problem (cid:48)s when Ruan -- Tian ([23]) and Kontsevich -- d The computation of E(g) is now very well understood from several different perspectives. The formula by Caporasso -- Harris [6], computes E(g) for all g and d. Since then, the computation of E(g) d has been studied from many different perspectives; these include (among others) the algorithm by Gathman ([10], [11]) and the method of virtual localization by Graber and Pandharipande ([15]) to compute the genus g Gromov-Witten invariants of CPn (although for n > 2 and g > 0, the Gromov-Witten invariants are not enumerative). More recently, the problem of computing E(g) d has been studied using the method of tropical geometry by Mikhalkin in [19] (using the results of that d 2010 Mathematics Subject Classification. 14N35, 14J45. 1 2 C. CHAUDHURI AND N. DAS paper, one can in principle compute E(g) A more general situation is as follows: let X be a projective manifold and and β ∈ H2(X; Z) a given homology class. Given cohomology classes µ1, . . . , µk ∈ H∗(X, Q), the k-pointed genus g for all g and d). Gromov-Witten invariant of X is defined to be d k(µk) (cid:94) (cid:2)M g,k(X, β)(cid:3)vir , (1.1) N (g) β,X (µ1, . . . , µk) := ev∗ 1(µ1) (cid:94) . . . (cid:94) ev∗ (cid:90) M g,k(X,β) (cid:2)M g,k(X, β)(cid:3)vir where M g,k(X, β) denotes the moduli space of genus g stable maps into X with k marked points representing β and evi denotes the ith evaluation map. For g = 0, this is a smooth, irreducible and proper Deligne-Mumford stack and has a fundamental class. However, for g > 0, M g,k(X, β) is not smooth or irreducible, hence it does not posses a fundamental class. Behrend, Behrend-Fantechi and Li-Tian, have however defined the virtual fundamental class ∈ H 2Θ(M g,k(X, β)), Θ := c1(T X) · β + (dim X − 3)(1 − g) + k; which is used to define the Gromov-Witten invariants (see [4],[5] and [18]). When all the µ1, . . . , µk represent the class Poincare dual to a point (and the degree of the cohomology class that is be- ing paired in (1.1), is equal to the virtual dimension of the moduli space), then we abbreviate N (g) β,X (µ1, . . . , µk) as N (g) β . The number of genus g curves of degree β in X, that pass through c1(T X)· β + (dim X − 3)(1− g) generic points is denoted by E(g) is not necessarily equal to N (g) β , i.e. the Gromov-Witten invariant is not necessarily enumerative (this happens for example when X := CP3 and g = 1). β . In general, E(g) β An important class of surfaces for which the enumerative geometry is particularly important are Fano surfaces, which are also called del-Pezzo surfaces (see section 4 for the definition of a del-Pezzo surface). When g = 0, it is proved in ([14], Theorem 4.1, Lemma 4.10) that for del-Pezzo surfaces N (0) β = E(0) β . β = E(g) In [25], Vakil generalizes the approach of Caporasso-Harris in [6] to compute the numbers E(g) for all g and β for del-Pezzo surfaces. It is also shown in ([25], Section 4.2) that all the genus β g Gromov-Witten invariants of del-Pezzo surfaces are enumerative (i.e. N (g) β ). The enumer- ative geometry of del-Pezzo surfaces has also been studied extensively by Abramovich and Bertram (in [3]). More recently, this question has been approached using methods of tropical geometry. In [24], M. Shoval and E. Shustin give a formula to compute all the genus g Gromov-Witten invariants of del-Pezzo surfaces using methods of tropical geometry. The genus one Gromov-Witten invariants of CPn can also be computed from a completely different method from the ones developed in [10], [11] and [15]. In [12], Getzler finds a relationship among certain codimension two cycles in M 1,4 and uses that to compute the genus one Gromov- Witten invariants of CP2 and CP3. In [9], using ideas from Physics, Eguchi, Hori and Xiong made a remarkable conjecture concerning the genus g Gromov-Witten invariants of projective manifolds; this is known as the Virasoro conjecture. The conjecture in particular produces an explicit formula (for CP2), which aprori looks very different from the formula obtained by Getzler (in [12]). for N (1) It is shown by Pandharipande (in [21]), that the formula obtained by Getzler for CP2 is equivalent to a completely different looking formula predicted in [9]. d In this paper, we extend the approach of Getzler to compute the genus one Gromov-Witten invariants of del-Pezzo surfaces. The formula we obtain has a completely different appearance from the one obtained by Vakil in [25]. We verify that our final numbers are consistent with the numbers he obtains (see section 8 for details). The Virasoro conjecture for projective manifolds (which is conjectured in [9]) has been a topic of active research in mathematics for the last twenty years. In [8], Dubrovin and Zhang compute ELLIPTIC GROMOV-WITTEN INVARIANTS OF DEL-PEZZO SURFACES 3 the genus one Gromov-Witten invariants of CP1 × CP1 by showing that it follows from the Vi- rasoro conjecture. We have verified that our numbers agree with all the numbers computed by them ([8], Page 463). They prove that the genus zero and genus one Virasoro Conjecture is true for all projective manifolds having semi-simple quantum cohomology. It is proved in [1] that the quantum cohomology of del-Pezzo surfaces is semi simple. It would be interesting to see if one can use the result of this paper and apply the method of [21] to obtain a formula for the genus one Gromov-Witten invariants of del-Pezzo surfaces, analogous to the one predicted for CP2 by Eguchi, Hori and Xiong (in [9]). That would give a direct confirmation of the Virasoro conjecture in genus one for del-Pezzo surfaces. A detailed survey of the Virasosro conjecture is given in [13]. The main result of this paper is the following: 2. Main Result Main Result. Let X be a del-Pezzo surface and β ∈ H2(X, Z) be a given effective homology class. We obtain a formula for N (1) (equation (3.1)) using Getzler's relation β Remark. We note that by ([25], Section 4.2), we conclude that N (1) that N (1) follows from ([26], Theorem 1.1). β = E(1) β β = E(1) β . Alternatively, we note Our formula for N (1) β . The latter can be computed via the algorithm given in [17] and [14]. The base case of our recursive formula are given by equations (3.2) and (3.3). We have written a C++ program that implements (3.1); it is available on our web page: is a recursive formula, involving N (0) β http://www.iiserpune.ac.in/~chitrabhanu/. 3. Recursive formula We will now give the recursive formula to compute N (1) β . First, we will develop some notation that is used throughout this paper. Let ξX κβ b2(X) dX := c1(T X), := ξX · β, := dim H2(X, Q), := ξX · ξX , for both the cohomology class and the divisor, where β ∈ H2(X, Z), the second betti number of X, the degree of X. Moreover, · is used for both the cup product in cohomology as well as cap product between a homology and a cohomology class. We are now ready to state the formula. First, let us define the following four quantities: (cid:88) T1 := β1+β2+β3=β (cid:88) β1+β2=β T2 := 2κβ2κ2 κβ − 2 β3(β1 · β2) κβ2 − 1, κβ3 − 1 (cid:19) (cid:18) (cid:18)(cid:0)4κβ1 + κβ2 − 2κβ3 (cid:18) (cid:19) (cid:20) (cid:18) κβ − 2 (cid:18)κβ − 2 (cid:19) (cid:18) 2κβ1κβ2 − κ2 dX (β1 · β2)(cid:0)4κβ1 + κβ2 (cid:19) (cid:1)(β2 · β3) − 3κβ2(β1 · β3) (cid:19) (cid:1) + 2κβ1κβ2 β2 − 3dX (β1 · β2) κβ1 − 1 4κ2 β2 2κβ2 κβ1 N (1) β1 N (0) β2 N (0) β3 , + (cid:1)(cid:19)(cid:21) (cid:0)2κβ1 − κβ2 N (1) β1 N (0) β2 , 4 (cid:88) β1+β2=β T3 := − 1 12 (cid:19) (cid:18) κβ − 2 κβ1 − 1 C. CHAUDHURI AND N. DAS (cid:20) (cid:18) κ2 β2(β1 · β2) κ2 β1 κβ1 − 2κβ2 − 6(β1 · β2) (cid:18) (cid:19) (cid:19)(cid:21) + κβ2(β1 · β1) 4κβ1 + κβ2) N (0) β1 N (0) β2 , (cid:18) (cid:19) N (0) β . 1 12 κ3 β (2 + b2(X))κβ − dX T4 := − The number N (1) β satisfies the following recursive relation: 6d2 X N (1) (3.1) We will now give the initial conditions for the recursion (3.1). Let X be P2 blown up at upto k = 8 points. Then the initial condition of the recursion is N (1) Ei β = T1 + T2 + T3 + T4. L = 0 N (1) (3.2) and = 0 Here L denotes the class of a line and Ei denotes the exceptional divisors. If X := P1 × P1, then Here e1 and e2 denote the class of [pt × P1] and [P1 × pt] respectively. The initial conditions (3.2) obtained from [17] and [14], give us the values of N (1) and (3.3), combined with the values of N (0) for any β. e2 = 0. e1 = 0 N (1) N (1) (3.3) and ∀i = 1 to k. β β Remark. We would like to mention that the formula (3.1) yields Getzler's recursion relation, equation (0.1) of [12], after some symmetrization of the summation indices of T1 and T3. A del-Pezzo surface X is a smooth projective algebraic surface with an ample anti-canonical divisor ξX . The degree of the surface is defined to be the self-intersection number 4. Del-Pezzo surfaces dX = ξX · ξX . This degree dX varies between 1 and 9. X can be obtained as a blow-up of P2 at k = 9− dX general points, except, when dX = 8 the surface can also be P1 × P1. If X has degree 9 − k and is not P1 × P1, then we have the blow up morphism Bl : X → P2. We denote by E1, . . . , Ek the exceptional divisors of Bl and by L the pull-back of the class of a hyperplane in P2. We have H 2(X, Z) = Z(cid:104)L, E1, . . . , Ek(cid:105), and L·L = 1, Ei·Ei = −1, L·Ei = Ei·Ej = 0 for all i, j ∈ {1, . . . , k} with i (cid:54)= j. The anti-canonical divisor is given by ξX = 3L − E1 − . . . − Ek. If X = P1 × P1, let e1 = pr∗ 2[pt], then ξX = 2e1 + 2e2, e1 · e2 = 1 and e2 · e1 = 1 whereas ei · ei = 0 for i = 1, 2. 1[pt] and e2 = pr∗ 5. Basic Strategy We will now recall the basic setup of [12], where Getzler computes the number N (1) d when X is CP2. First, let us consider the space M 1,4, the moduli space of genus one curves with four marked points. We shall be interested in certain S4 invariant codimension 2 boundary strata in M 1,4 which we list in Figure 1. In the figure we draw the topological type and the marked point distribution of the generic curve in each strata. We use the same nomenclature as [12] except for δ0,0 which was denoted by δβ in [12], (to avoid confusion between notations). See section 1 of [12] for a list of all the codimension 2 strata. There the strata are denoted by the dual graph of the generic curve. ELLIPTIC GROMOV-WITTEN INVARIANTS OF DEL-PEZZO SURFACES 5 δ0,3 δ2,2 δ0,0 δ2,4 δ0,4 δ2,3 δ3,4 Figure 1. Codimension 2 strata in M 1,4. These strata define cycles in H 4(M 1,4, Q). Let us now define the following cycle in H 4(M 1,4, Q), given by R := −2δ2,2 + 2 3 δ2,3 + 1 3 δ2,4 − δ3,4 − 1 6 δ0,3 − 1 6 δ0,4 + 1 3 δ0,0. The main result of [12] is that R = 0. This will subsequently be referred to as Getzler's relation. In [21], Pandharipande has shown that this relation, in fact, comes from a rational equivalence. Now we explain how to obtain our formula. Consider the natural forgetful morphism π : M 1,κβ +2(X, β) −→ M 1,4. mentary dimension; that will give us an equality of numbers and subsequently the formula. Let We shall pull-back the cycle R to H∗(M 1,κβ +2(X, β), Q) and intersect it with a cycle of comple- µ ∈ H 4(X, Q) be the class of a point. Define 1(ξX ) · . . . · ev∗ Z := ev∗ (cid:90) The class ξX is used since it is ample and hence numerically effective. Since R = 0 by Getzler's relation, we conclude that 4(ξX ) · ev∗ 5(µ) · . . . · ev∗ (cid:2)M 1,κβ +2(X, β)(cid:3)vir κβ +2(µ). (π∗ R · Z) · = 0. (5.1) M 1,κβ +2(X,β) We can also compute the left hand side of (5.1) using the composition axiom for Gromov-Witten invariants which will give us the recursive formula. •••••••••••••••••••••••••••• 6 C. CHAUDHURI AND N. DAS 6. Axioms for Gromov-Witten Invariants We shall make use of certain axioms for Gromov-Witten invariants. These are quite standard, see for example [7], however for completeness we list them here. We assume X is a smooth projective variety. Degree axiom: If deg µ1 + . . . + deg µn (cid:54)= 2n + 2κβ + 2(3 − dim X)(g − 1) then N (g) β,X (µ1, . . . , µn) = 0. Fundamental class axiom: If [X] is the fundamental class of X and 2g + n ≥ 4 or β (cid:54)= 0, then N (g) β,X ([X], µ1, . . . , µn−1) = 0. Divisor axiom: If D is a divisor of X and 2g + n ≥ 4. then N (g) β,X (D, µ1, . . . , µn−1) = (D · β)N (g) β,X (µ1, . . . , µn−1). Composition axiom: This is a bit complicated to write down, so we refer to [12], section 2.11. It is a combination of the splitting and reduction axioms of [17] section 2. We also need the following results which do not follow from the above axioms: (cid:90) X N (0) 0,X (µ1, µ2, µ3) = µ1 (cid:94) µ2 (cid:94) µ3, and N (1) 0,X (µ) = − 1 24 c1(T X) · µ. Now we are in a position to compute the left hand side of (5.1). Fix a homogeneous basis X γi (cid:94) γj and ((gij)) = ((gij))−1. For a cycle δ in 7. Intersection of cycles {γ1, . . . , γb(X)} of H∗(X, Q). Let gij = (cid:82) (cid:90) H∗(M g,n(X, β), Q), we introduce the following notation (cid:2)M g,n(X, β)(cid:3)vir . N δ β,X (µ1, . . . , µn) = δ · ev∗ 1(µ1)··· ev∗ n(µn) · M g,n(X,β) Let µ1 = . . . = µ4 = ξX , and µ5 = . . . = µκβ +2 = [pt] be the class of a point. If δ = π∗δ2,2, by the composition axiom β,X =N δ N δ β,X (µ1, . . . , µκβ +2) (cid:88) (cid:88) = β1+β2+β3=β A,B,C i,j,k,l gijgklN (1) β1,X (γi, γk, µαα ∈ A) × N (0) β2,X (γj, µαα ∈ B) × N (0) β3,X (γl, µαα ∈ C), where the second sum is over i, j, k, l ranging from 1 to b(X) and the first sum is over disjoint sets A, B, C satisfying A (cid:116) B (cid:116) C = {1, . . . , κβ + 2}, B ∩ {1, 2, 3, 4} = C ∩ {1, 2, 3, 4} = 2. Note that if β1, β2, β3 > 0, by the degree axiom the only non-trivial terms occur when A = κβ1,B = κβ2 + 1,C = κβ3 + 1. The limiting case β1 = 0 does not yield anything, however β2 = 0 or β3 = 0 have non-trivial contributions to the sum. When β3 = 0, β1, β2 > 0, the non-trivial contribution occurs precisely when C = 2, γl = [X], A = κβ1 − 1, γk = [pt], and B = κβ2 + 1. ELLIPTIC GROMOV-WITTEN INVARIANTS OF DEL-PEZZO SURFACES 7 Finally when β2 = β3 = 0, the only non-zero term occurs when B = C = 2, γl = γj = [X] and γk = γi = [pt]. Making use of the fact that for any σ, τ ∈ H∗(X, Q) b(X)(cid:88) b(X)(cid:88) gij(σ · γi)(γj · τ ) = (σ · τ ), we obtain the following expression i=1 j=1 β (cid:88) N π∗δ2,2 β,X =3(ξX · ξX )2N (1) (cid:88) β1+β2+β3=β + 3 + 6 β1+β2=β (cid:18) (cid:18) κβ − 2 (cid:88) (cid:88) κβ1 − 1 N π∗δ2,3 β,X = β1+β2+β3=β A,B,C i,j,k,l Next, let us consider the cycle δ2,3. We then have (cid:19) (β2 · ξX )2(β3 · ξX )2(β1 · β2)(β1 · β3)N (1) β1 κβ − 2 κβ2 − 1, κβ3 − 1 (cid:19) (ξX · ξX )(β1 · β2)(β2 · ξX )2N (1) β1 N (0) β2 . N (0) β2 N (0) β3 (7.1) gijgklN (1) β1,X (γi, µαα ∈ A) × N (0) β2,X (γj, γk, µαα ∈ B) × N (0) β3,X (γl, µαα ∈ C), where the sum is over sets A, B, C satisfying A (cid:116) B (cid:116) C = {1, . . . , κβ + 2}, A ∩ {1, 2, 3, 4} = B ∩ {1, 2, 3, 4} = 1. All the cases are similar to the previous calculation except, when β2 = 0. In this case we can either have B = 1,A = κβ1, γi = [pt] and γj = [X]; or B = 1,C = κβ3, γk = [X] and γl = [pt]. We get N π∗δ2,3 β,X =12 β1+β2+β3=β (cid:88) (cid:88) (cid:88) β1+β2=β β1+β2=β + 12 + 12 (cid:19) κβ − 2 κβ2 − 1, κβ3 − 1 (cid:18) (cid:19) (cid:18)κβ − 2 (cid:19) (cid:18) κβ − 2 (β1 · ξX )(β2 · ξX )3N (1) (cid:88) (cid:88) (β1 · ξX )(β2 · ξX ) κβ1 − 1 κβ1 β1 gijgklN (1) Moving on to δ2,4 we have N π∗δ2,4 β,X = β1+β2+β3=β A,B,C i,j,k,l β1,X (γi, µαα ∈ A) N (0) β2 . (cid:19) (β1 · ξX )(β2 · ξX )(β3 · ξX )2(β1 · β2)(β2 · β3)N (1) (cid:18) β1 N (0) β2 N (0) β3 (ξX · ξX )(β1 · β2) + (β1 · ξX )(β2 · ξX ) N (1) β1 N (0) β2 (7.2) × N (0) β2,X (γj, γk, µαα ∈ B) × N (0) β3,X (γl, µαα ∈ C), where the sum is over sets A, B, C satisfying A (cid:116) B (cid:116) C = {1, . . . , κβ + 2}, B ∩ {1, 2, 3, 4} = C ∩ {1, 2, 3, 4} = 2. 8 C. CHAUDHURI AND N. DAS Now there is no contribution when β2 = 0, however we have a non-trivial contribution when β1 = 0. We can use (6) to calculate this (cid:19) (β2 · ξX )2(β3 · ξX )2(β1 · β2)(β2 · β3)N (1) β1 N (0) β2 N (0) β3 N π∗δ2,4 β,X = 6 (cid:88) (cid:88) (cid:88) β1+β2+β3=β + 6 + 6 (cid:18) β1+β2=β β1+β2=β κβ2 − 1, κβ3 − 1 κβ − 2 (cid:18) (cid:18)κβ − 2 (cid:19) (cid:18) (cid:19)(cid:18) κβ − 2 κβ1 1 24 − κβ1 − 1 1 24 (cid:19) (cid:88) (cid:88) (ξX · β)3(ξX · ξX )N (0) β . + 6 − For δ3,4 we have N π∗δ3,4 β,X = gijgklN (1) β1,X (γi, µαα ∈ A) β1+β2+β3=β A,B,C i,j,k,l (β2 · ξX )2(ξX · ξX )(β1 · β2)N (1) (cid:19) (ξX · β1)3(β2 · ξX )2(β1 · β2)N (0) N (0) β2 β1 β1 N (0) β2 (7.3) where the first sum is over sets A, B, C satisfying × N (0) β2,X (γj, γk, µαα ∈ B) × N (0) β3,X (γl, µαα ∈ C), A (cid:116) B (cid:116) C = {1, . . . , κβ + 2}, The calculation is similar to the previous cases, so we omit the details. We obtain B ∩ {1, 2, 3, 4} = 1,C ∩ {1, 2, 3, 4} = 3. (cid:19) (β2 · ξX )(β3 · ξX )3(β1 · β2)(β2 · β3)N (1) β1 N (0) β2 N (0) β3 N π∗δ3,4 β,X = 4 κβ − 2 κβ2 − 1, κβ3 − 1 (cid:18) (cid:19) (cid:18)κβ − 2 (cid:18) κβ − 2 (cid:19) (β2 · ξX )3(β1 · ξX )N (1) (cid:19)(cid:18) (cid:18) κβ − 2 (β2 · ξX )4N (1) κβ1 − 1 N (0) β2 (cid:19) κβ1 β1 β1 1 24 − N (0) β2 (cid:88) (cid:88) (cid:88) (cid:88) β1+β2+β3=β β1+β2=β + 4 + 4 + 4 β1+β2=β (cid:19) (cid:18) β1+β2=β + 4 − 1 24 κβ1 − 1 (ξX · ξX )(β · ξX )3N (0) β . (cid:88) (cid:88) β1+β2=β A,B i,j,k,l N π∗δ0,3 β,X = 1 2 (ξX · β1)2(β2 · ξX )3(β1 · β2)N (0) β1 N (0) β2 (7.4) The remaining cycles all have 2 genus zero components so the calculations are simpler. We will first consider δ0,3: gijgklN (0) β1,X (γi, γj, γk, µαα ∈ A) where the first sum is over sets A, B satisfying A (cid:116) B = {1, . . . , κβ + 2}, A ∩ {1, 2, 3, 4} = 1. × N (0) β2,X (γl, µαα ∈ B), ELLIPTIC GROMOV-WITTEN INVARIANTS OF DEL-PEZZO SURFACES 9 The factor of 1 2 appears since the dual graph of a generic curve in δ0,3 has an automorphism of order 2. Neither β1 = 0, nor β2 = 0 has any non-trivial contribution so it is straight forward to see that N π∗δ0,3 β,X = (cid:88) β1+β2=β N π∗δ0,4 β,X = (cid:18) κβ − 2 (cid:88) κβ1 − 1 (cid:19) (β1 · ξX )(β2 · ξX )3(β1 · β2)(β1 · β1)N (0) (cid:88) β1 gijgklN (0) β1,X (γi, γj, γk, µαα ∈ A) 2 1 2 β1+β2=β A,B i,j,k,l The calculation for δ0,4 is a bit more subtle: N (0) β2 . (7.5) where the first sum is over sets A, B satisfying × N (0) β2,X (γl, µαα ∈ B), A (cid:116) B = {1, . . . , κβ + 2}, A ∩ {1, 2, 3, 4} = ∅. Contribution from β2 = 0 is 0. When β1 = 0, we must have A = ∅ which leads to N (0) β2 N π∗δ0,4 β,X = (cid:19) (β2 · ξX )4(β1 · β2)(β1 · β1)N (0) (cid:18) κβ − 2 (cid:88) 1 2 β1 (7.6) Finally, let us consider the cycle δ0,0: κβ1 − 1 β1+β2=β 1 2 + (2 + b2(X))(β · ξX )4N (0) β . (cid:88) (cid:88) gijgklN (0) β1+β2=β A,B i,j,k,l N π∗δ0,0 β,X = 1 2 β1,X (γi, γk, µαα ∈ A) where the first sum is over sets A, B satisfying × N (0) β2,X (γj, γl, µαα ∈ B), By an analogous calculation as the previous situations we have A (cid:116) B = {1, . . . , κβ + 2}, A ∩ {1, 2, 3, 4} = 2. (cid:88) β1+β2=β (cid:19) (cid:18) κβ − 2 κβ1 − 1 N π∗δ0,0 β,X = 3 2 (β1 · ξX )2(β2 · ξX )2(β1 · β2)2N (0) β1 N (0) β2 . (7.7) Now collecting all these terms and using relation (5.1) we obtain the desired formula (3.1). 8. Low degree checks We will now describe some concrete low degree checks that we have performed. Let Xk be a del-Pezzo surface obtained by blowing up P2 at k ≤ 8 points. It is a classical fact that N (1) dL+σ1E1+...+σrEr,Xk dL+σ1E1+...+σr−1Er−1,Xk−1 , = N (1) if σr is −1 (Qi [22]) or 0 (Theorem 1.3 of Hu [16]). We give a self contained reason for this assertion in our special case. Consider X1 which is P2 blown up at the point p. Let us consider the number N (1) ; this is the number of genus one curves in X1 representing the class dL− E1 and passing dL−E1,X1 through 3d − 1 generic points. Let C be one of the curves counted by the above number. The curve C intersects the exceptional divisor exactly at one point. Furthermore, since the 3d − 1 points are generic, they can be chosen not to lie in the exceptional divisor; let us call the points p1, p2, . . . , p3d−1. Hence, when we consider the blow down from X1 to P2, the curve C becomes a curve in P2 passing through p1, p2, . . . , p3d−1 and the blow up point p. We thus get a genus one, degree d curve in P2 passing through 3d points. There is a one to one correspondence between 10 C. CHAUDHURI AND N. DAS curves representing the class dL − E1 in X1 passing through 3d − 1 points and degree d curves in P2 passing through 3d points. Hence N (1) dL,P2. A similar argument holds when there are more than one blowup points. The same argument also shows that N (1) dL,P2; the same reasoning holds by taking a curve in the blowup and then considering its image under the blow down. The blow down gives a one to one correspondence between the two sets and hence, the corresponding numbers are the same. dL−E1,X1 = N (1) = N (1) dL+0E1,X1 We have verified this assertion in many cases. For instance we have verified that N (1) 5L−E1−E2,X2 = N (1) 5L−E1,X1, = N (1) 5L+0E1,X1 = N (1) 5L,P2 . dL+σ1E1+...+σrEr The reader is invited to use our program and verify these assertions. Hence without ambiguity we write N (1) P1 × P1; our numbers agree with the numbers he has listed in his paper (Page 463). Finally, in [25], Ravi Vakil has explicitly computed some N (g) Next, we note that in [8], Dubrovin has computed the genus one Gromov-Witten Invariants of for del-Pezzo surfaces (Page 78). dL+σ1E1+...+σrEr,Xr for N (1) . β Our numbers agree with the following numbers he has listed: = 225, N (1) and N (1) = 13775, N (1) = 240, N (1) 5L−2E1−2E2−2E3 = 20 N (1) N (1) 5L−2E1 5L−3E1 5L−3E1−2E2 5L−2E1−2E2−2E3−2E4 5L−3E1−2E2−2E3 = 1. = 20, Acknowledgements We would like to thank Ritwik Mukherjee for several fruitful discussions. The second author is indebted to Ritwik Mukherjee specially for suggesting the project and spending countless hours of time for discussions. The first author is also very grateful to ICTS for their hospitality and conducive atmosphere for doing mathematics research; he would specially like to acknowledge the program Integrable Systems in Mathematics, Condensed Matter and Statistical Physics (Code: ICTS/integrability2018/07) where a significant part of the project was carried out. We are also grateful to Ritwik Mukherjee for mentioning our result in the program Complex Algebraic Geometry (Code: ICTS/cag2018), which was also organized by ICTS. The first author was supported by the DST-INSPIRE grant IFA-16 MA-88 during the course of this research. Finally, the second author would like to thank Ritwik Mukherjee for supporting this project through the External Grant he has obtained, namely MATRICS (File number: MTR/2017/000439) that has been sanctioned by the Science and Research Board (SERB). References [1] Arend Bayer and Yuri I. Manin, (Semi)simple exercises in quantum cohomology, The Fano Conference, Univ. Torino, Turin, 2004, pp. 143 -- 173. 175 -- 203. [2] Somnath Basu and Ritwik Mukherjee, Enumeration of curves with one singular point, J. Geom. Phys. 104 (2016), [3] Dan Abramovich and Aaron Bertram, The formula 12 = 10 + 2 × 1 and its generalizations: counting rational curves on F2, Advances in algebraic geometry motivated by physics (Lowell, MA, 2000), Contemp. Math., vol. 276, Amer. Math. Soc., Providence, RI, 2001, pp. 83 -- 88. [4] K. Behrend, Gromov-Witten invariants in algebraic geometry, Invent. Math. 127 (1997), no. 3, 601 -- 617. [5] K. Behrend and B. Fantechi, The intrinsic normal cone, Invent. Math. 128 (1997), no. 1, 45 -- 88. [6] Lucia Caporaso and Joe Harris, Counting plane curves of any genus, Invent. Math. 131 (1998), no. 2, 345 -- 392. [7] David A. Cox and Sheldon Katz, Mirror symmetry and algebraic geometry, Mathematical Surveys and Mono- graphs, vol. 68, American Mathematical Society, Providence, RI, 1999. [8] Boris Dubrovin and Youjin Zhang, Frobenius manifolds and Virasoro constraints, Selecta Math. (N.S.) 5 (1999), no. 4, 423 -- 466. [9] Tohru Eguchi, Kentaro Hori, and Chuan-Sheng Xiong, Gravitational quantum cohomology, Internat. J. Modern Phys. A 12 (1997), no. 9, 1743 -- 1782. ELLIPTIC GROMOV-WITTEN INVARIANTS OF DEL-PEZZO SURFACES 11 [10] Andreas Gathmann, Relative Gromov-Witten invariants and the mirror formula, Math. Ann. 325 (2003), no. 2, 393 -- 412. [11] , The number of plane conics that are five-fold tangent to a given curve, Compos. Math. 141 (2005), no. 2, [12] E. Getzler, Intersection theory on M1,4 and elliptic Gromov-Witten invariants, J. Amer. Math. Soc. 10 (1997), 487 -- 501. no. 4, 973 -- 998. [13] , The Virasoro conjecture for Gromov-Witten invariants, Algebraic geometry: Hirzebruch 70 (Warsaw, 1998), Contemp. Math., vol. 241, Amer. Math. Soc., Providence, RI, 1999, pp. 147 -- 176. [14] L. Gottsche and R. Pandharipande, The quantum cohomology of blow-ups of P2 and enumerative geometry, J. Differential Geom. 48 (1998), no. 1, 61 -- 90. [15] T. Graber and R. Pandharipande, Localization of virtual classes, Invent. Math. 135 (1999), no. 2, 487 -- 518. [16] J. Hu, Gromov-Witten invariants of blow-ups along points and curves, Math. Z. 233 (2000), no. 4, 709 -- 739. [17] M. Kontsevich and Yu. Manin, Gromov-Witten classes, quantum cohomology, and enumerative geometry, Mirror symmetry, II, AMS/IP Stud. Adv. Math., vol. 1, Amer. Math. Soc., Providence, RI, 1997, pp. 607 -- 653. [18] Jun Li and Gang Tian, Virtual moduli cycles and Gromov-Witten invariants of algebraic varieties, J. Amer. Math. Soc. 11 (1998), no. 1, 119 -- 174. [19] Grigory Mikhalkin, Enumerative tropical algebraic geometry in R2, J. Amer. Math. Soc. 18 (2005), no. 2, 313 -- 377. [20] Ritwik Mukherjee, Enumerative geometry via topological computations, ProQuest LLC, Ann Arbor, MI, 2011. Thesis (Ph.D.) -- State University of New York at Stony Brook. [21] Rahul Pandharipande, A geometric construction of Getzler's elliptic relation, Math. Ann. 313 (1999), no. 4, 715 -- 729. [22] Xiaoxia Qi, A blow-up formula of high-genus Gromov-Witten invariants of symplectic 4-manifolds, Adv. Math. (China) 43 (2014), no. 4, 603 -- 607 (English, with English and Chinese summaries). [23] Yongbin Ruan and Gang Tian, A mathematical theory of quantum cohomology, J. Differential Geom. 42 (1995), no. 2, 259 -- 367. [24] Mendy Shoval and Eugenii Shustin, On Gromov-Witten invariants of del Pezzo surfaces, Internat. J. Math. 24 (2013), no. 7, 1350054, 44. [25] Ravi Vakil, Counting curves on rational surfaces, Manuscripta Math. 102 (2000), no. 1, 53 -- 84. [26] Aleksey Zinger, Reduced genus-one Gromov-Witten invariants, J. Differential Geom. 83 (2009), no. 2, 407 -- 460. School of Mathematics, IISER Pune E-mail address: [email protected] School of Mathematics, National Institute of Science Education and Research, Bhubaneswar (HBNI), Odisha 752050, India E-mail address: [email protected]
1506.01168
1
1506
2015-06-03T09:02:22
Numerical Adjunction Formulas for Weighted Projective Planes and Lattice Points Counting
[ "math.AG", "math.CO" ]
This paper gives an explicit formula for the Ehrhart quasi-polynomial of certain 2-dimensional polyhedra in terms of invariants of surface quotient singularities. Also, a formula for the dimension of the space of quasi-homogeneous polynomials of a given degree is derived. This admits an interpretation as a Numerical Adjunction Formula for singular curves on the weighted projective plane.
math.AG
math
NUMERICAL ADJUNCTION FORMULAS FOR WEIGHTED PROJECTIVE PLANES AND LATTICE POINTS COUNTING JOS´E IGNACIO COGOLLUDO-AGUST´IN, JORGE MART´IN-MORALES, AND JORGE ORTIGAS-GALINDO Abstract. This paper gives an explicit formula for the Ehrhart quasi-poly- nomial of certain 2-dimensional polyhedra in terms of invariants of surface quotient singularities. Also, a formula for the dimension of the space of quasi- homogeneous polynomials of a given degree is derived. This admits an in- terpretation as a Numerical Adjunction Formula for singular curves on the weighted projective plane. 1. Introduction This paper deals with the general problem of counting lattice points in a poly- hedron with rational vertices and its connection with both singularity theory of surfaces and Adjunction Formulas for curves in the weighted projective plane. In addition, we focus on rational polyhedra (whose vertices are rational points) as opposed to lattice polyhedra (whose vertices are integers). Our approach exploits the connexion between Dedekind sums (as originated from the work of Hirzebruch- Zagier [17]) and the geometry of cyclic quotient singularities, which has been pro- posed by several authors (see e.g. [22, 7, 10, 18, 27, 11, 4, 5]). According to Ehrhart [16], the number of integer points of a lattice (resp. ratio- nal) polygon P and its dilations dP = {dp p ∈ P} is a polynomial (resp. quasi- polynomial) in d of degree dimP referred to as the Ehrhart (quasi)-polynomial of P (cf. [12]). In this paper we focus on the Ehrhart quasi-polynomial of polygons of type (1) dDw = Dw,d = {(x, y, z) ∈ R3 x, y, z ≥ 0, w0x + w1y + w2z = d}, where w = (w0, w1, w2) are pairwise coprime and (2) Dw := {(x, y, z) ∈ R3 x, y, z ≥ 0, w0x + w1y + w2z = 1} is a rational polygon. In Theorem 1.1 we give an explicit formula for the Ehrhart quasi-polynomial of (1), which in Theorem 1.2 is shown to be an invariant of the quotient singularities of the weighted projective plane P2 w. Throughout this paper w0, w1, w2 are assumed to be pairwise coprime integers. Denote by w = (w0, w1, w2), ¯w = w0w1w2, and w = w0 + w1 + w2. Finally, the key ingredients to connect the arithmetical problem referred to above with the geometry of weighted projective planes come from the observation that (3) Lw(d) := #(Dw,d ∩ Z3) = h0(P2 w;O(d)), 2010 Mathematics Subject Classification. 32S05, 32S25, 52B20, 11F20. Key words and phrases. Quotient surface singularity, invariants of curve singularities, Ehrhart polynomial, rational polytope. All authors are partially supported by the Spanish Ministry of Education MTM2013-45710- C2-1-P and E15 Grupo Consolidado Geometr´ıa from the Gobierno de Arag´on. The second author is also supported by FQM-333 from Junta de Andaluc´ıa. 1 However, for a general d the generic curve of degree d in P2 w is not necessarily quasi-smooth (see [14, Lemma 5.4]). The final goal of this paper is to revisit (5) in the general (singular) case. Let us present the main results of this work. The first main statement shows an explicit formula for the Ehrhart quasi-polynomial Lw(d) of degree two of Dw,d in terms of d. Theorem 1.1. The Ehrhart quasi-polynomial Lw(d) for the polygon Dw in (2) satisfies Lw(d) = gw,d+w − ∆P (d + w). (cid:88) P∈Sing(P2 w) k(k − w) 2 J.I. COGOLLUDO, J. MART´IN, AND J. ORTIGAS that is, the dimension of the vector space of weighted homogeneous polynomials of degree d, and from a Numerical Adjunction Formula relating h0(P2 w;O(d)) with the genus of a curve in P2 w. To explain what we mean by Numerical Adjunction Formulas, assume a quasi- smooth curve C ⊂ P2 w of degree d exists. In that case, according to the classical Adjunction Formula one has the following equality relating canonical divisors on C and P2 (4) KC = (KP2 w w + C)C. Equating degrees on both sides of (4) and using the Weighted B´ezout's Theorem, one has 2g(C) − 2 = deg(KP2 w + C)C = deg(C) deg(KP2 w + C) ¯w = d(d − w) ¯w . Notice that the generic curve of degree k ¯w is smooth (see [14, Lemma 5.4]). In that case, one has (c.f. §4) (5) h0(P2 t(t − w) 2 ¯w where gw,t := w;O(k ¯w − w)) = Lw(k ¯w − w) = gw,k ¯w, + 1. The quadratic polynomial gw,k = + 1 in k is called the virtual genus (see [14, Definition 5.1]) and ∆P (k) is a periodic function of period ¯w which is an 2 ¯w invariant associated to the singularity P ∈ Sing(P2 w) (see Definition 3.8). The proof of Theorem 1.1 relies heavily on computations with Dedekind sums. The next result aims to show that the previous combinatorial number ∆P (k) has a geometric interpretation and can be computed via invariants of curve singularities on a singular surface. In order to do so we recall the recently defined invariant δP (f ) of a curve ({f = 0}, P ) on a surface with quotient singularity (see [14, Section 4.2]) and we define a new invariant κP (f ) in §3.1. Theorem 1.2. Let (f, P ) be a reduced curve germ at P ∈ X a surface cyclic quotient singularity. Then for any reduced germ f ∈ OX,P (k). ∆P (k) = δP (f ) − κP (f ) tion 2.3. The module OX,P (k) of k-invariant germs of X at P can be found in Defini- As an immediate consequence of Theorems 1.1 and 1.2 one has a method to com- pute Lw(d) by means of appropriate curve germs ({f = 0}, P ) on surface quotient singularities. In an upcoming paper we will study the ∆P (k)-invariant by means of singularity theory and intersection theory on surface quotient singularities in order to give a closed effective formula for the Ehrhart quasi-polynomial Lw(d). In fact, NUMERICAL ADJUNCTION FORMULAS... 3 Theorem 1.1 can also be seen as a version of [7], where an explicit interpretation of the correction term is given in Theorem 1.2. invariant κP . w relating h0(P2,OP2 Finally, we generalize the Numerical Adjunction Formula for a general singular w (d − w)), its genus g(C), and the newly defined curve C on P2 Theorem 1.3 (Numerical Adjunction Formula). Consider C = {f = 0} ⊂ P2 w an irreducible curve of degree d, then (cid:88) h0(P2 w,OP2 w (d − w)) = g(C) + κP (f ). P∈Sing(C) This paper is organized as follows. In §2 some basic definitions and preliminary results on surface quotient singularities, logarithmic forms, and Dedekind sums are given. In §3, after defining the three local invariants mentioned above, a proof of Theorem 1.2 is given. An introductory example is treated in §4 and finally, the main results Theorem 1.1 and Theorem 1.3 are proven in §5. 2. Definitions and Preliminaries In this section some needed definitions and results are provided. 2.1. V -manifolds and Quotient Singularities. We start giving some basic def- initions and properties of V -manifolds, weighted projective spaces, embedded Q- resolutions, and weighted blow-ups (for a detailed exposition see for instance [15, 2, 3, 19, 21]). Let us fix the notation and introduce several tools to calculate a special kind of embedded resolutions, called embedded Q-resolutions (see Definition 2.4), for which the ambient space is allowed to contain abelian quotient singularities. To do this, we study weighted blow-ups at points. Definition 2.1. A V -manifold of dimension n is a complex analytic space which admits an open covering {Ui} such that Ui is analytically isomorphic to Bi/Gi where Bi ⊂ Cn is an open ball and Gi is a finite subgroup of GL(n, C). We are interested in V -surfaces where the quotient spaces Bi/Gi are given by (finite) abelian groups. Let Gd ⊂ C∗ be the cyclic group of d-th roots of unity generated by ξd. Consider a vector of weights (a, b) ∈ Z2 and the action −→ (cid:55)→ (ξa The set of all orbits C2/Gd is called a cyclic quotient space of type (d; a, b) and it is denoted by X(d; a, b). Gd × C2 C2, d x, ξb (ξd, (x, y)) d y). (6) ρ The type (d; a, b) is normalized if and only if gcd(d, a) = gcd(d, b) = 1. If this is not the case, one uses the isomorphism (assuming gcd(d, a, b) = 1) (cid:16) (cid:2)(x(d,b), y(d,a))(cid:3) (d,a)(d,b) ; d X(d; a, b) −→ X (cid:2)(x, y)(cid:3) (cid:55)→ (cid:17) a (d,a) , b (d,b) , to normalize it. We present different properties of some important sheaves associated to a V - surface (see [8, §4] and [15]). Proposition 2.2 ([8]). Let OX be the structure sheaf of a V -surface X then, • If P is not a singular point of X then OX,P is isomorphic to the ring of convergent power series C{x, y}. also known as the module of k-invariant germs in X(d; a, b). d x, ξb dy) = ξk d h(x, y)}, OX,P (k) := {h ∈ C{x, y} h(ξa d−1(cid:77) C{x, y} = OX,P (k) k=0 Remark 2.1. Note that (7) 4 J.I. COGOLLUDO, J. MART´IN, AND J. ORTIGAS • If P is a singular point of X then OX,P is isomorphic to the ring of Gd- invariant convergent power series C{x, y}Gd . If no ambiguity seems likely to arise we simply write OP for the corresponding local ring or just O in the case P = 0. Definition 2.3. Let Gd be an arbitrary finite cyclic group, a vector of weights (a, b) ∈ Z2 and the action given in (6). Associated with X(d; a, b) one has the following OX,P -module: Remark 2.2 ([8]). Let l, k ∈ Z. Using the notation above one clearly has the following properties: • OX,P (k) = OX,P (d + k), • OX,P (l) ⊗ OX,P (k) ⊂ OX,P (l + k). called orbisheaves. These modules produce the corresponding sheaves OX (k) on a V -surface X also One of the main examples of V -surfaces is the so-called weighted projective plane >0 be a weight vector, that is, a triple of pairwise coprime positive integers. There is a natural action of the multiplicative (e.g. [15]). Let w := (w0, w1, w2) ∈ Z3 group C∗ on C3 \ {0} given by The universal geometric quotient of it is called the weighted projective plane of type w. C∗ (x0, x1, x2) (cid:55)−→ (tw0x0, tw1x1, tw2 x2). C3\{0} under this action is denoted by P2 w and Let us recall the adapted concept of resolution in this category. Definition 2.4 ([19]). An embedded Q-resolution of a hypersurface (H, 0) ⊂ (M, 0) in an abelian quotient space is a proper analytic map π : X → (M, 0) such that: (1) X is a V -manifold with abelian quotient singularities, (2) π is an isomorphism over X \ π−1(Sing(H)), (3) π−1(H) is a Q-normal crossing hypersurface on X (see [25, Definition 1.16]). Embedded Q-resolutions are a natural generalization of the usual embedded resolutions, for which some invariants, such as δ can be effectively calculated ([14]). As a key tool to construct embedded Q-resolutions of abelian quotient surface singularities we will recall toric transformations or weighted blow-ups in this context (see [20] as a general reference), which can be interpreted as blow-ups of m-primary ideals. the weighted blow-up π : (cid:98)X → X at a point P ∈ X with respect to w = (p, q). Let X be an analytic surface with abelian quotient singularities. Let us define Since it will be used throughout the paper, we briefly describe the local equations of a weighted blow-up at a point P of type (d; a, b) (see [19, Chapter 1] for further details). scribed as usual by covering stance (cid:98)U1 is of type X (cid:92) X(d; a, b)w into two charts (cid:98)U1 ∪ (cid:98)U2, where for in- (cid:92) (cid:19) X(d; a, b)w → X(d; a, b) can be de- ; 1, −q + a(cid:48)pb , with a(cid:48)a = b(cid:48)b ≡ 1 mod (d) and e (cid:18) pd e The birational morphism π = π(d;a,b),w : e = gcd(d, pb − qa). The first chart is given by NUMERICAL ADJUNCTION FORMULAS... 5 (cid:18) pd e X (8) (cid:19) (cid:2)(xe, y)(cid:3) ; 1, −q + a(cid:48)pb e −→ (cid:98)U1, (cid:55)→ (cid:2)((xp, xqy), [1 : y]w)(cid:3) (d;a,b) The second one is given analogously. −1 (d;a,b),w(0) is identified with P1 w(d; a, b) := P1 The singular points are cyclic and correspond to the origins of the two charts. The exceptional divisor E = π w/Gd. 2.2. Log-resolution logarithmic forms. All the preliminaries about De Rham cohomology for projective varieties with quotient singularities can be found in [25, Chapter 1] and the ones about C∞ log complex of quasi projective algebraic varieties in [13, §1.3]. Here we focus on the non-normal crossing Q-divisor case in weighted Let D be a Q-divisor in a surface X with quotient singularities. The complement of D will be denoted by XD. Let us fix π : Y −→ X a Q-resolution of the singularities of D so that the reduced Q-divisor D = π∗(D)red is a union of smooth Q-divisors on Y with Q-normal crossings. projective planes. Using the results in [25] we can generalize Definition 2.7 in [13] for a non-normal crossing Q-divisor in X. Definition 2.5. The sheaf π∗ΩY (log(cid:104)D(cid:105)) is called the sheaf of log-resolution loga- rithmic forms on X with respect to D. Remark 2.3. Note that the space Y in the previous definition is not smooth and we use the standard definition of logarithmic sheaf for V -manifolds and V -normal crossing divisors D due to Steenbrink [25]. In the sequel, a log-resolution logarithmic form with respect to a Q-divisor D and a Q-resolution π will be referred to as simply a logarithmic form, if D and π are known and no ambiguity is likely to arise. Remark 2.4. Let h be an analytic germ on X(d; a, b) where the type is normalized. automatically defines a logarithmic form with poles along Notice that ω = h dx ∧ dy xy xy, whereas expressions of the form ω = h unless h is so that ω is invariant under Gd. dx ∧ dy x might not even define a 2-form Also note that π∗ΩY (log(cid:104)D(cid:105)) depends, in principle, on the given resolution π. The following results shows that this is not the case. Proposition 2.6. The sheaves π∗Ω• Y (log(cid:104)D(cid:105)) of logarithmic forms on X with respect to the Q-divisor D do not depend on the chosen Q-resolution. Proof. Let Y and Y (cid:48) be two Q-resolutions of (X, D). After resolving (Y, ¯D) and (Y (cid:48), ¯D(cid:48)) and applying the strong factorization theorem for smooth surfaces, there exists a smooth surface Y obtained as a finite number of blow-ups of both Y and Y (cid:48) which is a common resolution of (Y, ¯D) and (Y (cid:48), ¯D(cid:48)). Since (cid:48) • Y (cid:48)(log(cid:104)D(cid:48)(cid:105)) = ρ ∗Ω and Ω (see [25, p. 351]) where D, D(cid:48), and D are the corresponding total preimages and ρ, ρ(cid:48) are the corresponding resolutions. The result follows, since • Y (log(cid:104) D(cid:105)) = π due to the commutativity of the diagram πρ = π(cid:48)ρ(cid:48). Notation 2.7. In the future, we will refer to such sheaves as logarithmic sheaves on D and they will be denoted simply as Ω• • Y (log(cid:104)D(cid:105)) = π∗ρ∗Ω • Y (log(cid:104)D(cid:105)) = ρ∗Ω (cid:48) ∗Ω • Y (cid:48)(log(cid:104)D(cid:48)(cid:105)) • Y (log(cid:104) D(cid:105)) = π (cid:48) (cid:48) ∗ρ ∗Ω • Y (log(cid:104) D(cid:105)) • Y (log(cid:104) D(cid:105)) (cid:3) Ω π∗Ω X (LR(cid:104)D(cid:105)). (cid:80)a−1 k=1 (cid:19) 1 (cid:19) (cid:80)b−1 (cid:18) 1 k=1 a−1(cid:88) (cid:18) 1 k=1 6 J.I. COGOLLUDO, J. MART´IN, AND J. ORTIGAS 2.3. Dedekind Sums. Let a, b, c, t be positive integers with gcd(a, b) = gcd(a, c) = gcd(b, c) = 1. The aim of this part (see [6] for further details) is to give a way to compute the cardinal of the following two sets: ∆1 := {(x, y) ∈ Z2≥0 ax + by ≤ t}, ∆2 := {(x, y, z) ∈ Z3≥0 ax + by + cz = t}. Note that #∆1 cannot be computed by means of Pick's Theorem unless t is divisible by a and b. Denote by L∆i(t) the cardinal of ∆i. Let us consider the following notation. Notation 2.8. If we denote by ξa := e 2iπ a , consider p{a,b,c}(t) (9) with := poly{a,b,c}(t) + 1 a (1−ξkb a )ξkt a 1 a )(1−ξkc (cid:80)c−1 k=1 + 1 b 1 (1−ξka b )(1−ξkc b )ξkt b + 1 c (1−ξka c )(1−ξkb c )ξkt c , 1 poly{a,b,c}(t) := t2 2abc + t 2 + 1 ac + 1 bc ab + 3(ab + ac + bc) + a2 + b2 + c2 12abc . Remark 2.5. Notice that in particular, one has (10) p{a,b,1}(t) = poly{a,b,1}(t)+ 1 a with poly{a,b,1}(t) = t2 2ab + t 2 b−1(cid:88) k=1 + 1 b 1 (1 − ξka b )(1 − ξk b )ξkt b , (1 − ξkb a )(1 − ξk a )ξkt a + 1 a + 1 b ab + 3(ab + a + b) + a2 + b2 + 1 12ab . Theorem 2.9 ([6]). One has the following result, L∆1(t) = p{a,b,1}(t) and L∆2 (t) = p{a,b,c}(t). Now we are going to define the Dedekind sums giving some properties which will be particularly useful for future results. See [23] and [6] for a more detailed exposition. Definition 2.10 ([23]). Let a, b be integers, gcd(a, b) = 1, b ≥ 1. The Dedekind sum s(a, b) is defined as follows (11) where the symbol ((x)) denotes (cid:19)(cid:19)(cid:18)(cid:18) j (cid:19)(cid:19) , b s(a, b) := (cid:18)(cid:18) ja b−1(cid:88) (cid:26) x − [x] − 1 j=1 b 2 0 ((x)) = if x is not an integer, if x is an integer, with [x] the greatest integer not exceeding x. The following result, referred to as a Reciprocity Theorem (see [6, Corollary 8.5] or [23, Theorem 2.1] for further details) will be key in what follows. Theorem 2.11 (Reciprocity Theorem, [6, 23]). Let a and b be two coprime integers. Then s(a, b) + s(b, a) = − 1 4 + 1 + a2 + b2 . 12ab NUMERICAL ADJUNCTION FORMULAS... 7 Let us express the sum (11) in terms of ath-roots of the unity (see for instance [6, Example 8.1] or [23, Chapter 2, (18b)] for further details). Proposition 2.12 ([6, 23]). Let a, b be integers, gcd(a, b) = 1, b ≥ 1, denote by ξb a primitive bth-root of the unity. The Dedekind sum s(a, b) can be written as follows: b−1(cid:88) k=1 s(a, b) = b − 1 4b − 1 b 1 b )(1 − ξk (1 − ξka b ) . Let us exhibit some useful properties of the Dedekind sum s(a, b). Since ((−x)) = −((x)) it is clear that and also s(−a, b) = −s(a, b) If we define a(cid:48) by a(cid:48)a ≡ 1 mod b then (cid:48) s(a , b) = s(a, b). s(a,−b) = s(a, b). Proposition 2.13 ([6, 23]). Let a, b, c be integers with gcd(a, b) = gcd(a, c) = gcd(b, c) = 1. Define a(cid:48) by a(cid:48)a ≡ 1 mod b, b(cid:48) by b(cid:48)b ≡ 1 mod c and c(cid:48) by c(cid:48)c ≡ 1 mod a. Then (cid:48) (cid:48) , a) + s(ca (cid:48) , b) + s(ab s(bc , c) = − 1 4 + a2 + b2 + c2 12abc . Definition 2.14 ([6]). Let a1, . . . , am, n ∈ Z, b ∈ Z>0, then the Fourier-Dedekind sum is defined as follows: b−1(cid:88) sn(a1, . . . , am; b) := ξkn b )(1 − ξka2 Let us see some interesting properties of these sums. 1 b (1 − ξka1 b k=1 )··· (1 − ξkam b ) . b Remark 2.6 ([6]). Let a, b, c ∈ Z then (1) For all n ∈ Z, sn(a, b; 1) = 0. (2) For all n ∈ Z, sn(a, b; c) = sn(b, a; c). (3) One has s0(a, 1; b) = −s(a, b) + b−1 4b . (4) If we denote by a(cid:48) the inverse of a modulo c, then s0(a, b; c) = −s(a(cid:48)b, c) + c−1 4c . With this notation we can express (9) and (10) as follows (12) (13) p{a,b,1}(t) = poly{a,1,b}(t) + s−t(a, 1; b) + s−t(1, b; a). p{a,b,c}(t) = poly{a,b,c}(t) + s−t(a, b; c) + s−t(b, c; a) + s−t(a, c; b). As a consequence of Zagier reciprocity in dimension 3 (see [6, Theorem 8.4]) one has the following result. Corollary 2.15 (Rademacher's Reciprocity Law, [6]). Substituting t = 0 in the previous expression one gets 1 − poly{a,b,c}(0) = s0(a, b; c) + s0(c, b; a) + s0(a, c; b) = − 1 4 + a2 + b2 + c2 12abc . 8 J.I. COGOLLUDO, J. MART´IN, AND J. ORTIGAS 3. Local algebraic invariants on quotient singularities In this section we study two local invariants of a curve in a V -surface, the delta invariant δP (C) and the dimension κP (C) (see Definitions 3.2 and 3.5) and a local invariant of the surface, ∆P (k) (see Definition 3.8), as well as the relation among them (see Theorem 1.2), so as to understand the right-hand side of the formula in Theorem 1.1. In [14, 21] we started extending the concept of Milnor fiber and Milnor number of a curve singularity allowing the ambient space to be a quotient surface singularity. A generalization of the local δ-invariant is also defined and described in terms of a Q-resolution of the curve singularity. All these tools allow for an explicit description of the genus formula of a curve defined on a weighted projective plane in terms of its degree and the local type of its singularities. Definition 3.1 ([14]). Let C = {f = 0} ⊂ X(d; a, b) be a curve germ. The Milnor fiber f w t of (C, [0]) is defined as follows, The Milnor number µw of (C, P ) is defined as follows, f w t := {f = t}/Gd. (cid:80) µw := 1 − χorb(f w t ). We recall that χorb(O) := 1G ∆(−1)dim ∆G∆ for an orbifold O with a finite CW -complex structure given by the cells ∆ and the finite group G acting on it, where G∆ denotes the stabilizer of ∆. Note that alternative generalizations of Milnor numbers can be found, for in- stance, in [1, 9, 26, 24]. The one proposed here seems more natural for quotient singularities, but more importantly, it allows for the existence of an explicit formula relating Milnor number, δ-invariant, and genus of a curve on a singular surface. We define the local invariant δ for curve singularities on X(d; a, b). Definition 3.2 ([14]). Let C be a reduced curve germ at [0] ∈ X(d; a, b), then we define δ (or δ0(C)) as the number verifying χorb(f w t ) = rw − 2δ, where rw is the number of irreducible branches of C at [0]. Remark 3.1. Note that the δ-invariant of a reduced curve (i.e. a reduced Q-divisor) on a surface with quotient singularity is not an integer number in general, but rather a rational number (see [14, Example 4.6]). However, in the case when C is in fact Cartier, δ0(C) is an integer number that has an interpretation as the dimension of the quotient ¯R/R where R is the coordinate ring of C and ¯R its normalization (see [14, Theorem 4.14]). An alternative definition of δ0(C) on normal surfaces in terms of a resolution can be found in [7]. Also note that rw can also be seen as the number of irreducible k-invariant factors of the defining equation f . For instance, the germ defined by f = (x2 − y4) in C2 is not irreducible since (x2 − y4) = (x − y2)(x + y2). However, f = 0 also defines a set of zeroes in X(2; 1, 1), which is irreducible (and hence rw = 1), since (x − y2) and (x + y2) are not k-invariant for any k (recall Definition 2.3). A recurrent formula for δ based on a Q-resolution of the singularity is provided in Theorem 3.3. Assume (f, 0) ⊂ X(d; a, b) and consider a (p, q)-blow-up π at the origin. Denote by ν0(f ) the (p, q)-multiplicity of f at 0 and e := gcd(d, pb − qa). As an inter- e E, where π∗(C) is the total pretation of ν = ν0(f ), we recall that π∗(C) = C + ν transform of C, C is its strict transform and E is the exceptional divisor. NUMERICAL ADJUNCTION FORMULAS... 9 We will use the following notation: (14) δ0,π(f ) = ν0(f ) 2dpq (ν0(f ) − p − q + e) , Theorem 3.3 ([14]). Let (C, [0]) be a curve germ on an abelian quotient surface singularity. Then (cid:88) Q≺[0] (15) δ(C) = δQ,π(p,q) (f ) where Q runs over all the infinitely near points of a Q-resolution of (C, [0]) and π(p,q) is a (p, q)-blow-up of Q, the origin of X(d; a, b). 3.1. Logarithmic Modules. For a given k ≥ 0, one has the module OP (k) of k-invariant germs (see Defini- tion 2.3), OP (k) := {h ∈ C{x, y} h(ξa d x, ξb dy) = ξk d h(x, y)}. Let {f = 0} be a germ in X(d; a, b). Note that if f ∈ OP (k), then one has the following OP -module OP (k − a − b) = {h ∈ C{x, y} h dx ∧ dy f is Gd-invariant}. Definition 3.4. Let D = {f = 0} be a germ in P ∈ X(d; a, b), where f ∈ OP (k). D denote the submodule of OP (k − a − b) consisting of all h ∈ (1) Let MLR OP (k − a − b) such that the 2-form ω = h dx ∧ dy f ∈ Ω2 X (LR(cid:104)D(cid:105)) (recall Notation 2.7). (2) Let Mnul D denote the submodule of MLR D consisting of all h ∈ MLR D such that the 2-form ω = h admits a holomorphic extension outside the strict transform (cid:98)f . X (LR(cid:104)D(cid:105)) f ∈ Ω2 dx ∧ dy (3) Any OP -module M ⊆ MLR D will be called logarithmic module. Definition 3.5. Let D = {f = 0} be a germ in P ∈ X(d; a, b). Let us define the following dimension, κP (D) = κP (f ) := dimC OP (s) Mnul D , for s = deg f − a − b. Remark 3.2. From the discussion in §2.2 note that κP (f ) turns out to be a finite number independent on the chosen Q-resolution. Intuitively, the number κP (f ) provides the minimal number of conditions required for a generic germ h ∈ OP (s) so that h ∈ Mnul D . Remark 3.3. It is known (see [13, Chapter 2]) that if f is a holomorphic germ in (C2, 0), then κ0(f ) = δ0(f ). 10 J.I. COGOLLUDO, J. MART´IN, AND J. ORTIGAS 3.2. The δ invariant in the general case of local germs. Let us start with the following constructive result which allows one to see any after performing a certain weighted blow-up. singularity on the quotient X(d; a, b) as the strict transform of some {g = 0} ⊂ C2 Remark 3.4. The Weierstrass division theorem states that given f, g ∈ C{x, y} with f y-general of order k, there exist q ∈ C{x, y} and r ∈ C{x}[y] of degree in y less than or equal to k − 1, both uniquely determined by f and g, such that g = qf + r. The uniqueness and the linearity of the action ensure that the division can be performed equivariantly for the action of Gd on C{x, y} (see (7)), i.e. if f, g ∈ O(l), for zero sets in C{x, y}Gd . then so are q and r. In other words, the Weierstrass preparation theorem still holds Let {f = 0} ⊂ (X(d; a, b), 0) be a reduced analytic germ. Assume (d; a, b) is a normalized type. After a suitable change of coordinates of the form X(d; a, b) → X(d; a, b), [(x, y)] (cid:55)→ [(x + λyk, y)] where bk ≡ a mod d, one can assume x (cid:45) f . Moreover, by Remark 3.4, f can be written in the form (16) f (x, y) = yr + aijxiyj ∈ C{x}[y] ∩ O(k). (cid:88) i>0, j<r For technical reasons, in the following results the space X(d; a, b) will be considered to be of type X(p;−1, q). Note that this is always possible. Lemma 3.6. Let f ∈ O(k) define an analytic germ on X(p;−1, q), gcd(p, q) = 1, such that x (cid:45) f . Then there exist g ∈ C{x, y} with x (cid:45) g such that g(xp, xqy) = xqrf (x, y). Moreover, f is reduced (resp. irreducible) if and only if g is. Proof. By the discussion after Remark 3.4 one can assume f ∈ C{x}[y] as in (16). We have −i + qj ≡ qr ≡ k mod p for all i, j so p(i + q(r − j)) and i + q(r − j) > 0. Consider g(x, y) = yr + i+q(r−j) p aijx (cid:88) i>0, j<r (cid:88) yj ∈ C{x}[y], (cid:88) yr +  . aijxiyj g(xp, xqy) = xqryr + aijxi+qryj = xqr Note that the strict transform passes only through the origin of the first chart. (cid:3) i>0, j<r i>0, j<r mark 3.3. The following Proposition 3.7 will be useful to give a generalization of Re- the following combinatorial number which generalizes(cid:0)d (cid:1): Before we state the result we need some notation. Given r, p, q ∈ Z>0 we define Note that(cid:0)d Proposition 3.7. Let be p, q, a, r ∈ Z>0 with gcd(p, q) = 1 and r1 = r + pa. (cid:1) = δ(1,1) r(qr − p − q + 1) δ(p,q) r := (17) 2 d 2 . 2p . Consider the following cardinal, A(p,q) r := #{(i, j) ∈ Z2 pi + qj ≤ qr; i, j ≥ 1}. Then, 1) If r = pa, one has δ(p,q) r = A(p,q) r . NUMERICAL ADJUNCTION FORMULAS... 11 2) The following equalities hold: (18) (19) = δ(p,q) r1−r + aqr, = A(p,q) r1−r + aqr. r r1 − δ(p,q) δ(p,q) r1 − A(p,q) A(p,q) − δ(p,q) r 3) The difference A(p,q) r r only depends on r modulo p. Proof. 1) To prove this fact it is enough to apply Pick's Theorem (see for example [6, §2.6]) noticing that the number of points on the diagonal without counting the ones in the axes is a − 1. Finally one gets, A(p,q) pa = a(pqa − p − q + 1) 2 = δ(p,q) pa . 2) Proving equation (18) is a simple and direct computation. To prove equa- tion (19), let us describe A(p,q) r1 , A(p,q) r and A(p,q) r1−r: r1 = #{(i, j) ∈ Z2 pi + qj ≤ qr + pqa; i, j ≥ 1}. A(p,q) Figure 1. A(p,q) r = #{(i, j) ∈ Z2 pi + qj ≤ qr; i, j ≥ 1} = #{(i, j) ∈ Z2 pi + qj ≤ qr + apq − apq; i, j ≥ 1} = #{(i, j) ∈ Z2 p(i + aq) + qj ≤ qr1; i ≥ 1, j ≥ 1} = #{(i, j) ∈ Z2 pi + qj ≤ qr1; i ≥ aq + 1, j ≥ 1}. r1−r = #{(i, j) ∈ Z2 pi + qj ≤ qr1 − qr; i, j ≥ 1} A(p,q) = #{(i, j) ∈ Z2 pi + q(j + r) ≤ qr1; i, j ≥ 1} = #{(i, j) ∈ Z2 p + qj ≤ qr1; i ≥ 1, j ≥ r + 1}. Using the decomposition shown in Figure 1 the claim follows. 3) Subtracting equations (18) and (19) the result holds. (cid:3) pa+r01rqaA(p,q)r1−rA(p,q)r1A(p,q)raqr 12 J.I. COGOLLUDO, J. MART´IN, AND J. ORTIGAS Definition 3.8. Let k ≥ 0 and P ∈ X(p;−1, q) = X. The ∆P (k)-invariant of X is defined as follows where r = q−1k mod p. ∆P (k) := A(p,q) r − δ(p,q) r , As a result of Proposition 3.7 one has the following result. Theorem 3.9. Let f1, f2 ∈ O(k) be two germs at [0] ∈ X(d; a, b). Then, κ0(f1) − κ0(f2) = δ0(f1) − δ0(f2). Proof. By Remark 3.4 and the discussion after it, we can assume that (cid:88) f(cid:96)(x, y) = yr(cid:96) + aijxiyj ∈ C{x}[y]. i>0≤j<r(cid:96) in X(p;−1, q) (p = d, q ≡ −ba−1 mod d). Consider g1 ∈ C{x, y} the reduced germ obtained after applying Lemma 3.6 to f1. Denote by π(p,q) the blowing-up at the origin. Note that νp,q(g1) = qr1, and thus δ(p,q) r1 = δπ(p,q) (g1) (see (17) and (14)). Consider the form ω := φ dx∧dy , φ ∈ C{x, y} and let us calculate the local g1 equations for the pull-back of ω after blowing-up the origin on C2, (20) φ dx ∧ dy g1 π(p,q)←− xνφ+p+q−1−qr1h g1 and Mnul dx ∧ dy f1 . Using the definitions of Mnul φ ∈ Mnul g1 ⇔ h ∈ Mnul f1 f1 (see Definition 3.4) this implies that and νφ + p + q − 1 − qr1 ≥ 0. Therefore φ(x, y) (cid:55)→ φ(xp, xqy) induces an isomorphism Mnul g1 ∼= A(p,q) r1 ∩ Mnul f1 , where A(p,q) of h. Since dimC r1 C{x,y} A(p,q) r1 := {h ∈ C{x, y} ordh +p+q−1−qr1 ≥ 0} and ordh is the (p, q)-order = A(p,q) r1 , one obtains (21) κ0(g1) = A(p,q) r1 + κ0(f1) On the other hand (see Remark 3.3 and Theorem 3.3), κ0(g1) = δ0(g1) = δπ(p,q)(f ) + δ0(f1) = δ(p,q) (22) r1 + δ0(f1). Therefore, from (21) and (22), (23) κ0(f1) = δ0(f1) + δ(p,q) r1 − A(p,q) r1 . Following a similar procedure we get, κ0(f2) = δ0(f2) + δ(p,q) (24) Notice that k ≡ qr1 ≡ qr2 mod p, which implies r1 ≡ r2 mod p since p and q are coprime. Therefore by Proposition 3.7 r1 − δ(p,q) A(p,q) r2 − δ(p,q) and finally from (23) and (24) it can be concluded that r1 = A(p,q) r2 − A(p,q) r2 r2 . , Proof of Theorem 1.2. The result follows directly from equation (23). κ0(f1) − κ0(f2) = δ0(f1) − δ0(f2). (cid:3) (cid:3) Remark 3.5. If (f, [0]) is a function germ on X = X(d; a, b), from Proposition 3.7(1) and Theorem 1.2, one has In particular if P is a smooth point of X, this generalizes Remark 3.3. κP (f ) = δP (f ). NUMERICAL ADJUNCTION FORMULAS... 13 Figure 2. 4. An introductory example Let us start this section with one basic illustrative example. Let us compute the number of solutions (a, b, c) ∈ Z3≥0 of the equation aw0 + bw1 + cw2 = k ¯w with w0, w1, w2 ∈ Z>0 and k ∈ Z≥0 fixed, or equivalently, the number of monomials in OP2 w of quasi-homogeneous degree k ¯ω. This number will be denoted by Lw(k ¯w) (recall (3)). Notice that this is equivalent to computing the number of non-negative integer solutions (a, b, c) to aw0 + bw1 = (kw01 − c)w2 (with wij := wiwj), which can be achieved by considering the following sets: A :=(cid:8)(a, b) ∈ Z2 B :=(cid:8)(a, 0) ∈ Z2 (cid:8)(0, b) ∈ Z2 >0 aw0 + bw1 = αw2, α = 0, . . . , kw01 >0 aw0 = αw2, α = 0, . . . , kw01 >0 bw1 = αw2, α = 0, . . . , kw01 ∪ If we denote by A = # A and B = # B, one has Lw(k ¯w) = A + B + 1. To compute A take two integers n0, n1 such that n0w0 + n1w1 = 1 with n1 > 0 and n0 ≤ 0 (this can always be done since the weights are pairwise coprime). There exists a positive integer λ satisfying a = n0αw2 + λw1 and − n0αw2 . This justifies the following definition: (see Figure 2) < λ < n1αw2 w1 w0 (cid:26) (cid:9) , (cid:9) (cid:9) . (cid:27) . Aα := # λ ∈ Z>0 − n0αw2 w1 < λ < n1αw2 w0 Note that by virtue of Pick's theorem the area of the triangle is equal to the number of natural points in its interior I plus one half the number of points in the Aααkw2−1kw1−1kw0−1y=−n0w2w1xkw01PP=(kw01,−n0w2w0k)QQ=(kw01,n1w2w1k)y=n1w2w0xPm=(mw1,−n0w2m),m=0,...,kw0Qm=(mw0,n1w2m),m=0,...,kw1QmPm 14 J.I. COGOLLUDO, J. MART´IN, AND J. ORTIGAS boundary minus one. The area of the triangle equals k2 ¯ω 2 , thus which implies A = I + (kw2 − 1) = k2 ¯ω 2 = I + kw 2 − 1, (cid:19) (cid:18) k2 ¯w 2 − kw 2 + 1 + (kw2 − 1) = 1 2 (k2 ¯w − kw) + kw2. It is easy to check that B = kw0 + kw1, then we have It is known that the genus of a smooth curve on P2 axes is w of degree d transversal w.r.t. the Lw(k ¯w) = k (k ¯w + w) + 1. 1 2 gw,d = d(d − w) 2 ¯w + 1. We want to find d such that Lw(k ¯w) = gw,d. To do that it is enough to solve the equation One finally gets that 1 2 k (k ¯w + w) + 1 = d(d − w) 2 ¯w + 1. Lw(k ¯w) = gw,w+k ¯w. The rest of this paper deals with the extension of this example when d is not necessarily a multiple of ¯w. 5. Proof of the main results Proof of Theorem 1.1. We will prove the equivalent formula Lw(d − w) = gw,d − From the definitions note that ∆P (d). P∈Sing(P2 w) (cid:88) Lw(d − w) = p{w0,w1,w2}(d − w) =: p{w}(d − w). Fix a point P ∈ Sing(P2 X(wi; wi+1, wi+2) = X(wi;−1, qi), where qi := −w −1 (indices are considered modulo 3). Define ri := w i+2d mod wi, then w) and describe for simplicity the local singularity as −1 i+1wi+2 mod wi, for i = 0, 1, 2 On the one hand from (13) and a direct computation one obtains A(wi,qi) ri = p{wi,qi,1}(qiri − wi − qi). 2(cid:88) Lw(d − w) − gw,d = −1 + poly{w}(0) + i=0 By Definition 2.14 and Corollary 2.15 one obtains sw−d(wi, wi+1; wi+2). 2(cid:88) i=0 (cid:0)sw−d(wi+1, wi+2; wi) − s0(wi+1, wi+2; wi)(cid:1) . (25) Lw(d − w) − gw,d = On the other hand, from (12), and straightforward computations one obtains δ(wi,qi) ri − A(wi,qi) ri (26) with wi + qi 2wiqi − poly{wi,qi,1}(0) = − (swi+qi−qiri(wi, 1; qi) + swi+qi−qiri(qi, 1; wi)) , swi+qi−qiri(qi, 1; wi) = 1 wi 1 wi )(1 − ξk (1 − ξkqi wi)ξk(qiri−wi−qi) wi , wi−1(cid:88) k=1 NUMERICAL ADJUNCTION FORMULAS... 15 and swi+qi−qiri(wi, 1; qi) = 1 qi = − which implies using Proposition 2.12 qi−1(cid:88) qi−1(cid:88) (1 − ξkwi k=1 qi k=1 (1 − ξ 1 qi 1 )(1 − ξk 1 −kwi qi , )(1 − ξk qi) qi)ξk(qiri−wi−qi) qi (27) swi+qi−qiri (wi, 1; qi) = swi(wi, 1; qi) = s(−wi, qi) − qi − 1 4qi . Since by hypothesis qi = −w obtains −1 i+1wi+2 mod wi and ri = w −1 i+2d mod wi, one (cid:80)wi−1 (cid:96)=1 (1−ξ (cid:96)qi wi )(1−ξ(cid:96) wi 1 )ξ (cid:96)(qiri−wi−qi) wi wi+2) 1 )(1−ξ(cid:96) wi (cid:96)(−w−1 wi i+1 )ξ d+w−1 i+1 wi+2) (1−ξ ¯(cid:96)wi+2 wi )(1−ξ 1 −¯(cid:96)wi+1 wi ¯(cid:96)(d−wi+2) wi )ξ swi+qi−qiri(qi, 1; wi) = 1 wi (cid:80)wi−1 (cid:96)=1 = 1 wi (1−ξ (cid:96)(−w−1 wi i+1 (cid:80)wi−1 ¯(cid:96)=1 1 wi (cid:96)=−wi+1 ¯(cid:96) = (cid:80)wi−1 ¯(cid:96)=1 (1−ξ ¯(cid:96)wi+2 wi 1 )(1−ξ ¯(cid:96)wi+1 wi ¯(cid:96)(d−w) wi )ξ = −sw−d(wi+1, wi+2; wi). (28) = − 1 wi Thus (29) for i = 0, 1, 2. sw−d(wi+1, wi+2; wi) = −swi+qi−qiri(qi, 1; wi), Using (25), (26), and (29), it only remains to show (30) − s0(wi+1, wi+2; wi) = wi + qi 2wiqi − poly{wi,1,qi}(0) − swi+qi−qiri(wi, 1; qi). For the left-hand side we use Remark 2.6(4) and obtain s0(wi+1, wi+2; wi) = −s(−qi, wi) + wi − 1 4wi = s(qi, wi) + wi − 1 4wi . For the right-hand side, using Corollary 2.15 and (27) we have, (cid:18) poly{wi,1,qi}(0)+swi+qi−qiri(wi, 1; qi) = 1−s0(qi, 1; wi)−s0(wi, 1; qi)+swi (wi, 1; qi), which, by Remark 2.6(3) and (27) becomes 1 + s(qi, wi) − wi − 1 4wi + s(wi, qi) − qi − 1 4qi + s(−wi, qi) − qi − 1 4qi . Combining these equalities into (30) one obtains the result. (cid:3) (cid:19) δ(wi,qi) ri − A(wi,qi) ri δP (f ) + (cid:80) P ∈Sing(C) κP (f ) (cid:17) (cid:125) . (cid:3) (cid:16) 2(cid:88) (cid:123)(cid:122) i=0 (cid:88) 16 J.I. COGOLLUDO, J. MART´IN, AND J. ORTIGAS Proof of Theorem 1.3. It is enough to apply [14, Theorem 5.7], Theorem 1.1 and recall the characterization of κP (f ) in the proof of Theorem 3.9 (see (23)). (cid:88) (cid:123)(cid:122) g(C) = gw,d − δP (f ) P∈Sing(C) 2(cid:88) (cid:16) i=0 = gw,d + (cid:124) δ(wi,qi) ri − A(wi,qi) ri Lw(d−w)  (cid:88) (cid:124) P∈Sing(C) (cid:17) (cid:125) − Remark 5.1. The second equality in the previous identity always holds and therefore w a reduced curve of degree d, then (recall Theorem 1.2) if one considers C ⊂ P2 (cid:88) Lw(d − w) = gw,d − δP (f ) + κP (f ). P∈Sing(C) P∈Sing(C) Let us see an example of the previous result. Example 5.1. Consider the polygon Dw := {(x, y, z) ∈ R3 w0x+w1y +w2z = 1}, for w = (w0, w1, w2) = (2, 3, 7). As an example, we want to obtain the Ehrhart quasi-polynomial Lw(d) for Dw. Note that, according to Theorem 1.1 1 84 d2 + 1 7 Lw(d) = d + a0(d), (cid:17) i=0 ∆i(d + 12) (cid:16)(cid:80)2 where a0(d) is a rational periodic number of period ¯w = 42. Moreover, a0(d) = 1 − , where ∆i has period wi and depends only on the singular point Pi = {xj = xk = 0} ({i, j, k} = {0, 1, 2}) in the weighted projective plane P2 w. In order to describe ∆i(d) we will introduce some notation. Given a list of rational numbers q0, . . . , qr−1 we denote by [q0, . . . , qr−1] the periodic function f : Z → Q whose period is r and such that f (i) = [q0, . . . , qr−1]i = qi. Using this notation it is easy to check that (cid:20) (cid:21) ∆0(d) = 0, and ∆1(d) = 0, d 1 3 , 1 3 . d (cid:20) (cid:21) 1 4 Finally, in order to obtain ∆2(d), one needs to compute both δ and K-invariants for the singular point P2 ∈ X = X(7; 2, 3). The following table can be obtained directly: Table 1. Local invariants at X(7; 2, 3) d ∆P δP κP Branches Equation 1 2/7 9/7 1 2 0 0 0 0 0 1 x(x3 + y2) 2 3/7 3/7 0 1 x 3 3/7 3/7 0 1 y 4 2/7 9/7 1 2 5 0 1 1 2 6 4/7 4/7 0 1 x(x + y3) xy x3 + y2 The typical way to obtain the first row is by applying Theorem 1.3 to a generic germ fd in OX (d). This is how the second and third rows in the previous table were obtained. The last two rows indicate the local equations and the number of branches of such a generic germ fd ∈ OX (d). NUMERICAL ADJUNCTION FORMULAS... 17 Let us detail the computations for the third column in Table 1 (case d = 1). One can write the generic germ f1 in OX (1) as x(x3 + y2). On the one hand, a (2, 3)-blow-up serves as a Q-resolution of X, and thus, using Theorem 3.3, δP (f1) = 8(8 − 2 − 3 + 7) + 2 · 7 · 2 · 3 3 − 1 2 · 3 = 9 7 . For a computation of κP one needs to study the quotient OX (3)/Mnul f1 . Notice that in the present case OX = C{x, y}G7 = C{x7, y7, x2y} and OX (3) is the OX - module generated by y and x5. In order to study Mnul f1 , consider a generic form (ay + bx5) dx ∧ dy f1 ∈ Ω2 X (LR(cid:104)f1(cid:105)), where a, b ∈ OX and its pull-back by a resolution of the singularity X(7; 2, 3). One obtains the following: (ay + bx5) dx ∧ dy x(x3 − y2) (31) x=u1 ¯v2 1 y=¯v3 1 1 , v1=¯v7 ←− 3¯v3 1(a + bu5 1¯v7 1) ¯v4 1 du1 ∧ d¯v1 ¯v8 1u1(u3 1 − 1) = Therefore (ay + bx5) /∈ Mnul f1 Definition 3.5, 3 7 (a + bu5 1v1) du1 ∧ dv1 v1u1(u3 1 − 1) . iff the function a ∈ OX is a unit. Hence, by κP (f1) = dimC OX (3) Mnul f1 = dimC < y >C= 1. Finally, ∆P (1) = δP (f1) − κP (f1) = (cid:20) 2 7 . (cid:21) , The rest of values in Table 1 can be computed analogously. Hence one obtains: ∆2(d) := 0, 2 7 , 3 7 , 3 7 , 2 7 , 0, 4 7 d (cid:20) (cid:20) and thus Lw(d) = (cid:21) (cid:20) For instance, if one wants to obtain Lw(54), note that (cid:2)0, 1 d − 54 = (cid:2)0, 4 (cid:3) (cid:2)0, 1 (cid:19) (cid:21) (cid:18) d − (cid:3) 0 = 0, and (cid:2)0, 4 (cid:3) 54 = (cid:2)0, 1 7 , 0, 2 (cid:18) 1 − 1 84 d2 + Thus 3 , 1 7 , 3 7 , 3 7 , 2 3 , 1 d + 1 7 1 3 1 3 1 4 4 7 0, 0, 0, 3 3 7 4 , , 0, , (cid:21) (cid:3) 54 = (cid:2)0, 1 (cid:3) 3 7 3 7 2 7 2 7 7 , 3 7 , 3 7 , 2 4 7 , , (cid:3) 7 , 0, 2 (cid:19) . d 0 = 0, 5 = 3 7 . Lw(54) = 542 + 1 84 1 7 54 + 1 − 3 7 = 43. References [1] E. Artal Bartolo, J. Fern´andez de Bobadilla, I. Luengo, and A. Melle-Hern´andez. Milnor number of weighted-Le-Yomdin singularities. Int. Math. Res. Not. IMRN, (22):4301 -- 4318, 2010. [2] E. Artal Bartolo, J. Mart´ın-Morales, and J. Ortigas-Galindo, Cartier and Weil divisors on varieties with quotient singularities, Internat. J. Math. 25 (2014), no. 11, 1450100 (20 pages). , Intersection theory on abelian-quotient V -surfaces and Q-resolutions, J. Singul. 8 [3] (2014), 11 -- 30. [4] T. Ashikaga, Toric modifications of cyclic orbifolds and an extended Zaiger reciprocity for Dedekind sums, http://www.mathsoc.jp/section/algebra/algsymp past/algsymp12 files/ Ashikaga.pdf. [5] T. Ashikaga and M. Ishizaka, Another form of the reciprocity law of Dedekind sum, http://eprints3.math.sci.hokudai.ac.jp/1849/1/pre908.pdf 18 J.I. COGOLLUDO, J. MART´IN, AND J. ORTIGAS [6] M. Beck and S. Robins. Computing the continuous discretely. Undergraduate Texts in Math- ematics. Springer, New York, 2007. Integer-point enumeration in polyhedra. [7] R. Blache. Riemann-Roch theorem for normal surfaces and applications. Abh. Math. Sem. Univ. Hamburg, 65:307 -- 340, 1995. [8] C.P. Boyer and K. Galicki. Sasakian geometry. Oxford Mathematical Monographs. Oxford University Press, Oxford, 2008. [9] J.-P. Brasselet, D. Lehmann, J. Seade, and T. Suwa. Milnor classes of local complete inter- sections. Trans. Amer. Math. Soc., 354(4):1351 -- 1371 (electronic), 2002. [10] M. Brion. Points entiers dans les polytopes convexes. Ast´erisque, (227):Exp. No. 780, 4, 145 -- 169, 1995. S´eminaire Bourbaki, Vol. 1993/94. [11] A. Buckley, M. Reid, and S. Zhou, Ice cream and orbifold Riemann-Roch, Izv. Ross. Akad. Nauk Ser. Mat. 77 (2013), no. 3, 29 -- 54. [12] Sheng Chen, Nan Li, and Steven V. Sam. Generalized Ehrhart polynomials. Trans. Amer. Math. Soc., 364(1):551 -- 569, 2012. [13] J.I. Cogolludo-Agust´ın. Topological invariants of the complement to arrangements of rational plane curves. Mem. Amer. Math. Soc., 159(756):xiv+75, 2002. [14] J.I. Cogolludo-Agust´ın, J. Ortigas-Galindo, and J. Mart´ın-Morales, Local invariants on quo- tient singularities and a genus formula for weighted plane curves, Int. Math. Res. Not. (2014), no. 13, 3559 -- 3581. [15] I. Dolgachev. Weighted projective varieties. In Group actions and vector fields (Vancouver, B.C., 1981), volume 956 of Lecture Notes in Math., pages 34 -- 71. Springer, Berlin, 1982. [16] E. Ehrhart. Polynomes arithm´etiques et m´ethode des poly`edres en combinatoire. Birkhauser Verlag, Basel, 1977. International Series of Numerical Mathematics, Vol. 35. [17] F. Hirzebruch and D. Zagier, The Atiyah-Singer theorem and elementary number theory, Publish or Perish, Inc., Boston, Mass., 1974, Mathematics Lecture Series, No. 3. [18] R. Laterveer. Weighted complete intersections and lattice points. Math. Z., 218(2):213 -- 218, 1995. [19] J. Mart´ın-Morales. Embedded Q-Resolutions and Yomdin-Le Surface Singularities. PhD the- sis, 2011. http://zaguan.unizar.es/record/6870. [20] M. Oka, Non-degenerate complete intersection singularity, Actualit´es Math´ematiques. [Cur- rent Mathematical Topics], Hermann, Paris, 1997. [21] J. Ortigas-Galindo. Algebraic and Topological Invariants of Curves and Surfaces with Quo- tient Singularities. PhD thesis, 2013. https://zaguan.unizar.es/record/11738. [22] J.E. Pommersheim. Toric varieties, lattice points and Dedekind sums. Math. Ann., 295(1):1 -- 24, 1993. [23] H. Rademacher and E. Grosswald. Dedekind sums. The Mathematical Association of America, Washington, D.C., 1972. The Carus Mathematical Monographs, No. 16. [24] J. Seade, M. Tibar, and A. Verjovsky. Milnor numbers and Euler obstruction. Bull. Braz. Math. Soc. (N.S.), 36(2):275 -- 283, 2005. [25] J.H.M. Steenbrink. Mixed Hodge structure on the vanishing cohomology. In Real and complex singularities (Proc. Ninth Nordic Summer School/NAVF Sympos. Math., Oslo, 1976), pages 525 -- 563. Sijthoff and Noordhoff, Alphen aan den Rijn, 1977. [26] Le D. Tr´ang. Some remarks on relative monodromy. In Real and complex singularities (Proc. Ninth Nordic Summer School/NAVF Sympos. Math., Oslo, 1976), pages 397 -- 403. Sijthoff and Noordhoff, Alphen aan den Rijn, 1977. [27] G. Urz´ua, Arrangements of curves and algebraic surfaces, J. Algebraic Geom. 19 (2010), no. 2, 335 -- 365. Departamento de Matem´aticas, IUMA, Universidad de Zaragoza, C. Pedro Cerbuna 12, 50009 Zaragoza, Spain E-mail address: [email protected] Centro Universitario de la Defensa-IUMA, Academia General Militar, Ctra. de Huesca s/n., 50090, Zaragoza, Spain E-mail address: [email protected],[email protected]
1910.06208
1
1910
2019-10-14T15:23:39
Some combinatorial aspects of generalised Bott-Samelson varieties
[ "math.AG", "math.CO" ]
We obtain two combinatorial results: an equality of Weyl groups and an inequality of roots, in the setting of generalised Bott-Samelson resolutions of minuscule Schubert varieties. These results are used in the companion paper [BK19] to describe minimal rational curves on these resolutions, and their relation to lines on the Schubert varieties.
math.AG
math
SOME COMBINATORIAL ASPECTS OF GENERALISED BOTT-SAMELSON VARIETIES MICHEL BRION AND S. SENTHAMARAI KANNAN Abstract. We obtain two combinatorial results: an equality of Weyl groups and an in- equality of roots, in the setting of generalised Bott-Samelson resolutions of minuscule Schu- bert varieties. These results are used in the companion paper [BK19] to describe minimal rational curves on these resolutions, and their relation to lines on the Schubert varieties. 1. Introduction The generalised Bott-Samelson varieties of the title are certain towers of locally trivial fi- brations with fibers being Schubert varieties. Bott-Samelson varieties, for which all fibers are projective lines, yield widely used desingularisations of Schubert varieties. Their generalisa- tions were introduced by Sankaran and Vanchinathan (see [SV94, SV95]), to construct small resolutions of Schubert varieties in the symplectic and orthogonal Grassmannian. They were then systematically studied by Perrin (see [Pe07]); he associated a quiver to any minuscule Schubert variety X, and he constructed generalised Bott-Samelson desingularisations of X in terms of this quiver. In particular, Perrin's "construction 1" yields all small resolutions of X (see [loc. cit., §5.4, §7.5]). The motivation for the present paper comes from our investigation of lines in minuscule Schubert varieties, and minimal rational curves (an intrinsic version of lines) on their gen- eralised Bott-Samelson resolutions, in the companion paper [BK19]. The main theorem of the latter paper (Theorem 4.9) yields a description of the families of minimal rational curves on the resolutions obtained by construction 1. Its proof combines geometric arguments with two combinatorial results, which are proved in the present paper. We now outline these two results by using notation defined in Section 2. Consider a semi-simple, simply-laced algebraic group G, a minuscule parabolic subgroup P ⊂ G, and a Schubert variety X(w) ⊂ G/P . Letbπ : bX(bw) → X(w) be the generalised Bott-Samelson desingularisation associated with a generalised reduced decomposition bw = (w1, . . . , wm) obtained by construction 1; then bX(bw) is equipped with a base pointbx and a locally trivial fibration bf : bX(bw) → X(w1) with fiber bX(w2, . . . , wm). Moreover, X(w1) ≃ G1/P1 for a and G1 acts on bX(bw) compatibly with its action on X(w1). In loose terms, our first result asserts that bx is fixed by a Levi subgroup of P1. This translates into an equality of Weyl semi-simple, simply-laced subgroup G1 ⊂ G and a minuscule parabolic subgroup P1 ⊂ G1, groups, obtained in Theorems 4.1, 4.8 and 4.14. In the above situation, it is easy to show that there is a unique simple root α such that 1 (α) is negative; then w−1(α) is a negative root as well (see [BK19, Lem. 5.1]). the root w−1 1 Our second result asserts that w−1(α) ≤ w−1 X(w) is smooth. See Propositions 4.7, 4.13 and 4.15. 1 (α) with equality if and only if w = w1, that is, We obtain both results via a case-by-case analysis. Types An and Dn present somewhat different features (see Remark 4.9). The exceptional types E6, E7 are handled via a classifi- cation of the corresponding generalised Bott-Samelson varieties, which yields many examples of such varieties. We informed Nicolas Perrin of preliminary versions of our work, and he came up with shorter, uniform proofs of the equality of Weyl groups and the root inequality. In turn, we obtained a variant of Perrin's proof of the root inequality, which is also uniform and perhaps more self-contained, and also a short, uniform proof of the equality of Weyl groups; these are presented in the final section of [BK19]. We believe that our original, case-by-case approach is still of interest, as it yields additional information (for example, the Weyl groups under consideration are generated by simple reflections in type An, but not in type Dn) as well as many examples. This paper is organized as follows. In Section 2, we gather notation and recall some basic facts on flag varieties and their Schubert varieties. In Section 3, we first survey the construction of generalised Bott-Samelson varieties, following [Pe07] and [BK19, §4.2]; then we collect some auxiliary results. Section 4, containing the proofs of the main results, forms the bulk of the paper. 2. Preliminaries on Schubert varieties 2.1. Flag varieties. Let G be a simply-connected semi-simple algebraic group over an alge- braically closed field. Let P be a parabolic subgroup of G. Let X = G/P , and let x = P/P be the base point. Choose a Borel subgroup B ⊂ P and a maximal torus T ⊂ B. Denote by X ∗(T ) the character group of T , and by R ⊂ X ∗(T ) the root system of (G, T ); then the subset R of roots of (B, T ) is a set of positive roots of R. Also, denote by R− the corresponding set of negative roots, and by S = {α1, . . . , αn} ⊂ R+ the set of simple roots. Given λ, µ ∈ X ∗(T ), we say that λ ≥ µ if λ − µ is a nonnegative integral linear combination of simple roots. Further, λ > µ if in addition λ − µ 6= 0. We also have the coroot system R∨ with simple roots α∨ n; these form a basis of the cocharacter lattice X∗(T ). The dual basis of the character lattice X ∗(T ) consists of the fundamental weights 1, . . . , n. More intrinsically, for any simple root α, we will denote by α the fundamental weight with value 1 at α∨, and 0 at all other simple coroots. 1 , . . . , α∨ The Weyl group W = NG(T )/T is generated by the associated simple reflections s1, . . . , sn. We denote by ℓ the corresponding length function on W . A reduced decomposition of w ∈ W is a sequence of simple reflections such that w = si1si2 · · · sik and ℓ(w) = k. For any w ∈ W , we denote by w ∈ NG(T ) a representative. Also, for any β ∈ R, we denote by Uβ ⊂ G the corresponding root subgroup. ew = (si1, si2, . . . , sik) 2 Consider the Levi decomposition P = Ru(P )L, where L is a connected reductive subgroup of G containing T ; then BL := B ∩ L is a Borel subgroup of L. Denote by RL ⊂ R the root system of (L, T ), with subset of positive roots R+ L = RL ∩ R+ and subset of simple roots I := RL ∩ S. Then P is generated by B and the sα, where α ∈ I; we write P = PI and L = LI. We denote the parabolic subgroup PI also by P S\I. With this notation, the parabolic subgroup P is maximal if and only if P = P α for some simple root α. Note that P is uniquely determined by the dominant weight :=Pi∈I i. We say that P is minuscule, if so is ; that is, h, β∨i ≤ 1 for all β ∈ R+. The minuscule dominant weights are exactly the sums of minuscule fundamental weights α associated with irreducible factors of R; the corresponding simple roots α will be called minuscule as well. 2.2. Schubert varieties. We keep the notation and assumptions of the previous subsection. The Weyl group WL = NL(T )/T is generated by the simple reflections sα, where α ∈ I; we also denote this group by WI. Let W I denote the subset of W consisting of those w such that w(β) ∈ R+ for all β ∈ I; equivalently, R+ I ⊂ w−1(R+). Then W I is a set of representatives of the coset space W/WI, consisting of the elements of minimal length in their right coset. Note that w ∈ W I has length 1 if and only if w = sα for some α ∈ S \ I. On the other hand, the unique element of maximal length in W I is wI 0 = w0w0,I, where w0 (respectively, w0,I) denotes the longest element of W (respectively, WI). For any w ∈ W , the point wx ∈ G/P is independent of the choice of the representative w; we thus denote this point by wx. Recall that the wx, where w ∈ W I, are exactly the T -fixed points in G/P ; moreover, G/P is the disjoint union of the B-orbits Bwx. The stabilizer Bwx is generated by T and the root subgroups Uβ, where β ∈ R+ ∩ w(R+); in particular, Bwx is smooth and connected. The closure of Bwx in G/P is the Schubert variety X(w); we have dim(X(w)) = dim(Bwx) = ℓ(w). Note that X(wI 0) = X. We say that the homogeneous space G/P is minuscule if so is P . Then the Schubert varieties X(w) ⊂ G/P and the Weyl group elements w ∈ W I are called minuscule as well. Note that every minuscule homogeneous space under G decomposes into a product of minuscule homogeneous spaces under simple factors of G, and every Schubert variety decomposes accordingly. Therefore, to study minuscule Schubert varieties, we may assume that G is simple; then P = P α for some minuscule simple root α. We may further reduce to the case where G is simply-laced (see [Pe07, Rem. 3.4]). From now on, we assume that G is simple and simply-laced. 3. Generalised Bott-Samelson varieties 3.1. Generalisations of Bott-Samelson varieties due to Perrin. Recall that the set of simple roots α such that sα occurs in a reduced decomposition of w is independent of the choice of a reduced decomposition, and called the support of w. We denote this set by Supp(w). The subgroup of G generated by the U±α, where α ∈ Supp(w), will be denoted by Gw; this is the derived subgroup of the Levi subgroup LSupp(w), and hence is a semi- simple subgroup of G, normalized by T and containing a representative of w. Since G is simply-laced, every simple factor of Gw is simply-laced as well. 3 Let I w := {α ∈ S w(α) ∈ R+}. Then I w is the largest subset of S such that w ∈ W I w. We denote PI w by P w. Consider the associated Schubert variety X(w) ⊂ G/P w, and denote by Pw the closed reduced subgroup of G consisting of those g such that gX(w) = X(w). Then Pw is a parabolic subgroup of G containing B, and hence we have Pw = PIw, where Iw := {α ∈ S sαw ≤ w}; here ≤ denotes the Bruhat order on W I w = W/WI w. Note that Pw ∩ Gw is a parabolic subgroup of Gw, and we have (3.1) X(w) = PwwP w/P w ≃ (Pw ∩ Gw)w(P w ∩ Gw)/(P w ∩ Gw) ⊂ Gw/(P w ∩ Gw). We say that w ∈ W is minuscule if so is G/P w; then one may readily check that Gw/(P w∩Gw) is a product of minuscule homogeneous varieties associated to simple factors of Gw. For any minuscule w ∈ W , we have X(w)sm = Pwwx by the main result of [BP99], where X(w)sm denotes the set of smooth points of X(w). In particular, X(w) is smooth if and only if it is homogeneous under Pw. Then Gw ⊂ Pw in view of [BK19, Lem. 4.8]. Next, let w = w1w′, where w1, w′ ∈ W satisfy P w1 ∩ Gw1 ⊂ Pw′; equivalently, we have I w1 ∩ Supp(w1) ⊂ Iw′. We may then define This is a projective variety equipped with an action of Pw1 ∩Gw1 and an equivariant morphism eX(w1, w′) := (Pw1 ∩ Gw1)w1(P w1 ∩ Gw1) ×P w1 ∩Gw1 X(w′). fw1,w′ : eX(w1, w′) −→ X(w1), a Zariski locally trivial fibration with fiber X(w′). If in addition ℓ(w) = ℓ(w1) + ℓ(w′), then w′ ∈ W P w and hence P w′ ⊃ P w. Thus, if P w is maximal and w′ 6= 1, then P w′ = P w. As a consequence, we obtain another equivariant morphism One may check that πw1,w′ is birational to its image X(w); it restricts to an isomorphism above the open orbit Bwx. Also, note that πw1,w′ : eX(w1, w′) −→ G/P w. Pw1 ∩ Gw1 ⊂ Pw ∩ Gw. This construction can be iterated, under certain additional assumptions that are discussed in detail in [Pe07, §5.2]. We now present some notions and results from [loc. cit.]: a finite sequence bw = (w1, . . . , wm) of elements of W is called a generalised reduced decomposition of w, if we have w = w1 · · · wm and ℓ(w) = ℓ(w1) + · · · + ℓ(wm). Such a decomposition is called good if in addition w is minuscule and we have I wi ∩ Supp(wi) ⊂ Iwi+1···wm ⊂ w⊥ (1 ≤ i ≤ m − 1), i ∪ Supp(wi) where w⊥ i is the set of simple roots α such that sα commutes with wi. Under these assumptions, (wi+1, . . . , wm) is a good generalised reduced decomposition of wi+1 · · · wm for i = 1, . . . , m − 1. Moreover, P wi ∩ Gwi ⊂ Pwi+1···wm for all such i. Given a good generalised reduced decomposition bw of w, we obtain a projective variety bX(bw) equipped with an action of Pw, a locally trivial fibration bf : bX(bw) −→ X(w1) 4 the equality of stabilizers Pw,bx = Pw,wx. Clearly, Pw,wx contains the maximal torus T . with fiber bX(w2, . . . , wm), and a birational morphism bπ : bX(bw) −→ X(w). Also, bX(bw) has a base point bx such that bπ(bx) = wx and bf (bx) = w1x1 with an obvious notation. Moreover,bπ is Pw-equivariant in view of [Pe07, §5.1]. As a consequence, we have Note that bX(bw) is smooth if and only if X(w1), . . . , X(wm) are smooth. Then we have Lemma 3.1. Let bw = (w1, . . . , wm) be a good generalised decomposition of w such that bX(bw) Gw1 ⊂ Pw1, and hence Gw1 ⊂ Pw ∩ Gw. Let Tw1 be the neutral component of T ∩ Gw1; then Tw1 is a maximal torus of Gw1. Further, we have an isomorphism of Weyl groups W (Gw1, Tw1) ≃ W (LSupp(w1), T ). This identifies W (Gw1, Tw1) with a subgroup of W . is smooth. (1) For any 1 ≤ i ≤ n such that si ∈ W (Gw1, Tw1), we have siX(w) = X(w). (2) We have the inclusion of stabilisers Gw1,bx ⊂ Gw1,w1x1. (3) Tw1 is a common maximal torus of both Gw1,bx and Gw1,w1x1. (4) We have W (Gw1,bx, Tw1) ⊂ W (Gw1,w1x1, Tw1) ⊂ W . (5) We have W (Gw1,bx, Tw1) = W (Gw1,wx, Tw1) ⊂ W . Proof. Proof of (1) follows from the fact that Gw1 ⊂ Pw ∩ Gw. Proof of (3) is easy. Proof of (2) follows from the fact that the morphism bf : bX(bw) −→ X(w1) is Gw1- equivariant and sendsbx to w1x1. Proof of (4): By (2) and (3), we have NGw1 ,bx(Tw1) ⊂ NGw1 ,w1 x1 (Tw1). Hence, we obtain W (Gw1,bx, Tw1) ⊂ W (Gw1,w1x1, Tw1). Clearly, W (Gw1,w1x1, Tw1) is isomorphic to a subgroup of W (Gw1, Tw1). Therefore, W (Gw1,w1x1, Tw1) is identified with a subgroup of W . Proof of (5): Since Gw1 ⊂ Pw ∩Gw and (Pw ∩Gw)bx = (Pw ∩Gw)wx, we have Gw1,bx = Gw1,wx. (cid:3) Thus, the Weyl groups are equal. 3.2. The quiver associated to a minuscule element w. We recall the following defini- tions and construction 1 of Perrin from [Pe07]. Let P = PI be a minuscule parabolic subgroup of G, and w ∈ W I. Choose a reduced decomposition ew = (si1, si2, . . . , sik) of w, and let βj = αij (1 ≤ j ≤ k). Definition 3.2. We define the successor s(i) (respectively, the predecessor p(i)) of an element i ∈ [1, k] by s(i) = min{j ∈ [1, k] : j > i and βj = βi} (respectively, by p(i) = max{j ∈ [1, k] : j < i and βj = βi}). Definition 3.3. We denote by Q ew the quiver whose set of vertices is the set [1, k] and whose arrows are given in the following way: there is an arrow from i to j if hβ∨ j , βii 6= 0 and i < j < s(i) (or only i < j if s(i) does not exist). 5 By [St97, Prop. 2.1], any two reduced decompositions of w differ only by commuting w. Therefore, we denote this quiver by Qw. relations. So, the quiver Q ew does not depend on the choice of a reduced decomposition ew of The quiver comes with a coloration of its vertices by simple roots via the map β : [1, k] −→ S such that β(j) = βj for all 1 ≤ j ≤ k. See [Pe07, §2.1] for more details. Definition 3.4. We denote by R ⊂ [1, k] × [1, k] the partial order on the vertices of the quiver Qw generated by the relations (i, j) ∈ R if there exists an arrow from i to j. Definition 3.5. We call peak any vertex of Qw minimal for the partial order R. The set of peaks is denoted by Peaks(Qw). We now explain a way of constructing good generalised reduced decompositions of w into a product of minuscule elements (wi)1≤i≤m (see [Pe07, §5.4]). Definition 3.6. Let A ⊂ Peaks(Qw). We denote by Qw(A) the full subquiver of Qw with vertices being those i of Qw such that (j, i) /∈ R for all j ∈ Peaks(Qw) \ A. Qw(A) is also the quiver of a minuscule Schubert variety. By [Pe07, Prop. 5.13], the quiver cQw(A) obtained from Qw by removing the vertices of To construct a partition of the quiver Qw of a minuscule element w into quivers (Qwi)1≤i≤m with each wi is a minuscule element, it suffices to give a partition of Peaks(Qw). Indeed, given such partition, (Ai)1≤i≤m, we define by induction a sequence (Qi)0≤i≤m with Q0 = Qw, (Qwi)1≤i≤m form a partition of Qw. Each quiver Qwi is associated to a minuscule element wi. Qi+1 =cQi(Ai+1), Qw1 = Qw(A1). We then denote by Qwi the quiver Qi−1(Ai). The quivers The generalised reduced decomposition bw = (w1, w2, . . . , wm) is good (see [Pe07, Prop. 5.15]). Construction 1 of Perrin. Choose any order {i1, i2, . . . , im} on Peaks(Qw) and set Aj = {ij} (1 ≤ j ≤ m). Throughout this paper, we only consider generalised reduced decompositions obtained by construction 1. Also, we use repeatedly Perrin's smoothness criterion: with the above in RSupp(wj) (see [Pe07, Thm. 7.11]). notation, bX(bw) is smooth if and only if, for 1 ≤ j ≤ m, the simple root β(ij) is minuscule Lemma 3.7. Let bw = (w1, w2, . . . , wm) be a generalised reduced decomposition of w obtained by construction 1. Then any simple reflection si ∈ W (Gw1,w1x1, Tw1) lies in W (Gw1,wx, Tw1). Proof. Since si ≤ w1 and siw1x1 = w1x1, we have αi 6= β(p) where Peaks(Qw1) = {p}. On the other hand, by construction 1, we have Peaks(Qw) ∩ Qw1 = {p}. Therefore, we have ℓ(siw) = ℓ(w) + 1. Also, by Lemma 3.1(1), we have siX(w) = X(w). Therefore, combining these two together, we obtain siwx = wx. (cid:3) Let i1 < i2 < . . . < im be the ordering of Peaks(Qw) induced by the standard increasing w obtained by construction 1 corresponding to this ordering of Peaks(Qw). ordering of integers. Let bw = (w1, w2, . . . , wm) be the generalised reduced decomposition of 6 Let (cid:22) be another ordering of Peaks(Qw). Let bw′ = (w′ m) be the generalised reduced decomposition of w obtained by construction 1 corresponding to this ordering of Peaks(Qw). Let 1 ≤ q ≤ m be the integer such that iq is the first peak in this ordering. Further, assume that q 6= 1. 2, . . . , w′ 1, w′ Then, we have Lemma 3.8. Every simple reflection se ≤ w′ for all 1 ≤ r ≤ q − 1. 1 commutes with every simple reflection sf ≤ wr Proof. By construction 1, for any vertex l of Qw(Peaks(Qw) \ {iq}) and for any vertex of i of Qw({iq}), we have (l, i) /∈ R. In particular, if l < i such that β(l) 6= β(i), then we have hβ(l)∨, β(i)i = 0. This implies the assertion. (cid:3) 4. Equality of Weyl groups and a root inequality Let G be as in Section 2. Throughout this section, we consider a minuscule parabolic subgroup P = PI, a Weyl group element w ∈ W I, and a generalised reduced decomposition bw = (w1, w2, . . . , wm) of w, obtained by construction 1 of Perrin. Gw. Thus, we may assume that Gw = G; then Supp(w) = S. In view of (3.1), we see that X(w) is a Schubert subvariety of a minuscule flag variety for 4.1. Equality of Weyl groups in type An. Let G be of type An, that is, G = SLn+1; then every fundamental weight is minuscule and hence the minuscule parabolic subgroups are exactly the maximal ones. The aim of this subsection is to prove the following: Theorem 4.1. For any w and bw as above, we have W (Gw1,bx, Tw1) = W (Gw1,w1x1, Tw1) and this group (viewed as a subgroup of W ) is generated by simple reflections. To prove this result, we first set up notation. We order the simple roots as in [Hu72, p.58]. Let P := P w; then P = P αr for a unique integer 1 ≤ r ≤ n. Further, w(i) < w(i + 1) for every 1 ≤ i ≤ r − 1. Let ai = w(i) − 1. Then i ≤ ai < ai+1 ≤ n for every 1 ≤ i ≤ r − 1. Let v = (sa1sa1−1 · · · s1)(sa2sa2−1 · · · s2) · · · (sar sar−1 · · · sr). Note that v ∈ W S\{αr}. Since w, v ∈ W S\{αr} and w(i) = v(i) for all 1 ≤ i ≤ r, we have w = (sa1sa1−1 · · · s1)(sa2sa2−1 · · · s2) · · · (sar sar −1 · · · sr). For integers bi ≤ ai, we let wbi,ai = saisai−1 · · · sbi. Since is a reduced decomposition of w, we see that (w1,a1, w2,a2, . . . , wr,ar) is a generalised reduced decomposition of w. ew = (sa1, sa1−1, . . . , s1, sa2, sa2−1, . . . , s2, . . . , sar , sar−1, . . . , sr) Next, we prove a succession of preliminary results: 7 Lemma 4.2. Let 1 ≤ i < k ≤ r. Let bs (i ≤ s ≤ k) be a sequence of integers such that s ≤ bs ≤ as and bs+1 = 1 + bs for all i ≤ s ≤ k − 1. Let w′ = wbi,aiwbi+1,ai+1 · · · wbk,ak. Then for any integer bi ≤ l ≤ ai − 1, we have slw′ = w′sj for some integer j 6= bk such that sj ≤ w′. Proof. Since bs+1 = 1 + bs for all i ≤ s ≤ k − 1 the pattern of wbi,aiwbi+1,ai+1 · · · wbk,ak is similar to that of wi,aiwi+1,ai+1 · · · wk,ak. Thus, we may assume that bs = s for all i ≤ s ≤ k. Since slsm = smsl for all m ≥ l + 2, slsl+1sl = sl+1slsl+1 and sl+1st = stsl+1 for all i ≤ t ≤ l − 1, we have slwi,ai = slsaisai−1 · · · si = sai · · · sisl+1. Therefore, we have slw′ = wi,aisl+1wi+1,ai+1 · · · wk,ak. Further, since i+1 ≤ l +1 ≤ ai+1 −1, by induction on k − i, we have sl+1wi+1,ai+1 · · · wk,ak = wi+1,ai+1 · · · wk,aksj for some integer j 6= k such that sj ≤ w′. Hence, we have slw′ = wi,aiwi+1,ai+1 · · · wk,aksj = w′sj. (cid:3) Lemma 4.3. Let i, k, as, bs (i ≤ s ≤ k) and w′ be as in Lemma 4.2. Further, assume that as+1 = 1 + as for all i ≤ s ≤ k − 1. Then for any integer bi ≤ l ≤ ak different from ai, we have slw′ = w′sj for some integer j 6= bk such that sj ≤ w′. Proof. As in the proof of Lemma 4.2, we may assume that bs = s for all i ≤ s ≤ k. If l ≤ ai − 1, then, we are done by Lemma 4.2. If l = ai+1, we first show that slwi,aiwi+1,ai+1 = wi,aiwi+1,ai+1si. For the base case ai = i and ai+1 = i + 1, we have slwi,aiwi+1,ai+1 = si+1sisi+1 = sisi+1si = wi,aiwi+1,ai+1si. If ai ≥ i + 1, then we obtain slwi,aiwi+1,ai+1 = sai+1saisai+1(sai−1 · · · si)(sai · · · si+1) = saisai+1(sai((sai−1 · · · si)(sai · · · si+1))) = saisai+1(wi,ai−1wi+1,aisi) = wi,aiwi+1,ai+1si by induction on length. Since siws,as = ws,assi for all s ≥ i + 2, this yields slw′ = wi,aiwi+1,ai+1siwi+2,ai+2 · · · wk,ak = w′si. The proofs for the cases l = as, i + 2 ≤ s ≤ k are similar. (cid:3) Lemma 4.4. Let i, k, as, bs (i ≤ s ≤ k) and w′ be as in Lemma 4.3. Then for any integer bk + 1 ≤ t ≤ ak, we have w′st = slw′ for some integer l 6= ai such that sl ≤ w′. Proof. Again, we may assume that bs = s for all i ≤ s ≤ k. Since, stsm = smst for all m ≤ t − 2, stst−1st = st−1stst−1 and sjst−1 = st−1sj for all j ≥ t + 1, we have wk,akst = saksak−1 · · · skst = st−1sak · · · sk. Therefore, we have w′st = wi,aiwi+1,ai+1 · · · wk−1,ak−1st−1wk,ak. Further, since k ≤ t − 1 ≤ ak − 1 = ak−1, by induction on k − i, we have wi,ai · · · wk−1,ak−1st−1 = slwi,ai · · · wk−1,ak−1 for some integer l 6= ai such that sl ≤ w′. 8 Hence, we obtain w′st = slwi,aiwi+1,ai+1 · · · wk,ak = slw′. (cid:3) Lemma 4.5. Let i, k, as, bs (i ≤ s ≤ k) and w′ be as in Lemma 4.3. Then for any integer bi ≤ t ≤ ak different from bk, we have w′st = slw′ for some integer l 6= ai such that sl ≤ w′. Proof. As in the proof of Lemma 4.2, we may assume that bs = s for all i ≤ s ≤ k. If t ≥ k + 1, then we are done by Lemma 4.4. If t = k − 1, we first show that wk−1,ak−1wk,akst = sak wk−1,ak−1wk,ak. For the base case k = ak and k − 1 = ak−1, we have wk−1,ak−1wk,akst = sk−1sksk−1 = sksk−1sk = sak wk−1,ak−1wk,ak. If ak−1 ≥ k, then we obtain wk−1,ak−1wk,akst = wk,ak−1wk+1,ak(sk−1sksk−1) = wk,ak−1wk+1,ak(sksk−1sk) = (wk,ak−1wk+1,aksk)(sk−1sk) = (sakwk,ak−1wk+1,ak)(sk−1sk) = sak wk−1,ak−1wk,ak by induction on length. Since wj,aj sak = sak wj,aj for all i ≤ j ≤ k − 2, we have w′st = sak w′. The proofs for the cases i ≤ t ≤ k − 2 are similar. (cid:3) of w = w1,a1w2,a2 · · · wr,ar . Let β1 = αa1, β2 = αa1−1, etc. Therefore, the peaks of the quiver Qw are the indices i such that ℓ(sβiw) = ℓ(w) − 1. Recall that ew = (sa1, sa1−1, . . . , s1, sa2, . . . , s2, . . . , sar , . . . , sr) is a reduced decomposition Let J(w) := {2 ≤ j ≤ r : aj − aj−1 ≥ 2}S{1} and let J(w) = m. By the above paragraph, peaks of the quiver Qw are indexed by the elements of J(w). So, Peaks(Qw) is identified with {ij ∈ [1, ℓ(w)] : j ∈ J(w)}. Then we have β(ij) = αaj for all j ∈ J(w). Let 1 = j1 < j2 < · · · < jm ≤ r be the standard increasing ordering of elements of J(w), that is; the ordering induced by the usual ordering of positive integers. Let bw = (w1, w2, . . . , wm) be the generalised reduced decomposition of w obtained by construction 1 corresponding to this ordering. By this construction , we have w1 = w1,a1w2,a2 · · · wj2−1,aj2−1. Note that ℓ(siw1) = ℓ(w1)−1 if and only if i = a1. Therefore, again by construction 1, we have Peaks(Qw1) = {p}, where β(p) = αa1. Thus, we have hα∨ at+1, αati 6= 0 for all 1 ≤ t ≤ j2 −2. Hence, we have at+1 = 1+at for all 1 ≤ t ≤ j2 − 2. We may now prove Theorem 4.1 in the case where the generalised reduced decomposition is associated to the above standard ordering of peaks. For the reader's convenience, we recall its statement: Lemma 4.6. With the above assumptions, we have W (Gw1,bx, Tw1) = W (Gw1,w1x1, Tw1) and this group is generated by simple reflections. 9 Proof. By Lemma 3.1(5), we have W (Gw1,bx, Tw1) = W (Gw1,wx, Tw1). Further, by Lemma 3.1(4), we have W (Gw1,bx, Tw1) ⊂ W (Gw1,w1x1, Tw1). Thus, it suffices to prove that W (Gw1,w1x1, Tw1) ⊂ W (Gw1,wx, Tw1) and the latter is generated by reflections. Recall that w1 = w1,a1w2,a2 · · · wj2−1,aj2 −1. We first prove the following Claim. W (Gw1,w1x1, Tw1) is generated by simple reflections. Let v ∈ W (Gw1, Tw1) be such that vw1 = w1u for some u ∈ W (Gw1, Tw1). We prove by induction on ℓ(u) that v is a product of simple reflections sj ≤ w1 such that sjw1 = w1st for some integer t 6= j2 − 1 with st ≤ w1. If ℓ(u) = 1, then we have u = st for some t 6= j2 − 1 such that st ≤ w1. Hence by Lemma 4.5, we have v = sl for some integer l 6= a1 such that sl ≤ w1. So, assume that ℓ(u) ≥ 2. Choose an integer t 6= j2 − 1 such that st ≤ w1 and ℓ(ust) = ℓ(u) − 1. We have vsw1(αt)w1 = vw1st = w1ust. Note that w1(αt) ∈ R+ \ v−1(R+). Therefore we have sw1(αt) ∈ W (Gw1, Tw1). Since ℓ(ust) = ℓ(u) − 1, by induction vsw1(αt) is a product of simple reflections sj ≤ w1 such that sjw1 = w1sp for some integer p 6= j2 − 1 such that sp ≤ w1. On the other hand, by Lemma 4.5, we have w1st = slw1 for some integer l 6= a1 such that sl ≤ w1. Therefore, we have w1(αt) = αl and so v is a product of simple reflections sj ≤ w1 such that sjw1 = w1sp for some integer p 6= j2 − 1 such that sp ≤ w1. This proves the claim. Now, let v ∈ W (Gw1,w1x1, Tw1). Then by the claim, v is a product of simple reflections sj ≤ w1 such that sjw1 = w1sp for some integer p 6= j2 − 1 such that sp ≤ w1. On the other hand, by Lemma 3.7, for any such sj, we have sjwx = wx and hence we have vwx = wx. This proves the desired inclusion. (cid:3) 1, w′ 2, . . . , w′ the elements of J(w) and let j ∈ J(w) be the first element in this ordering. To complete the proof of Theorem 4.1, we now consider an arbitrary ordering (cid:22) of Let m) be the generalised reduced decomposition obtained by construction 1 corresponding to this ordering of Peaks(Qw). Let 1 ≤ q ≤ m be the integer such that j = jq. Again by construction 1, we have wq = w′ 1) + ℓ(v). bw′ = (w′ Let bX(bw′) be the variety corresponding to bw′, with base pointbx′. 1v for some v ∈ W such that ℓ(wq) = ℓ(w′ Recall from Lemma 3.1 that W (Gw′ By arguing as in the beginning of the proof of Lemma 4.6, it suffices to prove the inclusion 1,bx′, Tw′ 1) ⊂ W (Gw′ 1, Tw′ 1) ⊂ W . 1,w′ 1x′ W (Gw′ 1,w′ 1x′ 1, Tw′ 1) ⊂ W (Gw′ 1,wx, Tw′ 1). The rest of the argument is also similar to that of Lemma 4.6. For completeness of proof, we give the details. If the ordering of the peaks of Qw is such that w′ 1 = w1, we are done. Otherwise, let k = jq+1 − 1. Then by construction 1, there exists a sequence l ≤ bl ≤ al (j ≤ l ≤ k) of 10 positive integers such that bl+1 = bl+1 for all j ≤ l ≤ k−1, and w′ Further, we have al+1 = al + 1 for all j ≤ l ≤ k − 1. 1 = wbj ,aj wbj+1,aj+1 · · · wbk,ak. So, w′ 1 satisfies the hypothesis of Lemma 4.5. Therefore, we can imitate the proof of Lemma 4.6. We first prove the following Claim : W (Gw′ Let v ∈ W (Gw′ 1,w′ 1x′ 1, Tw′ 1, Tw′ 1) be such that vw′ 1) is generated by simple reflections. induction on ℓ(u) that v is a product of simple reflections si ≤ w′ some integer t 6= bk with st ≤ w1. 1 = w′ 1u for some u ∈ W (Gw′ 1, Tw′ 1 such that siw′ 1). We prove by 1st for 1 = w′ If ℓ(u) = 1, then we have u = st for some t 6= bk such that st ≤ w′ 1. Hence by Lemma 4.5, we have v = sl for some integer l 6= aj such that sl ≤ w′ 1. So, assume that ℓ(u) ≥ 2. Choose an integer t 6= bk such that st ≤ w′ We have vsw′ 1 = vw′ have sw′ 1, Tw′ simple reflections si ≤ w′ 1(αt)w′ 1(αt) ∈ W (Gw′ 1st = w′ 1ust. Note that w′ 1). Since ℓ(ust) = ℓ(u) − 1, by induction vsw′ 1 such that siw′ 1 and ℓ(ust) = ℓ(u)−1. 1(αt) ∈ R+ \ v−1(R+). Therefore we 1(αt) is a product of 1sp for some integer p 6= bk such that sp ≤ w′ 1. 1 for some integer l 6= aj such that 1(αt) = αl and so v is a product of simple reflections si ≤ w′ 1 1st = slw′ 1 = w′ On the other hand, by Lemma 4.5, we have w′ 1. Therefore, we have w′ sl ≤ w′ such that siw′ 1 = w′ 1sp for some integer p 6= bk such that sp ≤ w′ 1. This proves the claim. Now, let v ∈ W (Gw′ 1 such that siw′ si ≤ w′ 1,w′ 1x′ 1 = w′ 1). Then by the claim, v is a product of simple reflections 1, Tw′ 1sp for some integer p 6= bk such that sp ≤ w′ 1. On the other hand, by Lemma 3.7, for any such si, we have siwx = wx and hence vwx = wx. This completes the proof of Theorem 4.1. 4.2. Root inequality in type An. We keep the notation of Subsection 4.1. In particular, w ∈ W denotes a minuscule element, and P := P w = P αr for a unique integer 1 ≤ r ≤ n. Recall that there exists a unique sequence 1 ≤ a1 < a2 < . . . < ar ≤ n of integers such that (w1,a1, w2,a2, . . . , wr,ar) is a generalised reduced decomposition of w. Also, recall that J(w) := {2 ≤ j ≤ r : aj − aj−1 ≥ 2}S{1} is identified with Peaks(Qw). Let (cid:22) be an arbitrary ordering of the elements of J(w). Let bw′ = (w′ m) be the generalised reduced decomposition of w obtained by construction 1 corresponding to this ordering. 2, . . . , w′ 1, w′ Then we have Proposition 4.7. Let α be the unique simple root such that (w′ (see [BK19, Lem. 5.1]). Then we have (w′ if and only if w = w′ 1)−1(α) ≥ w−1(α). Further, (w′ 1; that is, m = 1. 1)−1(α) is a negative root 1)−1(α) = w−1(α) Proof. We first claim that the above proposition holds for the standard increasing ordering 1 = j1 < j2 < · · · < jm ≤ r of the elements of J(w). by construction 1 corresponding to this ordering. By this construction, we have w1 = Let bw = (w1, w2, . . . , wm) be the generalised reduced decomposition of w obtained 11 w1,a1w2,a2 · · · wj2−1,aj2−1. Therefore, we have α = αa1. Further, we have On the other hand, we have w−1(αa1) = − w−1 1 (αa1) = − αi. aj2 −1Xi=1 aj2−1+r+1−j2Xi=1 αi. Thus, we have w−1 r + 1 − j2 ≥ 1. Hence we have w−1 1 (α) > w−1(α), proving the claim. 1 (α) ≥ w−1(α). Also, if m 6= 1, then we have j2 ≤ r and hence we have We now prove the proposition for an arbitrary ordering of the elements of J(w). let j ∈ J(w) be the first element in this ordering. Let 1 ≤ q ≤ m be the integer such that 1v for j = jq. By the claim, we may assume that q 6= 1. By construction 1, we have wq = w′ some v ∈ W such that ℓ(wq) = ℓ(w′ 1) + ℓ(v). Let k = jq+1 −1. Then by construction 1, there exists a sequence l ≤ bl ≤ al (j ≤ l ≤ k) of 1 = wbj ,aj wbj+1,aj+1 · · · wbk,ak. 1 commutes with wl positive integers such that bl+1 = bl+1 for all j ≤ l ≤ k−1, and w′ Further, we have al+1 = al + 1 for all j ≤ l ≤ k − 1. By Lemma 3.8, w′ for all 1 ≤ l ≤ q − 1. Therefore, we have bj ≥ j + 1. Now, we have α = αaj . Therefore, On the other hand, we have w−1 q (αaj ) = − (w′ 1)−1(αaj ) = − αi. akXi=bj akXi=j αi. Since bj ≥ j + 1, we obtain (4.1) (w′ 1)−1(αaj ) > w−1 q (αaj ). Applying the claim to the generalised reduced decomposition v = (wq, wq+1, · · · , wm) of v = wj,aj wj+1,aj+1 · · · wr,ar obtained by construction 1 for the standard increasing ordering of {jq, jq+1, · · · , jm}, we have (4.2) w−1 q (αaj ) ≥ v−1(αaj ). Again by Lemma 3.8, we see that saj commutes with wi for all 1 ≤ i ≤ q − 1. Therefore, we have (4.3) w−1(αaj ) = v−1(αaj ). Now, the proof of the lemma follows from (4.1), (4.2) and (4.3). (cid:3) 12 4.3. Equality of Weyl groups in type Dn. Let G be of type Dn, where n ≥ 4; then G = Spin2n. We order the simple roots as in [Hu72, p. 58]; then the minuscule simple roots are α1, αn−1 and αn. We will obtain a slightly weaker version of Theorem 4.1 in this setting: Theorem 4.8. We have W (Gw1,bx, Tw1) = W (Gw1,w1x1, Tw1). Remark 4.9. The group W (Gw1,w1x1, Tw1) is not necessarily generated by simple reflections. For example, take n = 8 and w = (s4s5s6s8)(s3s4s5s6s7)(s1s2s3s4s5s6s8). Then, we have i=4 αi. Then w1 = (s4s5s6s8)(s3s4s5s6s7) for the standard ordering of Peaks(Qw). Let β =P7 v = sβ ∈ W (Gw1,w1x1, Tw1) but v is not a product of simple reflections in this group. To show Theorem 4.8, it suffices to consider the cases where P = P α1, P = P αn, since there is an automorphism of the Dynkin diagram of G sending αn−1 to αn. We begin with the easy case where P = P α1. By arguing as in the beginning of the proof of Lemma 4.6, it suffices to prove that W (Gw1,w1x1, Tw1) ⊂ W (Gw1,wx, Tw1). If there is a unique peak of Qw, then by construction 1, we have w = w1 and so we are done. Otherwise, by the same construction, we have w = snsn−1sn−2 · · · s1. Since there is an automorphism of the Dynkin diagram of G sending αn to αn−1, without loss of generality, we may assume that w1 = sn and w2 = sn−1sn−2 · · · s1. Therefore, we have W (Gw1, Tw1) = W (Gw1, Tw1) = {1, sn}. Thus, identity is the only element of W (Gw1, Tw1) that fixes w1x1. So, we are done. We now turn to the case where P = P αn = PI, where I = S \ {αn}. For 1 ≤ i ≤ n − 1, 1 ≤ ai ≤ n − 2, let vi,ai =(saisai+1 · · · sn−2sn saisai+1 · · · sn−2sn−1 if i is odd, if i is even. Let vi,n = sn if i is odd, vi,n−1 = sn−1 if i is even. Lemma 4.10. The minimal representative wI 0 ∈ W I of the longest element w0 ∈ W is of the form wI 0 =(sn−1(vn−2,n−2vn−3,n−3 · · · v1,1) sn(vn−2,n−2vn−3,n−3 · · · v1,1) if n is odd, if n is even. Further, (sn−1, vn−2,n−2, vn−3,n−3, . . . , v1,1) (respectively, (sn, vn−2,n−2, vn−3,n−3, . . . , v1,1)) is a generalised reduced decomposition of wI 0 if n is odd (respectively, if n is even). Proof. By induction on n (=rank(G)), since there is an automorphism of the Dynkin diagram of G sending αn−1 to αn, the minimal representative v ∈ W (GS\{α1}, GS\{α1} ∩ T )S\{α1,αn−1} of the longest element in W (GS\{α1}, GS\{α1} ∩ T ) is of the form v =(sn−1(vn−2,n−2vn−3,n−3 · · · v2,2) sn(vn−2,n−2vn−3,n−3 · · · v2,2) if n is odd, if n is even. Again since there is an automorphism of the Dynkin diagram of G sending αn−1 to αn, the number of positive roots of the formPi6=1 miαi, with mn ≥ 1 is equal to ℓ(v). On the 13 ℓ(v1,1). other hand, the number of positive rootsPn Therefore, the number of positive roots of the form Pi miαi with mn ≥ 1 is equal to i=1 miαi, with mn ≥ 1 and m1 ≥ 1 is equal to ℓ(v) + ℓ(v1,1). Thus, we have ℓ(wI 0) = ℓ(v) + ℓ(v1,1). This implies the assertion. (cid:3) Let w ∈ W I be such that Supp(w) = S. By Lemma 4.10 there exists a unique increasing sequence 1 = a1 < a2 < . . . < ak ≤ n of integers such that w = vk,akvk−1,ak−1 · · · v1,a1. Fur- ther, again by Lemma 4.10, (vk,ak, vk−1,ak−1, . . . , v1,a1) is a generalised reduced decomposition of w. Let J(w) := {1 ≤ j ≤ k − 1 : hαaj+1, α∨ aj i = 0} ∪ {k} and let J(w) = m. The peaks of the quiver Qw are indexed by the elements of J(w). So, let {ij : j ∈ J(w)} be the peaks of Qw. Then we have β(ij) = αaj for all j ∈ J(w). Let j1 = k > · · · > jm be the standard decreasing ordering of the elements of J(w), that is; the ordering induced by the decreasing ordering corresponding to this ordering of Peaks(Qw) obtained by construction 1. of positive integers. Let bw = (w1, w2, . . . , wm) be the generalised reduced decomposition Lemma 4.11. Let 1 ≤ l ≤ k be a positive integer. (1) Assume that l is even. Then for any integer 1 + al ≤ i ≤ n − 1, we have sivl,al = vl,alsi−1. (2) Assume that l is odd. Then for any integer 1 + al ≤ i ≤ n − 2, we have sivl,al = vl,alsi−1, and snvl,al = vl,alsn−2. Proof. We consider the case where l is even. The proof of the case where l is odd is similar. Since sisj = sjsi for all 1 + al ≤ j ≤ i − 2, sisi−1si = si−1sisi−1 and si−1st = stsi−1 for all i + 1 ≤ t ≤ n − 1, we have sivl,al = vl,alsi−1. (cid:3) Let 1 ≤ r ≤ k be the least positive integer such that hαaj+1, α∨ aj i 6= 0 for all r ≤ j ≤ k − 1. Then by construction 1, we have w1 = vk,akvk−1,ak−1 · · · vr,ar . We may now prove Theorem 4.8 for the generalised reduced decomposition associated to the above standard ordering. By arguing as in the beginning of the proof of Lemma 4.6, it suffices to prove that W (Gw1,w1x1, Tw1) ⊂ W (Gw1,wx, Tw1). Let v ∈ W (Gw1, Tw1) be such that vw1x1 = w1x1. Then, we have vw1 = w1u for some u ∈ W (Gw1, Tw1) such that sn (cid:2) u if r is odd and sn−1 (cid:2) u if r is even. We prove by induction on ℓ(u) that uvr−1,ar−1vr−2,ar−2 · · · v1,a1 = vr−1,ar−1vr−2,ar−2 · · · v1,a1 mod W αn. 14 If ℓ(u) = 1, then we have u = st for some integer ar ≤ t ≤ n and t 6= n if r is odd and t 6= n − 1 if r is even. Using Lemma 4.11 repeatedly, for any such t, we see that stvr−1,ar−1vr−2,ar−2 · · · v1,a1 = vr−1,ar−1vr−2,ar−2 · · · v1,a1sl for some integer l 6= n. So, assume that ℓ(u) ≥ 2. Choose an integer ar ≤ t ≤ n such that t 6= n if r is odd and t 6= n − 1 if r is even such that ℓ(ust) = ℓ(u) − 1. Since ℓ(ust) = ℓ(u) − 1, by induction on ℓ(u), we see that ustvr−1,ar−1vr−2,ar−2 · · · v1,a1 = vr−1,ar−1vr−2,ar−2 · · · v1,a1 mod W αn. By the above discussion, we have stvr−1,ar−1vr−2,ar−2 · · · v1,a1 = vr−1,ar−1vr−2,ar−2 · · · v1,a1sl for some integer l 6= n. Hence, we have uvr−1,ar−1vr−2,ar−2 · · · v1,a1 = vr−1,ar−1vr−2,ar−2 · · · v1,a1 mod W αn. There- fore, we have vw = w1uvr−1,ar−1vr−2,ar−2 · · · v1,a1 = w mod W αn. Thus, we obtain vwx = wx. This completes the proof for the above (special) generalised reduced decomposition. Next, we turn to the general case. Recall that J(w) = {1 ≤ j ≤ k − 1 : hαaj+1, α∨ aj i = 0} ∪ {k} and let J(w) = m. Also, let (cid:22) be an arbitrary ordering of the elements of J(w) and let j ∈ J(w) be the first element in this ordering. Since Peaks(Qw) is indexed by J(w), this ordering of J(w) induces an ordering of Peaks(Qw). 1, w′ 2, . . . , w′ m) be the generalised reduced decomposition of w obtained by construction 1 corresponding to this ordering of Peaks(Qw). Let 1 ≤ q ≤ m be the integer such that j = jq. Again by construction 1, we have wq = w′ 1v for some v ∈ W such that ℓ(wq) = ℓ(w′ Let bw′ = (w′ Let bX(bw′) be the variety corresponding to this ordering. Recall that bX(bw) denotes the variety corresponding to the standard decreasing ordering of the elements of J(w). Recall from Lemma 3.1 that W (Gw′ 1) + ℓ(v). 1) ⊂ W (Gw′ 1) ⊂ W . 1,bx′, Tw′ 1, Tw′ 1x′ 1,w′ Proposition 4.12. Assume that q 6= 1. That is, w1 6= w′ 1. Then W (Gw′ 1,bx′, Tw′ 1) = W (Gw′ 1,w′ 1x′ 1, Tw′ 1) and this group is generated by simple reflections. Proof. As at the beginning of the proof of Lemma 4.6, it suffices to prove the inclusion W (Gw′ 1,w′ 1x′ 1, Tw′ 1) ⊂ W (Gw′ 1,wx, Tw′ 1). Recall the definition of Qw(A) for any A ⊂ Peaks(Qw) from Definition 3.6. By construction 1, for any vertex l of Qw(Peaks(Qw) \ {ij}) and for any vertex of i of Qw({ij}), we have (l, i) /∈ R. In particular, we have hβ(l)∨, β(i)i = 0. Thus, every simple reflection se ≤ w′ 1 commutes with every simple reflection sf ≤ wr for all 1 ≤ r ≤ q − 1. Hence, w′ 1 commutes 1 = sn we are done. with w1. Further, either sn ≤ w1 or sn−1 ≤ w1. Otherwise, we have sl (cid:2) w′ 1 is of type A. 1 for all n − 2 ≤ l ≤ n. Therefore the Dynkin diagram of Gw′ 1 = sn−1 or w′ If w′ Let h = jq+1. By construction 1, we have as+1 = as + 1 for all h + 1 ≤ s ≤ j − 1. Again, by construction 1, there exists a sequence as ≤ bs ≤ n − 3 (h + 1 ≤ s ≤ j) of positive integers with bs+1 = 1 + bs for all h + 1 ≤ s ≤ j − 1 such that (w′ 1)−1 = wah+1,bh+1wah+2,bh+2 · · · waj ,bj . 15 Therefore, by Lemma 4.3, for any integer ah+1 ≤ t ≤ bj different from bh+1, we have st(w′ 1)−1 = (w′ 1)−1sp for some integer p 6= aj such that sp ≤ w′. Hence we have Observation. For any ah+1 ≤ t ≤ bj different from bh+1, we have w′ 1st = spw′ 1 for some integer p 6= aj such that sp ≤ w′. Therefore, we can imitate the proof of Lemma 4.6. Claim. W (Gw′ Let v ∈ W (Gw′ 1) is generated by simple reflections. 1,w′ 1x′ 1, Tw′ 1, Tw′ 1) be such that vw′ induction on ℓ(u) that v is a product of simple reflections sp ≤ w′ some integer t 6= bh+1 with st ≤ w′ 1. 1 = w′ 1u for some u ∈ W (Gw′ 1, Tw′ 1 such that spw′ 1). We prove by 1st for 1 = w′ If ℓ(u) = 1, then we have u = st for some integer t 6= bh+1 such that st ≤ w′ 1. Hence by Observation, we have v = sp for some integer p 6= aj such that sp ≤ w′ 1. So, assume that ℓ(u) ≥ 2. Choose an integer t 6= bh+1 such that st ≤ w′ 1 and ℓ(ust) = ℓ(u) − 1. We have vsw′ 1(αt) ∈ W (Gw′ sw′ reflections sp ≤ w′ 1(αt)w′ 1, Tw′ 1 = vw′ 1). Since ℓ(ust) = ℓ(u) − 1, by induction vsw′ 1ust. Note that w′ 1(αt) ∈ R+ \ v−1(R+). Therefore 1(αt) is a product of simple 1st = w′ 1 such that spw′ 1 = w′ 1sl for some integer l 6= bh+1 such that sl ≤ w′ 1. On the other hand, by Observation, we have w′ 1 for some integer e 6= aj such 1(αt) = αe and so v is a product of simple reflections 1st = sew′ that se ≤ w′ sp ≤ w′ 1. Therefore, we have w′ 1 such that spw′ 1 = w′ 1sl for some integer l 6= bh+1 such that sl ≤ w′ 1. This proves the claim. Now, let v ∈ W (Gw′ 1 such that spw′ 1). Then by the claim, v is a product of simple reflections sp ≤ w′ 1. On the other hand, by Lemma 3.7, for any such sp, we have spwx = wx and hence we have vwx = wx. (cid:3) 1sl for some integer l 6= bh+1 such that sl ≤ w′ 1,w′ 1 = w′ 1, Tw′ 1x′ 4.4. Root inequality in type Dn. We keep the notation of Subsection 4.3 In particular, w ∈ W denotes a minuscule element, P := P w = P αr for some r = 1, n − 1, n, and (cid:22) an m) be the generalised 2, . . . , w′ 1, w′ reduced decomposition of w obtained by construction 1 for this ordering. Then we have arbitrary ordering of the elements of J(w). Let bw′ = (w′ Proposition 4.13. Let α be the unique simple root such that (w′ (see [BK19, Lem. 5.1]). Then we have (w′ if and only if w = w′ 1)−1(α) ≥ w−1(α). Further, (w′ 1; that is, m = 1. 1)−1(α) is a negative root 1)−1(α) = w−1(α) Proof. It suffices to consider the cases where P = P α1, P = P αn, since there is an automor- phism of the Dynkin diagram of G sending αn−1 to αn. Also, the case where P = P α1 is easy: if there is a unique peak of Qw, then by construction 1, we have w = w1 and so we are done. Otherwise, by the same construction, we have w = snsn−1sn−2 · · · s1. Using again the existence of an automorphism of the Dynkin diagram sending αn to αn−1, without loss of generality, we may assume that w1 = sn and w2 = 16 sn−1sn−2 · · · s1. Therefore, we have α = αn and w−1 w−1(αn). Hence, we are done. 1 (αn) = −αn > −(αn +Pn−2 l=1 αl) = Now, we turn to the case P = P αn. We adapt the argument of the proof of Proposition 4.7. Recall that by Lemma 4.10 there exists a unique increasing sequence 1 = a1 < a2 < . . . < ak ≤ n of integers such that w = vk,akvk−1,ak−1 · · · v1,a1. Claim. The proposition holds for the standard decreasing ordering j1 = k > j2 > · · · > jm of the elements of J(w). Here again, we may assume that m ≥ 2. Let r = j2 + 1. We consider the case where k is even. The proof for the case where k is odd is similar. By construction 1, we have w1 = vk,akvk−1,ak−1 · · · vr,ar . Further, we have α = αak . Therefore, we have w−1 l=ar αl) if r = k l=ar αl) if r = k − 1 −(Pn−1 −(Pn −(Pn−(k+1−r) 1 (α) = αl + 2(Pn−2 Subcase 1. r = 2. We have w−1(α) = −(Pn Subcase 2. r ≥ 3. We have Case 1. r = k. l=a2−1 αl). l=ar l=n−(k−r) αl) + αn−1 + αn) if r ≤ k − 2. . w−1(α) = −( Case 2. r = k − 1. We have w−1(α) = −( Case 3. r ≤ k − 2. We have w−1(α) = −( n−kXl=ak−k+1 n−kXl=ak−k+1 n−kXl=ak−k+1 αl + 2( αl + 2( αl + 2( n−2Xl=n−k+1 n−2Xl=n−k+1 n−2Xl=n−k+1 αl) + αn−1 + αn). αl) + αn−1 + αn). αl) + αn−1 + αn). This completes the proof of the claim. We now prove the proposition for an arbitrary ordering of the elements of J(w). Let j ∈ J(w) be the first element in this ordering. Let 1 ≤ q ≤ m be the integer such that j = jq. By the proof for the standard decreasing ordering, we may assume that q 6= 1. By construction 1, we have wq = w′ 1v for some v ∈ W such that ℓ(wq) = ℓ(w′ 1) + ℓ(v). Let h = jq+1. Then by construction 1, there exists a sequence al ≤ bl ≤ n (h + 1 ≤ l ≤ j) of integers such that bl+1 = bl + 1 for all h + 1 ≤ l ≤ j − 1, and we have (w′ 1)−1 = wah+1,bh+1wah+2,bh+2 · · · waj ,bj . Further, we have al+1 = al + 1 for all h + 1 ≤ l ≤ j − 1. By Lemma 3.8, w′ 1 commutes with wl for all 1 ≤ l ≤ q − 1. Therefore, we have bj ≤ n − 3. 17 Now, we have α = αaj . Therefore, (w′ 1)−1(αaj ) = − bjXi=ah+1 αi. On the other hand, we have w−1 q (α) =  l=ah+1 l=ah+1 −(Pn−1 −((Pn−2 −(Pn −(Pn+h−j l=ah+1 l=ah+1 Since bj ≤ n − 3, we have αl) if j = h + 1 and is even αl) + αn) if j = h + 1 and is odd αl) if h = j − 2 αl + 2(Pn−2 l=n+h+1−j) αl) + αn−1 + αn) if h ≤ j − 3. . (4.4) (w′ 1)−1(αaj ) > w−1 q (αaj ). Applying the claim to the generalised reduced decomposition v = (wq, wq+1, · · · , wm) of v = wj,aj wj+1,aj+1 · · · wr,ar obtained by construction 1 corresponding to the standard increas- ing ordering of {jq, jq+1, · · · , jm}, we obtain (4.5) w−1 q (αaj ) ≥ v−1(αaj ). Again by Lemma 3.8, we see that saj commutes with wi for all 1 ≤ i ≤ q − 1. Therefore, we have (4.6) w−1(αaj ) = v−1(αaj ). Now, the proof of the lemma follows from (4.4), (4.5) and (4.6). (cid:3) 4.5. Type E6. Let G be of type E6. We order the simple roots as in [Hu72, p.58]; then the minuscule simple roots are α1 and α6. Since there is an automorphism of the Dynkin diagram of G taking α1 to α6, it is sufficient to consider Schubert varieties in G/P , where P = P α1 = PI with I = S \ {α1}. A reduced decomposition of the longest element wI 0 ∈ W I is (s6, s5, s4, s3, s2, s4, s1, s3, s5, s4, s6, s5, s2, s4, s3, s1) and all the w ∈ W I are obtained by taking certain reduced subexpressions of the above one. For any such w, we define the standard ordering on Peaks(Qw) as the ordering induced by the standard increasing ordering on its vertices (viewed as positive integers). Using the smoothness criterion of [Pe07, Thm. 7.11], one may check that the w ∈ W I such that s6s5s2s4s3s1, Supp(w) = S and bX(bw) is smooth for this standard ordering are exactly the following: s6s5s4s3s2s4s1s3s5s4s6s5s2s4s3s1 = wI s4s1s3s5s4s6s5s2s4s3s1, 0. We now describe the varieties associated to the generalised reduced decompositions of these elements obtained by construction 1. s2s4s1s3s5s4s6s5s2s4s3s1, s1s3s4s6s5s2s4s3s1, s3s4s6s5s2s4s3s1, s4s6s5s2s4s3s1, If w = wI 0, then X(w) = G/P . Thus, there is a unique peak and bX(bw) = X(w). 18 In all other cases, there are two peaks and hence two decompositions, bw (for the standard ordering) and bw′ (for the nonstandard one). For w = s6s5s2s4s3s1, we have bw = (s6s5, s2s4s3s1). Thus, bf : bX(bw) −→ X(s6s5) ≃ P2 is a fibration with fiber X(s2s4s3s1) ≃ P4. Moreover, bw′ = (s2, s6s5s4s3s1). So, the morphism bf ′ : bX(bw′) −→ X(s2) ≃ P1 is a fibration with fiber X(s6s5s4s3s1) ≃ P5. For w = s4s6s5s2s4s3s1, we have bw = (s4s2, s6s5s4s3s1). Thus, bf : bX(bw) −→ X(s4s2) ≃ P2 is a fibration with fiber X(s6s5s4s3s1) ≃ P5. Also, bw′ = (s6, s4s5s2s4s3s1). The α4 is not minuscule in the Dynkin diagram of Supp(s4s5s2s4s3s1). Hence bX(bw′) is singular in view of For w = s3s4s6s5s2s4s3s1, we have bw = (s3s4s2, s6s5s4s3s1). Thus, the morphism bf : bX(bw) −→ X(s3s4s2) ≃ P3 is a fibration with fiber X(s6s5s4s3s1) ≃ P5. Also, bw′ = (s6, s3s4s5s2s4s3s1). The simple root α3 is not minuscule in the Dynkin diagram of Supp(s3s4s5s2s4s3s1). Hence bX(bw′) is singular. For w = s1s3s4s6s5s2s4s3s1, we have bw = (s1s3s4s2, s6s5s4s3s1). Thus, the morphism bf : bX(bw) −→ X(s1s3s4s2) ≃ P4 is a fibration with fiber X(s6s5s4s3s1) ≃ P5. Also, bw′ = (s6, s1s3s4s5s2s4s3s1). The simple root α1 is minuscule in the Dynkin diagram of Supp(s1s3s4s5s2s4s3s1). Hence bX(bw′) is smooth. The morphism bf ′ : bX(bw′) −→ X(s6) ≃ P1 is a fibration with fiber Gv/GvT P , where v = s1s3s4s5s2s4s3s1. Moreover, Gv is of type D5 For w = s4s1s3s5s4s6s5s2s4s3s1, we have bw = (s4s5s6, s1s3s4s5s2s4s3s1) = (s4s5s6, v), where v is as above. So, the morphism bf : bX(bw) −→ X(s4s5s6) ≃ P3 is a fibration with fiber Q8. Also, bw′ = (s1, s4s5s6s3s4s5s2s4s3s1). The simple root α4 is not minuscule in the Dynkin diagram of Supp(s4s5s6s3s4s5s2s4s3s1). Hence bX(bw′) is singular. For w = s2s4s1s3s5s4s6s5s2s4s3s1, we have bw = (s2s4s5s6, v). Thus, the morphism bf : bX(bw) −→ X(s2s4s5s6) ≃ P4 is a fibration with fiber Q8 again. Also, bw′ = of Supp(s4s5s6s3s4s5s2s4s3s1). Hence bX(bw′) is singular in this case, too. Finally, there is a unique element w ∈ W I for which bX(bw) is singular but bX(bw′) is smooth. Take w = s1s3s5s4s6s5s2s4s3s1. Then bw = (s1s3, s5s4s6s5s2s4s3s1). The simple root α5 is not minuscule in the Dynkin diagram of Supp(s5s4s6s5s2s4s3s1). Hence bX(bw) is singular. Also, bw′ = (s5s6, v). Hence bX(bw′) is smooth. The morphism bf ′ : bX(bw′) −→ X(s5s6) ≃ P2 is (s1, s2s4s5s6s3s4s5s2s4s3s1). The simple root α2 is not minuscule in the Dynkin diagram and Gv/Gv ∩ P is isomorphic to the quadric Q8. [Pe07, Thm. 7.11] again. a fibration with fiber Q8 once more. 4.6. Type E7. Let G be of type E7. Here 7 is the unique minuscule fundamental weight. Let P = P α7 = PI, where I = S \ {α7}. Then wI 0 admits the reduced decomposition (s7, s6, s5, s4, s2, s3, s1, s4, s5, s3, s4, s6, s5, s2, s4, s3, s7, s6, s5, s4, s1, s3, s2, s4, s5, s6, s7). Like in type E6, we obtain all the w ∈ W I by taking certain reduced subexpressions of the above one, and we define the standard ordering on the peaks of the associated quivers as the ordering induced by the standard increasing order on vertices. 19 Using again the smoothness criterion of [Pe07, Thm. 7.11], one may check that the w ∈ W I exactly the following: such that Supp(w) = S and bX(bw) is smooth for the standard ordering of Peaks(Qw) are s1s3s2s4s5s6s7, s4s1s3s2s4s5s6s7, s5s4s1s3s2s4s5s6s7, s6s5s4s1s3s2s4s5s6s7, s7s6s5s4s1s3s2s4s5s6s7, s3s7s6s5s4s1s3s2s4s5s6s7, s4s3s7s6s5s4s1s3s2s4s5s6s7, s2s4s3s7s6s5s4s1s3s2s4s5s6s7, s5s4s3s7s6s5s4s1s3s2s4s5s6s7, s5s2s4s3s7s6s5s4s1s3s2s4s5s6s7, s7s6s5s4s2s3s1s4s5s3s4s6s5s2s4s3s7s6s5s4s1s3s2s4s5s6s7 = wI 0. We now describe the varieties associated to all the generalised reduced decompositions of these elements obtained by construction 1. Note that w = wI ements except s5s2s4s3s7s6s5s4s1s3s2s4s5s6s7 have two peaks, and hence two decompositions, (s1, s4s3s2s4s5s6s7). The simple root α4 is not a minuscule root of the Dynkin diagram of 0 has a unique peak, and X(w) = G/P = bX(bw). All other Weyl group el- bw (for the standard ordering) and bw′ (for the nonstandard one). For w = s1s3s2s4s5s6s7, we have bw = (s1s3, s2s4s5s6s7). So, the morphism bf : bX(bw) −→ X(s1s3) ≃ P2 is a fibration with fiber X(s2s4s5s6s7) ≃ P5. Also, bw′ = (s2, s1s3s4s5s6s7). So, the morphism bf ′ : bX(bw′) −→ X(s2) ≃ P1 is a fibration with fiber X(s1s3s4s5s6s7) ≃ P6. For w = s4s1s3s2s4s5s6s7, we have bw = (s4s2, s1s3s4s5s6s7). Therefore, the morphism bf : bX(bw) −→ X(s4s2) ≃ P2 is a fibration with fiber X(s1s3s4s5s6s7) ≃ P6. Also, bw′ = Supp(s4s3s2s4s5s6s7). Therefore, bX(bw′) is singular. For w = s5s4s1s3s2s4s5s6s7, we have bw = (s5s4s2, s1s3s4s5s6s7). Thus, the morphism bf : bX(bw) −→ X(s5s4s2) ≃ P3 is a fibration with fiber X(s1s3s4s5s6s7) ≃ P6. Also, bw′ = (s1, s5s4s3s2s4s5s6s7). Since α5 is not a minuscule root of the Dynkin diagram of Supp(s5s4s3s2s4s5s6s7), we see that bX(bw′) is singular. For w = s6s5s4s1s3s2s4s5s6s7, we have bw = (s6s5s4s2, s1s3s4s5s6s7). So, the morphism bf : bX(bw) −→ X(s6s5s4s2) ≃ P4 is a fibration with fiber X(s1s3s4s5s6s7) ≃ P6. Also, bw′ = (s1, s6s5s4s3s2s4s5s6s7). The simple root α6 is not a minuscule root of the Dynkin diagram of Supp(s6s5s4s3s2s4s5s6s7). Therefore, bX(bw′) is singular. For w = s7s6s5s4s1s3s2s4s5s6s7, we have bw = (s7s6s5s4s2, s1s3s4s5s6s7). So, the morphism bf : bX(bw) −→ X(s7s6s5s4s2) ≃ P5 is a fibration with fiber X(s1s3s4s5s6s7) ≃ P6. Also, bw′ = (s1, s7s6s5s4s3s2s4s5s6s7). So, the morphism bf ′ : bX(bw′) −→ X(s1) ≃ P1 is a fibration with fiber Gv/GvT P , where v = s7s6s5s4s3s2s4s5s6s7. Moreover, Gv is of type D6 and Gv/GvT P is isomorphic to the quadric Q10. For w = s3s7s6s5s4s1s3s2s4s5s6s7, we have bw = (s3s1, v), where v is as above. Therefore, the morphism bf : bX(bw) −→ X(s3s1) ≃ P2 is a fibration with fiber Q10. Also, bw′ = (s7s6s5, s3s4s1s3s2s4s5s6s7). The simple root α3 is not a minuscule root of Supp(s3s4s1s3s2s4s5s6s7). Therefore, bX(bw′) is singular. 20 For w = s4s3s7s6s5s4s1s3s2s4s5s6s7, we have bw = (s4s3s1, v). Therefore, the mor- phism bf : bX(bw) −→ X(s4s3s1) ≃ P3 is again a fibration with fiber Q10. Also, we have bw′ = (s7s6, s4s3s5s4s1s3s2s4s5s6s7). The simple root α4 is not a minuscule root of Supp(s4s3s5s4s1s3s2s4s5s6s7). Therefore, bX(bw′) is singular. For w = s2s4s3s7s6s5s4s1s3s2s4s5s6s7, we have bw = (s2s4s3s1, v). Therefore, the mor- phism bf : bX(bw) −→ X(s2s4s3s1) ≃ P4 is still a fibration with fiber Q10. Also, we have bw′ = (s7s6, s2s4s3s5s4s1s3s2s4s5s6s7). The simple root α2 is not a minuscule root of Supp(s2s4s3s5s4s1s3s2s4s5s6s7). Therefore, bX(bw′) is singular. For w = s5s4s3s7s6s5s4s1s3s2s4s5s6s7, we have bw = (s5s4s3s1, v). Therefore, the mor- phism bf : bX(bw) −→ X(s2s4s3s1) ≃ P4 is a fibration with fiber Q10 once more. Also, we have bw′ = (s7, s5s4s3s6s5s4s1s3s2s4s5s6s7). The simple root α5 is not a minuscule root of Supp(s5s4s3s5s4s1s3s2s4s5s6s7). Therefore, bX(bw′) is singular. ordering, we have bw = (s5, s2s4s3s1, s7s6s5s4s3s2s4s5s6s7). Thus, bX(bw) is a tower of fibra- tions with fibers P1, P4 and Gv/GvT P , where v is as above. There is an ordering for which bw′ = (s2, s5s4s3s1, s7s6s5s4s3s2s4s5s6s7). In this case also, bX(bw′) is a tower of fibrations with fibers P1, P4 and Gv/GvT P . For all other orderings, bX(bw′) is singular. Finally, if bX(bw) is singular for some w ∈ W I, then bX(bw′) is also singular for any gener- alised reduced decomposition bw′ obtained by construction 1 corresponding to a nonstandard For w = s5s2s4s3s7s6s5s4s1s3s2s4s5s6s7, there are three peak elements. For the standard 4.7. Equality of Weyl groups and root inequality in exceptional types. Let G be of type E6 or E7. As in Section 4, we consider a minuscule parabolic subgroup P = PI of G, a ordering. Weyl group element w ∈ W I, and a generalised reduced decomposition bw = (w1, w2, . . . , wm) of w, obtained by construction 1 corresponding to any ordering of Peaks(Qw) such that bX(bw) is smooth. Then Theorem 4.1 adapts to this setting: Theorem 4.14. With the above notation, we have and this group (viewed as a subgroup of W ) is generated by simple reflections. W (Gw1,bx, Tw1) = W (Gw1,w1x1, Tw1) type A. Therefore, we can adapt the proof of Proposition 4.12. Proof. By descriptions of bX(bw) in 4.5 and 4.6, we see that the Dynkin diagram of Gw1 is of Also, by using the same descriptions, one may readily check the following: (cid:3) Proposition 4.15. Let α be the unique simple root such that w−1 [BK19, Lem. 5.1]). Then we have w−1 1 (α) ≥ w−1(α). Further, w−1 if w = w1; that is, m = 1. 1 (α) is a negative root (see 1 (α) = w−1(α) if and only [BK19] M. Brion, S. Senthamarai Kannan, Minimal rational curves on generalized Bott-Samelson varieties, in preparation. 21 References [BP99] M. Brion, P. Polo, Generic singularities of certain Schubert varieties, Math. Z. 231 (1999), 301 -- 324. [Hu72] J. E. Humphreys, Introduction to Lie algebras and representation theory, Springer-Verlag, Berlin, Heidelberg, New York, 1972. [Pe07] N. Perrin, Small resolutions of minuscule Schubert varieties, Compositio Math. 143 (2007), 1255 -- 1312. [SV94] P. Sankaran, P. Vanchinathan, Small resolutions of Schubert varieties in symplectic and orthogonal Grassmannians, Publ. RIMS, Kyoto Univ. 30 (1994), no. 3, 443 -- 458. [SV95] P. Sankaran, P. Vanchinathan, Small resolutions of Schubert varieties and Kazhdan-Lusztig polyno- mials, Publ. RIMS, Kyoto Univ. 31 (1995), no. 3, 465 -- 480. [St97] J. R. Stembridge, Minuscule elements of Weyl groups, J. Algebra 235 (2001), no. 2, 722 -- 745. 22
1911.05674
1
1911
2019-11-13T17:54:15
Betti numbers of stable map spaces to Grassmannians
[ "math.AG" ]
Let $\bar{M}_{0,n}(G(r,V), d)$ be the coarse moduli space of stable degree $d$ maps from $n$-pointed genus $0$ curves to a Grassmann variety $G(r,V)$. We provide a recursive method for the computation of the Hodge numbers and the Betti numbers of $\bar{M}_{0,n}(G(r,V), d)$ for all $n$ and $d$. Our method is a generalization of Getzler and Pandharipande's work for maps to projective spaces.
math.AG
math
Betti numbers of stable map spaces to Grassmannians Massimo Bagnarol Abstract Let M 0,n(G(r, V ), d) be the coarse moduli space of stable degree d maps from n-pointed genus 0 curves to a Grassmann variety G(r, V ). We provide a recursive method for the computation of the Hodge numbers and the Betti numbers of M 0,n(G(r, V ), d) for all n and d. Our method is a generalization of Getzler and Pandharipande's work for maps to projective spaces. Contents 1 Introduction 2 Composition structures on Grothendieck rings 2.1 Grothendieck groups of varieties . . . . . . . . . . . . . . . . . . 2.2 Grothendieck groups of mixed Hodge structures . . . . . . . . . . 2.3 The Hodge-Grothendieck characteristic . . . . . . . . . . . . . . . 3 Motivic study of spaces of genus 0 stable maps 3.1 Stratification by stable marked trees . . . . . . . . . . . . . . . . 3.2 Stable maps from genus 0 curves to Grassmannians . . . . . . . . 3.3 The HG-characteristic of the open stratum . . . . . . . . . . . . 4 Spaces of morphisms from P1 to Grassmannians 4.1 Quot compactification and its decomposition . . . . . . . . . . . 4.2 The classes of the strata in K0(Vark) . . . . . . . . . . . . . . . . 4.3 The class of Mord(P1, G(r, V )) in K0(Vark) . . . . . . . . . . . . 5 The Betti numbers of M 0,n(G(r, V ), d) 5.1 The HG-characteristic of M 0,0(G(2, 4), 2) 5.2 The HG-characteristic of M 0,1(G(2, 4), 2) . . . . . . . . . . . . . . . . . . . . . . . . . . References 1 Introduction 1 3 4 7 10 11 11 14 20 21 22 28 31 32 33 34 34 The relevance of Kontsevich's moduli spaces of stable maps (see [Kon95]) has by now become widespread, notably in connection with Gromov-Witten theory. These spaces are compactifications of spaces of morphisms from nonsingular 1 curves to a fixed variety, which are obtained by allowing the domain curves to have nodes. Let us briefly recall their definition in the genus 0 complex case, which is the one we deal with in this paper. For an exhaustive and general treatment, we refer to [BM96] and [FP97]. Let Y be a nonsingular projective C-variety, and let β ∈ H2(Y, Z) be a curve class. A stable map from an n-pointed genus 0 curve to Y of class β, or stable (Y, 0, n, β)-map for short, is a triple (C, x, f ), where C is a projective, connected, reduced, at worst nodal curve of arithmetic genus 0, x = (xi)1≤i≤n is an n-tuple of pairwise distinct regular points of C, and f : C → Y is a morphism such that f∗[C] = β. These data are subject to the following stability condition: if C′ is an irreducible component of C that is contracted to a point by f , then C′ has at least three special points (i.e., either marked points or nodes). Stable (Y, 0, n, β)-maps are naturally parametrized by a proper algebraic Deligne-Mumford stack M0,n(Y, β), with a projective coarse space M 0,n(Y, β). If Y is convex, then the stack M0,n(Y, β) is nonsingular and M 0,n(Y, β) is a normal variety which is locally the quotient of a nonsingular variety by a finite group. Furthermore, the boundary of M 0,n(Y, β), i.e., the locus of singular domain curves, is a divisor with normal crossings, up to a finite group quotient. It can be given a combinatorial description similar to that of the boundary of the moduli space M 0,n of stable n-pointed genus 0 curves. Despite the relevance of stable map spaces, their topology is still not well understood. For instance, their cohomology ring has been studied only in few cases (see [BO03] and [MM07]). In the case of genus 0 stable maps to a flag variety Y , it is known that the cohomology groups are generated by tautological classes (see [Opr06]). In fact, computing the Betti numbers of M 0,n(Y, β) is already a nontrivial problem. For Y = Pr, this problem was solved in [GP06] by Getzler and Pandharipande, who computed the Hodge and Betti numbers of M 0,n(Pr, d) by determining its Hodge-Grothendieck characteristic (referred to as Serre characteristic in [GP06]), i.e., a refined Sn-equivariant version of the E-polynomial. The aim of this paper is to extend the method of [GP06] to the case of stable maps to a finite dimensional Grassmann variety G(r, V ). For some special values of n and d, there are some computations of the Poincaré polynomial of these spaces in the literature (see, for instance, [Ló14]), which are performed using the Białynicki-Birula decomposition determined by a certain torus action. However, such computations are limited to n = 0 and d ≤ 3. Therefore, in order to determine the Hodge and Betti numbers of M 0,n(G(r, V ), d) for all n and d, we developed an alternative method that generalizes [GP06]. The outcome of the paper is a recursive algorithm through which the Hodge-Grothendieck characteristic e(M 0,n(G(r, V ), d)) can be computed, for any V, r, n, d. The main point of our method is the same as that of [GP06]. Using the combinatorial description of the boundary of M 0,n(G(r, V ), d) via stable marked trees, we recursively reduce the computation of the Hodge-Grothendieck charac- teristic of M 0,n(G(r, V ), d) to that of the open loci M0,m(G(r, V ), δ) (for δ ≤ d and m ≤ n + d − δ) where the domain curves are nonsingular. This reduction is obtained by translating the combinatorial properties of the spaces under consid- 0 (Var))[[q]] eration into recursive relations in the Grothendieck rings (Qn≥0 K Sn and (Qn≥0 K Sn 0 (MHS))[[q]], via the composition structures on those rings. 2 The Hodge-Grothendieck characteristic of M0,m(G(r, V ), δ) is determined by that of the space of degree δ morphism from P1 to G(r, V ), and by that of the configuration space F(P1, m), which has already been calculated in [Get95]. This way, the original problem boils down to the computation of e(Morδ(P1, G(r, V ))). Finally, by means of the Quot scheme compactification of Morδ(P1, G(r, V )) studied by Strømme [Str87], we provide a method to explicitly compute its Hodge-Grothendieck characteristic. Together with the previous results, this gives a recursive algorithm to determine e(M 0,n(G(r, V ), d)). The paper is organized as follows. In Section 2, we describe the composition structures on the rings (Qn≥0 K Sn 0 (Var))[[q]] and (Qn≥0 K Sn 0 (MHS))[[q]]. These structures, together with the stratification of M 0,n(G(r, V ), d) by strata corresponding to stable marked trees, are used in Section 3 to obtain suitable recursive relations in the above Grothendieck rings. The calculation of the class of Morδ(P1, G(r, V )) in the Grothendieck ring of varieties, which in turn determines its Hodge-Grothendieck characteristic, is the subject of Section 4. Section 5 contains some exemplifying applications of our algorithm. Notation and conventions. For all categories C and D such that C is small, [C, D] denotes the category of functors from C to D, with natural transformations of such functors as morphisms. For all n ∈ Z≥0, Sn denotes the symmetric group on n elements. For any field k, by a k-variety we mean a reduced separated scheme of finite type over k, not necessarily irreducible. Unless otherwise stated, the field C of complex numbers is assumed to be the ground field. In order to distinguish between set-theoretic and scheme-theoretic disjoint union, the first is denoted by ⊔, while the latter is denoted by ∐. For any Cartesian diagram X ×Y T T X Y and any morphism ψ : F → G of sheaves on X, the pullback of ψ to X ×Y T is denoted by ψT : FT → GT . In particular, if Spec(κ(y)) → Y is the morphism determined by a point y ∈ Y , then the pullback of ψ to Xy is denoted by ψy : Fy → Gy. Acknowledgements The paper is based on my Ph.D. thesis [Bag19], which was written while I was a student at SISSA (Internation School for Advanced Studies) in Trieste. I am deeply indebted to Fabio Perroni and Barbara Fantechi, my thesis advisors. I would also like to thank Orsola Tommasi and Dan Petersen for their helpful comments. 2 Composition structures on Grothendieck rings In this section, we study the algebraic setting where our calculations will take place, namely the Grothendieck rings of varieties and mixed Hodge structures 3 with symmetric group actions. A relevant role is played by their composition structures, which are essential to translate the combinatorial properties of stable map spaces into relations in these rings, as we will see in §3. Our exposition is mainly based on that of [GP06], which we slightly extend in order to provide a suitable setting for studying the Betti numbers of stable map spaces to any nonsingular, projective, convex variety. For further details and proofs, the interested reader is invited to consult the author's Ph.D. thesis [Bag19, §1]. 2.1 Grothendieck groups of varieties Let k be a field. For simplicity, we assume that k is algebraically closed. Definition 2.1. The Grothendieck group K0(Vark) of k-varieties is the abelian group generated by the isomorphism classes of k-varieties subject to the relations [X] = [Y ] + [X \ Y ] , where X is a k-variety and Y ⊆ X is a closed subvariety. The Grothendieck group K0(Vark) admits different presentations, especially if char(k) = 0 (see [Bit04]). The presentation we will mainly use is the following. Proposition 2.2. The Grothendieck group of k-varieties can be alternatively presented as the abelian group generated by the isomorphism classes of quasi- projective k-varieties subject to the relations [X] = [Y ] + [X \ Y ], where X is a quasi-projective k-variety and Y ⊆ X is a closed subvariety. The group K0(Vark) is actually a commutative ring, whose product is defined on generators by [X][Y ] := [X ×k Y ] . The class of the affine line A1 k in K0(Vark) is denoted by L. As proved in [Bor18], L is a zero divisor in K0(Vark) when char(k) = 0. Let us recall the following well-known properties of K0(Vark), which will be used in the next sections. Proposition 2.3. Let X be a k-variety. Assume that there is a decomposition X = Y1 ⊔ · · · ⊔ Ym, where each Yi is a locally closed subvariety of X. Then [X] = [Y1] + · · · + [Ym] in K0(Vark). Proposition 2.4. Let f : X → Y be a morphism of k-varieties. If f is a locally trivial fibration in the Zariski topology, with fiber F , then in K0(Vark). [X] = [Y ][F ] As a consequence of Proposition 2.3, the decomposition Pr k = A0 k ⊔ · · · ⊔ Ar k implies that [Pr k] = 1 + L + · · · + Lr. If G is a finite group, there are G-equivariant versions of the Grothendieck group of k-varieties. In this paper, we only deal with the case where k is the field C of complex numbers and G is the symmetric group Sn on n elements. Hereinafter, all varieties are understood to be over C, and the notation VarC is shortened to Var. 4 Definition 2.5. The Grothendieck group K Sn 0 (Var) of Sn-varieties is the abelian group generated by the isomorphism classes of quasi-projective varieties with an action of Sn, subject to the relations [X] = [Y ] + [X \ Y ] whenever Y is a closed Sn-invariant subvariety of X. When studying moduli spaces of stable maps to a nonsingular projective variety Y , an important role is played by the group of generalized power series (cid:0)Qn≥0 K Sn This group can be itself interpreted as a suitable Grothendieck group. 0 (Var)(cid:1)[[Υ]], where Υ ⊆ H2(Y, Z) is the monoid of curve classes of Y . Let S be the permutation groupoid, i.e., the category whose objects are the nonnegative integers n ∈ Z≥0, with S(m, n) = ∅ if m 6= n and S(n, n) = Sn. Definition 2.6. A Υ-graded S-variety is a functor from S× Υ to the category Var of quasi-projective varieties. A morphism of Υ-graded S-varieties is a natural transformation of such functors. The quasi-projectivity assumption is made to guarantee that quotients by finite group actions exist in Var. If X is a Υ-graded S-variety, X(n, β) denotes the image of (n, β) ∈ Z≥0 × Υ under X; by definition, X(n, β) is a quasi-projective variety with an Sn-action. 0 (VarΥ) of Υ-graded S-varieties is Definition 2.7. The Grothendieck group K S the abelian group generated by the isomorphism classes of Υ-graded S-varieties subject to the relations [X] = [Y ] + [X \ Y ] , where X is a Υ-graded S-variety, Y (n, β) is a closed Sn-invariant subvariety of X(n, β) for every (n, β), and (X \ Y )(n, β) := X(n, β) \ Y (n, β). A {0}-graded S-variety is simply called an S-variety, and the corresponding 0 (Var). A Z≥0-graded S-variety is called a Grothendieck group is denoted by K S graded S-variety. There are canonical isomorphisms K S 0 (VarΥ) ∼= K S 0 (Var)[[Υ]] ∼= Yn≥0 K Sn 0 (Var)![[Υ]] . The presentation of this group as the Grothendieck group of Υ-graded S-varieties endows it with some additional structures, as shown in [GP06]. Remark 2.8. The structures we are going to describe can be traced back to Joyal's theory of combinatorial species (see [Joy81] and [BLL97]). Furthermore, they are closely related with Kelly's formalism for operads of [Kel05]. For all n ∈ Z≥0, let Sn be the Υ-graded S-variety defined by Sn(i, β) = ∅ if (i, β) 6= (n, 0) and Sn(n, 0) = Spec(C) with the trivial Sn-action. The element [Sn] ∈ K S 0 (VarΥ) is denoted by sn. The category [S × Υ, Var] of Υ-graded S-varieties is symmetric monoidal, with the product given by Day's convolution (X ⊠ Y )(n, β) := aα+γ=β ai+j=n IndSn Si×Sj (X(i, α) × Y (j, γ)) and the identity object S0. Here, IndSn is defined by the left Kan extension along the inclusion of one-object categories Si × Sj → Sn. As Day's convolution Si×Sj 5 product preserves coproducts in both arguments, it induces a commutative ring structure on K S 0 (VarΥ), whose product is defined on generators by [X][Y ] := [X ⊠ Y ] . Let K0(VarΥ) ∼= K0(Var)[[Υ]] be the subring of K S elements [X] such that X(n, β) = ∅ for all n > 0. Its inclusion into K S induces a commutative K0(VarΥ)-algebra structure on K S the subalgebras 0 (VarΥ) generated by the 0(VarΥ) 0 (VarΥ). Furthermore, Fn :=* n [i=0(cid:8)[X] X(j, β) = ∅ if j < n − i or mβ < i(cid:9)+ 0 (VarΥ) is complete with respect form a decreasing filtration of K S to this filtration. Here, for any β 6= 0, mβ is the maximum integer m such that 0 (VarΥ), and K S the set {(β1, . . . , βm) ∈ Υm β =Pi βi, βi 6= 0 ∀ i} is nonempty, while m0 = 0. There is another partially defined monoidal structure on [S × Υ, Var]. Its tensor product is the composition (also known as substitution or plethysm) (X ◦ Y )(n, β) := aα+γ=βai≥0(cid:0)X(i, α) × Y ⊠i(n, γ)(cid:1)/Si , which is defined whenever Y (0, 0) = ∅, and its identity object is S1. Remark 2.9. The composition product provides a conceptual definition of an operad with values in Var: a Var-valued operad P with P(0) = ∅ is a monoid in ([S, Var], ◦, S1). For further details, see [Kel05]. By [GP06, §2], the composition product induces a well-defined associative operation ◦ : K S 0 (VarΥ) × F1 → K S 0 (VarΥ) such that [X] ◦ [Y ] := [X ◦ Y ] . As the functor −◦Y preserves coproducts and commutes with Day's convolution, this operation is a K0(VarΥ)-algebra homomorphism in the first argument. On the other hand, the functor X ◦ − does not preserve coproducts. For instance, if X = Sn then Sn ◦ (Y ∐ Y ′) ∼= ai+j=n (Si ◦ Y ) ⊠ (Sj ◦ Y ′) Finally, there is an endofunctor D of [S × Υ, Var] which is defined by (DX)(n, β) := X(n + 1, β) , where the Sn-action on X(n + 1, β) is the restriction of the given action of Sn+1 to its subgroup {σ ∈ Sn+1 σ(n + 1) = n + 1} ∼= Sn. This endofunctor preserves coproducts, and DSn = Sn−1 for all n ≥ 1. Moreover, D satisfies both Leibniz's rule D(X ⊠ Y ) ∼= (DX ⊠ Y ) ∐ (X ⊠ DY ) and the chain rule D(X ◦ Y ) ∼= (DX ◦ Y ) ⊠ DY . Thus, it induces a K0(VarΥ)-derivation D on K S 0 (VarΥ) such that D[X] := [DX]. Altogether, the structures we have introduced form a composition algebra 0 (VarΥ), in the sense of [GP06, Def. 1.1]. For the reader's con- structure on K S venience, we recall the relevant definition. 6 Definition 2.10. Let R be a commutative ring, and let A be a commutative alge- bra over R which is filtered by subalgebras A = F0 ⊇ F1 ⊇ . . . . A composition operation on A is an operation ◦ : A × F1 → A such that Fn◦Fm ⊆ Fnm, endowed with a derivation D such that D(Fn) ⊆ Fn−1, and a sequence of elements sn ∈ Fn for n ∈ Z≥0, satisfying the following axioms: (i) for fixed b ∈ F1, the map − ◦ b : A → A is an R-algebra endomorphism; (ii) (a ◦ b1) ◦ b2 = a ◦ (b1 ◦ b2) for any a ∈ A, b1, b2 ∈ F1; (iii) D(a ◦ b) = (Da ◦ b)(Db) for any a ∈ A, b ∈ F1; (iv) s0 = 1, s1 ◦ b = b for all b ∈ F1, a ◦ s1 = a for all a ∈ A, Dsn = sn−1, and sn ◦ (b + b′) =Pi+j=n(si ◦ b)(sj ◦ b′) for all b, b′ ∈ F1. An algebra with a composition operation is called a composition algebra. If it is also complete with respect to F•, then it is called a complete composition algebra. The previous discussion can now be summarized in the following result (see [GP06, Thm. 2.2 and §3]). Theorem 2.11. The Grothendieck group of Υ-graded S-varieties K S a complete composition algebra over K0(VarΥ). 0 (VarΥ) is Since the monoid of curve classes of a finite dimensional Grassmannian is Z≥0, the only nontrivial Υ that will appear in the next sections is Υ = Z≥0, for which we have K0(VarZ≥0 ) ∼= K0(Var)[[q]] and K S 0 (VarZ≥0) ∼= K S 0 (Var)[[q]] . Nevertheless, the level of generality of the above exposition is motivated by its relevance for the study of stable maps to different targets. 2.2 Grothendieck groups of mixed Hodge structures The constructions of §2.1 are not peculiar to the category Var. For instance, if Var is replaced by any abelian tensor category (A, ⊗, 11) whose tensor product is exact, a composition algebra structure can be analogously defined on the Grothendieck group K S 0 (A) of functors from S to A. In fact, if A is also Q-linear, then K S 0 (A) acquires some additional properties, which are inherited from the theory of linear representations of symmetric groups. Our primary interest is the case where A is the category MHS of mixed Hodge structures over Q. The relevant constructions are briefly recalled below. Our main reference is [GP06, §5]; for a more detailed exposition, the reader is referred to [Get95]. Since we are interested in studying the cohomology of stable map spaces, the case of functors S × Υ → A is made explicit. As in §2.1, Υ denotes the monoid of curve classes of the nonsingular projective variety Y which is the target of the stable maps under consideration. First, recall the usual definition of the Grothendieck group of an abelian category. 7 Definition 2.12. Let C be an abelian category, which is assumed to be essentially small in order to avoid set-theoretic issues. The Grothendieck group K0(C) of C is the abelian group generated by the isomorphism classes of objects of C, subject to the relations [B] = [A] + [C] whenever 0 → A → B → C → 0 is a short exact sequence in C. In particular, for any abelian category A, Definition 2.12 also applies to the functor categories [Sn, A], [S, A] and [S × Υ, A] ∼= [S, [Υ, A]]. The corresponding Grothendieck groups are denoted by K Sn 0 (AΥ), respectively. There are canonical isomorphisms 0 (A) and K S 0 (A), K S K S 0 (AΥ) ∼= K S 0 (A)[[Υ]] ∼= Yn≥0 K Sn 0 (A)![[Υ]] . Now, let (A, ⊗, 11) be a fixed abelian tensor category over Q (or any field of characteristic 0), with the tensor product ⊗ and the identity object 11, such that A is essentially small and ⊗ is exact. The tensor product ⊗ induces a commutative ring structure on K0(A), with the identity [11]. For all n ∈ Z≥0, let Sn : S × Υ → A be the functor defined by Sn(i, β) := 0 if (i, β) 6= (n, 0) and Sn(n, 0) := 11 with the trivial Sn-action. The element [Sn] ∈ K S 0 (AΥ) is denoted by sn. Day's convolution (A ⊠ B)(n, β) := Mα+γ=β Mi+j=n IndSn Si×Sj (A(i, α) ⊗ B(j, γ)) defines a symmetric monoidal structure on [S × Υ, A], with the identity object 0(AΥ), whose product is S0. This induces a commutative ring structure on K S defined on generators by [A][B] := [A ⊠ B] . For Υ = 0, the Grothendieck ring K S 0 (A) is isomorphic to the completion of Λ⊗K0(A), where Λ is the filtered ring of symmetric functions. This isomorphism sends [sn] ∈ K S 0 (A) to s(n) ⊗ 1, where s(n) = hn is the homogeneous symmetric function of degree n, i.e., the Schur function corresponding to the partition (n) ⊢ n. As a consequence, there is a ring isomorphism K S 0 (A) ∼= K0(A)[[s1, s2, . . . ]] . Remark 2.13. In fact, K S a λ-ring (see [Get95, Thm. 4.8]). 0 (A) is isomorphic to the completion of Λ ⊗ K0(A) as Let K0(AΥ) ∼= K0(A)[[Υ]] be the subring of K S ments [A] such that A(n, β) = 0 for all n > 0. Its inclusion into K S a commutative K0(AΥ)-algebra structure on K S filtered by the subalgebras 0(AΥ) generated by the ele- 0 (AΥ) induces 0 (AΥ) is 0 (AΥ). The algebra K S Fn :=* n [i=0(cid:8)[A] A(j, β) = 0 if j < n − i or mβ < i(cid:9)+ , and it is complete with respect to this filtration. 8 The composition (usually known as plethysm in this context) (A ◦ B)(n, β) := Mα+γ=βMi≥0 A(i, α) ⊗Si B⊠i(n, γ) , which is defined whenever B(0, 0) = 0, is the tensor product of a partially defined nonsymmetric monoidal structure on [S × Υ, A]. The identity object of this monoidal structure is S1. The functor −◦B preserves biproducts and commutes with Day's convolution, whereas this does not hold true for the functor A ◦ −. For example, if A = Sn then Sn ◦ (B ⊕ B′) ∼= Mi+j=n (Si ◦ B) ⊠ (Sj ◦ B′) . (1) By means of (1) and the isomorphism K S 0 (AΥ) ∼= (K0(A)[[s1, s2, . . . , ]])[[Υ]], one easily sees that the composition induces a unique associative operation ◦ : K S 0 (AΥ) × F1 → K S 0 (AΥ) such that [A] ◦ [B] := [A ◦ B] . This operation is a K0(AΥ)-algebra homomorphism in the first argument. With respect to the second argument, it is conveniently expressed via the elements 0 (AΥ), which are analogous to the power sum symmetric functions in Λ. pn ∈ K S These elements are recursively defined by p1 = s1 , nsn = pn + s1pn−1 + · · · + sn−1pn , and there is an isomorphism K S 0 (AΥ) ⊗Z Q ∼=(cid:0)(K0(A) ⊗Z Q)[[p1, p2, . . . ]])(cid:1)[[Υ]] . The convenience of considering the elements pn lies in the next result, which follows from [Mac95, §I.8]. Proposition 2.14. For any n ∈ Z>0, the map pn ◦ − is a K0(AΥ)-algebra homomorphism. Moreover, pn ◦ pm = pnm for all n, m. As we will see in §5, Proposition 2.14 provides a way to effectively compute a ◦ b for any a ∈ K S 0 (AΥ) and b ∈ F1. Now, in order to define a composition algebra structure on K S 0 (AΥ), the last item we need is a suitable derivation. As in the case of Var, we can consider the endofunctor D of [S × Υ, A] given by Since this endofunctor preserves biproducts and satisfies both Leibniz's rule (DA)(n, β) := A(n + 1, β) . D(A ⊠ B) ∼= (DA ⊠ B) ⊕ (A ⊠ DB) and the chain rule D(A ◦ B) ∼= (DA ◦ B) ⊠ DB , it induces a K0(AΥ)-derivation D on K S 0 (AΥ), which is defined on generators by D[A] := [DA]. When applied to the elements sn and pn, the derivation D has a simple form: for all n ≥ 1, Dsn = sn−1 and Dpn = δn,1, where δi,j is the Kronecker delta. As a result of the preceding discussion, the following theorem is obtained (see [GP06, Thm. 5.1]). 9 Theorem 2.15. The Grothendieck group K S algebra over K0(AΥ). 0 (AΥ) is a complete composition The category of mixed Hodge structures over Q is a Q-linear rigid abelian tensor category, thus all the above results can be applied to A = MHS. In view of the purpose of this paper, we will be specifically concerned with the case where Υ = Z≥0, in which we have K0(MHSZ≥0) ∼= K0(MHS)[[q]] and K S 0 (MHSZ≥0) ∼= K S 0 (MHS)[[q]] . Remark 2.16. The Grothendieck group K0(MHS) is in fact isomorphic to the Grothendieck group K0(HS) of pure Hodge structures over Q (see for instance [PS08, Cor. 3.9]). The isomorphism K0(MHS) → K0(HS) is defined on gen- i (V )]. In particular, this shows that the erators by [(V, W•, F •)] 7→ Pi∈Z[GrW information about the weight filtration is lost in K0(MHS). 2.3 The Hodge-Grothendieck characteristic Let X be a variety over C. By Deligne's theory of mixed Hodge structures, each compactly supported cohomology group H i c(X, Q) carries a natural mixed Hodge structure (H i Definition 2.17. The Hodge-Grothendieck characteristic (hereinafter referred to as HG-characteristic) e : K0(Var) → K0(MHS) is the ring homomorphism that is defined on generators by c(X, Q), W•, F •) over Q. e(X) :=Xi≥0 (−1)i(cid:2)(cid:0)H i c(X, Q), W•, F •(cid:1)(cid:3) . Here and henceforth, e(X) is a shortened notation for e([X]). Remark 2.18. The homomorphism e has different names in the literature. We follow the terminology of [PS08]. In [GP06], e is called Serre characteristic. The proof of the existence of e and its properties first appeared in [DK86]. For instance, the HG-characteristic of L = [A1 C] is [Q(−1)] ∈ K0(MHS). This class is denoted by L. The HG-characteristic is a refined version of the E-polynomial, which is indeed the composition of e : K0(Var) → K0(MHS) with the ring homomorphism K0(MHS) → Z[t, u, t−1, u−1] given by [(V, W•, F •)] 7→ Xp,q∈Z dimC(cid:0)Grp F GrW p+q(V ⊗Q C)(cid:1)tpuq . If the variety X carries an action of Sn, then the Sn-representation H i c(X, Q) is compatible with the mixed Hodge structure on it, i.e., it is a Sn-representation in the category MHS. This shows that Definition 2.17 can be adapted to the context of Υ-graded S-varieties. Definition 2.19. The HG-characteristic e : K S homomorphism which is defined on generators by 0 (MHSΥ) is the ring 0 (VarΥ) → K S e(X) :=Xi≥0 (−1)i(cid:2)H i c(X, Q)(cid:3) , where H i c(X, Q)(n, β) := (H i c(X(n, β), Q), W•, F •) for all n ∈ Z≥0 and β ∈ Υ. 10 One of the main properties of the HG-characteristic is that it preserves the structures we have introduced in the previous sections. More precisely, we have the following result (see [GP06, §5]). Proposition 2.20. The HG-characteristic e : K S homomorphism of complete composition algebras. 0 (VarΥ) → K S 0(MHSΥ) is a For any projective variety X that has at most finite quotient singularities, the knowledge of e(X) determines its Hodge numbers, because the mixed Hodge structure (H i(X, Q), W•, F •) on its i-th cohomology group is in fact pure of weight i (see [PS08, Thm 2.43]). In particular, this applies to the moduli spaces of genus 0 stable maps to a finite dimensional Grassmannian. Therefore, the next sections will be concerned with the computation of the HG-characteristic of these spaces. 3 Motivic study of spaces of genus 0 stable maps Let Y be a nonsingular, projective, convex variety, and let Υ ⊆ H2(Y, Z) be its monoid of curve classes. For any n ∈ Z≥0 and β ∈ Υ, the coarse moduli space M 0,n(Y, β) of stable (Y, 0, n, β)-maps is a normal projective variety with finite quotient singularities [FP97, Thm. 2]. The variety M 0,n(Y, β) admits a stratification by locally closed subvarieties indexed by stable (n, β)-trees. When Y is a finite-dimensional Grassmanian, this stratification plays a key role in the development of a recursive procedure to compute the HG-characteristic of M 0,n(Y, β). In this section, we follow the same approach as that adopted in [GP06, §4] for the case Y = Pr. 3.1 Stratification by stable marked trees In order to fix our notation, we first recall some definitions regarding graphs. Definition 3.1. A graph τ is a quadruple (Fτ , Vτ , ∂τ , jτ ), where Fτ is a finite set of flags, Vτ is a finite set of vertices, ∂τ : Fτ → Vτ is a map, and jτ : Fτ → Fτ is an involution. The elements of Eτ := {{f, f ′} ⊆ Fτ jτ (f ) 6= f ′} Fτ are called the edges of τ, and the elements of Lτ := {f ∈ Fτ jτ (f ) = f } are called the leaves (or tails) of τ. For each vertex v ∈ Vτ , its valence n(v) is the cardinality of Fτ (v) := ∂−1 τ ({v}). An isomorphism of graphs ϕ : τ ∼−→ σ is a pair (ϕF , ϕV ), where ϕF : Fτ → Fσ and ϕV : Vτ → Vσ are bijective maps such that the diagrams ϕF ϕV Fτ ∂τ Vτ Fσ ∂σ Vσ and ϕF ϕF Fτ jτ Fτ Fσ jσ Fσ commute. Definition 3.2. A tree is a graph τ such that (i) #Vτ = #Fτ + 1, and 11 (ii) for any v, v′ ∈ Vτ , there exists a sequence of edges connecting v and v′, i.e., there exist f1, . . . , f2k ∈ Fτ such that ∂τ (f1) = v, ∂τ (f2k) = v′, jτ (f2i+1) = f2i+2 for all i = 0, . . . , k − 1, and ∂τ (f2i) = ∂τ (f2i+1) for all i = 1, . . . , k − 1. In other words, a tree is a graph whose geometric realization is simply connected (see [BM96, Def. 1.2] for the geometric realization of a graph). Now, let Υ be the monoid of curve classes of a nonsingular, projective, convex variety Y . More generally, one can replace Υ by any semigroup with indecomposable zero. We consider trees that are marked by elements of Υ, and whose leaves are labelled by nonnegative integers. Definition 3.3. Let n ∈ Z≥0 and β ∈ Υ. An (n, β)-tree is a triple (τ, lτ , βτ ), where τ is a tree, lτ : Lτ → {1, . . . , n} is a bijective map, and βτ : Vτ → Υ is a map such that Pv∈Vτ βτ (v) = β. An isomorphism of (n, β)-trees (τ, lτ , βτ ) ∼−→ (σ, lσ, βσ) is an isomorphism ϕ : τ ∼−→ σ of the underlying graphs such that the diagrams Lτ ϕF Lτ Lσ lτ lσ {1, . . . , n} and Vτ βτ ϕV Υ Vσ βσ commute. For each stable (Y, 0, n, β)-map (C, x, f ), there is an associated (n, β)-tree (τ, lτ , βτ ), which is defined as follows. • Let ν : Cν → C denote the normalization of C. The flags of τ correspond to the closed points y ∈ Cν such that ν(y) is a special point of (C, x). • The vertices of τ correspond to the irreducible components of C, and ∂τ : Fτ → Vτ sends y to the component where ν(y) lies. • If ν(y) is a marked point, then jτ (y) := y. If ν(y) is a double point, then jτ (y) := y′, where y′ 6= y is the other closed point of Cν such that ν(y) = ν(y′). In particular, the edges of τ correspond to the nodes of C, and the leaves of τ correspond to the marked points of C. • The labelling lτ : Lτ → {1, . . . , n} of the leaves is determined by the markings x = (xi)1≤i≤n: lτ (y) := i if and only if ν(y) = xi. • The map βτ : Vτ → Υ sends each irreducible component C′ of C to the class of f C ′. The stability condition on (C, x, f ) corresponds to a stability condition on its dual (n, β)-tree. Definition 3.4. An (n, β)-tree τ is stable if for each v ∈ Vτ either βτ (v) 6= 0 or n(v) > 2. In particular, a (Y, 0, n, β)-map is stable if and only if its dual (n, β)-tree is stable. Furthermore, note that the dual (n, β)-trees of two isomorphic stable (Y, 0, n, β)-maps are isomorphic as well. 12 Conversely, given an isomorphism class [τ ] of stable (n, β)-trees, one can consider the locus M (τ ) in M 0,n(Y, β) that parametrizes those maps whose dual (n, β)-tree is isomorphic to τ. This locus is a locally closed subvariety of M 0,n(Y, β) of codimension #Eτ (see [FP97, §6]). For instance, the locus that corresponds to the isomorphism class of stable (n, β)-trees with a single vertex and n leaves is the open subvariety M0,n(Y, β), which parametrizes maps from nonsingular curves. If [τ ] is a different isomor- phism class, then M (τ ) lies in the boundary M 0,n(Y, β) \ M0,n(Y, β). The decomposition of M 0,n(Y, β) into the locally closed subvarieties M (τ ) is finite, as the next result shows. Lemma 3.5. The set Γ0,n(β) of isomorphism classes of stable (n, β)-trees is finite. Proof. The proof is the same as that of [GP06, Prop. 4.3] for Y = Pr, adapted to our more general case. For every (n, β)-tree τ, we have #Vτ = #Eτ + 1 = 1 2 Xv∈Vτ n(v) − n! + 1 , (n(v) − 2) = n − 2. In particular, the number of vertices v ∈ Vτ such thusPv∈Vτ that n(v) > 2 is bounded by n − 2. Now, the set (cid:8)(β1, . . . , βi ∈ Υi β1 + · · · + βi = β , βj 6= 0 ∀ j(cid:9) is empty for almost all i; let iβ be the greatest integer for which it is nonempty. If τ is stable, then the number of vertices v ∈ Vτ such that n(v) ≤ 2 is bounded by iβ. Therefore, we have #Fτ = Xv∈Vτ n(v) = n − 2 + 2(#Vτ ) ≤ 3n + iβ − 4 . Since the number of isomorphism classes of trees with a fixed number of flags is finite, it follows that Γ0,n(β) is a finite set. As a consequence of Lemma 3.5, the equality (cid:2)M 0,n(Y, β)(cid:3) = X[τ ]∈Γ0,n(β) [M (τ )] holds in K0(Var). The variety M 0,n(Y, β) has a natural left action of the symmetric group Sn, which is given by permutation of the marked points. If we consider its class in K Sn 0 (Var), then in general the above equation does not hold, because the loci M (τ ) may be not Sn-invariant. Only the weaker equalies (cid:2)M 0,n(Y, β)(cid:3) = hold in K Sn 0 (Var).  a[τ ]∈Γ0,n(β) M (τ )  = [M0,n(Y, β)] + a[τ ]∈Γ0,n(β), Eτ 6=∅  M (τ )  As a consequence of [FP97, §6] (see also [ACG11, §XII.10] for the analogous description in the case of moduli of pointed curves), the loci M (τ ) can be given 13 a description in terms of the gluing of stable maps from nonsingular curves. For this purpose, it is useful to work with stable maps from pointed curves whose marked points are labelled by an arbitrary finite set. Definition 3.6. For any finite set L and any β ∈ Υ such that either β 6= 0 or #L > 2, M0,L(Y, β) is the coarse moduli space that parametrizes equivalence classes of triples (C, ι, f ), where C is a nonsingular projective curve of genus 0, ι : L → C is a closed immersion, and f : C → Y is a morphism such that f∗([C]) = β. The variety M0,L(Y, β) is (non-canonically) isomorphic to M0,#L(Y, β), and it is equipped with a canonical evaluation morphism M0,L(Y, β) → Y L, which maps each geometric point [(C, ι, f )] to f ◦ ι : L → Y . Let τ be a stable (n, β)-tree. Since Fτ = Fv∈Vτ phisms M0,Fτ (v)(Y, βτ (v)) → Y Fτ (v) induce a morphism Fτ (v), the evaluation mor- M0,Fτ (v)(Y, βτ (v)) → Y Fτ . Yv∈Vτ There is also a natural injection Y Eτ ⊔Lτ → Y Fτ , which is given by composing with the surjective map Fτ → Eτ ⊔ Lτ that maps each flag to its orbit under jτ . Let M(cid:3)(τ ) be the fiber product M(cid:3)(τ ) Y Eτ ⊔Lτ M0,Fτ (v)(Y, βτ (v)) Qv∈Vτ Y Fτ with respect to these morphisms. The next result follows from [FP97, §6.2] (compare also with [ACG11, Prop. 10.11] in the case of moduli of curves). Proposition 3.7. For each [τ ] ∈ Γ0,n(Y, β), there is a canonical isomorphism M (τ ) ∼= M(cid:3)(τ )/ Aut(τ ) . Proof. By [BM96, Prop. 2.4] and the definition of M(cid:3)(τ ), there is a canonical gluing morphism M(cid:3)(τ ) → M (τ ). This morphism is invariant under auto- morphisms of τ, therefore it factors through M(cid:3)(τ )/ Aut(τ ). The resulting morphism M(cid:3)(τ )/ Aut(τ ) → M (τ ) is the normalization of M (τ ), which is an isomorphism because M (τ ) is already normal. 3.2 Stable maps from genus 0 curves to Grassmannians Let V be a fixed C-vector space of dimension k ∈ Z>0, and let r be an integer such that 0 < r < k. We consider stable maps with values in the Grassmannian G := G(r, V ) of r-dimensional quotients of V . Under the isomorphism H2(G, Z) ∼= Z, the monoid of curve classes of G is identified with Z≥0. Therefore, a curve class in G will be denoted by d ∈ Z≥0. Proposition 3.8. The evaluation morphisms eva : M g,n(G, d) → G are locally trivial fibrations in the Zariski topology. 14 Proof. For notational simplicity, we identify V with Ck, via the choice of a basis of V . Throughout all the proof, by a point we mean a closed point. For any k × k matrix A and any I, J ⊆ {1, . . . , k}, we denote by A(I,J) the submatrix of A whose rows are indexed by I and whose columns are indexed by J. For every I = {i1, . . . , ik−r} ⊆ {1, . . . , k}, let UI ∼= Ar(k−r) be the open affine subset of G whose points are the quotients [q : Ck → W ] such that ker(q) ∩ h{ej j /∈ I}i = {0}. If I c = {1, . . . , k} \ I, then UI is isomorphic to the closed subgroup HI of GL(k, C) consisting of those matrices A ∈ GL(k, C) such that A(I,I) = 11k−r, A(I c,I c) = 11r and A(I,I c) = 0. An isomorphism ϕ : UI → HI is defined as follows: for any [q : Ck → W ] ∈ UI, let Kq be the k × (k − r) matrix whose column vectors span ker(q) and whose I-th minor is 11k−r; then ϕ(q) ∈ HI is the matrix such that ϕ(q)(I c,I) is the I c-th minor of Kq. The restriction of the GL(k, C)-action on G yields an action of HI on G such that UI is HI-invariant. The HI-action on UI corresponds to the translation in Ar(k−r); in particular, it is free and transitive. Let F denote the fiber (eva)−1(0I ), where 0I ∈ UI is the point corresponding to 0 ∈ Ar(k−r). The isomorphism class of F does not depend on the choice of I: for any other J ⊆ {1, . . . , k} with #J = k − r, the action of row-switching matrices of GL(k, C) on G determines an isomorphism F ∼= (eva)−1(0J ). Since the UI's form an open cover of G, the proposition will be proved if we show that there is an isomorphism (eva)−1(UI ) ∼= UI × F . Let (π : C → T, x, f ) be a stable (G, g, n, d)-map over T such that f ◦ xa factors through UI. Then we can consider the morphism f0 : C → G that sends z ∈ C to ϕ((f ◦ xa ◦ π)(z))−1 · f (z); informally, we are translating f ◦ xa to 0I. The map f0 is a morphism of class d such that f0 ◦ xa factors through {0I }. As a consequence, the mapping (π : C → T, x, f ) 7→ (f ◦ xa : T → UI, (π : C → T, x, f0)) determines a canonical morphism ψ : (eva)−1(UI ) → UI × F . Conversely, let (π′ : C′ → T, x′, f ′) be a stable (G, g, n, d)-map over T such that f ′ ◦ x′ l factors through {0I}, and let h : T → UI be a morphism. Then the map fh : C → G defined by fh(z) := ϕ((h ◦ π′)(z)) · f ′(z) is a morphism of class d such that fh ◦ x′ a = h. Therefore, the mapping (h : T → UI, (π′ : C′ → T, x′, f ′)) 7→ (π′ : C′ → T, x′, fh) determines a canonical morphism ψ′ : UI × F → (eva)−1(UI ), which is clearly the inverse of ψ. The fibers of evn+1 : M 0,n+1(G, d) → G and of its restriction to M0,n+1(G, d) will play a special role in our treatment. Definition 3.9. If p ∈ G is any closed point, Φn,d is the fiber of the morphism evn+1 : M 0,n+1(G, d) → G over p. Similarly, Φn,d is the fiber of evn+1M0,n+1(G,d) over p. Since we are only interested in the isomorphism class of the fibers, the choice of p ∈ G in the above definition is arbitrary. By identifying Sn with {σ ∈ Sn+1 σ(n + 1) = n + 1}, the restriction of the Sn+1-action on M 0,n+1(G, d) (resp. M0,n+1(G, d)) yields an Sn-action on Φn,d 15 (resp. Φn,d). Let M, M , Φ and Φ be the following graded S-varieties: M (n, d) := M 0,n(G), d) , M (n, d) := M0,n(G, d) , Φ(n, d) := Φn,d , Φ(n, d) := Φn,d . The next result is a direct consequence of Proposition 3.8. Corollary 3.10. The equalities D[M ] = [G][Φ] and D[M ] = [G][Φ] hold in K S 0 (Var)[[q]]. Following a procedure analogous to [GP06, §4], we are going to find a formula 0 (Var)[[q]]. First, we need some that relates the classes of M , M , Φ and Φ in K S preliminary results. Let us fix a representative for each isomorphism class in Γ0,n(d). The set of these representatives is denoted by Γ0,n(d). For any τ ∈ Γ0,n(d), let us also fix representatives for each equivalence class in Vτ / Aut(τ ), Eτ / Aut(τ ) and Fτ / Aut(τ ); the corresponding sets are denoted by Vτ , Eτ and Fτ , respectively. Notice that Lτ / Aut(τ ) = Lτ . For any a ∈ Vτ ⊔ Eτ ⊔ Lτ ⊔ Fτ , Aut(τ, a) denotes the group of automorphisms of τ that fix a. Finally, let us introduce the sets 0,n(d) := {(τ, v) τ ∈ Γ0,n(d) , v ∈ Vτ } , ΓV 0,n(d) := {(τ, e) τ ∈ Γ0,n(d) , e ∈ Eτ } , ΓE 0,n(d) := {(τ, l) τ ∈ Γ0,n(d) , l ∈ Lτ } , ΓL 0,n(d) := {(τ, ξ) τ ∈ Γ0,n(d) , ξ ∈ Fτ } . ΓF The following result translates the combinatorial properties of the moduli spaces we are considering into equations in K S 0 (Var)[[q]] (cf. [GP06, Lemma 4.7]). Proposition 3.11. The equations qd qd qd qd ∞ ∞ Xd=0 Xd=0 Xd=0 Xd=0 ∞ ∞ ∞ ∞   a(τ,v)∈ΓV Xn=0   a(τ,e)∈ΓE Xn=0   a(τ,l)∈ΓL Xn=0   a(τ,ξ)∈ΓF Xn=0 ∞ ∞ M(cid:3)(τ )/ Aut(τ, v)  M(cid:3)(τ )/ Aut(τ, e)  = [G](s2 ◦ [Φ]) , M(cid:3)(τ )/ Aut(τ, l)  = s1D[M ] , M(cid:3)(τ )/ Aut(τ, ξ)  0,n(d) 0,n(d) 0,n(d) 0,n(d) = [M ] ◦ (s1 + [Φ]) , = [G][Φ]2 + s1D[M ] (2) (3) (4) (5) hold in K S 0 (Var)[[q]]. 16 Proof. The result is a direct consequence of Proposition 3.7, Proposition 3.8 and the definition of ◦. For the sake of geometric intuition, a geometric interpretation of the equalities (2) -- (5) is provided below. In what follows, (C, x, f ) denotes a stable (G, 0, n, d)-map with dual (n, d)-tree τ. (2) If v ∈ Vτ is a fixed vertex and Cv is the corresponding irreducible compo- nent of C, let ν : Cv → Cv be the normalization of Cv, let zi, for i ∈ Fτ (v), be the closed points of Cv that lie over the special points of Cv, and let fv = f Cv ◦ν. Then (C, x, f ) is obtained by gluing the stable map ( Cv, (zi)i∈Fτ (v), fv) with • a stable map at each zi such that ν(zi) is a node, and • a point at each zi such that ν(zi) is a marked point. (3) If e ∈ Eτ is a fixed edge and xe is the corresponding node of C, then (C, x, f ) is obtained by gluing two stable maps, where both of them have an additional marked point that is mapped to f (xe) ∈ G, at these additional points. (4) Let l ∈ Lτ be a fixed leaf, and let λ ∈ {1, . . . , n} be its label. Then the action of Sn that permutes x1, . . . , xn is the same as that induced by the action of its subgroup {σ ∈ Sn σ(λ) = λ} ∼= Sn−1 that permutes x1, . . . , xλ, . . . , xn. (5) Let ξ ∈ Fτ be a fixed flag, and let zξ be the corresponding point of C, where ν : C → C is the normalization. If ν(zξ) is a marked point of (C, x), then we can argue as for (4). This argument yields the second summand on the right-hand side of (5). If ν(zξ) is a node of C, then we can argue as for (3). However, while in the case of edges there may be automorphisms of (τ, e = {ξ, jτ (ξ)}) that exchange ξ and jτ (ξ), such automorphisms are not in Aut(τ, ξ). This explains the lack of the quotient by S2 in the first summand on the right-hand side of (5), as compared with (3). In order to prove the above-mentioned formula, we need two other lemmas. The first one is [GP06, Lemma 4.8]. Lemma 3.12. For every stable (n, d)-tree τ, there is a canonical injective map ι : Fτ ֒→ Vτ ⊔ Eτ ⊔ Lτ . The proof of the second one is straightforward. Lemma 3.13. Let X be a quasi-projective variety, and let A be a finite set. Let G be a finite group acting on both X and A. Then any choice of representatives for the equivalence classes in A/G determines an isomorphism (A × X)/G ∼= a[a]∈A/G X/Ga , where Ga = {g ∈ G ga = a} is the stabilizer subgroup of G with respect to the chosen representative a of [a]. We can now prove the main theorem of this section (cf. [GP06, Thm. 4.5]). Theorem 3.14. The equality [M ] = [M ] ◦ (s1 + [Φ]) + [G(r, V )](cid:0)s2 ◦ [Φ] − [Φ]2(cid:1) holds in K S 0 (Var)[[q]]. 17 Proof. By Lemma 3.13, we have the equalities 0,n(d) 0,n(d)   a(τ,v)∈ΓV   a(τ,e)∈ΓE   a(τ,l)∈ΓL 0,n(d) M(cid:3)(τ )/ Aut(τ, v)  =  a[τ ]∈Γ0,n(d) M(cid:3)(τ )/ Aut(τ, e) =  a[τ ]∈Γ0,n(d)  M(cid:3)(τ )/ Aut(τ, l)  =  a[τ ]∈Γ0,n(d) (Vτ × M(cid:3)(τ ))/ Aut(τ )  , (Eτ × M(cid:3)(τ ))/ Aut(τ )  (Lτ × M(cid:3)(τ ))/ Aut(τ )  . , in K Sn 0 (Var). The sum of the right-hand sides is equal to   a[τ ]∈Γ0,n(d) ((Vτ ⊔ Eτ ⊔ Lτ ) × M(cid:3)(τ ))/ Aut(τ )  . (6) The class (6) can be rewritten by means of Lemma 3.12. Indeed, let us consider the canonical injection ι : Fτ ֒→ Vτ ⊔ Eτ ⊔ Lτ . Since τ is a tree, we have #Fτ = #(Vτ ⊔ Eτ ⊔ Lτ ) − 1, thus the injection ι induces a canonical bijection Fτ ⊔ {•} ∼−→ Vτ ⊔ Eτ ⊔ Lτ . As a consequence, (6) equals which in turn equals ((Fτ ⊔ {•}) × M(cid:3)(τ ))/ Aut(τ )  ,   a[τ ]∈Γ0,n(d) M(cid:3)(τ )/ Aut(τ, i)  M(cid:3)(τ )/ Aut(τ )  =  a[τ ]∈Γ0,n(d)   a(τ,i)∈ΓF 0,n(d)   a[τ ]∈Γ0,n(d) +  a[τ ]∈Γ0,n(d) M(cid:3)(τ )/ Aut(τ )  , again by Lemma 3.13. Furthermore, Proposition 3.7 implies that M (τ )  =(cid:2)M 0,n(G, d)(cid:3) . Finally, the above sequence of equalities, together with Proposition 3.11, yields the equation [M ] ◦ (s1 + [Φ]) + [G](s2 ◦ [Φ]) + s1D[M ] = [G][Φ]2 + s1D[M ] + [M ] in K S 0 (Var)[[q]], from which the theorem follows. Corollary 3.15. The equality holds in K S 0 (Var)[[q]]. [G](cid:0)[Φ] − [Φ] ◦ (s1 + [Φ])(cid:1) = 0 18 Proof. We apply the derivation D to both sides of Theorem 3.14. By Corollary 3.10, the left-hand side becomes D[M ] = [G][Φ]. Since D is additive, it can be applied separately to each summand on the right-hand side. Using the properties of D and Corollary 3.10, we get D(cid:0)[M ] ◦ (s1 + [Φ])(cid:1) =(cid:0)D[M ] ◦ (s1 + [Φ])(cid:1) D(s1 + [Φ]) =(cid:0)([G][Φ]) ◦ (s1 + [Φ])(cid:1)(cid:0)1 + D[Φ](cid:1) = [G](cid:0)[Φ] ◦ (s1 + [Φ])(cid:1)(cid:0)1 + D[Φ](cid:1) , then D(cid:0)[G](s2 ◦ [Φ])(cid:1) = [G] D(s2 ◦ [Φ]) = [G](cid:0)s1 ◦ [Φ](cid:1) D[Φ] = [G][Φ] D[Φ] , and finally The combination of all the above equalities yields D(cid:0)[G][Φ]2(cid:1) = 2[G][Φ] D[Φ] . [G](cid:0)1 + D[Φ](cid:1)(cid:0)[Φ] − [Φ] ◦ (s1 + [Φ])(cid:1) = 0 . Since 1 + D[Φ] is not a zero divisor in K S 0 (Var)[[q]], this implies the corollary. As the HG-characteristic e : K S 0 (MHS)[[q]] is a morphism of complete composition algebras, Theorem 3.14 and Corollary 3.15 also imply the following result. 0 (Var)[[q]] → K S Corollary 3.16. The equalities e(M ) = e(M ) ◦(cid:0)s1 + e(Φ)(cid:1) + e(G)(cid:0)s2 ◦ e(Φ) − e(Φ)2(cid:1) and hold in K S 0 (MHS)[[q]]. e(Φ) = e(Φ) ◦(cid:0)s1 + e(Φ)(cid:1) The main upshot of Theorem 3.14 and its corollaries is that they provide a recursive procedure to compute the HG-characteristic of M 0,n(G, d) for all n and d. The first equation of Corollary 3.16 shows that e(M 0,n(G, d)) ∈ K Sn 0 (MHS) can be obtained from the following elements: (i) e(G) ∈ K0(MHS); (ii) e(Φm,δ) ∈ K Sm 0 (MHS) for all δ ≤ d and m ≤(n if δ < d n − 2 if δ = d ; (iii) e(M0,m(G, δ)) ∈ K Sm 0 (MHS) for all δ ≤ d and m ≤ n + d − δ. 19 Given V and r, [G] ∈ K0(Var) (and thus e(G)) is easily computable by means of the recursive equality [G] = [G(r, k)] = [G(r, k − 1)] + Lk−r[G(r − 1, k − 1)] , starting from the base cases [G(1, k)] = [G(k − 1, k)] = [Pk−1] = Li . k−1 Xi=0 By applying Corollary 3.10, the second equality of Corollary 3.16 becomes e(Φ) = D(e(M )) e(G) ◦(cid:0)s1 + e(Φ)(cid:1) . (7) Note that (7) yields a recursive algorithm for computing e(Φm,δ). In particular, it shows that e(Φm,δ) is determined by e(M0,i(G, ǫ)) for ǫ ≤ δ and i ≤ m+δ−ǫ+1. Therefore, the elements (ii) above can be obtained from the elements (iii). In conclusion, we see that the computation of e(M 0,n(G, d)) recursively boils down to the calculation of e(M0,m(G, δ)). 3.3 The HG-characteristic of the open stratum Motivated by the previous results, we shall now face the problem of comput- ing the HG-characteristic of the open locus M0,n(G, d) ⊆ M 0,n(G, d), which parametrizes those maps whose domain curve is isomorphic to P1. Let Mord(P1, G) be the variety that parametrizes morphisms from P1 to G of class d, and let F(P1, n) :=(cid:8)(x1, . . . , xn) ∈(cid:0)P1(cid:1)n xi 6= xj ∀ i 6= j(cid:9) be the configuration space of n distinct points in P1. The symmetric group Sn acts on F(P1, n) by permuting the n-tuples of points. If n + 3d ≥ 3, then there is a canonical Sn-equivariant isomorphism M0,n(G, d) ∼=(cid:0)Mord(P1, G) × F(P1, n)(cid:1)/ Aut(P1) , where Aut(P1) = PGL(2, C) acts on both Mord(P1, G) and F(P1, n) via its action on P1. In the remaining cases, namely for d = 0 and n ≤ 2, the stability condition implies that M0,n(G, 0) = ∅. Let F(P1) be the S-variety defined as F(P1)(n) := F(P1, n). The class of 0 (Var) can be expressed via the Exp and Log functions of [GP06, §3]. F(P1) in K S (8) Proposition 3.17 ([GP06, Thm. 3.2]). The equality holds in K S 0 (Var). [F(P1)] = 1 + Exp(cid:0)[P1] Log(s1)(cid:1) The corresponding formula for e(F(P1)) ∈ K S 0 (MHS) can be simplified using the properties of K S 0 (MHS). 20 Corollary 3.18 ([GP06, §5]). Let µ denote the Möbius function. Then the equality ∞ Yn=1 e(F(P1)) = (1 + p1) holds in K S 0 (MHS). (1 + pn)(1/n) Pkn µ(n/k)Lk The class of Aut(P1) in K0(Var) and its HG-characteristic are also easily computable. Indeed, we have [GL(2, C)] = L(L − 1)2 + L2(L − 1)2, therefore [Aut(P1)] = [PGL(2, C)] = [GL(2, C)] [C∗] = L3 − L and e(Aut(P1)) = L3 − L . However, since the quotient morphism Mord(P1, G) × F(P1, n) → M0,n(G, d) is not a locally trivial fibration in the Zariski topology, Proposition 2.4 cannot be used to express [M0,n(G, d)] ∈ K Sn 0 (Var) in terms of [F(P1, n)] and [Aut(P1)]. This problem is overcome by passing to K Sn 0 (MHS) via the HG-characteristic. Theorem 3.19 ([GP06, Thm. 5.4]). Let X be an S-variety, and let G be a connected algebraic group which acts on each X(n). Assume that the actions of Sn and G on each X(n) commute. If the action of G has finite stabilizers, then e(X) = e(G) e(X/G) in K S 0 (MHS). Theorem 3.19 implies that the equality e(Mord(P1, G)) e(F(P1, n)) = e(Aut(P1)) e(M0,n(G, d)) holds in K Sn 0 (MHS) whenever n + 3d ≥ 3. We write e(M0,n(G, d)) = e(Mord(P1, G)) e(F(P1, n)) e(Aut(P1)) , (9) meaning that e(M0,n(G, d)) is the unique element of the form e(X) such that e(Mord(P1, G)) e(F(P1, n)) = e(Aut(P1)) e(X). Equation (9) shows that in order to compute e(M0,n(G, d)) the only missing part is the computation of e(Mord(P1, G)) ∈ K0(MHS). This computation is dealt with in the next section. 4 Spaces of morphisms from P1 to Grassmannians The aim of this section is to show how [Mord(P1, G)] ∈ K0(Var) (and thus also e(Mord(P1, G)) ∈ K0(MHS)) can be explicitly determined. Throughout the section, we work over a fixed algebraically closed field k of characteristic 0. Schemes and morphisms between them are tacitly assumed to be over k; in particular, the product is understood to be the fiber product over k. Moreover, P1 = P1 k, V is a fixed k-vector space of dimension k, and G is the Grassmannian G(r, V )/k of r-dimensional quotients of V . 21 4.1 Quot compactification and its decomposition As shown in [Nit05], Mord(P1, G) is naturally realized as an open subscheme of the Hilbert scheme of P1 × G, by associating to each morphism f : P1 → G its graph Γf ⊆ P1 × G. A different compactification is provided by the Quot scheme Qd := Quot(t+1)r+d V ⊗OP1 /P1/k , which parametrizes equivalence classes of coherent quotients of V ⊗ OP1 of rank r and degree d, i.e., with Hilbert polynomial equal to (t + 1)r + d ∈ k[t]. Inside Qd, Mord(P1, G) is the open locus Qd corresponding to locally free quotients of V ⊗ OP1. The compactification Qd was studied in [Str87]. In particular, recall the following result. Theorem 4.1 ([Str87, Thm. 2.1]). The Quot scheme Qd is an irreducible, rational, nonsingular, projective variety of dimension kd + r(k − r). The subvariety Qd ⊆ Qd is the open dense stratum of a certain locally closed decomposition of Qd, which we shall now study. Definition 4.2. Let X be an algebraic scheme. A locally closed decomposition of X is a morphism f : Y → X of algebraic schemes that satisfies the following conditions: • the restriction of f to each connected component of Y is a locally closed immersion; • f is bijective on closed points. The starting point in constructing this decomposition of Qd is the following observation. Since any coherent sheaf F on P1 splits as the direct sum of its torsion subsheaf T (F ) and of the locally free sheaf F /T (F ), the geometric points of Qd parametrizing quotients that are not locally free correspond to morphisms P1 → G of lower degree, by considering only the locally free part of the quotients. This fact suggests the possibility of finding a decomposition of Qd such that each of its members correspond to morphisms P1 → G of some fixed degree δ ≤ d. Motivated by this observation, we consider the schemes Qδ = Quot(t+1)r+δ V ⊗OP1 /P1/k for any δ ∈ Z such that 0 ≤ δ ≤ d. Let [πδ : V ⊗ OP1×Qδ quotient on P1 × Qδ, let Eδ = ker(πδ), and let ιδ : Eδ → V ⊗ OP1×Qδ inclusion. Then we have the short exact sequence of coherent OP1×Qδ → Fδ] be the universal be the -modules 0 → Eδ ιδ−→ V ⊗ OP1×Qδ πδ−→ Fδ → 0 , (10) whose exactness is preserved by any base change T → Qδ. Let us also introduce the relative Quot schemes Rδ := Quotd−δ Eδ /P1×Qδ/Qδ over Qδ. 22 Proposition 4.3. The Quot scheme Rδ is projective and smooth over Qδ. In particular, Rδ is nonsingular. Proof. Since the projection P1 × Qδ → Qδ is a projective morphism, Rδ is a projective Qδ-scheme (see [Nit05]). If y ∈ Qδ is a closed point, let 0 → K → (Eδ)y → G be the corresponding short exact sequence of OP1-modules. Then Ext1(K, G) ∼= H1(P1, K∨ ⊗ G) = 0 , because G has 0-dimensional support. Therefore, from [Leh98] it follows that Rδ is a smooth Qδ-scheme. Finally, since Qδ is a smooth k-scheme, Rδ is smooth over k and thus non- singular, because k is a perfect field. There is a Cartesian diagram P1 × Rδ Rδ f ′ δ fδ P1 × Qδ P1 , Qδ Spec(k) where fδ : Rδ → Qδ is the structure morphism and f ′ δ = idP1 ×fδ. Let 0 → (Eδ)Rδ (ιδ )Rδ−−−−→ V ⊗ OP1×Rδ (πδ)Rδ−−−−→ (Fδ)Rδ → 0 be the pullback of (10) by f ′ quotient on P1 × Rδ, and let Hδ be the cokernel of ker(ρδ) ֒→ V ⊗ OP1×Rδ we obtain the following commutative diagram of coherent OP1×Rδ exact rows and columns: → Gδ] be the universal . Then -modules, with δ. Moreover, let [ρδ : (Eδ)Rδ 0 0 0 0 Kδ := ker(ρδ) (Eδ)Rδ ρδ Kδ 0 (ιδ)Rδ V ⊗ OP1×Rδ (πδ)Rδ (Fδ)Rδ ∼= 0 Gδ Hδ 0 0 . (11) Hδ/Gδ 0 0 0 The existence of the dotted arrows follows from the universal properties of ker- nels and cokernels. The exactness of all rows and columns is a consequence of the Four lemma and of the Nine lemma. Since both Gδ and Hδ/Gδ are flat over Rδ, Hδ is Rδ-flat. Thus, for any x ∈ Rδ the sequence of OP1 x-modules 0 → (Gδ)x → (Hδ)x → (Hδ/Gδ)x → 0 23 is exact. Since (Hδ/Gδ)x ∼= (Fδ)fδ (x), it follows that the Hilbert polynomial of (Hδ)x is d − δ + (t + 1)r + δ = (t + 1)r + d ∈ k[t] . Therefore, by the universal property of Qd, there exists a unique morphism gδ : Rδ → Qd such that the quotient V ⊗ OP×Rδ → Hδ is equivalent to the pullback of πd : V ⊗ OP1×Qd Remark 4.4. The diagram (11) is stable under any base change T → Rδ, because all sheaves appearing in it are flat over Rδ. In particular, for any x ∈ Rδ, the fiber (Gδ)x is (isomorphic to) a 0-dimensional subsheaf of (Hδ)x, so that there is an injective morphism from (Gδ)x into the torsion subsheaf T ((Hδ)x) of (Hδ)x. As a consequence, the length of T ((Hδ)x) is at least d − δ. → Fd by idP1 ×gδ. Let Qδ be the open locus in Qδ corresponding to locally free quotients of V ⊗ OP1; Qδ is the locus corresponding to Morδ(P1, G) inside Qδ. Let Rδ be the preimage f −1 (Qδ), with its open subscheme structure. δ Proposition 4.5. For each 0 ≤ δ ≤ d, the set-theoretic image gδ(Rδ) is the constructible subset of Qd whose closed points are the elements of (cid:8)y ∈ Qd(k) length T ((Fd)y) = d − δ(cid:9) . In particular, there is a set-theoretic decomposition Qd =F0≤δ≤d gδ(Rδ). Proof. For any closed point y of gδ(Rδ), let x ∈ Rδ(k) such that gδ(x) = y. Let us consider the pullback of (11) to the fiber P1 x: 0 0 0 (Kδ)x (Eδ)fδ (x) (Gδ)x (Kδ)x V ⊗ OP1 (Hδ)x 0 0 0 . 0 0 (Fδ)fδ (x) ∼= (Hδ/Gδ)x 0 0 0 By Remark 4.4, (Gδ)x is a subsheaf of T ((Hδ)x). Moreover, since x ∈ Rδ, the quotient sheaf (Hδ)x/(Gδ)x ∼= (Hδ/Gδ)x ∼= (Fδ)fδ (x) is locally free. As a consequence, (Gδ)x actually coincides with T ((Hδ)x). It follows that the length of T ((Hδ)x) is exactly d − δ. By construction of the morphism gδ, we have (Hδ)x ∼= (Fd)y, therefore the length of T ((Fd)y) is d − δ. Conversely, let y ∈ Qd be a closed point such that T = T ((Fd)y) has length d − δ. Then we have a quotient V ⊗ OP1 → (Fd)y → (Fd)y/T , 24 where (Fd)y/T is a locally free OP1-module of rank r and degree δ. Let z ∈ Qδ be the closed point corresponding to this quotient. There is a diagram 0 0 0 T (Eδ)z ρ 0 0 ker(ρ) ∼= (Ed)y V ⊗ OP1 (Fd)y 0 (Fδ)z ∼= (Fd)y/T 0 0 0 0 0 which commutes and has exact rows and columns; this is a consequence of the Four lemma and of the Nine lemma. In particular, T is a 0-dimensional quotient of (Eδ)z of length d − δ. Therefore, ρ : (Eδ)z → T corresponds to a closed point x ∈ Rδ such that fδ(x) = z. Moreover, we have gδ(x) = y, hence y ∈ gδ(Rδ), as claimed. Note that for any y ∈ Qd, the length of T = T ((Fd)y) cannot be greater than d. If it was, then from the quotient V ⊗ OP1 → (Fd)y/T we would get an invertible sheaf on P1 with negative degree and a non-zero section. Thus, every closed point y ∈ Qd lies in some gδ(Rδ). y By Chevalley's Theorem, each gδ(Rδ) is a constructible subset of Qd. Since k is algebraically closed and Qd is of finite type over k, the constructible subsets of Qd are uniquely determined by their intersection with Qd(k). Therefore, from the above discussion we deduce that Qd = Sδ gδ(Rδ) and gδ(Rδ) ∩ gǫ(Rǫ) = ∅ whenever δ 6= ǫ. We thus get a set-theoretic decomposition Qd =Fδ gδ(Rδ). For every 0 ≤ δ ≤ d, define Uδ := Qd \ Gǫ<δ gǫ(Rǫ) . In particular, U0 = Qd. Let us recursively construct morphisms jδ : Rδ → Uδ as follows. For δ = 0, we have R0 = R0. Indeed, if x ∈ R0(k) then the subsheaf (G0)x ⊆ T ((H0)x) has maximal length d, hence it coincides with T ((H0)x). ∼= (H0)x/(G0)x is locally free on P1, i.e., x ∈ R0(k). Let Therefore, (F0)f0(x) j0 := g0 : R0 → U0. Since the morphisms gδ are projective, the image g0(R0) is closed in Qd. Thus, U1 = Qd \ g0(R0) is an open subvariety of Qd. Now, for δ > 0, assume that we have already defined jδ−1 and proved that Uδ is an open subvariety of Qd. Since gδ(Rδ) ⊆ Uδ, gδRδ : Rδ → Qd factors through a morphism jδ : Rδ → Uδ. 25 Lemma 4.6. The commutative diagram Rδ jδ Uδ Rδ gδ Qd is a Cartesian diagram. As a consequence, jδ : Rδ → Uδ is projective. Proof. Let u T v Uδ Rδ gδ Qd be a commutative diagram of k-schemes of finite type. Let us consider the pull- back of (11) by u′ = idP1 ×u : P1 × T → P1 × Rδ. For any closed point t ∈ T , T ((u′∗Hδ)t) contains the subsheaf (u′∗Gδ)t, which has length d−δ. On the other hand, since gδ ◦u factors through Uδ, the length of T ((u′∗Hδ)t) cannot be greater than d−δ, thus it is exactly d−δ. Therefore, T ((u′∗Hδ)t) coincides with its sub- sheaf (u′∗Gδ)t. As a consequence, the sheaf (u′∗(Hδ/Gδ))t ∼= (u′∗Hδ)t/(u′∗Gδ)t is locally free. Now, the set {t ∈ T (u′∗(Hδ/Gδ))t is locally free} is an open subset of T , because u′∗(Hδ/Gδ) is T -flat. Since it contains all closed points of T , it is δ ◦ u′)∗Fδ)t ∼= (u′∗(Hδ/Gδ))t actually equal to T . Therefore, for any t ∈ T , ((f ′ is locally free. It follows that u : T → Rδ factors through a unique morphism h : T → Rδ. Since the immersion Uδ ֒→ Qd is a monomorphism, we also have jδ ◦ h = v. By Lemma 4.6, the image jδ(Rδ) is closed in Uδ. Thus, the complement Uδ \ jδ(Rδ) is an open subvariety of Qd, because Uδ ⊆ Qd is open. Note that we exactly have Uδ+1 = Uδ \ jδ(Rδ). Thus, we can proceed recursively and define jδ : Rδ → Uδ for all 0 ≤ δ ≤ d. The morphisms jδ determine the desired decomposition of Qd. Lemma 4.7. For every 0 ≤ δ ≤ d, jδ : Rδ → Uδ is a monomorphism. Proof. By [Gro67, Prop. 17.2.6], jδ is a monomorphism if and only if it is radicial (equivalently, universally injective) and formally unramified. Since jδ : Rδ → Uδ is a morphism of finite type of algebraic k-schemes, it suffices to prove that jδ is injective on closed points and unramified. Let x1, x2 ∈ Rδ be closed points such that jδ(x1) = jδ(x2). For each i, xi, we get the following commutative diagram with exact pulling-back (11) to P1 26 rows and columns: 0 0 0 0 0 (Kδ)xi ((Eδ)Rδ )xi (Gδ)xi (Kδ)xi V ⊗ OP1 (Hδ)xi 0 ((Fδ)Rδ )xi ∼= (Hδ/Gδ)xi 0 . 0 0 0 0 ∼= (Hδ)x2 that commutes As jδ(x1) = jδ(x2), there is an isomorphism (Hδ)x1 with the quotients V ⊗OP1 → (Hδ)xi. Equivalently, we have (Kδ)x1 = (Kδ)x2 as subsheaves of OP1. The sheaves ((Fδ)Rδ )xi are locally free, because fδ(xi) ∈ Qδ. Therefore, each T ((Hδ)xi) coincides with its subsheaf (Gδ)xi. As a consequence, we obtain that ((Eδ)Rδ )x1 = ((Eδ)Rδ )x2 as subsheaves of V ⊗ OP1, because both of them are the saturation of the same subsheaf (Kδ)xi. Furthermore, ∼= (Gδ)x2 which commutes with the quotients there is an isomorphism (Gδ)x1 ((Eδ)Rδ )xi → (Gδ)xi, i.e., x1 = x2. Thus, jδ is injective on closed points. Let us now prove that jδ is unramified. Since fδRδ : Rδ → Qδ is smooth, there is a short exact sequence of locally free ORδ-modules 0 → (fδRδ )∗ΩQδ /k → ΩRδ/k → ΩRδ /Qδ → 0 . Moreover, we have the exact sequence of ORδ -modules g∗ δ (ΩUδ /k) → ΩRδ /k → ΩRδ/Uδ → 0 . By [Leh98], for any closed point x ∈ Rδ, we have (12) (13) ((fδRδ )∗ΩQδ /k) ⊗ κ(x) ∼= Hom(cid:0)((Eδ)Rδ )x, ((Fδ)Rδ )x(cid:1)∨ ΩRδ /Qδ ⊗ κ(x) ∼= Hom((Kδ)x, (Gδ)x)∨ , δ (ΩUδ /k) ⊗ κ(x) ∼= Hom((Kδ)x, (Hδ)x)∨ . g∗ , The vector space Hom((Gδ)x, (Hδ)/Gδ)x) vanishes, because (Gδ)x is a torsion sheaf and (Hδ)/Gδ)x is torsion-free. Therefore, by means of the long exact Ext sequences associated to the pullback of (11) to P1 x, (12) implies that ΩRδ /k ⊗ κ(x) ∼= coker(α)∨ , where α = − ◦ (ρδ)x : Hom((Gδ)x, (Hδ)x) → Hom(((Eδ)Rδ )x, (Hδ)x). Moreover, the pullback of the morphism g∗ x is the dual of the canonical injection coker(α) ֒→ Hom((Kδ)x, (Hδ)x), hence it is surjective. As a consequence, we obtain the vanishing of ΩRδ /Uδ ⊗ κ(x). Since ΩRδ/Uδ ⊗ κ(x) = 0 for each closed point x ∈ Rδ and Rδ is of finite type over k, we have ΩRδ /Uδ = 0, i.e., jδ is unramified. δ (ΩUδ /k) → ΩRδ /k of (13) to P1 27 Theorem 4.8. The morphism j := a0≤δ≤d(cid:16)Rδ jδ−→ Uδ ֒→ Qd(cid:17) : a0≤δ≤d Rδ → Qd is a locally closed decomposition of Qd. Proof. By Proposition 4.5, since jδ(Rδ) = gδ(Rδ), j is bijective on closed points. From Lemma 4.6 and Lemma 4.7 it follows that jδ : Rδ → Uδ is a proper monomorphism, hence a closed immersion by [Gro67, Cor. 18.12.6]. Therefore, jδ−→ Uδ ֒→ Qd is a locally closed immersion and j is a each composition Rδ locally closed decomposition of Qd. For the purposes of this work, the importance of Theorem 4.8 lies in the subsequent corollary, which directly follows from Proposition 2.3. Corollary 4.9. The equality [Qd] = X0≤δ≤d [Rδ] holds in K0(Vark). 4.2 The classes of the strata in K0(Vark) Let us now study the relation between the classes [Rδ] and [Qδ] in K0(Vark). Let s := k − r. By Grothendieck's theorem on vector bundles on P1, for every y ∈ Qδ there is an isomorphism (Eδ)y ∼= s Mi=0 OP1 y (ai(y)) , where a1(y) ≤ · · · ≤ as(y) is a uniquely determined sequence of integers such that a1(y) + · · · + as(y) = −δ. Note that each ai(y) is ≤ 0, because (Eδ)y is a subsheaf of V ⊗ OP1 y . Let Aδ = {a ∈ Zs a1 ≤ · · · ≤ as ≤ 0 , a = −δ}. By [Sha77], for any a ∈ Aδ the locus Qa δ := {y ∈ Qδ (a1(y), . . . , as(y)) = a} is locally closed in Qδ, because Eδ is flat over Qδ. Let Ra f −1 δ δ ), with its reduced induced scheme structure. (Qa The main theorem of §4.2 is the following. δ be the preimage Theorem 4.10. For every a ∈ Aδ, let E a δ := s Mi=0 OP1(ai) . Then the restriction fδRa topology, with fiber Quotd−δ E a : Ra δ /P1/k. δ δ → Qa δ is a locally trivial fibration in the Zariski 28 The proof of Theorem 4.10 essentially relies on the following result. Proposition 4.11. Let T be a smooth scheme, and let E be a coherent locally free sheaf of rank s on P1 × T , flat over T . Assume that for every t ∈ T , there is an isomorphism s Et ∼= Mi=1 T , E is isomorphic to the pullback of Ls P1 × T → P1. OP1 t (ai) , where the integers a1 ≤ · · · ≤ as are independent of t. Then, Zariski locally on i=1 OP1 (ai) by the projection morphism Proof. Without loss of generality, we may assume that T is an integral scheme. Let p : P1 × T → P1 and q : P1 × T → T be the two projections. If H is a sheaf on P1 × T , we write H(a) for H ⊗ p∗OP1 (a). We will proceed by induction on s. If s = 1, let us consider the coherent OT -module F = q∗(E(−a1)). By the results on cohomology and base change, F is a locally free sheaf of rank h0(P1, OP1) = 1, which commutes with any base change. The counit of the adjunction q∗ ⊣ q∗ gives an isomorphism of invertible sheaves q∗F ∼−→ E(−a1). Therefore, we can cover T with open subsets over which q∗F trivializes, so that we get an isomorphism p∗OP1(a1) ∼−→ E. If s > 1, let F = q∗(E(−as)). By Grauert's theorem on base change, F is a locally free OT -module of rank s′ = #{i ai = as}. As above, the counit of the adjunction q∗ ⊣ q∗ gives a morphism of locally free sheaves q∗F → E(−as). On each fiber P1 → Et(−as), which is an injection of vector bundles. Therefore, we have an injection of locally free sheaves (q∗F )(as) → E with locally free quotient Q: t over t ∈ T , this is the evaluation H 0(P1 t , Et(−as))⊗OP1 t 0 → (q∗F )(as) → E → Q → 0 . (14) On each fiber P1 t , we have (q∗F )(as)t ∼= Miai=as OP1 t (ai) , Qt ∼= Miai6=as OP1 t (ai) . By the induction hypothesis, since the rank of Q is strictly less than s, we can cover T with open subsets U , such that QU is isomorphic to the pullback of OP1(ai) by the projection P1 × U → U . By shrinking T if necessary, we may thus assume that this happens globally on T . In order to prove that the sequence (14) splits (Zariski locally), consider Liai6=as Ext1(Q, (q∗F )(as)) ∼= H 1(cid:0)P1 × T, Q∨ ⊗ (q∗F )(as)(cid:1) . Let us apply the Leray spectral sequence H i(T, Rjq∗(−)) ⇒ H i+j(P1 × T, −) to the sheaf Q∨ ⊗ (q∗F )(as). Since dim(P1) = 1, Riq∗(Q∨ ⊗ (q∗F )(as)) = 0 for every i > 1. For i = 1, using the projection formula, we obtain R1q∗(Q∨ ⊗ (q∗F )(as)) ∼= F ⊗ R1q∗(Q∨(as)) . The sheaf Q∨(as) is the pullback of Ljaj 6=as OP1(as − aj), whose H 1 vanishes because as > aj, hence R1q∗(Q∨(as)) = 0. Finally, the last contribution to H 1(P1 × T, Q∨ ⊗ (q∗F )(as)) coming from the Leray spectral sequence is H 1(cid:0)T, q∗(Q∨ ⊗ (q∗F )(as))(cid:1) ∼= H 1(T, F ⊗ q∗(Q∨(as)) . 29 If U is an affine open subset of T , then H 1(U, F ⊗ q∗(Q∨(as)) = 0. It follows that thus the sequence (14) splits over U . By considering an affine open cover {Uλ}λ such that F is trivial on each Uλ, we get the desired isomorphism. H 1(cid:0)P1 × U, Q∨ ⊗ (q∗F )(as)(cid:1) = 0 , Proof of Theorem 4.10. By Proposition 4.11, there exists a Zariski open cover δ ) for all λ, where p : P1 × Uλ → P1 is {Uλ}λ of Qa the projection. Therefore, by the base change property of the Quot scheme, the preimage f −1 δ such that EδP1×Uλ δ is naturally isomorphic to (Uλ) ⊆ Ra ∼= p∗(E a δ Quotd−δ Eδ/P1×Qa δ /Qa δ ×Qa δ Uλ ∼= Quotd−δ p∗(E a δ )/P1×Uλ/Uλ ∼= Quotd−δ E a δ /P1/k ×Uλ . Thus fδRa δ : Ra δ → Qa δ is a locally trivial fibration. As a consequence of Theorem 4.10, we get the following result. Corollary 4.12. For each 0 ≤ δ ≤ d and each a ∈ Aδ, the equality [Rδ] = Xa∈Aδ holds in K0(Vark). [Qa δ ]hQuotd−δ δ /P1/ki E a Proof. The subschemes Ra Rδ. Therefore, we have δ , for a ∈ Aδ, give a locally closed decomposition of By Theorem 4.10, each fδRa : Ra Zariski topology, with fiber Quotd−δ E a that δ δ ] . [Ra [Rδ] = Xa∈Aδ δ is a locally trivial fibration in the δ → Qa δ /P1/k. Thus, from Proposition 2.4 it follows δ ]hQuotd−δ δ /P1/ki . E a [Ra δ ] = [Qa Combining the two equalities, we obtain the desired result. The problem of computing the class of Quotd−δ E a δ /P1/k in K0(Vark) was solved in [BFP19], where it was proved that it is independent of a ∈ Aδ. Proposition 4.13. For any integer 0 ≤ δ ≤ d, the equality (1 − Lm1+1) · · · (1 − Lms+1) (1 − L)s Ldm E a m=d−δ hQuotd−δ δ /P1/ki = Xm∈(Z≥0)s holds in K0(Vark), where dm =Ps Xj=0 Proof. Since [Pm] = m i=1(i − 1)mi. Lj = 1 − Lm+1 1 − L , 30 the result follows from the equality hQuotd−δ δ /P1/ki = Xm∈(Z≥0)s E a m=d−δ [Pm1] · · · [Pms] Ldm . of [BFP19, Prop. 4.5]. The class appearing in Proposition 4.13, which depends only on s = k − r and d − δ, will be denoted by Ωd−δ. 4.3 The class of Mord(P1, G(r, V )) in K0(Vark) By means of the previous results, we can now describe a method for computing the class [Qd] = [Mord(P1, G)] ∈ K0(Vark). Theorem 4.14. The equality ∞ [Qd]qd = ∞ Xd=0 Xd=0 [Qd]qd! ∞ Xd=0 Ωdqd!−1 holds in K0(Vark)[[q]]. Proof. Applying Corollary 4.12 and Proposition 4.13, we get [Rδ] = Xa∈Aδ [Qa δ ]hQuotd−δ E a δ /P1/ki = Ωd−δ Xa∈Aδ [Qa δ ]! = Ωd−δ[Qδ] . By Corollary 4.9, we thus have d d therefore [Rδ] = Ωd−δ[Qδ] , [Qd] = Xδ=0 [Qd]qd = ∞ Xd=0 Xd=0 ∞ Xδ=0 [Qd]qd! ∞ Xd=0 Ωdqd! . (15) Since Ω0 = 1, the last power series in (15) is invertible, thus (15) implies the theorem. Theorem 4.14 provides a recursive formula for [Qd] in terms of [Qδ] and Ωδ for δ ≤ d: ℧0 := 1 , ℧j := − ℧iΩj−i , [Qd] = j−1 Xi=0 d Xj=0 ℧j[Qd−j] . (16) The elements Ωδ have already been determined in Proposition 4.13. Thus, in order for (16) to be effective for computing [Qd], we only need to calculate [Qδ]. 31 By [Str87, §3], Qδ has a Białynicki-Birula decomposition determined by a torus action. Since each subvariety in this locally closed decomposition is isomorphic to some affine space Ai, we have dim(Qδ) kδ+r(k−r) [Qδ] = Xi=0 mδ,i[Ai] = Xi=0 mδ,iLi , (17) where mδ,i is the number of i-dimensional cells. A formula for the numbers mδ,i is provided in [Str87]. Let G ⊆ Zs×Zs+1×Zs be the subset whose elements are triples (a = (a1, . . . , as), b = (b0, . . . , bs), c = (c1, . . . , cs)) such that b0 = 0 ≤ a1 ≤ b1 ≤ a2 ≤ · · · ≤ bs−1 ≤ as ≤ δ = bs , 0 ≤ c1 ≤ · · · ≤ cs ≤ r . Then the following result holds. Proposition 4.15 ([Str87, Thm. 5.4]). For all 0 ≤ i ≤ kδ + r(k − r), mδ,i is equal to the number of elements (a, b, c) ∈ G such that the equality (aj + cj(1 + bj − bj−1)) = i s Xj=1 holds. By means of this proposition, (17) becomes an explicit formula for the class [Qδ] ∈ K0(Vark). This completely solves the problem of calculating [Qd]: after computing [Qδ] for all δ ≤ d via (17), [Qd] can be determined via (16) and Proposition 4.13. 5 The Betti numbers of M 0,n(G(r, V ), d) The previous results yield an algorithm to compute the HG-characteristic of M 0,n(G, d), which in turn determines its Hodge and Betti numbers. Let us outline how this algorithm works. (i) Apply the results of §4.3 to compute [Qδ] and e(Qδ) for all δ ≤ d. (ii) Calculate e(M0,m(G, δ)) for all δ ≤ d and m ≤ n + d − δ, by means of (9). (iii) Use the recursive procedure explained in §3.2 to get e(M 0,n(G, d)). Remark 5.1. As a consequence of the above procedure, the cohomology group H i(M 0,n(G, d)) vanishes for i odd, whereas its class in K0(MHS) is a multiple of Li/2 for i even. This agrees with the results of [Opr06]. In the remainder of the section, we examine some examples, in order to show the effectiveness of our algorithm in practice. 32 5.1 The HG-characteristic of M 0,0(G(2, 4), 2) Let us consider the Grassmannian G = G(2, 4), whose class in K0(Var) is [G(2, 4)] = (L2 + 1)(L2 + L + 1) . First, we compute Ωi for all 0 ≤ i ≤ 2 via Proposition 4.13: Ω0 = 1 , Ω1 = (L + 1)2 , Ω2 = L4 + 2L3 + 4L2 + 2L + 1 . Then we also have the following equalities: ℧0 = 1 , ℧1 = −(L + 1)2 , ℧2 = 2L(L2 + L + 1) . The classes [Qδ] for all 0 ≤ δ ≤ 2 are determined by Proposition 4.15: [Q0] = (L2 + 1)(L2 + L + 1) , [Q1] = (L + 1)2(L2 + 1)(L2 − L + 1)(L2 + L + 1) , [Q2] = (L2 + 1)(L2 + L + 1)(L4 + 1)(L4 + L3 + L2 + L + 1) . Combining the previous equalities, we get [Q0] = (L2 + 1)(L2 + L + 1) , [Q1] = L(L − 1)(L + 1)2(L2 + 1)(L2 + L + 1) , [Q2] = L3(L − 1)(L + 1)(L2 + 1)(L2 + L + 1)(L3 + L2 + L − 1) . The knowledge of e(Qd) allows us to calculate e(M0,n(G, d)). By (9) and Corollary 3.18, for all d > 0 we have e(M0,0(G, d)) = e(Qd) L3 − L and e(M0,1(G, d)) = e(Qd) L2 − L s1 , therefore e(M0,m(G, 0)) = e(M 0,m(G, 0)) = 0 e(M0,0(G, 1)) = e(M 0,0(G, 1)) = (L + 1)(L2 + 1)(L2 + L + 1) , e(M0,1(G, 1)) = e(M 0,1(G, 1)) = (L + 1)2(L2 + 1)(L2 + L + 1)s1 , e(M0,0(G, 2)) = L2(L2 + 1)(L2 + L + 1)(L3 + L2 + L − 1) . for all m ≤ 2 , By (7), we also have e(Φ0,0) = 0 and e(Φ0,1) = D(e(M0,1(G, 1))) e(G) = (L + 1)2 . Now, we can conclude our computation by means of Corollary 3.16. Since pi ◦ − is an algebra homomorphism, Corollary 3.16 and the equations s1 = p1 and 2s2 = p2 + p2 1 imply that e(M 0,0(G, 2)) = e(M0,0(G, 2)) + e(M0,1(G, 1)) s1 e(Φ0,1) + e(G) p2 ◦ e(Φ0,1) − e(Φ0,1)2 2 . (18) 33 Thus, the above calculations yield e(M 0,0(G, 2)) = L9 + 3L8 + 7L7 + 11L6 + 14L5 + 14L4 + 11L3 + 7L2 + 3L + 1 . As a consequence, the E-polynomial E(M 0,0(G, 2); t, u) is t9u9 + 3t8u8 + 7t7u7 + 11t6u6 + 14t5u5 + 14t4u4 + 11t3u3 + 7t2u2 + 3tu + 1 , in agreement with [Ló14, Thm. 3.1]. 5.2 The HG-characteristic of M 0,1(G(2, 4), 2) By (9) and Corollary 3.18, we have e(M0,3(G, 0)) = (L2 + 1)(L2 + L + 1)s3 , e(M0,2(G, 1)) = L(L + 1)(L2 + 1)(L2 + L + 1)(cid:0)(L − 1)s2 + s2 1(cid:1) , e(M0,1(G, 2)) = L2(L + 1)(L2 + 1)(L2 + L + 1)(L3 + L2 + L − 1)s1 . As a consequence, we get e(Φ1,1) = D(e(M0,2(G, 1))) e(G) + D(e(M0,3(G, 0))) e(G) s2 e(Φ0,1)s1 = (L + 1)3s1 . Using these equalities and those of §5.1, we can now compute e(M 0,1(G, 2)). Corollary 3.16 implies that e(M 0,1(G, 2)) is equal to e(M0,1(G, 2)) + e(M0,1(G, 1)) s1 e(Φ1,1) + {e(M0,2(G, 1))}s2 e(Φ0,1)s1 + 2{e(M0,2(G, 1))}s2 1 e(Φ0,1)s1 + e(M0,3(G, 0)) 2s3 e(Φ0,1)2s1 + e(M0,3(G, 0)) 2s3 (cid:0)p2 ◦ e(Φ0,1)(cid:1)s1 − e(G) e(Φ0,1) e(Φ1,1) , where {e(M0,2(G, 1))}s2 (resp. {e(M0,2(G, 1))}s2 ) is the coefficient of s2 (resp. s2 1) in e(M0,2(G, 1)). Therefore, the HG-characteristic of M 0,1(G, 2) is equal to 1 (L10 + 4L9 + 12L8 + 22L7 + 33L6 + 36L5 + 33L4 + 22L3 + 12L2 + 4L + 1)s1 . In particular, its E-polynomial E(M 0,1(G, 2); t, u) is t10u10 + 4t9u9 + 12t8u8 + 22t7u7 + 33t6u6 + 36t5u5 + 33t4u4 + 22t3u3 + 12t2u2 + 4tu + 1 . References [ACG11] E. Arbarello, M. Cornalba, and P. A. Griffiths, Geometry of algebraic curves. Vol. II, Grund. Math. Wiss., vol. 268, Springer, Heidelberg, 2011. [Bag19] M. Bagnarol, On the cohomology of moduli to Grassmannians, Ph.D. maps http://hdl.handle.net/20.500.11767/103198. stable thesis, SISSA, Trieste, 2019, spaces of 34 [BFP19] M. Bagnarol, B. Fantechi, and F. Perroni, On the motive of Quot schemes of zero-dimensional quotients on a curve, arXiv e-prints (2019), arXiv:1907.00826. [Bit04] F. Bittner, The universal Euler characteristic for varieties of charac- teristic zero, Comp. Math. 140 (2004), no. 4, 1011 -- 1032. [BLL97] F. Bergeron, G. Labelle, and P. Leroux, Combinatorial species and tree-like structures, Encycl. Math. Appl., Cambridge Univ. Press, Cambridge, UK, 1997. [BM96] K. Behrend and Yu. Manin, Stacks of stable maps and Gromov-Witten invariants, Duke Math. J. 85 (1996), no. 1, 1 -- 60. [BO03] K. Behrend and A. O'Halloran, On the cohomology of stable map spaces, Invent. Math. 154 (2003), no. 2, 385 -- 450. [Bor18] L. Borisov, The class of the affine line is a zero divisor in the grothendieck ring, J. Alg. Geom. 27 (2018), 203 -- 209. [DK86] V. I. Danilov and A. G. Khovanskiı, Newton polyhedra and an algo- rithm for calculating Hodge-Deligne numbers, Izv. Akad. Nauk SSSR Ser. Mat. 50 (1986), no. 5, 925 -- 945. [FP97] W. Fulton and R. Pandharipande, Notes on stable maps and quan- tum cohomology, Algebraic Geometry Santa Cruz 1995, Proc. Sym- pos. Pure Math., vol. 62, Amer. Math. Soc., Providence, RI, 1997, pp. 45 -- 96. [Get95] E. Getzler, Mixed Hodge structures of configuration spaces, arXiv e- prints (1995), alg -- geom/9510018. [GP06] E. Getzler and R. Pandharipande, The Betti numbers of M0,n(r, d), J. Alg. Geom. 15 (2006), no. 4, 709 -- 732. [Gro67] A. Grothendieck, Éléments de géométrie algébrique: IV. Étude locale des schémas et des morphismes de schémas, Quatrième partie, Publ. Math. IHÉS 32 (1967), 5 -- 361. [Joy81] A. Joyal, Une théorie combinatoire des séries formelles, Adv. Math. 42 (1981), no. 1, 1 -- 82. [Kel05] G. M. Kelly, On the operads of J.P. May, Repr. Theory Appl. Categ. 13 (2005), 1 -- 13. [Kon95] M. Kontsevich, Enumeration of rational curves via torus action, The Moduli Space of Curves (Texel Island, 1994), Progr. Math., vol. 129, Birkhäuser Boston, Cambridge, MA, 1995, pp. 335 -- 368. [Leh98] M. Lehn, On the cotangent sheaf of Quot-schemes, Internat. J. Math. 9 (1998), no. 4, 513 -- 522. [Ló14] A. López Martín, Poincaré polynomials of stable map spaces to Grass- mannians, Rend. Sem. Mat. Univ. Padova 131 (2014), 193 -- 208. 35 [Mac95] I. G. Macdonald, Symmetric functions and Hall polynomials, 2nd ed., Oxford Math. Mon., The Clarendon Press, Oxford Univ. Press, New York, 1995. [MM07] A. Mustaţa and M.A. Mustaţa, Intermediate moduli spaces of stable maps, Invent. Math. 167 (2007), no. 1, 47 -- 90. [Nit05] N. Nitsure, Construction of Hilbert and Quot schemes, Fundamental Algebraic Geometry, Math. Surveys Monogr., vol. 123, Amer. Math. Soc., Providence, RI, 2005, pp. 105 -- 137. [Opr06] D. Oprea, The tautological rings of the moduli spaces of stable maps to flag varieties, J. Alg. Geom. 15 (2006), no. 4, 623 -- 655. [PS08] C. Peters and J. H. M. Steenbrink, Mixed Hodge structures, Erg. der Math., vol. 52, Springer, Berlin, 2008. [Sha77] S. S. Shatz, The decomposition and specialization of algebraic families of vector bundles, Comp. Math. 35 (1977), no. 2, 163 -- 187. [Str87] S. A. Strømme, On parametrized rational curves in Grassmann vari- eties, Space Curves (Rocca di Papa, 1985), Lect. Notes Math., vol. 1266, Springer, Berlin, Heidelberg, 1987, pp. 251 -- 272. Massimo Bagnarol -- [email protected] 36
1305.7462
2
1305
2013-09-18T03:44:42
Likelihood Geometry
[ "math.AG", "math.CO", "math.ST", "math.ST" ]
We study the critical points of monomial functions over an algebraic subset of the probability simplex. The number of critical points on the Zariski closure is a topological invariant of that embedded projective variety, known as its maximum likelihood degree. We present an introduction to this theory and its statistical motivations. Many favorite objects from combinatorial algebraic geometry are featured: toric varieties, A-discriminants, hyperplane arrangements, Grassmannians, and determinantal varieties. Several new results are included, especially on the likelihood correspondence and its bidegree. These notes were written for the second author's lectures at the CIME-CIRM summer course on Combinatorial Algebraic Geometry at Levico Terme in June 2013.
math.AG
math
Likelihood Geometry June Huh and Bernd Sturmfels 3 1 0 2 p e S 8 1 ] G A . h t a m [ 2 v 2 6 4 7 . 5 0 3 1 : v i X r a Abstract We study the critical points of monomial functions over an algebraic subset of the probability simplex. The number of critical points on the Zariski closure is a topolog- ical invariant of that embedded pro jective variety, known as its maximum likelihood degree. We present an introduction to this theory and its statistical motivations. Many favorite ob jects from combinatorial algebraic geometry are featured: toric varieties, A- discriminants, hyperplane arrangements, Grassmannians, and determinantal varieties. Several new results are included, especially on the likelihood correspondence and its bidegree. This article represents the lectures given by the second author at the CIME- CIRM course on Combinatorial Algebraic Geometry at Levico Terme in June 2013. Introduction Maximum likelihood estimation (MLE) is a fundamental computational problem in statistics, and it has recently been studied with some success from the perspective of algebraic geometry. In these notes we give an introduction to the geometry behind MLE for algebraic statistical models for discrete data. As is customary in algebraic statistics [15], we shall identify such models with certain algebraic subvarieties of high-dimensional complex pro jective spaces. The article is organized into four sections. The first three sections correspond to the three lectures given at Levico Terme. The last section will contain proofs of new results. In Section 1, we start out with plane curves, and we explain how to identify the relevant punctured Riemann surfaces. We next present the definitions and basic results for likelihood geometry in Pn . Theorems 1.6 and 1.7 are concerned with the likelihood correspondence, the sheaf of differential 1-forms with logarithmic poles, and the topological Euler characteristic. The ML degree of generic complete intersections is given in Theorem 1.10. Theorem 1.15 shows that the likelihood fibration behaves well over strictly positive data. Examples of Grassmannians and Segre varieties are discussed in detail. Our treatment of linear spaces in Theorem 1.20 will appeal to readers interested in matroids and hyperplane arrangements. Section 2 begins leisurely, with the question Does watching soccer on TV cause hair loss? [34]. This leads us to conditional independence and low rank matrices. We study likelihood geometry of determinantal varieties, culminating in the duality theorem of Draisma and Rodriguez [14]. The ML degrees in Theorems 2.2 and 2.6 were computed using the software Bertini [6], underscoring the benefits of using numerical algebraic geometry for MLE. After a discussion of mixture models, highlighting the distinction between rank and nonnegative rank, we end Section 2 with a review of recent results in [1] on tensors of nonnegative rank 2. 1 Section 3 starts out with toric models [37, §1.22] and geometric programming [9, §4.5]. Theorem 3.2 identifies the ML degree of a toric variety with the Euler characteristic of the complement of a hypersurface in a torus. Theorem 3.7 furnishes the ML degree of a variety parametrized by generic polynomials. Theorem 3.10 characterizes varieties of ML degree 1 and it reveals a beautiful connection to the A-discriminant of [19]. We introduce the ML bidegree and the sectional ML degree of an arbitrary pro jective variety in Pn , and we explain how these two are related. Section 3 ends with a study of the operations of intersection, pro- jection, and restriction in likelihood geometry. This concerns the algebro-geometric meaning of the distinction between sampling zeros and structural zeros in statistical modeling. In Section 4 we offer precise definitions and technical explanations of more advanced concepts from algebraic geometry, including logarithmic differential forms, Chern-Schwartz- MacPherson classes, and schon very affine varieties. This enables us to present complete proofs of various results, both old and new, that are stated in the earlier sections. We close the introduction with a disclaimer regarding our overly ambitious title. There are many important topics in the statistical study of likelihood inference that should belong to “Likelihood Geometry” but are not covered in this article. Such topics include Watanabe’s theory of singular Bayesian integrals [45], differential geometry of likelihood in information geometry [5], and real algebraic geometry of Gaussian models [43]. We regret not being able to talk about these topics and many others. Our presentation here is restricted to the setting of [15, §2.2], namely statistical models for discrete data viewed as pro jective varieties in Pn . 1 First Lecture Let us begin our discussion with likelihood on algebraic curves in the complex pro jective plane P2 . We fix a system of homogeneous coordinates p0 , p1 , p2 on P2 . The set of real points ∆2 = (cid:8) (p0 , p1 , p2 ) ∈ R3 : p0 , p1 , p2 > 0 and p0 + p1 + p2 = 1 (cid:9). in P2 with sign(p0 ) = sign(p1 ) = sign(p2 ) is identified with the open triangle Given three positive integers u0 , u1 , u2 , the corresponding likelihood function is pu0 1 pu2 0 pu1 2 (cid:96)u0 ,u1 ,u2 (p0 , p1 , p2 ) = . (p0 + p1 + p2 )u0+u1+u2 H = (cid:8) (p0 : p1 : p2 ) ∈ P2 : p0p1p2 (p0 + p1 + p2 ) = 0 (cid:9) This defines a rational function on P2 , and it restricts to a regular function on P2\H, where is our arrangement of four distinguished lines. The likelihood function (cid:96)u0 ,u1 ,u2 is positive on the triangle ∆2 , it is zero on the boundary of ∆2 , and it attains its maximum at the point 1 ( p0 , p1 , p2 ) = u0 + u1 + u2 (cid:19) (cid:18) u0 The corresponding point ( p0 : p1 : p2 ) is the only critical point of the function (cid:96)u0 ,u1 ,u2 on the four-dimensional real manifold P2\H. To see this, we consider the logarithmic derivative − u0 + u1 + u2 − u0 + u1 + u2 − u0 + u1 + u2 u2 u1 . p0 + p1 + p2 p0 + p1 + p2 p0 + p1 + p2 p0 p1 p2 dlog((cid:96)u0 ,u1 ,u2 ) = (u0 , u1 , u2 ). (1.1) , , 2 We note that this equals (0, 0, 0) if and only if (p0 : p1 : p2 ) is the point ( p0 : p1 : p2 ) in (1.1). Let X be a smooth curve in P2 defined by a homogeneous polynomial f (p0 , p1 , p2 ). This curve plays the role of a statistical model, and our task is to maximize the likelihood function (cid:96)u0 ,u1 ,u2 over its set X ∩ ∆2 of positive real points. To compute that maximum algebraically, we examine the set of all critical points of (cid:96)u0 ,u1 ,u2 on the complex curve X \H. That set of critical points is the likelihood locus. Using Lagrange Multipliers from Calculus, we see that it consists of all points of X \H such that dlog((cid:96)u0 ,u1 ,u2 ) lies in the plane spanned by df and  = 0.  1 (1, 1, 1) in C3 . Thus, our task is to study the solutions in P2\H of the equations u0 p0 ∂ f ∂ p0 Suppose that X has degree d. Then, after clearing denominators, the second equation has degree d + 1. By B´ezout’s Theorem, we expect the likelihood locus to consist of d(d + 1) points in P2\H. This is indeed what happens when f is a generic polynomial of degree d. We define the maximum likelihood degree (or ML degree) of our curve X to be the cardi- nality of the likelihood locus for generic choices of u0 , u1 , u2 . Thus a general plane curve of degree d has ML degree d(d + 1). However, for special curves, the ML degree can be smaller. Theorem 1.1. Let X be a smooth curve of degree d in P2 , and a = #(X ∩ H) the number of its points on the distinguished arrangement. Then the ML degree of X equals d2 − 3d + a. f (p0 , p1 , p2 ) = 0 (1.2) 1 u2 p2 1 u1 p1 ∂ f ∂ p1 and det ∂ f ∂ p2 This is a very special case of Theorem 1.7 which identifies the ML degree with the signed Euler characteristic of X \H. For a general curve of degree d in P2 , we have a = 4d, and so d2 − 3d + a = d(d + 1) as predicted. However, the number a of points in X ∩ H can drop: Example 1.2. Consider the case d = 1 of lines. A generic line has ML degree 2. The line X = V (p0 + cp1 ) has ML degree 1 provided c (cid:54)∈ {0, 1}. The special line X = V (p0 + p1 ) has ML degree 0: (1.2) has no solutions on X \H unless u0 + u1 = 0. In the three cases, X \H is ♦ the Riemann sphere P1 with four, three, or two points removed. Example 1.3. Consider the case d = 2 of quadrics. A general quadric has ML degree 6. The (cid:18)2p0 (cid:19) Hardy-Weinberg curve, which plays a fundamental role in population genetics, is given by = 4p0p2 − p2 p1 f (p0 , p1 , p2 ) = det 1 . 2p2 p1 X ∩ H = (cid:8) (1 : 0 : 0), (0 : 0 : 1), (1 : −2 : 1) (cid:9). The curve has only three points on the distinguished arrangement: Hence the ML degree of the Hardy-Weinberg curve equals 1. This means that the maximum (cid:0) (2u0 + u1 )2 , 2(2u0 + u1 )(u1 + 2u2 ) , (u1 + 2u2 )2 (cid:1). likelihood estimate (MLE) is a rational function of the data. Explicitly, the MLE equals 3 1 4(u0 + u1 + u2 )2 ( p0 , p1 , p2 ) = (1.3) In applications, the Hardy-Weinberg curve arises via its parametric representation s2 p0 (s) = p1 (s) = 2s(1 − s) p2 (s) = (1 − s)2 (1.4) Here the parameter s is the probability that a biased coin lands on tails. If we toss that same biased coin twice, then the above formulas represent the following probabilities: p0 (s) = probability of 0 heads p1 (s) = probability of 1 head p2 (s) = probability of 2 heads Suppose now that the experiment of tossing the coin twice is repeated N times. We record the following counts, where N = u0 + u1 + u2 is the sample size of our repeated experiment: u0 = number of times 0 heads were observed u1 = number of times 1 head was observed u2 = number of times 2 heads were observed The MLE problem is to estimate the unknown parameter s by maximizing (cid:96)u0 ,u1 ,u2 = p0 (s)u0 p1 (s)u1 p2 (s)u2 = 2u1 s2u0+u1 (1 − s)u1+2u2 . The unique solution to this optimization problem is 2u0 + u1 Substituting this expression into (1.4) gives the estimator (cid:0)p0 (s), p1 (s), p2 (s)(cid:1) for the three s = . 2u0 + 2u1 + 2u2 ♦ probabilities in our model. The resulting rational function coincides with (1.3). The ML degree is also defined when the given curve X ⊂ P2 is not smooth, but it counts critical points of (cid:96)u only in the regular locus of X . Here is an example to illustrate this. Example 1.4. A general cubic curve X in P2 has ML degree 12. Suppose now that X is a cubic which meets H transversally but has one isolated singular point in P2\H. If the singular point is a node then the ML degree of X is 10, and if the singular point is a cusp then the ML degree of X is 9. The ML degrees are found by saturating the equations in ♦ (1.2) with respect to the homogenous ideal of the singular point. Moving beyond likelihood geometry in the plane, we shall introduce our ob jects in any dimension. We fix the complex pro jective space Pn with coordinates p0 , p1 , . . . , pn , represent- ing probabilities. We summarize the observed data in a vector u = (u0 , u1 , . . . , un ) ∈ Nn+1 , where ui is the number of samples in state i. The likelihood function on Pn given by u equals 1 · · · pun pu0 0 pu1 n (p0 + p1 + · · · + pn )u0+u1+···+un (cid:96)u = . 4 The unique critical point of this rational function on Pn is the data point itself: (u0 : u1 : · · · : un ). , u1 p1 (1.5) , . . . , un pn Moreover, this point is the global maximum of the likelihood function (cid:96)u on the probability simplex ∆n . Throughout, we identify ∆n with the set of all positive real points in Pn . The linear forms in (cid:96)u define an arrangement H of n + 2 distinguished hyperplanes in Pn . (cid:1) − u+ dlog((cid:96)u ) = (cid:0) u0 The differential of the logarithm of the likelihood function is the vector of rational functions · (1, 1, . . . , 1). Here p+ = (cid:80)n i=0 pi and u+ = (cid:80)n p0 p+ i=0 ui . The vector (1.5) represents a section of the sheaf of differential 1-forms on Pn that have logarithmic singularities along H. This sheaf is denoted Ω1Pn (log(H)). Our aim is to study the restriction of (cid:96)u to a closed subvariety X ⊆ Pn . We will assume that X is defined over the real numbers, irreducible, and not contained in H. Let Xsing denote the singular locus of X , and Xreg denote X \Xsing . When X serves as a statistical model, the goal is to maximize the rational function (cid:96)u on the semialgebraic set X ∩ ∆n . To solve this problem algebraically, we determine all critical points of the log-likelihood function log((cid:96)u ) on the complex variety X . Here we must exclude points that are singular or lie in H. Definition 1.5. The maximum likelihood degree of X is the number of complex critical points of the function (cid:96)u on Xreg \H, for generic data u. The likelihood correspondence LX is (cid:8)(p, u) : p ∈ Xreg \H and dlog((cid:96)u ) vanishes at p(cid:9). the universal family of these critical points. To be precise, LX is the closure in Pn × Pn of p ×Pn u for Pn×Pn to highlight that the first factor is the probability We sometimes write Pn space, with coordinates p, while the second factor is the data space, with coordinates u. The first part of the following result appears in [25, §2]. A precursor was [23, Proposition 3]. Theorem 1.6. The likelihood correspondence LX of any irreducible subvariety X in Pn p is p × Pn u . The map pr1 : LX → Pn an irreducible variety of dimension n in the product Pn p is a projective bund le over Xreg \H, and the map pr2 : LX → Pn u is generical ly finite-to-one. See Section 4 for a proof. The degree of the map pr2 : LX → Pn u to data space is the ML degree of X . This number has a topological interpretation as an Euler characteristic, provided suitable assumptions on X are being made. The relationship between the homology of a manifold and critical points of a suitable function on it is the topic of Morse theory. The study of ML degrees was started in [10, §2] by developing the connection to the sheaf X (log(H)) of differential 1-forms on X with logarithmic poles along H. It was shown in Ω1 [10, Theorem 20] that the ML degree of X equals the signed topological Euler characteristic (−1)dim X · χ(X \H), 5 provided X is smooth and the intersection H∩X defines a normal crossing divisor in X ⊆ Pn . A ma jor drawback of that early result was that the hypotheses are so restrictive that they essentially never hold for varieties X that arise from statistical models used in practice. From a theoretical point view, this issue can be addressed by passing to a resolution of singularities. However, in spite of existing algorithms for resolution in characteristic zero, these algorithms do not scale to problems of the sizes of interest in algebraic statistics. Thus, whatever computations we wish to do should not be based on resolution of singularities. The following result due to [25] gives the same topological interpretation of the ML degree. The hypotheses here are much more realistic and inclusive than those in [10, Theorem 20]. Theorem 1.7. If the very affine variety X \H is smooth of dimension d, then the ML degree of X equals the signed topological Euler characteristic of (−1)d · χ(X \H). The term very affine variety refers to a closed subvariety of some algebraic torus (C∗ )m . (cid:1). (p0 : · · · : pn ) (cid:55)−→ (cid:0) p0 Our ambient space Pn\H is a very affine variety because it has a closed embedding Pn\H −→ (C∗ )n+1 , pn p+ p+ The study of such varieties is foundational for tropical geometry. The special case when X \H is a Riemann surface with a punctures, arising from a curve in P2 , was seen in Theorem 1.1. We remark that Theorem 1.7 can be deduced from works of Gabber-Loeser [18] and Franecki- Kapranov [16] on perverse sheaves on algebraic tori. The smoothness hypothesis is essential for Theorem 1.7 to hold. If X is singular then, generally, neither X \H nor Xreg \H has its signed Euler characteristic equal to the ML degree of X . Varieties X that demonstrate this are the two singular cubic curves in Example 1.4. Conjecture 1.8. For any projective variety X ⊆ Pn of dimension d, not contained in H, (−1)d · χ(X \H) ≥ MLdegree (X ). In particular, the signed topological Euler characteristic (−1)d · χ(X \H) is nonnegative. , . . . , Analogous conjectures can be made in the slightly more general setting of [25]. In par- ticular, we conjecture that the inequality (−1)d · χ(V ) ≥ 0 holds for any closed d-dimensional subvariety V ⊆ (C∗ )m . Remark 1.9. We saw in Example 1.2 that the ML degree of a pro jective variety X can be 0. In all situations of statistical interest, the variety X ⊂ Pn intersects the open simplex ∆n in a subset that is Zariski dense in X . If that intersection is smooth then MLdegree(X ) ≥ 1. In fact, arguing as in [10, Proposition 11], it can be shown that for smooth X , MLdegree (X ) ≥ #(bounded regions of XR\H). Here a bounded region is a connected component of the semialgebraic set XR\H whose classical closure is disjoint from the distinguished hyperplane V (p+ ) in Pn . 6 If X is singular then the number of bounded regions of XR\H can exceed MLdegree (X ). For instance, let X ⊂ P2 be the cuspidal cubic curve defined by (p0+p1+p2 )(7p0−9p1−2p2 )2 = (3p0+5p1+4p2 )3 . The real part XR\H consists of 8 bounded and 2 unbounded regions, but the ML degree of X is 7. The bounded region that contains the cusp (13 : 17 : −31) has no other critical ♦ points for (cid:96)u . (1.6) D = u0 p0 ∂ g1 p0 ∂ p0 u1 p1 ∂ g1 p1 ∂ p1 un pn ∂ g1 pn ∂ pn In what follows we present instances that illustrate the computation of the ML degree. We begin with the case of generic complete intersections. Suppose that X ⊂ Pn is a complete intersection defined by r generic homogeneous polynomials g1 , . . . , gr of degrees d1 , d2 , . . . , dr . (cid:88) Theorem 1.10. The ML degree of X equals Dd1d2 · · · dr , where 2 · · · dir 1 di2 di1 r . i1+i2+···+ir ≤n−r Proof. By Bertini’s Theorem, the generic complete intersection X is smooth in Pn . All critical points of the likelihood function (cid:96)u on X lie in the dense open subset X \H. Consider   . the following (r + 2) × (n + 1)-matrix with entries in the polynomial ring R[p0 , p1 , . . . , pn ]: · · · (cid:20) u (cid:21) · · · · · · · · · J (p) ∂ g2 ∂ g2 ∂ g2 ∂ pn ∂ p1 ∂ p0 ... ... ... . . . · · · ∂ gr ∂ gr ∂ gr p0 pn p1 ∂ pn ∂ p1 ∂ p0 Let Y denote the determinantal variety in Pn given by the vanishing of its (r + 2) × (r + 2) minors. The codimension of Y is at most n − r, which is a general upper bound for ideals of maximal minors, and hence the dimension of Y is at least r. Our genericity assumptions ensure that the matrix J (p) has maximal row rank r + 1 for all p ∈ X . Hence a point p ∈ X lies in Y if and only if the vector u is in the row span of J (p). Moreover, by Theorem 1.6, (Xreg \H) ∩ Y = X ∩ Y is a finite subset of Pn , and its cardinality is the desired ML degree of X . Since X has dimension n− r, we conclude that Y has the maximum possible codimension, namely n − r, and that the intersection of X with the determinantal variety Y is proper. We note that Y is Cohen-Macaulay, since Y has maximal codimension n − r, and ideals of minors of generic matrices are Cohen-Macaulay. B´ezout’s Theorem implies MLdegree(X ) = degree(X ) · degree(Y ) = d1 · · · dr · degree(Y ). = p0 p1 pn (1.7) The degree of the determinantal variety Y equals the degree of the determinantal variety given by generic forms of the same row degrees. By the Thom-Porteous-Giambelli formula, 7 this degree is the complete homogeneous symmetric function of degree codim(Y ) = n − r evaluated at the row degrees of the matrix. Here, the row degrees are 0, 1, d1 , . . . , dr , and the value of that symmetric function is precisely D. We conclude that degree(Y ) = D . Hence the ML degree of the generic complete intersection X = V (g1 , . . . , gr ) equals D · d1d2 · · · dn . Example 1.11 (r = 1). A generic hypersurface of degree d in Pn has ML degree d · D = d + d2 + d3 + · · · + dn . Example 1.12 (r = 2, n = 3). A space curve that is the generic intersection of two surfaces ♦ of degree d and e in P3 has ML degree de + d2e + de2 . Remark 1.13. It was shown in [23, Theorem 5] that (1.6) is an upper bound for the ML degree of any variety X of codimension r that is defined by polynomials of degree d1 , . . . , dr . In fact, the same is true under the weaker hypothesis that X is cut out by polynomials of degrees d1 ≥ · · · ≥ dr ≥ dr+1 ≥ · · · ≥ ds , so X need not be a complete intersection. However, the hypothesis codim(X ) = r is essential in order for MLdegree(X ) ≤ (1.6) to hold. That codimension hypothesis was forgotten when this upper bound was cited in [15, Theorem 2.2.6] and in [37, Theorem 3.31]. Hence these two book references are not correct as stated. Here is a simple counterexample. Let n = 3 and d1 = d2 = d3 = 2. Then the bound (1.6) is the B´ezout number 8, and this is also the correct ML degree for a general complete intersection of three quadrics in P3 . Now let X be a general rational normal curve in P3 . The curve X is defined by three quadrics, namely, the 2 × 2-minors of a 2 × 3-matrix filled with general linear forms in p0 , p1 , p2 , p3 . Since X is a Riemann sphere with 15 punctures, ♦ Theorem 1.7 tells us that MLdegree(X ) = 13, and this exceeds the bound of 8. We now come to a variety that is ubiquitous in statistics, namely the model of inde- pendence for two binary random variables [15, §1.1]. This model is represented by Segre’s quadric surface X in P3 . By this we mean the surface defined by the 2 × 2-determinant: X = V (p00p11 − p01p10 ) ⊂ P3 . The surface X is isomorphic to P1 × P1 , so it is smooth, and we can apply Theorem 1.7 to H = (cid:8) p ∈ P3 : p00p01p10p11 (p00+p01+p10+p11 ) = 0(cid:9). find the ML degree. In other words, we seek to determine the Euler characteristic of the open complex surface X \H where To this end, we write X = P1 × P1 with coordinates (cid:0)(x0 : x1 ), (y0 : y1 )(cid:1). Our surface is X \H = (cid:0)P1 × P1(cid:1)\(cid:8)x0x1y0y1 (x0 + x1 )(y0 + y1 ) = 0(cid:9) parametrized by pij = xiyj , and hence = (cid:0)P1\{x0x1 (x0 + x1 ) = 0}(cid:1) × (cid:0)P1\{y0y1 (y0 + y1 ) = 0}(cid:1) = (cid:0)2-sphere\{three points}(cid:1) × (cid:0)2-sphere\{three points}(cid:1). Since the Euler characteristic is additive and multiplicative, χ(X \H) = (−1) · (−1) = 1. This means that the map u (cid:55)→ p from the data to the MLE is a rational function in each coordinate. The following “word problem for freshmen” is aimed at finding that function. 8 u = . (1.8) p = . p = 1 (u++ )2 (cid:18)u00 u01 (cid:19) Example 1.14. Do this exercise: A biologist friend of yours wishes to test whether two binary random variables are independent. She collects data and records the matrix of counts u10 u11 How to ascertain whether u lies close to the independence model X = V (p00p11 − p01p10 ) ? A statistician who recently started working in her lab explains that, as the first step in the (cid:18) p00 p01 (cid:19) analysis of her data, the biologist should calculate the maximum likelihood estimate (MLE) p10 p11 Can you help your friend by supplying the formula for p as a rational function in u? (cid:18)u0+ (cid:19) · (cid:0)u+0 u+1 (cid:1) . The solution to this word problem is as follows. The MLE is the rank 1 matrix u1+ We illustrate the concepts introduced above by deriving this well-known formula. The like- lihood correspondence LX of X = V (p00p11 − p01p10 ) is the subvariety of X × P3 defined by U · (p00 , p01 , p10 , p11 )T = 0, (1.9)  .  0 −u10 − u11 u00 + u01 0 u11 + u01 −u00 − u10 0 0 −u01 − u00 0 u11 + u10 0 −u01 − u11 u00 + u10 0 0 We urge the reader to derive (1.9) from Definition 1.5 using a computer algebra system. Note that the determinant of U vanishes identically. In fact, for generic uij , the matrix U has rank 3, so its kernel is spanned by a single vector. The coordinates of that vector are given by Cramer’s rule, and we find them to be equal to the rational functions in (1.8). The locus where the function u (cid:55)→ p is undefined consists of those u where the matrix rank of U drops below 3. A computation shows that the rank of U drops to 2 on the variety V (u00 + u10 , u01 + u11 ) ∪ V (u00 + u01 , u10 + u11 ), and it drops to 0 on the point V (u00 + u01 , u10 + u11 , u01 + u11 ). In particular, the likelihood ♦ function (cid:96)u given by that point u has infinitely many critical points in the quadric X . We note that all coefficients of the linear forms that define the exceptional loci in P3 u for the independence model are positive. This means that data points u with all coordinates positive can never be exceptional. We will prove in Section 4 that this usually holds. Let pr1 : LX → Pn p and pr2 : LX → Pn u be the pro jections from the likelihood correspondence to p-space and u-space respectively. We are interested in the fibers of pr2 over positive points u. where U is the matrix U = 9 Theorem 1.15. Let u ∈ Rn+1 >0 , and let X ⊂ Pn be an irreducible variety such that no singular points of any intersection X ∩ {pi = 0} lies in the hyperplane at infinity {p+ = 0}. Then 1. the likelihood function (cid:96)u on X has only finitely many critical points in Xreg \H; 2. if the fiber pr−1 2 (u) is contained in Xreg , then its length equals the ML degree of X . The hypothesis concerning “no singular point” will be satisfied for essentially all statistical models of interest. Here is an example which shows that this hypothesis is necessary. Example 1.16. We consider the smooth cubic curve X in P2 that is defined by f = (p0 + p1 + p2 )3 + p0p1p2 . The ML degree of the curve X is 3. Each intersection X ∩ {pi = 0} is a triple point that lies on the line at infinity {p+ = 0}. The fiber pr−1 2 (u) of the likelihood fibration over the ♦ positive point u = (1 : 1 : 1) is the entire curve X . If u is not positive in Theorem 1.15, then the fiber of pr2 over u may have positive dimension. We saw an instance of this at the end of Example 1.14. Such resonance loci have been studied extensively when X is a linear subspace of Pn . See [11] and references therein. The following cautionary example shows that the length of the scheme-theoretic fiber of LX → Pn u over special points u in the open simplex ∆n may exceed the ML degree of X . Example 1.17. Let X be the curve in P2 defined by the ternary cubic f = p2 (p1 − p2 )2 + (p0 − p2 )3 . This curve intersects H in 8 points, has ML degree 5, and has a cuspidal singularity at P := (1 : 1 : 1). The prime ideal in R[p0 , p1 , p2 , u0 , u1 , u2 ] for the likelihood correspondence LX is minimally generated by five polynomials, having degrees (3, 0), (2, 2), (3, 1), (3, 1), (3, 1). They are ob- tained by saturating the two equations in (1.2) with respect to (cid:104)p0p2 (cid:105) ∩ (cid:104)p0 − p1 , p2 − p1 (cid:105). The scheme-theoretic fiber of pr1 over a general point of X is a reduced line in the u-plane, L := (cid:8) (u0 : u1 : u2 ) ∈ P2 : (2u0 − u1 − u2 )2 = 0 (cid:9). while the fiber of pr1 over P is the double line The reader is invited to verify the following assertions using a computer algebra system: 2 (u) consists of 5 reduced points in Xreg \H. u , then pr−1 (a) If u is a general point of P2 (b) If u is a general point on the line L, then the locus of critical points pr−1 2 (u) consists of 4 reduced points in Xreg \H and the reduced point P . (c) If u is the point (1 : 1 : 1) ∈ L, then pr−1 2 (u) is a zero-dimensional scheme of length 6. This scheme consists of 3 reduced points in Xreg \H and P counted with multiplicity 3. 10 In particular, the fiber in (c) is not algebraically equivalent to the general fiber (a). This example illustrates one of the difficulties classical geometers had to face when formulating the “principle of conservation of numbers”. See [17, Chapter 10] for a modern treatment. ♦ It is instructive to examine classical varieties from pro jective geometry from the likeli- hood perspective. For instance, we may study the Grassmannian in its Plucker embedding. Grassmannians are a nice test case because they are smooth, so that Theorem 1.7 applies. Example 1.18. Let X = G(2, 4) denote the Grassmannian of lines in P3 . In its Plucker embedding in P5 , this Grassmannian is the quadric hypersurface defined by p12p34 − p13p24 + p14p23 = 0. (1.10)  .  u12 As in (1.7), the critical equations for the likelihood function (cid:96)u are the 3 × 3-minors of u34 u24 u23 u14 u13 p34 p24 p23 p14 p13 p12 p12p34 −p13p24 p14p23 p14p23 −p13p24 p12p34 By Theorem 1.6, the likelihood correspondence LX is a five-dimensional subvariety of P5 ×P5 . The cohomology class of this subvariety can be represented by the bidegree of its ideal: (1.11) BX (p, u) = 4p5 + 6p4u + 6p3u2 + 6p2u3 + 2pu4 . (1.12) This is the multidegree, in the sense of [33, §8.5], of LX with respect to the natural Z2 -grading on the polynomial ring R[p, u]. We can use [33, Proposition 8.49] to compute the bidegree from the prime ideal of LX . Its leading coefficient 4 is the ML degree of X . Its trailing coefficient 2 is the degree of X . The polynomials BX (p, u) will be studied in Section 3. The prime ideal of LX is computed from the equations in (1.10) and (1.11) by saturation with respect to H. It is minimally generated by the following eight polynomials in R[p, u]: (a) one polynomial of degree (2, 0), namely the Plucker quadric, (cid:19) (cid:18) p12 − p34 (b) six polynomials of degree (1, 1), given by 2 × 2-minors of p14 − p23 p13 − p24 (cid:18) p12+p13+p23 u12 − u34 u13 − u24 u14 − u23 p23+p24+p34 p13+p14+p34 p12+p14+p24 u12+u13+u23 u12+u14+u24 u13+u14+u34 u23+u24+u34 and (cid:19) , (c) one polynomial of degree (2, 1), for instance 2u24p12p34 + 2u34p13p24 + (u23 + u24 + u34 )p14p24 −(u13 + u14 + u34 )p2 24 − (u12 + 2u13 + u14 − u24 )p24p34 . 11 For a fixed positive data vector u > 0, these six polynomials in (b) reduce to three linear equations, and these cut out a plane P2 inside P5 . To find the four critical points of (cid:96)u on X = G(2, 4), we must then intersect the two conics (a) and (c) in that plane P2 . of the manifold G(r, m)\H obtained by removing (cid:0)m (cid:1) + 1 distinguished hyperplane sections. r )−1 is the signed Euler characteristic The ML degree of the Grassmannian G(r, m) in P(m r It would be very interesting to find a general formula for this ML degree. At present, we only know that the ML degree of G(2, 5) is 26, and that the ML degree of G(2, 6) is 156. By Theorem 1.7, these numbers give the Euler characteristic of G(2, m)\H for m ≤ 6. ♦ We end this lecture with a discussion of the delightful case when X is a linear subspace of Pn , and the open variety X \H is the complement of a hyperplane arrangement. In this context, following Varchenko [44], the likelihood function (cid:96)u is known as as the master function, and the statement of Theorem 1.7 was first proved by Orlik and Terao in [36]. We assume that X has dimension d, is defined over R, and does not contain the vector 1 = (1, 1, . . . , 1). We can regard X as a (d + 1)-dimensional linear subspace of Rn+1 . The orthogonal complement X ⊥ with respect to the standard dot product is a linear space of dimension n − d in Rn+1 . The linear space X ⊥ + 1 spanned by X ⊥ and the vector 1 has dimension n − d + 1 in Rn+1 , and hence can be viewed as subspace of codimension d in Pn u . In our next formula, the operation (cid:63) is the Hadamard product or coordinatewise product. Proposition 1.19. The likelihood correspondence LX in Pn × Pn is defined by p ∈ X and u ∈ p (cid:63) (X ⊥ + 1). (1.13) The prime ideal of LX is obtained from these constraints by saturation with respect to H. Proof. If all pi are non-zero then u ∈ p (cid:63) (X ⊥ + 1) says that (cid:1) u/p := (cid:0) u0 un u1 p0 p1 pn lies in the subspace X ⊥ + 1. Equivalently, the vector obtained by adding a multiple of (1, 1, . . . , 1) to u/p is perpendicular to X . We can take that vector to be the differential (1.5). Hence (1.13) expresses the condition that p is a critical point of (cid:96)u on X . The intersection X ∩ H is an arrangement of n + 2 hyperplanes in X (cid:39) Pd . For special choices of the subspace X , it may happen that two or more hyperplanes coincide. Taking {p+ = 0} as the hyperplane at infinity, we view X ∩H as an arrangement of n+ 1 hyperplanes in the affine space Rd . A region of this arrangement is bounded if it is disjoint from {p+ = 0}. , . . . , , Theorem 1.20. The ML degree of X is the number of bounded regions of the real affine hyperplane arrangement X ∩H in Rd . The bidegree of the likelihood correspondence LX is the h-polynomial of the broken circuit complex of the rank d+1 matroid associated with X ∩ H. We need to explain the second assertion. The hyperplane arrangement X ∩ H consists of the intersections of the n + 2 hyperplanes in H with X (cid:39) Pd . We regard these as hyperplanes through the origin in Rd+1 . They define a matroid M of rank d + 1 on n + 2 elements. We identify these elements with the variables x1 , x2 , . . . , xn+2 . For each circuit C of M let mC = 12 ((cid:81) i∈C xi )/xj where j is the smallest index such that xj ∈ C . The broken circuit complex of M is the simplicial complex with Stanley-Reisner ring R[x1 , . . . , xn+2 ]/(cid:104) mC : C circuit of M (cid:105). See [33, §1.1] for Stanley-Reisner basics. The Hilbert series of this graded ring has the form h0 + h1z + · · · + hdz d (1 − z )d+1 . What is being claimed in Theorem 1.20 is that the bidegree of LX equals BX (p, u) = (h0ud + h1pud−1 + h2p2ud−2 + · · · + hdpd ) · pn−d Equivalently, this is the class of LX in the cohomology ring H ∗ (Pn × Pn ; Z) = Z[p, u]/(cid:104)pn+1 , un+1 (cid:105). There are several (purely combinatorial) definitions of the invariants hi of the matroid M . For χM (q + 1) = q · (cid:16) (cid:17) instance, they are coefficients of the following specialization of the characteristic polynomial: h0qd − hd−1qd−1 + · · · + (−1)d−1h1q + (−1)dh0 Theorem 1.20 was used in [26] to prove a conjecture of Dawson, stating that the sequence h0 , h1 , . . . , hd is log-concave, when M is representable over a field of characteristic zero. The first assertion in Theorem 1.20 was proved by Varchenko in [44]. For definitions and characterizations of the characteristic polynomial χ, and many pointers to matroid basics, we refer to [35]. A proof of the second assertion was given by Denham et al. in a slightly different setting [13, Theorem 1]. We give a proof in Section 4 following [25, §3]. The ramification locus of the likelihood fibration pr2 : LX → Pn u is known as the entropic discriminant [40]. Example 1.21. Let d = 2 and n = 4, so X is a plane in P4 , defined by two linear forms (1.14) (1.15) . (1.16) c10p0 + c11p1 + c12p2 + c13p3 + c14p4 = 0, c20p0 + c21p1 + c22p2 + c23p3 + c24p4 = 0. Following Theorem 1.20, we view X ∩ H as an arrangement of five lines in the affine plane { p ∈ X : p0 + p1 + p2 + p3 + p4 (cid:54)= 0 } (cid:39) C2 . Hence, for generic cij , the ML degree of X is equal to 6, the number of bounded regions of  ≤ 3.  u0 this arrangement. The condition u ∈ p (cid:63) (X ⊥ + 1) in Proposition 1.19 translates into u4 u3 u2 u1 p4 p3 p2 p1 p0 c14p4 c13p3 c12p2 c11p1 c10p0 c24p4 c23p3 c22p2 c21p1 c20p0 The 4 × 4-minors of this 4 × 5-matrix, together with the two linear forms defining X , form a system of equations that has six solutions in P4 , for generic cij . All solutions have real coordinates. In fact, there is one solution in each bounded region of X \H. The likelihood correspondence LX is the fourfold in P4 × P4 given by the equations (1.16) and (1.17). (1.17) rank 13 We now illustrate the second statement in Theorem 1.20. Suppose that the real numbers cij are generic, so M is the uniform matroid of rank three on six elements. The Stanley- Reisner ring of the broken circuit complex of M equals R[x1 , x2 , x3 , x4 , x5 , x6 ]/(cid:104)x2x3x4 , x2x3x5 , x2x3x6 , . . . , x4x5x6 (cid:105). The Hilbert series of this graded algebra is 1 + 3z + 6z 2 h0 + h1z + h2z 2 (1 − z )3 (1 − z )3 We conclude that the bidegree (1.14) of the likelihood correspondence LX equals BX (p, u) = 6p4 + 3p3u + p2u2 . = . For special choices of the coefficients cij in (1.16), some triples of lines in the arrangement X ∩ H may meet in a point. For such matroids, the ML degree drops from 6 to some integer between 0 and 5. We recommend it as an exercise to the reader to explore these cases. For instance, can you find explicit cij so that the ML degree of X equals 3? What are the prime ideal and the bidegree of LX in that case? How can the ML degree of X be 0 or 1? ♦ It would be interesting to know which statistical model X in Pn defines the likelihood correspondence LX which is a complete intersection in Pn × Pn . When X is a linear subspace of Pn , this question is closely related to the concept of freeness of a hyperplane arrangement. Proposition 1.22. If the hyperplane arrangement X ∩ H in X is free, then the likelihood correspondence LX is an ideal-theoretic complete intersection in Pn × Pn . Proof. For the definition of freeness see §1 in the paper [11] by Cohen, Denman, Falk and Varchenko. The proposition is implied by their [11, Theorem 2.13] and [11, Corollary 3.8]. Using Theorem 1.20, this provides a likelihood geometry proof of Terao’s theorem that the characteristic polynomial of a free arrangement factors into integral linear forms [41]. 2 Second Lecture In our newspaper we frequently read about studies aimed at proving that a behavior or food causes a certain medical condition. We begin the second lecture with an introduction to statistical issues arising in such studies. The “medical question” we wish to address is Does Watching Soccer on TV Cause Hair Loss? We learned this amusing example from [34, §1]. In a fictional study, 296 British sub jects aged between 40 to 50 were interviewed about their hair length and how many hours per week they watch soccer (a.k.a. “football”) on TV. Their responses are summarized in the following contingency table of format 3 × 3:   lots of hair medium hair 51 45 28 30 15 27 14 little hair 33 29 38 ≤ 2 hrs 2–6 hrs ≥ 6 hrs U = For instance, 29 respondents reported having little hair and watching between 2 and 6 hours of soccer on TV per week. Based on these data, are these two random variables independent, or are we inclined to believe that watching soccer on TV and hair loss are correlated? On first glance, the latter seems to be the case. Indeed, being independent means that the data matrix U should be close to a rank 1 matrix. However, all 2 × 2-minors of U are strictly positive, indeed by quite a margin, and this suggests a positive correlation. However, this interpretation is deceptive. A much better explanation of our data can be given by identifying a certain hidden random variable. That hidden variable is gender. Indeed, suppose that among the respondents 126 were males and 170 were females. Our 3 48 36 18  +  . data matrix U is then the sum of the male table and the female table, maybe as follows: 9 15 4 12 20 24 18 9 7 21 35 8 6 3 Both of these tables have rank 1, hence U has rank 2. Hence, the appropriate null hypothesis H0 for analyzing our situation is not independence but it is conditional independence: U = (2.1) H0 : Soccer on TV and Hair Loss are Independent given Gender. P = p1n p2n ... pmn And, based on the data U , we most definitely do not reject that null hypothesis. The key feature of the matrix U above was that it has rank 2. We now define low rank matrix models in general. Consider two discrete random variables X and Y having m and  p11  n states respectively. Their joint probability distribution is written as an m × n-matrix · · · p12 · · · p22 p21 ... ... . . . · · · pm1 pm2 whose entries are nonnegative and sum to 1. Here pij represents the probability that X is in state i and Y is in state j . The of all probability distributions is the standard simplex ∆mn−1 of dimension mn − 1. We write Mr for the manifold of rank r matrices in ∆mn−1 . The matrices P in M1 represent independent distributions. Mixtures of r independent distributions correspond to matrices in Mr . As always in applied algebraic geometry, we can make any problem that involves semi-algebraic sets progressively easier by three steps: • disregard inequalities, • replace real numbers with complex numbers, • replace affine space by pro jective space. In our situation, this leads us to replacing Mr with its Zariski closure in complex pro jective space Pmn−1 . This Zariski closure is the pro jective variety Vr of complex m×n matrices of rank ≤ r. Note that Vr is singular along Vr−1 . The codimension of Vr is (m − r)(n − r). It is a non-trivial exercise to write the degree of Vr in terms of m, n, r . Hint: [33, Example 15.2]. 15 Suppose now that i.i.d. samples are drawn from an unknown joint distribution on our  u11  . two random variables X and Y . We summarize the resulting data in a contingency table · · · u1n u12 · · · u2n u22 u21 ... ... ... . . . · · · umn um1 um2 The entries of the matrix U are nonnegative integers whose sum is u++ . (cid:19) m(cid:89) (cid:18) n(cid:89) The likelihood function for the contingency table U is the following function on ∆mn−1 : u++ u11u12 · · · umn j=1 i=1 U = puij ij . (cid:55)→ P Assuming fixed sample size, this is the likelihood of observing the data U given an unknown probability distribution P in ∆mn−1 . In what follows we suppress the multinomial coefficient. (cid:81)n (cid:81)m Furthermore, we regard the likelihood function as a rational function on Pmn−1 , so we write j=1 puij ij i=1 pu++ ++ (cid:96)U = . (2.2) We wish to find a low rank probability matrix P that best explains the data U . Maximum likelihood estimation means solving the following optimization problem: Maximize (cid:96)U (P ) sub ject to P ∈ Mr . The optimal solution P is a rank r matrix. This is the maximum likelihood estimate for U . For r = 1, the independence model, the maximum likelihood estimate P is obtained from the data matrix U by the following formula, already seen for m = n = 2 in (1.8). Multiply  u1+  · (cid:0)u+1 u+2 the vector of row sums with the vector of column sums and divide by the sample size: (cid:1) . u2+ ... um+ Statisticians, scientists and engineers refer to such a formula as an “analytic solution”. In our view, it would be more appropriate to call this an “algebraic solution”. After all, we are here using algebra not analysis. Our algebraic solution for r = 1 reveals the following points: • The MLE P is a rational function of the data U . • The function U (cid:55)→ P is an algebraic function of degree 1. • The ML degree of the independence model V1 equals 1. (u++ )2 · 1 · · · u+n P = (2.3) We next discuss the smallest case when the ML degree is larger than 1. 16 Example 2.1. Let m = n = 3 and r = 2. Our MLE problem is to maximize 33 )/pu++ 32 pu33 31 pu32 23 pu31 22 pu23 21 pu22 13 pu21 12 pu13 11 pu12 (cid:96)U = (pu11 ++ sub ject to the constraints P ≥ 0 and rank(P ) = 2, where P = (pij ) is a 3×3-matrix of un- knowns. The equations that characterize the critical points of this optimization problem are p11p22p33 − p11p23p32 − p12p21p33 +p12p23p31 + p13p21p32 − p13p22p31 det(P ) = = 0  u11  and the vanishing of the 3 × 3-minors of the following 3 × 9-matrix: u33 u32 u31 u23 u22 u21 u13 u12 p33 p11 p12 p13 p21 p22 p23 p31 p32 p11a11 p12a12 p13a13 p21a21 p22a22 p33a33 p31a31 p32a32 p33a33 where aij = ∂ det(P ) is the cofactor of pij in P . For random positive data uij , these equations ∂ pij have 10 solutions with rank(P ) = 2 in P8\H. Hence the ML degree of V2 is 10. If we regard the uij as unknowns, then saturating the above determinantal equations with respect to H ∪ V1 yields the prime ideal of the likelihood correspondence LV2 ⊂ P8 × P8 . See Example 4.8 for the bidegree and other enumerative invariants of the 8-dimensional variety LV2 . ♦ Recall from Definition 1.5 that the ML degree of a statistical model (or a pro jective variety) is the number of critical points of the likelihood function for generic data. Theorem 2.2. The known values for the ML degrees of the determinantal varieties Vr are (5, 5) (4, 6) (m, n) = (3, 3) (3, 4) (3, 5) (4, 4) (4, 5) 1 1 1 1 1 1 1 58 10 26 191 843 3119 6776 843 3119 61326 191 1 1 1 6776 1 1 1 1 r = 1 r = 2 r = 3 r = 4 r = 5 The numbers 10 and 26 were computed back in 2004 using the symbolic software Singular, and they were reported in [23, §5]. The bold face numbers were found in 2012 in [22] using the numerical software Bertini. In what follows we shall describe some of the details. Remark 2.3. Each determinantal variety Vr is singular along the smaller variety Vr−1 . (cid:81)pij = 0}. According to Conjecture 1.8, the ML degree above provides a Hence, the very affine variety Vr \H is singular for r ≥ 2, so Theorem 1.7 does not apply. Here, H = {p++ lower bound for the signed topological Euler characteristic of Vr \H. The difference between the two numbers reflect the nature of the singular locus Vr−1\H inside Vr \H. For plane curves that have nodes and cusps, we encountered this issue in Examples 1.4 and 1.17. We begin with a geometric description of the likelihood correspondence. An m× n-matrix P is a regular point in Vr if and only if rank(P ) = r. The tangent space TP is a subspace of dimension rn+rm−r2 in Cm×n . Its orthogonal complement T ⊥ P has dimension (m−r)(n−r). The partial derivatives of the log-likelihood function log((cid:96)U ) on Pmn−1 are − u++ ∂ log((cid:96)U ) uij ∂ pij pij p++ = . 17 Proposition 2.4. An m × n-matrix P of rank r is a critical point for log((cid:96)U ) on Vr if and (cid:21) (cid:20) uij only if the linear subspace T ⊥ P contains the matrix − u++ p++ pij i=1,...,m j=1,...,n In order to get to the numbers in Theorem 2.2, the geometric formulation was replaced in [22] with a parametric representation of the rank constraints. The following linear algebra formulation worked well for non-trivial computations. Assume m ≤ n. Let P1 , R1 , L1 and Λ (cid:18) R1−In−r (cid:18) P1 (cid:19) (cid:19) be matrices of unknowns of formats r × r, r × (n−r), (m−r) × r, and (n−r) × (m−r). Set (cid:1), P = L = (cid:0)L1 −Im−r P1R1 , L1P1 L1P1R1 where Im−r and In−r are identity matrices. In the next statement we use the symbol (cid:63) for the Hadamard (entrywise) product of two matrices that have the same format. Proposition 2.5. Fix a general m × n data matrix U . The polynomial system P (cid:63) (R · Λ · L)T + u++ · P = U , and R = consists of mn equations in mn unknowns. For generic U , it has finitely many complex solutions (P1 , L1 , R1 , Λ). The m×n-matrices P resulting from these solutions are precisely the critical points of the likelihood function (cid:96)U on the determinantal variety Vr .   . We next present the analogue to Theorem 2.2 for symmetric matrices · · · · · · · · · . . . · · · Such matrices, with nonnegative coordinates pij that sum to 1, represent joint probability distributions for two identically distributed random variables with n states. The case n = 2 and r = 1 is the Hardy-Weinberg curve, which we discussed in detail in Example 1.3. p1n p2n p3n ... 2pnn 2p11 p12 p13 ... p1n P = p12 2p22 p23 ... p2n p13 p23 2p33 ... p3n Theorem 2.6. The known values for ML degrees of symmetric matrices of rank at most r (mixtures of r independent identical ly distributed random variables) are n = 2 1 1 6 5 4 3 1 1 1 1 6 37 270 2341 1 37 1394 ? ? 270 1 1 2341 r = 1 r = 2 r = 3 r = 4 r = 5 18 At present we do not know the common value of the ML degree for n = 6 and r = 3, 4. In what follows we take a closer look at the model for symmetric 3 × 3-matrices of rank 2. Example 2.7. Let n = 3 and r = 2, so X is a cubic hypersurface in P5 . The likelihood correspondence LX is a five-dimensional subvariety of P5 × P5 having bidegree BX (p, u) = 6p5 + 12p4u + 15p3u2 + 12p2u3 + 3pu4 . The bihomogeneous prime ideal of LX is minimally generated by 23 polynomials, namely: • One polynomial of bidegree (3, 0); this is the determinant of P . • Three polynomials of degree (1, 1). These come from the underlying toric model (cid:18) 2p0 + p1 + p2 (cid:19) {rank(P ) = 1}. As suggested in Proposition 3.5, they are the 2 × 2-minors of p2 + p4 + 2p5 p1 + 2p3 + p4 2u0 + u1 + u2 u1 + 2u3 + u4 u2 + u4 + 2u5 • One polynomial of degree (2, 1), • three polynomial of degree (2, 2), • nine polynomials of degree (3, 1), • six polynomials of degree (3, 2). It turns out that this ideal represents an expression for the MLE P in terms of radicals in U . We shall work this out for one numerical example. Consider the data matrix U with . u11 = 10, u12 = 9, u13 = 1, u22 = 21, u23 = 3, u33 = 7. For this choice, all six critical points of the likelihood function are real and positive: p33 p23 p22 p13 p12 p11 0.1037 0.3623 0.0186 0.3179 0.0607 0.1368 0.1084 0.2092 0.1623 0.3997 0.0503 0.0702 0.0945 0.2554 0.1438 0.3781 0.4712 0.0810 0.1794 0.2152 0.0142 0.3052 0.2333 0.0528 0.1565 0.2627 0.0125 0.2887 0.2186 0.0609 0.1636 0.1517 0.1093 0.3629 0.1811 0.0312 log (cid:96)U (p) −82.18102 −84.94446 −84.99184 −85.14678 −85.19415 −87.95759 The first three points are local maxima in ∆5 and the last three points are local minima. These six points define an algebraic field extension of degree 6 over Q. One might expect that the Galois group of these six points over Q is the full symmetric group S6 . If this were the case then the above coordinates could not be written in radicals. However, that expectation is wrong. The Galois group of the likelihood fibration pr2 : LX → P5 U given by the 3 × 3 symmetric problem is a subgroup of S6 isomorphic to the solvable group S4 . 19 To be concrete, for the data above, the minimal polynomial for the MLE p33 equals 33 − 4125267629399052p5 9528773052286944p6 33 + 713452955656677p4 −63349419858182p3 33 − 75369770028p33 + 744139872 = 0. 33 33 + 3049564842009p2 We solve this equation in radicals as follows: (cid:0) 14779904193 211433981207339 ζ (cid:1) ω1ω 2 12 (ζ − ζ 2 ) ω2 − 66004846384302 227664 + 1 16427 19221271018849 ω 2 2 + p33 = 211433981207339 ζ 2 − 14779904193 2 + 1 2 ω3 , 2 = (cid:0) 150972770845322208 ζ 2(cid:1) + where ζ is a primitive third root of unity, ω 2 1 = 94834811/3, and 1248260766912 + (cid:0) 212309132509 4242035935404 ζ 2(cid:1)ω2 − 150972770845322208 ζ − 5992589425361 5992589425361 97163 ω 3 40083040181952 ω1 , 2 + (cid:0) 17063004159 422867962414678 ζ (cid:1) ω1ω 2 4242035935404 ζ − 212309132509 2409 3 = 5006721709 ω 2 20272573168 ω1ω2 − 158808750548335 422867962414678 ζ 2 − 17063004159 76885084075396 ω 2 2 . The explanation for the extra symmetry stems from the duality theorem below. It furnishes an involution on the set of six critical points that allows us to express them in radicals. ♦ The tables in Theorems 2.2 and 2.6 suggest that the columns will always be symmetric. This fact was conjectured in [22] and subsequently proved by Draisma and Rodriguez in [14]. Theorem 2.8. Fix m ≤ n and consider the determinantal varieties Vi for either general or symmetric matrices. Then the ML degrees for rank r and for rank m−r+1 coincide. In fact, the main result in [14] establishes the following more precise statement. Given a data matrix U of format m × n, we write ΩU for the m × n-matrix whose (i, j ) entry equals uij · ui+ · u+j . (u++ )3 Theorem 2.9. Fix m ≤ n and U an m × n-matrix with strictly positive integer entries. There exists a bijection between the complex critical points P1 , P2 , . . . , Ps of the likelihood function (cid:96)U on Vr and the complex critical points Q1 , Q2 , . . . , Qs of (cid:96)U on Vm−r+1 such that P1 (cid:63) Q1 = P2 (cid:63) Q2 = · · · = Ps (cid:63) Qs = ΩU . Thus, this bijection preserves reality, positivity, and rationality. The key to computing the ML degree tables and to formulating the duality conjectures in [22], was the use of numerical algebraic geometry. The software Bertini allowed for the computation of thousands of instances in which the formula of Theorem 2.9 was confirmed. Bertini is numerical software, based on homotopy continuation, for finding all complex solutions to a system of polynomial equations (and much more). The software is available at [6]. The developers, Daniel Bates, Jonathan Hauenstein, Andrew Sommese, Charles Wampler, have just completed a new textbook [7] on the mathematics behind Bertini. For the past two decades, algebraic geometers have increasingly employed computational methods as a tool for their research. However, these computations have almost always been symbolic (and hence exact). They relied on Grobner-based software such as Singular or Macaulay2. Algebraists often feel a certain discomfort when asked to trust a numerical computation. We encourage discussion about this issue, by raising the following question. 20 Example 2.10. In the rightmost column of Theorem 2.6, it is asserted that the solution to a certain enumerative geometry problem is 2341. Which of these would you trust most: • the output of a symbolic computation? • the output of a numerical computation? • a proof written by an algebraic geometer? In the authors’ view, it always pays off to be critical and double-check all computations, ♦ regardless of how they were carried out. And, this applies to all three of the above. One of the big advantages of numerical algebraic geometry over Grobner bases when it comes to MLE is the separation between Preprocessing and Solving. For any particular variety X ⊂ Pn , such as X = Vr , we preprocess by solving the likelihood equations once, for a generic data set U0 chosen by us. The coordinates of U0 may be complex (rather than real) numbers. We can chose them with stable numerics in mind, so as to compute all critical points up to high accuracy. This step can take a long time, but the output is highly reliable. After solving the equations once, for that generic U0 , all subsequent computations for any other data set U are very fast. In particular, the computation is fully parallelizable. If we have m processors at our disposal, where m = MLdegree (X ), then each processor can track one of the paths. To be precise, homotopy continuation starts from the critical points of (cid:96)U0 and transform them into the critical points of (cid:96)U . Geometrically speaking, for fixed X , the homotopy amounts to walking on the sheets of the likelihood fibration pr2 : LX → Pn u . To illustrate this point, here are the timings (in seconds) that were reported in [22] for the determinantal varieties X = Vr . Those computations were carried out in Bertini on a 64-bit Linux cluster with 160 processors. The first row is the preprocessing time for solving the equations once. The second row is the time needed to solve any subsequent instance: (m, n, r) Preprocessing Solving (4, 4, 2) (4, 4, 3) (4, 5, 2) (4, 5, 3) (5, 5, 2) (5, 5, 4) 348555 146952 2902 1938 427 257 4 4 20 20 83 83 This table suggests that combining numerical algebraic geometry with existing tools from computational statistics might lead to a viable tool for certifiably solving MLE problems. We are now at the point where it is essential to offer a disclaimer. The low rank model Mr does not correctly represent the notion of conditional independence. The model we should have used instead is the mixture model Mixr . By definition, Mixr is the set of probability distributions P in ∆mn−1 that are convex combinations of r independent distributions, each taken from M1 . Equivalently, the mixture model Mixr consists of all matrices P = A · Λ · B , (2.4) where A is a nonnegative m×r-matrix whose rows sum to 1, Λ is a nonnegative r×r diagonal matrix whose entries sum to 1, and B is a nonnegative r×n-matrix whose columns sum to 1. The formula (2.4) expresses Mixr as the image of a trilinear map between polytopes: (cid:55)→ P. φ : (∆m−1 )r × ∆r−1 × (∆n−1 )r → ∆mn−1 , The following result is well-known; see e.g. [15, Example 4.1.2]. (A, Λ, B ) 21 P = · 1 8 Proposition 2.11. Our low rank model Mr is the Zariski closure of the mixture model Mixr in the probability simplex ∆mn−1 . If r ≤ 2 then Mixr = Mr . If r ≥ 3 then Mixr (cid:40) Mr . The point here is the distinction between the rank and the nonnegative rank of a non- negative matrix. Matrices in Mr have rank ≤ r and matrices in Mixr have nonnegative rank ≤ r. Thus elements of Mr \Mixr are matrices whose nonnegative rank exceeds its rank. 1 1 0 0  Example 2.12. The following 4 × 4-matrix has rank 3 but nonnegative rank 4: 0 1 1 0 0 0 1 1 1 0 0 1 This is the slack matrix of a regular square. It is an element of M3\Mix3 . ♦ Engineers and scientists care more about Mixr than Mr . In many applications, nonneg- ative rank is more relevant than rank. The reason can be seen in (2.1). In such a low-rank decomposition, we do not want the female table or the male table to have a negative entry. This raises the following important questions: How to maximize the likelihood function (cid:96)U over Mixr ? What are the algebraic degrees associated with that optimization problem? Statisticians seek to maximize the likelihood function (cid:96)U on Mixr by using the expectation- maximization (EM) algorithm in the space (∆m−1 )r ×∆r−1 × (∆n−1 )r of parameters (A, Λ, B ). In each iteration, the EM algorithm strictly decreases the Kul lback-Leibler divergence from · U . The hope in 1 the current model point P = φ(A, Λ, B ) to the empirical distribution u++ running the EM algorithm for given data U is that it converges to the global maximum P on Mixr . For a presentation of the EM algorithm for discrete algebraic models see [37, §1.3]. A study of the geometry of this algorithm for the mixture model Mixr is undertaken in [29]. If the EM algorithm converges to a point that lies in the interior of the parameter poly- tope, and is non-singular with respect to φ, then that point will be among the critical points on Mr . These are characterized by Proposition 2.4. However, since Mixr is properly con- tained in Mr , it frequently happens that the true MLE P lies on the boundary of Mixr . In that case, P is not a critical point of (cid:96)U on Mr , meaning that ( P , U ) is not in the likelihood correspondence on Vr . Such points will never be found by the method described above. In order to address this issue, we need to identify the divisors in the variety Vr ⊂ Pmn−1 that appear in the algebraic boundary of Mixr . By this we mean the irreducible components W1 , W2 , . . . , Ws of the Zariski closure of ∂Mixr . Each of these Wi has codimension 1 in Vr . Once the Wi are identified, one would need to examine their ML degree, and also the ML degree of the various strata Wi1 ∩ · · · ∩Wis in which (cid:96)U might attain its maximum. At present we do not have this information even in the smallest non-trivial case m = n = 4 and r = 3. p11 p12 p13 p14  ·  0  = Example 2.13. We illustrate this issue by describing one of the components W of the  0  algebraic boundary for the mixture model Mix3 when m = n = 4. Consider the equation a12 a13 0 p21 p22 p23 p24 a22 a23 b21 0 p31 p32 p33 p34 a33 a31 b31 0 p41 p42 p43 p44 a41 a42 b12 0 b32 b13 b23 b33 b14 b24 0 22 This parametrizes a 13-dimensional subvariety W of the hypersurface V3 = {det(P ) = 0} in P15 . The variety W is a component in the algebraic boundary of Mix3 . To see this, we choose the aij and bij to be positive, and we note that P lies outside Mix3 when precisely one of the 0 entries gets replaced by −. The prime ideal of W in Q[p11 , . . . , p44 ] is obtained by eliminating the 17 unknowns aij and bij from the 16 scalar equations. A direct computation with Macaulay 2 shows that the variety W is Cohen-Macaulay of codimension-2. By the Hilbert-Burch Theorem, it is defined by the 4 × 4-minors of the 4 × 5-matrix. This following p11 p12 p13 p14  . specific matrix representation was suggested to us by Aldo Conca and Matteo Varbaro: 0 0 p21 p22 p23 p24 p34 (p11p22 − p12p21 ) p31 p32 p33 p34 p41 p42 p43 p44 p41 (p12p24 − p14p22 ) + p44 (p11p22 − p12p21 ) Tte algebraic boundary of Mix3 consists of precisely 304 irreducible components, namely the 16 coordinate hyperplanes and 288 hypersurfaces that are all isomorphic to W . This is proved in the forthcoming paper [29]. At present, we do not know the ML degree of W . ♦ The definition of rank varieties and mixture models extends to m-dimensional tensors P of arbitrary format d1 × d2 × · · · × dm . We refer to Landsberg’s book [30] for an introduction to tensors and their rank. Now, Vr is the variety of tensors of borderrank ≤ r, the model Mr is the set of all probability distributions in Vr , and the model Mixr is the subset of tensors of nonnegative rank ≤ r. Unlike in the matrix case m = 2, the mixture model for borderrank r = 2 is already quite interesting when m ≥ 3. We state two theorems that characterize our ob jects. The set-theoretic version of Theorem 2.14 is due to Landsberg and Manivel [31]. The ideal-theoretic statement was proved more recently by Raicu [38]. Theorem 2.14. The variety V2 is defined by the 3 × 3-minors of al l flattenings of P . i∈A di rows and (cid:81) and writing the tensor P as an ordinary matrix with (cid:81) Here, flattening means picking any subset A of [n] = {1, 2, . . . , n} with 1 ≤ A ≤ n − 1 j (cid:54)∈A dj columns. Theorem 2.15. The mixture model Mix2 is the subset of supermodular distributions in M2 . This theorem was proved in [1]. Being supermodular means that P satisfies a natural family of quadratic binomial inequalities. We explain these for m = 3, d1 = d2 = d3 = 2. Example 2.16. We consider 2 × 2 × 2 tensors. Since secant lines of the Segre variety P1×P1×P1 fill all of P7 , we have that V2 = P7 and M2 = ∆7 . The mixture model Mix2 is an interesting, full-dimensional, closed, semi-algebraic subset of ∆7 . By definition, Mix2 is the image of a 2-to-1 map φ : (∆1 )7 → ∆7 analogous to (2.4). The branch locus is the 2×2×2-hyperdeterminant, which is a hypersurface in P7 of degree 4 and ML degree 13. The analysis in [1, §2] represents the model Mix2 as the union of four toric cel ls. One of these toric cells is the set of tensors satisfying p111p222 ≥ p121p212 p111p222 ≥ p112p221 p112p222 ≥ p122p212 p121p222 ≥ p122p221 p111p122 ≥ p112p121 p111p212 ≥ p112p211 p111p222 ≥ p211p122 p211p222 ≥ p212p221 p111p221 ≥ p121p211 (2.5) 23 Figure 1: A 3-dimensional slice of the 7-dimensional model of 2×2×2 tensors of nonnegative rank ≤ 2. Each toric cell is bounded by 3 quadrics and contains a vertex of the tetrahedron. A nonnegative 2×2×2-tensor P in ∆7 is supermodular if it satisfies these inequalities, possibly after label swapping 1 ↔ 2. We visualize Mix2 by restricting to the 3-dimensional subspace H given by p111 = p222 , p112 = p221 , p121 = p212 and p211 = p122 . The intersection H ∩ ∆7 is a tetrahedron, and we consider H ∩ Mix2 inside that tetrahedron. The restricted model H ∩ Mix2 is shown on the left in Figure 1. It consists of four toric cells as shown on the right side. The boundary is given by three quadratic surfaces, shown in red, green and blue, and which are obtained from either the first or the second row in (2.5) by restriction to H . The boundary analysis suggested in Example 2.13 turns out to be quite simple in the present example. All boundary strata of the model Mix2 are varieties of ML degree 1. One such boundary stratum for Mix2 is the 5-dimensional toric variety X = V (p112p222 − p122p212 , p111p122 − p112p121 , p111p222 − p121p212 ) ⊂ P7 . As a preview for what is to come, we report its ML bidegree and its sectional ML degree: BX (p, u) = p7 + 2p6u + 3p5u2 + 3p4u3 + 3p3u4 + 3p2u5 , SX (p, u) = p7 + 14p6u + 30p5u2 + 30p4u3 + 15p3u4 + 3p2u5 . (2.6) In the next section, we shall study the class of toric varieties and the class of varieties having ♦ ML degree 1. Our variety X lies in the intersection of these two important classes. 3 Third Lecture In our third lecture we start out with the likelihood geometry of embedded toric varieties. Fix a (d+1) × (n+1) integer matrix A = (a0 , a1 , . . . , an ) of rank d+1 that has (1, 1, . . . , 1) as its last row. This matrix defines an effective action of the torus (C∗ )d on pro jective space Pn : t × (p0 : p1 : · · · : pn ) (cid:55)−→ (ta0 · p0 : ta1 · p1 : · · · : tan · pn ). (C∗ )d × Pn −→ Pn , 24 A = p/c = Here ai is the column vector ai with the last entry 1 removed. We also fix c = (c0 , c1 , . . . , cn ) ∈ (C∗ )n+1 , viewed as a point in Pn . Let Xc be the closure in Pn of the orbit (C∗ )d · c. This is a pro jective toric variety of dimension d, defined by the pair (A, c). The ideal that defines Xc (cid:19) (cid:18) p0 is the familiar toric ideal IA as in [15, §1.3], but with p = (p0 , . . . , pn ) replaced by pn p1 . , . . . , , cn c1 c0 0 3 0 1  Example 3.1. Fix d = 2 and n = 3. The matrix 0 0 3 1 1 1 1 1 specifies the following family of toric surfaces of degree three in P3 : 3 · p0p1p2 − c0c1c2 · p3 Xc = {(c0 : c1x3 2 : c3x1x2 ) : (x1 , x2 ) ∈ (C∗ )2} = V (c3 1 : c2x3 3 ). (cid:18) p3 (cid:19)3 Of course, the prime ideal of any particular surface Xc is the principal ideal generated by − p0 p1 p2 . c0 c1 c2 c3 ♦ How does the ML degree of Xc depend on the parameter c = (c0 , c1 , c2 , c3 ) ∈ (C∗ )4? We shall express the ML degree of the toric variety Xc in terms of the complement of a hypersurface in the torus (C∗ )d . The pair (A, c) define the sparse Laurent polynomial f (x) = c0 · xa0 + c1 · xa1 + · · · + cn · xan . Theorem 3.2. The ML degree of the d-dimensional toric variety Xc ⊂ Pn is equal to (−1)d Xc\H (cid:39) (cid:8)x ∈ (C∗ )d : f (x) (cid:54)= 0(cid:9). times the Euler characteristic of the very affine variety For generic c, the ML degree agrees with the degree of Xc , which is the normalized volume of the d-dimensional lattice polytope conv(A) obtained as the convex hul l of the columns of A. (3.1) (3.2) Proof. We first argue that the identification (3.2) holds. The map x (cid:55)−→ p = (c0 · xa0 : c1 · xa1 : · · · : cn · xan ) defines an injective group homomorphism from (C∗ )d into the dense torus of Pn . Its image is equal to the dense torus of Xc , so we have an isomorphism between (C∗ )d and the dense torus of Xc . Under this isomorphism, the affine open set {f (cid:54)= 0} in (C∗ )d is identified with the affine open set {p0 + · · · + pn (cid:54)= 0} in the dense torus of Xc . The latter is precisely Xc\H. Since (C∗ )d is smooth, we see that Xc\H is smooth, so our first assertion follows from Theorem 1.7. The second assertion is a consequence of the description of the likelihood correspondence LXc via linear sections of Xc that is given in Proposition 3.5 below. 25 Example 3.3. We return to the cubic surface Xc in Example 3.1. For a general parameter vector c, the ML degree of Xc is 3. For instance, the surface V (p0p1p2 − p3 3 ) ⊂ P3 has ML degree 3. However, the ML degree of Xc drops to 2 whenever the plane curve defined by 1 + c2x3 f (x1 , x2 ) = c0 + c1x3 2 + c3x1x2 has a singularity in (C∗ )2 . For instance, this happens for c = (1 : 1 : 1 : −3). The 3 ) ⊂ P3 has ML degree 2. ♦ corresponding surface V (27p0p1p2 + p3 The isomorphism (3.2) has a nice interpretation in terms of Convex Optimization. Namely, it implies that maximum likelihood estimation for toric varieties is equivalent to global mini- mization of posynomials, and hence to the most fundamental case of Geometric Programming. We refer to [9, §4.5] for an introduction to posynomials and geometric programming. We write · for the 1-norm on Rn+1 , we set b = Au, and we assume that c = (c0 , c1 , . . . , cn ) is in Rn+1 >0 . Maximum likelihood estimation for toric models is the problem pu sub ject to p ∈ Xc ∩ ∆n . pu Maximize (3.3) Setting pi = ci · xai as above, this problem becomes equivalent to the geometric program f (x)u sub ject to x ∈ Rd Minimize (3.4) >0 . xb By construction, f (x)u/xb is a posynomial whose Newton polytope contains the origin. Such a posynomial attains a unique global minimum on the open orthant Rd >0 . This can be seen by convexifying as in [9, §4.5.3]. This global minimum of (3.4) corresponds to the solution of (3.3), which exists and is unique by Birch’s Theorem [37, Theorem 1.10].  0 3 0 1  and u = (0, 0, 0, 1). Example 3.4. Consider the geometric program for the surfaces in Example 3.1, with 0 0 3 1 1 1 1 1 The problem (3.4) is to find the global minimum, over all positive x = (x1 , x2 ), of the function f (x1 , x2 ) 2 + c2x−1 1x−1 1 x−1 = c0x−1 1 x2 2 + c1x2 2 + c3 . x1x2 This is equivalent to maximizing p3/p+ sub ject to p ∈ V (c3 3 · p0p1p2 − c0c1c2 · p3 3 ) ∩ ∆3 . ♦ We now describe the toric likelihood correspondence LXc in Pn × Pn associated with the pair (A, c). This is the likelihood correspondence of the toric variety Xc ⊂ Pn defined above. Proposition 3.5. On the open subset (Xc\H) × Pn , the toric likelihood correspondence LXc (cid:18)p/c · AT (cid:19) is defined by the 2 × 2-minors of the 2 × (d+1)-matrix u/c · AT Here the notation p/c is as in (3.1). In particular, for any fixed data vector u, the critical points of (cid:96)u are characterized by a linear system of equations in p restricted to Xc . (3.5) A = . 26 Proof. This is an immediate consequence of Birch’s Theorem [37, Theorem 1.10]. 1 − Example 3.6. The Hardy-Weinberg curve of Example 1.3 is the subvariety Xc = V (p2 (cid:18)0 1 2 (cid:19) 4p0p2 ) in the pro jective plane P2 . As a toric variety, this plane curve is given by A = and c = (1, 2, 1). 2 1 0 (cid:18) p1 + 2p2 (cid:18)2p0 (cid:19) (cid:19) The likelihood correspondence of Xc is the surface in P2 × P2 given by 2p0 + p1 p1 u1 + 2u2 2u0 + u1 2p2 p1 Note that the second determinant equals the determinant of the 2 × 2-matrix (3.5) times 4. Saturating (3.6) with respect to p0 + p1 + p2 reveals two further equations of degree (1, 1): = det det = 0. (3.6) and (u1 + 2u2 )p1 = 2(2u0 + u1 )p2 . 2(u1 + 2u2 )p0 = (2u0 + u1 )p1 For fixed u, these equations have a unique solution in P2 , given by the formula in (1.3). ♦ Toric varieties are rational varieties that are parametrized by monomials. We now ex- amine those varieties that are parametrized by generic polynomials. Understanding these is useful for statistics since many widely used models for discrete data are given in the form f : Θ → ∆n , where Θ is a d-dimensional polytope and f is a polynomial map. The coordinates f0 , f1 , . . . , fn are polynomial functions in the parameters θ = (θ1 , . . . , θd ) satisfying f0 + f1 + · · · + fn = 1. Such models include the mixture models in Proposition 2.11, phylogenetic models, Bayesian networks, hidden Markov models, and many others arising in computational biology [37]. The model specified by the polynomials f0 , . . . , fn is the semialgebraic set f (Θ) ⊂ ∆n . Theorem 3.7. Let f0 , f1 , . . . , fn be polynomials of degrees b0 , b1 , . . . , bn satisfying (cid:80) fi = 1. We study its Zariski closure X = f (Θ) in Pn . Finding its equations is hard and interesting. The ML degree of the variety X is at most the coefficient of z d in the generating function (1 − z )d Equality holds when the coefficients of f0 , f1 , . . . , fn are generic relative to (cid:80) fi = 1. (1 − z b0 )(1 − z b1 ) · · · (1 − z bn ) . Proof. This is the content of [10, Theorem 1]. Example 3.8. We examine the case of quartic surfaces in P3 . Let d = 2, n = 3, pick random affine quadrics f1 , f2 , f3 in two unknowns and set f0 = 1 − f1 − f2 − f3 . This defines a map f : C2 → C3 ⊂ P3 . 27 The ML degree of the image surface X = f (C2 ) in P3 is equal to 25 since (1 − z )2 (1 − 2z )4 = 1 + 6z + 25z 2 + 88z 3 + · · · The rational surface X is a Steiner surface (or Roman surface). Its singular locus consists of three lines that meet in a point P . To understand the graph of f , we observe that the linear span of {f0 , f1 , f2 , f3} in C[x, y ] has a basis {1, L2 , M 2 , N 2} where L, M , N represent lines in C2 . Let l denote the line through M ∩ N parallel to L, m the line through L ∩ N parallel to M , and n the line through L ∩ M parallel to N . The map C2 → X is a bijection outside these three lines, and it maps each line 2-to-1 onto one of the lines in Xsing . The fiber over the special point P on X consists of three points, namely, l ∩ m, l ∩ n and m ∩ n. If the quadric f0 were also picked at random, rather than as 1 − f1 − f2 − f3 , then we would still get a Steiner surface X ⊂ P3 . However, now the ML degree of X increases to 33. On the other hand, if we take X to be a general quartic surface in P3 , so X is a smooth K3 surface of Picard rank 1, then X has ML degree 84. This is the formula in Example 1.11 evaluated at n = 3 and d = 4. Here X \H is the generic quartic surface in P3 with five plane sections removed. The number 84 is the Euler characteristic of that open K3 surface. In the first case, X \H is singular, so we cannot apply Theorem 1.7 directly to our Steiner surface X in P3 . However, we can work in the parameter space and consider the smooth very affine surface C2\V (f0f1f2f3 ). The number 25 is the Euler characteristic of that surface. It is instructive to verify Conjecture 1.8 for our three quartic surfaces in P3 . We found χ(X \H) = 38 25 = MLdegree (X ), > χ(X \H) = 49 33 = MLdegree (X ), > χ(X \H) = 84 84 = MLdegree (X ). = The Euler characteristics of the three surfaces were computed using Aluffi’s method [2]. ♦ We now turn to the following question: which projective varieties X have ML degree one? This question is important for likelihood inference because a model having ML degree one means that the MLE p is a rational function in the data u. It is known that Bayesian networks and decomposable graphical models enjoy this property, and it is natural to wonder which other statistical models are in this class. The answer to this question was given by the first author in [27]. We shall here present the result of [27] from a slightly different angle. Our point of departure is the notion of the A-discriminant, as introduced and stud- ied by Gel’fand, Kapranov and Zelevinsky in [19]. We fix an r × m integer matrix A = (cid:8)(ta1 : ta2 : · · · : tam ) ∈ Pm−1 : t ∈ (C∗ )r (cid:9) (a1 , a2 , . . . , am ) of rank r which has (1, 1, . . . , 1) in its row space. The Zariski closure of is an (r−1)-dimensional toric variety YA in Pm−1 . We here intentionally changed the notation relative to that used for toric varieties at the beginning of this section. The reason is that d and n are always reserved for the dimension and embedding dimension of a statistical model. The dual variety Y ∗ A is an irreducible variety in the dual pro jective space (Pm−1 )∨ whose : x1 · ta1 + x2 · ta2 + · · · + xm · tam = 0 (cid:9). (cid:8) t ∈ (C∗ )r coordinates are x = (x1 : x2 : · · · : xm ). We identify points x in (Pm−1 )∨ with hypersurfaces (3.7) 28 The dual variety Y ∗ A is the Zariski closure in (Pm−1 )∨ of the locus of all hypersurfaces (3.7) that are singular. Typically, Y ∗ A is a hypersurface. In that case, Y ∗ A is defined by a unique (up to sign) irreducible polynomial ∆A ∈ Z[x1 , x2 , . . . , xm ]. The homogeneous polynomial ∆A is called the A-discriminant. Many classical discriminants and resultants are instances (cid:19) (cid:18)3 2 1 0 of ∆A . So are determinants and hyperdeterminants. This is the punch line of the book [19]. YA = (cid:8)(1 : t : t2 : t3 ) t ∈ C(cid:9) ⊂ P3 . Example 3.9. Let m = 4, r = 2, and A = . The associated toric variety is 0 1 2 3 A that is dual to the curve YA is a surface in (P3 )∨ . The surface Y ∗ The variety Y ∗ A parametrizes all planes that are tangent to the curve YA . These represent univariate cubics the twisted cubic curve x1 + x2 t + x3 t2 + x4 t3 that have a double root. Here the A-discriminant is the classical discriminant 4 − 18x1x2x3x4 + 4x1x3 2x4 − x2 ∆A = 27x2 1x2 3 + 4x3 2x2 3 . A in P3 defined by this equation is the discriminant of the univariate cubic. ♦ The surface Y ∗ Theorem 3.10. Let X ⊆ Pn be a projective variety of ML degree 1. Each coordinate pi of the rational function u (cid:55)→ p is an alternating product of linear forms in u0 , u1 , . . . , un . The paper [27] gives an explicit construction of the map u (cid:55)→ p as a Horn uniformization. A precursor was [28]. We explain this construction. The point of departure is a matrix A as above. We now take ∆A to be any non-zero homogenous polynomial that vanishes on the A of the toric variety YA . If Y ∗ dual variety Y ∗ A is a hypersurface then ∆A is the A-discriminant. First, we write ∆A as a Laurent polynomial by dividing it by one of its monomials: · ∆A = 1 − c0 · xb0 − c1 · xb1 − · · · − cn · xbn . 1 (3.8) monomial This expression defines an m × (n + 1) integer matrix B = (b0 , . . . , bn ) satisfying AB = 0. Second, we define X to be the rational subvariety of Pn that is given parametrically by = ci · xbi pi p0 + p1 + · · · + pn The defining ideal of X is obtained by eliminating x1 , . . . , xm from the equations above. Then X has ML degree 1, and, by [27], every variety of ML degree 1 arises in this manner. (cid:1). X = V (cid:0) 9p1p2 − 8p0p3 , p2 Example 3.11. The following curve in P3 happens to be a variety of ML degree 1: 0 − 12(p0+p1+p2+p3 )p3 (cid:1) − (cid:0) 1 (cid:1) − (cid:0)− 4 (cid:1) − (cid:0)− 4 · ∆A = 1 − (cid:0) 2 This curve comes from the discriminant of the univariate cubic in Example 3.9: 3 27 27 27 29 1 monomial for i = 0, 1, . . . , n. (cid:1). x2 2x2 3 1x2 x2 4 (3.9) x2x3 x1x4 x3 2 x2 1x4 x3 3 x1x2 4 p0 = where p3 = 1 27 2x2 x2 3 1x2 x2 4 bij ui )bkj . 2 3 x2x3 x1x4 x3 3 x1x2 4 We derived the curve X from the four parenthesized monomials via the formula (3.9). The maximum likelihood estimate for this model is given by the products of linear forms p2 = − 4 p1 = − 4 27 27 x3 2 x2 1x4 x1 = −u0 − u1 − 2u2 − 2u3 x2 = u0 + 3u2 + 2u3 x4 = −u0 − 2u1 − u2 − 2u3 x3 = u0 + 3u1 + 2u3 These expressions are the alternating products of linear forms promised in Theorem 3.10. ♦ We now give the formula for pi in general. This is the Horn uniformization of [19, §9.3]. Corollary 3.12. Let X ⊂ Pn be the variety of ML degree 1 with parametrization (3.9) derived from a scaled A-discriminant (3.8). The coordinates of the MLE function u (cid:55)→ p are pk = ck · m(cid:89) n(cid:88) ( i=0 j=1 It is not obvious (but true) that p0 + p1 + · · · + pn = 1 holds in the formula above. In light of its monomial parametrization, our variety X is toric in Pn\H. In general, it is not toric in Pn , due to appearances of the factor (p0 + p1 + · · · + pn ) in equations for X . Interestingly, there are numerous instances when this factor does not appear and X is toric also in Pn . One toric instance is the independence model X = V (p00p11 − p01p10 ), whose MLE was derived in Example 1.14. What is the matrix A in this case? We shall answer this question for a slightly larger example, which serves as an illustration for decomposable graphical models. Example 3.13. Consider the conditional independence model for three binary variables given by the graph •—–•—–•. We claim that this graphical model is derived from  . d c1 c0 b11 b10 b01 b00 a00 a10 a01 a11 1 1 1 1 1 1 1 1 1 1 1 0 0 1 0 0 0 0 0 0 1 1 x 0 1 0 0 0 0 0 1 1 0 0 y 0 0 1 0 0 1 1 0 0 0 0 z 0 1 0 1 1 0 0 0 0 0 w 0 (cid:8)(x, y , z , w) ∈ (C∗ )4 (a00+a10 )x+(a01+a11 )y+(b00+b01 )z+(b10+b11 )w+c0xz+c1yw+d = 0(cid:9) The discriminant of the corresponding family of hypersurfaces equals ∆A = c0c1d − a01 b10c0 − a11 b10c0 − a01b11c0 − a11b11c0 −a00 b00c1 − a10 b00c1 − a00 b01c1 − a10b01c1 . We divide this A-discriminant by its first term c0c1d to rewrite it in the form (3.8) with n = 7. The parametrization of X ⊂ P7 given by (3.9) can be expressed as aij · bj k for i, j, k ∈ {0, 1}. cj · d A = pij k = (3.10) 30 (cid:1) ⊂ P7 . X = V (cid:0)p000p101 − p001p100 , p010p111 − p011p110 This is indeed the desired graphical model •—–•—–• with implicit representation The linear forms used in the Horn uniformization of Corollary 3.12 are aij = uij+ bj k = u+j k cj = u+j+ d = u+++ Substituting these expressions into (3.10), we obtain uij+ · u+j k for i, j, k ∈ {0, 1}. u+j+ · u+++ This is the formula in Lauritzen’s book [32] for MLE of decomposable graphical models. ♦ pij k = We now return to the likelihood geometry of an arbitrary d-dimensional pro jective variety X in Pn , as always defined over R and not contained in H. We define the ML bidegree of X to be the bidegree of its likelihood correspondence LX ⊂ Pn × Pn . This is a binary form BX (p, u) = (b0 · pd + b1 · pd−1u + · · · + bd · ud ) · pn−d , where b0 , b1 , . . . , bd are certain positive integers. By definition, BX (p, u) is the multidegree [33, §8.5] of the prime ideal of LX , with respect to the natural Z2 -grading on the polynomial ring R[p, u] = R[p0 , . . . , pn , u0 , . . . , un ]. Equivalently, the ML bidegree BX (p, u) is the class defined by LX in the cohomology ring H ∗ (Pn × Pn ; Z) = Z[p, u]/(cid:104)pn+1 , un+1 (cid:105). We already saw some examples, for the Grassmannian G(2, 4) in (1.12), for arbitrary linear spaces in (1.14), and for a toric model of ML degree 1 in (2.6). We note that the bidegree BX (p, u) can be computed conveniently using the command multidegree in Macaulay2. To understand the geometric meaning of the ML bidegree, we introduce a second poly- nomial. Let Ln−i be a sufficiently general linear subspace of Pn of codimension i, and define si = MLdegree (X ∩ Ln−i ). We define the sectional ML degree of X to be the polynomial SX (p, u) = (s0 · pd + s1 · pd−1u + · · · + sd · ud ) · pn−d , Example 3.14. The sectional ML degree of the Grassmannian G(2, 4) in (1.10) equals SX (p, u) = 4p5 + 20p4u + 24p3u2 + 12p2u3 + 2pu4 . Thus, if H1 , H2 , H3 denote generic hyperplanes in P5 , then the threefold G(2, 4) ∩ H1 has ML degree 20, the surface G(2, 4)∩H1 ∩H2 has ML degree 24, and the curve G(2, 4)∩H1 ∩H2 ∩H3 has ML degree 12. Lastly, the coefficient 2 of pu4 is simply the degree of G(2, 4) in P5 . ♦ 31 BX (p, u) = Conjecture 3.15. The ML bidegree and the sectional ML degree of any projective variety X ⊂ Pn , not lying in H, are related by the fol lowing involution on binary forms of degree n: u · SX (p, u − p) − p · SX (p, 0) u − p , u · BX (p, u + p) + p · BX (p, 0) SX (p, u) = . u + p This conjecture is a theorem when X \H is smooth and its boundary is schon. See Theorem 4.6 below. In that case, the ML bidegree is identified, by [25, Theorem 2], with the Chern-Schwartz-MacPherson (CSM) class of the constructible function on Pn that is 1 on X \H and 0 elsewhere. Aluffi proved in [4, Theorem 1.1] that the CSM class of an locally closed subset of Pn satisfies such a log-adjunction formula. Our formula in Conjecture 3.15 is precisely the homogenization of Aluffi’s involution. The combination of [4, Theorem 1.1] and [25, Theorem 2] proves Conjecture 3.15 in cases such as generic complete intersections (Theorem 1.10) and arbitrary linear spaces (Theorem 1.20). In the latter case, it can also be verified using matroid theory. Conjecture 3.15 says that this holds for any X , indicating a deeper connection between likelihood correspondences and CSM classes. We note that BX (p, u) and SX (p, u) always share the same leading term and the same trailing term, and this is compatible with our formulas. Both polynomials start and end like MLdegree (X ) · pn + · · · + degree (X ) · pcodim(X )udim(X ) . We now illustrate Conjecture 3.15 by verifying it computationally for a few more examples. Example 3.16. Let us examine some cubic fourfolds in P5 . If X is a generic hypersurface of degree 4 in P5 then its sectional ML degree and ML bidegree satisfy the conjectured formula: SX (p, u) = 1364p5 + 448p4u + 136p3u2 + 32p2u3 + 3pu4 , BX (p, u) = 1364p5 + 341p4u + 81p3u2 + 23p2u3 + 3pu4 . Of course, in algebraic statistics, we are more interested in special hypersurfaces that are statistically meaningful. One such instance was seen in Example 2.7. The mixture model 2p11  = 0. for two identically distributed ternary random variables is the fourfold X ⊂ P5 defined by p12 p13 The sectional ML degree and the ML bidegree of this determinantal fourfold are p13 p23 2p33 (3.11) det p12 2p22 p23 SX (p, u) = 6p5 + 42p4u + 48p3u2 + 21p2u3 + 3pu4 BX (p, u) = 6p5 + 12p4u + 15p3u2 + 12p2u3 + 3pu4 . For the toric fourfold X = V (p11p22p33 − p12p13p23 ), ML bidegree and sectional ML degree are BX (p, u) = 3p5 + 3p4u + 3p3u2 + 3p2u3 + 3pu4 , SX (p, u) = 3p5 + 12p4u + 18p3u2 + 12p2u3 + 3pu4 . Now, taking X = V (p11p22p33+p12p13p23 ) instead, the leading coefficient 3 changes to 2. ♦ 32 Remark 3.17. Conjecture 3.15 is true when Xc is a toric variety with c generic, as in Theorem 3.2. Here we can use Proposition 3.5 to infer that all coefficients of BX are equal d(cid:88) to the normalized volume of the lattice polytope conv(A). In symbols, for generic c, we have i=0 BXc (p, u) = degree (Xc ) · pn−iui . It is now an exercise to transform this into a formula for the sectional ML degree SXc (p, u). In general, it is hard to compute generators for the ideal of the likelihood correspondence. 12p0 3p1  = 0. Example 3.18. The following submodel of (3.11) was featured prominently in [23, §1]: 2p2 3p3 3p1 2p2 3p3 12p4 2p2 This cubic threefold X is the secant variety of a rational normal curve in P4 , and it represents the mixture model for a binomial random variable (tossing a biased coin four times). It takes several hours in Macaulay2 to compute the prime ideal of the likelihood correspondence LX ⊂ P4 × P4 . That ideal has 20 minimal generators one in degree (1, 1), one in degree (3, 0), five in degree (3, 1), ten in degree (4, 1) and three in degree (3, 2). After passing to a Grobner basis, we use the formula in [33, Definition 8.45] to compute the bidegree of LX : (3.12) det BX (p, u) = 12p4 + 15p3u + 12p2u2 + 3pu3 . We now intersect X with random hyperplanes in P4 , and we compute the ML degrees of the intersections. Repeating this experiment many times reveals the sectional ML degree of X : SX (p, u) = 12p4 + 30p3u + 18p2u2 + 3pu3 . The two polynomials satisfy our transformation rule, thus confirming Conjecture 3.15. We note that Conjecture 1.8 also holds for this example: using Aluffi’s method [2], we find χ(X \H) = −13. ♦ Our last topic is the operation of restriction and deletion. This is a standard tool for complements of hyperplane arrangements, as in Theorem 1.20. It was developed in [25] for arbitrary very affine varieties, such as X \H. We motivate this by explaining the distinction between structural zeros and sampling zeros for contingency tables in statistics [8, §5.1.1]. Returning to the “hair loss due to TV soccer” example from the beginning of Section 2, let us consider the following questions. What is the difference between the data set   lots of hair medium hair 15 0 20 24 10 12 little hair 9 12 6 ≤ 2 hrs 2–6 hrs ≥ 6 hrs U = 33 and the data set U = ≤ 2 hrs 2–6 hrs ≥ 6 hrs little hair 5 6 8  lots of hair medium hair  ? 10 0 9 3 9 7 How should we think about the zero entries in row 1 and column 2 of these two contingency tables? Would the rank 1 model M1 or the rank 2 model M2 be more appropriate? The first matrix U has rank 2 and it can be completed to a rank 1 matrix by replacing the zero entry with 18. Thus, the model M1 fits perfectly except for the structural zero in row 1 and column 2. It seems that this zero is inherent in the structure of the problem: planet Earth simply has no people with medium hair length who rarely watch soccer on TV. The second matrix U also has rank two, but it cannot be completed to rank 1. The model M2 is a perfect fit. The zero entry in U appeared to be an artifact of the particular group that was interviewed in this study. This is a sampling zero. It arose because, by chance, in this cohort nobody happened to have medium hair length and watch soccer on TV rarely. We refer to the book of Bishop, Feinberg and Holland [8, Chapter 5] for an introduction. We now consider an arbitrary pro jective variety X ⊆ Pn , serving as our statistical model. Suppose that structural zeros or sampling zeros occur in the last coordinate un . Following [39, Theorem 4], we model structural zeros by the pro jection πn (X ). This model is the variety in Pn−1 that is the closure of the image of X under the rational map (p0 : p1 : · · · : pn−1 : pn ) (cid:55)−→ (p0 : p1 : · · · : pn−1 ). πn : Pn (cid:57)(cid:57)(cid:75) Pn−1 , Which pro jective variety is a good representation for sampling zeros? We propose that sampling zeros be modeled by the intersection X ∩ {pn=0}. This is now to be regarded as a subvariety in Pn−1 . In this manner, both structural zeros and sampling zeros are modeled by closed subvarieties of Pn−1 . Inside that ambient Pn−1 , our standard arrangement H consists of n + 1 hyperplanes. Usually, none of these hyperplanes contains X ∩ {pn=0} or πn (X ). It would be desirable to express the (sectional) ML degree of X in terms of those of the intersection X ∩ {pn = 0} and the pro jection πn (X ). As an alternative to the ML degree of the pro jection πn (X ) into Pn−1 , here is a quantity in Pn that reflects the presence of structural zeros even more accurately. We denote by MLdegree (X un=0 ) the number of critical points p = ( p0 : p1 : · · · : pn−1 : pn ) of (cid:96)u in Xreg \H for those data vectors u = (u0 , u1 , . . . , un−1 , 0) whose first n coordinates ui are positive and generic. Conjecture 3.19. The maximum likelihood degree satisfies the inductive formula MLdegree (X ) = MLdegree (X ∩ {pn=0}) + MLdegree (X un=0 ), (3.13) provided X and X ∩ {pn=0} are reduced, irreducible, and not contained in their respective H. We expect that an analogous formula will hold for the sectional ML degree SX (p, u). The intuition behind equation (3.13) is as follows. As the data vector u moves from a general 34 u to a general point on the hyperplane {un = 0}, the corresponding fiber pr−1 point in Pn 2 (u) of the likelihood fibration splits into two clusters. One cluster has size MLdegree (X un=0 ) and stays away from H. The other cluster moves onto the hyperplane {pn = 0} in Pn p , where it approaches the various critical points of (cid:96)u in that intersection. This degeneration is the perfect scenario for a numerical homotopy, e.g. in Bertini, as discussed in Section 2. These homotopies are currently being studied for determinantal varieties by Elizabeth Gross and Jose Rodriguez [20]. The formula (3.13) has been verified computationally for many examples. Also, Conjecture 3.19 is known to be true in the slightly different setting of [25], under a certain smoothness assumption. This is the content of [25, Corollary 3.2]. Example 3.20. Fix the space P8 of 3 × 3-matrices as in §2. For the rank 2 variety X = V2 , the formula (3.13) reads 10 = 5 + 5. For the rank 1 variety X = V1 , it reads 1 = 0 + 1. ♦ Example 3.21. If X is a generic (d, e)-curve in P3 , then MLdegree (X ) = d2e + de2 + de and X ∩ {p3 = 0} = (d · e distinct points). Computations suggest that MLdegree (X u3=0 ) = d2e + de2 and MLdegree (π3 (X )) = d2e + de2 . To derive the second equality geometrically, one may argue as follows. Both curves X ⊂ P3 2 (d2e + de2 ) − 2de + 1. Subtracting this from and π3 (X ) ⊂ P2 have degree de and genus 1 2 (de − 1)(de − 2) of a plane curve of degree de, we find that π3 (X ) has the expected genus 1 2 d(d − 1)e(e − 1) nodes. Example 1.4 suggests that each node decreases the ML degree of a 1 plane curve by 2. Assuming this to bet the case, we conclude MLdegree (π3 (X )) = de(de + 1) − d(d − 1)e(e − 1) = d2e + de2 . Here we are using that a general plane curve of degree de has ML degree de(de + 1). ♦ This example suggests that, in favorable circumstances, the following identity would hold: MLdegree (X un=0 ) = MLdegree (πn (X )). (3.14) However, this is certainly not true in general. Here is a particularly telling example: Example 3.22. Suppose that X is a generic surface of degree d in P3 . Then = d + d2 + d3 , MLdegree (X ) MLdegree (X ∩ {p3 = 0}) = d + d2 , MLdegree (X u3=0 ) d3 , = 1. = MLdegree (π3 (X )) Indeed, for most hypersurfaces X ⊂ Pn , the same will happen, since πn (X ) = Pn−1 . ♦ As a next step, one might conjecture that (3.14) holds when the map is birational and the center (0 : · · · : 0 : 1) of the pro jection does not lie on the variety X . But this also fails: 35 (cid:19) (cid:18) p0 + p1 − p2 Example 3.23. Let X be the twisted cubic curve in P3 defined by the 2 × 2-minors of p0 − 6p1 + 8p2 2p0 − p2 + 9p3 2p0 − p2 + 9p3 p0 − 6p1 + 8p2 . 7p0 + p1 + 2p2 The ML degree of X is 13 = 3 + 10, and X intersects {p3 = 0} in three distinct points. The pro jection of the curve X into P2 is a cuspidal cubic, as in Example 1.4. We have MLdegree (X u3=0 ) = 10 and MLdegree (π3 (X )) = 9. It is also instructive to compare the number 13 = −χ(X \H) with the number 11 one gets in Theorem 3.7 for the special twisted cubic curve with d = 1, n = 3 and b0 = b1 = b2 = b3 = 3. ♦ There are many mysteries still to be explored in likelihood geometry, even within P3 . 4 Characteristic Classes We start by giving an alternative description of the likelihood correspondence which reveals its intimate connection with the theory of Chern classes on possibly noncompact varieties. An important role will be played by the Lie algebra and cotangent bundle of the algebraic torus (C∗ )n+1 . This section ties our discussion to the work of Aluffi [2, 3, 4] and Huh [25, 26, 27]. In particular, we introduce and explain Chern-Schwartz-MacPherson (CSM) classes. And, most importantly, we present proofs for Theorems 1.6, 1.7, 1.15, and 1.20. Let X ⊆ Pn be a closed and irreducible subvariety of dimension d, not contained in our n(cid:88) H = (cid:8)(p0 : p1 : · · · : pn ) ∈ Pn p0 · p1 · · · pn · p+ = 0 }, distinguished arrangement of n + 2 hyperplanes, i=0 Let ϕi denote the restriction of the rational function pi/p+ to X \H. The closed embedding ϕ : X \H −→ (C∗ )n+1 , ϕ = (ϕ0 , . . . , ϕn ), shows that the variety X \H is very affine. Let x be a smooth point of X \H. We define γx : TxX −→ Tϕ(x) (C∗ )n+1 −→ g := T1 (C∗ )n+1 (4.1) to be the derivative of ϕ at x followed by that of left-translation by ϕ(x)−1 . Here g is the Lie algebra of the algebraic torus (C∗ )n+1 . In local coordinates (x1 , . . . , xd ) around the smooth (cid:33) (cid:32) point x, the linear map γx is represented by the logarithmic Jacobian matrix p+ = pi . ∂ log ϕi ∂xj 0 ≤ i ≤ n, 1 ≤ j ≤ d. , The linear map γx in (4.1) is injective because ϕ is injective. We write q0 , . . . , qn for the coordinate functions on the torus (C∗ )n+1 . These functions define a C-linear basis of the dlog(q0 ), . . . , dlog(qn ) ∈ H 0(cid:16) (cid:17) (cid:39) g∨ (cid:39) Cn+1 . dual Lie algebra g∨ corresponding to differential forms (C∗ )n+1 , Ω1 (C∗ )n+1 36 (4.2) ui · dlog(ϕi )(x). ui · dlog(qi ) u = (u0 , . . . , un ) ∈ Cn+1(cid:111) g∨ (cid:39) (cid:110) n(cid:88) We fix this choice of basis of g∨ , and we identify P(g∨ ) with the space of data vectors Pn u : . i=0 (x, u) (cid:55)−→ n(cid:88) Consider the vector bundle homomorphism defined by the pullback of differential forms Xreg \H −→ Ω1 γ ∨ : g∨ Xreg \H , i=0 Xreg \H is the trivial vector bundle over Xreg\H modeled on the vector space g∨ . The Here g∨ induced linear map γ ∨ x between the fibers over a smooth point x is dual to the injective linear map γx : TxX −→ g. Therefore γ ∨ is surjective and ker(γ ∨ ) is a vector bundle over Xreg \H. This vector bundle has positive rank n − d + 1, and hence its pro jectivization is nonempty. Proof of Theorem 1.6. Under the identification P(g∨ ) (cid:39) Pn u , the pro jective bundle P(ker γ ∨ ) LX ∩ (cid:16) (cid:17) ⊆ Pn corresponds to the following constructible subset of dimension n: p × Pn (Xreg \H) × Pn u . u Therefore its Zariski closure LX is irreducible of dimension n, and pr1 : LX → Pn p is a pro jective bundle over Xreg \H. The likelihood vibration pr2 : LX → Pn u is generically finite- to-one because the domain and the range are algebraic varieties of the same dimension. (cid:101)X Our next aim is to prove Theorem 1.15. For this we fix a resolution of singularities π−1 (Xreg \H) π Xreg \H / Pn , / X where π is an isomorphism over Xreg \H, the variety (cid:101)X is smooth and pro jective, and the complement of π−1 (Xreg \H) is a simple normal crossing divisor in (cid:101)X with irreducible com- ponents D1 , . . . , Dk . Each ϕi lifts to a rational function on (cid:101)X which is regular on π−1 (X \H). If u = (u0 , . . . , un ) is an integer vector in Zn+1 , then these functions satisfy n(cid:88) i=0 If u ∈ Cn+1\Zn+1 then ordDj ((cid:96)u ) is the complex number defined by the equation (4.3) for j = 1, . . . , k . We write Hi := {pi = 0} and H+ := {p+ = 0} for the n + 2 hyperplanes in H. boundary of (cid:101)X such that π(Dj ) ⊆ H. Then the fol lowing three statements hold: Lemma 4.1. Suppose that X ∩ Hi is smooth along H+ , and let Dj be a divisor in the (cid:40) if π(Dj ) ⊆ Hi , positive 1. If π(Dj ) (cid:42) H+ then ordDj (ϕi ) is if π(Dj ) (cid:42) Hi . zero ordDj ((cid:96)u ) = ui · ordDj (ϕi ). (4.3) 37 / /     / / 2. If π(Dj ) ⊆ H+ then −ordDj (ϕi ) is (cid:40) positive nonnegative if π(Dj ) (cid:42) Hi , if π(Dj ) ⊆ Hi . 3. In each of the above two cases, ordDj (ϕi ) is non-zero for at least one index i. i and H (cid:48) Proof. Write H (cid:48) + for the pullbacks of Hi and H+ to X respectively. Note that i )) is positive if Dj is contained in π−1 (H (cid:48) ordDj (π ∗ (H (cid:48) i ) and otherwise zero. Since i )) − ordDj (π ∗ (H (cid:48) ordDj (ϕi ) = ordDj (π ∗ (H (cid:48) + )), this proves the first and second assertion, except for the case when π(Dj ) ⊆ Hi ∩ H+ . In this + shows that π(Dj ) ⊆ Xreg and the order of case, our assumption that H (cid:48) i is smooth along H (cid:48) vanishing of H (cid:48) i along π(Dj ) is 1. Therefore −ordDj (ϕi ) = ordDj (π ∗ (H (cid:48) + )) − 1 ≥ 0. The third assertion of Lemma 4.1 is derived by the following set-theoretic reasoning: • If π(Dj ) ⊆ H+ , then π(Dj ) (cid:42) Hi for some i because (cid:84)n • If π(Dj ) (cid:42) H+ , then π(Dj ) ⊆ Hi for some i because π(Dj ) ⊆ H is irreducible. i=0 Hi = ∅. From Lemma 4.1 and equation (4.3) we deduce the following result. In Lemmas 4.2 and 4.3 we retain the hypothesis from Lemma 4.1 which coincides with that in Theorem 1.15. Lemma 4.2. If π(Dj ) ⊆ H and u ∈ Rn+1 is strictly positive, then ordDj ((cid:96)u ) is nonzero. >0 Consider the sheaf of logarithmic differential 1-forms Ω1(cid:101)X (log D), where D is the sum likelihood function (cid:96)u on (cid:101)X defines a global section of this sheaf: of the irreducible components of π−1 (H). If u is an integer vector, then the corresponding n(cid:88) ui · dlog(ϕi ) ∈ H 0(cid:0) (cid:101)X , Ω1(cid:101)X (log D)(cid:1). dlog((cid:96)u ) = i=0 If u ∈ Cn+1\Zn+1 then we define the global section dlog((cid:96)u ) by the above expression (4.4). Lemma 4.3. If u ∈ Rn+1 is strictly positive, then dlog((cid:96)u ) does not vanish on π−1 (H). >0 Proof. Let x ∈ π−1 (H) and D1 , . . . , Dl the irreducible components of D containing x, with local equations g1 , . . . , gl on a small neighborhood G of x. Clearly, l ≥ 1. By passing to a smaller neighborhood if necessary, we may assume that Ω1(cid:101)X (log D) trivializes over G, and l(cid:88) ordDj ((cid:96)u ) · dlog(gj ) + ψ , j=1 dlog((cid:96)u ) = (4.4) where ψ is a regular 1-form. Since the dlog(gj ) form part of a free basis of a trivialization of Ω1(cid:101)X (log D) over G, Lemma 4.2 implies that dlog((cid:96)u ) is nonzero on π−1 (H) if u ∈ Rn+1 >0 . 38 cd (cid:90) (cid:0)Ω1(cid:101)X (log D)(cid:1) = (−1)d · χ(cid:0) (cid:101)X \π−1 (H)(cid:1). Proof of Theorem 1.7. In the notation above, the logarithmic Poincar´e-Hopf theorem states (cid:101)X See [3, Section 3.4] for example. If X \H is smooth, then Lemma 4.3 shows that, for generic (cid:8)x ∈ X \H dlog((cid:96)u )(x) = 0(cid:9). u, the zero-scheme of the section (4.4) is equal to the likelihood locus Since the likelihood locus is a zero-dimensional scheme of length equal to the ML degree of X , the logarithmic Poincar´e-Hopf theorem implies Theorem 1.7. contains a curve. Let C and (cid:101)C denote the closures of that curve in X and (cid:101)X respectively. Let Proof of Theorem 1.15. Suppose that the likelihood locus {x ∈ Xreg \H dlog((cid:96)u )(x) = 0} π ∗ (H) · (cid:101)C is rationally equivalent to zero in (cid:101)X . It then follows from the Pro jection Formula π ∗ (H) be the pullback of the divisor H ∩ X of X . If u ∈ Rn+1 then Lemma 4.3 implies that >0 that H · C is also rationally equivalent to zero in Pn . But this is impossible. Therefore the likelihood locus does not contain a curve. This proves the first part of Theorem 1.15. 2 (u) is contained in X \H for a strictly positive For the second part, we first show that pr−1 vector u. This means there is no pair (x, u) ∈ LX with x ∈ H which is a limit of the form xt ∈ Xreg \H, (x, u) = lim (xt , ut ), dlog((cid:96)ut )(xt ) = 0. If there is such a sequence (xt , ut ), then we can take its limit over (cid:101)X to find a point (cid:101)x ∈ (cid:101)X t→0 such that dlog((cid:96)u )((cid:101)x) = 0, but this would contradict Lemma 4.3. 2 (u) is contained in Xreg , and hence in Xreg \H. By Now suppose that the fiber pr−1 2 (u) is contained the smooth variety (LX )reg . Furthermore, by Theorem 1.6, this fiber pr−1 2 (u) is a zero-dimensional subscheme of (LX )reg . The the first part of Theorem 1.15, pr−1 assertion on the length of the fiber now follows from a standard result on intersection theory on Cohen-Macaulay varieties. More precisely, we have MLdegree(X ) = (U1 · . . . · Un )LX = (U1 · . . . · Un )(LX )reg = deg(pr−1 2 (u)), where the Ui are pullbacks of sufficiently general hyperplanes in Pn u containing u, and the two terms in the middle are the intersection numbers defined in [17, Definition 2.4.2]. The fact that (LX )reg is Cohen-Macaulay is used in the last equality [17, Example 2.4.8]. Remark 4.4. If X is a curve, then the zero-scheme of the section (4.4) is zero-dimensional for generic u, even if X \H is singular. Furthermore, the length of this zero-scheme is at least −χ(X \H) ≥ −χ(cid:0) (cid:101)X \π−1 (H)(cid:1) ≥ MLdegree (X ). as large as ML degree of X . Therefore This proves that Conjecture 1.8 holds for d = 1. 39 Next we give a brief description of the Chern-Schwartz-MacPherson (CSM) class. For a gentle introduction we refer to [3]. The group C (X ) of constructible functions on a complex algebraic variety X is a subgroup of the group of integer valued functions on X . It is generated by the characteristic functions 1Z of all closed subvarieties Z of X . If f : X → Y 1Z (cid:55)−→ (cid:16) (cid:17) is a morphism between complex algebraic varieties, then the pushforward of constructible y (cid:55)−→ χ(cid:0)f −1 (y) ∩ Z (cid:1), functions is the homomorphism f∗ : C (X ) −→ C (Y ), If X is a compact complex manifold, then the characteristic class of X is the Chern class of the tangent bundle c(T X ) ∩ [X ] ∈ H∗ (X ; Z). A generalization to possibly singular or noncompact varieties is provided by the Chern-Schwartz-MacPherson class, whose existence was once a conjecture of Deligne and Grothendieck. In the next definition, we write C for the functor of constructible functions from the category of complete complex algebraic varieties to the category of abelian groups. y ∈ Y . Definition 4.5. The CSM class is the unique natural transformation cSM : C −→ H∗ such that cSM (1X ) = c(T X ) ∩ [X ] ∈ H∗ (X ; Z) when X is smooth and complete. The uniqueness follows from the naturality, the resolution of singularities over C, and the requirement for smooth and complete varieties. We highlight two properties of the CSM class which follow directly from Definition 4.5: (4.5) (4.6) χ(U ) = 1. The CSM class satisfies the inclusion-exclusion relation cSM (1U ∪U (cid:48) ) = cSM (1U ) + cSM (1U (cid:48) ) − cSM (1U ∩U (cid:48) ) ∈ H∗ (X ; Z). (cid:90) 2. The CSM class captures the topological Euler characteristic as its degree: cSM (1U ) ∈ Z. X Here U and U (cid:48) are arbitrary constructible subsets of a complete variety X . What kind of information on a constructible subset is encoded in its CSM class? In likelihood geometry, U is a constructible subset in the complex pro jective space Pn , and we identify cSM (1U ) with its image in H∗ (Pn , Z) = Z[p]/(cid:104)pn+1 (cid:105). Thus cSM (1U ) is a polynomial of degree ≤ n is one variable p. To be consistent with the earlier sections, we introduce a homogenizing variable u, and we write cSM (1U ) as a binary form of degree n in (p, u). n(cid:88) The CSM class of U carries the same information as the sectional Euler characteristic χ(U ∩ Ln−i ) · pn−iui . i=0 Here Ln−i is a generic linear subspace of codimension i in Pn . Indeed, it was proved by Aluffi in [4, Theorem 1.1] that cSM (1U ) is the transform of χsec (1U ) under a linear involution on binary forms of degree n in (p, u). In fact, our involution in Conjecture 3.15 is nothing but the signed version of the Aluffi’s involution. This is explained by the following result. χsec (1U ) = 40 Theorem 4.6. Let X ⊂ Pn be closed subvariety of dimension d that is not contained in H. If the very affine variety X \H is schon then, up to signs, the ML bidegree equals the CSM class and the sectional ML degree equals the sectional Euler characteristic. In symbols, cSM (1X \H ) = (−1)n−d · BX (−p, u) and χsec (1X \H ) = (−1)n−d · SX (−p, u). Proof. The first identity is a special case of [25, Theorem 2], here adapted to Pn minus n + 2 hyperplanes, and the second identity follows from the first by way of [4, Theorem 1.1]. To make sense of the statement in Theorem 4.6, we need to recall the definition of schon. This term was coined by Tevelev in his study of tropical compactifications [42]. Let U be an arbitrary closed subvariety of the algebraic torus (C∗ )n+1 . In our application, U = X \H. We consider the closures U of U in various (not necessarily complete) normal toric varieties Y with dense torus (C∗ )n+1 . The closure U is complete if and only if the support of the fan of Y contains the tropicalization of U [42, Proposition 2.3]. We say that U is a tropical compactification of U if it is complete and the multiplication map m : (C∗ )n+1 × U −→ Y , (t, x) (cid:55)−→ t · x is flat and surjective. Tropical compactifications exist, and they are obtained from toric varieties Y defined by sufficiently fine fan structures on the tropicalization of U [42, §2]. The very affine variety U is called schon if the multiplication is smooth for some tropical compactification of U . Equivalently, U is schon if the multiplication is smooth for every tropical compactification of U , by [42, Theorem 1.4]. Two classes of schon very affine varieties are of particular interest. The first is the class of complements of essential hyperplane arrangements. The second is the class of nondegenerate hypersurfaces. What we need from the schon hypothesis is the existence of a simple normal crossings compactification which admits sufficiently many differential one-forms which have logarithmic singularities along the boundary. For complements of hyperplane arrangements, such a compactification is provided by the wonderful compactification of De Concini and Pro- cesi [12]. For nondegenerate hypersurfaces, and more generally for nondegenerate complete intersections, the needed compactification has been constructed by Khovanskii [24]. We illustrate this in the setting of likelihood geometry by a d-dimensional linear subspace of X ⊂ Pn . The intersection of X with distinguished hyperplanes H of Pn is an arrangement of n + 2 hyperplanes in X (cid:39) Pd , defining a matroid M of rank d + 1 on n + 2 elements. Proposition 4.7. If X is a linear space of dimension d then the CSM class of X \H in Pn is d(cid:88) i=0 (−1)ihiud−ipn−d+i . cSM (1X \H ) = where the hi are the signed coefficients of the shifted characteristic polynomial in (1.15). Proof. This holds because the recursive formula for a triple of arrangement complements cSM (1U1 ) = cSM (1U − 1U0 ) = cSM (1U ) − cSM (1U0 ), 41 agrees with the usual deletion-restriction formula [35, Theorem 2.56]: χM1 (q + 1) = χM (q + 1) − χM0 (q + 1). Here our notation is as in [25, §3]. We now use induction on the number of hyperplanes. Proof of Theorem 1.20. The very affine variety X \H is schon when X is linear. Hence the asserted formula for the ML bidegree of X follows from Theorem 4.6 and Proposition 4.7. Rank constraints on matrices are important both in statistics and in algebraic geometry, and they provide a rich source of test cases for the theory developed here. We close our dis- cussion with the enumerative invariants of three hypersurfaces defined by 3× 3-determinants. It would be very interesting to compute these formulas for larger determinantal varieties. Example 4.8. We record the ML bidegree, the CSM class, the sectional ML degree, and the sectional Euler characteristic for three singular hypersurfaces seen earlier in this paper. These p × Pn examples were studied already in [23]. The classes we present are elements of H ∗ (Pn u ) and of H ∗ (Pn p ; Z) respectively, and they are written as binary forms in (p, u) as before. • The 3 × 3 determinantal hypersurface in P8 (Example 2.1) has 10p8 + 24p7u + 33p6u2 + 38p5u3 + 39p4u4 + 33p3u5 + 12p2u6 + 3pu7 , BX (p, u) = cSM (1X \H ) = −11p8 + 26p7u − 37p6u2 + 44p5u3 − 45p4u4 + 33p3u5 − 12p2u6 + 3pu7 , SX (p, u) = 11p8 + 182p7u + 436p6u2+518p5u3+351p4u4+138p3u5+30p2u6+3pu7 , χsec (1X \H ) = −11p8 + 200p7u − 470p6u2+542p5u3−357p4u4+138p3u5−30p2u6+3pu7 . • The 3 × 3 symmetric determinantal hypersurface in P5 (Example 2.7) has BX (p, u) = 6p5 + 12p4u + 15p3u2 + 12p2u3 + 3pu4 , cSM (1X \H ) = 7p5 − 14p4u + 19p3u2 − 12p2u3 + 3pu4 , SX (p, u) = 6p5 + 42p4u + 48p3u2 + 21p2u3 + 3pu4 , χsec (1X \H ) = 7p5 − 48p4u + 52p3u2 − 21p2u3 + 3pu4 . • The secant variety of the rational normal curve in P4 (Example 3.18) has 12p4 + 15p3u + 12p2u2 + 3pu3 , BX (p, u) = cSM (1X \H ) = −13p4 + 19p3u − 12p2u2 + 3pu3 , 12p4 + 30p3u + 18p2u2 + 3pu3 , SX (p, u) = χsec (1X \H ) = −13p4 + 34p3u − 18p2u2 + 3pu3 . In all known examples, the coefficients of BX (p, u) are less than or equal to the absolute value of the corresponding coefficients of cSM (1X \H ), and similary for SX (p, u) and χsec (1X \H ). That this inequality holds for the first coefficient is Conjecture 1.8 which relates the ML degree of a singular X to the signed Euler characteristic of the very affine variety X \H. ♦ Acknowledgments: We thank Paolo Aluffi and Sam Payne for helpful communications, and the Mathematics Department at KAIST, Daejeon, for hosting both authors in May 2013. Bernd Sturmfels was supported by NSF (DMS-0968882) and DARPA (HR0011-12-1-0011). 42 References [1] E. Allmann, J. Rhodes, B. Sturmfels and P. Zwiernik: Tensors of nonnegative rank two, Linear Algebra and its Applications, Special Issue on Statistics, to appear. [2] P. Aluffi: Computing characteristic classes of projective schemes, Journal of Symbolic Computation 35 (2003) 3–19. [3] P. Aluffi: Characteristic classes of singular varieties, Topics in Cohomological Studies of Algebraic Varieties, 1–32, Trends in Mathematics, Birkhauser, Basel, 2005. [4] P. Aluffi: Euler characteristics of general linear sections and polynomial Chern classes, Rend. Circ. Mat. Palermo. 62 (2013) 3–26. [5] S. Amari and H. Nagaoka: Methods of Information Geometry, Translations of Mathe- matical Monographs 191, American Math. Society, 2000. [6] D.J. Bates, J.D. Hauenstein, A.J. Sommese, and C.W. Wampler: Bertini: Software for Numerical Algebraic Geometry, www.nd.edu/∼sommese/bertini, 2006. [7] D.J. Bates, J.D. Hauenstein, A.J. Sommese, and C.W. Wampler: Numerical ly Solving Polynomial Systems with the Software Package Bertini, to be published by SIAM, 2013. [8] Y. Bishop, S. Fienberg, and P. Holland: Discrete Multivariate Analysis: Theory and Practice, Springer, New York, 1975. [9] S. Boyd and L. Vandenberghe: Convex Optimization, Cambridge University Press, 2004. [10] F. Catanese, S. Ho¸sten, A. Khetan and B. Sturmfels: The maximum likelihood degree, American Journal of Mathematics 128 (2006) 671–697. [11] D. Cohen, G. Denham, M. Falk, and A. Varchenko: Critical points and resonance of hyperplane arrangements, Canadian Journal of Mathematics 63 (2011) 1038–1057. [12] C. De Concini and C. Procesi: Wonderful models of subspace arrangements, Selecta Mathematica. New Series 1 (1995) 459–494. [13] G. Denham, M. Garrousian, and M. Schulze: A geometric deletion-restriction formula, Advances in Mathematics 230 (2012) 1979–1994. [14] J. Draisma and J. Rodriguez: Maximum likelihood duality for determinantal varieties, to appear in International Mathematics Research Notices, arXiv:1211.3196. [15] M. Drton, B. Sturmfels and S. Sullivant: Lectures on Algebraic Statistics, Oberwolfach Seminars, Vol 39, Birkhauser, Basel, 2009. [16] J. Franecki and M. Kapranov, The Gauss map and a noncompact Riemann-Roch formula for constructible sheaves on semiabelian varieties, Duke Math. J. 104 (2000) 171–180. 43 [17] W. Fulton: Intersection Theory, Second edition. Ergebnisse der Mathematik und ihrer Grenzgebiete. A Series of Modern Surveys in Mathematics 2, Springer, Berlin, 1998. [18] O. Gabber and F. Loeser, Faisceaux pervers l-adiques sur un tore, Duke Math. J. 83 (1996) 501–606. [19] I.M. Gel’fand, M. Kapranov and A. Zelevinsky: Discriminants, Resultants, and Multi- dimensional Determinants, Birkhauser, Boston, 1994. [20] E. Gross and J. Rodriguez: Maximum likelihood geometry in the presence of structural and sampling zeros, in preparation. [21] P. Hacking: Homology of tropical varieties, Collectanea Mathem. 59 (2008) 263–273. [22] J. Hauenstein, J. Rodriguez and B. Sturmfels: Maximum likelihood for matrices with rank constraints, arXiv:1210.0198. [23] S. Ho¸sten, A. Khetan and B. Sturmfels: Solving the likelihood equations, Foundations of Computational Mathematics 5 (2005) 389–407. [24] A. Hovanskiı: Newton polyhedra and toroidal varieties, Akademija Nauk SSSR. Funkcional’nyi Analiz i ego Prilozenija 11 (1977) 56–64. [25] J. Huh: The maximum likelihood degree of a very affine variety, Compositio Math. 149 (2013) 1245–1266. [26] J. Huh: h-vectors of matroids and logarithmic concavity, preprint, arXiv:1201.2915. [27] J. Huh: Discriminants, Horn uniformization, and varieties with maximum likelihood degree one, arXiv:1301.2732. [28] M. Kapranov: A characterization of A-discriminantal hypersurfaces in terms of the logarithmic Gauss map, Math. Annalen 290 (1991) 277–285. [29] K. Kub jas, E. Robeva and B. Sturmfels: Nonnegative matrix rank and the EM algo- rithm, in preparation. [30] J.M. Landsberg: Tensors: Geometry and Applications, Graduate Studies in Mathemat- ics, 128. American Mathematical Society, Providence, RI, 2012. [31] J.M. Landsberg and L. Manivel: On ideals of secant varieties of Segre varieties, Found. Comput. Math. 4 (2004) 397–422. [32] S. Lauritzen: Graphical Models, Oxford University Press, 1996. [33] E. Miller and B. Sturmfels: Combinatorial Commutative Algebra, Graduate Texts in Mathematics, 227, Springer, New York, 2004. [34] D. Mond, J. Smith, and D. van Straten: Stochastic factorizations, sandwiched sim- plices and the topology of the space of explanations, R. Soc. Lond. Proc. Ser. A Math. Phys. Eng. Sci. 459 (2003) 2821–2845. 44 [35] P. Orlik and H. Terao: Arrangements of Hyperplanes, Grundlehren der Mathematischen Wissenschaften 300, Springer-Verlag, Berlin, 1992. [36] P. Orlik and H. Terao: The number of critical points of a product of powers of linear functions, Inventiones Mathematicae 120 (1995) 1–14. [37] L. Pachter and B. Sturmfels: Algebraic Statistics for Computational Biology, Cambridge University Press, 2005. [38] C. Raicu: Secant varieties of Segre–Veronese varieties, Algebra and Number Theory 6 (2012) 1817–1868. [39] F. Rapallo: Markov bases and structural zeros, J. Symbolic Comput. 41 (2006) 164–172. [40] R. Sanyal, B. Sturmfels and C. Vinzant: The entropic discriminant, Advances in Math- ematics 244 (2013) 678–707. [41] H. Terao, Generalized exponents of a free arrangement of hyperplanes and the Shepherd- Todd-Brieskorn formula, Invent. Math. 63 (1981) 159–179. [42] J. Tevelev: Compactifications of subvarieties of tori, American Journal of Mathematics 129 (2007) 1087–1104. [43] C. Uhler: Geometry of maximum likelihood estimation in Gaussian graphical models, Annals of Statistics 40 (2012) 238–261. [44] A. Varchenko: Critical points of the product of powers of linear functions and families of bases of singular vectors, Compositio Math. 97 (1995) 385–401. [45] S. Watanabe: Algebraic Geometry and Statistical Learning Theory, Monographs on Applied and Computational Mathematics 25, Cambridge University Press, 2009. Authors’ addresses: June Huh, Department of Mathematics, University of Michigan, Ann Arbor, MI 48109, USA, [email protected] Bernd Sturmfels, Department of Mathematics, University of California, Berkeley, CA 94720, USA, [email protected] 45
1706.07354
1
1706
2017-06-22T15:06:13
Second Chern numbers of vector bundles and higher adeles
[ "math.AG", "math.NT" ]
We give a construction of the second Chern number of a vector bundle over a smooth projective surface by means of adelic transition matrices for the vector bundle. The construction does not use an algebraic $K$-theory and depends on the canonical $\mathbb{Z}$-torsor of a locally linearly compact $k$-vector space. Analogs of certain auxiliary results for the case of an arithmetic surface are also discussed.
math.AG
math
Second Chern numbers of vector bundles and higher adeles D. V. Osipov Abstract We give a construction of the second Chern number of a vector bundle over a smooth projective surface by means of adelic transition matrices for the vector bundle. The construction does not use an algebraic K -theory and depends on the canonical Z -torsor of a locally linearly compact k -vector space. Analogs of certain auxiliary results for the case of an arithmetic surface are also discussed. 1 Introduction In [16] A.N. Parshin constructed Chern classes of vector bundles on a scheme Y which is finite type over the field Q using higher adeles. In particular, Chern classes, which he constructed, were in H m(Y, Ωm Y ) . Taking the higher residues when m = dim Y , we obtain the Chern numbers, see [16, § 4.3]. This construction can be carried out when Y is a scheme over any field k , but because of the higher residues the values of the Chern numbers of vector bundles will be in the image of the ring Z in the field k . Thus, if char k = p > 0 , then we will obtain the Chern numbers only modulo p . Much later there appeared adelic constructions of second Chern classes on certain two-dimensional regular schemes be means of K2 -groups. In particular, R.Ya. Budylin in [3] constructed the second Chern classes of vector bundles of rank 2 on a smooth algebraic surface Y over any perfect field using K2 -groups of rational adeles on Y . Besides, T. Chinburg, G. Pappas and M. J. Taylor gave in [5] a construction of the second Chern classes of vector bundles of arbitrary rank on a regular two-dimensional scheme Y with projective morphism of relative dimension 1 to the spectrum of a Dedekind ring by means of K2 -adeles on Y originated from [9]. In this paper we provide a quite elementary construction of the second Chern numbers of vector bundles on a smooth projective surface X over a perfect field k . This construc- tion does not use algebraic K -theory, but uses only Z -torsors and central extensions of a group GLn(AX) by the group Z , where AX is the adelic ring of X , which is also called the Parshin-Belinson adeles of X . More exactly, any locally linearly compact vector space over a field k gives a canonical Z -torsor of dimension theories. The adelic space AX has a filtration given by divisors This work is supported by the Russian Science Foundation under grant 14-50-00005. 1 on X with the quotient spaces being locally linearly compact vector spaces over k . The same is also true for An X for any integer n ≥ 0 . Therefore from the action of the group X we obtain a canonical central extensions ^GLn(AX) GLn(AX ) on the k -vector space An and then \GLn(AX) of this group by the group Z . The trivializations of a vector bundle at scheme points of X give transition matrices which are elements of GLn(AX) and satisfy the cocycle condition. Using canonical splittings of the central extension \GLn(AX) over certain subgroups of GLn(AX ) , we obtain lifts of these transition matrices to \GLn(AX) , where their product is an element over 1 ∈ GLn(AX) , i.e. it belongs to the subgroup Z . This is the second Chern number of the vector bundle, see Theorem 1. The advantage of our approach is similarity to the constructions from [15], where an "analytic" proof of the Riemann-Roch theorem for linear bundles on a smooth projective surface X over a finite field was given. One of the main ingredients in this proof was the definition of the intersection index of two divisors on X via the commutator of lifts of certain elements from the group A∗ X to a central extension which is similar to the central extension ^GL1(AX) . We note that the Noether formula was not obtained in [15]. Therefore one of the first expected applications of our construction of the second Chern numbers will be the proof of the Noether formula in the spirit of [15]. The next direction for the applications is the transfer of our constructions to the case of an arithmetic surface such that the fibres over Archimedean points of the base are taken into account. In particular, in the case of an arithmetic surface X over Spec Z and the adelic ring Aar X which includes an adelic object of the fibre over ∞ -point of Spec Z , we prove in this paper in Proposition 5 splittings of central extensions ^GLn(Aar X) and \GLn(Aar X) . These splittings are analogs of splittings considered above for the construction of second Chern number of a vector bundle over an algebraic surface. The central extensions ^GLn(Aar X) are central extensions by the group of positive real numbers R∗ X ) over certain subgroups of GLn(Aar X) and \GLn(Aar + and were also considered in [13]. The paper is organized as follows. In Section 2.1 we recall certain facts on the Parshin- Beilinson adeles of an algebraic surface X . In section 2.2 we recall the notion of Z -torsor of dimension theories for a locally linearly compact k -vector space. In section 2.3 we give a construction of the central extension ^GLn(A∆) , where A∆ is the adelic ring which depends on a subset ∆ of all pairs x ∈ C , where x is a point and C is an irreducible curve on X . In section 2.4 we connect the commutator of lifts of elements from A∗ X to ^GL1(AX ) with the intersection index of divisors on X by proving a result which was given without proof in [15], see Proposition 2. In section 3.1 we give a construction of the central extension \GLn(A∆) and prove some properties of this central extension, see Proposition 3. In section 3.2 we prove canonical splittings of the central extensions ^GLn(AX ) and \GLn(AX) over certain subgroups, see Proposition 4. In section 3.3 we give a construction of the second Chern number, see Theorem 1. In section 4 we prove certain results on splittings of central extensions in the case of an arithmetic surface, see Proposition 5. 2 2 Central extension and intersection index of divisors 2.1 Parshin-Beilinson adeles Let X be a smooth algebraic surface over a perfect field k . Let AX be the Parshin- Beilinson adelic ring of X (see, for example, a survey in [11]). X such that C contains x . Let Kx,C = Ql Let x ∈ C be a pair, where x is a point on X , and C is an irreducible curve on i=1 Ki , where an index i corresponds to a formal irreducible branch Ci of the curve C in the formal neighbourhood of x (i.e. Ci , where Ox,X is the completion of the local ring Ox,X of x on X ), and = C Spec Ox,X lSi=1 is a two-dimensional local field that is the completion of the fraction field Frac Ox,X Ki with respect to the discrete valuation given by Ci . We note that AX ⊂ Yx∈C Kx,C, where the product is over all pairs x ∈ C described as above. (1) Let ∆ be a subset in the set of all pairs x ∈ C described as above. There are the following subrings of the ring Qx∈C Kx,C : A∆ = AX ∩ Y{x∈C}∈∆ Kx,C, OA∆ = AX ∩ Y{x∈C}∈∆ OKx,C , (2) i=1 OKi , and OKi is the discrete valuation ring of the field Ki . Clearly, if ∆ is the set of all pairs x ∈ C , then A∆ = AX . Moreover, if ∆ is a single pair x ∈ C , then A∆ = Kx,C . where OKx,C =Ql Let D =Pi aiCi be a divisor on X . (Here ai ∈ Z and Ci is an irreducible curve on X for any i ). We call ai = νCi(D) for any i . We define OA∆(D) = AX ∩ Y{x∈C}∈∆ t−νC (D) C OKx,C , where tC = 0 is an equation of an irreducible curve C on some open subset of X . (The definition of OA∆(D) does not depend on the choice of tC .) We note (see [7, prop. 2.1.5]) that if ∆ = ∆1 ∪ ∆2 and ∆1 ∩ ∆2 = ∅ , then A∆ = A∆1 × A∆2, OA∆ = OA∆1 × OA∆2 . Hence we obtain for any integer n ≥ 1 GLn(A∆) = GLn(A∆1) × GLn(A∆2). (3) 2.2 Dimension theories Our first goal is to construct central extensions ^GLn(A∆) and \GLn(A∆) of the group GLn(A∆) by the group Z . These central extensions are similar to central extensions 3 + and \GLn(A∆)R∗ ^GLn(A∆)R∗ from [13, § 3]. (More close relation will be given in Section 4 below.) The main tool for this construction is a Z -torsor Dim of dimension theories on a locally linearly compact k -vector space V (or, in other words, on 1 -Tate k -vector space V ). This Z -torsor was defined by M. Kapranov in [8]. + We recall the definition of Dim(V ) . A dimension theory d on V is a map from the set of all open linearly compact k -subspaces of V to the group Z such that d(U2) = d(U1) + dimk(U2/U1) whenever U2 ⊃ U1 are two open linearly compact k -subspaces of V . (We note that dimk(U2/U1) < ∞ .) The set of all dimension theories on V is denoted by Dim(V ) . The group Z acts on Dim(V ) by adding constant maps. This makes Dim(V ) into a Z -torsor. We consider an exact sequence of k -vector spaces 0 −→ V1 −→ V2 −→ V3 −→ 0, (4) where Vi ( 1 ≤ i ≤ 3 ) are locally linearly compact k -vector spaces and all the maps in the above sequence are continuous. Besides, let V1 be a closed subspace of V2 , and topology on V3 coincides with the quotient topology. In this case, there is a canonical isomorphism Dim(V1) ⊗Z Dim(V3) −→ Dim(V2) (5) given as d1 ⊗ d3 7→ d2 , where d2(U) = d1(U ∩ V1) + d3(U/(U ∩ V1)) for a linearly compact subspace U of V2 . 2.3 Central extension ^GLn(A∆) By construction, A∆ = lim −→ D1 lim ←− D2≤D1 OA∆(D1)/OA∆(D2), and the k -vector space OA∆(D1)/OA∆(D2) is a locally linearly compact k -vector space for any divisors D2 ≤ D1 on X . Besides, for any divisors D1 ≥ D2 ≥ D3 on X the corresponding exact sequence 0 −→ OA∆(D2)/OA∆(D3) −→ OA∆(D1)/OA∆(D3) −→ OA∆(D1)/OA∆(D2) −→ 0 has the same properties as the exact sequence (4). This means that A∆ , and corre- spondingly An ∆ , is a complete C2 -vector space over k (or a 2 -Tate vector space over from GLn(A∆) such that k ) from [12]. In particular, for any elements g1 and g2 g1On A∆ we have that the k -vector space g2On A∆ is a locally linearly compact with the induced and quotient topology from a locally linearly compact k - vector space OA∆(D1)n/OA∆(D2)n for appropriate divisors D1 ≥ D2 on X . Therefore a Z -torsor A∆ ⊂ g2On A∆/g1On (6) (7) Dim(g1On A∆ g2On A∆) def= Dim(g2On A∆/g1On A∆) is well-defined. We define also Dim(g2On A∆ g1On A∆) def= Dim(g2On A∆/g1On A∆)∨, 4 where the sign ∨ means the dual Z -torsor. Now for any elements g1 and g2 from GLn(A∆) a Z -torsor Dim(g1On A∆) is canonically defined by the following prop- erty (using that there is an element g3 from GLn(A∆) such that g3On A∆ , where i = 1 and i = 2 ). For any elements g1, g2, g3 from GLn(A∆) there is a canonical isomorphism of Z -torsors A∆ ⊂ giOn A∆ g2On Dim(g1On A∆ g2On A∆) ⊗Z Dim(g2On A∆ g3On A∆) −→ Dim(g1On A∆ g3On A∆). (8) Any element g from GLn(A∆) defines an isomorphism of Z -torsors for any elements g1, g2 from GLn(A∆) : Dim(g1On A∆ g2On A∆) −→ Dim(gg1On A∆ gg2On A∆) , where d 7−→ g(d). We obtain a central extension 0 −→ Z −→ ^GLn(A∆) θ−→ GLn(A∆) −→ 1, (9) where the group ^GLn(A∆) is defined as the set of all pairs (g, d) , where g ∈ GLn(A∆) and d ∈ Dim(On A∆) , with the multiplication law given as A∆ gOn (g1, d1)(g2, d2) = (g1g2, d1 ⊗ g1(d2)), and θ((g, d)) = g . The following lemma is an important property which follows from the construction and formulas (3) and (5) (compare also with the proof of [13, Prop. 2]). Lemma 1. If ∆ = ∆1 ∪ ∆2 such that ∆1 ∩ ∆2 = ∅ , then the central extension ^GLn(A∆) is the Baer sum (i.e. it corresponds to the sum of 2 -cocycles) of central extensions ^GLn(A∆1) , where p1 and p2 are projections in decomposition (3). p∗ 1 ^GLn(A∆1) and p∗ 2 2.4 Commutator of the lift of elements and intersection index Using central extension (9) when n = 1 , for arbitrary elements f, g from A∗ an element from Z : ∆ we define hf, gi∆ def= [ f , g] = f g f −1g−1, where elements f , g are from ^GL1(A∆) such that θ( f ) = f and θ(g) = g . The element hf, gi∆ does not depend on the choice of f , g . The map h·, ·i∆ is a bilinear and alternating map from A∗ ∆ to Z . From Lemma 1 we have the following property (under conditions and notations of this lemma): ∆×A∗ hf, gi∆ = hp1(f ), p1(g)i∆1 + hp2(f ), p2(g)i∆2. (10) If ∆ coincides with the set of all pairs x ∈ C on X , then we will use also notation h·, ·iX for the map h·, ·i∆ . 5 Let K = k′((u))((t)) be a two-dimensional local field, where k′ ⊃ k is a finite extension of fields . By νK(·, ·) : K ∗ × K ∗ → Z we denote a bilinear and alternating map given as (11) where f, g ∈ K ∗ , the maps νK : K ∗ → Z and ν ¯K : ¯K ∗ = k′((u))∗ → Z are discrete valuations, and π : OK → ¯K is the natural homomorphism. νK(f, g) def= [k′ : k] · ν ¯K(cid:0)π(f νK (g)g−νK(f ))(cid:1) , Remark 1. There is another explicit formula for the expression νK(f, g) given as the product of the number [k′ : k] and the determinant of 2 × 2 -matrix of discrete valuations of rank 2 for the elements f and g . See this and another properties of the map νK(·, ·) in [16, § 2.2] and, for example, in [6, § 8.1]. For Kx,C = Ql x,C × Kx,C → Z as νx,C : K ∗ i=1 Ki , where Ki is a two-dimensional local field, we define a map νx,C(f, g) def= lXi=1 νKi(fi, gi), (12) where f, g are from K ∗ K ∗ x,C to K ∗ i . x,C , and fi, gi are corresponding projections of elements f, g from Proposition 1. 1. Let ∆ be a single pair x ∈ C . In this case h·, ·i∆ = −νx,C(·, ·) 2. For any set ∆ of pairs x ∈ C (as in the beginning of the paper) we have hf, gi∆ = X{x∈C}∈∆ hfx,C, gx,Cix∈C, (13) where f, g are from A∗ elements f, g from A∗ mula (13) contains only a finite number of non-zero terms. ∆ , the elements fx,C, gx,C are corresponding projections of ∆ to K ∗ x,C (see formulas (1) and (2)), and the sum in for- Proof. 1. Let Kx,C =Ql i=1 Ki , where Ki is a two-dimensional local field. Using a direct analog of formula (10) we reduce the statement to the following: hfi, giiKi = −νKi(fi, gi) , where the map h·, ·iKi is constructed by the central extension which is obtained as the restriction of the central extension ^GL1(Kx,C) from the group GL1(Kx,C) to the subgroup GL1(Ki) . Now this statement follows from Theorem 1 of [10]. (We note that there is a misprint with the sign in the statement and in the last line of the proof of Theorem 1 from [10].) 2. Let ∆2 be the set of all pairs x ∈ C from ∆ such that fx,COKx,C = OKx,C and gx,COKx,C = OKx,C . Let ∆1 be the complement set to ∆1 inside the set ∆ . By construction, the central extensions ^GL1(A∆2) and ^GL1(Kx,C) , where {x ∈ C} ∈ ∆2 , split. Therefore hf∆2, g∆2i∆2 = 0 , where f∆i, g∆i ( i = 1, 2 ) are corresponding projections of elements f, g from A∗ ∆i , and hfx,C, gx,Cix,C = 0 when {x ∈ C} ∈ ∆2 . Besides, from formulas (11) and (12) it follows that νx,C(fx,C, gx,C) = 0 when {x ∈ C} ∈ ∆2 . ∆ to A∗ 6 Therefore from formula (10) we obtain hf, gi∆ = hf∆1, g∆1i∆1 . Thus we can change ∆ to ∆1 in formula (13). From construction of the set ∆1 we have that the set of irreducible curves C which appear in pairs x ∈ C from ∆1 is finite. Again by formula (10) we can restrict ourself to a fixed irreducible curve C , i.e. we consider a set ∆ such that a curve C is fixed for pairs x ∈ C from ∆ . Since f ∈ A∆ and f −1 ∈ A∆ , from adelic conditions we obtain that there is a finite set of integers such that νKx,C (fx,C) belongs to this set when x runs over all smooth points on C from pairs {x ∈ C} ∈ ∆ . (If x is a smooth point on C , then Kx,C is a two-dimensional local field with the discrete valuation νKx,C .) The same is true for the element g ∈ A∗ ∆ , but with possibly another finite set. Therefore, subdividing the set ∆ into a finite number of subsets and using formula (10) we will suppose that ∆ satisfies conditions of one of the following two cases. In the former case, the set ∆ consists of one pair x ∈ C (when x is a singular point on C ), and therefore formula (13) is tautological and we will not consider this case further. In the remaining case, the integers νKx,C (fx,C) and νKx,C (gx,C) do not change when x runs over all smooth points on C such that {x ∈ C} ∈ ∆ . Let tC = 0 be an equation of the irreducible curve C on some open subset of X . Then using bilinear and alternating property of both hand sides of formula (13), and also the above properties of the set ∆ , we obtain that it is enough to consider two cases: 1) f and g are from O∗ A∆ and g = tC . In the first case, fx,C and gx,C are from O∗ Kx,C for all pairs x ∈ C from ∆ . Therefore, by construction, central extensions ^GL1(A∆) and ^GL1(Kx,C) , where {x ∈ C} ∈ ∆ , split. Hence hf, gi∆ = 0 and hfx,C, gx,Cix,C = 0 when {x ∈ C} ∈ ∆ , and formula (13) follows. In the second case, the A∆ ; 2) f ∈ O∗ right hand side of formula (13) equals to P{x∈C}∈∆ −νx,C(fx,C, tC) by the first statement of this proposition, and this sum contains only a finite number of non-zero terms by formulas (11) and (12) and adelic conditions on f . On the other hand, by definition of h·, ·i∆ we have hf, t−1 C i∆ = d(π(f )−1(U) − d(U) , where π is the natural homomorphism OA∆ → OA∆/tCOA∆ , d is a dimension theory on OA∆/tCOA∆ and U is an open linearly compact k -subspace in OA∆/tCOA∆ . (Compare with the calculation of case 2 in the proof of Theorem 1 of [10].) Fixing an open set U as the product of rings of integers of one-dimensional local fields, and dimension theory d such that d(U) = 0 , it is easy to see that d(π(f )−1(U) − d(U) =P{x∈C}∈∆ νx,C(fx,C, tC) . Thus we obtain formula (13) in this case. For a surface X , an irreducible curve C ⊂ X , and a point x ∈ X , let KC be the completion of the field k(X) of rational functions on X with respect to the discrete valuation given by C , let Kx = k(X) · Ox,X be a subring of the fraction field Frac Ox,X . Let D be a divisor on X . For an irreducible curve C ⊂ X let jD C be an equation of the divisor D after the restriction to Spec KC . For any point y ∈ C we have an embedding KC ⊂ Ky,C . It is easy to check that a collection {jD C } , where C runs over the set of all irreducible curves on X , defines a well-defined element from A∗ X under the natural diagonal embedding C ∈ K ∗ QC⊂X KC ֒→Qy∈C Ky,C . 7 For a point x ∈ X let jD x be an equation of the divisor D after the restriction to Spec Kx . For any irreducible curve E ∋ x we have an embedding Kx ⊂ Kx,E . It is easy to check that a collection {jD x } , where x runs over the set of all points of X , defines a well-defined element from A∗ X under the natural diagonal embedding x ∈ K ∗ Qx∈X Kx ֒→Qx∈E Kx,E . Using the definition of the intersection index of divisors given by A.N. Parshin in [16, § 2.2] by means of sum of local maps νx,C , we immediately obtain from Proposition 1 the following proposition. (We note that the analog of this proposition was used without written proof in [15].) Proposition 2. Let S and T be divisors on a smooth projective surface X , and (S, T ) ∈ Z be their intersection index. We have h{jS x }, {jT C}iX = −(S, T ). 3 Second Chern numbers 3.1 Central extension \GLn(A∆) For any ∆ which is a subset of all pairs x ∈ C , where C is an irreducible curve on X . We have natural isomorphism of groups GLn(A∆) = SLn(A∆) ⋊ A∗ ∆, where the group A∗ ∆ is embedded into the upper left corner of the group GLn(A∆) and acts on the group SLn(A∆) by conjugation, i.e. by inner automorphisms h 7→ aha−1 , where a ∈ A∗ ∆ and h ∈ SLn(A∆) . By means of the central extension (9) the action of the group A∗ ∆ is lifted to the action on the group θ−1(SLn(A∆)) (by inner automorphisms of the group ^GLn(A∆) . We define a group \GLn(A∆) def= θ−1(SLn(A∆)) ⋊ A∗ ∆, whose natural homomorphism to GLn(A∆) gives a central extension 0 −→ Z −→ \GLn(A∆)−→GLn(A∆) −→ 1, which, by construction, splits over the subgroup A∗ ∆ of GLn(A∆) . Remark 2. To construct central extension \GLn(A∆) we used an embedding of A∗ ∆ to GLn(A∆) as a 7→ diag(a, 1, . . . , 1) , where a ∈ A∗ ∆ . Since an inner automorphism of the group GLn(A∆) induces a canonical automorphism of the group that is a central extension of GLn(A∆) , another embedding a 7→ diag(1, . . . , a, . . . , 1) of A∗ ∆ to GLn (into other place on the diagonal) produces a construction of the central extension which is canonically isomorphic to the central extension \GLn(A∆) (compare also with Remark 3 from [13]). 8 Remark 3. From formula (3) and Lemma 1 we obtain the property which is similar to the statement of Lemma 1 when we replace the central extensions ^GLn(A∆) , ^GLn(A∆1) and ^GLn(A∆2) to \GLn(A∆) , \GLn(A∆1) and \GLn(A∆2) correspondingly (compare with [13, Prop. 2]). The analogy with the next proposition (and with remark after them) can be found in [1, § A5] and [5, Appendix], where it was considered a central extension of a group GLn(A) by a group K2(A) for a ring A with the property SK1(A) = 0 . We note that it is not clear how to deduce the next proposition (and remark after them) from [1, § A5] and [5, Appendix]. We consider a central extension 0 −→ Z −→ \A∗ ∆ × A∗ ∆ −→ A∗ ∆ × A∗ ∆ −→ 1, (14) where \A∗ ∆ × A∗ ∆ def= A∗ ∆ × Z as a set, and with the multiplication law given as ∆ × A∗ (f, g; r)(f ′, g′; r′) def= (f f ′, gg′; r + r′ + hf ′, gi∆), where f, g, f ′, g′ are from A∗ ∆ , and r, r′ are from Z . For any a ∈ A∗ ∆ we denote by φ1(a) the element from \GLn(A∆) which equals to the canonical section of the central extension \GLn(A∆) over the subgroup A∗ ∆ applied to the element a . For any integer l such that 1 ≤ l ≤ n we denote φl(a) = ΦlaΦ−1 , where Φl is l a lift to \GLn(A∆) of the matrix from GLn(A∆) which acts as transposition on standard coordinates of An ∆ permuting the first and the l -th coordinates. Clearly, φl(a) does not depend on a lift of such matrix, and the image of φl(a) under the standard homomorphism to GLn(A∆) equals to diag(1, . . . , a, . . . , 1) with the element a is located on the l -th place of the diagonal. Proposition 3. 1. We fix integers 1 ≤ i < j ≤ n and embed the group A∗ ∆ × A∗ ∆ into the group GLn(A∆) as (f, g) 7→ diag(1, . . . , f, . . . , g, . . . , 1) , where elements f and g from A∗ ∆ are located on i -th and j -th places on the diagonal. We obtain that the restriction of the central extension \GLn(A∆) to the subgroup A∗ ∆ is isomorphic to the central extension (14) via the map ∆ × A∗ where r is from Z , which is a subgroup of the center of the group \GLn(A∆) . rφi(f )φj(g) 7−→ (f, g; r), 2. For positive integers n1 and n2 such that n = n1 + n2 we consider a subgroup Pn1,n2(A∆) def= (cid:26)(cid:18)GLn1(A∆) 0 GLn2(A∆)(cid:19)(cid:27) ⊂ GLn(A∆) ∗ : Pn1,n2(A∆) → GLni(A∆) be the projections, where i = 1 and i = 2 . Let pi We obtain that the restriction of the central extension \GLn(A∆) to the subgroup 1( \GLn1(A∆)) , Pn1,n2(A∆) is isomorphic to the Baer sum of central extensions p∗ 2( \GLn2(A∆)) and (det(p1) × det(p2))∗ \A∗ ∆ × A∗ p∗ ∆ . 9 Proof. 1. For any element a ∈ A∗ and integer i such that 1 ≤ i ≤ n , we denote di(a) = diag(1, . . . , a, . . . , 1) ∈ GLn(A∆) , where a is located on i -th place in the diagonal matrix. It is enough to prove the following equality inside the group \GLn(A∆) for any elements f, g, f ′, g′ from GLn(A∆) : φi(f )φj(g)φi(f ′)φj(g′) = hf ′, gi∆ · φi(f f ′)φj(gg′). Clearly, this equality follows from an equality φj(g)φi(f ′) = hf ′, gi∆ · φi(f ′)φj(g) . Thus, we have to prove that [φj(g), φi(f ′)] = hf ′, gi∆ or [φi(f ′), φj(g)] = hg, f ′i∆ . Applying the conjugation by Φi , we obtain that it is enough to prove an equality [φ1(f ′), Φi · φj(g) · Φ−1 under the standard homomorphism to GLn(A∆) equals to dj(g) . Since the commutator of lifts of two elements does not depend on the choice of lifts of these elements to the central exten- sion, we have that [φ1(f ′), Φi · φj(g) · Φ−1 ] = [φ1(f ′), φj(g)] . Further, using the bilinear property of the commutator of lifts of commuting elements, we obtain ] = hg, f ′i∆ . We note that the image of Φi · φj(g) · Φ−1 i i i [φ1(f ′), φj(g)] = [φ1(f ′), φ1(g)φ1(g)−1φj(g)] = = [φ1(f ′), φ1(g)] · [φ1(f ′), φ1(g)−1φj(g)] = [φ1(f ′), φ1(g)−1φj(g)]. (15) We denote by h the image of the element φ1(g)−1φj(g) under the homomorphism to GLn(A∆) . Since h belongs to the subgroup SLn(A∆) , by construction of the group \GLn(A∆) the last commutator in formula (15) equals to commutator [^d1(f ′), h] computed in the group ^GLn(A∆) , where ^d1(f ′) and eh are lifts of elements d1(f ′) and h from the group GLn(A∆) to the group ^GLn(A∆) . We obtain in the group ^GLn(A∆) an equality [^d1(f ′), h] = [^d1(f ′), ^d1(g−1) · ]dj(g)] = [^d1(f ′), ^d1(g−1)] · [^d1(f ′), ]dj(g)]. Now from construction of the group ^GLn(A∆) we obtain that the commutator of the lift of diagonal matrices can be calculated componentwise (first, separately for each place on the diagonal, and then multiply the results). Therefore we obtain [^d1(f ′), ^d1(g−1)] = hg, f ′i∆ , and since j 6= 1 , we have [^d1(f ′), ]dj(g)] = 1 . 2. We embed the group A∗ ∆ into the group Pn1,n2(A∆) in the following way: (f, g) 7→ diag(f, . . . , g, . . . , 1) , where elements f and g from A∗ ∆ are located on the first and (n1 + 1) -th places on the diagonal, and other places of the diagonal are occupied by 1 . We can write the group Pn1,n2(A∆) as a semidirect product: ∆ × A∗ (cid:26)(cid:18)SLn1(A∆) 0 SLn2(A∆)(cid:19)(cid:27) ⋊ (A∗ ∗ ∆ × A∗ ∆). (16) According to Construction 1.7 from [2], a central extension \G ⋊ H of a semidirect product G ⋊ H by a group A is equivalent to the following data: 1) a central extension G of G by A ; 2) a central extension H of H by A ; 3) an action of H on G → G , 10 lifting the action of H on G . We note that central extensions G and H are obtained as restrictions of the central extension \G ⋊ H to the subgroups G and H correspondingly. To prove the second statement of Proposition 3, we apply the above construction ∆ and to the case of semidirect product given by formula (16), where H = A∗ ∆ × A∗ SLn2(A∆)(cid:19)(cid:27) . Then condition 2) of the construction follows from ∗ G = (cid:26)(cid:18)SLn1(A∆) 0 the first statement of Proposition 3. To obtain condition 1) we note that the restriction of the central extension ^GLn(A∆) to the subgroup Pn1,n2(A∆) is canonically isomorphic ^GLn2(A∆) . This fact easily to the Baer sum of central extensions p∗ 1 follows from a natural action of Pn1,n2(A∆) on an exact triple (exact triple of complete C2 -vector spaces over k ) ^GLn1(A∆) and p∗ 2 0 −→ An1 ∆ −→ An ∆ −→ An2 ∆ −→ 0, (17) from formula (5), and from the construction of the group ^GLn(A∆) . To finish the checking of condition 2) we note that the restrictions of the central extensions ^GLn(A∆) and \GLn(A∆) to the subgroup SLn(A∆) coincide (or canonically isomorphic). To obtain condition 3) we note that an action of the group H on G → G comes from the action by ∆ (lifted to ^GLn(A∆) ) on ^GLn(A∆) restricted as conjugations of elements from A∗ central extension to Pn1,n2(A∆) . Besides, it is important that the group H naturally acts on exact triple (17), and therefore the action of H on G is compatible with the action on the corresponding Baer sum with respect to projections p1 and p2 . ∆ × A∗ Remark 4. 1. From the first statement of Proposition 3 it is easy to obtain the fol- lowing generalization. For any integer k such that 1 ≤ k ≤ n we consider a central extension 0 −→ Z −→ \(A∗ ∆)k −→ (A∗ ∆ × . . . A∗ ∆)k −→ 1, ∆ with A∗ (18) ∆ beeing taken k times, ∆)k is the direct product A∗ where (A∗ and \(A∗ ∆)k def= (A∗ ∆)k × Z as a set, where the group multiplication law is given as (f1, . . . , fk; r)(f ′ 1, . . . , f ′ k; r′) def= (f1f ′ 1, . . . , fkf ′ k; r + r′ +Xi<j hf ′ i , fji∆), 1, . . . , f ′ k are from A∗ ∆ , and r, r′ are from Z . We fix integers where f1, . . . , fk, f ′ 1 ≤ j1 < . . . jk ≤ n , and consider an embedding of the group (A∗ ∆)k to the group GLn(A∆) given as: (f1, . . . , fk) is mapped to the diagonal matrix diag(a1, . . . , an) , where aji = fi (for 1 ≤ i ≤ k ), and al = 1 otherwise. Then the restriction of the central extension \GLn(A∆) to the subgroup (A∗ ∆)k is isomorphic to the central extension (18) via the map rφj1(f1) · . . . · φjk(fk) 7−→ (f1, . . . , fk; r). 2. The central extension \GLn(A∆) canonically splits over A∗ ∆ , where this group is embedded into the i -th place of the diagonal, via the map a 7→ φi(a) . From this 11 fact and the second statement of Proposition 3 we obtain that the central extension \GLn(A∆) canonically splits over the subgroup Un =    1 0 1 ∗ 1 .     3.2 Canonical splittings Now we give the generalization of non-commutative reciprocity laws from [13, § 3.5] when X is a smooth algebraic surface over k . We recall (see the end of Section 2.4) that we have diagonal embeddings YC⊂X KC ֒→ Yx∈C Kx,C and Yx∈X Kx ֒→ Yx∈C Kx,C. There are the following subrings of the adelic ring AX : KC) ∩ AX , AX,01 = (YC⊂X where the intersection is taken inside the ring Qx∈C Kx,C . AX,02 = (Yx∈X Kx) ∩ AX , and AX,12 = OAX , (19) Proposition 4. 1. For any set ∆ of pairs x ∈ C the central extensions ^GLn(A∆) and \GLn(A∆) canonically split over the subgroup GLn(OA∆) of the group GLn(A∆) . 2. The central extensions ^GLn(AX) and \GLn(AX ) canonically split over the subgroup GLn(AX,02) of the group GLn(AX ) . 3. Suppose that X is projective. Then the central extensions ^GLn(AX) and \GLn(AX) canonically split over the subgroup GLn(AX,01) of the group GLn(AX ) . 4. Splittings of the central extension \GLn(AX ) from statements 2-3 coincide over the subgroup GLn(k(X)) . The analogous results are also true for the splittings from statements 1-2 and the subgroup GLn(AX,12 ∩ AX,02) , and for the splittings from statements 1, 3 and the subgroup GLn(AX,12 ∩ AX,01) . 5. Under the same conditions as in statements 1-3, the central extension \A∗ ∆ × A∗ X,01 ×A∗ X,02 and A∗ X,02 ×A∗ formula (14)) splits over the subgroups O∗ via the map (f, g) 7→ (f, g; 0) . ∆ (see X,01 A∆ ×O∗ A∆ , A∗ 6. For the central extension \GLn(AX ) restricted to a subgroup Pn1,n2(AX) ⊂ GLn(AX) , splittings from statements 1-3 and 5 are compatible with respect to the isomorphism constructed in the second statement of Proposition 3. Proof. 1. The splittings follow from the constructions of the central extensions ^GLn(A∆) and \GLn(A∆) , since for any element f ∈ GLn(OA∆) we have f On A∆ = On A∆ . 12 2. First we prove that the central extension ^GLn(AX) splits over the subgroup GLn(AX,02) . We note that for any two divisors D1 ≥ D2 the subspace (OAX (D2) ∩ AX,02)/(OAX (D1) ∩ AX,02) ⊂ OAX (D2)/OAX (D1) is an open linearly compact k -vector space. Hence for any g1, g2 ∈ GLn(AX ) such that g2On AX the subspace AX ⊃ g1On (g2On AX ∩ An X,02)/(g1On AX ∩ An X,02) ⊂ g2On AX /g1On AX is an open linearly compact k -vector space. We define dg1,g2 ∈ Dim(g2On the rule dg1,g2((g2On a well-defined element dg1,g2 ∈ Dim(g1On Now it is easy to see that the map AX ) by X,02)) = 0 . Using formulas (6)-(8), we obtain AX ) for any elements g1, g2 ∈ GLn(AX) . X,02)/(g1On AX g2On AX /g1On AX ∩ An AX ∩ An GLn(AX,02) −→ ^GLn(A∆) : g 7−→ (g, d1,g) (20) is a group splitting. To prove the splitting of the central extension \GLn(AX ) over the subgroup GLn(AX,02) we note that GL(n, AX,02) = SL(n, AX,02) ⋊ A∗ X,02 , the conjuga- tion by the element (a, d1,a) ∈ ^GLn(A∆) does not change the section over the group SL(n, AX,02) which was constructed in formula (20). By construction, this gives the splitting of \GLn(AX) over GLn(AX,02) , where we take the trivial section over A∗ X,02 . For any a ∈ A∗ X,02 . AX g2On AX g2On 3. The idea for the proof of this statement is analogous to the proof of statement 2, but instead of element dg1,g2 ∈ Dim(g1On AX ) we have to use another element d′ g1,g2 ∈ Dim(g1On AX ) which is constructed by means of the following property. For any two divisors D1 ≥ D2 the subspace U = (OAX (D2) ∩ AX,01)/(OAX (D1) ∩ AX,01) of the space V = OAX (D2)/OAX (D1) is a discrete k -vector subspace such that V /U is a linearly compact k -vector space. (It is important that on the projective curve C the field of rational functions k(C) is a discrete subspace inside the adelic ring of C , and the quotient space is a linearly compact k -vector space that follows, for example, from the adelic complex on the curve C and the fact that k -vector spaces H 0(C, OC) and H 1(C, OC) are finite-dimensional over k .) Hence for any g1, g2 ∈ GLn(AX) such that g2On AX we have an exact triple of k -vector spaces AX ⊃ g1On 0 −→ Y −→ W −→ W/Y −→ 0, AX ∩ An AX /g1On X,01)/(g1On where W = g2On AX is a locally linearly compact k -vector space, k -vector space AX ∩ An Y = (g2On X,01) is a discrete subspace in induced topology, and the space W/Y endowed with the quotient topology is a linearly compact k -vector space. Using formulas (4)-(5) we define an element d′ g1,g2 ∈ Dim(W ) as dY ⊗ dW/Y , where dY ∈ Dim(Y ) is defined as dY ((0)) = 0 (here (0) is the zero subspace), and dW/Y ∈ Dim(W/Y ) is defined as dW/Y (W/Y ) = 0 . To finish we proceed further as in the proof of statement 2. 13 4. The group SLn(k(X)) is perfect. Therefore any two sections of the central exten- sion ^GLn(AX) restricted to the group SLn(k(X)) coincide. Hence, two sections of the central extension \GLn(AX) restricted to the group GLn(k(X)) coincide, because over the subgroup k(X)∗ : a 7→ diag(a, 1, . . . , 1) two sections are trivial by constructions in the proof of statements 2-3. of extension ^GLn(AX) the central Various splittings subgroup GLn(AX,01 ∩ AX,12) and over the subgroup GLn(AX,02 ∩ AX,12) coincide, because for any element f ∈ GLn(OA∆) we have f On A∆ , and then we have to use the con- structions of the splittings. Hence, again by construction, the same is true for the central extension \GLn(AX ) . A∆ = On over the 5. By statements 1-3 the central extension ^GL1(A∆) splits over O∗ extension ^GL1(AX) splits over A∗ hf, gi∆ = 0 for f, g ∈ O∗ for f, g ∈ A∗ A∆ , the central X,01 (when X is projective). Hence X,02 and when X is projective ∆ . 6. This statement follows from constructions of sections in proofs of statements 1-3 and of the second statement of Proposition 3. It is important that together with exact sequence (17) we can write exact sequences X,01 . Now we finish by the definition of the multiplication law in \A∗ ∆ × A∗ A∆ , and hf, giX = 0 for f, g ∈ A∗ X,02 and over A∗ 0 −→ An1 X,02 −→ An X,02 −→ An2 X,02 −→ 0 and 0 −→ An1 X,01 −→ An X,01 −→ An2 X,01 −→ 0 and the corresponding groups Pn1,n2(AX)∩GLn(AX,02) and Pn1,n2(AX)∩GLn(AX,01) act on these sequences. Besides, concerning the central extension \A∗ X × A∗ X , we note that for any integer l such that 1 ≤ l ≤ n if an element a is from A∗ X,ij with ij equal to 12 or 02 or 01 , then the element φl(a) ∈ \GLn(AX ) (see its definition before Proposition 3) equals to a section over the element diag(1, . . . , a, . . . , 1) , where this section is constructed in statements 1, 2 or 3 correspondingly, and a is located on the l -th place of the diagonal in diag(1, . . . , a . . . , 1) . This is because the matrix of transposition of coordinates which was used to construct φl(a) belongs to any of the groups: GLn(AX,12) , GLn(AX,02) and GLn(AX,01) . Hence, the map diag(f, . . . , g, . . . , 1) 7→ φ1(f )φn1+1(g) equals to a section from the proof of statements 1, 2 and 3 when f, g ∈ A∗ X,ij , where ij equal to 12 or 02 or 01 correspondingly (compare with the proof of the second statement of Proposition 3). 6(iii)] (after T. Fimmel Remark 5. For a smooth projective surface X we have AX,02 ∩ AX,01 = k(X) , see [4, Th. obtain GLn(k(X)) = GLn(AX,02) ∩ GLn(AX,01) . Hence we can reformulate the statement 4 of Proposition 4 in the following way: splittings of the central extension \GLn(AX) from statements 1-3 coincide over the intersections of corresponding subgroups. and A.N. Parshin). Therefore we 3.3 Second Chern number Now we give a construction of the second Chern number for a vector bundle on a smooth algebraic surface X over k . 14 Let E be a locally free sheaf of OX -modules of rank n on X . Follow [16] we introduce transition matrices for E . For any point x ∈ X let ex be a basis of the free Ox,X -module Ox,X . For any irreducible curve C on X let eC be a basis of the free OKC -module E ⊗OX E ⊗OX OKC , where OKC is the discrete valuation ring of the field KC . Let e0 be a a basis of the free k(X) -module E ⊗OX k(X) . These expressions can be considered as completions of the stalks of E at scheme points of X . Each of above basis consists of n elements. For any point x ∈ X we have the transition matrix αx ∈ GLn(Kx) defined as e0 = αxex . For any irreducible curve C on X we have the transition matrix αC ∈ GLn(KC) defined as e0 = αCeC . For any pair x ∈ C we have the transition matrix αx,C ∈ GLn(OKx,C ) defined as ex = αx,CeC . via the diagonal embedding. When we vary irreducible curves C on X , we obtain the When we vary points x ∈ X , we obtain the matrix α02 = {αx} ∈ GLn(Qx∈C Kx,C) matrix α01 = {αC} ∈ GLn(Qx∈C Kx,C) via the diagonal embedding. When we vary pairs x ∈ C , we obtain the matrix α21 = {αx,C} ∈ GLn(Qx∈C Kx,C) . We define α20 = α−1 21 . We have an evident equality: 02 , α10 = α−1 01 , α12 = α−1 α02α21α10 = 1, (21) where 1 is the identity matrix. If we change the basis: {ex} 7−→ α2{ex} , {eC} 7−→ α1{eC} , e0 7−→ α0e0 , where α2 ∈ GLn(Qx∈X GLn(AX,01 ∩ AX,12) , and α0 ∈ GLn(k(X)) , then we obtain the change of matrices: Ox,X) = GLn(AX,02 ∩ AX,12) , α1 ∈ GLn(QC⊂X OKC ) = α02 7−→ α0α02α−1 2 , α21 7−→ α2α21α−1 1 , α10 7−→ α1α10α−1 0 . (22) It is easy to see that α01 and α02 are from GLn(AX ) , because by formula (22) we can change the basis e0 , {ex} and {eC} to a more convenient basis, for example, to take a trivialization of E on some open cover of X in Zariski topology, and then e0 equal to the trivialization of E on a fixed open subset from this open cover, and ex , eC also come from the trivialization of E on this open cover of X . Hence and using formula (21) we obtain that α21 ∈ GLn(AX) . Therefore we have that α02 ∈ GLn(AX,02) , α21 ∈ GLn(AX,12) , α10 ∈ GLn(AX,01). Let cα02 ∈ \GLn(AX) be the canonical section applied to α02 and which was construct- ed in statement 2 of Proposition 4. Let cα21 ∈ \GLn(AX) be the canonical section applied to α21 and which was constructed in statement 1 of Proposition 4. Let cα02 ∈ \GLn(AX) be the canonical section applied to α02 and which was constructed in statement 3 of Proposition 4. Theorem 1. Let E be a locally free sheaf of OX -modules of rank n on a smooth pro- jective surface X over a perfect field k . 15 and the statement 4 of Proposition 4. on the choose of basis e0 , {ex} and {eC} of E . 1. An expression cα02 cα21 cα10 ∈ \GLn(AX) gives an element from Z and does not depend 2. An expression cα02 cα21 cα10 equals to the second Chern number c2(E) of E . Proof. 1. Since the image of cα02 cα21 cα10 in GLn(AX ) equals to α02α21α10 = 1 , we obtain that cα02 cα21 cα10 ∈ Z . The independence on the choice of basis follows from formula (22) 2. It is known that for any locally free sheaf F of OX -modules of rank more than 1 there is a smooth surface Y and a morphism f : Y → X , where Y is obtained by means of chain of of blow-ups of points, such that there is a locally free subsheaf F1 ⊂ f ∗F with (f ∗F )/F1 is again a locally free subsheaf of OY -modules. (Indeed, it is enough to find a section s ∈ H 0(Y, f ∗E) such that s(y) 6= 0 for any point y ∈ Y , then the sheaf M = (f ∗E)/(OY · s) is locally free, because TorOy,Y (k(y), My) = 0 for any point y ∈ Y .) Therefore from the general theory of the Chern classes it follows that the number c2(E) coincides with the second Chern number c2(E) on a smooth projective algebraic surface if and only if the following conditions are satisfied: 1 1) c2(L) = 0 for any locally free sheaf L of rank 1 ; 2) c2(N ) = c2(π∗(N )) , where N is a locally free sheaf and π is a blow-up of a point; 3) for any exact sequence of locally free sheaves 0 −→ E1 −→ E2 −→ E3 −→ 0 (23) we have c2(E2) = c2(E1) + c2(E3) + (det(E1), det(E3)) , where (·, ·) is the intersection index of two divisors which are rational sections of corresponding invertible sheaves. X . construction, \GL1(AX ) = Z × A∗ In our case c2 equals to cα02 cα21 cα10 . The first condition is satisfied, because, by To check the second condition we note that if π : Y → X is a blow-up of a point x ∈ X , then AY = AX × A∆ with the set ∆ which consists of all pairs y ∈ R , where π(R) = x . By the first statement of this theorem, c2 does not depend on the choice of the basis. Therefore we choose the special basis. We fix a trivialization of E on an open neighbourhood of x on X . This trivialization gives us the same basis e0 , ex , eR and ey , where y ∈ R , for E and π∗E . We identify the other basis for E and π∗E . The decomposition GLn(AY ) = GLn(AX) × GLn(A∆) implies the canonical embedding γ : GLn(AX) ֒→ GLn(AY ) . From construction of the central extension we have canonical isomorphism (compare also with Remark 3): δ : \GLn(AX ) −→ γ∗( \GLn(AY ). It is easy to see that from our choice of the basis for E and π∗E we have γ(α02,E ) = α02,π∗E , γ(α21,E) = α21,π∗E , γ(α10,E) = α10,π∗E, where we put an additional index E or π∗E to specify a locally free sheaf. Besides, from the construction of the central extension and the splittings we obtain δ([α02,E ) = \α02,π∗E , δ([α21,E) = \α21,π∗E) , δ([α10,E) = \α10,π∗E, 16 where we consider elements \α02,π∗E , \α21,π∗E and \α10,π∗E as elements from the group γ∗( \GLn(AY ) . This finishes the checking of the second condition. Since we can take the basis compatible with exact sequence (23) and c2 does not depend on the choice of the basis, the third condition for c2 follows from statements 5 and 6 of Proposition 4, the second statement of Proposition 3 and the following fact. Let C and D be invertible sheaves on a smooth projective surface X , and α02,C , α21,C , α10,C , α02,D , α21,D , α10,D be transition matrices, in fact elements from A∗ X , for sheaves C and D correspondingly (after the choice of basis for these sheaves). Then there is an equality in the group \A∗ X × A∗ X : (α02,C, α02,D; 0)(α21,C, α21,D; 0)(α10,C, α10,D; 0) = (C, D) ∈ Z. (24) We prove this equality now. The left hand side of expression (24) equals to (α02,C · α21,C, α02,D · α21,D; hα21,C, α02,DiX) · (α10,C, α10,D; 0) = = (α01,C, α01,D, hα21,C, α02,DiX)· (α10,C, α10,D; 0) = (1, 1; hα21,C, α02,DiX) +hα10,C, α01,DiX) Since by statements 1-3 of Proposition 4 the central extension ^GL1(AX) canonically splits over subgroups A∗ X,02 , we have hα10,C, α01,DiX = 0 . Therefore it is enough to prove that hα21,C, α02,DiX = (C, D) . We have X,01 and A∗ X,12 , A∗ hα21,C, α02,DiX = hα02,D, α12,CiX = hα02,D, α12,CiX + hα21,D, α12,CiX = hα01,D, α12,CiX = = hα01,D, α01,CiX + hα01,D, α12,CiX = hα01,D, α02,CiX = (C, D), where the last equality follows from Proposition 2. Thus we have proved the theorem. 4 Case of arithmetic surface Now we will give analogs of certain statements of Proposition 4 for arithmetic surfaces. We note that Proposition 4 is one of key propositions used in Theorem 1. By an arithmetic surface we mean here a two-dimensional integral regular scheme of finite type over Z with the proper surjective morphism to Spec Z . For an arithmetic surface X there is an adelic arithmetic ring Aar X introduced in [14, Example 11] (see also explanations in [13, § 3.4]): Aar X def= AX × AX,∞ , where the ring AX,∞ def= AXQ b⊗ R = lim −→ D2 lim ←− D1≥D2 (AXQ(D2)/AXQ(D1)) ⊗Q R, and AXQ is the adelic ring of the curve XQ = X ×Spec Z Spec Q , which is the generic fibre, D1 and D2 are divisors on the curve XQ . 17 + + X)R∗ and \GLn(Aar The central extensions ^GLn(Aar of the group GLn(Aar X)R∗ X) by the (multiplicative) group of positive real numbers R∗ + were constructed in [13]. The con- structions of these central extensions can be done similarly to constructions of central extensions ^GLn(A∆) and \GLn(A∆) from sections 2.3 and 3.1, but we have to use that Aar X is an object of the category C ar from [14, § 5] (instead of the category C2 which 2 we used before). Correspondingly, instead of locally linearly compact k -vector spaces we have to use locally compact Abelian groups, and instead of Z -torsor of dimension theories Dim(V ) for a locally linearly compact k -vector space V we have to use R∗ + -torsor µ(W ) of Haar measures for a locally compact Abelian group W . We note that the similar construction can be done for an algebraic surface over a finite field Fq . In this case a locally linearly compact Fq -vector space V is also an Abelian locally compact group. A homomorphism Z → R∗ + : a 7→ qa induces the map Dim(V ) → µ(V ) of corresponding torsors. This gives the homomorphism from the central extensions ^GLn(AX) and \GLn(AX) to the central extensions ^GLn(Aar X)R∗ and \GLn(Aar correspondingly. + X)R∗ + Similarly to Lemma 1 we have that the cental extension ^GLn(Aar X)R∗ + is the Baer and ^GLn(AX,∞)R∗ sum of the central extensions ^GLn(AX)R∗ extensions are obtained by restrictions of the central extension ^GLn(Aar GLn(AX ) and GLn(AX,∞) of the group GLn(Aar similarly to Remark 3, the central extension \GLn(Aar extensions \GLn(AX)R∗ and \GLn(AX,∞)R∗ to subgroups X) . Analogously to this statement and is the Baer sum of the central , where the last two central X)R∗ X)R∗ . + + + + + + We have a subring AX,∞(0) = lim ←− D≤0 (AXQ(0)/AXQ(D)) ⊗Q R of the ring AX,∞ , where 0 is zero divisor on XQ , and D is a divisor on XQ which is less or equal than 0 . + The ring AX,02 (with the definition as in formula (19)) is a subring of the ring AX . to the X ) as g 7→ g × 1 coincide (or and Besides, the restrictions of the central extensions ^GLn(Aar subgroup GLn(AX,02) embedded to the group GLn(Aar canonically isomorphic) with the restrictions of the central extensions ^GLn(AX)R∗ \GLn(AX )R∗ to the subgroup GLn(AX,02) of the group GLn(AX) . and \GLn(Aar X )R∗ X)R∗ + + + The ring AX,01 (with the definition as in formula (19)) is a subring of the ring AX . Besides, there is a homomorphism from the ring AX,01 to the ring Aar X induced by the natural embedding KC ֒→ AX,∞ for any "horizontal" curve C on X (or, in other words, C is an integral one dimensional subscheme of X which maps surjectively onto Spec Z ), see more explanations in [13, § 3.4-§ 3.5]. Thus we consider AX,01 as a subring of Aar X . The last embedding induces the embedding GLn(AX,01) ֒→ GLn(Aar X , where AX,01 is mapped in the both parts of Aar X ) . We obtain the following proposition, which contains analogs of statements 1-3 from Proposition 4 and generalizes Theorem 1 from [13]. 18 Proposition 5. Let X be an arithmetic surface. The central extensions ^GLn(Aar and \GLn(Aar GLn(AX,02) and GLn(AX,01) of the group GLn(Aar X)R∗ canonically split over the subgroups GLn(AX,12) × GLn(AX,∞(0)) , X )R∗ + + X) . The proof of this proposition is completely similar to the proof of analogous statements of Proposition 4. We note only that the splitting over the subgroup GLn(AX,02) is enough to prove for the central extensions ^GLn(AX)R∗ and \GLn(AX )R∗ . + + References [1] Beilinson A. A., Schechtman V. V., Determinant bundles and Virasoro algebras Comm. Math. Phys., 118 (1988), no. 4, pp. 651-701. [2] Brylinski J.-L, Deligne P., Central extensions of reductive groups by K2 , Publ. Math. Inst. Hautes ´Etudes Sci. No. 94 (2001), pp. 5-85. [3] Budylin R. Ya., Adelic construction of the Chern class. (Russian) Mat. Sb. 202 (2011), no. 11, 75-96; translation in Sb. Math. 202 (2011), no. 11-12, 1637-1659. [4] Budylin R. Ya., Gorchinskiy S. O., Intersections of adelic groups on a surface, (Rus- sian) Mat. Sb. 204 (2013), no. 12, 3-14; translation in Sb. Math. 204 (2013), no. 11-12, pp. 1701-1711. [5] Chinburg T., Pappas, G., Taylor M. J., Higher adeles and non-abelian Riemann-Roch Adv. Math. 281 (2015), pp. 928-1024. [6] Gorchinskiy S. O., Osipov D. V., A higher-dimensional Contou-Carrre symbol: local theory. (Russian) Mat. Sb. 206 (2015), no. 9, 21-98; translation in Sb. Math., 2015, 206 (9), 1191-1259. [7] Huber A., On the Parshin-Beilinson Adeles for Schemes, Abh. Math. Sem. Univ. Hamburg, 61 (1991), pp. 249-273. [8] Kapranov M., Semiinfinite symmetric powers, e-print arXiv:math/0107089 [math.QA]. [9] Osipov, D. V., Adelic constructions of direct images of differentials and symbols. (Russian) Mat. Sb. 188 (1997), no. 5, 59-84; translation in Sb. Math. 188 (1997), no. 5, 697-723. [10] Osipov D. V., Central extensions and reciprocity laws on algebraic surfaces, (Russian) Mat. Sb. 196:10 (2005), pp. 111-136; translation in Sb. Math. 196:10 (2005), pp. 1503- 1527. 19 [11] Osipov D. V., n -dimensional local fields and adeles on n -dimensional schemes, Sur- veys in Contemporary Mathematics, Edited by N. Young, Y. Choi; Cambridge Uni- versity Press, London Mathematical Society Lecture Note Series, No. 347 (2007), pp. 131-164. [12] Osipov D., Adeles on n -dimensional schemes and categories Cn , International Jour- nal of Mathematics, vol. 18, no. 3 (2007), pp. 269-279. [13] Osipov D. V., Unramified two-dimensional Langlands correspondence, Izvestiya RAN: Ser. Mat, 2013, 77:4, pp. 73-102; english translation in Izvestiya: Mathematics, 2013, 77:4, pp. 714-741. [14] Osipov D. V., Parshin A. N., Harmonic analisys on local fields and adelic spaces. II, Izvestiya RAN: Ser. Mat., 2011, 75:4, pp. 91-164; english translation in Izvestiya: Mathematics, 2011, 75:4, pp. 749-814. [15] Osipov D. V., Parshin A. N., Harmonic analisys and the Riemann-Roch theorem, Doklady Akademii Nauk, 2011, Vol. 441, No. 4, pp. 444-448; english translation in Doklady Mathematics, 2011, Vol. 84, No. 3, pp. 826-829. [16] Parshin A. N., Chern classes, adeles and L -functions, J. Reine Angew. Math., 341 (1983), pp. 174-192. Steklov Mathematical Institute of Russsian Academy of Sciences ul. Gubkina 8, Moscow, 119991 Russia E-mail: d−[email protected] 20
1212.0269
2
1212
2013-12-23T05:56:43
On certain duality of N\'eron-Severi lattices of supersingular K3 surfaces
[ "math.AG" ]
Let X and Y be supersingular K3 surfaces defined over an algebraically closed field. Suppose that the sum of their Artin invariants is 11. Then there exists a certain duality between their N\'eron-Severi lattices. We investigate geometric consequences of this duality. As an application, we classify genus one fibrations on supersingular K3 surfaces with Artin invariant 10 in characteristic 2 and 3, and give a set of generators of the automorphism group of the nef cone of these supersingular K3 surfaces. The difference between the automorphism group of a supersingular K3 surface X and the automorphism group of its nef cone is determined by the period of X. We define the notion of genericity for supersingular K3 surfaces in terms of the period, and prove the existence of generic supersingular K3 surfaces in odd characteristics for each Artin invariant larger than 1.
math.AG
math
ON CERTAIN DUALITY OF N´ERON-SEVERI LATTICES OF SUPERSINGULAR K3 SURFACES SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA Abstract. Let X and Y be supersingular K3 surfaces defined over an alge- braically closed field. Suppose that the sum of their Artin invariants is 11. Then there exists a certain duality between their N´eron-Severi lattices. We in- vestigate geometric consequences of this duality. As an application, we classify genus one fibrations on supersingular K3 surfaces with Artin invariant 10 in characteristic 2 and 3, and give a set of generators of the automorphism group of the nef cone of these supersingular K3 surfaces. The difference between the automorphism group of a supersingular K3 surface X and the automorphism group of its nef cone is determined by the period of X. We define the notion of genericity for supersingular K3 surfaces in terms of the period, and prove the existence of generic supersingular K3 surfaces in odd characteristics for each Artin invariant larger than 1. 1. Introduction A K3 surface X defined over an algebraically closed field k is said to be super- singular (in the sense of Shioda) if the rank of its N´eron-Severi lattice SX is 22. Supersingular K3 surfaces exist only when the base field k is of positive charac- teristic. Let X be a supersingular K3 surface in characteristic p > 0. Artin [1] proved that the discriminant group of SX is a p-elementary abelian group of rank 2σ, where σ is a positive integer such that σ ≤ 10. This integer σ is called the Artin invariant of X. The isomorphism class of the lattice SX depends only on p and σ (Rudakov and Shafarevich [19]). Moreover supersingular K3 surfaces with Artin invariant σ form a (σ − 1)-dimensional family, and a supersingular K3 surface with Artin invariant 1 in characteristic p is unique up to isomorphisms (Ogus [16], [17], Rudakov and Shafarevich [19]). Recently many studies of supersingular K3 surfaces in small characteristics with Artin invariant 1 have appeared. For example, for p = 2, Dolgachev and Kondo [6], Katsura and Kondo [9], Elkies and Schutt [8]; for p = 3, Katsura and Kondo [10], Kondo and Shimada [12], Sengupta [20]; and for p = 5, Shimada [24]. On the other hand, geometric properties of supersingular K3 surfaces with big Artin invariant are not so much known (e.g. Rudakov and Shafarevich [18], [19], Shioda [25], Shimada [22], [23]). In this paper, we present some methods to investigate supersingular K3 surfaces with big Artin invariant by means of the following simple observation. Let Xp,σ be 2000 Mathematics Subject Classification. 14J28, 14G17. The first author was partially supported by JSPS Grant-in-Aid for Scientific Research (S) No.22224001. The second author was partially supported by JSPS Grants-in-Aid for Scientific Research (B) No.20340002. 1 2 SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA a supersingular K3 surface in characteristic p with Artin invariant σ, and let Sp,σ denote its N´eron-Severi lattice. Lemma 1.1. Suppose that σ + σ′ = 11. Then Sp,σ′ is isomorphic to S∨ S∨ p,σ(p) is the lattice obtained from the dual lattice S∨ symmetric bilinear form with p. p,σ(p), where p,σ of Sp,σ by multiplying the Lemma 1.1 is proved in Section 3. We use this duality between Sp,σ and Sp,σ′ in the study of genus one fibrations and the automorphism groups of supersingular K3 surfaces. First, we apply Lemma 1.1 to the classification of genus one fibrations. Note that the N´eron-Severi lattice SY of a K3 surface Y is a hyperbolic lattice. The orthogonal group O(SY ) of SY contains the stabilizer subgroup O+(SY ) of a positive cone of SY ⊗ R as a subgroup of index 2. Definition 1.2. Let Y be a K3 surface, and let φ : Y → P1 be a genus one fibration. We denote by fφ ∈ SY the class of a fiber of φ. Let ψ : Y → P1 be another genus one fibration on Y . We say that φ and ψ are Aut-equivalent if there exist g ∈ Aut(Y ) and ¯g ∈ Aut(P1) such that φ ◦ g = ¯g ◦ ψ holds, while we say that φ and ψ are lattice equivalent if there exists g ∈ O+(SY ) such that f g φ = fψ. We denote by E(Y ) the set of lattice equivalence classes of genus one fibrations on Y , and by [φ] ∈ E(Y ) the lattice equivalence class containing φ. Many combinatorial properties of a genus one fibration φ : Y → P1 depend only on the lattice equivalence class [φ]. See Proposition 4.1. Moreover, when σ = 10, the classification of genus one fibrations by Aut-equivalence seems to be too fine (see Remark 7.6). Therefore, we concentrate upon the study of lattice equivalence classes. Using Lemma 1.1, we prove the following: Theorem 1.3. Suppose that σ + σ′ = 11. Then there exists a canonical one-to-one correspondence [φ] 7→ [φ′] between E(Xp,σ) and E(Xp,σ′ ). We say that a genus one fibration is Jacobian if it admits a section. Theorem 1.4. Suppose that a genus one fibration φ : Xp,σ → P1 is a Jacobian fibration, and let φ′ : Xp,σ′ → P1 be a genus one fibration on Xp,σ′ with σ′ = 11 − σ such that [φ′] ∈ E(Xp,σ′ ) corresponds to [φ] ∈ E(Xp,σ) by Theorem 1.3. Then φ′ does not admit a section. Elkies and Schutt [8] proved the following: Theorem 1.5 ([8]). Any genus one fibration on Xp,1 admits a section. Thus we obtain another proof of [7, Proposition 12.1]: Corollary 1.6 ([7]). There exist no Jacobian fibrations on Xp,10. By an ADE-type, we mean a finite formal sum of the symbols Ai (i ≥ 1), Di (j ≥ 4) and Ek (k = 6, 7, 8) with non-negative integer coefficients. For a genus one fibration φ : Y → P1 on a K3 surface Y , we have the ADE-type of reducible fibers of φ. This ADE-type depends only on the lattice equivalence class [φ] ∈ E(Y ) (see Proposition 4.1). Therefore we can use R[φ] to denote the ADE-type of the reducible fibers of φ. SUPERSINGULAR K3 SURFACES 3 From the classification of lattice equivalence classes of genus one fibrations of X2,1 by Elkies and Schutt [8], and that of X3,1 by Sengupta [20], we obtain the classification of lattice equivalence classes of genus one fibrations on X2,10 and X3,10. In particular, we obtain the list of ADE-types R[φ′] of the reducible fibers of genus one fibrations φ′ on X2,10 and X3,10. See Theorems 4.8 and 4.9. In Elkies and Schutt [8] and Sengupta [20] mentioned above, they also obtained explicit defining equations of the Jacobian fibrations. Note that the lattice equiva- lence classes of all extremal (quasi-) elliptic fibrations (i.e., Jacobian fibrations with Mordell-Weil rank zero) on supersingular K3 surfaces are classified in Shimada [21]. As the second application of Lemma 1.1, we investigate the automorphism group of the nef cone of a supersingular K3 surface. For a K3 surface Y , let Nef(Y ) ⊂ SY ⊗ R denote the nef cone. We denote by Aut(Nef(Y )) ⊂ O+(SY ) the group of isometries of SY that preserve Nef(Y ). Since Aut(Xp,σ) acts on Sp,σ faithfully (Rudakov and Shafarevich [19, Section 8, Proposition 3]), we have (1.1) Aut(Xp,σ) ⊂ Aut(Nef(Xp,σ)) ⊂ O+(Sp,σ). Using the description of Aut(X2,1) by Dolgachev and Kondo [6], and that of Aut(X3,1) by Kondo and Shimada [12], we give a set of generators of Aut(Nef(X2,10)) and Aut(Nef(X3,10)) in Theorems 6.4 and 6.8, respectively. Suppose that p is odd. We fix a lattice N isomorphic to Sp,σ. Then a quadratic space (N0, q0) of dimension 2σ over Fp is defined by N0 := pN ∨/pN and q0(px mod pN ) := px2 mod p (x ∈ N ∨). (1.2) We fix a marking η : N →∼ Sp,σ for a supersingular K3 surface X := Xp,σ defined over k. Then Aut(Nef(X)) acts on (N0, q0), and the period K(X,η) ⊂ N0 ⊗ k of the marked supersingular K3 surface (X, η) is defined as the Frobenius pull-back of the kernel of the natural homomorphism N ⊗ k → SX ⊗ k → H 2 DR(X/k) In virtue of Torelli theorem for supersingular K3 surfaces by (see Section 7). Ogus [16], [17], the subgroup Aut(X) of Aut(Nef(X)) is equal to the stabilizer sub- group of the period K(X,η). In particular, the index of Aut(Xp,σ) in Aut(Nef(Xp,σ)) is finite. On the other hand, the classification of 2-reflective lattices due to Nikulin [15] implies that Aut(Nef(Xp,σ)) is infinite. Hence, at least when p is odd, Aut(Xp,σ) is an infinite group. See Sections 5 and 7 for details. Moreover, Lieblich and Maulik [13] proved that, if p > 2, then Aut(Xp,σ) is finitely generated and its action on Nef(Xp,σ) has a rational polyhedral fundamental domain. We say that a supersingular K3 surface X is generic if there exists a marking η : N →∼ SX such that the isometries of (N0, q0) that preserve the period K(X,η) ⊂ N0 ⊗ k are only scalar multiplications (see Definition 7.4). Using the surjectivity of the period mapping proved by Ogus [17], we prove the following: Theorem 1.7. Suppose that p is odd and σ > 1. Then there exist an algebraically closed field k and a supersingular K3 surface X with Artin invariant σ defined over k that is generic. We observe that, if X3,10 is generic, then the index of Aut(X3,10) in Aut(Nef(X3,10)) is very large (see Remark 7.6). An analogous result for a generic complex Enriques surface was obtained by Barth and Peters [2]. 4 SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA This paper is organized as follows. In Section 2, we fix notation and terminologies about lattices and K3 surfaces. In Section 3, Lemma 1.1 is proved by means of the fundamental results of Rudakov and Shafarevich [19] on the N´eron-Severi lattices of supersingular K3 surfaces. In Section 4, we study genus one fibrations on supersingular K3 surfaces, and prove Theorems 1.3 and 1.4. Moreover, the bijections E(Xp,1) ∼= E(Xp,10) for p = 2 and 3 are given explicitly in Tables 4.1 and 4.2. In Section 5, we review the classical method to investigate the orthogonal group of a hyperbolic lattice by means of a chamber decomposition of the associated hyperbolic space, and fix some notation and terminologies. We then apply this method to the nef cone of a supersingular K3 surface. In Section 6, we give a set of generators of Aut(Nef(X2,10)) and Aut(Nef(X3,10)). In Section 7, we review the theory of the period mapping and Torelli theorem for supersingular K3 surfaces in odd characteristics due to Ogus [16], [17], and describe the relation between Aut(Xp,σ) and Aut(Nef(Xp,σ)). In Section 8, we prove Theorem 1.7. Convention. We use Aut to denote automorphism groups of lattice theoretic objects, and Aut to denote automorphism groups of geometric objects on K3 sur- faces. 2. Preliminaries 2.1. Lattices. A Q-lattice is a free Z-module L of finite rank equipped with a non- degenerate symmetric bilinear form h·, ·iL : L × L → Q. We omit the subscript L in h·, ·iL if no confusions will occur. If h·, ·iL takes values in Z, we say that L is a lattice. For x ∈ L ⊗ R, we call x2 := hx, xi the norm of x. A vector in L ⊗ R of norm n is sometimes called an n-vector. A lattice L is said to be even if x2 ∈ 2Z holds for any x ∈ L. Let L be a free Z-module of finite rank. A submodule M of L is primitive if L/M is torsion free. A non-zero vector v ∈ L is primitive if the submodule of L generated by v is primitive. Let L be a Q-lattice of rank r. For a non-zero rational number m, we denote by L(m) the free Z-module L with the symmetric bilinear form hx, yiL(m) := mhx, yiL. The signature of L is the signature of the real quadratic space L ⊗ R. We say that L is negative definite if the signature of L is (0, r), and L is hyperbolic if the signature is (1, r − 1). A Gram matrix of L is an r × r matrix with entries hei, eji, where {e1, . . . , er} is a basis of L. The determinant of a Gram matrix of L is called the discriminant of L. For an even lattice L, the set of (−2)-vectors is denoted by R(L). A negative definite even lattice L is called a root lattice if L is generated by R(L). Let R be an ADE-type. The root lattice of type R is the root lattice whose Gram matrix is the Cartan matrix of type R. Suppose that L is negative definite. By the ADE-type of R(L), we mean the ADE-type of the root sublattice hR(L)i of L generated by R(L). (See, for example, Bourbaki [4] for the classification of root lattices.) Let L be an even lattice and let L∨ := Hom(L, Z) be identified with a submodule of L ⊗ Q with the extended symmetric bilinear form. We call this Q-lattice L∨ the dual lattice of L. The discriminant group of L is defined to be the quotient L∨/L, and is denoted by AL. We define the discriminant quadratic form of L qL : AL → Q/2Z SUPERSINGULAR K3 SURFACES 5 by qL(x mod L) := x2 mod 2Z. The order of AL is equal to the discriminant of L up to sign. We say that L is unimodular if AL is trivial, while L is p-elementary if AL is p-elementary. An even 2-elementary lattice L is said to be of type I if qL(x mod L) ∈ Z/2Z holds for any x ∈ L∨. Note that L is p-elementary if and only if pG−1 L is an integer matrix, where GL is a Gram matrix of L. Let O(L) denote the orthogonal group of a lattice L, that is, the group of iso- morphisms of L preserving h·, ·iL. We assume that O(L) acts on L from right, and the action of g ∈ O(L) on v ∈ L ⊗ R is denoted by v 7→ vg. Similarly O(qL) denotes the group of isomorphisms of AL preserving qL. There is a natural homomorphism O(L) → O(qL). Let L be a hyperbolic lattice. A positive cone of L is one of the two connected components of { x ∈ L ⊗ R x2 > 0 }. Let PL be a positive cone of L. We denote by O+(L) the group of isometries of L that preserve PL. We have O(L) = O+(L) × {±1}. For a vector v ∈ L ⊗ R with v2 < 0, we put (v)⊥ := { x ∈ PL hx, vi = 0 }, which is a real hyperplane of PL. An isometry g ∈ O+(L) is called a reflection with respect to v or a reflection into (v)⊥ if g is of order 2 and fixes each point of (v)⊥. An element r of R(L) defines a reflection sr : x 7→ x + hx, rir with respect to r. We denote by W (L) the subgroup of O+(L) generated by the set of these reflections {sr r ∈ R(L)}. It is obvious that W (L) is normal in O+(L). 2.2. K3 surfaces. Let Y be a K3 surface, and let SY denote the N´eron-Severi lattice of Y . A smooth rational curve on Y is called a (−2)-curve. We denote by P(Y ) ⊂ SY ⊗ R the positive cone containing an ample class of Y . Recall that the nef cone Nef(Y ) of Y is defined by Nef(Y ) := { x ∈ SY ⊗ R hx, [C]i ≥ 0 for any curve C on Y }, where [C] ∈ SY is the class of a curve C ⊂ Y . Then Nef(Y ) is contained in the closure P(Y ) of P(Y ) in SY ⊗ R. We put Nef ◦(Y ) := Nef(Y ) ∩ P(Y ) = { x ∈ Nef(Y ) x2 > 0 }. The following is well-known. See, for example, Rudakov and Shafarevich [19, Sec- tion 3]. Proposition 2.1. (1) We have Nef(Y ) = { x ∈ SY ⊗ R hx, [C]i ≥ 0 for any (−2)-curve C on Y }. (2) If v ∈ SY is contained in P(Y ), then there exists g ∈ W (SY ) such that vg ∈ Nef(Y ). 3. N´eron-Severi lattices of supersingular K3 surfaces Let Xp,σ be a supersingular K3 surface with Artin invariant σ in characteristic p > 0. Then the isomorphism class of the N´eron-Severi lattice Sp,σ of Xp,σ depends only on p and σ, and is characterized as follows (see Rudakov-Shafarevich [19, Sections 3,4 and 5] for the proof). 6 SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA Theorem 3.1 ([19]). (1) The lattice Sp,σ is an even hyperbolic p-elementary lattice of rank 22 with discriminant −p2σ. Moreover, S2,σ is of type I. (2) Suppose that N is an even hyperbolic p-elementary lattice of rank 22 with discriminant −p2σ. When p = 2, we further assume that N is of type I. Then N is isomorphic to Sp,σ. Using this theorem, we can prove Lemma 1.1 easily. , and that S∨ p,σ of S∨ p,σ ⊂ Sp,σ. Therefore S∨ Proof of Lemma 1.1 . It is enough to show that S∨ lattice of discriminant −p2σ′ elementary, we have pS∨ a Gram matrix of Sp,σ. Then the determinant of the Gram matrix pG−1 is equal to p22 · det(Gp,σ)−1 = −p2σ′ −p2σ′ Suppose that p is odd. Then, for any ξ ∈ S∨ hpξ, pξiSp,σ = phξ, ξiS∨ p = 2. Then, for any ξ ∈ S∨ I. Therefore S∨ 2,σ(2) is even. Moreover, for any η ∈ (S∨ hη, ηiS2,σ (1/2) ∈ Z, because S2,σ is even. Therefore S∨ p,σ(p) is an even p-elementary 2,σ(2) is of type I. Since Sp,σ is p- p,σ(p) is a lattice. Let Gp,σ be p,σ(p) p,σ(p) is p,σ(p) is p-elementary. p,σ, we have pξ ∈ Sp,σ and hence p,σ(p) is even. Suppose that ∈ Z, because S2,σ is of type 2,σ(2))∨ = S2,σ(1/2), we have 2,σ(2) is of type I. (cid:3) . Therefore the discriminant of S∨ . Since p(pG−1 p,σ)−1 = Gp,σ is an integer matrix, S∨ p,σ (p) is even. Therefore S∨ 2,σ, we have hξ, ξiS∨ 2,σ Corollary 3.2. Suppose that σ + σ′ = 11. Then there exists an embedding of Z- p,σ(p) ∼= Sp,σ′ . modules j : Sp,σ ֒→ Sp,σ′ that induces an isomorphism of lattices S∨ This embedding induces an isomorphism j∗ : O(Sp,σ) →∼ O(Sp,σ′ ). Moreover such an embedding j is unique up to compositions with elements of O(Sp,σ′ ). Remark 3.3. Suppose that v ∈ Sp,σ satisfies v2 ≥ 0. Then, by Proposition 2.1(2), we can choose j : Sp,σ ֒→ Sp,σ′ in Corollary 3.2 in such a way that j(v) ∈ Nef(Xp,σ′ ). 4. Genus one fibrations Let Y be a K3 surface defined over an algebraically closed field of arbitrary characteristic. Recall that fφ ∈ SY is the class of a fiber of a genus one fibration φ : Y → P1, E(Y ) is the set of lattice equivalence classes of genus one fibrations on Y , and [φ] ∈ E(Y ) is the class containing φ. We summarize properties of a genus one fibration φ : Y → P1 that depends only on the class [φ]. See Sections 3 and 4 of Rudakov and Shafarevich [19], and Shioda [26] for the proof. (1) The fibration φ admits a section if and only if there exists a (−2)-vector z ∈ SY such that hfφ, zi = 1. (2) Note that fφ ∈ SY is primitive of norm 0, and that hfφi⊥/hfφi is an even negative definite lattice, where hfφi⊥ is the orthogonal complement in SY of the lattice hfφi of rank 1 generated by fφ. The ADE-type of the reducible fibers of φ is equal to the ADE-type of the set R(hfφi⊥/hfφi) of (−2)-vectors in hfφi⊥/hfφi. (3) Suppose that φ admits a section Z ⊂ Y . Then fφ and [Z] ∈ SY generate an even unimodular hyperbolic lattice Uφ of rank 2 in SY . Let Kφ denote the orthog- onal complement of Uφ in SY . We have an orthogonal direct-sum decomposition SY = Uφ ⊕ Kφ, SUPERSINGULAR K3 SURFACES 7 and the lattice hfφi⊥/hfφi is isomorphic to Kφ. Then the Mordell-Weil group of φ is isomorphic to Kφ/hR(Kφ)i, where hR(Kφ)i is the root sublattice of Kφ generated by the (−2)-vectors in Kφ. (4) In characteristic 2 or 3, φ is quasi-elliptic if and only if hR(Kφ)i is p- elementary of rank 20. As a corollary, we obtain the following: Proposition 4.1. Suppose that genus one fibrations φ : Y → P1 and ψ : Y → P1 on Y are lattice-equivalent. Then the following hold. (1) The fibration φ admits a section if and only if so does ψ. (2) The ADE-type of the reducible fibers of φ is equal to that of ψ. (3) Suppose that φ and ψ admit a section. Then the Mordell-Weil groups for φ and for ψ are isomorphic. (4) In characteristic 2 or 3, the fibration φ is quasi-elliptic if and only if so is ψ. Definition 4.2. For a hyperbolic lattice S, we put eE(S) := {v ∈ S ⊗ Q v 6= 0, v2 = 0}/Q× and E(S) := eE(S)/O(S). Remark 4.3. Let a positive cone PS of S be fixed. Then each element of eE(S) is represented by a unique non-zero primitive vector v ∈ S of norm 0 that is contained in the closure P S of PS in S ⊗ R. In Sections 3 and 4 of Rudakov and Shafarevich [19], the following is proved: Proposition 4.4. Let v be a non-zero vector of SY . Then there exists a genus one fibration φ : Y → P1 such that v = fφ if and only if v is primitive, v2 = 0, and v ∈ Nef(Y ). Combining Propositions 2.1, 4.4 and Remark 4.3, we obtain the following: Corollary 4.5. The map φ 7→ fφ induces a bijection from E(Y ) to E(SY ). From now on, we work over an algebraically closed field of characteristic p > 0. Proof of Theorem 1.3. Consider the embedding j : Sp,σ ֒→ Sp,σ′ in Corollary 3.2. Then j is unique up to O(Sp,σ′), induces a bijection from eE(Sp,σ) to eE(Sp,σ′ ), and induces an isomorphism O(Sp,σ) ∼= O(Sp,σ′ ). Hence it induces a canonical bijection from E(Sp,σ) to E(Sp,σ′ ). (cid:3) We denote this canonical one-to-one correspondence from E(Xp,σ) to E(Xp,σ′ ) by [φ] 7→ [φ′]. Remark 4.6. Let a genus one fibration φ : Xp,σ → P1 be given, and let φ′ : Xp,σ′ → P1 be a representative of [φ′]. Then we can choose the embedding j : Sp,σ ֒→ Sp,σ′ p,σ(p) ∼= Sp,σ′ in such a way that j(fφ) is a scalar multiple of fφ′ by a inducing S∨ positive integer. Theorem 4.7. Suppose that a genus one fibration φ : Xp,σ → P1 admits a section. : Xp,σ′ → P1 does not admit a Then the corresponding genus one fibration φ′ section. Moreover the ADE-type of the reducible fibers of φ′ is equal to the ADE- type of R(K ∨ φ (p)). 8 SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA No. RN 4A5 + D4 6D4 2A7 + 2D5 2A9 + D6 4D6 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 R[φ] 4A5 5D4 2A7 + D5 2A1 + 2A9 2A1 + 3D6 A11 + D7 A11 + D7 + E6 A11 + D7 + E6 A3 + A11 + E6 4E6 3D8 A15 + D9 A17 + E7 D10 + 2E7 D10 + 2E7 2D12 D16 + E8 D16 + E8 3E8 D24 3E6 D4 + 2D8 A15 + D5 3A1 + A17 3A1 + D10 + E7 D6 + 2E7 D8 + D12 D4 + D16 D12 + E8 D4 + 2E8 D20 σ = 1 MWtor [3, 6] [2, 2, 2, 2] [8] [10] [2, 2, 2] [4] [6] [3] [2, 2] [4] [6] [2, 2] [2] [2] [2] [1] [1] [1] σ = 10 rank(MW) R[φ′] 0 0 1 0 0 2 0 2 0 0 0 0 0 0 0 0 0 0 0 0 A1 2A1 2A1 A2 3A1 A2 4A1 5A1 A3 A3 6A1 8A1 D4 12A1 D4 20A1 Table 4.1. Genus one fibrations on X2,1 and X2,10 Proof. Let z ∈ Sp,σ be the class of a section of φ. We choose j : Sp,σ ֒→ Sp,σ′ as in Remark 4.6. Since U ∨ φ = Uφ, we see that j(fφ) is primitive in Sp,σ′ and hence j(fφ) = fφ′ . We have an isomorphism Sp,φ′ ∼= Uφ(p) ⊕ K ∨ φ (p) such that fφ′ and j(z) form a basis of Uφ(p). Since there are no vectors v ∈ Uφ(p) ⊕ K ∨ φ (p) with hv, fφ′i = 1, the fibration φ′ does not admit a section. Moreover the lattice hfφ′i⊥/hfφ′i is isomorphic to K ∨ (cid:3) φ (p). The list of lattice equivalence classes of genus one fibrations on X2,1 and X3,1 were obtained by Elkies and Schutt [8] and by Sengupta [20], respectively. From their results, we obtain the following results on supersingular K3 surfaces with Artin invariant 10: Theorem 4.8. There exist 18 lattice equivalence classes of genus one fibrations on X2,10. The ADE-type R[φ′] of the reducible fibers of each [φ′] ∈ E(X2,10) is given at the last column of Table 4.1. Theorem 4.9. There exist 52 lattice equivalence classes of genus one fibrations on X3,10. The ADE-type R[φ′] of the reducible fibers of each [φ′] ∈ E(X3,10) is given at the last column of Table 4.2. In Table 4.1 (resp. Table 4.2), the lists E(X2,1) and E(X2,10) (resp. E(X3,1) and E(X3,10)) are presented. Two lattice equivalence classes in the same row are the pair of [φ] ∈ E(Xp,1) and its corresponding partner [φ′] ∈ E(Xp,10). The ADE-type R[φ] of the reducible fibers of φ, and the torsion MWtor and the rank of the Mordell- Weil group of φ are also given. (Recall that φ is Jacobian for any [φ] ∈ E(Xp,1) by Elkies and Schutt [8].) The meaning of the entry RN is explained in the proof of Theorems 4.8 and 4.9. No. RN 3A8 3A8 4D6 4D6 2A9 + D6 2A9 + D6 2A9 + D6 2A9 + D6 4E6 4E6 12A2 8A3 6A4 6D4 4A5 + D4 4A5 + D4 4A5 + D4 4A6 4A6 2A7 + 2D5 2A7 + 2D5 2A7 + 2D5 2A7 + 2D5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 A11 + D7 + E6 25 A11 + D7 + E6 26 A11 + D7 + E6 27 A11 + D7 + E6 28 A11 + D7 + E6 29 A11 + D7 + E6 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 A15 + D9 A15 + D9 A15 + D9 A17 + E7 A17 + E7 A17 + E7 D10 + 2E7 D10 + 2E7 D10 + 2E7 D10 + 2E7 D16 + E8 D16 + E8 D16 + E8 A24 D24 2A12 2A12 3D8 3D8 2D12 2D12 3E8 3E8 SUPERSINGULAR K3 SURFACES 9 R[φ] 10A2 6A3 2A1 + 4A4 4D4 A2 + 3A5 3A5 + D4 2A2 + 2A5 + D4 3A6 2A3 + 2A6 4A1 + 2A7 A1 + A7 + 2D5 2A1 + A4 + A7 + D5 2A4 + 2D5 A2 + 2A8 2A5 + A8 3D6 2A3 + 2D6 2A9 A3 + A9 + D6 A3 + A6 + A9 2A6 + D6 A2 + 3E6 4A2 + 2E6 A2 + A11 + D7 A11 + E6 2A2 + A11 + D4 A5 + D7 + E6 2A2 + A8 + D7 A8 + D4 + E6 A6 + A12 2A9 2A1 + 2D8 2D5 + D8 A3 + A15 A9 + D9 A12 + D6 A2 + A17 A11 + E7 A5 + A14 A2 + D10 + E7 2A5 + D10 D4 + 2E7 A5 + D7 + E7 D6 + D12 2D9 2A2 + 2E8 2E6 + E8 2A2 + D16 D10 + E8 D13 + E6 A18 D18 σ = 1 MWtor [3, 3, 3, 3] [4, 4] [5] [2, 2] [3] [2, 6] [2] [7] [1] [2, 4] [4] [2] [1] [3] [1] [2, 2] [2, 2] [5] [2] [1] [1] [3] [3, 3] [4] [3] [6] [1] [1] [1] [1] [1] [2, 2] [2] [4] [1] [1] [3] [1] [1] [2] [2, 2] [2] [2] [2] [1] [1] [1] [2] [1] [1] [1] [1] rank(MW) σ = 10 R[φ′] 0 2 2 4 3 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 0 0 0 3 1 2 1 2 2 2 2 2 2 2 2 1 2 1 1 0 2 1 2 2 0 0 0 2 1 2 2 0 0 0 0 0 A1 0 A1 0 0 A1 0 0 A1 0 2A1 0 2A1 A1 0 0 A2 0 A2 2A1 0 A1 0 0 A1 0 2A1 0 2A1 A1 0 A1 + A2 A1 0 A1 + A2 0 2A1 0 2A1 0 2A2 0 2A2 2A1 0 A1 2A1 Table 4.2. Genus one fibrations on X3,1 and X3,10 10 SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA Proof of Theorems 4.8 and 4.9. By Theorem 4.7, it is enough to calculate the ADE- type of R(K ∨ φ (p)) for p = 2, 3 and [φ] ∈ E(Xp,1). The lattices Kφ are calculated in Elkies and Schutt [8] and Sengupta [20]. We put T := the root lattice of type(D4 2A2 if p = 2, if p = 3. Then, for each [φ] ∈ E(Xp,1), there exist a Niemeier lattice Nφ and a primitive embedding of T into Nφ such that Kφ is isomorphic to the orthogonal complement of T in Nφ. The entry RN in Tables 4.1 and 4.2 indicates the ADE-type of R(Nφ). From a Gram matrix of Kφ, we can calculate the ADE-type of R(K ∨ φ (p)) by the algorithm described in [23, Section 4] or [24, Section 3]. (cid:3) Corollary 4.10. There exist no quasi-elliptic fibrations on X3,10. Remark 4.11. Rudakov and Shafarevich [19, Section 5] showed that there exists a quasi-elliptic fibration on X2,σ for any σ. The quasi-elliptic fibration on X2,10 (No. 18 of Table 4.1) was discovered by Rudakov and Shafarevich [18, Section 4]. 5. Chamber decomposition of a positive cone Let S be an even hyperbolic lattice, and let PS ⊂ S ⊗ R be a positive cone. In this section, we review a general method to find a set of generators of a subgroup of O+(S) by means of a chamber decomposition of PS, which was developed by Vinberg [27], Conway [5] and Borcherds [3]. Any real hyperplane in PS is written in the form (v)⊥ by some vector v ∈ S ⊗ R with negative norm. We denote by HS the set of real hyperplanes in PS, which is canonically identified with { v ∈ S ⊗ R v2 < 0 }/R×. For a subset V of {v ∈ S ⊗ R v2 < 0}, we denote by V ∗ ⊂ HS the image of V by v 7→ (v)⊥. A closed subset D of PS is called a chamber if the interior D◦ of D is non-empty and there exists a set ∆D of vectors v ∈ S ⊗ R with v2 < 0 such that D = { x ∈ PS hx, vi ≥ 0 for all v ∈ ∆D }. A hyperplane (v)⊥ of PS is called a wall of D if D◦ ∩ (v)⊥ = ∅ and D ∩ (v)⊥ contains an open subset of (v)⊥. When D is a chamber, we always assume that the set ∆D is minimal in the sense that, for any v ∈ ∆D, there exists a point x ∈ PS such that hx, vi < 0 and hx, v′i ≥ 0 for any v′ ∈ ∆D \ {v}, that is, the projection ∆D → ∆∗ D is bijective and every hyperplane (v)⊥ ∈ ∆∗ D is a wall of D. For a chamber D, we put Aut(D) := { g ∈ O+(S) Dg = D }. A chamber D is said to be fundamental if the following hold: (i) PS is the union of all Dg, where g runs through O+(S), and (ii) if D◦ ∩ Dg 6= ∅, then g ∈ Aut(D). Let F be a family of hyperplanes in PS with the following properties: (a) F is locally finite in PS , and (b) F is invariant under the action of O+(S) on HS. SUPERSINGULAR K3 SURFACES 11 Then the closure of each connected component of PS \[F (v)⊥ is a chamber, which we call an F -chamber. Suppose that D is an F -chamber. Then Dg is also an F -chamber for any g ∈ O+(S) by the property (b) of F , and D satisfies the property (ii) in the definition of fundamental chambers. Moreover, D satisfies the property (i) if and only if every F -chamber is equal to Dg for some g ∈ O+(S). For each wall (v)⊥ ∈ ∆∗ D of an F -chamber D, there exists a unique F -chamber D′ distinct from D such that D ∩D′ ∩(v)⊥ contains an open subset of (v)⊥. We say that D′ is adjacent to D along (v)⊥, and that (v)⊥ is the wall between the adjacent chambers D and D′. Proposition 5.1. An F -chamber D is fundamental if and only if, for each v ∈ ∆D, there exists gv ∈ O+(S) such that Dgv is adjacent to D along (v)⊥. Proof. The 'only if ' part is obvious. We prove the 'if ' part. It is enough to show that, for an arbitrary F -chamber D′, there exists g ∈ O+(S) such that D′ = Dg. Since the family F of hyperplanes is locally finite in PS, there exists a finite chain of F -chambers D0 = D, D1, . . . , DN = D′ such that Di and Di+1 are adjacent. We show, by induction on N , that there exists a sequence of vectors v1, . . . , vN in ∆D such that Di = Dgvi ···gv1 holds for i = 1, . . . , N . The case N = 0 is trivial. Suppose that N > 0. Let (w)⊥ be the wall between DN −1 and DN , and let vN ∈ ∆D be the vector such that the wall (vN )⊥ of D is mapped to the wall (w)⊥ of DN −1 by gvN −1 · · · gv1. Then we have DN = DgvN ···gv1 . (cid:3) Remark 5.2. If an F -chamber is fundamental, then any F -chamber is fundamental. Let G be a subset of F that is invariant under the action of O+(S). Then G is locally finite, and any G-chamber is a union of F -chambers. If an F -chamber is fundamental, then any G-chamber is also fundamental. Proposition 5.3. Let D be an F -chamber and let C be a G-chamber such that D ⊂ C. Suppose that D is fundamental. For v ∈ ∆D, let gv ∈ O+(S) be an isometry such that Dgv is adjacent to D along (v)⊥. We put Then Aut(C) is generated by Aut(D) and Γ. Γ := { gv v ∈ ∆D, (v)⊥ /∈ G }. Proof. If gv ∈ Γ, then Dgv is contained in C because the wall (v)⊥ between D and Dgv does not belong to G, and hence gv ∈ Aut(C). Therefore the subgroup hAut(D), Γi of O+(S) generated by Aut(D) and Γ is contained in Aut(C). To prove Aut(C) ⊂ hAut(D), Γi, it is enough to show that, for any g ∈ Aut(C), there exists a sequence γ1, . . . , γN of elements of Γ such that Dg = DγN ···γ1 . There exists a sequence of F -chambers D0 = D, D1, . . . , DN = Dg such that each Di is contained in C and that Di+1 is adjacent to Di for i = 0, . . . , N − 1. Suppose that we have constructed γ1, . . . , γi ∈ Γ such that Di = Dγi...γ1 holds. The wall (w)⊥ between Di and Di+1 does not belong to G. Let vi+1 be an element of ∆D such that the wall (vi+1)⊥ of D is mapped to the wall (w)⊥ of Di by γi . . . γ1. Since G is invariant under the action of O+(S), we have (vi+1)⊥ /∈ G and hence γi+1 := gvi+1 is an element of Γ. Then Di+1 = Dγi+1γi···γ1 holds. (cid:3) 12 SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA Remark 5.4. Let D and C be as in Proposition 5.3. Let v and v′ be elements of ∆D. Suppose that the wall (v)⊥ of D is mapped to the wall (v′)⊥ of D by h ∈ Aut(D). Then Dhgv′ h−1 D be a subset of ∆D such that the subset ∆′∗ D is a complete set of representatives of the orbit decomposition of ∆∗ D by the action of Aut(D). Then Aut(C) is generated by Aut(D) and {gv v ∈ ∆′ is adjacent to D along (v)⊥ . Let ∆′ D of ∆∗ D, (v)⊥ /∈ G}. Considering the case G = ∅, we obtain the following: If D is fundamental, then O+(S) is Corollary 5.5. Let D be an F -chamber. generated by Aut(D) and the isometries gv that map D to its adjacent chambers. Example 5.6. Recall that W (S) ⊂ O+(S) is the subgroup generated by {sr r ∈ R(S)}. Any R(S)∗-chamber is fundamental, because every r ∈ R(S) defines a reflection It follows that O+(S) is equal to the semi-direct product of W (S) and the sr. automorphism group Aut(D) of an R(S)∗-chamber D. In particular, we have Aut(D) ∼= O+(S)/W (S). Let L be an even unimodular hyperbolic lattice, and let ι : S ֒→ L be a primitive embedding. Let PL be the positive cone of L that contains ι(PS). We denote by Tι the orthogonal complement of S in L, and by v 7→ vS the orthogonal projection L ⊗ R → S ⊗ R . Since L is a submodule of S∨ ⊕ T ∨ ι , the image of L by v 7→ vS is contained in S∨. We assume the following: (5.1) the natural homomorphism O(Tι) → O(qTι ) is surjective. Then we have the following: Proposition 5.7. For any g ∈ O+(S), there exists g ∈ O+(L) such that ι(vg) = ι(v)g holds for any v ∈ S ⊗ R. Proof. See Nikulin [14, Proposition 1.6.1]. (cid:3) A hyperplane (r)⊥ of PL defined by a (−2)-vector r ∈ R(L) intersects ι(PS) if and only if r2 S < 0. We put R(L, ι) := { rS r ∈ R(L) and r2 S < 0 } ⊂ S∨. Since Tι is negative definite, we have −2 ≤ r2 S for any r ∈ R(L). Since S∨ is discrete in S ⊗ R, the family of hyperplanes R(L, ι)∗ is locally finite in PS. By Proposition 5.7, if r ∈ R(L) satisfies rS ∈ R(L, ι), then, for any g ∈ O+(S), we have rg S = (rg)S ∈ R(L, ι). Therefore R(L, ι) is invariant under the action of O+(S). Note that R(L, ι) contains R(S), and that R(S) is obviously invariant under the action of O+(S). Therefore, by Proposition 5.3, we can obtain a set of generators of the automorphism group Aut(C) of an R(S)∗-chamber C if we find an R(L, ι)∗-chamber D contained in C, show that D is fundamental, calculate the group Aut(D), and find isometries of S that map D to its adjacent chambers. Let L26 denote an even hyperbolic unimodular lattice of rank 26, which is unique up to isomorphisms. The walls of an R(L26)∗-chamber D ⊂ L26 ⊗ R and the group Aut(D) ⊂ O+(L26) were determined by Conway [5]. Then Borcherds [3] determined the structure of O+(S) for some even hyperbolic lattices S of rank < 26 by embedding S into L26 in such a way that Tι is a root lattice. SUPERSINGULAR K3 SURFACES 13 Kondo [11] first applied the Borcherds method to the study of the automorphism group of a generic Jacobian Kummer surface. Later Dolgachev and Kondo [6] applied the Conway-Borcherds method to the study of the automorphism group of X2,1, and Kondo and Shimada [12] applied it to X3,1. We say that an even hyperbolic lattice S is 2-reflective if the index of W (S) in O+(S) is finite, or equivalently, if the automorphism group of an R(S)∗-chamber is finite (see Example 5.6). Nikulin [15] classified all 2-reflective lattices of rank ≥ 5. It turns out that there are no 2-reflective lattices of rank > 19. Let Y be a K3 surface with the N´eron-Severi lattice SY and the positive cone P(Y ) containing an ample class. Then the closed subset Nef ◦(Y ) = Nef(Y ) ∩ P(Y ) of P(Y ) is an R(SY )∗-chamber by Proposition 2.1(1), and hence we have Aut(Nef(Y )) = Aut(Nef ◦(Y )) ∼= O+(SY )/W (SY ). Therefore Aut(Nef(Xp,σ)) is infinite for any supersingular K3 surface Xp,σ. 6. The groups Aut(Nef(X2,10)) and Aut(Nef(X3,10)) 6.1. The group Aut(Nef(X2,10)). By Lemma 1.1, the result of Dolgachev and Kondo [6], and the method of the previous section, we obtain a set of generators of Aut(Nef(X2,10)). First we recall the results of [6]. As a projective model of X2,1, we consider the minimal resolution X of the inseparable double cover Y → P2 of P2 defined by w2 = x0x1x2(x3 0 + x3 1 + x3 2). Note that the projective plane P2(F4) defined over F4 contains 21 points p1, . . . , p21 and 21 lines ℓ1, . . . , ℓ21. The inseparable double cover Y has 21 ordinary nodes over the 21 points in P2(F4) and hence X has 21 disjoint (−2)-curves. We denote by e1, . . . , e21 ∈ S2,1 the classes of these (−2)-curves, by h ∈ S2,1 the class of the pullback of a line on P2, and by f1, . . . , f21 ∈ S2,1 the classes of the proper transforms of the 21 lines in P2(F4). Then S2,1 is generated by the (−2)-vectors e1, . . . , e21, f1, . . . , f21. The vector has the property wM := 1 3 21Xi=1 (ei + fi) wM ∈ SX , w2 M = 14, hwM , eii = hwM , fii = 1. The complete linear system associated with the line bundle corresponding to wM defines an embedding of X into P2 × P2, and its image XM ⊂ P2 × P2 is defined by (x0y2 0 + x1y2 1 + x2y2 x2 0y0 + x2 1y1 + x2 2 = 0, 2y2 = 0. Six points on P2(F4) are said to be general if no three points of them are collinear. There exist 168 sets of general six points in P2(F4). If I = {pi1, . . . , pi6 } is a set of general six points, then the (−1)-vector 1 2 (ei1 + · · · + ei6 ) cI := h − is contained in S∨ 2,1. Note that each cI defines a reflection x 7→ x + 2hx, cI icI 14 SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA in O+(S2,1) because cI ∈ S∨ 2,1. Let P(X2,1) be the positive cone of S2,1 containing an ample class. and let ∆(X2,1) be the set consisting of e1, . . . , e21, f1, . . . , f21 and the (−1)-vectors cI defined above. We define a chamber D(X2,1) in P (X2,1) by D(X2,1) := { x ∈ P (X2,1) hx, vi ≥ 0 for all v ∈ ∆(X2,1) }. Then, for each v ∈ ∆(X2,1), the hyperplane (v)⊥ is a wall of D(X2,1). Moreover the ample class wM is contained in the interior of D(X2,1). Recall that L26 is the even unimodular hyperbolic lattice of rank 26. There exists a primitive embedding ι : S2,1 ֒→ L26, which is unique up to O(L26). The orthogonal complement Tι of S2,1 in L26 is isomorphic to the root lattice of type D4, and hence satisfies the hypothesis (5.1). Proposition 6.1. The chamber D(X2,1) is an R(L26, ι)∗-chamber contained in the R(S2,1)∗-chamber Nef ◦(X2,1). An isometry g ∈ O+(S2,1) belongs to Aut(D(X2,1)) if and only if wg M = wM . Thus we can apply Proposition 5.3 to the pair of chambers D(X2,1) and Nef ◦(X2,1) for the study of Aut(Nef(X2,1)) and Aut(X2,1). We have the following elements in Aut(X2,1) and O+(S2,1). Since Aut(X2,1) is naturally embedded in O+(S2,1), we use the same letter to denote an element of Aut(X2,1) and its image in O+(S2,1). • The action of PGL(3, F4) on P2 induces automorphisms of the inseparable double cover Y of P2, and hence automorphisms of X2,1. Their action on S2,1 preserves D(X2,1). • The interchange of the two factors of P2 × P2 preserves XM ⊂ P2 × P2, and hence it induces an involution sw ∈ Aut(X2,1), which we call the switch. Its action on S2,1 preserves D(X2,1). • For each set I of general six points in P2(F4), the linear system of plane curves of degree 5 that pass through the points of I and are singular at each point of I defines a birational involution of P2, and this involution lifts to an involution of Y . Hence we obtain an involution CrI ∈ Aut(X2,1), which we call a Cremona automorphism of X2,1. The action of CrI on S2,1 is the reflection with respect to cI ∈ S∨ 2,1. • The Frobenius action of Gal(F4/F2) on XM induces an isometry Fr of S2,1, which preserves D(X2,1). • The (−2)-vectors ei and fi defines the reflections sei and sfi . By the reflections CrI , sei and sfi , we see that the chamber D(X2,1) is fundamental. Theorem 6.2 ([6]). (1) The projective automorphism group Aut(X2,1, wM ) of XM ⊂ P2 × P2 is generated by PGL(3, F4) and the switch sw. (2) The group Aut(D(X2,1)) is generated by Aut(X2,1, wM ) and Fr. (3) The automorphism group Aut(X2,1) is generated by Aut(X2,1, wM ) and the 168 Cremona automorphisms CrI . (4) The group Aut(Nef(X2,1)) is generated by Aut(X2,1) and Fr. (5) The group O+(S2,1) is generated by Aut(Nef(X2,1)) and the 21+21 reflections sei and sfi . We then study Aut(Nef(X2,10)). By Corollary 3.2, we have an embedding j : S2,1 ֒→ S2,10 SUPERSINGULAR K3 SURFACES 15 2,1(2) ∼= S2,10. Composing j with some element of W (S2,10) × {±1}, that induces S∨ we can assume that j(wM ) is contained in Nef(X2,10) (Proposition 2.1(2)). The isomorphism j∗ : O+(S2,1) →∼ O+(S2,10) induced by j is denoted by The j(R(L26, ι))∗-chamber j(D(X2,1)) is fundamental, and we have Aut(j(D(X2,1))) = Aut(D(X2,1))′. g 7→ g′. Lemma 6.3. The set j(R(L26, ι)) contains R(S2,10). Hence the j(R(L26, ι))∗- chamber j(D(X2,1)) is contained in the R(S2,10)∗-chamber Nef ◦(X2,10). Proof. It is enough to show that, if v ∈ S∨ that is, there exists u ∈ T ∨ ι the submodule L26 of S∨ L26/(S2,1⊕Tι) of (S∨ 2,1 satisfies v2 = −1, then v ∈ R(L26, ι), such that u2 = −1 and that u + v is contained in ι . By Nikulin [14, Proposition 1.4.1], the submodule ι )/(S2,1⊕Tι) = AS2,1 ⊕ATι is the graph of an isomorphism 2,1 ⊕ T ∨ 2,1⊕T ∨ qS2,1 ∼= −qTι. Hence it is enough to show that, for any ¯u ∈ ATι with qTι (¯u) = 1, there exists such that u2 = −1 and u mod Tι = ¯u. Since Tι is a root lattice of type u ∈ T ∨ ι D4, we can confirm this fact by direct computation. The set of (−1)-vectors in T ∨ ι consists of 24 vectors, and its image by the natural projection T ∨ ι → ATι is the set of all non-zero elements of ATι (cid:3) ∼= F2 2. The set of walls of j(D(X2,1)) is equal to {(j(ei))⊥ i = 1, . . . , 21} ∪ {(j(fi))⊥ i = 1, . . . , 21} ∪ {(j(cI ))⊥ I is a set of general six points}. Note that the 21 + 21 vectors j(ei) and j(fi) are of norm −4 and the 168 vectors j(cI ) are of norm −2. Note also that neither (j(ei))⊥ nor (j(fi))⊥ are contained in R(S2,10)∗, because there are no rational numbers λ such that (−4)λ2 = −2. By Proposition 5.3, Theorem 6.2 and Lemma 6.3, we obtain the following: Theorem 6.4. The group Aut(Nef(X2,10)) is generated by PGL(3, F4)′, sw′, Fr′, s′ ei and s′ fi . 6.2. The group Aut(Nef(X3,10)). By the same argument as above, we obtain a set of generators of Aut(Nef(X3,10)) from the result of Kondo and Shimada [12]. We consider the Fermat quartic surface 0 + x4 XFQ : x4 2 + x4 1 + x4 3 = 0 in characteristic 3. Then XFQ is isomorphic to X3,1. The surface XFQ contains 112 lines, and their classes l1, . . . , l112 span S3,1. We denote by hFQ ∈ S3,1 the class of a hyperplane section of XFQ. There exists a primitive embedding ι : S3,1 ֒→ L26, which is unique up to O(L26). The orthogonal complement Tι is isomorphic to the root lattice of type 2A2, and hence satisfies the hypothesis (5.1). We calculated an R(L26, ι)∗-chamber D(X3,1) that contains hFQ in its interior, and found 648 vectors uj ∈ S∨ 3,1 of norm −2/3 such that the walls of D(X3,1) consist of the 112 hyperplanes (li)⊥, the 648 hyper- planes (uj)⊥ and the 5184 hyperplanes (wk)⊥. Note that the R(L26, ι)∗-chamber 3,1 of norm −4/3, and 5184 vectors wk ∈ S∨ 16 SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA D(X3,1) is contained in the R(S3,1)∗-chamber Nef ◦(X3,1), because hFQ ∈ D(X3,1)◦. Moreover, since 28 hFQ =P li, the following holds: Proposition 6.5. An isometry g ∈ O+(S3,1) belongs to Aut(D(X3,1)) if and only if hg FQ = hFQ. We have the following elements in Aut(X3,1) and O+(S3,1). Note that, for a po- larization h ∈ S3,1 of degree 2, we have the deck transformation τ (h) ∈ Aut(X3,1) of the generically finite morphism X3,1 → P2 of degree 2 induced by the the complete linear system associated with h. • The subgroup PGU(4, F9) of PGL(4, k) = Aut(P3) acts on XFQ. Its action on S3,1 preserves D(X3,1). Moreover, the action of PGU(4, F9) on S∨ 3,1 is transitive on each of the set of 112 vectors li, the set of 648 vectors uj and the set of 5184 vectors wk. • There exists a polarization h648 ∈ S3,1 of degree 2 such that the deck trans- formation τ (h648) ∈ Aut(X3,1) maps D(X3,1) to an R(L26, ι)∗-chamber adjacent to D(X3,1) along one of the 648 walls (uj)⊥. • There exists a polarization h5184 ∈ S3,1 of degree 2 such that the deck trans- formation τ (h5184) ∈ Aut(X3,1) maps D(X3,1) to an R(L26, ι)∗-chamber adjacent to D(X3,1) along one of the 5184 walls (wk)⊥. • The Frobenius action of Gal(F9/F3) on XFQ gives rise to an element Fr ∈ Aut(D(X3,1)) of order 2. • We have the reflections sli with respect to the classes li of the 112 lines on XFQ. Thus D(X3,1) is fundamental, and hence we have the following: Theorem 6.6 ([12]). (1) The projective automorphism group Aut(X, hFQ) of the Fermat quartic surface XFQ ⊂ P3 is equal to PGU(4, F9). (2) The group Aut(D(X3,1)) is generated by Aut(X, hFQ) and Fr. (3) The automorphism group Aut(X3,1) is generated by Aut(X, hFQ) and the two involutions τ (h648) and τ (h5184). (4) The group Aut(Nef(X3,1)) is generated by Aut(X3,1) and Fr. (5) The group O+(S3,1) is generated by Aut(Nef(X3,1)) and the 112 reflections sli . By Corollary 3.2, we have an embedding j : S3,1 ֒→ S3,10 3,1(3) ∼= S3,10. By Proposition 2.1(2), we can assume that j(hFQ) is that induces S∨ contained in Nef(X3,10). The isomorphism j∗ : O+(S3,1) →∼ O+(S3,10) induced by j is denoted by g 7→ g′. The j(R(L26, ι))∗-chamber j(D(X3,1)) is fundamental, and Aut(j(D(X3,1))) is equal to Aut(D(X3,1))′. Lemma 6.7. The set j(R(L26, ι)) contains R(S3,10). Hence the j(R(L26, ι))∗- chamber j(D(X3,1)) is contained in the R(S3,10)∗-chamber Nef ◦(X3,10). ι such that u2 = −4/3 and that u + v is contained in L26 ⊂ S∨ 3,1 satisfies v2 = −2/3, then there exists Proof. It is enough to show that, if v ∈ S∨ u ∈ T ∨ ι . For this, it suffices to prove that, for any ¯u ∈ ATι with qTι(¯u) = −4/3, there exists such that u2 = −4/3 and u mod Tι = ¯u. Since Tι is a root lattice of type u ∈ T ∨ ι 2A2, we can confirm this fact by direct computation. (cid:3) 3,1 ⊕ T ∨ SUPERSINGULAR K3 SURFACES 17 The set of walls of j(D(X3,1)) is equal to {(j(li))⊥ i = 1, . . . , 112} ∪ {(j(uj))⊥ j = 1, . . . , 648} ∪ {(j(wk))⊥ k = 1, . . . , 5184}. Note that the vectors j(li) are of norm −6, the vectors j(uj) are of norm −4, and the vectors j(wk) are of norm −2. Note also that neither (j(li))⊥ nor (j(uj))⊥ are contained in R(S3,10)∗. By Proposition 5.3, Theorem 6.6 and Lemma 6.7, we obtain the following: Theorem 6.8. The group Aut(Nef(X3,10)) is generated by PGU(4, F9)′, Fr′, s′ li and τ (h648)′. 7. Torelli theorem for supersingular K3 surfaces We review the theory of period mapping and Torelli theorem for supersingular K3 surfaces in odd characteristics by Ogus [16], [17]. Throughout this section, we assume that p is odd. We summarize results on quadratic spaces over finite fields. Let Fq be a finite extension of Fp. There exist exactly two isomorphism classes of non-degenerate quadratic forms in 2σ variables x1, . . . , x2σ over Fq. They are represented by (7.1) f+ := x1x2 + · · · + x2σ−1x2σ, f− := x2 1 + cx1x2 + x2 and (7.2) where c is an element of Fq such that t2 + ct+ 1 ∈ Fq[t] is irreducible. The quadratic form f+ (resp. f−) is called neutral (resp. non-neutral ). We denote by O(F2σ q , fǫ) the group of the self-isometries of the quadratic space (F2σ 2 + x3x4 + · · · + x2σ−1x2σ, q , fǫ). Let N be an even hyperbolic p-elementary lattice of rank 22 with discriminant −p2σ. We define a quadratic space (N0, q0) over Fp by (1.2). It is known that q0 is non-degenerate and non-neutral. We denote by O(N0, q0) the group of the self- isometries of (N0, q0). Note that the scalar multiplications in O(N0, q0) are only ±1. Let k be a field of characteristic p. We put where Fk is the Frobenius map of k. ϕ := idN0 ⊗ Fk : N0 ⊗ k → N0 ⊗ k, Definition 7.1. A subspace K of N0 ⊗ k with dim K = σ is said to be a charac- teristic subspace of (N0, q0) if K is totally isotropic with respect to the quadratic form q0 ⊗ k and dim(K ∩ ϕ(K)) = σ − 1 holds. Suppose that k is algebraically closed. Let X be a supersingular K3 surface with Artin invariant σ defined over k. An isomorphism η : N →∼ SX of lattices is called a marking of X. We fix a marking η of X. The composite of the marking η and the Chern class map SX → H 2 DR(X/k) defines a linear homo- morphism ¯η : N ⊗ k → H 2 DR(X/k). It is known that Ker ¯η is contained in N0 ⊗ k, and is totally isotropic with respect to q0 ⊗ k. We put K(X,η) := ϕ−1(Ker ¯η), 18 SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA and call K(X,η) the period of the marked supersingular K3 surface (X, η). Then it is proved by Ogus [16], [17] that K(X,η) is a characteristic subspace of (N0, q0). We denote by η∗ : O(SX ) →∼ O(N ) the isomorphism induced by the marking η, and let ¯η∗ : O(SX ) → O(N0, q0) be the composite of η∗ with the natural homomorphism O(N ) → O(N0, q0). As a corollary of Torelli theorem by Ogus [17, Corollary of Theorem II′′], we have the following: Corollary 7.2. Let η be a marking of X. Then, as a subgroup of O+(SX ), the automorphism group Aut(X) of X is equal to { g ∈ Aut(Nef(X)) K ¯η∗(g) (X,η) = K(X,η) }. In particular, the index of Aut(X) in Aut(Nef(X)) is at most O(N0, q0)/{±1}. Since Aut(Nef(X)) is infinite, we obtain the following: Corollary 7.3. The automorphism group Aut(X) is infinite. Definition 7.4. We say that a supersingular K3 surface X with Artin invariant σ is generic if there exists a marking η for X such that the subgroup (7.3) { γ ∈ O(N0, q0) K γ (X,η) = K(X,η) } of O(N0, q0) consists of only scalar multiplications ±1. If X is generic, then the subgroup (7.3) consists of only scalar multiplications for any marking η. The existence of generic supersingular K3 surfaces with Artin invariant > 1 (Theorem 1.7) is proved in the next section. Recall that ASX is the discriminant group of SX , and qSX : ASX → Q/2Z is the discriminant quadratic form. We will regard ASX as a 2σ-dimensional vector space over Fp. Note that the image of qSX is contained in (2/p)Z/2Z. We define ¯qSX : ASX → Fp by ¯qSX (x mod SX ) := p · qSX (x) mod p. Then we obtain a quadratic space (ASX , ¯qSX ) over Fp. Note that we can recover qSX from ¯qSX . We have natural homomorphisms (7.4) O(SX ) → O(qSX ) ∼= O(ASX , ¯qSX ) →→ PO(ASX , ¯qSX ) := O(ASX , ¯qSX )/{±1}. Let η : N ∨ →∼ S∨ X be the isomorphism induced by a marking η. Then the map px mod pN ∈ N0 7→ η(x) mod SX ∈ ASX (x ∈ N ∨) induces an isomorphism of quadratic spaces from (N0, q0) to (ASX , ¯qSX ). By Corol- lary 7.2, we obtain the following: Corollary 7.5. Suppose that X is generic. Then Aut(X) is equal to the kernel of the homomorphism obtained by restricting (7.4) to Aut(Nef(X)) ⊂ O(SX). Φ : Aut(Nef(X)) → PO(ASX , ¯qSX ) SUPERSINGULAR K3 SURFACES 19 Remark 7.6. Suppose that X3,10 is generic. By the standard algorithm of combina- torial group theory, we can obtain a finite set of generators of Aut(X3,10) from the generators of Aut(Nef(X3,10)) given in Theorem 6.8. However, a naive application of the algorithm would be inexecutable, because, when p = 3 and σ = 10, the order of O(N0, q0) is 236 · 390 · 56 · 73 · 112 · 133 · 17 · 19 · 37 · 412 · 61 · 73 · 193 · 547 · 757 · 1093 · 1181, which is about 7.886 × 1090. In fact, we can show that there exists a genus one fibration φ′ on X3,10 whose lattice equivalence class contains at least 6531840 Aut- equivalence classes, by giving the class fφ of a Jacobian fibration on X3,1 = XFQ explicitly. 8. Existence of generic supersingular K3 surfaces We prove Theorem 1.7. For the proof, we recall the construction by Ogus [16] of the scheme M parameterizing characteristic subspaces of the 2σ-dimensional quadratic space (N0, q0) over Fp. This scheme M plays the role of the period domain for supersingular K3 surfaces. We continue to assume that p is odd. Let Grass(ν, N0) denote the Grassmannian variety of ν-dimensional subspaces of N0, and let Isot(ν, q0) be the subscheme of Grass(ν, N0) parameterizing ν- dimensional totally isotropic subspaces of (N0, q0). We put Gen := Isot(σ, q0), where Gen is for "generatrix". Note that Isot(ν, q0) is defined over Fp for any ν. Let k be a field of characteristic p. For a subspace L of N0 ⊗ k with dimension ν, we denote by [L] the k-valued point of Grass(ν, N0) corresponding to L. We then have the following: (1) If ν < σ, then Isot(ν, q0) is geometrically connected. (2) The scheme Gen⊗Fp2 has two connected components Gen+ and Gen−, each of which is geometrically connected. Since q0 is non-neutral, the action of Gal(Fp2 /Fp) interchanges the two connected components. (3) Let K and K ′ be two σ-dimensional totally isotropic subspaces of (N0, q0)⊗ k. Suppose that dim(K ∩ K ′) = σ − 1. Then the k-valued points [K] and [K ′] belong to distinct connected components of Gen. (4) Suppose that k is algebraically closed. Then, for each k-valued point [L] of the scheme Isot(σ − 1, q0), there exist exactly two σ-dimensional totally isotropic subspaces of (N0, q0) ⊗ k that contain L. (5) Let P be the subscheme of Gen × Gen parameterizing pairs (K, K ′) such that dim(K ∩ K ′) = σ − 1. Then the scheme P ⊗ Fp2 has two connected components, each of which is isomorphic to Isot(σ − 1, q0) over Fp2. The action of Gal(Fp2/Fp) interchanges the two connected components. Consider the graph id × ϕ : Gen → Gen × Gen of the Frobenius morphism Gen → Gen given by K 7→ ϕ(K). The subscheme M of Gen that parametrizes the characteristic subspaces of (N0, q0) is defined by the fiber product M ֒→ Gen ↓ id × ϕ ↓ (cid:3) ֒→ Gen × Gen. P 20 SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA Ogus [16] proved the following: Theorem 8.1. The scheme M defined over Fp is smooth and projective of dimen- sion σ − 1. The scheme M ⊗ Fp2 has two connected components M+ = M ∩ Gen+ and M− = M ∩ Gen−, each of which is geometrically connected. The action of Gal(Fp2 /Fp) interchanges M+ and M−. Proof of Theorem 1.7. Let κ be an algebraic closure of the function field of the scheme M+ over Fp2 , and let [Kκ] be the geometric generic point of M+. By the surjectivity of the period mapping for supersingular K3 surfaces (Ogus [17, Theorem III′′]), there exist a supersingular K3 surface X of Artin invariant σ defined over κ and a marking η : N →∼ SX such that K(X,η) = Kκ. We prove that this X is generic, that is, Gκ := { γ ∈ O(N0, q0) K γ κ = Kκ } is equal to {±1}. Note that the closure of the point [Kκ] coincides with M+. Therefore we have the following: If a field k contains Fp2, then the action of Gκ leaves K invariant for any k-valued point [K] of M+. Suppose that σ ≥ 3. Let u be an arbitrary non-zero isotropic vector of N0. We will prove that u is an eigenvector of Gκ. Let b0 : N0 × N0 → Fp denote the symmetric bilinear form obtained from q0. There exists a vector v ∈ N0 such that q0(v) = 0 and b0(u, v) = 1, and hence (N0, q0) has an orthogonal direct- sum decomposition N0 = U ⊥ ⊕ U, where U is the subspace spanned by u and v. Repeating this procedure and noting that q0 is non-neutral, we obtain a basis a1, . . . , a2σ of N0 with u = a2σ such that q0(x1a1 + · · · + x2σa2σ) is equal to the quadratic polynomial f− in (7.2). Let α and ¯α = αp be the roots in Fp2 of the irreducible polynomial t2 + ct + 1 ∈ Fp[t]. We consider the basis (8.1) b(−1) 1 b(−1) i := αa1 + a2, := a2i−1, b(1) 1 b(1) i := ¯αa1 + a2, := a2i and (i = 2, . . . , σ) of N0 ⊗ Fp2. Note that each b(±1) i is isotropic, and that b0(b(α) i , b(β) j ) = 0 if i 6= j, b0(b(1) i , b(−1) i We put E := {1, −1}σ. ) =((4 − c2)/2 if i = 1, if i ≥ 2. 1/2 For e = (ε1, . . . , εσ) ∈ E, we denote by Ke the linear subspace of N0 ⊗ Fp2 spanned by It is obvious that Ke is isotropic. Moreover, since b(ε1) 1 , . . . , b(εσ ) σ . ϕ(b(ε) 1 ) = b(−ε) 1 and ϕ(b(ε) i ) = b(ε) i if i ≥ 2, we have dim(Ke ∩ ϕ(Ke)) = σ − 1. Therefore Ke is a characteristic subspace of (N0, q0). Suppose that e and e′ ∈ E differ only at one component. Then we have SUPERSINGULAR K3 SURFACES 21 dim(Ke ∩Ke′ ) = σ − 1, and hence the Fp2-valued points [Ke] and [Ke′ ] of M belong to distinct connected components. We put E+ := { e ∈ E the number of −1 in e is even }, 1 := (1, . . . , 1) ∈ E+. Interchanging α and ¯α if necessary, we can assume that [K1] is an Fp2-valued point of M+, and hence [Ke] is an Fp2-valued point of M+ for any e ∈ E+. It follows that Ke is invariant under the action of Gκ for any e ∈ E+. Let b(α) be an arbitrary element among the basis (8.1). Recall that we have assumed σ ≥ 3. Therefore, for each element b(β) , there exists e(j, β) = (ε1, . . . , εσ) ∈ E+ such that εi = α and εj 6= β. Since among the basis (8.1) that is distinct from b(α) j i i \(j,β)6=(i,α) Ke(j,β) = hb(α) i i is invariant under the action of Gκ, we see that b(α) ticular, the isotropic vector u = a2σ = b(1) of Gκ. Let is an eigenvector of Gκ. In par- σ given at the beginning is an eigenvector i λ(α) i : Gκ → F× p2 be the character defined by b(α) is an isotropic vector of N0 for any choice of α, β ∈ {±1}, and hence is an eigenvector of Gκ. Therefore we have . Suppose that i, j ≥ 2 and i 6= j. Then b(α) i + b(β) j i (8.2) i = λ(β) λ(α) j if i, j ≥ 2 and i 6= j. Since the cardinality of {x2 x ∈ Fp} is (p + 1)/2, there exist ξ, η ∈ Fp such that (4 − c2) + ξ2 + η2 = 0. Then b(1) 1 + b(−1) 1 + ξ(b(1) 2 + b(−1) 2 ) + η(b(1) 3 + b(−1) 3 ) is also an isotropic vector of N0, and hence is an eigenvector of Gκ. Therefore we have (8.3) λ(1) 1 = λ(−1) 1 = λ(1) 2 = λ(−1) 2 or λ(1) 1 = λ(−1) 1 = λ(1) 3 = λ(−1) 3 . Combining (8.2) and (8.3), we see that all the characters λ(α) other. Thus Gκ consists of only scalar multiplications. i are equal to each Suppose that σ = 2. In this case, the scheme M coincides with Isot(2, q0), which is the scheme parametrizing lines on the smooth quadratic surface Q0 = {q0 = 0} in the projective space P∗N0 = Grass(1, N0). Hence M+ and M− correspond to the two rulings of Q0. Let g be an element of Gκ. Then g leaves every line in the ruling of Q0 corresponding to M+ invariant. Since g is defined over Fp and Gal(Fp2 /Fp) interchanges M+ and M−, we see that g also leaves every line in the other ruling of Q0 invariant. Therefore g fixes every point of Q0, and hence every point of P∗N0. (cid:3) 22 SHIGEYUKI KOND ¯O AND ICHIRO SHIMADA References [1] M. Artin. Supersingular K3 surfaces. Ann. Sci. ´Ecole Norm. Sup. (4), 7:543 -- 567 (1975), 1974. [2] W. Barth and C. Peters. Automorphisms of Enriques surfaces. Invent. Math., 73(3):383 -- 411, 1983. [3] R. E. Borcherds. Automorphism groups of Lorentzian lattices. J. Algebra, 111(1):133 -- 153, 1987. [4] N. Bourbaki. ´El´ements de math´ematique. Fasc. XXXIV. Groupes et alg`ebres de Lie. Chapitre IV: Groupes de Coxeter et syst`emes de Tits. Chapitre V: Groupes engendr´es par des r´eflexions. Chapitre VI: syst`emes de racines. Actualit´es Scientifiques et Industrielles, No. 1337. Hermann, Paris, 1968. [5] J. H. Conway. The automorphism group of the 26-dimensional even unimodular Lorentzian lattice. J. Algebra, 80(1):159 -- 163, 1983. [6] I. Dolgachev and S. Kond¯o. A supersingular K3 surface in characteristic 2 and the Leech lattice. Int. Math. Res. Not., (1):1 -- 23, 2003. [7] T. Ekedahl and G. van der Geer. Cycle classes on the moduli of K3 surfaces in positive characteristic, 2011. arXiv:1104.3024 [8] N. Elkies and M. Schutt. Genus 1 fibrations on the supersingular K3 surface in characteristic 2 with Artin invariant 1, 2012. arXiv:1207.1239v1. [9] T. Katsura and S. Kond¯o. A note on a supersingular K3 surface in characteristic 2. In Geometry and Arithmetic, Series of Congress Reports, pages 243 -- 255. European Math. Soc., 2012. [10] T. Katsura and S. Kond¯o. Rational curves on the supersingular K3 surface with Artin invari- ant 1 in characteristic 3. J. Algebra, 352:299 -- 321, 2012. [11] S. Kond¯o. The automorphism group of a generic Jacobian Kummer surface. J. Algebraic Geom., 7(3):589 -- 609, 1998. [12] S. Kond¯o and I. Shimada. The automorphism group of a supersingular K3 surface with Artin invariant 1 in characteristic 3. Int. Math. Res. Notices (2012) doi: 10.1093/imrn/rns274 [13] M. Lieblich and D. Maulik. A note on the cone conjecture for K3 surfaces in positive char- acteristic, 2011. arXiv:1102.3377 [14] V. V. Nikulin. Integer symmetric bilinear forms and some of their geometric applications. Izv. Akad. Nauk SSSR Ser. Mat., 43(1):111 -- 177, 238, 1979. English translation: Math USSR-Izv. 14 (1979), no. 1, 103 -- 167 (1980). [15] V. V. Nikulin. Quotient-groups of groups of automorphisms of hyperbolic forms by subgroups generated by 2-reflections. Algebro-geometric applications. In Current problems in mathemat- ics, Vol. 18, pages 3 -- 114. Akad. Nauk SSSR, Vsesoyuz. Inst. Nauchn. i Tekhn. Informatsii, Moscow, 1981. [16] A. Ogus. Supersingular K3 crystals. In Journ´ees de G´eom´etrie Alg´ebrique de Rennes (Rennes, 1978), Vol. II, volume 64 of Ast´erisque, pages 3 -- 86. Soc. Math. France, Paris, 1979. [17] A. Ogus. A crystalline Torelli theorem for supersingular K3 surfaces. In Arithmetic and geometry, Vol. II, volume 36 of Progr. Math., pages 361 -- 394. Birkhauser Boston, Boston, MA, 1983. [18] A. N. Rudakov and I. R. Shafarevich. Supersingular K3 surfaces over fields of characteristic 2. Izv. Akad. Nauk SSSR Ser. Mat., 42(4):848 -- 869, 1978. Reprinted in I. R. Shafarevich, Collected Mathematical Papers, Springer-Verlag, Berlin, 1989, pp. 614 -- 632. [19] A. N. Rudakov and I. R. Shafarevich. Surfaces of type K3 over fields of finite characteristic. In Current problems in mathematics, Vol. 18, pages 115 -- 207. Akad. Nauk SSSR, Vsesoyuz. Inst. Nauchn. i Tekhn. Informatsii, Moscow, 1981. Reprinted in I. R. Shafarevich, Collected Mathematical Papers, Springer-Verlag, Berlin, 1989, pp. 657 -- 714. [20] T. Sengupta. Elliptic fibrations on supersingular K3 surface with Artin invariant 1 in char- acteristic 3. preprint, 2012, arXiv:1105.1715. [21] I. Shimada. Rational double points on supersingular K3 surfaces. Math. Comp., 73(248):1989 -- 2017 (electronic), 2004. [22] I. Shimada. Supersingular K3 surfaces in characteristic 2 as double covers of a projective plane. Asian J. Math., 8(3):531 -- 586, 2004. SUPERSINGULAR K3 SURFACES 23 [23] I. Shimada. Supersingular K3 surfaces in odd characteristic and sextic double planes. Math. Ann., 328(3):451 -- 468, 2004. [24] I. Shimada. Projective models of the supersingular K3 surface with Artin invariant 1 in characteristic 5. Preprint, 2012, arXiv:1201.4533 [25] T. Shioda. Supersingular K3 surfaces with big Artin invariant. J. Reine Angew. Math., 381:205 -- 210, 1987. [26] T. Shioda. On the Mordell-Weil lattices. Comment. Math. Univ. St. Paul., 39(2):211 -- 240, 1990. [27] `E. B. Vinberg. Some arithmetical discrete groups in Lobacevskiı spaces. In Discrete subgroups of Lie groups and applications to moduli (Internat. Colloq., Bombay, 1973), pages 323 -- 348. Oxford Univ. Press, Bombay, 1975. E-mail address: [email protected] Graduate School of Mathematics, Nagoya University, Nagoya, 464-8602, JAPAN E-mail address: [email protected] Department of Mathematics, Graduate School of Science, Hiroshima University, Higashi-Hiroshima, 739-8526, JAPAN
1811.05865
1
1811
2018-11-14T15:51:48
Mixed Hodge-Riemann bilinear relations and $m$-positivity
[ "math.AG", "math.CV" ]
Motivated by our previous work on Hodge-index type inequalities, we give a form of mixed Hodge-Riemann bilinear relation by using the notion of $m$-positivity, whose proof is an adaptation of the works of Timorin and Dinh-Nguy\^{e}n. This mixed Hodge-Riemann bilinear relation holds with respect to mixed polarizations in which some satisfy particular positivity condition, but could be degenerate along some directions. In particular, it applies to fibrations of compact K\"ahler manifolds.
math.AG
math
MIXED HODGE-RIEMANN BILINEAR RELATIONS AND M-POSITIVITY JIAN XIAO Abstract. Motivated by our previous work on Hodge-index type inequalities, we give a form of mixed Hodge-Riemann bilinear relation by using the notion of m-positivity, whose proof is an adaptation of the works of Timorin and Dinh-Nguyen. This mixed Hodge-Riemann bilinear relation holds with respect to mixed polarizations in which some satisfy particular positivity condition, but could be degenerate along some directions. In particular, it applies to fibrations of compact Kahler manifolds. Contents Introduction 1. 2. Preliminaries 3. Proof of the main result 4. Further remarks References 1 3 5 8 9 8 1 0 2 v o N 4 1 ] . G A h t a m [ 1 v 5 6 8 5 0 . 1 1 8 1 : v i X r a 1. Introduction 1.1. The classical and mixed HRR. Let X be a compact Kahler manifold of dimension n, and let ω be a Kahler class on X. Let 0 ≤ p, q ≤ p + q ≤ n be integers. Denote Ω = ωn−p−q, which is a strictly positive class in H n−p−q,n−p−q(X, C). Associated to Ω, one could define the following quadratic form Q on H p,q(X, C): Q(Φ, Ψ) = iq−p(−1)(p+q)(p+q+1)/2Ω · Φ · Ψ. The classical Hodge-Riemann bilinear relation theorem (HRR) (see e.g. [Dem12], [Voi07]) states that Q is positive definite on the primitive subspace P p,q(X, C), which is defined as follows: P p,q(X, C) = {Φ ∈ H p,q(X, C)Ω · ω · Φ = ωn−p−q+1 · Φ = 0}. As a corollary, one could get the classical Hard Lefchetz theorem (HL), i.e., the linear map Ω :H p,q(X, C) → H n−q,n−p(X, C), Φ 7→ Ω · Φ is an isomorphism. Furthermore, by the classical HRR, one could also get the Lefchetz decomposition theorem (LD), that is, there is an orthogonal decomposition H p,q(X, C) = P p,q(X, C) ⊕ ω ∧ H p−1,q−1(X, C) with respect to Q, with the convention that H p−1,q−1(X, C) = {0} if either p = 0 or q = 0. In [DN06, Cat08] (see also [Gro90, Tim98]), the classical HRR is greatly generalized to the mixed situation. More precisely, let ω1, ..., ωn−p−q+1 be Kahler classes on X. Denote Ω = ω1 · ... · ωn−p−q. We call Φ ∈ H p,q(X, C) primitive with respect to Ω · ωn−p−q+1, if Ω · ωn−p−q+1 · Φ = 0. The mixed HRR states that the corresponding quadratic form Q defined by Ω is positive definite on the subspace of primitive classes. This mixed HRR then implies the mixed versions of HL and LD. The reader can find some applications of this mixed HRR in complex geometry in [DS04,DS05]. One can also find some related topics and applications in the other contexts in [EW14], [Wil16], [AHK18], [McM93], [Tim99], [dCM02] and the references therein. 1 2 JIAN XIAO 1.2. The main result. Our aim is to weaken the positivity of (1, 1) classes, which we also call polarizations, in the definition of Ω such that the HRR still holds. The key positivity notion is m-positivity. Let ω be a reference Kahler metric on X of dimension n. Then a real smooth (1, 1) form bα is called m-positive (1 ≤ m ≤ n) with respect to ω, if bαk ∧ ωn−k > 0 holds for any 1 ≤ k ≤ m and any point on X. It is well-known that bα is n-positive if and only if it is strictly positive. This kind of positivity plays an important role in the study of Hessian equations, as it gives the natural solution set of such PDEs. A cohomology class α ∈ H 1,1(X, R) is called m-positive with respect to the metric ω, if it has a smooth representative which is m-positive in the pointwise sense. In our previous work [Xia18], inspired by the Hodge-index type inequalities obtained by complex Hessian equations, we proved a Hodge-index type theorem for classes of type (1, 1) by using m- positivity and Garding's theory of hyperbolic polynomials. More precisely, let α1, ..., αm−1 be m- positive classes, and denote Ω = ωn−m · α1 · ... · αm−2, then the HRR holds with respect to Ω, where the primitive subspace is defined by Ω · αm−1. This is the main motivation of our work. We generalize this result to arbitrary (p, q) classes, by assuming further that the αj are semipositive. Note that: if an (1, 1) form is semipositive, then it is m-positive with respect to ω if and only if it has at least m positive eigenvalues. Thus our positivity assumption is that every αj has a semipositive smooth representative, which has at least m positive eigenvalues at every point of X. Theorem A. Let X be a compact Kahler manifold of dimension n, and let ω be a Kahler class on X. Let p, q, m be integers such that 0 ≤ p, q ≤ p + q ≤ m ≤ n. Let α1, ..., αm−p−q+1 ∈ H 1,1(X, R). Assume that every αj has a smooth representative which is semipositive and has at least m positive eigenvalues at every point. Denote Ω = ωn−m · α1 · α2 · ... · αm−p−q. Let P p,q(X, C) = {Φ ∈ H p,q(X, C)Ω · αm−p−q+1 · Φ = 0} be the primitive subspace defined by Ω · αm−p−q+1. Then • (HRR) the quadratic form Q(Φ, Ψ) = iq−p(−1)(p+q)(p+q+1)/2Ω · Φ · Ψ is positive definite on P p,q(X, C). This implies that • (HL) the map Ω : H p,q(X, C) → H n−q,n−p(X, C), Φ 7→ Ω · Φ is an isomorphism. • (LD) the space H p,q(X, C) has an orthogonal decomposition H p,q(X, C) = P p,q(X, C) ⊕ αm−p−q+1 ∧ H p−1,q−1(X, C) with respect to Q, and dim P p,q(X, C) = dim H p,q(X, C) − dim H p−1,q−1(X, C), where we use the convention that H p−1,q−1(X, C) = {0} if either p = 0 or q = 0. Remark 1.1. In general, the HRR, HL and LD theorems are not true if Ω is an arbitrary class in H n−p−q,n−p−q(X, C), even if the class has a smooth strictly positive representative (see e.g. [BS02, Section 9], [DN13, Remark 2.9]). Thus Theorem A provides a sufficient condition on Ω such that HRR, HL and LD hold true. MIXED HODGE-RIEMANN BILINEAR RELATIONS AND M-POSITIVITY 3 1.3. Relative HRR. Interesting examples are given by the holomorphic fibrations between compact Kahler manifolds. Corollary A. Let f : X → Y be a holomorphic submersion from a compact Kahler manifold of dimension n to a compact Kahler manifold of dimension m. Let p, q, m be non-negative integers such that p + q ≤ m. Assume that ωX is a Kahler class on X and ωY1, ..., ωYm−p−q+1 are Kahler classes on Y , then the HRR holds with respect to where the primitive space is defined by ΩX,Y · f ∗ωYm−p−q+1. ΩX,Y = ωn−m X · f ∗ωY1 · ... · f ∗ωYm−p−q , This is true because every f ∗ωYk is m-positive with respect to some Kahler metric on X and semipositive. In some sense, this can be seen as a relative version of the classical and mixed HRR. Remark 1.2. When f is just a surjective holomorphic map, the f ∗ωYj is m-positive at generic points, that is, it has a smooth representative which is m-positive on a Zariski open set. In this general setting, we are not quite sure what kind of condition could be proposed to f such that the "relative" HRR holds if and only if the condition on f holds. Note that the relative HRR or HL is not true for a general map. Nevertheless, for general holomorphic maps, see Section 4.2 for some discussions. We intend to address this in a future work. To our knowledge, there are only some known results (see [dCM02]) when f : X → Y is a semi-small map between projective varieties and the ωYj are given by the same ample line bundle on Y . For this particular case, f is generically finite, thus n = m. 1.4. About the proof. To prove the main result, we mainly follow the approach of Timorin [Tim98] and Dinh-Nguyen [DN06,DN13]. First, we apply Timorin's inductive argument to establish the linear version of our HRR. To this end, we need to study the positivity of restrictions of m-positive forms. This is the only place where we need to assume further that the αj are semipositive, not only m- positive. Once the linear HRR is proved, then we could apply the ddc-method to reduce the global case to the local case. 2. Preliminaries 2.1. Restriction of m-positive forms. In this section, we consider the problem on positivity of the restrictions of m-positive forms. Let Λ1,1 R (Cn) be the space of real (1, 1) forms on Cn with constant coefficients. Lemma 2.1. Let ω be a Kahler metric on Cn with constant coefficients. Assume that α ∈ Λ1,1 is m-positive with respect to ω. Then R (Cn) • for any 1 ≤ k ≤ m − 1 and any hyperplane H ⊂ Cn, we have αk • for k = m, if we assume further that α is semipositive, then there is a proper subspace S(α) H ∧ ωn−k−1 > 0 on H. H such that for any v ∈ Cn \ S(α), we have αm Hv ∧ ωn−m−1 Hv > 0 on Hv, where Hv = {z ∈ Cnv · z = 0} is the hyperplane defined by v. Proof. Let a = (a1, ..., an) be a non-zero vector in Cn. Let Ha be the hyperplane defined by a. Denote the equation of Ha by the same notation Ha(z) = a · z. Since a 6= 0, we could find linear functions w1, ..., wn−1 of z such that (w1, ..., wn−1, Ha) give a new coordinate system of Cn. Write α = α0 + β0, where β0 is the sum of (1, 1) forms containing either dHa or dHa. Then α0 = αHa. Similarly, we write ω = ωHa + β1. It is easy to see that (1) αk ∧ ωn−k−1 ∧ idHa ∧ dHa = αk Ha ∧ ωn−k−1 Ha ∧ idHa ∧ dHa. In particular, if we denote V = in−1dw1 ∧ d ¯w1 ∧ ... ∧ dwn−1 ∧ d ¯wn−1, then (1) yields (2) αk ∧ ωn−k−1 ∧ idHa ∧ dHa V ∧ idHa ∧ dHa αk Ha ∧ ωn−k−1 Ha V . = 4 JIAN XIAO For simplicity, we just write (2) as (3) αk ∧ ωn−k−1 ∧ idHa ∧ dHa = αk Ha ∧ ωn−k−1 Ha . We first prove the first statement. For 1 ≤ k ≤ m − 1, we can write αk ∧ ωn−k−1 ∧ idHa ∧ dHa = ωn−m ∧ ωm−k−1 ∧ αk ∧ ∧idHa ∧ dHa. Since ω, α are m-positive with respect to ω, by the theory of hyperbolic polynomials (see e.g. [Gar59], [Hor94, Chapter 2], [Xia18, Lemma 3.8]), αk ∧ ωn−k−1 is strictly positive. Thus for any non-zero semipositive (1, 1) form β, Letting β = idHa ∧ dHa and using (3) finish the proof. Next we consider the second statement. Assume that αk ∧ ωn−k−1 ∧ β > 0. αm ∧ ωn−m−1 =Xi,j Φij \dzi ∧ d¯zj, where \dzi ∧ d¯zj is the (n − 1, n − 1) form omitting dzi, d¯zj such that \dzi ∧ d¯zj ∧ idzi ∧ d¯zj > 0. After dividing a volume form, we have αm ∧ ωn−m−1 ∧ idHa ∧ dHa =Xij Φijaiaj. As we have assumed further that α is semipositive, the Hermitian matrix [Φij] is semipositive and non-zero. This implies that the set S(α) = {(a1, ..., an) ∈ CnXij Φijaiaj = 0} is a proper linear subspace of Cn. Thus for any v ∈ Cn \ S(α), we have By (3), this is equivalent to αm ∧ ωn−m−1 ∧ idHv ∧ dHv > 0. αm Hv ∧ ωn−m−1 Hv > 0. This finishes the proof of the second statement. (cid:3) Remark 2.2. Take a coordinate system such that ω = iPn j=1 λjdzj ∧ d¯zj. If α is only m-positive with respect to ω, then (3) implies that αHe is m-positive with respect to ωHe for any j=1 dzj ∧ d¯zj, α = iPn To verify this, we only need to observe that e ∈ {(a1, ..., an)ai = ±1}. αk ∧ ωn−k−1 ∧ idHe ∧ dHe = αk ∧ ωn−k. In particular, this shows that there is an open set O1 (independent of α) of hyperplanes containing all the He such that, for any H ∈ O1, αH is m-positive with respect to ωH. By induction, this yields that there is an open set Om (independent of α) of linear subspaces of dimension m such that, for any V ∈ Om, αV is Kahler on V . MIXED HODGE-RIEMANN BILINEAR RELATIONS AND M-POSITIVITY 5 2.2. Existence of orthogonal bases. In order to prove the linear HRR by induction, we need the following existence result. Lemma 2.3. Let Hv1, ..., Hvk ⊂ Cn be hyperplanes, then there is an orthonormal basis (e1, ..., en) such that every ei ∈ Cn \ ∪k i=1Hvi. Proof. We first take e1 ∈ Cn \ (∪k i=1Hvi[ ∪k Cvi). i=1 Then for every i, Hvi ∩ He1 is a hyperplane in He1. By induction, there is an orthonormal basis (e2, ..., en) such that every ei ∈ He1 \ k[i=1 (Hvi ∩ He1). The vectors e1, ..., en give the desired basis. (cid:3) 3. Proof of the main result 3.1. The local case. We first adapt the arguments of Timorin [Tim98] to give a form of HRR in the linear case. Roughly speaking, Timorin's proof goes by induction: • assume that HRR holds for dim ≤ n − 1, then it can be used to prove HL for dim = n; • the HL for dim = n yields HRR for dim = n. Theorem 3.1. Let ω ∈ Λ1,1 0 ≤ p, q ≤ p + q ≤ m ≤ n. Assume that α1, ..., αm−p−q+1 ∈ Λ1,1 and semipositive. Denote R (Cn) be a Kahler metric on Cn. Let p, q be arbitrary integers satisfying R (Cn) are m-positive with respect to ω Ω = ωn−m ∧ α1 ∧ ... ∧ αm−p−q. Then the HRR holds with respect to Ω, where the primitive space is defined by Ω ∧ αm−p−q+1. In the linear case, the quadratic form Q is defined by Q(Φ, Ψ) = Ω ∧ Φ ∧ Ψ/Γ, where Γ is a fixed volume form of Cn. Note that when n = m, Theorem 3.1 is exactly the linear version of mixed HRR recalled in the introduction. Remark 3.2. It is clear that the linear version of HRR is equivalent to the HRR on a compact complex torus. We denote the space of complex (p, q) forms on Cn with constant coefficients by Λp,q. As stated above, Theorem 3.1 will be proved by induction on dimensions. Lemma 3.3. Assume Theorem 3.1 holds for dim = n − 1. Then the HL holds for dim = n, i.e., Ω : Λp,q → Λn−q,n−p is an isomorphism. Proof. When p + q = m or m = n, it is the classical HL recalled in the introduction. Thus we only need to consider the case when p + q < m < n. Without loss of generalities, we can assume ω = iPn We only need to prove that the map defined by Ω is injective. Assume that Φ ∈ Λp,q satisfies j=1 dzj ∧ d¯zj. Ω ∧ Φ = 0, then for any hyperplane H, ωn−m−1 H ∧ ωH ∧ α1H ∧ ... ∧ αm−p−qH ∧ ΦH = 0. This implies that ΦH is primitive on H, where αm−p−q+1 = ωH on H. By Lemma 2.1, for every general hyperplane H = Hv, where v ∈ Cn \ m−p−q[j=1 S(αj), the restrictions αj H are m-positive with respect to ωH. The restrictions αj H are also semipositive. Denote c = iq−p(−1)(p+q)(p+q+1)/2. By induction, the HRR holds in lower dimensions, thus (4) QH(ΦH, ΦH) = cωn−m−1 H ∧ α1H ∧ ... ∧ αm−p−qH ∧ ΦH ∧ ΦH ≥ 0. 6 JIAN XIAO By the same argument for (1), (4) is equivalent to (5) cωn−m−1 ∧ α1 ∧ ... ∧ αm−p−q ∧ Φ ∧ Φ ∧ idH ∧ dH ≥ 0. Note that every S(αj) is contained in a hyperplane, by Lemma 2.3 we can take an orthonormal basis e1, ..., en such that every ek ∈ Cn \ m−p−q[j=1 S(αj). Applying (5) to Hej and taking the sum over j imply (6) by using that cωn−m ∧ α1 ∧ ... ∧ αm−p−q ∧ Φ ∧ Φ ≥ 0, iXj dHej ∧ dHej = ω. By the assumption Ω ∧ Φ = 0, (6) is an equality, thus QHej (ΦHej , ΦHej ) = 0 for every j. By induction, this yields that ΦHej = 0 for every j. We claim that this implies Φ = 0. Without loss of generalities, we can assume Hej (z) = zj. The claim can be proved by contradiction. Assume Φ = XI=p,J=q ΦIJ dzI ∧ d¯zJ 6= 0, then there is a term ΦIJ dzI ∧ d¯zJ 6= 0. Since p + q < m < n, there must exist some j such that the multi-indexes I, J do not contain j. This implies ΦHej 6= 0, a contradiction. This finishes the proof of the lemma. (cid:3) Remark 3.4. Without the assumption on the semipositivity of αj, we associate αj to the following open set P (αj) = {v ∈ Cnαm j ∧ ωn−m−1 ∧ idHv ∧ dHv > 0}. Then by Lemma 2.1, it is clear that αj Hv It is unclear to us whether the intersection is m-positive with respect to ωHv on Hv for every v ∈ P (αj). m−p−q\j=1 P (αj) always contains an othonormal basis. If this was true, then we could apply the induction as above and remove the semipositivity assumption. By Lemma 2.3, this holds when the αj are also semipositive. Lemma 3.5. In the same setting as Theorem 3.1, assume that the HL holds for dim = n, then the quadratic form Q defined by Ω = ωn−m ∧ α1 ∧ ... ∧ αm−p−q is non-degenerate on Λp,q. Proof. This follows directly from Lemma 3.3. (cid:3) Another consequence of the HL in dimension n is the following LD. Lemma 3.6. In the same setting as Theorem 3.1, assume that the HL holds for dim = n, then the space Λp,q has a Q-orthogonal direct sum decomposition where we use the convention that Λp−1,q−1 = {0} when p = 0 or q = 0. Moreover, Λp,q = P p,q ⊕ (αm−p−q+1 ∧ Λp−1,q−1), dim P p,q = dim Λp,q − dim Λp−1,q−1, which is independent of Ω and αm−p−q+1. MIXED HODGE-RIEMANN BILINEAR RELATIONS AND M-POSITIVITY 7 Proof. By the assumption, the map Ω ∧ α2 Thus, the map αm−p−q+1 : Λp−1,q−1 → Λp,q is injective, and m−p−q+1 : Λp−1,q−1 → Λn−q+1,n−p+1 is an isomorphism. We also get that P p,q ∩ (αm−p−q+1 ∧ Λp−1,q−1) = {0}. Ω ∧ αm−p−q+1 : Λp,q → Λn−q+1,n−p+1 is an isomorphism when restricted to αm−p−q+1 ∧ Λp−1,q−1. The kernel of Ω ∧ αm−p−q+1 is exactly given by P p,q. On the hand, note that dim(αm−p−q+1 ∧ Λp−1,q−1) = dim Λp−1,q−1 = dim Λn−q+1,n−p+1. Thus, dim Λp,q = dim P p,q + dim αm−p−q+1 ∧ Λp−1,q−1. The orthogonality follows directly from the definition of P p,q. (cid:3) Proof of Theorem 3.1. Assume that the HRR holds in dim = n − 1, then by Lemma 3.3 the HL holds in dim = n. Thus we could apply Lemmas 3.5 and 3.6. Theorem 3.1 will follow from a homotopy argument as in [Tim98]. Consider the deformation Ωt = ωn−m ∧ ((1 − t)α1 + tω) ∧ ... ∧ ((1 − t)αm−p−q + tω), 0 ≤ t ≤ 1. The primitive space P p,q is defined by Ωt ∧ ((1 − t)αm−p−q+1 + tω). Note that every (1 − t)α1 + tω is m-positive and semipositive, thus all the quadratic forms Qt are non-degenerate and all the primitive subspaces P p,q have the same dimension. When t = 1, Q1 is positive definite by the classical HRR. Thus Q0 is also positive definite, since Qt never becomes degenerate in the course of deformation. t t Therefore, the HRR holds in dim = n. This finishes the proof. (cid:3) The following local estimate is important when reducing the global case to the local case. Recall that the space Λp,q admits an inner product defined by hΦ, Ψi =XI,J ΦIJ ΨIJ , where Φ =PI,J ΦIJ dzI ∧ d¯zJ , Ψ =PI,J ΨIJ dzI ∧ d¯zJ . The norm is given by Φ2 = hΦ, Φi. Lemma 3.7. Using the same notations as Theorem 3.1, there are positive constants c1, c2 such that Φ2 ≤ c1Q(Φ, Φ) + c2Ω ∧ αm−p−q+1 ∧ Φ2, ∀Φ ∈ Λp,q. Proof. By Theorem 3.1 and Lemma 3.6, the proof is the same as that of [DN06, Proposition 2.2] (see also [DN13, Proposition 2.8]). (cid:3) 3.2. The global case. Following [DN06, DN13], the global case (Theorem A) can be reduced to the local case (Theorem 3.1) by solving a ddc-equation. We denote the space of smooth complex (p, q) forms on X by Λp,q(X, C). a Kahler metric and smooth m-positive and semipositive forms in the corresponding classes, then for Lemma 3.8. In the same setting as Theorem A, use the notations bω and bα1, ...,bαm−p−q+1 to denote any smooth form bΦ ∈ Λp,q(X, C) such that its class {bΦ} ∈ P p,q(X, C), there is a smooth form bF such that Proof. Using Lemma 3.7, this follows from [DN13, Propositions 3.1, 3.2] (see also [DN06]). (cid:3) bΩ ∧bαm−p−q+1 ∧ ddcbF = bΩ ∧bαm−p−q+1 ∧ bΦ. Now we could prove Theorem A by using Lemma 3.8. (8) Proof of Theorem A. Assume that Φ ∈ P p,q(X, C) and let bΦ be a smooth representative of the class Φ. Then by Lemma 3.8, there is a smooth form bF such that Thus bΦ − ddcbF is a primitive (p, q) form with respect to bΩ ∧ bαm−p−q+1. bΩ ∧bαm−p−q+1 ∧ (bΦ − ddcbF ) = 0. Applying Theorem 3.1, we have (7) (9) cbΩ ∧ (bΦ − ddcbF ) ∧ (bΦ − ddcbF ) ≥ 0 8 JIAN XIAO at every point. By Stokes formula, (10) where c = iq−p(−1)(p+q)(p+q+1)/2. Q(Φ, Φ) = cZ bΩ ∧ (bΦ − ddcbF ) ∧ (bΦ − ddcbF ) ≥ 0, Moreover, if Q(Φ, Φ) = 0, then we have equalities everywhere. In particular, (9) is an equality at every point. By Theorem 3.1 again, this yields on X. Thus Φ = 0 in H p,q(X, C). This finishes the proof of the global HRR. bΦ − ddcbF = 0 As an immediate consequence, we get the global HL. We only need to check that Ω : H p,q(X, C) → H n−q,n−p(X, C) is injective. Assume that Φ ∈ H p,q(X, C) satisfyies Ω · Φ = 0, then Φ is primitive and Q(Φ, Φ) = 0. By HRR, Φ = 0, which finishes the proof of the HL. Finally, the global LD follows from similar arguments as Lemma 3.6. This finishes the proof of Theorem A. (cid:3) 4. Further remarks 4.1. Abstract Hodge-Riemann forms. In [DN13], Dinh-Nguyen gave an abstract version of HRR on compact Kahler manifolds by introducing the notion of Hodge-Riemann forms. By Theorem A, a more general abstract mixed HRR can be established. In the sequel, the αj are assumed to have the same positivity as above, i.e., m-positivity and semipositivity. Definition 4.1. (analogous to [DN13, Definition 2.1]) A real (k, k) form Ω ∈ Λk,k, k = n − p − q, is called a Lefschetz form for the bidegree (p, q) if the map Ω : Λp,q → Λn−q,n−p, Φ 7→ Ω ∧ Φ is an isomorphism. Assume p + q ≤ m ≤ n, a real (k, k) form is said to be a Hodge-Riemann form for the bidegree (p, q) if there is a continuous deformation Ωt ∈ Λk,k with 0 ≤ t ≤ 1, Ω0 = Ω and Ω1 = ωn−m ∧ α1 ∧ ... ∧ αm−p−q such that Ωt ∧ α2r m−p−q+1 is a Lefschetz form for the bidegree (p − r, q − r) for every r = 0, 1 and 0 ≤ t ≤ 1. Definition 4.2. Let X be a compact Kahler manifold of dimension n. Then Ω ∈ H k,k(X, R) is called a Hodge-Riemann class, if it has a representative which is a Hodge-Riemann form at every point. Analogous to [DN13], we have: Theorem 4.3. Let X be a compact Kahler manifold of dimension n, and let Ω be a Hodge-Riemann class. Then the HRR holds with respect to Ω, where the primitive space is given by Ω · αm−p−q+1. 4.2. Generalized semi-small maps. Let f : X → Y be a proper surjective holomorphic map between two complex spaces. For every integer k define The spaces Y k are analytic subvarieties of Y , whose disjoint union is Y . Recall that f is called a semi-small map in the sense of Goresky-MacPherson if Y k = {y ∈ Y dim f −1(y) = k}. dim Y k + 2k ≤ dim X for every k ≤ dim X/2. Note that a semi-small map must be generically finite. Semi-small maps can be generalized as follows. MIXED HODGE-RIEMANN BILINEAR RELATIONS AND M-POSITIVITY 9 Definition 4.4. We call a proper surjective holomorphic map f : X → Y relatively semi-small if dim Y k + 2k ≤ 2 dim X − dim Y for every k. Equivalently, f is relatively semi-small if and only if there are no irreducible analytic subvarieties T ⊂ X such that 2 dim T − 2 dim X + dim Y > dim f (T ). In particular, when dim X = dim Y , we get a semi-small map. In [dCM02], a line bundle is called lef if the Kodaira map of its multiple induces a semi-small map. Corresponding to relatively semi-small maps, this can be generalized as follows. Definition 4.5. A line bundle L on a projective manifold X is called relatively lef if kL is generated by its global sections for some positive integer k and the corresponding morphism is a relatively semi-small map. Analogous to [dCM02], it is easy to get the following result. φkL : X → Y = φkL(X) Proposition 4.6. Let f : X → Y be a surjective holomorphic map from a compact Kahler manifold of dimension n to a projective variety Y of dimension m. Let ω be a Kahler class on X, and let A1, ..., Am−p−q be ample line bundles on Y . If the relative HL holds with respect to for any 0 ≤ p, q ≤ p + q ≤ m, then f is a relatively semi-small map. ωn−m · f ∗A1 · ... · f ∗Am−p−q Proof. Otherwise, assume that f is not relatively semi-small. Then there is an irreducible analytic subvariety T ⊂ X such that 2 dim T − 2n + m > dim f (T ). Let {T } ∈ H n−dim T,n−dim T (X, R) be the fundamental class of T . The class f ∗A1 · ... · f ∗Am−2(n−dim T ) can be represented by an analytic cycle that does not intersect T , thus its intersection with {T } is zero. In particular, ωn−m · f ∗A1 · ... · f ∗Am−2(n−dim T ) · {T } = 0, which implies that the relative HL does not hold for the bidegree (n − dim T, n − dim T ). This finishes the proof. (cid:3) Remark 4.7. In the beautiful paper [dCM02], de Cataldo-Migliorini proved that L is lef on a projective manifold of dimension n if and only if the HL holds with respect to Ln−p−q for every 0 ≤ p, q ≤ p + q ≤ n. Moreover, they proved that for semi-small maps, the deep Decomposition Theorem of Beilinson, Bernstein, Deligne and Gabber [BBD82] is equivalent to the non-degeneracy of certain intersection forms (i.e., HRR) associated with a stratification. They applied this result to give a new proof of the Decomposition Theorem for the direct image of the constant sheaf. In our setting, we expect that φkL : X → Y = φkL(X) is relatively semi-small if and only if the relative HL holds with respect to ωn−m · Lm−p−q for every 0 ≤ p, q ≤ p + q ≤ m. We intend to discuss it elsewhere. Furthermore, we expect that the generalized form of Corollary A might be applied to study the topology of these maps. References [AHK18] Karim Adiprasito, June Huh, and Eric Katz, Hodge theory for combinatorial geometries, Ann. of Math. (2) 188 (2018), no. 2, 381 -- 452. MR 3862944 [BBD82] A. A. Beılinson, J. Bernstein, and P. Deligne, Faisceaux pervers, Analysis and topology on singular spaces, I [BS02] (Luminy, 1981), Ast´erisque, vol. 100, Soc. Math. France, Paris, 1982, pp. 5 -- 171. MR 751966 Bo Berndtsson and Nessim Sibony, The ∂-equation on a positive current, Invent. Math. 147 (2002), no. 2, 371 -- 428. MR 1881924 [Cat08] Eduardo Cattani, Mixed Lefschetz theorems and Hodge-Riemann bilinear relations, Int. Math. Res. Not. IMRN (2008), no. 10, Art. ID rnn025, 20. MR 2429243 [dCM02] Mark Andrea A. de Cataldo and Luca Migliorini, The hard Lefschetz theorem and the topology of semismall maps, Ann. Sci. ´Ecole Norm. Sup. (4) 35 (2002), no. 5, 759 -- 772. MR 1951443 10 JIAN XIAO [Dem12] Jean-Pierre Demailly, Complex analytic and differential geometry. online book, available at www-fourier. ujf- grenoble. fr/ demailly/manuscripts/agbook. pdf, Institut Fourier, Grenoble (2012). [DN06] Tien-Cuong Dinh and Viet-Anh Nguyen, The mixed Hodge-Riemann bilinear relations for compact Kahler manifolds, Geom. Funct. Anal. 16 (2006), no. 4, 838 -- 849. [DN13] Tien-Cuong Dinh and Viet-Anh Nguyen, On the Lefschetz and Hodge-Riemann theorems, Illinois J. Math. 57 [DS04] [DS05] (2013), no. 1, 121 -- 144. MR 3224564 Tien-Cuong Dinh and Nessim Sibony, Groupes commutatifs d'automorphismes d'une vari´et´e kahl´erienne com- pacte, Duke Math. J. 123 (2004), no. 2, 311 -- 328. MR 2066940 , Green currents for holomorphic automorphisms of compact Kahler manifolds, J. Amer. Math. Soc. 18 (2005), no. 2, 291 -- 312. MR 2137979 [EW14] Ben Elias and Geordie Williamson, The Hodge theory of Soergel bimodules, Ann. of Math. (2) 180 (2014), no. 3, 1089 -- 1136. MR 3245013 [Gar59] Lars Garding, An inequality for hyperbolic polynomials, J. Math. Mech. 8 (1959), 957 -- 965. MR 0113978 [Gro90] Misha Gromov, Convex sets and Kahler manifolds, Advances in Differential Geometry and Topology, ed. F. Tricerri, World Scientific, Singapore (1990), 1 -- 38. [Hor94] Lars Hormander, Notions of convexity, Progress in Mathematics, vol. 127, Birkhauser Boston, Inc., Boston, MA, 1994. MR 1301332 [McM93] Peter McMullen, On simple polytopes, Invent. Math. 113 (1993), no. 2, 419 -- 444. MR 1228132 [Tim98] V. A. Timorin, Mixed Hodge-Riemann bilinear relations in a linear context, Funktsional. Anal. i Prilozhen. 32 (1998), no. 4, 63 -- 68, 96. MR 1678857 [Tim99] , An analogue of the Hodge-Riemann relations for simple convex polyhedra, Uspekhi Mat. Nauk 54 (1999), no. 2(326), 113 -- 162. MR 1711255 [Voi07] Claire Voisin, Hodge theory and complex algebraic geometry. I, english ed., Cambridge Studies in Advanced Mathematics, vol. 76, Cambridge University Press, Cambridge, 2007, Translated from the French by Leila Schneps. MR 2451566 [Wil16] Geordie Williamson, The Hodge theory of the Hecke category, arXiv preprint, arXiv:1610.06246 (2016). [Xia18] Jian Xiao, Hodge-index type inequalities, hyperbolic polynomials and complex Hessian equations, arXiv preprint, arXiv:1810.04662 (2018). Tsinghua University, Beijing, China Email: [email protected]
1807.01932
2
1807
2018-11-07T16:32:39
Hodge ideals for Q-divisors: birational approach
[ "math.AG" ]
We develop the theory of Hodge ideals for Q-divisors by means of log resolutions, extending our previous work on reduced hypersurfaces. We prove local (non-)triviality criteria and a global vanishing theorem, as well as other analogues of standard results from the theory of multiplier ideals, and we derive a new local vanishing theorem. The connection with the V-filtration is analyzed in a sequel.
math.AG
math
HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH MIRCEA MUSTAT¸ A AND MIHNEA POPA Abstract. We develop the theory of Hodge ideals for Q-divisors by means of log resolutions, extending our previous work on reduced hypersurfaces. We prove local (non-)triviality criteria and a global vanishing theorem, as well as other analogues of standard results from the theory of multiplier ideals, and we derive a new local vanishing theorem. The connection with the V -filtration is analyzed in a sequel. 8 1 0 2 v o N 7 ] . G A h t a m [ 2 v 2 3 9 1 0 . 7 0 8 1 : v i X r a Contents A. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. Hodge ideals via log resolutions, and first properties . . . . . 1. A brief review of Hodge modules . . . . . . . . . . . . . . . . . . . . . 2. Filtered D-modules associated to Q-divisors . . . . . . . . . . . . . . 3. The case of smooth divisors . . . . . . . . . . . . . . . . . . . . . . . . 4. Definition of Hodge ideals for Q-divisors . . . . . . . . . . . . . . . . 5. A global setting for the study of Hodge ideals . . . . . . . . . . . . . . 6. A complex associated to simple normal crossing divisors . . . . . . . . 7. The Hodge ideals of simple normal crossing divisors . . . . . . . . . . 8. Computation in terms of a log resolution . . . . . . . . . . . . . . . . 9. The ideal I0(D) and log canonical pairs . . . . . . . . . . . . . . . . . C. Local study and global vanishing theorem . . . . . . . . . . . . 10. Generation level of the Hodge filtration, and examples . . . . . . . . 11. Non-triviality criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . 12. Vanishing theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. Restriction, subadditivity, and semicontinuity theorems . . . 13. Restriction theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 14. Semicontinuity theorem . . . . . . . . . . . . . . . . . . . . . . . . . 15. Subadditivity theorem . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 5 5 7 13 16 17 18 21 24 26 27 27 32 37 41 41 43 44 45 2010 Mathematics Subject Classification. 14F10, 14J17, 32S25, 14F17. MM was partially supported by NSF grant DMS-1701622; MP was partially supported by NSF grant DMS-1700819. 1 2 M. MUSTAT¸ A AND M. POPA A. Introduction In this paper we continue the study of Hodge ideals initiated in [MP16], [MP18a], by considering an analogous theory for arbitrary Q-divisors. The emphasis here is on a birational definition and study of Hodge ideals, while the companion paper [MP18b] is devoted to a study based on their connection with the V -filtration, inspired by [Sai16]. Both approaches turn out to provide crucial information towards a complete understanding of these objects. Let X be a smooth complex variety. If D is reduced divisor on X, the Hodge ideals Ik(D), with k ≥ 0, are defined in terms of the Hodge filtration on the DX-module OX(∗D) of functions with poles of arbitrary order along D. Indeed, this DX-module underlies a mixed Hodge module on X, and therefore comes with a Hodge filtration F•OX(∗D), which satisfies Fk OX(∗D) = Ik(D) ⊗ OX(cid:0)(k + 1)D(cid:1), for all k ≥ 0. See [MP16] for details, and for an extensive study of the ideals Ik(D). Our goal here is to provide a similar construction and study in the general case. A natural device for dealing with the fact that fractional divisors are not directly related to Hodge theory is to use new objects derived from covering constructions. Let D be an arbitrary effective Q-divisor on X. Locally, we can write D = αH, for some α ∈ Q>0 and H = div(h), the divisor of a nonzero regular function; we also denote by Z the support of D. A well-known construction associates to this data a twisted version of the localization D-module above, namely M(h−α) := OX(∗Z)h−α, that is the rank 1 free OX(∗Z)-module with generator the symbol h−α, on which a derivation D of OX acts by D(wh−α) :=(cid:18)D(w) − αw D(h) h (cid:19) h−α. It turns out that this DX-module can be endowed with a natural filtration FkM(h−α), with k ≥ 0, which makes it a filtered direct summand of a D-module underlying a mixed Hodge module on X; see §2. This plays a role analogous to the Hodge filtration, and just as in the reduced case one can show that FkM(h−α) ⊆ OX(kZ)h−α. This is done in §3 and §4, by first analyzing the case when Z is a smooth divisor (in this case, if ⌈D⌉ = Z, then the inclusion is in fact an equality). It is therefore natural to define the k-th Hodge ideal of D by the formula FkM(h−α) = Ik(D) ⊗OX OX (kZ)h−α. Similarly to [MP16], one of our main goals here is to study Hodge ideals of Q- divisors by means of log resolutions. To this end, let f : Y → X be a log resolution of the pair (X, D) that is an isomorphism over U = X r Z, and denote g = h ◦ f . There is a filtered isomorphism (cid:0)M(h−α), F(cid:1) ≃ f+(cid:0)M(g−α), F(cid:1). HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 3 Denoting G = f ∗D and E = Supp(G), so that E is a simple normal crossing divisor, it turns out that there exists a complex on Y : C • g−α(−⌈G⌉) : 0 → OY (−⌈G⌉) ⊗OY DY → OY (−⌈G⌉) ⊗OY Ω1 Y (log E) ⊗OY DY → . . . → OY (−⌈G⌉) ⊗OY ωY (E) ⊗OY DY → 0, which is placed in degrees −n, . . . , 0, whose differential is described in §6. This com- plex has a natural filtration given, for k ≥ 0, by subcomplexes Fk−nC • g−α(−⌈G⌉) := 0 → OY (−⌈G⌉) ⊗ Fk−nDY → → OY (−⌈G⌉) ⊗ Ω1 Y (log E) ⊗ Fk−n+1DY → · · · → OY (−⌈G⌉) ⊗ ωY (E) ⊗ FkDY → 0. Extending [MP16, Proposition 3.1], we show in Proposition 6.1 and Proposition 7.1 that there is a filtered quasi-isomorphism g−α(−⌈G⌉), F(cid:1) ≃(cid:0)Mr(g−α), F(cid:1), (cid:0)C • where Mr(g−α) is the filtered right DY -module associated to M(g−α). Thus one g−α(−⌈G⌉), F(cid:1) as a concrete representative for computing the filtered D- can use (cid:0)C • module pushforward of (cid:0)Mr(g−α), F(cid:1), hence for computing the ideals Ik(D). More R0f∗Fk−n(cid:0)C • DY →X(cid:1) ≃ h−αωX(kZ) ⊗OX Ik(D). See Theorem 8.1 for a complete picture regarding this push-forward operation. g−α(−⌈G⌉) ⊗DY precisely, we have This fact, together with special properties of the filtration on D-modules underlying mixed Hodge modules, leads to our main results on Hodge ideals, which are collected in the following: Theorem A. In the set-up above, the Hodge ideals Ik(D) satisfy: (i) I0(D) is the multiplier ideal I(cid:0)(1 − ǫ)D(cid:1), so in particular I0(D) = OX if and only if the pair (X, D) is log canonical; see §9. (ii) If Z has simple normal crossings, then Ik(D) = Ik(Z) ⊗ OX (Z − ⌈D⌉), while Ik(Z) can be computed explicitly as in [MP16, Proposition 8.2]; see §7. In particular, if Z is smooth, then Ik(D) = OX(Z − ⌈D⌉) for all k; cf. also Corollary 11.12. (iii) The Hodge filtration is generated at level n − 1, where n = dim X, i.e. FℓDX ·(cid:0)Ik(D) ⊗ OX (kZ)h−α(cid:1) = Ik+ℓ(D) ⊗ OX(cid:0)(k + ℓ)Z(cid:1)h−α for all k ≥ n − 1 and ℓ ≥ 0; see §10. (iv) There are non-triviality criteria for Ik(D) at a point x ∈ D in terms of the multiplicity of D at x; see §11. (v) If X is projective, Ik(D) satisfy a vanishing theorem analogous to Nadel Vanishing for multiplier ideals; see §12. 4 M. MUSTAT¸ A AND M. POPA (vi) If Y is a smooth divisor in X such that ZY is reduced, then Ik(D) satisfy Ik(DY ) ⊆ Ik(D) · OY , with equality when Y is general; see §13 for a more general statement. (vii) If X → T is a smooth family with a section s : T → X, and D is a relative divisor on X that satisfies a suitable condition (see §14 for the precise statement) then {t ∈ T Ik(Dt) 6⊆ m q s(t)} is an open subset of T , for each q ≥ 1. (viii) If D1 and D2 are Q-divisors with supports Z1 and Z2, such that Z1 + Z2 is also reduced, then the subadditivity property holds; see §15 for a more general statement. Ik(D1 + D2) ⊆ Ik(D1) · Ik(D2) For comparison, the list of properties of Hodge ideals in the case when D is reduced is summarized in [Pop18, §4]. While much of the story carries over to the setting of Q-divisors -- besides of course the connection with the classical Hodge theory of the complement U = X r D, which only makes sense in the reduced case -- there are a few significant points where the picture becomes more intricate. For instance, the bounds for the generation level of the Hodge filtration can become worse. Moreover, we do not know whether the inclusions Ik(D) ⊆ Ik−1(D) continue to hold for arbitrary Q-divisors. New phenomena appear as well: unlike in the case of multiplier ideals, for rational numbers α1 < α2, usually the ideals Ik(α1Z) and Ik(α2Z) cannot be compared for k ≥ 1; see for instance Example 10.5. It turns out however that most of these issues disappear if one works modulo the ideal of the hypersurface, at least for rational multiples of a reduced divisor. This, as well as other basic facts, is addressed in the sequel [MP18b], which studies Hodge ideals from a somewhat different point of view, namely by comparing them to the (microlocal) V -filtration induced on OX by h. This is inspired by the work of Saito [Sai16] in the reduced case. In the statement below we summarize some of these properties, which complement the results in Theorem A, but which we do not know how to obtain with the methods of this paper. Theorem B. [MP18b] Let D = αZ, where Z is a reduced divisor and α ∈ Q>0. Then the following hold: (1) Ik(D) + OX(−Z) ⊆ Ik−1(D) + OX(−Z) for all k. (2) If α ∈ (0, 1], then Ik(D) = OX ⇐⇒ k ≤ eαZ − α, where eαZ is the negative of the largest root of the reduced Bernstein-Sato polynomial of Z. (3) If Ik−1(D) = OX (we say that (X, D) is (k−1)-log canonical), then Ik+1(D) ⊆ Ik(D). (4) Fixing k, there exists a finite set of rational numbers 0 = c0 < c1 < · · · < cs < cs+1 = 1 such that for each 0 ≤ i ≤ s and each α ∈ (ci, ci+1] we have Ik(αZ) · OZ = Ik(ci+1Z) · OZ = constant HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 5 and such that Ik(ci+1Z) · OZ ( Ik(ciZ) · OZ . Going back to the description of Hodge ideals by means of log resolutions, the strictness of the Hodge filtration for the push-forwards of (summands of) mixed Hodge modules leads to the following local Nakano-type vanishing result for Q-divisors: Corollary C. Let D be an effective Q-divisor on a smooth variety X of dimension n, and let f : Y → X be a log resolution of (X, D) that is an isomorphism over X r Supp(D). If E = (f ∗D)red, then Rqf∗(cid:0)Ωp Y (log E) ⊗OY OY (−⌈f ∗D⌉)(cid:1) = 0 for p + q > n. Note that for p = n this is the local vanishing for multiplier ideals [Laz04, Theorem 9.4.1], since E − ⌈f ∗D⌉ = −[(1 − ǫ)f ∗D] for 0 < ǫ ≪ 1. In general, the statement extends the case of reduced D in [Sai07, Corollary 3] (cf. also [Sai16, §A.5]). Unlike [MP16, Theorem 32.1] regarding that case, at the moment we are unable to prove this corollary via more elementary methods. A different series of applications, given in [MP18b], uses the results proved in this paper together with the relationship between Hodge ideals of Q-divisors and the V - B (called the minimal exponent of Z). For instance, the triviality criterion proved here filtration, in order to describe the behavior of the invariant eαZ described in Theorem as Proposition 11.2 leads to a lower bound [MP18b, Corollary D] for eαZ in terms of and to restriction and semicontinuity statements for eαZ, in analogy with well-known invariants on a log resolution, addressing a question of Lichtin and Koll´ar. Moreover, the results in Theorem A (vi) and (vii), and Corollary 11.11, lead to effective bounds properties of log canonical thresholds; for details see [MP18b, §6]. B. Hodge ideals via log resolutions, and first properties Let X be a smooth complex algebraic variety. Given an effective Q-divisor D on X, our goal is to attach to D ideal sheaves Ik(D) for k ≥ 0; when D is a reduced divisor, these will coincide with the Hodge ideals in [MP16]. 1. A brief review of Hodge modules. A key ingredient for the definition of our invariants is Saito's theory of mixed Hodge modules. In what follows, we give a brief presentation of the relevant objects, and recall a few facts that we will need. For details, we refer to [Sa90]. Given a smooth n-dimensional complex algebraic variety X, we denote by DX the sheaf of differential operators on X. This carries the increasing filtration F•DX by order of differential operators. A left or right D-module is a left, respectively right, DX -module, which is quasi-coherent as an OX-module. There is an equivalence between the categories of left and right D-modules, which at the level of OX-modules is given by M → N := M ⊗OX ωX and N → HomOX (ωX, N ). 6 M. MUSTAT¸ A AND M. POPA For example, this equivalence maps the left D-module OX to the right D-module ωX. For a thorough introduction to the theory of D-modules, we refer to [HTT08]. A filtered left (or right) D-module is a D-module M, together with an increasing filtration F = F•M that is compatible with the order filtration on DX and which is good, in a sense to be defined momentarily. A morphism of filtered D-modules is required to be compatible with the filtrations. The equivalence between left and right D-modules extends to the categories of filtered modules, with the convention that Fp−n(M ⊗OX ωX) = FpM ⊗OX ωX. A filtration F•M on a coherent D-module M is good if the corresponding graded object grF DX. We note that every coherent D-module admits a good filtration, but this is far from being unique. • M :=Lk FkM/Fk−1M is locally finitely generated over grF • We now come to the key objects in Saito's theory, the mixed Hodge modules from [Sa90]. Such an object is given by the data M = (M, F, P, ϕ, W ), where: i) (M, F ) is a filtered D-module, with M a holonomic left (or right) D-module, with regular singularities; F is the Hodge filtration of M. ii) P is a perverse sheaf of Q-vector spaces on X. iii) ϕ is an isomorphism between PC = P ⊗Q C and DR(M), i.e. the perverse sheaf corresponding to M via the Riemann-Hilbert correspondence. iv) W is a finite, increasing filtration on (M, F, P, ϕ), the weight filtration of the mixed Hodge module. For a such an object to be a mixed Hodge module, it has to satisfy a complicated set of conditions of an inductive nature, which we do not discuss here. The main reference for the basic definitions and results of this theory is [Sa90]; see also [Sai17] for an introduction. Given a mixed Hodge module (M, F, P, ϕ, W ), we say that the filtered D-module (M, F ) is a Hodge D-module (or that it underlies a mixed Hodge module). In fact, this is the only piece of information that we will be concerned with in this article. The basic example of a mixed Hodge module is QH X[n], the trivial one. In this case, the filtered D-module is the structure sheaf OX , with the filtration such that grF OX = 0 p for all p 6= 0. The corresponding perverse sheaf is QX[n] and the weight filtration is such that grW p OX = 0 for p 6= n. The mixed Hodge modules on X form an Abelian category, denoted MHM(X). Morphisms in this category are strict with respect to both the Hodge and the weight Mixed Hodge modules satisfy Grothendieck's 6 operations formalism. The relevant fact for us is that to every morphism f : X → Y of smooth complex algebraic varieties filtration. The corresponding bounded derived category is denoted Db(cid:0)MHM(X)(cid:1). we have a corresponding push-forward functor f+ : Db(cid:0)MHM(X)(cid:1) → Db(cid:0)MHM(Y )(cid:1) (this is denoted by f∗ in [Sa90]). Moreover, if g : Y → Z is another such morphism, we have a functorial isomorphism (g ◦ f )+ ≃ g+ ◦ f+. HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 7 Regarding the push-forward functor for mixed Hodge modules, we note that on the level of D-modules, it coincides with the usual D-module push-forward. Moreover, if f : X → Y is proper and if we denote by FM(DX ) the category of filtered D-modules on X (here it is convenient to work with right D-modules), then Saito defined in [Sai88] a functor f+ : Db(cid:0)FM(DX )(cid:1) → Db(cid:0)FM(DY )(cid:1). This is compatible with the usual direct image functor for right D-modules and it is used to define the push-forward between the derived categories of mixed Hodge modules at the level of filtered complexes. With a slight abuse of notation, if (M, F ) underlies a mixed Hodge module M on X and if f : X → Y is an arbitrary morphism, then we write f+(M, F ) for the object in Db(cid:0)FM(DY )(cid:1) underlying f+M . An important feature of the push-forward of Hodge D-modules with respect to proper morphisms is strictness. This says that if f : X → Y is proper and (M, F ) underlies a mixed Hodge module on X, then f+(M, F ) is strict as an object in Db(cid:0)FM(DY )(cid:1) (and moreover, each H if+(M, F ) underlies a Hodge DY -module). This means that the natural mapping (1.1) is injective for every i, k ∈ Z. Taking FkH if+(M, F ) to be the image of this map, we get the filtration on H if+(M, F ). DX→Y ) DX→Y )(cid:1) −→ Rif∗(M Rif∗(cid:0)Fk(M L ⊗DX L ⊗DX The push-forward with respect to open embeddings is more subtle. For example, suppose that Z is an effective divisor on the smooth variety X and j : U = X rZ ֒→ X is the corresponding open immersion. Recall that OX(∗Z) is the push-forward j∗OU ; on a suitable affine open neighborhood V of a given point in X, this is given by localizing OX (V ) at an equation defining Z ∩ V in V . OX(∗Z) has a natural left D-module structure induced by the canonical D-module structure on OX. In fact, as such we have OX(∗Z) ≃ j+OU (in general, for a DU -module M, the D-module push-forward j+M agrees with j∗M, with the induced DX-module structure). We thus see that OX(∗Z) carries a canonical filtration such that the corresponding filtered D-module underlies j+QH U [n]. This filtration is the one that leads to the Hodge ideals studied in [MP16]. 2. Filtered D-modules associated to Q-divisors. Let X be a smooth complex algebraic variety, with dim(X) = n. The ideals we associate to effective Q-divisors on X arise from certain Hodge D-modules. The D-modules themselves have been extensively studied: these are the D-modules attached to rational powers of functions on X. We proceed to recall their definition. Consider a nonzero h ∈ OX(X) and β ∈ Q. We denote by Z the reduced divisor on X with the same support as H = div(h) and let j : U = X r Supp(Z) ֒→ X be the inclusion map. We consider the left DX -module M(hβ), which is a rank 1 free OX(∗Z)-module with generator the symbol hβ, on which a derivation D of OX acts by D(whβ ) :=(cid:18)D(w) + w β · D(h) h (cid:19) hβ. 8 M. MUSTAT¸ A AND M. POPA We will denote the corresponding right DX -module by Mr(hβ). This can be described as hβωX(∗Z), an OX-module isomorphic to ωX(∗Z), and such that if x1, . . . , xn are local coordinates, then (hβwdx1 · · · dxn)∂i = −hβ(cid:18) ∂w ∂xi ∂h ∂xi(cid:19) dx1 · · · dxn β h · + w for every i with 1 ≤ i ≤ n. Remark 2.1. When β ∈ Z, we have a canonical isomorphism of left DX -modules (2.1) M(hβ) ≃ OX(∗Z), whβ → w · hβ, where on the localization OX(∗Z) we consider the natural DX -module structure in- duced from OX. Note that OX (∗Z) is also the D-module push-forward j+OU . Remark 2.2. For every positive integer m, we have a canonical isomorphism of left DX -modules M(hβ) ≃ M(cid:0)(hm)β/m(cid:1), whβ → w(hm)β/m. Remark 2.3. We can define, more generally, left D-modules M(hβ1 r ), for nonzero regular functions h1, . . . , hr ∈ OX (X) and rational numbers β1, . . . , βr. If ℓi i , then we have an are positive integers such that βi/ℓi = β for all i and if h =Qi hℓi isomorphism of left DX -modules 1 · · · hβr M(hβ1 1 · · · hβr r ) ≃ M(hβ). Remark 2.4. If r is an integer, then we have an isomorphism of left DX -modules M(hβ) → M(hr+β), whβ → (wh−r)hr+β. Let now D be an effective Q-divisor on X. We denote by Z the reduced divisor with the same support as D. As above, we put U = X r Z and let j : U ֒→ X be the inclusion map. We first assume that we can write D = α · div(h) for some nonzero h ∈ OX(X) and α ∈ Q>0 (this is of course always the case locally). To this data we can associate the DX-module M(h−α); later it will be more convenient to consider equivalently (according to Remark 2.4) the DX -module M(h1−α). This depends on the choice of h; however, if we replace h by hm and α by α/m, for some positive integer m, the D-module does not change (see Remark 2.2). In particular, we may always assume that α = 1/ℓ, for a positive integer ℓ. Remark 2.5. Suppose that D′ is a Q-divisor with the same support as D and such that D−D′ = div(u), for some u ∈ OX(X). Suppose that we can write D′ = 1 ℓ ·div(h′) for some h′ ∈ OX(X) and some positive integer ℓ. In this case we can also write D = 1 ℓ · div(h), where h = uℓh′, and we have an isomorphism of DX -modules (2.2) M(h−1/ℓ) → M(h′−1/ℓ), gh−1/ℓ → gu−1h′−1/ℓ. HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 9 Our first goal is to show that M(h−α) is canonically a filtered DX -module. Let ℓ be a positive integer such that ℓα ∈ Z. Consider the finite ´etale map p : V → U , where V = Spec OU [y]/(yℓ − h−ℓα). Note that this fits in a Cartesian diagram (2.3) in which V p U j W q X, W = Spec OX [z]/(zℓ − hℓα), such that the map V → W pulls z back to y−1 = yℓ−1hℓα. Lemma 2.6. We have an isomorphism of left DX -modules (2.4) j+p+OV ≃ ℓ−1Mi=0 M(h−iα), with the convention that the first summand is OX (∗Z). Proof. Since p is finite ´etale, it follows that we have a canonical isomorphism τ : p∗DU ≃ DV , and for every DV -module M we have p+M ≃ p∗M, with the action of DU in- duced via the isomorphism τ . By mapping gyi to gh−iα, where g is a section of OX and 0 ≤ i ≤ ℓ − 1, we obtain an isomorphism of OX -modules as in (2.4). In order to see that this is an isomorphism of DX -modules, consider a local derivation D of OX and note that since yℓ = h−ℓα, by identifying D with its pull-back to V we have D(yi) = iyi−1D(y) = −iαyi D(h) h , which via our map corresponds to D(h−iα). This implies the assertion. (cid:3) It follows from the lemma that the right-hand side of (2.4) is the D-module corre- V [n]. In particular, it sponding to the mixed Hodge module push-forward (j ◦ p)+QH carries a canonical structure of filtered D-module. Remark 2.7. Let's see what happens if we replace ℓ by a multiple mℓ. Let pℓ : Vℓ → U and pmℓ : Vmℓ → U be the corresponding ´etale covers. Note that Vmℓ = Spec OU [y]/(yℓm − h−ℓmα) decomposes as a disjoint union of m copies of Vℓ, and thus we have an isomorphism of filtered DX -modules (and a corresponding isomorphism of mixed Hodge modules) where we write i = ℓc + d, with 0 ≤ c ≤ m − 1 and 0 ≤ d ≤ ℓ − 1. (2.5) If η is a primitive root of 1 of order ℓm, and if on each side of (2.5) we consider the decompositions (2.4), then the isomorphism maps j+(pmℓ)+OVmℓ ≃(cid:0)j+(pℓ)+OVℓ(cid:1)⊕m. h−iα →(cid:0)ηish−cℓα · h−dα(cid:1)0≤s≤m−1, 10 M. MUSTAT¸ A AND M. POPA We can interpret the isomorphism in (2.4) in terms of a suitable µℓ-action, where µℓ is the group of ℓ-th roots of 1 in C∗. Note that we have a natural action of µℓ on W such that via the corresponding action on OW , an element λ ∈ µℓ maps zi to λizi. If we let µℓ act trivially on X, then q is an equivariant morphism (in fact, q is the quotient morphism with respect to the µℓ-action). It is clear that q−1(Z) is fixed by the µℓ-action and we have an induced µℓ-action on W r q−1(Z) = V . This in turn induces a µℓ-action on j+p+OV and the isomorphism in (2.4) corresponds to the isotypic decomposition of j+p+OV , such that every λ ∈ µℓ acts on M(h−iα) by multiplication with λ−i. Lemma 2.8. The filtration on j+p+OV is preserved by the µℓ-action. Therefore we have an induced filtration on each M(h−iα) such that (2.4) is an isomorphism of filtered D-modules. Proof. One way to see this is by using a suitable equivariant resolution of W . Let W ′ be the disjoint union of the irreducible components of W and q′ : W ′ → W the canonical morphism. It is clear that the µℓ-action on W induces an action on W ′ such that q′ is equivariant. Since V is contained in the smooth locus of W , it has an open immersion into W ′. We use equivariant resolution of singularities to construct a µℓ-equivariant morphism ϕ : Y → W ′ that is an isomorphism over V and such that (q ◦ q′ ◦ ϕ)∗(Z) is a divisor with simple normal crossings. Let g = q ◦ q′ ◦ ϕ. If E is the reduced, effective divisor supported on g−1(Z), then we have an isomorphism of filtered D-modules (induced by a corresponding isomorphism of mixed Hodge modules) (2.6) j+p+OV ≃ g+ej+OV ≃ g+OY (∗E), whereej : Y r Supp(E) ֒→ Y is the inclusion map. We can deduce the assertion in the lemma from an explicit computation of the filtration on j+p+OV via the isomorphism (2.6), as follows. First, since we deal with D-module push-forward, it is more convenient to work with right D-modules. We will thus compute g+ωY (∗E), where ωY (∗E) is the filtered right D-module corresponding to OY (∗E). Since E is a simple normal crossing divisor, ωY (∗E) has a resolution by a complex C • of filtered right DY -modules 0 −→ C −n −→ · · · −→ C 0 −→ 0, where C i = Ωi+n Y (log E) ⊗OY DY , with the filtration given by Fk−nC i = Ωi+n Y (log E) ⊗OY Fk+iDY . For a description of the maps in this complex, see the beginning of §6 below; a proof of the fact that it resolves ωY (∗E) is given in [MP16, Proposition 3.1]. We can thus compute Fkg+ωY (∗E) as the image of the injective map R0g∗(cid:0)Fk(C • ⊗DY DY →X)(cid:1) → R0g∗(C • ⊗DY DY →X) = g+ωY (∗E). HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 11 Since g is equivariant and the action of µℓ on Y induces an action on E (in fact, it fixes E), the above description implies that each Fkg+ωY (∗E) is preserved by the µℓ-action. (cid:3) Remark 2.9. We note that the filtration on j+p+OV induces the canonical filtration on the first summand OX (∗Z). Indeed, on U we have a morphism of mixed Hodge modules QH V [n]. Applying j+ and only considering the underlying filtered D-modules, we obtain a morphism j+OU → j+p+OV , which is an isomorphism onto the first summand. U [n] → p+QH Definition 2.10. Given α > 0, choose ℓ ≥ 2 such that ℓα ∈ Z. In this case M(h−α) appears as the second summand in the decomposition (2.4). We define the filtration FkM(h−α) for k ≥ 0 to be the filtration induced from the canonical filtration on j+p+OV . It is straightfor- ward to see, using the discussion in Remark 2.7 that this filtration does not change if we replace ℓ by a multiple; therefore it is independent of ℓ. Moreover, we note that if α is an integer, using the same Remark 2.7, the isomorphism M(h−α) ≃ OX (∗Z) is an isomorphism of filtered D-modules. In this definition, a priori different covers have to be considered for each of the summands M(h−iα). However, we have: Lemma 2.11. With the filtration defined above, the isomorphism (2.4) is an isomor- phism of filtered D-modules. Proof. By Lemma 2.8, we only need to show that for every i with 0 ≤ i ≤ ℓ − 1, the filtration induced on M(h−iα) by that on j+p+OV coincides with the one given in the above definition. For i = 0 this follows from Remark 2.9. If i > 0, consider the cover used to define the filtration on M(h−iα), namely p′ : V ′ = SpecOU [y]/(yℓ − hiαℓ) → U. Note that we have a finite morphism ψ : V → V ′ of varieties over U , that pulls- back y to yi. We have a canonical morphism of mixed Hodge modules QH V ′[n] → ψ+QH + and passing to the underlying filtered D-modules, we obtain a morphism of filtered D-modules j+p′ OV ′ → j+p+OV that is the identity on + the summand M(h−iα). This proves our claim. (cid:3) V [n]. Applying j+p′ Remark 2.12. It is clear from definition that for every α > 0 and every positive integer m, the isomorphism M(h−α) → M(cid:0)(hm)−α/m(cid:1), is an isomorphism of filtered D-modules. ghm → g(hm)−α/m Remark 2.13. In the setting of Remark 2.5, the isomorphism (2.2) is an isomorphism of filtered DX -modules. This is clear if ℓ = 1, hence we assume ℓ ≥ 2. Let p : V → U and p′ : V ′ → U be the canonical projections, where V = Spec OU [y]/(yℓ − h) and V ′ = Spec OU [y]/(yℓ − h′). 12 M. MUSTAT¸ A AND M. POPA We have an isomorphism ϕ : V ′ → V of schemes over U , where ϕ∗(y) = uy. This induces an isomorphism of filtered DX -modules j+p+OV ≃ j+p′ + OV ′, which via the identifications given by Lemma 2.6 is the direct sum ℓ−1Mi=0 M(h−i/ℓ) ≃ ℓ−1Mi=0 M(h′−i/ℓ) of the isomorphisms (2.2). For i = 1, we obtain our assertion. A special case of the above remark implies that for every α > 0 the isomorphism M(h−α) → M(h−α−1), gh−α → (gh)h−α−1 is an isomorphism of filtered D-modules. We use this to put a structure of filtered D- module on M(hβ) for every β ∈ Q, such that for every r ∈ Z, we have an isomorphism of filtered D-modules M(hβ) → M(hβ−r), ghβ → (ghr)hβ−r. For example, we have have an isomorphism of filtered D-modules M(h0) ≃ OX(∗Z). Remark 2.14. Suppose that h, ¯h ∈ OX(X) are nonzero, and α, ¯α ∈ Q>0 are such that we have the equality of Q-divisors α · div(h) = ¯α · div(¯h). Let ℓ be a positive integer such that ℓα, ℓ ¯α ∈ Z. In this case there is g ∈ O ∗ X (X) such that hℓα = g¯hℓ ¯α. Suppose now that there exists G ∈ OX(X) such that Gℓ = g. (For example, this holds after pulling-back to the ´etale cover Spec OX [z]/(zℓ − g).) In this case we have an isomorphism of filtered DX-modules given by Φ : M(h−α) −→ M(¯h− ¯α) Φ(wh−α) = wG−1¯h− ¯α. Indeed, this follows from the definition of the filtrations and the isomorphism of schemes over U ϕ : SpecOU [y]/(yℓ − ¯h−ℓ ¯α) → SpecOU [y]/(yℓ − h−ℓα) that pulls-back y to G−1y. Remark 2.15. It is clear that the filtration on M(h−α) is compatible with restriction to open subsets. More generally, it is compatible with smooth pullback, as follows. Suppose that h ∈ OX (X) is nonzero and α ∈ Q. If ϕ : Y → X is a smooth morphism and g = h ◦ ϕ, then there is an isomorphism of DY -modules M(g−α) ≃ ϕ∗M(h−α), such that for every k we have FkM(g−α) ≃ ϕ∗FkM(h−α). HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 13 Indeed, choose ℓ ≥ 2 such that ℓα ∈ Z and consider the Cartesian diagram pY p VY ψ V UY U jY j Y ϕ X, where j and p are as in Lemma 2.6 and jY and pY are the corresponding morphisms for Y and g. Note that we have a base-change theorem that gives (2.7) ϕ!j+p+QH V [n] ≃ (jY )+(pY )+ψ!QH V [n] (see [Sa90, (4.4.3)]). Moreover, since ϕ is smooth, if d = dim(Y ) − dim(X), then for every filtered D-module (M, F ) underlying a mixed Hodge module M , the filtered D-module underlying ϕ!M is (ϕ∗M, F )[d], where Fk(ϕ∗M) = ϕ∗(FkM) (see [Sai88, 3.5]). This also applies to ψ; in particular, we have ψ!QH [n + 2d]. By decomposing both sides of (2.7) with respect to the µℓ-action, we obtain our assertion. V [n] ≃ QH VY 3. The case of smooth divisors. Our goal now is to describe the filtrations introduced in the previous section when Z is a smooth divisor. We will then use this to define Hodge ideals for arbitrary Q-divisors. The key result in the smooth case is the following: Lemma 3.1. Let ψ : Y = Spec C[t] −→ X = Spec C[x] be the map given by ψ∗(x) = tℓ. If Z is the divisor on Y defined by t, then we have an isomorphism of filtered DX-modules ψ+OY (∗Z) ≃ ℓ−1Mj=0 Mj, where Mj ≃ DX/DX (∂xx − j ℓ ) and FkMj is generated over OX by the classes of 1, ∂x, . . . , ∂k x. Moreover, if we consider on Y the µℓ-action such that every λ ∈ µℓ maps t to λt, then Mj is the component of ψ+OY (∗Z) on which every λ ∈ µℓ acts by multiplication with λj. Proof. As usual, it is easier to do the computation for the filtered right D-module ωY (∗Z) corresponding to OY (∗Z). Note that this is filtered quasi-isomorphic to the complex A• : 0 −→ DY w−→ ωY (Z) ⊗OY DY −→ 0, t ⊗ t∂t; see e.g. placed in degrees −1 and 0, where w(1) = dt [MP16, Proposition 3.1]. Since ψ is finite, the functor ψ∗ is exact on quasi-coherent OY -modules, hence ψ+ωY (∗Z) is computed by the 0-th cohomology of the complex B• = ψ∗(A• ⊗DY DY →X), with the obvious induced filtration. The definition of w immediately implies that w ⊗ 1DY →X is injective. Note that dt dx x and t∂t = ℓx∂x. t = 1 ℓ 14 M. MUSTAT¸ A AND M. POPA In order to describe the complex B•, note that any element of B−1 can be uniquely j=0 tjPj, with Pj ∈ DX. Similarly, any element in B0 can be uniquely j=0 tj dx x Qj, with Qj ∈ DX. Moreover, if τ is the differential in B•, then written as Pℓ−1 written as Pℓ−1 τ  ℓ−1Xj=0 tjPj  = ℓ−1Xj=0 tj dx x (x∂x + j ℓ )Pj , where we use the fact that t∂ttj = tjt∂t + jtj. eigenspace decomposition In other words, we have have an B• ≃ ℓ−1Mj=0 B• j , where B• j is identified with the complex 0 → DX → DX → 0, with the differential mapping P to (x∂x + j isomorphic to ℓ )P . It follows that B• is filtered quasi- where the filtration on the j-th component is such that j ℓ DX /(x∂x + ℓ−1Mj=0 Fk−1(cid:18)DX /(x∂x + )DX , )DX(cid:19) j ℓ is the OX -submodule generated by the classes of 1, ∂x, . . . , ∂k acts on the jth factor in the above decomposition by multiplication with λj. x. Moreover, every λ ∈ µℓ The assertion in the lemma now follows immediately from the explicit description of the equivalence between the categories of left and right D-modules on X = A1. Indeed, recall that if τ is the C-linear endomorphism of the Weyl algebra Γ(A1, DA1 ) such that τ (P Q) = τ (Q)·τ (P ) for all P and Q, and such that τ (t) = t and τ (∂t) = −∂t, then the left D-module N corresponding to a right D-module M is isomorphic to M itself, with scalar multiplication given via the map τ . Moreover, for filtered D- modules, via this isomorphism FkN corresponds to Fk−1M . In particular, we see that if M = DX/P · DX, then N ≃ DX /DX · τ (P ), and we obtain the statement. (cid:3) In what follows, we denote by ⌈α⌉ the smallest integer that is ≥ α. For a Q-divisor D =Pr i=1 aiDi, we put ⌈D⌉ =Pr i=1⌈ai⌉Di. Corollary 3.2. If h ∈ OX(X) is nonzero and such that the support Z of div(h) is smooth (possibly disconnected), then for every α ∈ Q>0 the filtration on M(h−α) is given by FkM(h−α) = OX(cid:0)(k + 1)Z − ⌈D⌉(cid:1)h−α where D = α · div(h), and FkM(h−α) = 0 if k < 0. if k ≥ 0, HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 15 Proof. We first reduce to the case when Z = div(h). We can check the assertion in the proposition locally, hence we may assume that Z = div(g), for some g ∈ OX (X), and h = ugm, for some u ∈ O ∗ X(X). Furthermore, by Remark 2.15, it is enough to prove the assertion after passing to a surjective ´etale cover, hence we may assume that u = vm for some v ∈ O ∗ X(X). After replacing g by vg, we may thus assume that h = gm. In this case we have an isomorphism of filtered D-modules M(h−α) ≃ M(g−mα), hence we may and will assume that div(h) = Z. We consider the smallest positive integer ℓ such that m := ℓα ∈ Z. If ℓ = 1, then the assertion follows from the formula for the filtration on OX(∗Z) when Z is smooth; see [MP16, Proposition 8.2]. Therefore from now on we assume ℓ > 1. The morphism h : X → A1 is smooth over some open neighborhood of 0. Using Remark 2.15, we see that in order to prove the corollary, we may assume that X = A1 and h = x, the standard coordinate on A1. Consider the Cartesian diagram j0 j V p U W g X, where j0 : V = Spec C[x, x−1, y]/(yℓ − x−m) → W = Spec C[x, z]/(zℓ − xm), j∗ 0 (z) = y−1. Let ϕ : fW = Spec C[t] → W be the normalization, given by ϕ∗(x) = tℓ and ϕ∗(z) = tm. (Here we use that ℓ and m are relatively prime.) Note that ϕ is an isomorphism over V , hence we have an open embedding ι : V ֒→ fW , with complement the smooth divisor T defined by t (in fact, if a and b are integers such that am + bℓ = 1, then ι∗(t) = y−axb). We thus have j+p+OV ≃ ψ+ι+OV ≃ ψ+OfW (∗T ), where ψ = g ◦ ϕ. We apply Lemma 3.1 for ψ. Note that ϕ is a µℓ-equivariant morphism if we let each λ ∈ µℓ act on t by multiplication with λa. By considering the behavior with respect to the µℓ-action, we see that in the decomposition given by the lemma, we have Mj ≃ M(x−α) if and only if ja ≡ −1 (mod ℓ), that is, j ≡ −m (mod ℓ). Suppose first that α < 1, in which case the condition for j is that j = ℓ − m. As a reality check, note that we indeed have an isomorphism DX /DX (∂xx − ℓ − m ℓ ) ≃ M(x−α) that maps the class of 1 to x−α. The formula for the filtration on M(hα) now follows from Lemma 3.1. When α > 1, we put m = ⌈α⌉ − 1, and use the fact from Remark 2.4, namely that we have an isomorphism of filtered modules M(x−α) → M(x−α+m), gx−α → (gx−m)x−α+m, 16 M. MUSTAT¸ A AND M. POPA to reduce the assertion to the case α ∈ (0, 1). This completes the proof of the corollary. (cid:3) 4. Definition of Hodge ideals for Q-divisors. In general, we obtain an upper bound for the terms in the filtration on M(h−α) by restricting to the open subset where the support of div(h) is smooth, as follows. Proposition 4.1. Given a nonzero h ∈ OX(X) and a positive rational number α, for every k ≥ 0 we have where D = α · div(h) and Z = Supp(D), while FkM(h−α) = 0 for k < 0. FkM(h−α) ⊆ OX(cid:0)(k + 1)Z − ⌈D⌉(cid:1)h−α, Proof. Let ι : X0 → X be an open immersion such that the codimension of its im- age in X is ≥ 2 and ZX0 is smooth (though possibly disconnected). Note that our constructions are compatible with restrictions to open subsets. Moreover, since M(h−α) is clearly torsion-free, it follows that Fk := FkM(h−α) is torsion free, hence the canonical map Fk → ι∗(cid:0)FkX0(cid:1) is injective. Therefore it is enough to prove the assertion on X0, hence we may assume that Z is smooth. However, in this case the assertion follows from Corollary 3.2. (cid:3) We can now define the Hodge ideals for Q-divisors. Let X be a smooth complex algebraic variety and Z a reduced effective divisor on X. Given an effective Q-divisor D with Supp(D) = Z, we define coherent ideals sheaves Ik(D) in OX as follows. Suppose first that there is a nonzero h ∈ OX (X), with H = div(h), and a positive rational number α such that D = αH. It turns out to be more convenient to work with the DX-module M(hβ), where β = 1 − α. Recall that we have a filtered isomorphism M(h−α) → M(hβ), wh−α → (wh−1)hβ, and therefore, if k ≥ 0, it follows from Proposition 4.1 that there is a unique coherent ideal Ik(D) such that FkM(hβ) = Ik(D) ⊗OX OX(cid:0)kZ + H(cid:1)hβ (note that we always have ⌈D⌉ ≥ Z). The definition is independent of the choice of indeed, using Remark 2.15, it is enough to check this after the pullback α and h: by a suitable ´etale surjective map, hence we deduce the independence assertion using Remark 2.14. This implies that the general case of the definition follows by covering X with suitable affine open subsets such that D can be written as above in each of them. Note that when D = Z we have β = 0, and so the ideals Ik(D) are the Hodge ideals studied in [MP16]. Remark 4.2. From the definition and the filtration property, it follows that we always have the inclusion OX(−Z) · Ik−1(D) ⊆ Ik(D) for k ≥ 1. We note that for the reduced divisor Z, we have the more subtle inclusions Ik(Z) ⊆ Ik−1(Z) for k ≥ 1 HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 17 (see [MP16, Proposition 13.1]). We do not know however whether this holds for arbitrary Q-divisors D, and in fact we suspect that this is not the case. (Note that it does hold when D has simple normal crossings support by Proposition 7.1. It is also shown to hold when D has an isolated weighted homogeneous singularity in the upcoming [Zha18].) However, when D = αZ these inclusions do hold modulo the ideal OX (−Z), see [MP18b, Corollary B]. More precisely, we have Ik(αZ) + OX (−Z) ⊆ Ik−1(αZ) + OX(−Z) for k ≥ 1. This implies in particular that if Ik(αZ) = OX for some k ≥ 1, then Ik−1(αZ) = OX . Remark 4.3. According to Proposition 4.1, we also have ideals I ′ k(D) given by k(D) ⊗OX which are related to Ik(D) by the formula FkM(h−α) = I ′ OX(cid:0)(k + 1)Z − ⌈D⌉(cid:1)h−α, Ik(D) = I ′ k(D) ⊗OX OX (Z − ⌈D⌉). The following periodicity property often allows us to reduce our study to the case ⌈D⌉ = Z. Lemma 4.4. If D′ is an integral divisor with Supp(D′) ⊆ Supp(D), then Ik(D + D′) = Ik(D) ⊗OX OX(−D′). In particular with B = D + Z − ⌈D⌉ satisfying ⌈B⌉ = Z. Ik(D) = Ik(B) ⊗ OX(Z − ⌈D⌉), Proof. Using the notation in Remark 4.3, the equivalent statement I ′ k(D + D′) = I ′ k(D) follows from the definition and Remark 2.13. (cid:3) Remark 4.5. Note that Ik(D) ⊆ OX(Z − ⌈D⌉) for all k, and so if ⌈D⌉ 6= Z, then one can never have Ik(D) = OX . It is however still interesting to ask whether Ik(B) = OX . 5. A global setting for the study of Hodge ideals. We now consider a setting in which we can define global filtered DX -modules that are locally isomorphic to the (cid:0)M(h−α), F(cid:1) discussed in the previous sections. Let X be a smooth variety and D = 1 ℓ H a Q-divisor, where H is an integral divisor and ℓ is a positive integer. The extra assumption we make here is that there is a line bundle M such that We denote by U the complement of Z = Supp(H) and by j the inclusion U ֒→ X. OX(H) ≃ M ⊗ℓ. Let s ∈ Γ(X, M ⊗ℓ) be a section whose zero locus is H. Since s does not vanish on U , we may consider the section s−1 ∈ Γ(cid:0)U, (M −1)⊗ℓ(cid:1). Let p : V → U be the ´etale cyclic cover corresponding to s−1, hence V ≃ Spec(cid:0)OU ⊕ M ⊕ . . . ⊕ M ⊗(ℓ−1)(cid:1). 18 M. MUSTAT¸ A AND M. POPA We consider the filtered DX -module M = j+p+OV , that underlies a mixed Hodge module. The µℓ-action on V , where λ ∈ µℓ acts on M ⊗i by multiplication with λ−i, induces an eigenspace decomposition M = ℓ−1Mi=0 Mi, where λ ∈ µℓ acts on Mi by multiplication with λ−i. We consider on each Mi the induced filtration. Note that if X0 is an open subset of X such that we have a trivialization M X0 ≃ OX0, and if via the corresponding trivialization of M ⊗ℓX0, the restriction sX0 cor- responds to h0 ∈ OX(X0), then we have isomorphisms of filtered DX0-modules Mi ≃ M(h−i/ℓ 0 ) for 0 ≤ i ≤ ℓ − 1. We also see that the filtration on M is the direct sum filtration, since this holds locally. Moreover, we have isomorphisms of OX0-modules MiX0 ≃ OX (∗Z)X0, which glue to isomorphisms of OX -modules Mi ≃ M ⊗i ⊗OX OX (∗Z) = j∗j∗M ⊗i. Via these isomorphisms, it follows from the definition of Hodge ideals (see also Remark 4.3) that we have FkMi ≃ M ⊗i ⊗OX I ′ k (i/ℓ · H) ⊗OX OX ((k + 1)Z − ⌈i/ℓ · H⌉) ≃ ≃ M ⊗i(−H) ⊗OX Ik (i/ℓ · H) ⊗OX OX (kZ + H) . 6. A complex associated to simple normal crossing divisors. We now discuss a complex that, as we will see later, gives a filtered resolution of Mr(h−α) by filtered induced DX -modules in the case when h defines a simple normal crossing divisor. Let X be a smooth, n-dimensional, complex variety, h ∈ OX (X) nonzero, and α a nonzero rational number (we allow α to be either positive or negative). Let D = α · div(h). We denote by Z the support of D, and assume that it has simple normal crossings. Associated to Z we have the following complex of right DX-modules: C • : 0 → DX → Ω1 X(log Z) ⊗OX DX → · · · → Ωn X(log Z) ⊗OX DX → 0, placed in degrees −n, . . . , 0. We denote by Di : C i → C i+1 its differentials. x1, . . . , xn are local coordinates on X, then If Di(η ⊗ P ) = dη ⊗ P + nXi=1 (dxi ∧ η) ⊗ ∂xiP. In fact C • is a filtered complex, where Fp−nC i = Ωi+n X (log Z) ⊗OX Fp+iDX . HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 19 This filtered complex is quasi-isomorphic to the filtered right DX -module ωX(∗Z) corresponding to the filtered left DX -module OX(∗Z) (see [MP16, Proposition 3.1], and [Sa90, Proposition 3.11(ii)] for a more general statement). Given h and α as above, we also consider the filtered complex C • h−α consisting of the same sheaves, but with differential C i h−α → C i+1 h−α given by It is easy to see that this is indeed a filtered complex. Di −(cid:0)(α · dlog(h) ∧ •) ⊗ 1DX(cid:1).1 Suppose now that we also have an effective divisor T supported on Z. It is not hard to check that the formula for the map h−α → C i+1 C i h−α induces also a map h−α(−T ) := OX (−T ) ⊗OX Ωi+n C i h−α(−T ) := OX(−T ) ⊗OX Ωi+1+n → C i+1 X X (log Z) ⊗OX DX (log Z) ⊗OX DX . This is due to the fact that if locally T = div(u) and η is a local section of Ωi+n X (log Z), then we can write d(uη) = ud(η)+u·dlog(u)∧η. We thus obtain a filtered subcomplex h−α(−T ) of C • C • h−α with OX(−T ). Proposition 6.1. If no coefficient of D − T lies in Z<0, then the complex C • h−α. We emphasize that this is not obtained by tensoring C • h−α(−T ) is filtered quasi-isomorphic to (cid:0)h−αωX(∗Z), G•(cid:1), where Gk−nh−αωX(∗Z) = 0 if k < 0, G−nh−αωX(∗Z) = h−αωX(Z − T ) and Gk−nh−αωX(∗Z) = G−nh−αωX(∗Z) · FkDX if k > 0. Proof. It is immediate to check that the differential induced on grF h−α(−T ) does become equal to the differential Di twisted with the identity on OX(−T ), and therefore for every p we have p C • grF p C • h−α(−T ) = OX (−T ) ⊗OX grF p C •. In particular, we have H iFpC • h−α(−T ) = 0 for every p ∈ Z and i ∈ Z r {0}, by the result in [MP16] quoted above. Consider now the morphism of right DX - modules ϕ : C 0 h−α(−T ) = ωX(Z − T ) ⊗OX DX −→ h−αωX(∗Z) given by ϕ(w ⊗ η ⊗ Q) = (h−αwη)Q. We first check that this morphism is surjective. We do this locally, hence we may assume that we have a system of coordinates x1, . . . , xn on X such that OX (−Z) is 1In related settings, for instance involving the de Rham complex of M(h−α), this type of complex can already be found in the literature; see for instance [Bjo93, §6.3.11]. 20 M. MUSTAT¸ A AND M. POPA generated by x1 · · · xr and OX (−T ) by xβ1 1 · · · xar r , where u is an everywhere nonvanishing function, and define αi = αai and γi = αi − βi for all i. Note for later use that r . We also write h = uxa1 1 · · · xβr α · dlog(h) = α u du + rXi=1 αi dxi xi . The surjectivity of ϕ follows from the fact that (6.1) where Im(ϕ) = (h−αxβ1 1 · · · xβr r η) · DX = h−αωX(∗Z), η = dlog(x1) ∧ . . . ∧ dlog(xr) ∧ dxr+1 ∧ . . . ∧ dxn, and the second equality in (6.1) is a consequence of the fact that −γi − 1 6∈ Z≥0 for all i, by assumption. In order to complete the proof of the proposition it is enough to show that, for every k ≥ 0, the following sequence is exact: OX(−T ) ⊗ Ωn−1 X (log Z) ⊗ Fk−1DX ψk−→ OX (−T ) ⊗ ωX(Z) ⊗ Fk DX ϕk−→ ϕk−→ Gk−nh−αωX(∗Z) −→ 0 where ϕk is the restriction of ϕ to the (k − n)-th level of the filtration and ψk is the restriction of the differential of C • h−α(−T ). The surjectivity of ϕk is an immediate consequence of the surjectivity of ϕ and the definition of the filtration on hαωX(∗Z). Keeping the above notation for the local coordinates on X, it follows from the definition of ψk that Im(ψk) = rYj=1 xβj j ⊗ η ⊗(cid:0) rXi=1(cid:0)xi∂i − γi − nXi=r+1(cid:0)∂i − ∂u ∂xi + · α u(cid:1) · Fk−1DX(cid:1) ∂u ∂xi · αxi u (cid:1) · Fk−1DX + and it is straightforward to see that this is contained in Ker(ϕk). We now prove by induction on k that if ϕk(xβ1 r ⊗ η ⊗ P ) = 0 for some P ∈ FkDX , then xβ1 1 · · · xβr r ⊗ η ⊗ P ∈ Im(ψk). Note that the case k = 0 is trivial. Let's write P = ≥0. After subtracting suitable terms from P , we may assume that whenever cu,v 6= 0, we have ui = 0 for i > r. Furthermore, note that if ui, vi > 0 for some i ≤ r, then we can write Pu,v cu,v∂uxv, where u and v vary over Zn 1 · · · xβr ∂uxv = (xi∂i − γi − ∂u ∂xi · αxi u )A + B, with both A and B of order ≤ k − 1. Therefore we may also assume that whenever cu,v 6= 0 and u :=Pi ui = k, we have (6.2) uivi = 0 for 1 ≤ i ≤ n. HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 21 Since (h−αxβ1 1 · · · xβr r η)∂uxv = (non-zero constant2) · (h−αxβ1 1 · · · xβr r η)xv−u, and since (6.2) implies that for every (u, v) and (u′, v′) with u = k and cu,v, cu′,v′ 6= 0 we have xv−u 6= xv′−u′ , we conclude that in fact P ∈ Fk−1DX , hence we are done by induction. (cid:3) 7. The Hodge ideals of simple normal crossing divisors. In this section we show that the Hodge ideals of divisors with simple normal crossing support essentially depend only on the support of the divisor, and therefore can be computed as in [MP16, §8]. Proposition 7.1. Let X be a smooth variety, and D an effective divisor on X with simple normal crossing support Z. Then for all k we have Ik(D) = Ik(Z) ⊗OX OX(Z − ⌈D⌉). Proof. Equivalently, we need to show that I ′ k(D) = Ik(Z) for all k. The assertion is local, hence we may assume that we have coordinates x1, . . . , xn on X such that Z = H1 + · · · + Hr, where Hi is defined by xi = 0. The morphism X → Cr given by (x1, . . . , xr) is smooth, hence using Remark 2.15 we see that it is enough to prove the i=1 αiHi, where Hi = div(xi) and αi > 0. Let ℓ be the smallest positive integer such that all ai := ℓαi are integers. The assertion to be proved is trivial when ℓ = 1, hence from now on we assume ℓ ≥ 2. Consider the Cartesian diagram proposition when X = Spec C[x1, . . . , xn] and D = Pn j0 j V p U W g X, where j0 : V = Spec C[x±1 · · · x−an n 1 → W = Spec C[x1, . . . , xn, z]/(zℓ − xa1 1 · · · xan n ), n , y]/(yℓ − x−a1 1 , . . . , x±1 ) with j∗ 0 (z) = y−1. We will make use of some standard facts about cyclic covers with respect to simple normal crossing divisors, exploiting the toric variety structure on the normalization of W . For basic facts regarding toric varieties, we refer to [Ful93]. Let N be the lattice Zn and M its dual. We also consider the lattice N ′ = {(v1, . . . , vn+1) ∈ Zn+1 a1v1 + · · · + anvn = ℓvn+1} and its dual M ′ = Zn+1/Z · (a1, . . . , an, −ℓ). Note that we have an injective lattice map N ′ → N , with finite cokernel, induced by the projection onto the first n components, and the dual map M → M ′ is again injective, with finite cokernel. In fact, we have an isomorphism M ′/M ≃ Z/ℓZ that maps the class of (u1, . . . , un+1) ∈ M ′ to the class of un+1 in Z/ℓZ. 2Here we use again the fact that −γi − 1 6∈ Z≥0 for all i. 22 M. MUSTAT¸ A AND M. POPA We thus have an isomorphism N ′ in NR = Rn gives the toric variety X = Cn. As a cone in N ′ R ≃ NR = Rn. The strongly convex cone σ = Rn ≥0 R, σ gives an affine toric variety fW , and the lattice map N ′ → N corresponds to a toric map ψ : fW → X. Note that we have a morphism of O(X)-algebras O(W ) → O(fW ) that maps xi to the element of C[σ∨ ∩ M ′] corresponding to the class of the i-th element of the standard basis of Zn, and z to the class of (0, . . . , 0, 1). It is easy to check that if we denote by ⌊γ⌋ the largest integer ≤ γ, then (7.1) O(fW ) = M0≤j≤ℓ−1 O(X)x−⌊jα1⌋ 1 · · · x−⌊jαn⌋ n zj, embedding ι : V ֒→ Y such that f ◦ ι = j ◦ p. The support EY of Y r ι(V ) is the sum of all prime toric divisors on Y . and consequently to deduce that O(fW ) is integral over O(W ). As the coordinate ring of a toric variety, O(fW ) is normal, hence it is the integral closure of O(W ) in its field of fractions. Moreover, since fW is a toric variety, we may choose a toric resolution of singularities Y → fW , and let f : Y → X be the composition. Since the map Y → W is an isomorphism over the complement of g−1(P Hi), it follows that there is an open The action of the torus TM ′ = Spec C[M ′] on fW induces an action of the finite group Spec C[M ′/M ] ≃ Spec C[Z/ℓZ] = µℓ on fW . This is the action induced on the normalization fW by the µℓ-action on W that we discussed in §2. In particular, the toric resolution Y → fW is automatically equivariant. Note that in the decomposition (7.1), an element λ ∈ µℓ acts on the summand corresponding to j by multiplication with λj. The equality f ◦ ι = j ◦ p implies that we have an isomorphism of filtered DX - modules j+p+OV ≃ f+ι+OV = f+OY (∗EY ). As usual, in order to compute the push-forward of OY (∗EY ), it is more convenient to work with right D-modules. Recall that there is a complex of right DY -modules A• = A• Y : 0 → DY → Ω1 Y (log EY ) ⊗OY DY → · · · → ωY (EY ) ⊗OY DY → 0 located in degrees −n, . . . , 0, that is filtered quasi-isomorphic to ωY (∗EY ); see the be- ginning of §6. Since Y is a toric variety, we have a canonical isomorphism Ω1 Y (log EY ) ≃ M ′ ⊗Z OY (see [Ful93, Section 4.3]). We will also consider the corresponding complex on X: A• X : 0 −→ DX −→ M ⊗Z DX −→ · · · −→ ∧nM ⊗Z DX −→ 0. It follows from the definition that, forgetting about the filtration, we have f+ωY (∗EY ) = Rf∗(A• ⊗DY DY →X). Note that DY →X = f ∗DX as OY -modules, hence the projection formula implies Rif∗(Ap−n ⊗DY DY →X) ≃ ∧pM ′ ⊗Z Rif∗OY ⊗OX DX = 0 HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 23 for i > 0, since f is the composition of a finite map with a toric resolution. Therefore f+ωY (∗EY ) is represented by the complex B•, where Bp−n = ∧pM ′ ⊗Z ψ∗OfW ⊗OX DX . In order to describe the differential of this complex, it is convenient to use the iso- morphism MQ ≃ M ′ Q and the decomposition (7.1). With a little care, it follows from the definitions that if we put j = ∧pMQ ⊗Q OX · x−⌊jα1⌋ Bp−n zj ⊗OX then B• decomposes as the direct sum of the subcomplexes B• 1. Furthermore, if we identify each Bp−n differential · · · x−⌊jαn⌋ n 1 j in the obvious way with Ap−n DX , j , for 0 ≤ j ≤ ℓ − X , then the δp−n Bj : ∧p MQ ⊗Q DX → ∧p+1MQ ⊗Q DX is given by δp−n Bj where δAX is the differential on A• = δp−n AX X and + (wj ∧ −) ⊗ IdDX , wj = (wj,1, . . . , wj,n), with wj,i = jαi − ⌊jαi⌋. It follows from Proposition 6.1 that we have a morphism j → Mr(xwj,1 B0 1 · · · xwj,n n ) that induces a quasi-isomorphism j → Mr(xwj,1 B• 1 · · · xwj,n n ) (see also Remark 2.3). We now bring the filtrations into the picture. It follows from Saito's strictness results (see the discussion in §1; cf. also [MP16, §4, §6]) that Arguing as above, we deduce that Fkf+ωY (∗EY ) = Im(cid:0)Rf∗Fk(A• ⊗DY Fkf+ωY (∗EY ) = Im(cid:0)f∗Fk(A• ⊗DY DY →X) → Rf∗(A• ⊗DY DY →X) → f∗(A• ⊗DY DY →X)(cid:1). DY →X)(cid:1). In other words, (f+ωY (∗EY ), F ) is represented by the filtered complex B•, and using Proposition 6.1, we conclude that f+ωY (∗EY ) ≃ ℓ−1Mj=0 Mr(xwj,1 1 · · · xwj,n n ), where the filtration on Mr(xwj,1 ) is given by · · · xwj,n n · · · xwj,n ) = xwj,1 n 1 · · · xwj,n n ωX(Z) and 1 F−nMr(xwj,1 wj,1 wj,n n 1 · · · x 1 Fk−nMr(x ) = F−nMr(x wj,1 1 · · · x wj,n n ) · FkDX for k ≥ 1. 24 M. MUSTAT¸ A AND M. POPA By comparing the µℓ-actions, we conclude that the summand Mr(x−α1 · · · x−αn ) on which an element λ ∈ µℓ acts by multiplication with λ−1 corresponds to j = ℓ − 1. Therefore the filtration on M(x−α1 ) is given by · · · x−αn n 1 F−nMr(x−α1 Fk−nMr(x−α1 1 · · · x−αn n 1 ) = (x−α1 n · · · x−αn 1 n ) = F−nMr(x−α1 )x⌈α1⌉ 1 · · · x⌈αn⌉ n ωX(Z) and 1 · · · x−αn It is now a straightforward computation to see that I ′ · · · x−αn n the monomials Qn coincides with Ik(Z) according to [MP16, Proposition 8.2], completing the proof of the proposition. (cid:3) i , where 0 ≤ ci ≤ k for all i and Pi ci = (n − 1)k. This i=1 xci k(D) is the ideal generated by ) · FkDX for k ≥ 1. n 1 8. Computation in terms of a log resolution. We use the results of the previous two sections in order to describe Hodge ideals of Q-divisors in terms of log resolutions. Let X be a smooth variety, h ∈ OX (X) a nonzero function, H = div(h), and α ∈ Q>0. We are interested in computing Ik(D), where D = αH. As always, let Z = Supp(D) and β = 1 − α. Let f : Y → X be a log resolution of the pair (X, D) that is an isomorphism over U = X r Z, and denote g = h ◦ f ∈ OY (Y ). We fix a positive integer ℓ such that ℓα ∈ Z. As usual, we consider p : V = Spec OU [y]/(yℓ − h−ℓα) −→ U and the inclusion j : U ֒→ X. By assumption, we also have an open immersion ι : U ֒→ Y such that f ◦ ι = j. By considering the decompositions of j+p+OV ≃ f+ι+p+OV into isotypical components, we conclude that we have a filtered isomorphism (8.1) M(h−α) ≃ f+M(g−α). We now denote G = f ∗D, and consider on Y the complex introduced in §6: C • g−α(−⌈G⌉) : 0 → OY (−⌈G⌉) ⊗OY DY → OY (−⌈G⌉) ⊗OY Ω1 Y (log E) ⊗OY DY → · · · → OY (−⌈G⌉) ⊗OY ωY (E) ⊗OY DY → 0, where E = (f ∗D)red. This is placed in degrees −n, . . . , 0, and if x1, . . . , xn are local coordinates on Y , then its differential is given by η ⊗ Q → dη ⊗ Q + nXi=1 (dxi ∧ η) ⊗ ∂iQ −(cid:0)α · dlog(g) ∧ η(cid:1) ⊗ Q. Theorem 8.1. With the above notation, the following hold: i) For every p 6= 0 and every k ∈ Z, we have and Rpf∗(cid:0)C • Rpf∗Fk(cid:0)C • g−α(−⌈G⌉) ⊗DY g−α(−⌈G⌉) ⊗DY DY →X(cid:1) = 0 DY →X(cid:1) = 0. HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 25 ii) For every k ∈ Z, the natural inclusion induces an injective map R0f∗Fk(cid:0)C • g−α(−⌈G⌉) ⊗DY iii) We have a canonical isomorphism DY →X(cid:1) ֒→ R0f∗(cid:0)C • g−α(−⌈G⌉) ⊗DY DY →X(cid:1). R0f∗(cid:0)C • g−α(−⌈G⌉) ⊗DY DY →X(cid:1) ≃ Mr(h−α) that induces for every k ∈ Z an isomorphism R0f∗Fk−n(cid:0)C • g−α(−⌈G⌉) ⊗DY DY →X(cid:1) ≃ h−αωX(cid:0)(k + 1)Z − ⌈D⌉(cid:1) ⊗OX I ′ k(D) ≃ ≃ hβωX(cid:0)kZ + H(cid:1) ⊗OX Ik(D). Proof. It follows from Lemma 2.8, and from the definition of its filtration, that Mr(g−α) is a direct summand of a right Hodge D-module on Y . By Saito's strictness of the filtration of (push-forwards of) such D-modules, it follows that for all k, p ∈ Z the canonical map Rpf∗Fk(cid:0)Mr(g−α) L ⊗DY is injective, and its image is equal to DY →X(cid:1) → Rpf∗(cid:0)Mr(g−α) DY →X(cid:1) L ⊗DY FkRpf∗(cid:0)Mr(g−α) L ⊗DY DY →X(cid:1) (see the discussion in §1). On the other hand, note that if write G = α · div(g) =Pi αiEi, then −⌈αi⌉ + αi 6∈ Z<0 for all i. We may thus apply Proposition 6.1 for the divisor G, with T = ⌈G⌉. Using Proposition 7.1 as well, we see that C • g−α(−⌈G⌉) is filtered quasi-isomorphic to Mr(g−α), hence Rpf∗(cid:0)C • Rpf∗Fk(cid:0)C • g−α(−⌈G⌉) ⊗DY g−α(−⌈G⌉) ⊗DY Rpf∗(cid:0)Mr(g−α) L ⊗DY DY →X(cid:1) ≃ Rpf∗(cid:0)Mr(g−α) DY →X(cid:1) ≃ Rpf∗Fk(cid:0)Mr(g−α) DY →X(cid:1) ≃ H pf+Mr(g−α), L ⊗DY DY →X(cid:1) and DY →X(cid:1). L ⊗DY Finally, by the definition of push-forward for right D-modules we have and by (8.1) this is 0 if p 6= 0, and is canonically isomorphic to Mr(h−α) if p = 0. The assertions in the proposition follow by combining all these facts. (cid:3) Remark 8.2 (Local vanishing). The statement in Theorem 8.1 i) is a generalization of the Local Vanishing theorem for multiplier ideals [Laz04, Theorem 9.4.1], in view of the calculation in Proposition 9.1 below. As a consequence of the vanishing statements in Theorem 8.1(i), provided by strict- ness, we deduce the following local Nakano-type vanishing result, first obtained by Saito [Sai07, Corollary 3] when D is reduced; cf. Corollary C in the Introduction and the discussion following it. 26 M. MUSTAT¸ A AND M. POPA Corollary 8.3. Let D be an effective Q-divisor on the smooth variety X and f : Y → X a log resolution of (X, D) that is an isomorphism over X r Supp(D). If E = (f ∗D)red, then Rqf∗(cid:0)OY (−⌈f ∗D⌉) ⊗OY Ωp Y (log E)(cid:1) = 0 for p + q > n = dim(X). Proof. We argue by descending induction on p, the case p > n being trivial. Suppose now that p ≤ n and q > n − p. After possibly replacing X by suitable open subsets, we may assume that D = α · div(h). We may thus apply Theorem 8.1 to deduce that if C • = F−n(cid:16)C • g−α(−⌈f ∗D⌉) ⊗DY DY →X(cid:17) [p − n], then (8.2) Rjf∗C • = 0 for j > n − p. Note that by definition, we have C i = OY (−⌈f ∗D⌉) ⊗OY Ωp+i Y (log E) ⊗OY f ∗FiDX for 0 ≤ i ≤ n − p. Consider the spectral sequence Ei,j 1 = Rjf∗C i ⇒ Ri+jf∗C •. It follows from (8.2) that E0,q ∞ = 0. Now by the projection formula we have (8.3) Ei,j 1 = Rjf∗(cid:0)OY (−⌈f ∗D⌉ ⊗OY Ωp+i Y (log E)(cid:1) ⊗OX FiDX . In particular, it follows from the inductive hypothesis that for every r ≥ 1 we have = 0, hence Er,q−r+1 Er,q−r+1 = 0 as well. On the other hand, we clearly have E−r,q+r−1 = 0, since this is a first-quadrant spectral sequence. We thus conclude that 1 r r E0,q r = E0,q r+1 for all r ≥ 1, hence E0,q 1 = E0,q ∞ = 0. Using (8.3) again, we conclude that Rqf∗(cid:0)OY (−⌈f ∗D⌉) ⊗OY Ωp Y (log E)(cid:1) = 0. (cid:3) 9. The ideal I0(D) and log canonical pairs. We now use Theorem 8.1 in order to relate I0(D) to multiplier ideals. Recall that for a Q-divisor B, one denotes by I(B) the associated multiplier ideal; see [Laz04, Ch.9] for the definition and basic properties. Proposition 9.1. If f : Y → X is a log resolution of (X, D) that is an isomorphism over X r D, and E = (f ∗D)red, then I0(D) ≃ f∗OY(cid:0)KY /X + E − ⌈f ∗D⌉(cid:1) = I(cid:0)(1 − ǫ)D(cid:1) for 0 < ǫ ≪ 1. HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 27 Proof. The first equality follows from Theorem 8.1, together with the fact that the term F−nC • g−α(−⌈f ∗D⌉) consists of ωY (E − ⌈f ∗D⌉) placed in degree 0. The second equality then follows from the definition of multiplier ideals and the fact that if A is an effective divisor with support E, then E − ⌈A⌉ = −⌊(1 − ǫ)A⌋ for 0 < ǫ ≪ 1. (cid:3) As in [MP16] in the case of reduced divisors, we obtain therefore that for every Q-divisor D we have that I0(D) = OX if and only if the pair (X, D) is log canonical, which leads to the following: Definition 9.2. The pair (X, D) is k-log canonical if I0(D) = · · · = Ik(D) = OX.3 Note however that by Remark 4.3, the triviality of any Ik(D) is possible only if k(D). ⌈D⌉ = Z; in general it is more suitable to focus on the triviality of the ideals I ′ We therefore introduce also: Definition 9.3. The pair (X, D) is reduced k-log canonical if I ′ 0(D) = · · · = I ′ k(D) = OX, or equivalently I0(D) = · · · = Ik(D) = OX(Z − ⌈D⌉). Example 9.4. Let Z have an ordinary singularity, i.e. an isolated singular point whose projectivized tangent cone is smooth, of multiplicity m. If D = αZ with 0 < α ≤ 1, then (X, D) is k−log canonical ⇐⇒ k ≤ [ n m − α]. See Corollary 11.8 and Remark 11.9. C. Local study and global vanishing theorem 10. Generation level of the Hodge filtration, and examples. As above, we consider a divisor D = αH, with H = div(h) for some nonzero h ∈ OX(X) and α ∈ Q>0. We denote by Z the support of D, and β = 1 − α. By construction, the filtration on M(hβ) is compatible with the order filtration on DX . This means that for every k, ℓ ≥ 0 we have (10.1) (10.2) or equivalently for every k ≥ 0 we have FℓDX ·(cid:0)Ik(D) ⊗ OX (kZ + H)hβ(cid:1) ⊆ Ik+ℓ(D) ⊗ OX ((k + ℓ)Z + H)hβ, F1DX ·(cid:0)Ik(D) ⊗ OX (kZ + H)hβ(cid:1) ⊆ Ik+1(D) ⊗ OX ((k + 1)Z + H)hβ. 3We note that by the results in [MP18b, §5], at least in the case of divisors of the form D = αZ, with α ∈ Q>0, this condition is equivalent simply to Ik(D) = OX (cf. Remark 4.2). 28 M. MUSTAT¸ A AND M. POPA By working locally, we may assume that we also have an equation g for Z. With this notation, condition (10.2) is equivalent to the following two conditions: (10.3) g · Ik(D) ⊆ Ik+1(D) and for every derivation Q of OX and every w ∈ Ik(D), we have (10.4) g · Q(w) − kw · Q(g) − αgw · Q(h) h ∈ Ik+1(D). We now turn to the problem of describing the generation level of the filtration on M(hβ). Recall that one says that the filtration is generated at level k if FℓDX · FkM(hβ) = Fk+ℓM(hβ) for all ℓ ≥ 0, or in other words if equality is satisfied in (10.1). This is of course equivalent to having F1DX · FpM(hβ) = Fp+1M(hβ) for all p ≥ k. Suppose now that we are in the setting of Theorem 8.1. Theorem 10.1. The filtration on M(hβ) is generated at level k if and only if Rqf∗(cid:0)Ωn−q Y (log E) ⊗OY OY (−⌈f ∗D⌉)(cid:1) = 0 In particular, the filtration is always generated at level n − 1. for q > k. Proof. The proof follows almost verbatim that of [MP16, Theorem 17.1]. It is more convenient to work equivalently with M(h−α), and in fact with the associated right DX -module Mr(h−α). It is enough to show that (10.5) Fk−nMr(h−α) · F1DX = Fk−n+1Mr(h−α) if and only if Rk+1(cid:0)f∗Ωn−k−1 Y (log E) ⊗OY OY (−⌈f ∗D⌉)(cid:1) = 0. The inclusion "⊆" in (10.5) always holds of course by the definition of a filtration, hence the issue is the reverse inclusion. With the notation in §6, for every p let where g = h ◦ f . Consider the morphism of complexes C • p := Fp(cid:0)C • g−α(−⌈f ∗D⌉) ⊗DY DY →X(cid:1), Φk : C • k−n ⊗f −1OX f −1F1DX −→ C • k+1−n induced by right multiplication, and let T • = Ker(Φk). Using Theorem 8.1, we see that (10.5) holds if and only if the morphism (10.6) R0f∗C • k−n ⊗OX F1DX −→ R0f∗C • k+1−n induced by Φk is surjective. For every m ≥ 0, let Rm be the kernel of the morphism induced by right multipli- cation FmDX ⊗OX F1DX −→ Fm+1DX . HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 29 Note that this is a surjective morphism of locally free OX-modules, hence Rm is a locally free OX -module and for every p we have T p = OY (−⌈f ∗D⌉) ⊗OY Ωn+p Y (log E) ⊗f −1OX f −1Rk+p. Consider the first-quadrant hypercohomology spectral sequence Ep,q 1 = Rqf∗T p−n =⇒ Rp+q−nf∗T •. The projection formula gives Rqf∗T p−n ≃ Rqf∗(cid:0)OY (−⌈f ∗D⌉) ⊗OY Ωp Y (log E)(cid:1) ⊗OX Rk+p−n, and this vanishes for p + q > n by Corollary 8.3. We thus deduce from the spectral sequence that Rjf∗T • = 0 for all j > 0. We first consider the case when k ≥ n and show that (10.5) always holds. Indeed, in this case Φk is surjective. It follows from the projection formula and the long exact sequence in cohomology that we have an exact sequence R0f∗C • k−n ⊗OX F1DX → R0f∗C • k+1−n → R1f∗T •. We have seen that R1f∗T • = 0, hence the morphism in (10.6) is surjective. Suppose now that 0 ≤ k < n. Let B• ֒→ C • k+1−n be the subcomplex given by k+1−n for all p 6= −k − 1 and B−k−1 = 0. Note that we have a short exact Bp = C p sequence of complexes (10.7) 0 −→ B• −→ C • k+1−n −→ C −k−1 k+1−n[k + 1] −→ 0. It is clear that Φk factors as C • k−n ⊗f −1OX f −1F1DX Φ′ k−→ B• ֒→ C • k+1−n. k) = T •. As before, since R1f∗T • = 0, we k is surjective and Ker(Φ′ Moreover, Φ′ conclude that morphism induced by Φ′ k: R0f∗C • k−n ⊗OX F1DX → R0f∗B• is surjective. This implies that (10.6) is surjective if and only if the morphism (10.8) R0f∗B• → R0f∗C • k+1−n is surjective. The exact sequence (10.7) induces an exact sequence R0f∗B• → R0f∗C • k+1−n → Rk+1f∗C −k−1 k+1−n → R1f∗B•. We have seen that R2f∗T • = 0, and we also have R1f∗(cid:0)C • k−n ⊗f −1OX f −1F1DX(cid:1) = 0. This follows as above, using the projection formula, the hypercohomology spectral sequence, and Corollary 8.3. We deduce from the long exact sequence associated to 0 −→ T • −→ C • k−n ⊗f −1OX f −1F1DX −→ B• −→ 0 30 M. MUSTAT¸ A AND M. POPA that R1f∗B• = 0. Putting all of this together, we conclude that (10.6) is surjective if and only if Rk+1f∗C −k−1 k+1−n = 0. Since by definition we have Rk+1f∗C −k−1 k+1−n = Rk+1f∗(cid:0)OY (−⌈f ∗D⌉) ⊗OY Ωn−k−1 Y (log E)(cid:1), this completes the proof of the first assertion in the proposition. The second assertion follows from the first, since all fibers of f have dimension < n. (cid:3) Example 10.2 (Nodal curves). If X is a smooth surface and Z is a reduced curve on X, defined by h ∈ O(X), such that Z has a node at x ∈ X and no other singularities, then the filtration on M(hβ ) is generated at level 0. Indeed, let f : Y → X be the blow-up of X at x, with exceptional divisor F . This is a log resolution of (X, Z), hence our assertion follows if we show that (10.9) R1f∗(cid:0)ΩY (log E) ⊗OY OY (−⌈αf ∗Z⌉)(cid:1) = 0, where E = eZ + F . Note that f ∗Z = eZ + 2F and we may assume that 0 < α ≤ 1. If 1 2 < α ≤ 1, then ⌈αf ∗Z⌉ = f ∗Z and (10.9) follows from [MP16, Theorem B] using the projection formula. On the other hand, if 0 < α ≤ 1 2 , then ⌈αf ∗Z⌉ = E and the vanishing follows from the fact that the pair (X, Z) is log canonical, using [GKKP11, Theorem 14.1] (though, in this case, one could also check this directly). Once we know that the filtration on M(hβ) is generated at level 0, it is straight- forward to check that Ik(αZ) = m k x, for all 0 < α ≤ 1 and k ≥ 0, where mx is the ideal defining x in X. Unlike in the case when D is a reduced integral divisor, when the filtration F•OX (∗D) is generated at level n − 2 by [MP16, Theorem B], in general it is not possible to im- prove the bound given by Proposition 10.1. Example 10.3 (Optimal generation level). It can happen that on a surface X the filtration on M(hβ) is not generated at level 0. Suppose, for example, that X = A2 and Z = L1 + L2 + L3, where L1, L2, and L3 are 3 lines passing through the origin. If f : Y → X is the blow-up of the origin and E = (f ∗Z)red, then we write E = F + G1 + G2 + G3, where F is the exceptional divisor and the Gi are the strict transforms of the Li. Let D = αZ with 0 < α ≪ 1, so that ⌈f ∗D⌉ = E. If H 1(cid:0)Y, ΩY (log E) ⊗OY OY (−⌈f ∗D⌉)(cid:1) = H 1(cid:0)Y, ΩY (log E) ⊗OY were zero, then it would follow from the standard exact sequence OY (−E)(cid:1) 0 → ΩY (log E) ⊗OY OY (−E) → ΩY → ΩF ⊕ 3Mi=1 ΩGi → 0 that the map H 0(Y, ΩY ) → H 0(F, ΩF ) ⊕ 3Mi=1 H 0(Gi, ΩGi) HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 31 is surjective. In particular, we would deduce that the map H 0(X, ΩX ) → 3Mi=1 H 0(Li, ΩLi) is surjective. It is an easy exercise to see that this is not the case. Note that the non- vanishing of H 1(cid:0)Y, ΩY (log E) ⊗OY OY (−E)(cid:1) is not inconsistent with the Steenbrink- type vanishing in [GKKP11, Theorem 14.1], since the pair (X, Z) is not log-canonical. Example 10.4 (Quasi-homogeneous isolated singularities). For the class of quasi-homogeneous isolated singularities (such as those in the examples above), the generation level for the filtration on M(hβ) can be detected by the Bernstein-Sato polynomial. Before formulating this more precisely, we recall some definitions. Sup- pose that Z is a hypersurface in X defined by h ∈ OX(X). The Bernstein-Sato polynomial of Z is the non-zero monic polynomial bh ∈ C[s] of smallest degree such that we locally have a relation of the form for some nonzero P ∈ DX[s]. bh; moreover, all the roots of bh are negative rational numbers. bh(s)hs = P (s) • hs+1 If Z is non-empty, it is known that (s + 1) divides In this case, one defines eαh = −λ, where λ is the largest root of the reduced Bernstein-Sato polynomial ebh = bh(s)/(s + 1). Note thatebh has degree 0 if and only if Z is smooth, and in this case one makes the convention that eαh = ∞. The statement is that if Z = div(h) is reduced and has a unique singular point at x, which is a quasi-homogeneous singularity, and D = αZ, then the generation level k0 of the filtration on M(hβ) (i.e. the smallest k such that the filtration is generated at level k) is This was proved by Saito [Sai09, Theorem 0.7] when D is reduced, i.e. for α = 1, and was extended to the general case by Zhang [Zha18].4 k0 = ⌊n −eαh − α⌋. fact that the generation level is equal to 1. §4.1]. Just as an illustration, for h = xy(x+y), which describes the previous example, Note that for such singularities there is an explicit formula for eαh; see e.g. [Sai09, we have eαh = 2/3, and so for α small (more precisely 0 < α ≤ 1/3) we recover the a smooth surface and Z =Pr Let D =Pr Example 10.5 (Incomensurability of higher Hodge ideals). Suppose that X is i=1 Di is a reduced effective divisor on X. Let f : Y → X be a log resolution of (X, Z) that is an isomorphism over X rZ, and put E = (f ∗Z)red. i=1(1 − ai)Di be a divisor with 0 ≤ ai ≪ 1 for all i, so that ⌈f ∗D⌉ = f ∗Z. In this case we have R1f∗(cid:0)OY (−⌈f ∗D⌉) ⊗ Ω1 Y (log E)(cid:1) = 0 by the projection formula and [MP16, Theorem B], and so the filtration is generated at level 0. It follows from the discussion at the beginning of the section (see (10.3) 4Moreover, based on calculations of Saito, Zhang shows in loc. cit. that all Hodge ideals of Q-divisors associated to such singularities can be computed explicitly. 32 M. MUSTAT¸ A AND M. POPA and (10.4)) that if g is a local equation of Z, and D = αZ, with α ≤ 1 and close to 1, then Ik+1(D) is generated by g · Ik(D) and {h · Q(w) − (α + k)w · Q(h) w ∈ Ik(D), Q ∈ DerC(OX )} . For example, if X = C2 and Z is the cusp defined by x2 + y3, then for D = αZ with α ≤ 1 and close to 1 we have I0(D) = (x, y), I1(D) = (x2, xy, y3), and I2(D) = (x3, x2y2, xy3, y4 − (2α + 1)x2y). Note in particular that if D1 = α1Z and D2 = α2Z, with α1 < α2 both close to 1, then there is no inclusion between the ideals I2(D1) and I2(D2). This is in contrast with the picture for multiplier ideals, where for any Q-divisors D1 ≤ D2 one has I0(D2) ⊆ I0(D1); see [Laz04, Proposition 9.2.32(i)]. It is not hard to check however that I2(D1) = I2(D2) mod x2 + y3, and that this is part of a general phenomenon where the picture is well behaved after modding out by a defining equation for the hypersurface; this follows from the connection with the V -filtration, see [MP18b, Corollary B]. Remark 10.6. If the filtration is generated at level k, then Ik+1(D) is generated by the terms appearing on the left hand side of conditions (10.3) and (10.4). A simple calculation shows then that in this case, for every j ≥ 1 and every x ∈ X, we have multxIk+j(D) ≥ multxIk+j−1(D) + multxZ − 1. In particular, we have multxIk+j(D) ≥ multxIk(D) + j · (multxZ − 1). Since the filtration is always generated at level n − 1 by Proposition 10.1, we obtain the following consequence. Corollary 10.7. If D is an effective Q-divisor on the smooth variety X, with support Z, and if Z is singular at some x ∈ X, then Ij(D)x 6= OX,x for all j ≥ n. In fact, if m = multxZ, then multxIj(D) ≥ (j − n + 1)(m − 1) for all j ≥ n. 11. Non-triviality criteria. The following is the analogue of [MP16, Theorem 18.1] in the setting of Q-divisors. Let D be an effective Q-divisor on the smooth variety X, with Z = Supp(D), and let ϕ : X1 → X be a projective morphism with X1 smooth, such that ϕ is an isomorphism over X r Z. We denote Z1 = (ϕ∗Z)red and TX1/X = Coker(TX1 → ϕ∗TX ). Theorem 11.1. With the above notation, the following hold: i) We have an inclusion ϕ∗(cid:0)Ik(ϕ∗D) ⊗OX1 OX1(KX1/X + k(Z1 − ϕ∗Z))(cid:1) ⊆ Ik(D). HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 33 ii) If J is a coherent ideal on X such that J · TX1/X = 0, then J k · Ik(D) ⊆ ϕ∗(cid:0)Ik(ϕ∗D) ⊗OX1 OX1(KX1/X + k(Z1 − ϕ∗Z))(cid:1). Proof. We may assume that D = α · div(h), for some α ∈ Q>0 and some nonzero h ∈ OX (X). Let ψ : Y → X1 be a log resolution of (X1, ϕ∗D) that is an isomorphism over X1 r ϕ−1(Z). We put f = ϕ ◦ ψ and E = (f ∗Z)red. With the notation in §6, consider the filtered complex C • = C • g = h ◦ f . We have an inclusion of complexes g−α(−⌈f ∗D⌉), where A• = C • ⊗DY DY →X1 ֒→ B• = C • ⊗DY DY →X. Note that this is an injection due to the fact that OY (−⌈f ∗D⌉) and Ωq Y (log E) are locally free sheaves of OY -modules, while all the maps FpDY →X1 → FpDY →X are generically injective morphisms of locally free OY -modules. Consider, for any integer k, the short exact sequence of complexes 0 −→ Fk−nA• −→ Fk−nB• −→ M• −→ 0. Applying Rf∗ and taking the corresponding long exact sequence, we obtain a short exact sequence R0f∗Fk−nA• ι−→ R0f∗Fk−nB• −→ R0f∗M •. If β = 1 − α, it follows from Theorem 8.1 that Rf∗Fk−nB• = R0f∗Fk−nB• ≃ hβ OX(cid:0)KX + kZ + H(cid:1) ⊗OX Ik(D) and Rg∗Fk−nA• = R0g∗Fk−nA• ≃ hβ OX1(cid:0)KX1 + kZ1 + ϕ∗H(cid:1) ⊗OX1 Therefore, after tensoring by OX (−H), the map ι induces a map Ik(ϕ∗D). (11.1) ϕ∗(cid:0)Ik(ϕ∗D) ⊗ OX1(KX1 + kZ1)(cid:1) → Ik(D) ⊗OX OX(cid:0)KX + kZ(cid:1). Finally, the map ι is compatible with restriction to open subsets of X. By restricting to an open subset X0 in the complement of Z, such that f is an isomorphism over X0, we see that the map in (11.1) is the identity on ωX0. We thus deduce the assertion in i) by tensoring (11.1) with OX(cid:0) − KX − kZ(cid:1). Furthermore, we see that the assertion in ii) follows if we show that J k · R0f∗M • = 0. Since M p = OY (−⌈f ∗D⌉) ⊗OY Ωn+p Y (log E) ⊗OY ψ∗(ϕ∗Fk+pDX/Fk+pDX1 ), it is enough to show that under our assumption we have ϕ∗Fj DX · J j ⊆ Fj DX1 for all j ≥ 0. This is proved in [MP16, Lemma 18.6]. (cid:3) We first use Theorem 11.1 in order to give a triviality criterion for Hodge ideals in terms of invariants of a fixed resolution of singularities. We use this in turn in order to bound the largest root of the reduced Bernstein-Sato polynomial (i.e. eαh defined in Example 10.4) in terms of such invariants, in [MP18b, Corollary D]. 34 M. MUSTAT¸ A AND M. POPA Proposition 11.2. Let Z be a reduced divisor on the smooth variety X, and let D = αZ, with α ∈ Q>0. Let f : Y → X be a log resolution of (X, Z) that is an isomorphism over X r Z and such that the strict transform eZ of Z is smooth. We define integers ai and bi by the expressions aiFi and KY /X = biFi, mXi=1 f ∗Z = eZ + mXi=1 bi + 1 ai where F1, . . . , Fm are the prime exceptional divisors. If (11.2) ≥ k + α for 1 ≤ i ≤ m, then Ik(D) = OX(cid:0)(1 − ⌈α⌉)Z(cid:1). In particular, if 0 < α ≤ 1, then Ik(D) = OX . Ik(D) = Ik(D′) ⊗ OX(cid:0)(1 − ⌈α⌉)Z(cid:1). Since the inequalities (11.2) clearly also hold if Proof. If D′ = α′Z, where α′ = α + 1 − ⌈α⌉, then it follows from Lemma 4.4 that we replace α by α′, it follows that it is enough to treat the case 0 < α ≤ 1. First, note that since f ∗D has simple normal crossings, by Proposition 7.1 we have where E = (f ∗Z)red = eZ +Pm (11.3) where Ik(f ∗D) = Ik(E) ⊗ OY(cid:0) mXi=1 (1 − ⌈αai⌉)Fi(cid:1), f∗(cid:0)Ik(E) ⊗ OY (F )(cid:1) ֒→ Ik(D), mXi=1(cid:0)bi + k + 1 − kai − ⌈αai⌉(cid:1)Fi. F := i=1 Fi. We apply Theorem 11.1 i) to obtain the inclusion smooth, it follows from the description of Hodge ideals of simple normal crossing divisors in [MP16, Proposition 8.2] that we have On the other hand, since E = Z +Pm i=1 Fi has simple normal crossings and eZ is mXi=1 Fi) ⊆ Ik(E). OY (−k · Note that the inequalities in (11.2) imply bi + 1 ≥ kai + ⌈αai⌉ for all i, hence the i=1 Fi is effective We thus deduce using (11.3) that we have divisor F − k ·Pm OX = f∗OY ֒→ Ik(D). Remark 11.3. More generally, suppose that we write Z = Pr an effective Q-divisor D =Pr j=1 Zj, and consider j=1 αjZj supported on Z. For simplicity, let us assume that 0 < αj ≤ 1 for all j. If f is a log resolution as in Proposition 11.2, and we write (cid:3) f ∗Zj = fZj + mXi=1 aj i Fi HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 35 for all j (so that ai =Pr j=1 aj i ), then the same proof gives Ik(D) = OX if bi + 1 ≥ kai + mXi=1 αjaj i for all i. We now turn our attention to non-triviality criteria for the Hodge ideals Ik(D) in terms of the multiplicity of D, and of its support Z, along a given subvariety. Corollary 11.4. Let D be an effective Q-divisor on the smooth variety X, and let Z be the support of D. If W is an irreducible closed subset of X of codimension r such that multW Z = a and multW D = b, and if q is a non-negative integer such that then Ik(D) ⊆ I (q) W , the q-th symbolic power of IW . In particular, if b + ka > q + r + 2k − 1, multW D > q + r + 2k − 1 k + 1 , then Ik(D) ⊆ I (q) W . Proof. After possibly restricting to a suitable open subset of X meeting W , we may assume that W is smooth. The first assertion in the corollary follows by applying Theorem 11.1(ii) to the blow-up ϕ : X1 → X along W . Note that we may take J = IW by [MP16, Example 18.7], while Ik(ϕ∗D) ⊆ OX1(Z1 − ⌈ϕ∗D⌉). The last assertion follows thanks to the fact that by assumption we have a ≥ b. (cid:3) Remark 11.5. An interesting consequence of the above corollary is that if Z is a reduced divisor on the smooth, n-dimensional variety X, k is a positive integer, and x ∈ X is a point such that multxZ ≥ 2 + n k , then Ik(D) is non-trivial at x for every effective Q-divisor D with support Z (no matter how small the coefficients). Example 11.6 (Ordinary singularities, I). Let X be a smooth variety of dimen- sion n, and Z a reduced divisor with an ordinary singularity at x ∈ X (recall that this means that the projectivized tangent cone of Z at x is smooth), for instance a cone over a smooth hypersurface. If D = αZ, with α a rational number satisfying 0 < α ≤ 1, then multxZ ≤ =⇒ Ik(D)x = OX,x. n k + α Note that the converse of this statement will be proved in Corollary 11.8 below. Indeed, the assumption implies that after possibly replacing X by an open neigh- borhood of x, the blow-up f : Y → X of X at x gives a log resolution of (X, Z). Let E = F + eZ, where F is the exceptional divisor of f and eZ is the strict transform of Z. If m = multxZ, then we deduce from Theorem 11.1 that f∗(cid:0)Ik(ϕ∗D) ⊗OY OY (KY /X + k(E − f ∗Z))(cid:1) 36 M. MUSTAT¸ A AND M. POPA = f∗(cid:0)Ik(ϕ∗D) ⊗OY OY ((n − 1 + k − km)F )(cid:1) ⊆ Ik(D). Now since ϕ∗D is supported on the simple normal crossings divisor E, by Proposition 7.1 we have where we use the fact that ⌈α⌉ = 1. Moreover, by [MP16, Proposition 8.2] we have Ik(ϕ∗D) = Ik(E) ⊗OY OY ((1 − ⌈αm⌉)F ), Now by assumption Ik(E) =(cid:0)OY (−eZ) + OY (−F )(cid:1)k ⊇ OY (−kF ). n − km − ⌈αm⌉ ≥ 0, hence we deduce Ik(D) = OX. Example 11.7 (Ordinary singularities, II). With considerable extra work, one can say more in the ordinary case. We keep the notation of the previous example, and assume that x is a singular point of Z, hence m ≥ 2. If k is a positive integer such that (k − 1)m + ⌈αm⌉ < n and k ≤ n − 2, then we have Ik(D) = mkm+⌈αm⌉−n x in a neighborhood of x, where mx is the ideal defining x (with the convention that j x = OX if j ≤ 0). The argument is similar to that in [MP16, Proposition 20.7], so m we omit it. In what follows we make use of some general properties of Hodge ideals that will be proved in Ch.D, namely the Restriction and Semicontinuity Theorems. Corollary 11.8. If X is a smooth n-dimensional variety, Z is a reduced divisor with an ordinary singularity of multiplicity m ≥ 2 at x ∈ X, and D = αZ with 0 < α ≤ 1, then Ik(D)x = OX,x ⇐⇒ m ≤ n . k + α Proof. The "if" part follows directly from Example 11.6. For the converse, we need to show that if mx is the ideal defining x and m > n k+α , then Ik(D) ⊆ mx. We may assume that Z is defined in X by h ∈ OX (X). Let r ≥ 0 be such that n + r = 1 + · · · + ym mk + ⌈mα⌉ − 1 and consider the divisor Z ′ in X × Cr defined by h + ym r , where y1, . . . , yr are the coordinates on Cr. It is easy to check that Z ′ is reduced and has an ordinary singularity at (x, 0). By the Restriction Theorem (see Theorem 13.1 and Remark 13.4 below), we have Ik(αZ) ⊆ Ik(αZ ′) · OX , where we consider X embedded in X × Cr as X × {0}. After replacing X and Z by X ′ and Z ′, we may thus assume that n = mk + ⌈mα⌉ − 1. If k ≤ n − 2, then we may apply Example 11.7 to conclude that Ik(D) ⊆ mx. Otherwise we have k ≥ n − 1 = mk + ⌈mα⌉ − 2, which easily implies m = 2, k = 1, and α ≤ 1 2 , hence n = 2. Since Z has an ordinary singularity at x, it follows that it must be a node, and in this case we have I1(αZ) = mx by Example 10.2. (cid:3) HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 37 Remark 11.9. One can give an alternative argument, arguing as follows. Suppose that Z is a reduced divisor in X, defined by h ∈ OX(X). It is shown in [MP18b, Z has an ordinary singularity at x ∈ X, of multiplicity m ≥ 2, then after replacing X m (see [Sai16, §2.5]), and we recover Corollary C] that for 0 < α ≤ 1, we have Ik(αZ) = OX if and only if k ≤ eαh − α. If by a suitable neighborhood of x, we have eαh = n the assertion in Corollary 11.8. Question 11.10. Is it true that if X is a smooth n-dimensional variety, Z is a reduced divisor on X, D is an effective Q-divisor with support Z, and for a point x ∈ Zsing we have then Ik(D) ⊆ mx? k · multxZ + multxD > n, This would be a natural improvement of Corollary 11.4, and it does hold when D is reduced by [MP16, Corollary 21.3]. We may of course assume that ⌈D⌉ = Z, since otherwise the inclusion is trivial (see Remark 4.3). At the moment we have: Corollary 11.11. Question 11.10 has a positive answer if D is of the form D = αZ. Proof. We may assume that α ≤ 1 and, arguing as in the proof of [MP16, Theorem E], we construct a reduced divisor F on X × U , for a smooth variety U , such that for t ∈ U general the divisor Ft = F X×{t} is reduced, with an ordinary singularity at (x, t) of multiplicity m = multxZ, and for some t0 ∈ U , the isomorphism X ≃ X ×{t0} maps D to Ft0 . In this case Corollary 11.8 implies that Ik(Ft) vanishes at (x, t) for t ∈ U general, and the Semicontinuity Theorem (see Theorem 14.1 below) implies that Ik(Ft0 ) vanishes at (x, t0). (cid:3) This allows us in particular to provide an analogue of [MP16, Theorem A]: Corollary 11.12. If D is of the form D = αZ, then Z is smooth ⇐⇒ Ik(D) = OX (Z − ⌈D⌉) for all k. Proof. It suffices to assume 0 < α ≤ 1, in which case the condition becomes Ik(D) = OX for all k. By Corollary 11.11 however, if multxZ ≥ 2, then Ik(D) ⊆ mx for all k > n (cid:3) 2 − α. 12. Vanishing theorem. As usual, we consider an effective Q-divisor D with support Z, on the smooth variety X. In this section we assume that X is projective, and prove a vanishing theorem for Hodge ideals, extending [MP16, Theorem F] as well as Nadel Vanishing for Q-divisors. We start by choosing a positive integer ℓ such that ℓD is an integral divisor, and further assume that there exists a line bundle M on X such that (12.1) M ⊗ℓ ≃ OX(ℓD), so that the setting of §5 applies. We note that this can always be achieved after passing to a finite flat cover of X. 38 M. MUSTAT¸ A AND M. POPA Theorem 12.1. Let X be a smooth projective variety of dimension n and D an effective Q-divisor on X such that (12.1) is satisfied. Let L be a line bundle on X such that L + Z − D is ample. For some k ≥ 0, assume that the pair (X, D) is reduced (k − 1)-log-canonical, i.e. I0(D) = · · · = Ik−1(D) = OX(Z − ⌈D⌉).5 Then we have: (1) If k ≤ n, and L(pZ − ⌈D⌉) is ample for all 2 ≤ p ≤ k + 1, then H i(cid:0)X, ωX ⊗ L((k + 1)Z) ⊗ Ik(D)(cid:1) = 0 H 1(cid:0)X, ωX ⊗ L((k + 1)Z) ⊗ Ik(D)(cid:1) = 0 X ⊗ L((k − j + 2)Z − ⌈D⌉)(cid:1) = 0 for all 1 ≤ j ≤ k. for all i ≥ 2. Moreover, holds if H j(cid:0)X, Ωn−j (2) If k ≥ n + 1, then Z must be smooth by Corollary 10.7, and so Ik(D) = OX(Z − ⌈D⌉) by Corollary 3.2. In this case, if L is a line bundle such that L((k + 1)Z − ⌈D⌉) is ample, then 5Recall from Definition 9.3 that equivalently this means I ′ 0(D) = · · · = I ′ k−1(D) = OX . By convention the condition is vacuous when k = 0. 6When k ≥ 1, the condition of U being affine is in fact implied by the positivity condition, since D + Z − ⌈D⌉ is then an ample divisor with support Z. H i(cid:0)X, ωX ⊗ L((k + 1)Z) ⊗ Ik(D)(cid:1) = 0 for all i > 0. (3) If U = X r Z is affine (e.g. if D or Z are ample), then (1) and (2) also hold with L = M (−Z), assuming that M (pZ − ⌈D⌉) is ample for 1 ≤ p ≤ k.6 Proof. We use the notation in §5 and Remark 4.3. filtered left DX -module In particular, we consider the M1 = M ⊗OX OX(∗Z), which we know is a direct summand in a filtered D-module underlying a mixed Hodge module on X. Its filtration satisfies FkM1 ≃ M (−Z) ⊗ OX(cid:0)(k + 2)Z − ⌈D⌉(cid:1) ⊗ I ′ k(D). Note also that since L + Z − D is ample, there exists an ample line bundle A on X such that L ≃ M (−Z) ⊗ A. Let's prove (1), i.e. consider the case k ≤ n. The statement is equivalent to the vanishing of the cohomology groups H i(cid:0)X, ωX ⊗ L((k + 2)Z − ⌈D⌉) ⊗ I ′ k(D)(cid:1) = 0 Since I ′ k−1(D) = OX, we have a short exact sequence 0 −→ ωX ⊗ L((k + 1)Z − ⌈D⌉) −→ ωX ⊗ L((k + 2)Z − ⌈D⌉) ⊗ I ′ k(D) −→ −→ ωX ⊗ A ⊗ grF k M1 −→ 0. By taking the corresponding long exact sequence in cohomology and using Kodaira vanishing, we see that the vanishing we are aiming for is equivalent to the same statement for H i(cid:0)X, ωX ⊗ A ⊗ grF k M1(cid:1). HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 39 We now consider the complex C • :=(cid:0) grF −n+k DR(M1) ⊗ A(cid:1)[−k]. p(D), this can be identified with a complex of the Given the hypothesis on the ideals I ′ form (cid:2)Ωn−k X ⊗ L(2Z − ⌈D⌉) −→ Ωn−k+1 X ⊗ L ⊗ OZ (3Z − ⌈D⌉) −→ · · · placed in degrees 0 up to k. Saito's Vanishing theorem [Sa90, §2.g] gives X ⊗ L ⊗ OZ(cid:0)(k + 1)Z − ⌈D⌉(cid:1) −→ ωX ⊗ A ⊗ grF k M1(cid:3) · · · −→ Ωn−1 (12.2) Hj(X, C •) = 0 for all j ≥ k + 1. We use the spectral sequence Ep,q 1 = H q(X, C p) =⇒ Hp+q(X, C •). The vanishing statements we are interested in are for the terms Ek,i will in fact show that 1 with i ≥ 1. We (12.3) This implies that Ek,i r = Ek,i r+1, for all r ≥ 1. Ek,i 1 = Ek,i ∞ = 0, where the vanishing follows from (12.2) since i ≥ 1, and this gives our conclusion. We are thus left with proving (12.3). Now on one hand we always have Ek+r,i−r+1 = 0 because C k+r = 0. On the other hand, we will show that under our hypothesis we have Ek−r,i+r−1 = 0 as well, allowing us to conclude. To this end, note first that if r > k this vanishing is clear, since the complex C • starts in degree 0. If k = r, we have = 0, from which we infer that Ek−r,i+r−1 1 r r E0,i+k−1 1 = H i+k−1(cid:0)X, Ωn−k X ⊗ L(2Z − ⌈D⌉)(cid:1). If i ≥ 2 this is 0 by Nakano vanishing, while if i = 1 it is 0 because of our hypothesis. Finally, if k ≥ r + 1, we have which sits in an exact sequence Ek−r,i+r−1 1 X ⊗ L ⊗ OZ ((k − r + 2)Z − ⌈D⌉)(cid:1), = H i+r−1(cid:0)X, Ωn−r X ⊗ L((k − r + 2)Z − ⌈D⌉)(cid:1) −→ Ek−r,i−r+1 −→ H i+r(cid:0)X, Ωn−r X ⊗ L((k − r + 1)Z − ⌈D⌉)(cid:1). −→ 1 H i+r−1(cid:0)X, Ωn−r We again have two cases: (1) If i ≥ 2, we deduce that Ek−r,i+r−1 (2) If i = 1, using Nakano vanishing we obtain a surjective morphism = 0 by Nakano vanishing. 1 H r(cid:0)X, Ωn−r X ⊗ L((k − r + 2)Z − ⌈D⌉)(cid:1) −→ Ek−r,i+r−1 1 , and if the extra hypothesis on the term on the left holds, then we draw the same conclusion as in (1). 40 M. MUSTAT¸ A AND M. POPA The same argument proves (3), once we replace Saito Vanishing (12.2) by the vanishing Hi(cid:0)X, grF k DR(M1)(cid:1) = 0 for all i > 0 and all k, which in turn is implied by the same statement for the DX -module M underlying a Hodge D-module, in which M1 is a direct summand. Furthermore, this is implied by the vanishing of the perverse sheaf cohomology Indeed, by the strictness property for direct images (see e.g. [MP16, Example 4.2]), for (M, F ) we have the decomposition H i(cid:0)X, DR(M)(cid:1) = 0 for all i > 0. H i(cid:0)X, DR(M)(cid:1) ≃Mq∈Z Hi(cid:0)X, grF −q DR(M)(cid:1). Recall now from §5 that M ≃ j+N , where N underlies a Hodge D-module on U , and j : U ֒→ X is the inclusion. Denoting P = DR(M), we then have P ≃ j∗j∗P , and so it suffices to show that H i(U, j∗P ) = 0 for all i > 0. But this is a consequence of Artin vanishing (see e.g. [Dim04, Corollary 5.2.18]), since U is affine. Finally, the assertion in (2) follows from Kodaira vanishing, using the long exact sequence in cohomology associated to the short exact sequence 0 → ωX⊗L(cid:0)(k+1)Z−⌈D⌉(cid:1) → ωX⊗L(cid:0)(k+2)Z−⌈D⌉(cid:1) → ωZ⊗L(cid:0)(k+1)Z−⌈D⌉(cid:1)Z → 0. (cid:3) Remark 12.2. We expect the statement of the theorem to hold even without assum- ing the existence of M (i.e. of an ℓ-th root of the line bundle OX(ℓD)). This is known for k = 0, when the statement follows from Nadel Vanishing, see [Laz04, Theorem 9.4.8]. However, at the moment we do not know how to show this for k ≥ 1. Remark 12.3 (Toric varieties). As in [MP16, Corollary 25.1], when X is a toric variety the Nakano-type vanishing requirement in Theorem 12.1(1) is automatically satisfied thanks to the Bott-Danilov-Steenbrink vanishing theorem. A stronger result in this setting is proved in [Dut18]. Remark 12.4 (Projective space, abelian varieties). As in [MP16, Theorem 25.3 and 28.2], appropriate statements on Pn and abelian varieties work without the extra assumptions of reduced log canonicity and Nakano-type vanishing in Theorem 12.1. More precisely, keeping the notation at the beginning of the section, we have: Variant 12.5. Let D be an effective Q-divisor on Pn which is numerically equivalent to a hypersurface of degree d ≥ 1. If ℓ ≥ d − n − 1, then H i(cid:0)Pn, OPn (ℓ) ⊗ OPn(kZ) ⊗ Ik(D)(cid:1) = 0 for all i > 0. Note that the positivity condition in Theorem 12.1 is satisfied, since for every effective Q-divisor D 6= 0 in Pn we have deg⌈D⌉ < deg D + deg Z. HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 41 Variant 12.6. If X is an abelian variety and D is an ample Q-divisor on X, then H i(X, M (kZ) ⊗ Ik(D) ⊗ α) = 0 for all i > 0 and α ∈ Pic0(X). Note that on an abelian variety every effective Q-divisor is nef, and the ampleness of D is equivalent to that of any divisor whose support is equal to that of D. The proofs are completely similar to those in loc. cit., replacing OX(∗D) in the reduced case by M1 in the proof above, and noting that since M1 is a filtered direct summand in j+p+OV as in §5, the vanishing properties we use continue to hold. D. Restriction, subadditivity, and semicontinuity theorems In this part of the paper we provide Q-divisor analogues of the results in [MP18a]. This extends well-known statements in the setting of multiplier ideals; further discus- sion and references regarding these can be found in loc. cit. 13. Restriction theorem. We begin with the Q-divisor version of the Restriction Theorem: Theorem 13.1. Let D be an effective Q-divisor, with support Z, on the smooth variety X, and let Y be a smooth irreducible divisor on X such that Y 6⊆ Z. If we denote DY = DY , ZY = ZY , and Z ′ Y = (ZY )red, then for every k ≥ 0 we have (13.1) OY(cid:0) − k(ZY − Z ′ Y )(cid:1) · Ik(DY ) ⊆ Ik(D) · OY . In particular, if ZY is reduced, then for every k ≥ 0 we have (13.2) Ik(DY ) ⊆ Ik(D) · OY . Moreover, if Y is sufficiently general (e.g. a general member of a basepoint-free linear system), then we have equality in (13.2). Remark 13.2. Note that when D is a reduced divisor we have DY = ZY , and DY − Z ′ Y is an integral divisor with support in Z ′ Y . Therefore Lemma 4.4 gives Ik(DY ) = OX(cid:0) − (DY − Z ′ Y )(cid:1) · Ik(Z ′ Y ), hence the statement in the theorem coincides with that of [MP18a, Theorem A]. Proof of Theorem 13.1. The argument follows the proof of [MP18a, Theorem A], with a simplification observed in [Sai16], hence we only give the outline of the proof. Since the statement is local, we may assume that D = α · div(h) for some nonzero We obtain, in particular, an isomorphism of filtered right DX -modules M (k) =(cid:0)M, F•−kM, K ⊗Q Q(k)(cid:1). (cid:0)H1i!Mr(h−α), F•(cid:1) ≃(cid:0)Mr(h−α Y ), F•+1(cid:1). 42 M. MUSTAT¸ A AND M. POPA h ∈ OX (X). Consider the following commutative diagram with Cartesian squares: i′′ i′ i VY p′ UY j′ Y V p U j X, where p and j are as in diagram (2.3), while i is the inclusion of Y in X. Note that if n = dim(X), we have a canonical base-change isomorphism i!(j ◦ p)+QH V [n] ≃ (j′ ◦ p′)+i ′′ ! QH V [n] proved in [Sa90, 4.4.3]. We also have a canonical isomorphism ′′ ! i QH V [n] = (QH VY [n − 1])(−1)[−1] (see for instance [Sai88, §3.5]). Here we use the Tate twist notation, which for a mixed Hodge module M = (M, F•M, K) is given by Recall now that if (VαM)α∈Q is the V -filtration on M = Mr(h−α) corresponding to the smooth hypersurface Y ⊆ X, then there is a canonical morphism σ : grV 0 M → grV −1M ⊗OX OX(Y ) such that H1i!M ≃ coker(σ), with the Hodge filtration on the right-hand side induced by the Hodge filtration on M. We refer to [MP18a, §2] for details. One defines a morphism η : FkgrV −1M = FkV−1M FkV<−1M −→ FkM ⊗OX OY that maps the class of u ∈ FkV−1M = FkM∩V−1M to the class of u in FkM⊗OX OY . After tensoring η with OX (Y ), the resulting morphism vanishes on the image of the restriction of σ to FkgrV 0 M, hence we obtain an induced morphism Y ) ≃ FkH1i!M ≃ Fkcoker(σ) → FkM ⊗OX Fk+1Mr(h−α OY (Y ). (13.3) Applying this with k replaced by k − n, it follows from the definition of Hodge ideals and the formula for the equivalence between left and right D-modules that we have a morphism Ik(DY ) ⊗OY ωY(cid:0)kZ ′ Y + div(hY )(cid:1) → Ik(D) ⊗OX ωX(cid:0)kZ + div(h)(cid:1) ⊗OX OY (Y ). HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 43 By tensoring this with ω−1 map Ik(D) ⊗OX Y (cid:0) − kZY − div(hY )(cid:1) and composing with the canonical OY → Ik(D) · OY , we obtain a canonical morphism ϕ : OY(cid:0) − k(ZY − Z ′ Y )(cid:1) ⊗ Ik(DY ) → Ik(D) · OY . Note that all constructions are compatible with restrictions to open subsets and when restricting to Z = X r U , the above morphism can be identified with the identity map on OY . Therefore the morphism ϕ is compatible with the two inclusions in OY , and we deduce the inclusion in (13.1). Suppose now that Y is general, so that ZY = Z ′ Y and Y is non-characteristic with respect to M. For example, this condition holds if Y is transversal to the strata in a Whitney stratification of Z (see [DMST06, §2]); in particular, it holds if Y is a general member of a basepoint-free linear system. We may assume that Y is defined by a global equation t ∈ OX(X). In this case, it follows from [Sai88, Lemme 3.5.6] that grV OY . It is now straightforward to check that the morphism (13.3) is an isomorphism, hence ϕ is an isomorphism, and we thus have equality in (13.2). (cid:3) 0 M = 0 and grV −1M = M ⊗OX We deduce the following analogue of inversion of adjunction: Corollary 13.3. With the notation of Theorem 13.1, if ZY is reduced and Ik(DY )x = OY,x for some x ∈ Y , then Ik(D)x = OX,x. Remark 13.4. If D is an effective Q-divisor, with support Z, on the smooth variety X, and Y is a smooth subvariety of X such that Y 6⊆ Z and ZY is reduced, then for every k ≥ 0 we have Ik(DY ) ⊆ Ik(D) · OY . This follows by writing Y locally as a transverse intersection of r smooth divisors on X and applying repeatedly the inclusion (13.2). Remark 13.5. With the notation in Theorem 13.1, let Y1, . . . , Yr be general elements in a basepoint-free linear system on X, where r ≤ n = dim(X). If W = Y1 ∩ · · · ∩ Yr, then for every k ≥ 0 we have Ik(DW ) = Ik(D) · OW . Indeed, if Wi = Y1 ∩ · · · ∩ Yi, and if (Sβ)β are the strata of a Whitney stratification of Z, then it follows by induction on i that we have a Whitney stratification of ZWi with strata (Sβ ∩ Wi)β. Moreover, Yi+1 is transversal to each such stratum. We may thus apply the theorem to each divisor DWi and smooth hypersurface Yi+1 ∩ Wi ⊆ Wi, to conclude that Ik(DW ) = Ik(D) · OW . 14. Semicontinuity theorem. The same argument as in [MP18a, §5], based on the Restriction Theorem (in this case Theorem 13.1 above), gives the following semicon- tinuity statement. The set-up is as follows: let f : X → T be a smooth morphism of relative dimension n between arbitrary varieties X and T , and s : T → X a morphism such that f ◦ s = idT . Let D be an effective Q-Cartier Q-divisor on X, relative over T (that is, we can write D locally as αH, for an effective divisor H and a positive 44 M. MUSTAT¸ A AND M. POPA rational number α, with H flat over T ). We assume that we have an effective divisor Z on X, relative over T , with Supp(Z) = Supp(D), and such that for every t ∈ T , the restriction Zt to the fiber Xt = f −1(t) is reduced. For every x ∈ X, we denote by mx the ideal defining x in Xf (x). Theorem 14.1. With the above notation, for every q ≥ 1, the set is open in T . Vq :=(cid:8)t ∈ T Ik(Dt) 6⊆ m q s(t)(cid:9), 15. Subadditivity theorem. The calculation for I2 in Example 10.5 shows that the inclusion Ik(D1 + D2) ⊆ Ik(D1) cannot hold for arbitrary Q-divisors D1 and D2. However, with an appropriate as- sumption on the support, we have the following stronger subadditivity statement: Theorem 15.1. If D1 and D2 are effective Q-divisors on the smooth variety X, whose supports Z1 and Z2 satisfy the property that Z1 + Z2 is reduced, then for every k ≥ 0 we have Ii(D1) · Ij(D2) · OX(−jZ1 − iZ2) ⊆ Ik(D1) · Ik(D2). Ik(D1 + D2) ⊆ Xi+j=k Note first that, for every i and j, the inclusion FiM(h−α) ⊆ Fi+jM(h−α) implies the inclusion (15.1) OX (−jZ) · Ii(D) ⊆ Ii+j(D). This gives the second inclusion in the statement above. To prove the first inclusion, as in the proof of [MP18a, Theorem B] it is enough to show the following:7 Proposition 15.2. Let X1 and X2 be smooth varieties and let Di be effective Q- divisors on Xi, with support Zi, for i = 1, 2. If Bi = p∗ i Di, where pi : X1 × X2 → Xi are the canonical projections, then for every k ≥ 0 we have Ik(B1 + B2) = Xi+j=k(cid:0)Ii(D1)OX1(−jZ1) · OX1×X2(cid:1) ·(cid:0)Ij(D2)OX2(−iZ2) · OX1×X2(cid:1). Proof. By Remark 2.2, we can assume that there exist regular functions h1 on X1 and h2 on X2, together with α ∈ Q>0, such that Ii(D1) and Ij(D2) are defined by Mr(h−α 2 ), respectively. The statement follows precisely as in [MP18a, Proposition 4.1], as long as we show that there is a canonical isomorphism of filtered D-modules 1 ) and Mr(h−α (cid:0)Mr((p∗ 1h1 · p∗ 2h2)−α), F(cid:1) ≃(cid:0)Mr(h−α 1 ) ⊠ Mr(h−α 2 ), F(cid:1), 7Indeed, the Restriction Theorem applies in the form given in Remark 13.4 for the diagonal embedding X ֒→ X × X, since we are assuming that Z1 + Z2 is reduced. HODGE IDEALS FOR Q-DIVISORS: BIRATIONAL APPROACH 45 where the filtration on the right hand side is the exterior product of the filtrations on the two factors. But this is a consequence of the canonical isomorphism of mixed Hodge modules j∗p∗QH V2[n2], V1×V2[n1 + n2] ≃ j1∗p1∗QH V1[n1] ⊠ j2∗p2∗QH with the obvious notation as in (2.3) for i = 1, 2, together with Lemma 2.8. (cid:3) References [Bjo93] J.-E. Bjork, Analytic D -modules and applications, Mathematics and its Applications, Kluwer Academic Publishers, 1993. [Dim04] A. Dimca, Sheaves in topology, Universitext, Springer-Verlag, Berlin, 2004. [DMST06] A. Dimca, P. Maisonobe, M. Saito, and T. Torrelli, Multiplier ideals, V -filtrations and transversal sections, Math. Ann. 336 (2006), no. 4, 901 -- 924. [Dut18] Y. Dutta, Vanishing for Hodge ideals on toric varieties, in preparation (2018). [EV92] H. Esnault and E. Viehweg, Lectures on vanishing theorems, DMV Seminar, vol. 20, Birkhauser, 1992. [Ful93] W. Fulton, Introduction to toric varieties, Annals of Mathematics Studies, vol. 131, Princeton University Press, Princeton, NJ, 1993. The William H. Roever Lectures in Geometry. [GKKP11] D. Greb, S. Kebekus, S. J. Kov´acs, and T. Peternell, Differential forms on log canonical spaces, Publ. Math. Inst. Hautes ´Etudes Sci. 114 (2011), 87 -- 169. [HTT08] R. Hotta, K. Takeuchi, and T. Tanisaki, D-modules, perverse sheaves, and representation theory, Birkhauser, Boston, 2008. [Laz04] R. Lazarsfeld, Positivity in algebraic geometry II, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 49, Springer-Verlag, Berlin, 2004. [MP16] M. Mustat¸a and M. Popa, Hodge ideals, preprint arXiv:1605.08088, to appear in Memoirs of the AMS (2016). [MP18a] M. Mustat¸a and M. Popa, Restriction, subadditivity, and semicontinuity theorems for Hodge ideals, Int. Math. Res. Not. 11 (2018), 3587 -- 3605. [MP18b] M. Mustat¸a and M. Popa, Hodge ideals for Q-divisors, V -filtration, and minimal exponent, preprint arXiv:1807.01935 (2018). [Pop18] M. Popa, D -modules in birational geometry, to appear in the Proceedings of the ICM, Rio de Janeiro (2018). [PS13] M. Popa and C. Schnell, Generic vanishing theory via mixed Hodge modules, Forum Math. Sigma 1 (2013), 60 pp. [Sai88] M. Saito, Modules de Hodge polarisables, Publ. Res. Inst. Math. Sci. 24 (1988), no. 6, 849 -- 995. [Sa90] M. Saito, Mixed Hodge modules, Publ. Res. Inst. Math. Sci. 26 (1990), no. 2, 221 -- 333. [Sai07] M. Saito, Direct image of logarithmic complexes and infinitesimal invariants of cycles, Algebraic cycles and motives. Vol. 2, London Math. Soc. Lecture Note Ser., vol. 344, Cambridge Univ. Press, Cambridge, 2007, pp. 304 -- 318. [Sai09] M. Saito, On the Hodge filtration of Hodge modules, Mosc. Math. J. 9 (2009), no. 1, 161 -- 191. [Sai16] M. Saito, Hodge ideals and microlocal V -filtration, preprint arXiv:1612.08667 (2016). [Sai17] M. Saito, A young person's guide to mixed Hodge modules, Hodge theory and L2-analysis, Adv. Lect. Math. (ALM), vol. 39, Int. Press, Somerville, MA, 2017, pp. 517 -- 553. [Vie82] E. Viehweg, Vanishing theorems, J. Reine Angew. Math. 335 (1982), 1 -- 8. [Zha18] M. Zhang, Hodge filtration for Q-divisors with quasi-homogeneous isolated singularities, in preparation (2018). 46 M. MUSTAT¸ A AND M. POPA Department of Mathematics, University of Michigan, Ann Arbor, MI 48109, USA E-mail address: [email protected] Department of Mathematics, Northwestern University, 2033 Sheridan Road, Evanston, IL 60208, USA E-mail address: [email protected]
1807.07888
1
1807
2018-07-20T15:18:37
De Rham epsilon factors for flat connections on higher local fields
[ "math.AG" ]
This note is a companion to the author's "Higher de Rham epsilon factors". Using Grayson's binary complexes and the formalism of $n$-Tate spaces we develop a formalism of graded epsilon lines, associated to flat connections on a higher local field of characteristic $0$. The definition is based on comparing a Higgs complex with a de Rham complex on the same underlying vector bundle.
math.AG
math
De Rham epsilon factors for flat connections on higher local fields Michael Groechenig Abstract This note is a companion to the author's Higher de Rham epsilon factors. Using Grayson's binary complexes and the formalism of n-Tate spaces we develop a formalism of graded epsilon lines, associated to flat connections on a higher local field of characteristic 0. The definition is based on comparing a Higgs complex with a de Rham complex on the same underlying vector bundle. Contents 1 Linear differential equations on higher local fields 1.1 Definitions and basic properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Cyclic vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Formal de Rham cohomology is finite-dimensional . . . . . . . . . . . . . . . . . . . . . . 2 De Rham epsilon-factors: definition and basic properties 2.1 Recapitulation on Tate objects: categorical preliminaries . . . . . . . . . . . . . . . . . . . 2.2 A reformulation of BBE's theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Higher local fields and closed 1-forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The variation of epsilon-factors in families 3.1 Epsilon-factors for epsilon-nice families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 The epsilon-crystal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 2 2 4 4 4 6 7 7 8 8 9 1 Linear differential equations on higher local fields This section is devoted to a study of flat connections (E,∇) on the higher local field Fn = k((t1))··· ((tn)), where k denotes as always a field of characteristic 0. A lot of the material will be familiar to the expert. We include the often well-known proofs to demonstrate that the behaviour of differential equations on Fn is not any different from the theory of D-modules on algebraic varieties. We do not lay claim to originality in this subsection. 1.1 Definitions and basic properties Definition 1.1. Let (E,∇) be a flat connection on Fn. The corresponding de Rham complex is the chain X ]. For i ∈ Z we denote the i-th complex of Fn-vector spaces Ω• cohomology group of this complex by H i∇(E). / ··· ∇ / / E ⊗ Ωn ∇ = [E ∇ / ∇ / / E ⊗ Ω1 X This project has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sk(cid:32)lodowska-Curie Grant Agreement No. 701679. 1 1.2 Cyclic vectors The proof of the following lemma follows the lecture notes on algebraic D-modules by Braverman– Chmutova [BC, Lecture 3]. We denote by D the ring of formal differential operators on F1 = k((t)), which we define to be the free k-algebra F1(cid:104)∂(cid:105) modulo the relation [∂, t] = 1. It is clear that the centre is given by Z(D) = k (relying heavily on the assumption that k has characteristic 0). Lemma 1.2 (Cyclic Vector Lemma). Let (E,∇) be a flat connection on F1 = k((t)). Then, as a D- module, there exists a cyclic vector s ∈ E, that is, a generator Ds = E. Proof. The proof begins by observing that D does not have any proper two-sided ideals, thus is a simple k-algebra. Indeed, let I ⊂ D be a two-sided ideal, distinct from the trivial ideal D. It is clear that the intersection F1 ∩ I must be {0}. We will use the tautological filtration on D (also known as the order of differential operators) to establish that I must be the zero ideal. If there is a non-zero element in I, then there is a minimal integer r ≥ 1, such that I contains an element p of order ≤ r. Since p cannot be central in D (as Z(D) = k) we obtain the existence of either a differential operator q of order ≤ 1, such that [p, q] (cid:54)= 0. The definition of two-sided ideals shows that [p, q] ∈ I. Hence it is a non-zero element of I of order ≤ r − 1 which contradicts the minimality assumption on r. Since E is a finite-dimensional F1-vector space, it has finite length as a F1-module and therefore also as a D-module. We assume by induction that the existence of a cyclic vector has already been established for D-modules of smaller length than the one of E (the case of length 1 being automatically true). There exists a short exact sequence of D-modules 0 / F / E φ / / M / 0, such that F has length 1. Let m ∈ M be a cyclic vector for M . We assume by contradiction that φ−1(m) does not contain a cyclic vector for E. Then we have Ds∩ F = {0}. Let p ∈ D be a differential operator, such that pm = 0, and f ∈ F an arbitrary element of F . We have φ(s + f ) = m, and therefore also D(s + f ) ∩ F = {0}. Since p(s + f ) = pf ∈ D(s + f ) ∩ F , we must have that pf = 0. Therefore, the ideal Ann(m) contains the two-sided ideal given by Ann(F ). This leaves us with two options to consider, Ann(F ) equal to {0} or D. The first case is not possible as then F would be isomorphic to D and hence dimF1 F = ∞ which is impossible. We conclude that Ann(m) = D which implies M = 0 and therefore simplicity of E. Corollary 1.3. Let ∇ be a flat connection on the trivial rank m bundle F . There exist elements a0, . . . , am−1 ∈ F1, such that the k-vector space of ∇-flat sections is isomorphic to the k-vector space of solutions to the ordinary differential equation y(m) + am−1y(m−1) + ··· + a0y = 0. ⊕m 1 Proof. Let s be a cyclic vector. The F1-dimension of E = F of the (m + 1)-tuple s, ∂s,··· ∂ms. Therefore we can write ∂ms = −(cid:80)m−1 ai ∈ F1. We conclude that the D-module E is equivalent to the quotient D/(∂m +(cid:80)m−1 is m, which implies linear dependence i=0 ai∂is, for certain elements ⊕m 1 i=0 ai∂i). Furthermore the cyclic vector lemma for F1 implies the existence of cyclic vectors for Fn. Corollary 1.4. Let (E,∇) be a flat connection on Fn. Then E is cyclic as a module over Dn = DFn . Proof. The ring Dn contains a subring D(cid:48) ⊂ Dn which is generated by Fn = Fn−1((tn)) and ∂n = ∂ . The ring D(cid:48) is canonically isomorphic to DFn−1((tn))/Fn−1 . As a D(cid:48)-module, E is cyclic by virtue of Lemma 1.2. This implies that E is cyclic as a D-module. ∂tn 1.3 Formal de Rham cohomology is finite-dimensional Lemma 1.5. An m-tuple of formal Laurent series y1, . . . ym ∈ k((t)) is k-linearly independent if and only if the Wronskian W (y1, . . . , ym) = (y(j−1) )1≤i,j≤m is non-zero. i Proof. It is clear that linear dependence of the m-tuple y1, . . . , ym implies vanishing of the Wronskian. We prove the converse by induction on m, anchored to the case m = 1 which is a tautology. Let us assume that the statement has been proven for (m − 1)-tuples and that y1 (cid:54)= 0. This allows us to divide ) and does not affect every element of the m-tuple by y1 (which changes the Wronskian by a factor of y k-linear independence. Henceforth we assume without loss of generality y1 = 1 and compute −m 1 0 = W (1, y2, . . . , ym) = W (y (cid:48) 2, . . . , y (cid:48) m). 2 / / / scalars bearing witness to this fact, i.e.,(cid:80)m of a scalar λ1 ∈ k, such that λ1 · 1 +(cid:80)m m is k-linear independent. Let λ2, . . . , λm ∈ k be By assumption this shows that the (m−1)-tuple y(cid:48) 2, . . . , y(cid:48) i=2 λiy(cid:48) i = 0. Integrating this equation we obtain the existence i=2 λiyi = 0. Since we assumed y1 = 1 this establishes a linear dependence and hence concludes the proof. Corollary 1.6. Let a0, . . . , am−1 ∈ k((t)). The k-vector space of solutions to the ordinary differential equation y(m) + am−1y(m−1) + ··· + a0y = 0 is of dimension at most m. Proof. Let y1, . . . , ym+1 be solutions to the ordinary differential equation above. We claim that they are k-linearly dependent. As we have seen above this is equivalent to vanishing of the Wronkian W (y1, . . . , ym+1) = (y(j−1) + ··· + a0yi = 0 (which holds for all i by assumption) describes a F1-linear relation between the rows of the matrix (y(j−1) Proposition 1.7. Let (E,∇) be a flat connection on F1 = k((t)). The k-vector spaces ker(∇) and coker(∇) are finite-dimensional. )1≤i,j≤m+1. The F1-linear relation y(m) )1≤i,j≤m+1. This implies vanishing of the Wronskian. i + am−1y(m−1) i i i Proof. We begin by showing finite-dimensionality of the kernel. As we have seen in Subsection 1.2, there exists an m-th order linear differential equation y(m) + am−1y(m−1) + ··· + a0y = 0 whose solutions form a k-vector space isomorphic to ker(∇). Corollary 1.6 shows that this k-vector space is finite-dimensional. Finite-dimensionality of the cokernel is shown by dualising. Since the k-vector spaces E and E ⊗ Ω1 F1 are infinite-dimensional, it helps to endow them with the t-adic topology. This is the topology induced by the natural valuation on F1 = k((t)) (which is discrete on k). With respect to this topoology, E is a so-called linearly locally compact k-vector space in the sense of Lefschetz (also known as Tate k- vector spaces). The residue pairing yields an isomorphism of the topological dual E(cid:48) with the Tate k-vector space E∨ ⊗ Ω1 F1 (where E∨ refers to the K1-linear dual). With respect to this isomorphism, the dual of the connection ∇ can be seen to be the map induced by the dual connection. We conclude finite-dimensionality of the cokernel of ∇ from finite-dimensionality of the kernel of the dual connection ∇∨. Corollary 1.8. Let (E,∇) be a flat connection on Fn. For all degrees i ∈ Z the de Rham cohomology groups H i∇(E) are finite-dimensional k-vector spaces. Before giving the proof it seems appropriate to summarise the underlying ideas. We argue by induction on n, the case n = 0 being a tautology. Let us assume that the assertion has been verified for flat connections on Fn−1. There is a canonical inclusion of fields Fn−1 (cid:44)→ Fn, whose geometric counterpart is projection along the tn-coordinate. The de Rham cohomology groups H i∇(E) can be computed from a variant of the Leray spectral sequence relative to this map. Since Fn = Fn−1((tn)) we can apply the finiteness result of Proposition 1.7 for the n = 1 case and the induction hypothesis to conclude the result. Proof. Let (E,∇) be a flat connection on Fn = Fn−1((tn)). We denote by Ω1 subspace of Ω1 Fn , spanned by the elements dtn. By definition it is orthogonal to the subspace the Fn-linear Fn/Fn−1 (Ω1 Fn−1 )Fn (cid:44)→ Ω1 Fn , generated by the elements dt1, . . . , dtn−1. By taking exterior powers and tensor products we obtain Fn-linear subspaces (Ωp and (Ωp )Fn ⊂ Ωp )Fn ⊗Fn Ω1 / E ⊗ (Ω1 Fn/Fn−1 ⊂ Ωp+1 . Fn−1 )Fn , ∇(cid:48)(cid:48) : E Fn / E ⊗ Ω1 Fn/Fn−1 . More Fn−1 There are natural differential operators ∇(cid:48) : E Fn−1 Fn generally we may form the double complex E ∇(cid:48) E ⊗ (Ω1 Fn−1 )Fn ∇(cid:48) ··· ∇(cid:48) / E ⊗ (Ωn−1 Fn−1 )Fn ∇(cid:48)(cid:48) E ⊗ Ω1 ∇(cid:48)(cid:48) ∇(cid:48) / E ⊗ (Ω1 Fn−1 )Fn ⊗ Ω1 ∇(cid:48) / ··· ∇(cid:48) / E ⊗ (Ωn−1 Fn−1 Fn/Fn−1 Fn/Fn−1 Note that ∇ = ∇(cid:48) + ∇(cid:48)(cid:48) and ∇(cid:48)∇(cid:48)(cid:48) = −∇(cid:48)(cid:48)∇(cid:48). We conclude that the de Rham complex Ω• isomorphic to the totalisation of this double complex. The associated spectral sequence is Fn/Fn−1 ∇(E) is ∇(cid:48)(cid:48) )Fn ⊗ Ω1 . 2 = H p∇(cid:48) (H q∇(cid:48)(cid:48) (E)) ⇒ H p+q∇ (E). Ep,q 3 / / / /   / /   /   / / / By Proposition 1.7, H q∇(cid:48)(cid:48) (E) is finite-dimensional, since Fn = Fn−1((t)). The induction hypothesis implies finite-dimensionality of H p∇(cid:48) (H q∇(cid:48)(cid:48) (E)). Therefore, every object appearing on the second page of our spectral sequence is finite-dimensional over k. Convergence of the spectral sequence (see [Sta, Tag 0132]) implies that the formal de Rham cohomology groups H i∇(E) are of finite dimension over k. 2 De Rham epsilon-factors: definition and basic properties Given a flat connection (E,∇) on the higher local field Fn = k((t1))··· ((tn)) we define a formalism of epsilon-factors. Our theory depends on an Fn-linearly independent n-tuple of 1-forms ν = (ν1, . . . , νn) ∈ Fn and produces a graded (or super) line εν (E,∇). This graded line is naturally obtained from a Ω1 homotopy point of the K-theory spectrum K(k). Our treatment begins with a recapitulation of the case n = 1, which is due to Beilinson–Bloch– Esnault [BBE02]. We reformulate their definition in terms of Grayson's binary complexes [Gra12]. This perspective allows us a glimpse at the higher-dimensional generalisation. We then define epsilon-lines in arbitrary dimension n, by using Grayson's iterated binary complexes, finite-dimensionality of formal de Rham cohomology (Corollary 1.8), and crucially, almost-commutativity of the ring of differential operators. 2.1 Recapitulation on Tate objects: categorical preliminaries The notion of Tate objects in exact categories goes back to Beilinson's [Bei87] and Kato's [Kat00], and is inspired by Lefschetz's theory of linearly locally compact vector spaces. We refer the reader to [Pre11] and [BGW14] for a detailed introduction to the theory and an overview of the history of Tate objects. To an idempotent complete exact category C (see [Buh10] for an account of the general theory of exact categories) one defines an exact category of admissible Ind objects Inda C. It is the full subcategory of the category of Ind objects Ind C, whose objects can be represented as a formal colimit colimK Xk, where for F1 ≤ k2 ∈ K the resulting morphism XF1 / Xk2 is an admissible monomorphism in the sense of exact categories. The dual construction yields Proa C, an exact category of admissible Pro objects. The exact category Inda Proa C contains an extension closed full subcategory Tateel(C) whose objects are referred to as elementary Tate objects. By definition, every elementary Tate object V belongs to an admissible short exact sequence L (cid:44)→ V (cid:16) D, where L ∈ Proa C and D ∈ Inda C. One also says that L ⊂ V is a lattice. The exact category of Tate objects Tate(C) is defined to be the idempotent closure of Tateel(C). The categorical quotient Inda(C)/C was studied by Schlichting in [Sch04] as the suspension category SC. We denote its idempotent closure by Calk(C) and refer to it as the category of Calkin objects. There / Calk(C) given by sending V to V /L where L is a lattice in V . In fact, is a natural functor Tate(C) one has Tateel(C)/ Proa(C) (cid:39) Inda(C)/C. Remark 2.1. The most important input category C in this paper is Pf (R), that is, the exact category of finitely generated projective R-modules. Here we denote by R a commutative ring (with unit). In this case we will simply write Tate(R), in reference to Tate(Pf (R)). 2.2 A reformulation of BBE's theory For the purpose of this subsection we let (E,∇) denote a flat connection on F = F1 = k((t)). Recall that E is an F -vector space and that ∇ : E F is a k-linear map, satisfying the Leibniz identity. Since F has a natural structure of a Tate vector space over k, so does E, and ∇ can be easily seen to be a morphism in this category. Furthermore this almost defines an isomorphism of linearly locally compact vector spaces (in a technical sense). Lemma 2.2. A connection ∇ as above induces an isomorphism in the localised category Tateel(k)/ Proa(k). / E ⊗ Ω1 Proof. Consider the diagram of abstract k-vector spaces E ∇ / E ⊗ Ω1 F (1) K E/K C 4 / / / ! ! ! ! / " " " " /  @ @ , : : where the diagonals are short exact sequences. The kernel K and cokernel C are finite-dimensional by virtue of Proposition 1.7. Furthermore, the k-vector spaces E and E ⊗ Ω1 F are endowed with a linearly locally compact topology (that is, they are Tate vector spaces). We claim that the diagram (1) is a diagram in category of Tate vector spaces. This amounts to showing that all arrows represent continuous maps of topological vector spaces. Continuity of ∇ has already been asserted and can be checked using an explicit presentation in terms of coordinates. It remains to verify that the diagonal morphisms are continuous. The k-linear map i : K (cid:44)→ E is certainly continuous when endowing K with the discrete topology. Since K is finite-dimensional, this is the only possibility within the category of linearly locally compact k-vector spaces. Furthermore we assert that i has a continuous retraction r. To construct r one chooses a lattice L ⊂ E, such that i(K) ⊂ L. By definition of the category of linearly locally compact vector / L. Moreover we know that the inclusion K (cid:44)→ L has a retraction spaces there exists a retraction r(cid:48) : E r(cid:48)(cid:48), since the category of linearly compact k-vector spaces is contravariantly equivalent to the category of discrete k-vector spaces (which is a split abelian category). We define r = r(cid:48)(cid:48) ◦ r(cid:48). This shows that E/K is a direct summand of E and hence establishes continuity of the morphism E / E/K. Using the observation of the proof of Proposition 1.7 that the k-linear topological dual of the map F is equivalent to the dual connection ∇∨, we obtain continuity of the maps on the right / E ⊗ Ω1 E hand side of diagram (1). In the category Tateel(k)/ Proa(k) we have that morphisms with finite-dimensional kernel or cokernel F are isomorphisms in this are isomorphisms. In particular we see that E localisation. Commutativity of the triangle in (1) implies the assertion. / E/K and E/K / E ⊗ Ω1 In the article [Gra12], Grayson approaches higher algebraic K-groups using the notion of (iterated) binary complexes. A binary complex in an additive category C is a complex with two differentials, respectively a pair of complexes sharing the same objects. More precisely it is defined to be a collection / X j+1 for all j ∈ Z, such that for i = 1, 2 and all of objects (X j)j∈Z and morphisms d1,j, d2,j : X j j ∈ Z we have di,j+1 ◦ di,j = 0. Definition 2.3. Let C denote an additive category. We denote by BinCh(C) the category whose objects are binary complexes ((X j; d1,j, d2,j)j∈Z), and morphisms ((X j; d1,j, d2,j)j∈Z) / ((Y j; d1,j, d2,j)j∈Z) are given by tuples (f j)j∈Z, such that di,j ◦ f j = f j+1 ◦ di,j for i = 1, 2 and j ∈ Z. Definition 2.4. Let (E,∇) be a flat connection on F = k((t)) and ν ∈ Ω1 the length 2 binary complex, given by F . We define Ω• ∇,ν (E) to be [ E ∇ / ν / E ⊗ Ω1 X ] ∇,ν : Loc(F ) in the exact category Calk(k). For fixed ν this defines an exact functor Ω• / BinCh (Tate(k)). One says that a binary complex (X•; d1, d2) in an exact category C is exact (or synonymously acyclic), if both differentials define an exact complex in C in the sense of [Buh10, Definition 8.8]. We denote the corre- sponding extension-closed subcategory by BinChex(C). Grayson shows in [Gra12] that there exists a canon- / ΩK(C). In the article he assumes that C admits a ical morphism of connective spectra K(BinChex(C)) calculus of long exact sequences, but this assumption is not needed to define the aforementioned morphism /ΩK(C) but only in proving the main result [Gra12, Theorem 4.3] that K(Ch≥0 is a fibre sequence of connected spectra. Lemma 2.5. Assume that ν is a non-zero 1-form on F . Then for every flat connection (E,∇), the binary complex Ω• ∇,ν (E) is exact in the Calkin category. In particular we have an exact functor /K(BinChex(C)) ex(C)) Consequently we have for ν ∈ Ω1 Ω∇,ν : Loc(F ) F1 \ {0} a well-defined map of K-theory spectra / BinChex(Calk(k)). which we call the spectral ε-factor. εν : K(LocF1 ) / K(k), 5 / / / / / / / / / / / / / / / 2.3 Higher local fields and closed 1-forms Let us denote by LocFn the exact category of pairs (E,∇), where E is a finite rank Fn-vector space, and ∇ a formal flat connection on E. In this subsection we will define for a Fn-linearly independent tuple of closed 1-forms ν = (ν1, . . . , νn) on Fn a map of spectra εν : K(LocFn ) / K(k) which generalises the epsilon-factor formalism for the 1-local field k((t)). One could argue that the assumption that the forms ν1, . . . , νn are closed is a feature of the higher-dimensional case, since it is automatically true for n = 1. However we believe that it should not be needed in order to have a well-defined εν (see the companion article [Gro]). Definition 2.6. Let (E,∇) be a de Rham local system on Fn and ν ∈ (Ω1 Fn )n denote an Fn-linearly independent tuple of closed 1-forms. We define a multi-complex (whose totalisation is the formal de Rham complex) which is supported on the cube {0, 1}n which we identify with the power set P({1, . . . , n}). For M ⊂ {1, . . . , n} we use the notation where M = {i1 < ··· < iq}. Furthermore, for every inclusion j : M (cid:44)→ N = M ∪ {i}, we let ΩM,ν∇ (E) = E ⊗ Fn(cid:104)νi1 ∧ ··· ∧ νiq(cid:105) ⊂ E ⊗ Ωq , Fn ∇j : ΩM,ν∇ (E) / ΩN,ν∇ (E) be the component of the connection ∇ : E⊗Ωq and ΩN,ν∇ (E). Fn /E⊗Ωq+1 Fn with respect to the direct summands ΩM,ν∇ (E) It is important to emphasise that without the assumption that ν is an n-tuple of closed forms, this Indeed anti-commutativity of the resulting squares is definition would not produce a multicomplex. guaranteed by this assumption. The proof of the following lemma is based on an observation on n-Tate objects. There is a natural embedding e : C (cid:44)→ Tate(C). This yields n distinct ways to embed Taten−1(C) into Taten(C) which we denote by e1, . . . , en. Lemma 2.7. The multicomplex of Definition 2.6 is acyclic in the category Calkn(k). Proof. A direct computation involving writing out power series in the variables t1, . . . , tn allows one to infer the following claim. Claim 2.8. Let V =(cid:80)i j=1 ai belongs to the full subcategory ∂ ∂tj , where aj ∈ Fn, and ai (cid:54)= 0. Then the kernel and cokernel of the map ∇V : E / E ei(Taten−1) (cid:44)→ Taten . Applying this lemma to the vector fields given by the dual basis of ν we deduce a similar statement for the differentials of the multicomplex refining the de Rham complex of E. In particular, since ei factors through the kernel of Taten Definition 2.9. For (E,∇) ∈ LocFn and ν ∈ (Ω1 Fn )n a linearly independent n-tuple of closed 1-forms, we define a binary multicomplex supported on {0, 1}n in the sense of Grayson. It is constructed by adding extra differentials to the multicomplex Ω / Calkn, we deduce the lemma. ∇(E). For every inclusion j : M (cid:44)→ N = M ∪ {i}, we let (cid:4) νi : ΩM,ν∇ (E) be the component of the morphism ∧νi : E ⊗ Ωq ΩM,ν∇ (E) and ΩN,ν∇ (E). This binary multi-complex will be denoted by E(cid:4) Definition 2.10. We have an exact functor Eν : LocFn binary multicomplex E(cid:4) Definition 2.11. We define the map of spectra εν : K(Locn) ∇,ν (E). Fn Fn / ΩN,ν∇ (E) / E ⊗ Ωq+1 with respect to the direct summands ∇,ν (E). / (BinChex)n(Calkn(k)), sending (E,∇) to the / K(k) as the composition K(Eν ) K(Locn) / K((BinChex)n(Calkn(k))) / K(k), where the second map combines Grayson's K((BinChex)n(Calkn(k))) alence K(Taten(k)) (cid:39) K(k) (see [Sai15]). /ΩnK(Calkn(k)) with Saito's equiv- 6 / / / / / / / / / / / / n/Fn be a finite extension of n-fields with last residue fields k(cid:48)/k, and ν ∈ (Ω1 2.4 Induction Let F (cid:48) independent n-tuple. Observe that we have Ω1 to a de Rham local system on Fn. The resulting exact functor is denoted by Ind : LocF (cid:48) will be referred to as induction with respect to F (cid:48) Proposition 2.12. There is a commutative diagram of spectra Fn ⊂ Ωn F (cid:48) . A de Rham local system (E,∇) on F (cid:48) n/Fn. n n Fn )n be an Fn-linearly n gives rise / LocFn and K(LocF (cid:48) n ) Ind K(LocFn ) εν K(k(cid:48)) εν / K(k). (2) Proof. Note that every finite F (cid:48) n-vector space E gives rise to an n-Tate k(cid:48)-vector space. However, since k(cid:48)/k is a finite field extension, we can also view E as an n-Tate k-vector space. Furthermore, the diagram of exact categories Pf (F (cid:48) n) Pf (Fn) commutes strictly. We obtain the commutative diagram above by applying the K-theory functor K(−) to the following strictly commutative diagram of exact categories Taten(k(cid:48)) / Taten(k) LocF (cid:48) n Ind Eν ex Calkn(k(cid:48)) BinChn LocFn Eν / BinChn ex Calkn(k) and using the natural map K(BinChn ex Calkn(k)) / K(k). 2.5 Duality In this subsection we will prove the proposition below. We refer the reader to [BGHW18] for the definition of duality for higher Tate objects. Proposition 2.13. There is a commutative diagram of spectra (−)∨ (−)∨ K(LocFn ) εν K(k) K(LocFn ) ε−ν / K(k). (3) Let E be a finite-dimensional Fn-vector space. We denote by E∨ its Fn-linear dual. Since E can be seen as an n-Tate k-vector space, it is also possible to consider the k-linear dual in the sense of n-Tate spaces, which we denote by E(cid:48). The following lemma relates the Fn-linear and n-Tate dual in a canonical fashion. It is the analogue of Serre duality for n-fields. Its proof is a straight-forward extension of the well-known case where n = 1. Fn which is induced by the higher residue pairing Res(−,−) : E×(E∨⊗Ωn Lemma 2.14. For every finite-dimensional Fn-vector space E there is a natural isomorphism of n-Tate k- vector spaces E (cid:39) E∨⊗Ωn /k. is For p > 0 we define Ω Fn . With respect to canonically isomorphic to Ω this notation we obtain the following isomorphisms from Lemma 2.14 as an immediate consequence. )(cid:48) (cid:39) E∨ ⊗ Ωn−p Corollary 2.15. We have natural isomorphisms of n-Tate vector spaces (E ⊗ Ωp = Ωn−p , by virtue of the twisted pairing ∧ : Ωp It is easy to see that the Fn-linear dual of Ωp Fn ⊗ Ωn−p ⊗ (Ωn Fn )∨. −p Fn −p Fn Fn ) / Ωn Fn Fn Fn Fn . Fn With this isomorphism at hand we ask ourselves what the n-Tate dual of the map E⊗Ωp /E⊗Ωp+1 is. The answer to this question is given by the next corollary. We denote by ∇∨ the dual connection on E∨. Fn Fn ∇ / 7 / / /     / / /     / / /     / / / /     / / / Corollary 2.16. With respect to the isomorphism of Corollary 2.15 we have ∇(cid:48) = −∇∨ : E ∨ ⊗ Ωn−p−1 Fn / E ∨ ⊗ Ωn−p Fn . Proof. For s ∈ E and t ∈ E∨ we have the relation (∇s, t) + (s,∇∨t) = d(s, t). Applying the higher residue to both sides, we obtain Res(∇s, t) + Res(s,∇∨t) = 0, since the right hand side is exact. This shows ∇(cid:48)t = −∇∨t for all t ∈ E∨. For a complex A• in an exact category C we denote by ι∗A• the complex obtained by replacing / Ai+1 by −di. We can construct an isomorphism A• (cid:39) ι∗A• in terms of a all differentials di : Ai commutative ladder. ··· di / Ai di di+1 Ai+1 / ··· (−1)i (−1)i+1 ··· −di / Ai −di / Ai+1 −di+1 / / ··· However there is a second choice of an isomorphism, which replaces all vertical arrows (−1)i by (−1)i+1. So we see that there is a µ2-torsor of natural isomorphisms A• (cid:39) ι∗A•. The same remark applies to binary complexes. As a consequence one sees that the proof of the proposition below produces not just one commutative square as in (3), but two (we remind the reader that in the context of ∞-categories, commutativity of diagrams is an extra structure and not a property). Proof of Proposition 2.13. Corollary 2.16 implies that the n-Tate dual of the binary multicomplex E(cid:4) is given by ι∗E(cid:4) of exact categories ∇,ν (E) ∇,−ν (E∨). Using one the natural isomorphisms above we obtain a of commutative diagram (−)∨ LocFn Eν LocFn E−ν Using the natural map of spectra K(BinChn ex Calkn(k)) / K(k) we conclude the proof of the proposition. BinChn ex Calkn(k) (−)∨ / BinChn ex Calkn(k). 3 The variation of epsilon-factors in families 3.1 Epsilon-factors for epsilon-nice families As before we let k be a field of characteristic 0. We denote by R a commutative k-algebra, and use F R n as shorthand for the commutative ring R((t1))··· ((tn)). As an R-module it carries the structure of an n-Tate R-module, and hence every finitely generated projective F R n -module admits a natural realisation in the exact category Taten(R). Given E ∈ Pf (F R ΩF R n /R, we investigate when it is possible to define an epsilon-factor εν (E,∇) in the K-theory spectrum K(R). Definition 3.1. Let (E,∇) ∈ LocR(Fn) be a de Rham local system, such that the formal de Rham (cid:4),R∇ (E) is an acyclic multi-complex in Calkn(R). We say that (E,∇) is a epsilon-nice (or multi-complex Ω blissful) R-family of flat connections. The corresponding full subcategory of LocR(Fn) will be denoted by LocR bliss(Fn). Since formation of the de Rham complex ΩR∇(E) is an exact functor, and acyclic complexes form a bliss(Fn) is an exact and idempotent n ), and a relative connection ∇ : E fully exact and idempotent complete subcategory, we see that LocR complete category. / E ⊗F R n Remark 3.2. (a) In [BBE02] such R-families of flat connections are called nice. (b) Intuitively speaking the irregularity type of the connection stays constant in epsilon-nice R-families. However, just as in loc. cit. we prefer to define epsilon-nice families in the abstract manner above. For epsilon-nice families one can define a (spectral) de Rham epsilon factor, for every R-linearly tuple of closed relative 1-forms (ν1, . . . , νn) ∈ Ω1 Definition 3.3. We denote by E(cid:4) K-theory functor we obtain a morphism of spectra εR ν the exact functor LocR Fn(cid:98)⊗kR. bliss(Fn) ν : K(LocR bliss(Fn)) / BinChn ex Calkn(R). Applying the / K(R). 8 / / / / /     / / / / /     / / / / / 3.2 The epsilon-crystal Epsilon-factors in families are naturally endowed with a partial connection (meaning that we cannot covariantly derive along all vector fields, but only along a fields belonging to a subbundle). Since we are working with homotopy points of K-theory spectra this is best formalised using a crystalline approach. Before stating the main result of this subsection we have to introduce some notation. Definition 3.4. (a) For a scheme S we denote by SdR the presheaf on the category of schemes given by T (cid:55)→ S(T red). We have a natural transformation S / Sred. (b) A functor H : Sch / Sp extends to a functor H : PrSh(SchN is) / Sp, such that for a sheaf F , we have H(F ) (cid:39) limT /F H(T ). In particular we get a well-defined spectrum H(SdR) for every scheme S. We call it the spectrum of F -crystals over S. (c) We refer to the spectrum K(SdR) as the spectrum of K-crystals of S. Homotopy points thereof will be referred to as K-crystals defined over S. / B Using the natural map of Nisnevich sheaves K Z Gm we see that every K-crystal on S gives rise to a (graded) line with a flat connection on S. We will now show that the formalism of ε-factors has a natural crystalline refinement. This is the higher rank generalisation of the epsilon connection defined in in section 3 of [BBE02] (furthermore they did not consider K-crystals). The proof is omitted, as it is based on the (P1,∞)-invariance property of algebraic K-theory detailed in the companion paper [Gro] to construct the epsilon connection. Proposition 3.5. Let (E,∇) be an object of Loc(Fn), S an affine k-scheme and ν = (ν1, . . . , νn) a basis of Ω1 / K(SdR), which renders the diagram n /S. There exists a map crεν : K(Loc(Fn)) F S K(Loc(Fn)) crεν / K(SdR) εν K(S) commutative. Concluding remarks In [Gro] the author introduced a theory of de Rham epsilon factors for D-modules on higher-dimensional varieties which is supported on points and gives rise to an epsilon-crystal. The definition in loc. cit. is a continuation of Patel's theory of de Rham epsilon factors introduced in [Pat12]. We expect these two notions of de Rham epsilon factors to be equivalent. Acknowledgements The author thanks Oliver Braunling, H´el`ene Esnault, Javier Fres´an, Markus Roeser and Jesse Wolfson for many interesting conversations about epsilon factors, differential equations and Tate objects. References [BBE02] [BC] [Bei87] A. Beilinson, S. Bloch, and H. Esnault, -factors for Gauss-Manin determinants, Mosc. Math. J. 2 (2002), no. 3, 477–532, Dedicated to Yuri I. Manin on the occasion of his 65th birthday. MR 1988970 (2004m:14011) Alexander Braverman and Tatyana Chmutova, Lectures on algebraic D-modules, http:// www.math.harvard.edu/~gaitsgde/grad_2009/Dmod_brav.pdf. A. Beilinson, How to glue perverse sheaves, K-theory, arithmetic and geometry (Moscow, 1984–1986), Lecture Notes in Math., vol. 1289, Springer, Berlin, 1987, pp. 42–51. MR 923134 (89b:14028) [BGHW18] O. Braunling, M. Groechenig, A. Heleodoro, and J. Wolfson, On the normally ordered tensor product and duality for Tate objects, Theory and Applications of Categories 33 (2018), no. 13, 296–349. 9 / / / / / / & &   [BGW14] O. Braunling, M. Groechenig, and J. Wolfson, Tate Objects in Exact Categories (with appendix by Jan Stovicek and Jan Trlifaj), arXiv:1402.4969, 02 2014. [Buh10] [Gra12] Theo Buhler, Exact categories, Expo. Math. 28 (2010), no. 1, 1–69. MR 2606234 (2011e:18020) Daniel Grayson, Algebraic K-theory via binary complexes, Journal of the American Mathe- matical Society 25 (2012), no. 4, 1149–1167. [Gro] M. Groechenig, Higher de Rham epsilon factors, https://arxiv.org/abs/1807.03190. [Kat00] [Pat12] [Pre11] [Sai15] [Sch04] [Sta] Kazuya Kato, Existence theorem for higher local fields, Invitation to higher local fields (Munster, 1999), Geom. Topol. Monogr., vol. 3, Geom. Topol. Publ., Coventry, 2000, pp. 165– 195. MR 1804933 (2002e:11173) D. Patel, De Rham Epsilon-factors, Invent. Math. 190 (2012), no. 2, 299–355. MR 2981817 Luigi Previdi, Locally compact objects in exact categories, Internat. J. Math. 22 (2011), no. 12, 1787–1821. MR 2872533 Sho Saito, On Previdi's delooping conjecture for K-theory, Algebra Number Theory 9 (2015), no. 1, 1–11, arXiv:1203.0831. M. Schlichting, Delooping the K-theory of exact categories, Topology 43 (2004), no. 5, 1089– 1103. MR 2079996 (2005k:18023) Stacks Project Authors, Stacks project, http://math.columbia.edu/algebraic geometry/stacks- git. E-mail: [email protected] Address: FU Berlin, Arnimallee 3, 14195 Berlin 10
0804.1627
3
0804
2018-01-12T13:18:06
Counting conics in complete intersections
[ "math.AG" ]
We count the number of conics through two general points in complete intersections when this number is finite and give an application in terms of quasi-lines.
math.AG
math
COUNTING CONICS IN COMPLETE INTERSECTIONS LAURENT BONAVERO, ANDREAS H ORING Abstract. We count the number of conics through two general points in complete intersections when this number is finite and give an application in terms of quasi-lines. 1. Introduction Let X be a complex projective manifold of dimension n. A quasi-line l in X is a smooth rational curve f : P1 ֒→ X such that f ∗TX is the same as for a line in Pn, i.e. is isomorphic to OP1(2) ⊕ OP1(1)⊕n−1. Let X be a smooth projective variety containing a quasi-line l. Following Ionescu and Voica [IV03], we denote by e(X, l) the number of quasi-lines which are deformations of l and pass through two given general points of X. We denote by e0(X, l) the number of quasi-lines which are deformations of l and pass through a general point x of X with a given general tangent direction at x. Note that one always has e0(X, l) ≤ e(X, l), but in general the inequality may be strict [IN03, p.1066]. 1.1. Theorem. Let X ⊂ Pn+r be a general smooth n-dimensional complete intersec- tion of multi-degree (d1, . . . , dr). Assume moreover that d1 + · · · + dr = n + 1 2 + r. Then (1) the family of conics contained in X is a nonempty, smooth and irreducible component of the Chow scheme C(X), (2) a general conic C contained in X is a quasi-line of X and e0(X, C) = e(X, C) = 1 2 r (di − 1)!di!. Yi=1 The numerical assumption d1 + · · · + dr = (n + 1)/2 + r assures that if C is a conic in X, then −KX · C = n + 1. This numerical condition is of course necessary for a curve to be a quasi-line. Note that varieties appearing in our theorem are Fano varieties of dimension n and index (n+1)/2; they are well known to be the boundary Fano varieties with Picard number one being conic-connected (see [IR07], Theorem 2.2). Using a degeneration argument, one can strengthen parts of the statement. 1.2. Corollary. Let X ⊂ Pn+r be a smooth n-dimensional complete intersection of multi-degree (d1, . . . , dr). If d1 + · · · + dr = (n + 1)/2 + r, the variety X contains a conic that is a quasi-line. Date: September 22, 2009. Key-words : conics, quasi-lines, complete intersections. A.M.S. classification : 14N10, 14M10. 1 obtained the formula e(X, l) = 1 2 r Yi=1 (di − 1)!di! as a consequence of his computation 2 LAURENT BONAVERO, ANDREAS H ORING By a theorem of Ionescu [Ion05], we obtain an immediate application of the theorem to formal geometry. Before stating it, let us recall that a subvariety Y of a variety X is G3 in X if the ring K(XY ) of formal-rational functions of X along Y is equal to K(X). 1.3. Corollary. Let X ⊂ Pn+r be a general smooth n-dimensional complete intersec- tion of multi-degree (d1, . . . , dr) such that d1 + · · ·+ dr = (n + 1)/2+ r. Then any general conic C contained in X is G3 in X. In particular, if (X, C) and (X ′, C ′) are two such pairs such that the formal completions XC and X ′ C ′ are isomorphic as formal schemes, there exists an isomorphism from X to X ′ sending C to C ′. When this note was almost finished, we learned from L. Manivel that A. Beauville had of the quantum cohomology algebra H ∗(X, Q) of a complete intersection [Bea95]. We provide here a completely elementary proof. We end this note by mentioning a similar question where no elementary proof seems to be known. We want to thank Fr´ed´eric Han, Paltin Ionescu and Laurent Manivel for their interest in our work and helpful discussions. The first author express his warm gratitude to the organisers of the Hanoı conference, with a special mention to H´el`ene Esnault, for their kind invitation. We also thank the referee for numerous detailed remarks. 2. Proofs We start by explaining the enumerative argument in the simplest case. 2.1. A well known example. Suppose that X = {s = 0} is a smooth cubic threefold in P4. A general conic C in X is a quasi-line [BBI00, Thm.3.2]. The basic idea of our proof is that counting conics in X through p and q can be reduced to counting 2-planes π through p and q such that the restriction sπ is a product of a polynomial of degree two and some residual polynomial. We will explain how to do this in general below, in the case of the cubic threefold we can use a geometric construction. It is a classical fact that the lines in X form an irreducible smooth family of dimension two and that there are exactly six lines passing through a general point of X [AK77, Prop.1.7]. Fix now two general points p and q in X, then the line [pq] intersects X in a third point u. For every line l ⊂ X through u there exists a unique plane πl containing l and [pq]. The intersection X ∩ πl is the union of l and a residual conic C. Since l does not pass through p and q, the conic C passes through p and q. Vice versa the linear span of a conic C ⊂ X passing through p and q is a 2-plane πC containing the line [pq]. Since C does not pass through u, the residual line passes through u. Thus the conics through p and q are in bijection with the lines through u, so e(X, C) = 6. Suppose now that we are in the general situation of Theorem 1.1. We always assume that X ⊂ Pn+r is a general smooth n-dimensional complete intersection of multi-degree (d1, . . . , dr) with di ≥ 2 for all i and d1 + · · · + dr = n + 1 2 + r. Let l ⊂ X be a smooth rational curve contained in X. Then −KX · l = (n + r + 1 − (d1 + · · · + dr)) deg(l) = n + 1 2 deg(l) COUNTING CONICS IN COMPLETE INTERSECTIONS 3 therefore −KX · l = n + 1 if and only if l is a conic. 2.2. The main step. For any general points p and q of X, there exists a conic con- tained in X passing through p and q. Fix two distinct points in Pn+r, say p = [1 : 0 : · · · : 0] and q = [0 : 0 : · · · : 1]. Suppose that X is a general complete intersection with equations (s1 = 0) ∩ (s2 = 0) ∩ · · · ∩ (sr = 0) passing through p and q, where each si ∈ H 0(Pn+r, OPn+r (di)) is general among sections vanishing at p and q. Suppose there is a conic C contained in X, passing through p and q and let πC the projective 2-plane generated by C. If sC denotes the equation defining C in πC, there exists for each i = 1, . . . , r a si ∈ H 0(πC, OπC (di − 2)) (defining the residual curve) such that (si)πC = sC · si. Since X is general, it does not contain the 2-plane πC [DM98, Thm. 2.1]. Therefore (si)πC and si are not zero for at least one i. Conversely, let π be a projective 2-plane containing p and q and assume there exists a non-zero sC ∈ H 0(π, Oπ(2)) vanishing at p and q and, for each i = 1, . . . , r, there exists a si ∈ H 0(π, Oπ(di − 2)) such that (si)π = sC · si, then the conic (sC = 0) is obviously contained in X. Consider now the projective space of dimension n + r − 2 parametrizing the projective 2-planes in Pn+r containing p and q. Fixing homogeneous coordinates [a1 : · · · : an+r−1] on this space, let π[a1:···:an+r−1] = {[x : za1 : · · · : zan+r−1 : y] [x : z : y] ∈ P2} be such a 2-plane. Then (si)π[a1:...:an+r−1](x, z, y) = di k Xk=0 Xa=0 si a,kxayk−azdi−k where si The equation of an irreducible conic in this plane that passes through p and q is a,k is a homogeneous polynomial of degree di − k in the variables a1, . . . , an+r−1. sC = s2z2 + s1xz + s′ 1yz + xy. So for each i = 1, . . . , r, the equation (si)π = sC · si can be written explicitly di k Xk=0 Xa=0 a,kxayk−azdi−k = (s2z2 + s1xz + s′ si 1yz + xy) × di−2 k Xk=0 Xa=0 a,kxayk−azdi−2−k. si Thus we have to solve the equations si a,k = s2si a,k + s1si a−1,k−1 + s′ 1si a,k−1 + si a−1,k−2 for any 0 ≤ k ≤ di and 0 ≤ a ≤ k. Let us first solve this system (whose unknown variables are s2, s1, s′ and the si 1 defining the conic a,k's defining the residual curve) for each i separately. Note that X passes 4 LAURENT BONAVERO, ANDREAS H ORING 0,di = si a−1,di−2's. 0,di−3 and s′ a−1,di−2 for 1 ≤ a ≤ di − 1 provides the si di,di = 0. Therefore writing the di − 1 equations a,k's (this determines the residual curve (si = 0) !). through p and q if and only if si a,di = si si Considering the equations corresponding to (a, k) = (0, di − 1) and (di − 1, di − 1) allows to find s2 and s1. Considering then the equations corresponding to (a, k) = (1, di − 1) and (0, di − 2) gives si 1 (in particular this determines the conic, if it exists !). Write down successively the equations for (a, k), a = 1, . . . , k − 1, k = di − 1, . . . , 2 to find all the si Therefore, the r systems have a common solution if and only if the remaining equations for each system are satisfied and the corresponding conic is the same for each i. For each i, the remaining equations are "universal formulas" (meaning the coefficients just depend on the equations defining X) corresponding to (a, k) = (0, di − 3), . . . , (0, 0) and (a, k) = (di − 2, di − 2), . . . , (1, 1). This gives 2di − 4 equations of respective degrees 3, . . . , di and 2, 3, . . . , di − 1 in the variables a1, . . . , an+r−1. The 3r − 3 equations saying that the conic is the same for each i = 1, . . . , r are 2r − 2 equations of degree 1 and r − 1 equations of degree 2 in the variables a1, . . . , an+r−1. Altogether, using the relation d1 + · · · + dr = (n + 1)/2 + r, this gives exactly n + r − 2 homogeneous equations in the variables a1, · · · , an+r−1. We therefore get at least one solution. Moreover since X is general, the coefficients si a,k appearing in the initial equations are general. Since they completely determine the remaining n + r − 2 homogeneous equations, these equations are general. Thus the space of solutions is smooth and of the expected dimension, so there are exactly 1 2 by Bezout's theorem. r (di − 1)!di! solutions Yi=1 Let us briefly indicate how the same method gives the number of conics contained in X, passing through p and tangent to the line (pq). With the above notations, we have si di−1,di = si di,di = 0 and we have to solve the r systems di k Xk=0 Xa=0 si a,kxayk−azdi−k = (s2z2 + s1xz + s′ 1yz + y2) × di−2 k Xk=0 Xa=0 si a,kxayk−azdi−2−k which means a,k = s2si si a,k + s1si a−1,k−1 + s′ 1si a,k−1 + si a,k−2 for any 0 ≤ k ≤ di and 0 ≤ a ≤ k. The remaining details are left to the reader. 2.3. The space of conics is irreducible. Let G(2, n + r) be the Grassmannian of projective 2-planes contained in Pn+r and E be the tautological rank 3-bundle on G(2, n + r). The Hilbert scheme of conics in Pn+r is the projectivisation2 of S2E∗. Denote by ϕ : P(S2E∗) → G(2, n + r) the natural map. We have an exact sequence on P(S2E∗) : (∗) 0 → r Mi=1 ϕ∗Sdi−2E∗ ⊗ OP(S2E ∗)(−1) → r Mi=1 ϕ∗SdiE∗ → Q → 0 defining a vector bundle Q of rank n + 1 + 3r. Since X is a complete intersection (s1 = 0) ∩ (s2 = 0) ∩ · · · ∩ (sr = 0), the si's induce by restriction to 2-planes, pull-back and projection onto Q a section of Q whose zero locus Z is precisely the set of conics 2 In this article we follow the convention that the projectivisation of a vector bundle E is the variety of lines of E. COUNTING CONICS IN COMPLETE INTERSECTIONS 5 contained in X. Since E∗ is globally generated, the images of sections (s1, . . . , sr) give a vector space V ⊆ H 0(P(S2E∗), Q) that globally generates Q. Applying Bertini's theorem to this subspace we see that the zero locus of a general section in V is smooth. Since X is supposed to be a general complete intersection, Z is smooth and proving its irreducibility reduces to showing that h0(Z, OZ ) = 1. By the Koszul resolution of OZ , it is enough to show that for any 1 ≤ j ≤ rk Q hj(P(S2E∗), ∧jQ∗) = 0. Using the exact sequence (∗), this easily reduces to showing that for any 1 ≤ j ≤ rk Q and any 0 ≤ k ≤ j, H k(P(S2E∗), ∧k(⊕r i=1ϕ∗SdiE) ⊗ Sj−k((⊕r i=1ϕ∗Sdi−2E) ⊗ OP(S2E ∗)(1))) = 0. Since the higher direct images with respect to ϕ vanish, it is sufficient to show that for any 1 ≤ j ≤ rk Q and for any 0 ≤ k ≤ j, we have H k(G(2, n + r), ∧k(⊕r i=1SdiE) ⊗ Sj−k((⊕r i=1Sdi−2E) ⊗ S2E))) = 0. This will follow from Bott's theorem applied on G(2, n + r). Indeed, using Schur functor notation, let SbE be an irreducible factor appearing in the decomposition of ∧k(⊕r i=1Sdi−2E) ⊗ S2E) where b = (b1, b2, b3) is a triple of integers b1 ≥ b2 ≥ b3 ≥ 0. By the Littlewood-Richardson rule, we get i=1SdiE) ⊗ Sj−k((⊕r (∗∗) b2 + b3 ≥ k − r and b3 ≥ k − (d1 + . . . + dr) − r = k − (n + 1)/2 − 2r. On the other hand by Bott's theorem, the whole cohomology of SbE vanishes except maybe in the following cases : (1) k = n + r − 2 and (b1, b2, b3) = (b1, 0, 0) with b1 ≥ n + r − 1, (2) k = n + r − 2 and (b1, b2, b3) = (b1, 1, 0) with b1 ≥ n + r − 1, (3) k = n + r − 2 and (b1, b2, b3) = (b1, 1, 1) with b1 ≥ n + r − 1, (4) k = 2(n + r − 2) with b2 ≥ n + r and b3 = 0, 1, 2, (5) k = 3(n + r − 2) with b3 ≥ n + r + 1. The case n = 3 has been dealt with by Badescu, Beltrametti and Ionescu [BBI00], so we may assume n ≥ 5 since n is odd. In the first three cases, we get k − r = n − 2 ≥ 3 > b2 + b3 = 0, 1, 2, which is excluded by (∗∗). In case (4), since n ≥ 5, we get k − (n + 1)/2 − 2r = 3(n − 3)/2 > b3 = 0, 1, 2, which is again excluded by (∗∗). Case (5) is also excluded since we are only interested in the situation where k ≤ rk Q = n + 1 + 3r, but 3(n + r − 2) > n + 1 + 3r when n ≥ 5. We obtain the following corollary of the proof. 2.1. Corollary. Let X ⊂ Pn+r be a general smooth n-dimensional complete intersec- tion of multi-degree (d1, . . . , dr). Assume moreover that d1 + · · · + dr ≤ n + 1 2 + r and n ≥ 5. Then the family of conics contained in X is a nonempty, smooth and irreducible component of the Chow scheme C(X). Let us also mention that Harris, Roth and Starr have shown the irreducibility of the space of smooth rational curves of arbitrary degree e for general hypersurfaces of low degree d [HRS04]. 6 LAURENT BONAVERO, ANDREAS H ORING 2.4. Conics are quasi-lines. By the first step, there exists a conic C passing through two general points. Such a conic is necessarily smooth: a line d contained in X and passing through a general point satisfies TXd ≃ OP1(2) ⊕ OP1(1)⊕ n−3 2 ⊕ O ⊕ n+1 P1 2 , so an easy dimension count shows that two general points are not connected by a chain of two lines. Thus C smooth and its deformations with a fixed point cover a dense open subset in X. This implies that the normal bundle NC/X is ample [Deb01, Prop.4.10] and since −KX · C = n + 1, the curve C is a quasi-line. 2.5. Proof of the Corollary 1.3. The irreducibility of the variety of conics gives e0(X, l) = e(X, l) = 1 2 r Yi=1 (di − 1)!di!. The equality e0(X, l) = e(X, l) implies that general conics are G3 in X [Ion05, Cor. 4.6], in particular [Ion05, Cor. 4.7, Cor.1.9] apply. 3. A similar question Using exactly the same method as developed in §2.3 , one can prove the following result, left to the reader. 3.1. Proposition. Let Xd ⊂ Pn+1 be a general smooth n-dimensional hypersurface of degree d. Then, for n ≥ 7 and d ≤ n + 1, the family of conics contained in Xd is a nonempty, smooth and irreducible component of dimension 3n − 2d + 1 of the Chow scheme C(Xd). In the case of d = n + 1, there is a finite number of conics passing through a general point of Xn+1. Let us denote by Nn+1 this number. It seems that there are no known elementary method to compute this number. A general formula comes from the cal- culation of some Gromov-Witten invariants using mirror symmetry and an ordinary differential equation introduced by Givental. The following lines were written while reading [JNS04] and [Jin05]. 3.2. Proposition. (Coates, Givental - Jinzenji, Nakamura, Suzuki) Let Xn ⊂ Pn be a general smooth hypersurface of degree n in Pn. Let Nn be the number of conics passing through a general point of Xn. Then (2n)! 2n+1 − Nn = (n!)2 2 . Let us briefly explain where this result comes from. If a, b, c et d are four integers, let hOaObOcid be the Gromov-Witten invariant counting the number (possibly infinite) of rational curves of degree d contained in Xn and meeting 3 general subspaces of Pn, of respective codimension a, b and c. When a, b or c are equal to 1, each such rational curve has to be counted d times since the intersection of a degree d curve intersects a general hyperplane in d points. Since a general line meets Xn in n points, we get that Nn = hO1O1On−1i2/4n. In [Jin05] are introduced some constants Ln+1,n,d , called "structure constants of the quantum cohomology ring of Xn". They satisfy the following formula: m n−1 Xm=0 n−1 Ln+1,n,1 m wm = n Yj=1 (jw + (n − j)) COUNTING CONICS IN COMPLETE INTERSECTIONS 7 and n−2 Xm=0 Ln+1,n,2 m wm = n−2 Xj2=0 j2 j1 Xj1=0 Xj0=0 Ln+1,n,1 j1 j2+1 wj1−j0 (cid:18) 1 + w Ln+1,n,1 2 (cid:19)j2−j1 . It is also shown in [Jin05] that for every integer m, 0 ≤ m ≤ n − 2, we have Ln+1,n,2 hO1On−1−mOm+1i2/n. Then the proposition follows by evaluating the wn−2 coefficient in the second formula above, the wn−1 coefficient in the first and putting w = 2. = m References [AK77] [BBI00] [Bea95] [Deb01] [DM98] [HRS04] [IN03] [Ion05] [IR07] [IV03] [Jin05] Allen B. Altman and Steven L. Kleiman. Foundations of the theory of Fano schemes. Com- positio Math., 34(1):3 -- 47, 1977. Lucian Badescu, Mauro C. Beltrametti, and Paltin Ionescu. Almost-lines and quasi-lines on projective manifolds. In Complex analysis and algebraic geometry, 1 -- 27, Walter de Gruyter, Berlin, 2000. Arnaud Beauville. Quantum cohomology of complete intersections. Matematicheskaya Fizika, Analiz, Geometriya, 2:384 -- 398, 1995. Olivier Debarre. Higher-dimensional algebraic geometry. Universitext. Springer-Verlag, New York, 2001. Olivier Debarre and Laurent Manivel. Sur la vari´et´e des espaces lin´eaires contenus dans une intersection compl`ete. Math. Ann., 312:549 -- 574, 1998. Joe Harris, Mike Roth and Jason Starr. Rational curves on hypersurfaces of low degree. J. Reine Angew. Math., 571:73 -- 106, 2004. Paltin Ionescu and Daniel Naie. Rationality properties of manifolds containing quasi-lines. Internat. J. Math., 14(10):1053 -- 1080, 2003. Paltin Ionescu. Birational geometry of rationally connected manifolds via quasi-lines. In Projective varieties with unexpected properties, 317 -- 335, Walter de Gruyter, Berlin, 2005. Paltin Ionescu and Francesco Russo. Conic-connected Manifolds. To appear in Crelle's Jour- nal, http://arxiv.org/abs/math/0701885. Paltin Ionescu and Cristian Voica. Models of rationally connected manifolds. J. Math. Soc. Japan, 55(1):143 -- 164, 2003. Masao Jinzenji. Coordinate Change of Gauss-Manin System and Generalized Mirror Trans- formation. Int. J. Mod. Phys., A20:2131 -- 2156, 2005. [JNS04] Masao Jinzenji, Iku Nakamura and Yasuki Suzuki. Conics on a Generic Hypersurface. http://arxiv.org/abs/math/0412527. Laurent Bonavero. Institut Fourier, UMR 5582, Universit´e de Grenoble 1, BP 74. 38402 Saint Martin d'H`eres. France. e-mail : [email protected] Andreas Horing. Universit´e Paris 6, Institut de Math´ematiques de Jussieu, Equipe de Topologie et G´eom´etrie Alg´ebrique, 175, rue du Chevaleret, 75013 Paris, France. e-mail : [email protected]
1307.0724
1
1307
2013-07-02T15:13:19
Approximation on Nash sets with monomial singularities
[ "math.AG" ]
This paper is devoted to the approximation of differentiable semialgebraic functions by Nash functions. Approximation by Nash functions is known for semialgebraic functions defined on an affine Nash manifold M, and here we extend it to functions defined on Nash subsets X of M whose singularities are monomial. To that end we discuss first "finiteness" and "weak normality" for such sets X. Namely, we prove that (i) X is the union of finitely many open subsets, each Nash diffeomorphic to a finite union of coordinate linear varieties of an affine space and (ii) every function on X which is Nash on every irreducible component of X extends to a Nash function on M. Then we can obtain approximation for semialgebraic functions and even for certain semialgebraic maps on Nash sets with monomial singularities. As a nice consequence we show that m-dimensional affine Nash manifolds with divisorial corners which are class k semialgebraically diffeomorphic, for k>m^2, are also Nash diffeomorphic.
math.AG
math
APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ Abstract. This paper is devoted to the approximation of differentiable semialgebraic functions by Nash functions. Approximation by Nash functions is known for semialgebraic functions defined on an affine Nash manifold M , and here we extend it to functions defined on Nash sets X ⊂ M whose singularities are monomial. To that end we discuss first finiteness and weak normality for such sets X. Namely, we prove that (i) X is the union of finitely many open subsets, each Nash diffeomorphic to a finite union of coordinate linear varieties of an affine space, and (ii) every function on X which is Nash on every irreducible component of X extends to a Nash function on M . Then we can obtain approximation for semialgebraic functions and even for certain semialgebraic maps on Nash sets with monomial singularities. As a nice consequence we show that m-dimensional affine Nash manifolds with divisorial corners which are class k semialgebraically diffeomorphic, for k > m2, are also Nash diffeomorphic. 1. Introduction The importance of approximation in the Nash setting arises from the fact that this category is highly rigid to work with; for instance, it does not admit partitions of unity, a usual and fruitful tool when available. On the other hand, there exist finite differentiable semialgebraic partitions of unity and therefore approximation becomes interesting as a bridge between the differentiable semialgebraic category and the Nash one. There are already relevant results concerning absolute approximation in the Nash setting (e.g. Effroymson's approximation theorem [E, §1]). An even more powerful tool is relative approximation that allows approximation having a stronger control over certain subsets. In the '80s Shiota developed a thorough study of Nash manifolds and Nash sets (see [Sh] for the full collected work); among other things, he devised approximation on an affine Nash manifold relative to a Nash submanifold. Our purpose is to generalize this type of results developing Nash approximation on an affine Nash manifold relative to a Nash subset; of course, as one can expect we need some conditions concerning the singularities on the Nash subset and we will focus on those whose singularities are of monomial type. This will require a careful preliminary study of this type of singularities in the Nash context. As an interesting application of our results, we prove that the Nash classification of affine m-dimensional Nash manifolds with (divisorial) corners is equivalent to the Ck semialgebraic classification for k > m2. Recall that a set X ⊂ Rn is semialgebraic if it is a Boolean combination of sets defined by polynomial equations and inequalities. A semialgebraic set M ⊂ Rn is called an (affine) Nash manifold if it is moreover a smooth submanifold of (an open subset of) Rn; as in this paper all manifolds are affine we often drop the word affine. A function f : U → R on an open semialgebraic subset U ⊂ M is Nash if it is smooth and semialgebraic, i.e., its graph is semialgebraic. We denote by N(U ) the ring of Nash functions on U . Similarly, we define the ring Sν (U ) of Cν semialgebraic functions on U for ν ≥ 1. The orthogonal projection of a Nash manifold M ⊂ Rn into the tangent space at one of its points is a local diffeomorphism at the point; as it is a semialgebraic map we see that a semialgebraic subset M ⊂ Rn is an affine Nash manifold of dimension m if and only if every point x ∈ M has an open neighborhood U in Rn equipped with a Nash diffeomorphism (u1, . . . , un) : U → Rn that maps x to the origin and such that U ∩ M = {um+1 = ··· = un = 0}. Date: July 23, 2018. 2000 Mathematics Subject Classification. Primary 14P20; Secondary 58A07, 32C05. Authors supported by Spanish GAAR MTM2011-22435. 1 2 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ Even more, the Nash manifold M can actually be covered with finitely many open sets of this type (see [FGR, 2.2] and [Sh, I.3.9]). This kind of finiteness property permeates all the theory of semialgebraic sets and functions and plays a relevant role in the present paper. We are interested in Nash sets, that is, the zero sets of Nash functions f : M → R. It is well known that the set Smooth(Z) of smooth points of a semialgebraic set Z, that is, the points x ∈ Z where the germ Zx equals the germ of an affine Nash submanifold, is a dense open semialgebraic subset of Z (see [St]). We will focus on Nash sets whose singular points (i.e., points which are not smooth) are of a specific form. Definition 1.1. Let X ⊂ M be a set and let x ∈ X. The germ Xx is a monomial singularity if there is a neighborhood U of x in M equipped with a Nash diffeomorphism u : U → Rm with u(x) = 0 that maps X ∩ U onto a union of coordinate linear varieties. That is, there is a (finite) family Λ of subsets of indices λ = {ℓ1,··· , ℓr} of possibly different cardinality r ≤ m such that X ∩ U =[λ∈Λ {uλ = 0} where u = (u1, . . . , um) and {uλ = 0} denotes {uℓ1 = ··· = uℓr = 0}. For simplicity we assume that there are no immersed components, that is, if λ, λ′ ∈ Λ are different then λ * λ′ and λ′ * λ. This assures that the germs {uλ = 0}x, λ ∈ Λ, are the irreducible components of the germ Xx. We call Λ a type of the monomial singularity Xx; we also say that X has a monomial singularity of type Λ at x. Of course, a different Nash diffeomorphism u′ may provide a different type Λ′ and therefore a monomial singularity has several types. Thus, two types Λ and Λ′ will be called equivalent if the union of the coordinate linear varieties given by Λ is Nash diffeomorphic to the one given by Λ′ as germs in the origin. In Section 3 we show that this equivalence is in fact global via linear isomorphisms and the resulting classes can be characterized arithmetically (Proposition 3.12). Nash monomial singularity germs are a generalization of Nash normal crossings germs, that is, locally Nash diffeomorphic to a finite union of coordinate linear hyperplanes (see also Section 3). Nash normal crossings have been studied extensively due to their relevance in desingularization problems. In [FGR] several finiteness properties of Nash normal crossings were established and our first objective in Sections 3 to 5 is to extend them to the much more general setting here. Specifically, in Section 3 we study in depth the concept of type and in Section 5 we prove the following semialgebraic property of the monomial singular locus (see 4.B). Proposition 1.2. Let X ⊂ M be a Nash set. Fix a type Λ and put T (Λ) = {x ∈ X : X has a monomial singularity of type Λ at x}. Then, the set T (Λ) is semialgebraic. In Section 5 we analyze the sets of interest in our discussion. Definition 1.3. A Nash set X ⊂ M has monomial singularities in M if all germs Xx, x ∈ X, are monomial singularities. In fact, we will see that any closed semialgebraic set whose germs are all monomial singularities is a coherent Nash set with monomial singularities (see Lemma 5.1). Now, the main purpose in Section 5 is the following finiteness property (see the proof in 5.5). Theorem 1.4. Let X ⊂ M be a Nash set with monomial singularities. Then X can be covered with finitely many open sets U of M each one equipped with a Nash diffeomorphism u : U → Rm that maps X ∩ U onto a union of coordinate linear varieties. To prove the above result we will study first Nash functions on Nash sets with monomial singularities. APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 3 Definition 1.5. Let Z be a semialgebraic subset of M . We say that a function f : Z → R is a Nash function if there exists an open semialgebraic neighborhood U of Z and a Nash extension F : U → R of f . We denote by N(Z) the ring of Nash functions of Z. If X is a Nash set then we say that a function f : X → R is a c-Nash or cN function if its restriction to each irreducible component is a Nash function. We denote by cN(X) the ring of cN functions. Similarly, we define the ring Sν (X) of Cν semialgebraic functions on X and the ring cSν(X) of cSν semialgebraic functions on X for ν ≥ 1 (cN or cSν functions on X are clearly continuous). We point out that if Z is a semialgebraic subset of M and f : Z → R is a function with a Nash extension to a non necessarily semialgebraic open neighborhood of Z (see the definition in 2.A) then there is a Nash extension to a semialgebraic one [FGR, 1.3]. We will show in 5.2 that by means of analytic coherence of Nash sets with monomial singularities we have the following notorious weak normality property : Theorem 1.6. If X is a Nash set with monomial singularities then N(X) = cN(X). The ring N(X) has been revealed crucial to develop a satisfactory theory of irreducibility and irreducible components for the semialgebraic setting [FG]; as one can expect such theory extends Nash irreducibility and can be based as well on the ring of analytic functions on the semialgebraic set. Once we have presented these properties of Nash sets with monomial singularities we approach in Sections 6, 7 and 8 the approximation problems. First the ring Sν (M ) of Cν semialgebraic functions is equipped with an Sν Whitney topology via Sν tangent fields (see 2.C). The fact that we have Sν bump functions as well as finite Sν partitions of unity makes a crucial difference between Sν and Nash functions and the existence of these glueing functions justify our interest in approximation. In particular, given a closed semialgebraic set Z of M we can extend any Sν function on Z to M and therefore Sν (Z) carries the quotient topology making the restriction map ρ : Sν(M ) → Sν (Z) a quotient map. If Z is a closed semialgebraic set and T is a semialgebraic subset of Rb a map f = (f1, . . . , fb) : Z → T ⊂ Rb is Sν if each component fi is Sν and Sν(Z, T ) inherits the topology from the product Sν(Z, Rb) = Sν(Z) × ··· × Sν(Z). Obviously, approximation problems focus mainly on maps f : Z → T where either Z is not compact or T is not a Nash manifold. For, by the Stone-Weierstrass approximation theorem [N2, 1.6.2], if Z is compact any Cν map f : Z → T ⊂ Rb can be Cν approximated by a polynomial map g : Z → Rb. Therefore, if T is a Nash manifold and g is close enough to f then g(Z) is contained in a tubular neighborhood which retracts onto T . In the '80s, Shiota carried a systematic study of approximation over Nash manifolds and showed that if Z ⊂ M is a (non necessarily compact) closed semialgebraic set then any function in Sν(Z) can be approximated by Nash functions (see Proposition 2.C.5). Of course, it follows that if N is an affine Nash manifold then using a Nash tubular neighborhood of N any map in Sν (Z, N ) can be also approximated by Nash maps in N(Z, N ) (see Proposition 2.D.3). However, if the codomain is not an affine Nash manifold then the situation gets more complicated. As we have announced before, our purpose in this paper is to obtain this kind of approximation for Sν maps between certain Nash sets with monomial singularities. Specifically, we say that a Nash set with monomial singularities X is a Nash monomial crossings if in addition the irreducible components of X are Nash manifolds. Again this mimics the concept of normal crossing divisor [FGR, 1.8]. Our main result concerning approximation is the following, which is proved in Section 8 (see 8.4). Theorem 1.7. Let X ⊂ M be a Nash set with monomial singularities and let Y ⊂ N be a Nash monomial crossings. Let m = dim(M ), n = dim(N ) and q = m(cid:0)(cid:0) n [n/2](cid:1) − 1(cid:1) where [n/2] denotes the integer part of n/2. If ν ≥ q then every Sν map f : X → Y that preserves irreducible components can be Sν−q approximated by Nash maps g : X → Y . 4 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ To prove this we need to rework approximation for Sν functions over Nash set with monomial singularities. Indeed, we start in Section 6 with a result concerning extension of Sν functions, which is well-known for Nash submanifolds (see Proposition 6.2): If X ⊂ M is a Nash set with monomial singularities, then there are continuous extension linear maps θ : Sν(X) → Sν(M ) such that θ(h)X = h. This is stronger than the mere existence of extensions. This extension result will be used to show in Section 7 that Nash functions on Nash sets with monomial singularities can be relatively approximated (and not only absolutely, Proposition 6.2). Namely: An Sν function F : M → R whose restriction to X is Nash can be Sν−m approximated by Nash functions H : M → R that coincide with F on X. Section 9 is devoted to the classification of affine Nash manifolds with corners, and our approxi- mation results will be crucial to compare Sν and Nash classifications. This somehow complements Shiota's results on Cν classification of Nash manifolds [Sh, VI.2.2]. An (affine) Nash manifold with corners is a semialgebraic set Q ⊂ Ra that is a smooth sub- manifold with corners of (an open subset of) Ra. In [FGR] it is proved that any Nash manifold with corners Q ⊂ Ra is a closed semialgebraic subset of a Nash manifold M ⊂ Ra of the same dimension and the Nash closure X in M of the boundary ∂Q is a Nash normal crossings. Note that we can define naturally Sν functions and maps and their topologies via the closed inclusion of Q in M ; of course this does not depend on the affine Nash manifold M (see 2.E for further details). A Nash manifold with corners Q has divisorial corners if it is contained in a Nash manifold M as before such that the Nash closure X of ∂Q in M is a normal crossing divisor (as one can expect this is not always the case and a careful study can be found in [FGR, 1.12]). Theorem 1.8. Let Q1 and Q2 be two m-dimensional affine Nash manifolds with divisorial cor- ners. If Q1 and Q2 are Sν diffeomorphic for some ν > m2 then they are Nash diffeomorphic. All the basic notions and the notation used in this paper will be explained in the following section. The reader can proceed directly to Section 3 and refer to the Preliminaries when needed. 2. Preliminaries In this section we introduce the concepts and notation needed in the sequel. First, we adopt the following conventions: M ⊂ Ra and N ⊂ Rb are affine Nash manifolds of respective dimensions m and n; X and Y are Nash subsets of an affine Nash manifold and their irreducible components are denoted by Xi and Yj. The semialgebraic sets will be denoted by S, T, Z; on the other hand, Q ⊂ Ra is an affine Nash manifold with corners, ∂Q is its boundary and Int(Q) = Q \ ∂Q is its interior. Also Sν and Nash functions on a semialgebraic set will be denoted by f, g, h, . . . and their extensions to a larger semialgebraic set by F, G, H, . . . Homomorphisms between rings of Nash functions, Sν functions, etc. are denoted using greek letters like θ, γ, . . . but we reserve as usual ε and δ to denote (small) positive real numbers or positive semialgebraic functions involved in approximation of functions. Let us recall some general properties of semialgebraic sets. Semialgebraic sets are closed under Boolean combinations and by quantifier elimination they are also closed under projections. In other words, any set defined by a first order formula in the language of ordered fields is a semial- gebraic set [BCR, p. 28, 29]. Keeping this in mind, the basic topological constructions as closures, interiors or borders of semialgebraic sets are again semialgebraic. Also images and preimages of semialgebraic sets by semialgebraic maps are again semialgebraic. The dimension dim(Z) of a semialgebraic set Z is the dimension of its algebraic Zariski closure [BCR, §2.8]. The local dimension dim(Zx) of Z at a point x ∈ Z is the dimension dim(U ) of a small enough open semialgebraic neighborhood U of x in Z. The dimension of Z coincides with the maximum of APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 5 those local dimensions. For any fixed d the set of points x ∈ Z such that dim(Zx) = d is a closed semialgebraic subset. An important property of compact semialgebraic sets is that they can be triangulated [BCR, §9.2]. More relevant for us here is that given a Nash manifold M ⊂ Ra and some semialgebraic subsets Z1, . . . , Zs ⊂ M , as a consequence of this triangulability property via a one-point com- pactification of M , there is a stratification G of M compatible with Z1, . . . , Zs. That is, there exists a finite collection G of disjoint semialgebraic subsets of M called strata whose union is M and which are affine Nash manifolds Nash diffeomorphic to some Rd with the following properties: (1) The sets Z1 . . . , Zs are unions of strata, (2) The closure in M of a stratum of G is a finite union of strata of G. In particular, one deduces that if Σ, Γ are strata of G, then either Σ ⊂ Γ or Σ∩Γ = ∅. Therefore, if GΣ = {Γ ∈ G : Σ ⊂ Γ}, then the semialgebraic set SΓ ∈GΣ Γ is an open semialgebraic neighborhood of Σ in M ; this fact will be used freely along the sequel. (3) Every stratum Γ is connected at every point x ∈ Γ , that is, x has a basis of neighborhoods V in M with connected intersection V ∩ Γ . This, together with the fact that Γ is an affine Nash an x of the germ Γx is an irreducible analytic germ of manifold, implies that the analytic closure Γ dimension dim(Γ ). (4) Suppose the stratum Γ is adherent to another stratum Σ, that is, Σ ⊂ Γ . Then Σ has a basis of neighborhoods V in M with connected intersection V ∩ Γ . Following the terminology above, we say that Γ is connected at Σ. We refer to [FGR, 2.3] for a more detailed presentation of the triangulation theorem and the deduction of the preceding stratification. 2.A. Basics on the Nash category. The open semialgebraic subsets of M are a base of the topology and therefore Nash functions define a sheaf that we will denote by N. In particular, the sheaf N induces a notion of Nash function f : U → R over an arbitrary open subset U of M possibly not semialgebraic. In case U is an open subset of Ra, Nash means that f is smooth and there exists a nonzero polynomial P (x, t) ∈ R[x, t] = R[x1, . . . , xa, t] such that P (x, f (x)) = 0 for all x ∈ U . If U is semialgebraic this definition coincides with the one in the Introduction. We recall that both N(M ) and Nx are noetherian for any x ∈ M (see [BCR, 8.2.11, 8.7.18]) and therefore it makes sense to consider the Nash closure globally and locally. Moreover, Nash functions are analytic and therefore Nash manifolds and Nash sets are analytic sets [BCR, 8.1.8]. The ring of germs of analytic functions Ox is always noetherian even thought the ring of global analytic functions O(M ) is noetherian only in case M is compact. However, any arbitrary intersection of global analytic sets is a global analytic set and so the local and the global analytic closure are well-defined [N1, Ch. V, Prop. 16]. We have: Fact 2.A.1. Let Z ⊂ M be a semialgebraic set. Then: (i) [FG, 2.10] The Nash and the analytic closures of Z coincide. In particular, if Z is global analytic then it is a Nash set. (ii) [FGR, 2.8] If Z is a Nash set, then its Nash irreducible components are also its global analytic irreducible components. Since M is a semialgebraic subset of Ra and because of Definition 1.5 we apparently have two definitions of Nash function f : M → R. Of course, both notions are equal because every affine Nash manifold has a Nash tubular neighborhood. Fact 2.A.2. [BCR, 8.9.5] Let M ⊂ Ra be a Nash manifold. Then there exists an open semialge- braic neighborhood U of M in Ra and a Nash retraction ρ : U → M . 6 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ 2.B. Coherence and Nash functions. In the Introduction we have provided the definition of Nash function on a semialgebraic subset of an affine Nash manifold. Now for a Nash set X it will be useful to recover this definition via global sections of a certain sheaf. Of course, the first choice should be the following. Let I be the sheaf of N-ideals given by Ix = I(Xx), the germs of Nash functions vanishing on X. The support of I is X and for any x ∈ M we have (N/I)x = Nx/Ix = N(Xx). However, as in the analytic setting, the sheaf N/I has in general a bad local-global behaviour and the ring of global sections is not necessarily N(X). Cartan showed that in the analytic context this can be solved using coherence. Let N be an analytic manifold and O its sheaf of analytic germs. Recall that a sheaf of O-modules F is coherent if (i) F is of finite type, i.e., for every x ∈ N there is an open neighborhood U and a surjective morphism OnU → FU and (ii) any morphism OpU → FU has a kernel of finite type. By [C, Prop. 4] the sheaf O is coherent and therefore a sheaf of O-ideals is coherent if and only if it is of finite type. The famous Theorems A and B describe the local-global behaviour of coherent sheaves: The stalks are spanned by the global sections and each p-cohomology group of the sheaf with p > 0 is trivial [C, Thm. 3]. In particular, for any coherent locally analytic set Y ⊂ N , i.e., locally described by finitely many analytic equations and such that its sheaf IO of O-ideals of germs of analytic functions vanishing on Y is coherent, we have the following properties. Fact 2.B.1. [C, Prop. 14, 15] Every coherent locally analytic set Y of an analytic manifold N is (global) analytic. Moreover, if Y is irreducible then it has pure dimension. Back to the Nash setting, even for sheaves of N-ideals, coherence is not enough (see the Intro- duction of [CRS3]) because semialgebraicity involves a finiteness phenomenon that mere coherence does not capture. Thus we say that a sheaf F of N-ideals is finite if there is a finite open semial- gebraic covering {Ui} of M such that each FUi is generated by finitely many Nash functions over Ui. Since N is coherent as a sheaf of N-ideals [Sh, I.6.6] it follows that a finite sheaf of N-ideals is coherent. In [CS] and [CRS2] the authors prove that this is the correct notion for sheaves of N-ideals (we remark that in general for N-modules the right notion is strong coherence [CRS3]). We suggest the survey [CRS1] as a general reference. Fact 2.B.2. Let F be a finite sheaf of N-ideals and x ∈ M . Then (A) the stalk Fx is generated by global sections, and (B) every global section σ of N/F is represented by a global Nash function. This result assures that if X is a Nash subset of M and I(X) stands for its (finitely generated) ideal {f ∈ N(M ) : f (X) = 0}, then the sheaf of N-ideals given by Jx = I(X)Nx is the biggest finite sheaf of ideals whose support is X (in fact, as we will see later, is the biggest coherent one). Moreover, let us see that there is a natural isomorphism between the ring N(X) and the ring of global sections of N/J. Indeed, by [Sh, I.6.5] we have Γ(U, J) = I(X)N(U ) for any open semialgebraic subset U of M . Then by Fact 2.B.2.(B) we deduce Γ(U, N/J) = N(U )/I(X)N(U ). On the other hand, the support of N/J is X and therefore for any open semialgebraic set U containing X we have Γ(X, N/J) ≡ Γ(U, N/J) ≡ Γ(M, N/J). The fact that N(M )/I(X) ≡ Γ(M, N/J) ≡ Γ(U, N/J) ≡ N(U )/I(X)N(U ) means that for every Nash function g : U → R there is a Nash function h : M → R such that g − hU ∈ I(X)N(U ) and therefore g = h on X. In particular, every function f ∈ N(X) has a Nash extension to M , so that the restriction homomorphism N(M ) → N(X) f 7→ fX (2.1) is surjective and hence N(X) = N(M )/I(X) ≡ Γ(X, N/J), as required. On the other hand, we also need to describe cN(X) as a ring of sections of a sheaf. Let X1, . . . , Xs be the irreducible components of X and for every 1 ≤ i, j ≤ s let Ji and Jij be respectively the sheaves of N-ideals given by Ji,x = I(Xi)Nx and Jij,x = I(Xi ∩ Xj)Nx. We define APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 7 the morphism induced by φ : N/J1 × . . . × N/Js →Yi<j (f1, . . . , fs) 7→ (fi − fj)i<j. N/Jij Consider the kernel sheaf ker(φ). Using equation (2.1) we define a multiple restriction map cN(X) → N(M )/I(X1) × ··· × N(M )/I(Xs) : f 7→ (fX1, . . . , fXs) that provides an isomorphism between cN(X) and the global sections Γ(M, ker(φ)) ⊂ N(M )/I(X1) × ··· × N(M )/I(Xs). Indeed, any global section of ker(φ) ⊂ N/J1 × . . . × N/Js is represented by global Nash functions (g1, . . . , gs). Since (gi − gj)x ∈ I(Xi ∩ Xj)Nx for any x ∈ Xi ∩ Xj we deduce that gi = gj on Xi ∩ Xj and therefore the function defined by g(x) = gi(x) if x ∈ Xi is c-Nash on X, as required. Roughly speaking, the coherent sheaf N/J captures the global notions concerning Nash func- tions and N/I the local ones. Hence it is natural to ask when I = J and for the answer we look at the analytic functions. Consider the ideal IO(X) of all analytic functions vanishing in X. Now, if X is coherent as an analytic set then by Theorem A of Cartan the stalks IO,x are generated by the global sections Γ(M, IO) = IO(X). Since IO(X) is generated by I(X) (see [FGR, 2.8]) it follows that JxOx = I(X)Ox = IO(X)Ox = IO,x ⊃ IxOx and therefore JxOx = IxOx. The monomorphism Nx ֒→ Ox is faithful flat [BCR, 8.3.2], so that Jx = Ix. Then I = J and therefore (N/J)x = (N/I)x = Nx/Ix = N(Xx). In particular, since by definition J is coherent as N-sheaf we have that X is also coherent as a Nash set. Moreover, since IO,x = IxOx [BCR, 8.6.9] it is easy to prove that Nash coherence implies analytic coherence. Summarizing we have: Proposition 2.B.3. A Nash set X is coherent as an analytic set if and only if it so as a Nash set. Furthermore, if X is coherent then for any semialgebraic open subset U of M we have I(X)N(U ) = I(X ∩ U ) (2.2) where I(X ∩ U ) = {f ∈ N(U ) : fX∩U = 0}. Indeed, let J′ be the sheaf of NU -ideals given by x = I(X∩U )Nx. Since X∩U is a Nash subset of U , we know that N(X∩U ) = Γ(X∩U, NU /J′) = J′ N(U )/I(X ∩ U ). Moreover, X ∩ U is also coherent in U and hence J′ = IU , so that NU /J′ = (N/I)U = (N/J)U . Thus N(X ∩ U ) = Γ(X ∩ U, NU /J′) = Γ(U, N/J) = N(U )/I(X)N(U ). We finish this discussion with a well-known fact included for the sake of completeness. Proposition 2.B.4. Let X be a coherent Nash set and let X1, . . . , Xs be its Nash irreducible components. Then, for every x ∈ X, each germ Xi,x is the union of some Nash irreducible com- ponents of the germ Xx. In particular, the irreducible components X1, . . . , Xs are also coherent. Proof. By Fact 2.A.1 the Nash irreducible components X1, . . . , Xs are also the irreducible compo- nents in the analytic sense. Since X is a (global) analytic set we can consider its complexification eX whose irreducible components are eX1, . . . , eXr (see [N1, Ch.V, Prop.15,17]). The irreducibility of each Xi implies that each germ eXi,x is pure dimensional [N1, Ch.IV, Cor.4]. The pure di- mensionality of the sets eXi implies, by the Identity Principle, that no irreducible component of the germ eXi,x is contained in eXj,x if i 6= j. This shows that the decompositions into irreducible components of eX1,x, . . . , eXs,x provide the decomposition of eXx into irreducible components. On the other hand, fix a point x ∈ M and consider the Nash irreducible components Y1, . . . , Yr of the germ Xx. These are also the analytic irreducible components [BCR, 8.3.2] and therefore they are coherent [N1, Ch.V, Prop.6]. Furthermore, their complexifications eYi are the irreducible compo- nents of the complexification fXx (see [N1, Ch.V, Prop.2]). Now, it follows from the coherence of X that the germ eXx of the complexification eX is the complexification fXx of the germ Xx (see 8 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ Yj. Thus, Xi,x is coherent because it is a finite union of coherent germs [C, Prop.13]. [C, Prop.12]). Therefore each eXi,x is the union of some eYj; hence, each Xi,x is the union of some 2.C. Spaces of differentiable semialgebraic functions. Let M ⊂ Ra be an affine Nash manifold. Recall that for every integer ν ≥ 1 we denote by Sν(M ) the set of all semialgebraic functions f : M → R that are differentiable of class ν. We equip Sν (M ) with the following Sν semialgebraic Whitney topology [Sh, II.1, p. 79 -- 80]. Let ξ1, . . . , ξr be semialgebraic Sν−1 tangent fields on M that span the tangent bundle of M . For every continuous semialgebraic function ε : M → R+ we denote by Uε the set of all functions g ∈ Sν (M ) such that (cid:3) g < ε and ξi1 ··· ξiµ(g) < ε for 1≤ i1, . . . , iµ ≤ r, 1≤ µ≤ ν. These sets Uε form a basis of neighborhoods of the zero function for a topology in Sν (M ) that does not depend on the choice of the tangent fields. It is well-known that each Sν (M ) is a Hausdorff topological ring [Sh, II.1.6], but neither a topological vector space nor a Fr´echet space. Note that the obvious inclusions Sν(M ) ⊂ Sµ(M ), ν > µ, are continuous. Moreover, since semialgebraic smooth functions on M are Nash by definition, we have N(M ) =Tν Sν (M ). The first important result is that the inclusion N(M ) ⊂ Sν(M ) is dense. Fact 2.C.1. [Sh, II.4.1, p. 123] Every semialgebraic Sν function on M can be approximated in the Sν topology by Nash functions. It is clear that given a semialgebraic set Z ⊂ M the zero-ideal I ν (Z) = {f ∈ Sν (M ) : fZ ≡ 0} is closed. Indeed, if f (z) 6= 0 for some z ∈ Z, no function vanishing on Z can be closer than the constant semialgebraic function ε = f (z) > 0 to f . The closedness of I ν(Z) in Sν(M ) is the standard assumption to equip Sν (M )/I ν (Z) with the quotient topology. Furthermore, with this topology the continuous quotient map Sν (M ) → Sν(M )/I ν (Z) is also open: if U is an open set in Sν(M ), its saturation U Z = [f ∈ U (f + I ν(Z)) = [h∈I ν (Z) (U + h) is open as translations are homeomorphism of the ring Sν (M ). Now, if the semialgebraic set Z is closed in M we can identify Sν (Z) with the quotient Sν(M )/I ν (Z). Indeed, every function in Sν (Z) can be extended to M : we can use bump Sν functions to see that a semialgebraic Sν function f : U → R defined on an open semialge- braic neighborhood U of Z coincides on a perhaps smaller neighborhood with a semialgebraic Sν function F : M → R. Therefore, we have a topology on Sν (Z) such that the restriction map ρ : Sν(M ) → Sν(Z) is an open quotient homomorphism. In general, for any two semialgebraic sets Z ⊂ Z ′ the composition of the restrictions Sν(M ) → Sν(Z) and Sν (Z) → Sν(Z ′) coincide with the restriction Sν(M ) → Sν (Z ′), so that: (2.C.2) Given two closed semialgebraic sets Z ′ ⊂ Z the restriction map Sν (Z) → Sν (Z ′) is an open quotient map. In several situations we have to work with a semialgebraic set N of M which is in turn a closed affine Nash submanifold of M . We already pointed out that the intrinsic notion of semialgebraic Sν map on N as manifold equals the one coming from M as semialgebraic set because of the existence of Nash tubular neighborhoods. Furthermore, the topology in Sν (N ) coming from the tangent fields of N and the one as a closed semialgebraic subset of M are equal: Proposition 2.C.3. Let N ⊂ M be affine Nash manifolds with N closed in M . Then the restriction homomorphism Sν(M ) → Sν(N ) is open when the Sν topologies are defined through tangent fields. APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 9 Proof. Consider Sν (M ) and Sν (N ) endowed with the topologies defined through tangent fields. It is known that then the restriction is continuous [Sh, II.1.5, p. 83], and we are to see it is open. The key fact is that there is a continuous linear map θ : Sν (N ) → Sν (M ) with θ(h)N = h (see [Sh, II.2.14, p. 107]). Let F be a semialgebraic Sν function on M and let f = FN . We must see that if g ∈ Sν(N ) is close enough to f , then there is a semialgebraic Sν extension G of g to M arbitrarily close to F . But given a θ as above, the function G = F + θ(g) − θ(f ) is close to F when g is close to f . (cid:3) Note that in the preceding result we are using the existence of a continuous extension linear map Sν(N ) → Sν (M ) when N is a closed Nash submanifold of M . This property is stronger than mere extension. In Proposition 6.2 we will prove that such extension linear maps exist for Nash sets with monomial singularities. Remark 2.C.4. Observe that all we put so far can be done for Sν manifolds, that is, semialgebraic sets which are differentiable submanifolds of class ν (see [Sh, I.3.11(i), p. 30]). Instead of Nash tubular neighborhood we have to use bent tubular neighborhoods [Sh, II.6.1, p. 135]. In the case of a Nash set X ⊂ M it is also natural to consider the ring cSν (X): if X1, . . . , Xs are the irreducible components of X then f ∈ cSν (X) means fXi ∈ Sν (Xi) for 1 ≤ i ≤ s. The multiple restriction homomorphism cSν(X) → Sν(X1)×···× Sν(Xs) is injective and therefore we consider in cSν (X) the topology induced by the product topology of Sν (X1) × ··· × Sν(Xs) . Note that the image of cSν(X) in Sν (X1) × ··· × Sν(Xs) is closed. Indeed, this image is the kernel of the homomorphism Sν (X1) × ··· × Sν (Xs) →Yi<j Sν (Xi ∩ Xj) : (f1, . . . , fs) 7→ (fiXi∩Xj − fjXi∩Xj )i<j. This homomorphism is continuous because Sν (M ) is a topological ring and therefore its kernel is closed. Moreover, the inclusion γ : Sν(X) ֒→ cSν(X) : f 7→ f is continuous, but one cannot expect it to be surjective in general. Since the embedded image of cSν (X) in Sν (X1) × ··· × Sν (Xs) is closed, if γ is a homeomorphism then the multiple restric- tion homomorphism Sν(X) → Sν (X1) × ··· × Sν (Xs) is a closed embedding. As we will see in Proposition 6.2 this happens whenever X is a Nash set with monomial singularities. Finally, we point out that Shiota's approximation for affine Nash manifolds implies readily Indeed, since the restriction homomorphism a rather general absolute approximation result. Sν(M ) → Sν (Z) is onto and continuous, from Fact 2.C.1 we obtain: Proposition 2.C.5. If Z is a closed semialgebraic set of M then every Sν function f : Z → R can be Sν approximated by Nash functions. 2.D. Spaces of differentiable semialgebraic maps. Let M ⊂ Ra be a Nash manifold. Let Z ⊂ M and T ⊂ Rb be semialgebraic sets. A semialgebraic map f : Z → T is Sν when so are its components: we write f = (f1, . . . , fb) : Z → T ⊂ Rb and then check if the functions fk : Z → R are Sν . We denote Sν (Z, T ) the set of all Sν maps Z → T . We suppose henceforth that Z is closed in M in order to endow Sν (Z, T ) with a topology. We use the canonical inclusions Sν(Z, T ) ⊂ Sν(Z, Rb) = Sν (Z, R) × ··· × Sν (Z, R) : f 7→ (f1, . . . , fb). b The product has of course the product topology, and then Sν (Z, T ) is equipped with the subspace topology. Roughly speaking, g is close to f when its components gk are close to the components fk of f . This definition is the one used in the case of semialgebraic Sν manifolds [Sh, II.1.3, p. 80]. If T is contained in another semialgebraic subset T ′, the inclusion Sν(Z, T ) ⊂ Sν(Z, T ′) is an embedding. If X is a Nash set, we say again that a semialgebraic map f = (f1, . . . , fb) : X → T 10 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ is cSν when so are its components fk and we denote by cSν (X, T ) the set of all such maps. If X1, . . . , Xs are the irreducible components of X then the multiple restriction homomorphism cSν(X, T ) → Sν (X1, T ) × ··· × Sν (Xs, T ) is injective and therefore we consider in cSν (X, T ) the topology induced by the product topology of Sν (X1, T ) × ··· × Sν(Xs, T ). Again, the image of cSν(X, T ) in Sν (X1, T ) × ··· × Sν (Xs, T ) is closed. Some basics on spaces of functions pass immediately to spaces of maps. For instance, given two closed semialgebraic sets Z ′ ⊂ Z ⊂ M , the restriction map Sν(Z, T ) → Sν (Z ′, T ) : f 7→ f ′ = fZ ′ is continuous. On the other hand, we know that if T = R this restriction is an open quotient, from which the same follows for T open in Rb (as then Sν (Z, T ) is open in Sν (Z, Rb)). It is well-known that already for topological reasons surjectivity fails in general: for instance, the identity S1 → S1 has no continuous extension R2 → S1. We want to study in which situations the restriction map is at least open. A useful fact will be that composite on the left is continuous. Proposition 2.D.1. Let Z ⊂ M be a closed semialgebraic set and let T ⊂ Rb be a locally compact semialgebraic set. Let h : T → T ′ ⊂ Rc be an Sν map of semialgebraic sets. Then the map is continuous for the Sν topologies. h∗ : Sν(Z, T ) → Sν(Z, T ′) : f 7→ h ◦ f Proof. Since T is locally compact, it is closed in some open semialgebraic set U ⊂ Rb, and there is a semialgebraic Sν map H : U → Rc that extends h. We have the following commutative diagram: Sν (Z, T ) j ❄ Sν (Z, U ) ρ ✻ Sν(M, U ) j ′ ❄ h∗−→ Sν (Z, T ′) H∗−→ Sν (Z, Rc) H∗−→ Sν (M, Rc) ✻ ρ where j and j′ stand for the canonical embeddings and ρ for the restriction quotients. First we explore the lower square. Here we know that the lower H∗ is continuous (the Nash manifolds case [Sh, II.1.5, p. 83]), hence the composition ρ◦ H∗ is continuous too. The latter map coincides with H∗ ◦ ρ, which is thus continuous. But ρ is a quotient map (the target is open in an affine space), hence the middle H∗ is continuous. Now we turn to the upper square. As we have just seen that H∗ is continuous, the composite H∗ ◦ j is continuous. But this map coincides with j′ ◦ h∗, which is consequently continuous. As j′ is an embedding, h∗ is continuous. (cid:3) Back to restrictions we deduce the following (the same result is true if we have a Sν manifold instead of a Nash manifold). Proposition 2.D.2. Let Z ′ ⊂ Z be two closed semialgebraic sets, and let N ⊂ Rb be a Nash manifold. Then the restriction Sν (Z, N ) → Sν(Z ′, N ) : f 7→ fZ ′ is an open map. Proof. Let U be an open semialgebraic tubular neighborhood of N equipped with a Nash retrac- tion η : U → N (that is η(y) = y for y ∈ N ). We must see that for any open neighborhood V of a function f ∈ Sν (Z, N ), every close enough approximation g′ ∈ Sν(Z ′, N ) of fZ ′ has an k of g′ are extension g ∈ V. By definition, g′ is close to fZ ′ if and only if the components g′ close to those fkZ ′ of fZ ′. But since the restriction of functions is open, if g′ k is close enough to fkZ ′, then it has an extension gk to Z close to fk. Thus the gk's are the components of an extension g : Z → Rb of g′ which is close to f , and we can take it close enough for its image to be contained in U . To obtain a map into N we take h = η ◦ g : Z → N . This settles the matter APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 11 because composition on the left is continuous (Proposition 2.D.1), and so, h(= η ◦ g) is close to f (= η ◦ f ), as wanted. (cid:3) Nash tubular neighborhoods also help to get absolute approximation for maps into Nash man- ifolds. Proposition 2.D.3. Let Z ⊂ M be a closed semialgebraic set and let N ⊂ Rb be a Nash manifold. Every Sν map f : Z → N can be Sν approximated by Nash maps. Proof. Pick first a Nash retraction η : W → N defined on an open semialgebraic tubular neigh- borhood W ⊂ Rb of N . Let us write f = (f1, . . . , fb) : Z → N ⊂ Rb. By Proposition 2.C.5 every component fk : Z → R can be Sν approximated by a Nash function gk : Z → R. Thus the Sν map g = (g1, . . . , gk) : Z → Rb approximates f , and for a closed enough approximation, we have g(Z) ⊂ W . By Proposition 2.D.1, η ◦ g approximates f (= η ◦ f ). (cid:3) 2.E. Nash manifolds with corners. As in the case of Nash manifolds, a semialgebraic set Q ⊂ Ra is an m-dimensional Nash manifold with corners if every point x ∈ Q has an open neighborhood U of Ra equipped with a Nash diffeomorphism (u1, . . . , un) : U → Ra that maps x to the origin and with U ∩ Q = {u1 ≥ 0, . . . , ur ≥ 0, um+1 = 0, . . . , un = 0} for some 0 ≤ r ≤ m. In [FGR] several properties of affine Nash manifolds with corners are established: for instance, and as in the case of Nash manifolds, that Q can be covered with finitely many open sets of this type. The most relevant property for our purposes here is the following. Recall that the boundary of Q is ∂Q = Q \ Smooth(Q), where Smooth(Q) is the set of smooth points of Q. Fact 2.E.1. [FGR, 1.12, 6.5] Let Q ⊂ Ra be a Nash manifold of dimension m with corners. Then there exists a Nash manifold M ⊂ Ra of the same dimension m such that Q is a closed semialgebraic subset of M and the boundary ∂Q is a closed semialgebraic set whose Nash closure X in M is a normal crossings at every point and satisfies X ∩ Q = ∂Q. We will call such an M a Nash envelope of Q. Note that the boundary ∂Q defined above coincides with the topological boundary of Q in M and that the interior as manifold Int(Q) = Q\ ∂Q coincides with the topological interior of Q in M . Moreover, if M ′ ⊂ M is an open subset that contains Q, then M ′ is also a Nash envelope of Q. This is the usual procedure to obtain a Nash envelope with additional properties: to drop the closed semialgebraic set C = M\M ′. Thus, replacing M by M ′ or making M smaller amounts to drop a closed semialgebraic set C ⊂ M disjoint from Q. All in all, we see that the concept of Nash envelope works as a germ at Q. The proof of the preceding fact uses the following result, which we will also need in the sequel: Fact 2.E.2. [FGR, 1.2] Let Z ⊂ Ra be a locally compact semialgebraic set such that for each x of the germ Zx is smooth of constant dimension m. Then Z is a x ∈ Z the analytic closure Z closed subset of a Nash manifold M ⊂ Ra of dimension m. an Now we want to define Sν functions for affine Nash manifolds with corners. Since an affine Nash manifold with corners Q ⊂ Ra is locally compact, Q is closed in the open semialgebraic subset V = Ra \ (Q \ Q) of Ra. Consequently, we define Sν (Q) and its topology using the closed inclusion of Q in V , as we did in Section 2.C. Let us see that this topology does not depend on the neighborhood (notice that the argument for maps into Rb is similar). Indeed, let W ⊂ V be another open semialgebraic neighborhood of Q and consider an Sν partition of unity ϕ1, ϕ2 : V → R subordinated to the covering {V \ Q, W}. Then the maps ΨV W : Sν(V ) → Sν(W ) : f 7→ ϕ2f and ΨW V : Sν (W ) → Sν(V ) : g 7→ ϕ1 + ϕ2g are continuous; 12 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ moreover, ϕ2f = f and ϕ1 + ϕ2g = g on Q. Thus, we have the commuting diagram Sν (V ) ρV Sν (Q) ΨV W / ΨW V / Sν(W ) ρW Id / Id / Sν (Q) where ρV : Sν (V ) → Sν (Q) and ρW : Sν(W ) → Sν (Q) are the open quotient homomorphisms. The identity map from left to right is continuous if and only if Id ◦ ρV is continuous if and only if ρW ◦ ΨV W is continuous, which is true. The continuity of the identity map from right to left is similar and hence the topology does not depend on the neighborhood. On the other hand, we could define Sν(Q) and its topology via the closed inclusion of Q into any Nash envelope M . Using a Nash retraction ρ : V → M of M it follows that both definitions coincide (note that M is closed in V and therefore we can apply Proposition 2.C.3). In general, if Q ⊂ Ra is a Nash manifold with corners and M ⊂ Ra a Nash envelope of Q, then the Nash closure X of the boundary ∂Q is not a Nash normal crossing divisor, i.e., its irreducible components need not to be Nash manifolds. In [FGR, 1.12] there is a full characterization of Nash manifolds with corners for which this is true for M small enough: such a Q is called here a Nash manifold with divisorial corners. To progress further we introduce the following notion: a face of a Nash manifold with corners Q ⊂ Ra is the (topological) closure of a connected component of Smooth(∂Q); of course, ∂Q is the union of all the faces. This notion of faces are used to characterize divisorial corners. We state that characterization as it is more convenient here: Fact 2.E.3. [FGR, 1.12, 6.4, 6.5] Let Q ⊂ Ra be a Nash manifold with divisorial corners and let M ⊂ Ra be a Nash envelope of Q. Let X be the Nash closure of ∂Q in M . Then, if we make M small enough, we have: (1) X is a normal crossing divisor in M . (2) Let D1, . . . , Dr be the distinct faces of Q and let X1, . . . , Xr be their Nash closure in M . Then the distinct irreducible components of X are X1, . . . , Xr. (3) All faces Di are again Nash manifolds with divisorial corners. (4) Xi is a Nash envelope of Di satisfying Xi ∩ Q = Di for 1 ≤ i ≤ r. (5) The number of faces of Q that contain a given point x ∈ ∂Q coincides with the number of connected components of the germ Smooth(∂Qx). Note that (3) and (5) are intrinsic and do not depend on M . We finish this section with an iterated construction of faces that fits properly in the framework of this paper. 2.F. Iterated faces of Nash manifolds with divisorial corners. Let Q be a connected Nash manifold with divisorial corners. By Fact 2.E.3(3) the iterated faces are manifolds with divisorial corners. We start with one single m-face : Q itself. Then by descending induction, for d < m a d-face is a face of a (d+1)-face. Clearly, a d-face has dimension d. The induction ends at some d = m0 ≥ 0. In particular, the m0-faces are Nash manifolds. These data are the same for all Nash envelopes M . Next, for a given Nash envelope M of Q, we will consider the Nash closures of the iterated faces. Notice how these Nash closures vary when shrinking M to M ′ ⊂ M containing Q: if Z is the Nash closure in M of a d-face D then the Nash closure Z ′ of D in M ′ is the irreducible component of Z ∩ M ′ containing D.   o o   o o APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 13 Moreover, we can apply Fact 2.E.3 in each step of the construction and therefore: Lemma 2.F.1. For M small enough we can assume that if D is a d-face then its Nash closure Z in M is a Nash envelope of D such that the Nash closure of ∂D is a normal crossing divisor in Z and Z ∩ Q = D. Proof. Indeed, note first that if it holds for some d-face D and we shrink M to M ′ ⊃ Q then D retains the property. This is a consequence of the way Nash closures vary when we shrink M . Now we argue by descending induction on d. For d = m the statement reduces to Fact 2.E.3(1). Now, let D be a d-face, d < m, which is a face of a (d + 1)-face D′. By induction the Nash closure Z ′ of D′ in M ′ is a Nash envelope of D′ satisfying Z ′ ∩ Q = D′. We apply again 2.E.3(4) to D′ ⊂ Z ′ and find a closed semialgebraic set C ′ ⊂ Z ′ \ D′ such that the Nash closure Z of D in Z ′ \ C ′ is a Nash envelope of D with Z ∩ D′ = D. Moreover, since Z ′ ∩ Q = D′ we have that C ′ ∩ Q = ∅ and therefore we can replace M by M ′ = M \ C ′. The Nash closure of D in M ′ is the same Z and D ⊂ Z ∩ Q ⊂ Z ∩ Z ′ ∩ Q ⊂ Z ∩ D′ = D, so that Z ∩ Q = D. Finally, we apply 2.E.3(1) to D ⊂ Z and we get a closed semialgebraic set C ⊂ Z \ D such that the Nash closure of ∂D in Z \ C is a normal crossing divisor. Since Z ∩ Q = D we have that C ∩ Q = ∅ and therefore we can replace M ′ by M ′′ = M ′ \ C. (cid:3) Let us look at this construction locally, that is, M = Rm and Q = {x1 ≥ 0, . . . , xs ≥ 0} with 0 ≤ s ≤ m − m0. In this case the d-faces are {xi1 = . . . = xim−d = 0} ∩ Q for 1 ≤ i1, . . . , im−d ≤ s and the Nash closures of such a face is {xi1 = ··· = xim−d = 0}. We see that any intersection of Nash closures of faces is again the Nash closure of a face. Again in the general setting Q ⊂ M , let us fix x ∈ Q. We can make the same construction of iterated faces for the germ Qx, which can be clearly described in the local model. More relevant is the following: Lemma 2.F.2. The distinct (Nash closures of the) d-faces of the germ Qx are the germs at x of the (resp. Nash closures of the) d-faces of Q. Proof. We argue by descending induction on d. For d = m it is obvious, so suppose d < m. By x of the (d + 1)-faces D′ of Q and consequently induction the (d + 1)-faces of Qx are the germs D′ the d-faces of Qx are the faces of those D′ x. Let D1, . . . , Ds be the faces of a fixed (d + 1)-face D′. We claim that the germs D1,x, . . . , Ds,x are the faces of the germ D′ x. It is clear that each Di,x is a union of faces of D′ x has exactly s faces. This is Fact 2.E.3(5), and we are done. x, hence what we claim is that D′ Finally, let Z be the Nash closure of a d-face D. The Nash closure Yx of the germ Dx is (cid:3) included in Zx and since by 2.F.1 both are smooth of the same dimension, Yx = Zx. Proposition 2.F.3. In the setting above, if M is small enough then for any two Nash closures Z1 and Z2 of iterated faces we have that Z1 ∩ Z2 is a Nash manifold and any irreducible component of Z1 ∩ Z2 meeting Q is again the Nash closure of some iterated face. Proof. Consider Nash closures Z1 and Z2 of iterated faces D1 and D2. By 2.F.2, for every x ∈ Q we have that Z1,x and Z2,x are Nash closures of iterated faces of Qx and therefore their intersection Z1,x∩ Z2,x = (Z1∩ Z2)x is the Nash closure of a face of Qx. In particular, for every x ∈ Q we have that (Z1 ∩ Z2)x is smooth. Thus, the closed semialgebraic set C = (Z1 ∩ Z2) \ Smooth(Z1 ∩ Z2) does not meet Q and we shrink M to M ′ = M \ C. Let Z ′ 2 be the Nash closures of D1 and D2 in M ′. For all x ∈ Z ′ 2,x = Z1,x ∩ Z2,x = (Z1 ∩ Z2)x is smooth. 2, the germ Z ′ 1,x ∩ Z ′ 1 and Z ′ 1 ∩ Z ′ 14 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ Now, let Y be a connected component of Z1 ∩ Z2 such that there is x ∈ Y ∩ Q. Since (Z1∩ Z2)x is smooth we have that Yx = (Z1 ∩ Z2)x. As before, Z1,x ∩ Z2,x = (Z1 ∩ Z2)x is the Nash closure of a face of Qx and therefore, by 2.F.2, we have that (Z1 ∩ Z2)x = Zx where Z is the Nash closure of an iterated face. In particular, since Yx = (Z1 ∩ Z2)x = Zx and both Y and Z are connected Nash manifolds we deduce Y = Z. (cid:3) 3. Monomial singularity types As said in the Introduction a set X ⊂ M has a monomial singularity at x ∈ X if the point has a neighborhood U in M equipped with a Nash diffeomorphism u = (u1, . . . , um) : U → Rm such that u(x) = 0 and X ∩ U =[λ∈Λ{uλ = 0}, where {uλ = 0} = {uℓ1 = ··· = uℓr = 0}, for a certain type Λ (see Definition 1.1), that is, the germ Xx is a monomial singularity of type Λ. Note that any u as above maps each irreducible component Xi of Xx onto some coordinate linear variety Lλ(i) = {uλ(i) = 0}, and so each Xi is non-singular, as well as any intersection Xi1 ∩ ··· ∩ Xip. Furthermore, the derivative dxu : TxM → Rm maps the tangent space TxXi onto that of Lλ(i), which is the same Lλ(i). Thus, v = (dxu)−1 ◦ u : U → Ra is a diffeomorphism onto its image that maps Xx onto its tangent cone, that is, the union of the tangent spaces of its irreducible components; henceforth, by abuse of notation, tangent cone will also mean the collection of those tangent spaces. In case X ∩ U = {uℓ1 = 0} ∪ ··· ∪ {uℓr = 0}, 1 ≤ ℓ1, . . . , ℓr ≤ m, we have a Nash normal crossings [FGR]. After the obvious linear change of coordinates we can assume that X ∩ U = {u1 = 0} ∪ ··· ∪ {ur = 0}. That is, in the context of Nash normal crossings, the number of hyperplanes determines the type up to linear isomorphism. For monomial singularities the characterization of the type is far more involved. Our first aim will be to understand when a family of linear varieties of Rm is linearly isomorphic to a family of coordinate linear varieties. To that end, let L = {L1, . . . , Ls} be a family of linear varieties of Rm. Henceforth, every time we consider a family of linear varieties we assume there are no immersions. For each subset I ⊂ {1, . . . , s} and each 1 ≤ p ≤ s we denote and L(p) = X#I=p LI = \j∈I Lj LI . We set L(s+1) = {0}. For each I ⊂ {1, . . . , s} with #I = p we also define VI = L(p+1) ∩ LI and we denote with WI any supplement of VI in LI . We will use this notation consistently in all what follows. The following equations hold: (3.1) L(p) = L(p+1) +P#I=p WI =Ps k=pP#J=k WJ . (3.2) P#J=p, J6=I LJ ∩ LI ⊂ VI and LI ⊃PJ⊃I WJ . (3.3) dim(WI ) = dim(LI ) − dim(VI ) ≤ dim(LI ) − dim(cid:16)P#J=p (3.4) dim(L(p)) ≤ dim(L(p+1)) +P#I=p(cid:16) dim(LI ) − dim(cid:16)P#J=p Indeed, 3.2 and 3.3 are obvious. To prove 3.1 we pick u ∈ L(p) and write u =P#I=p uI where uI ∈ LI. Next, write uI = vI + wI where vI ∈ VI and wI ∈ WI . Since P#I=p VI ⊂ L(p+1) we LJ ∩ LI(cid:17). LJ ∩ LI(cid:17)(cid:17). J6=I J6=I APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 15 have u = X#I=p vI + X#I=p wI ∈ L(p+1) + X#I=p WI The last inequality 3.4 gives way to the following notion. and so L(p) = L(p+1) +P#I=p WI, as required. Finally 3.4 follows from 3.2 and 3.3. Definition 3.5. We say that L is an extremal family if the inequality 3.4 is an equality for all p = 1, . . . , s. J6=I LJ ∩ LI . k=p L#J=k WJ . Clearly, this notion does not depend on the ordering the varieties are listed in the family L. Lemma 3.6. Suppose L extremal. Then for p = 1, . . . , s and I ⊂{1, . . . , s} with #I = p we have: (1) L(p) = L(p+1) ⊕L#I=p WI =Ls (2) VI =P#J=p (3) LI =LJ⊃I WJ . In particular, L(1) =LJ WJ and if we choose a basis BJ for each WJ , then any extension of the union of the BJ 's to a basis B of Rm satisfies that B ∩ LI is a basis of LI for every I. Proof. (1) By 3.1 we have L(p) = L(p+1) +P#I=p WI and therefore it is enough to check that dim(L(p)) = dim(L(p+1)) +P#I=p dim(WI ). Now, by 3.2 and 3.3 and since the family is extremal dim(L(p)) ≤ dim(L(p+1)) + X#I=p dim(WI ) ≤ we have dim(L(p+1)) + X#I=p(cid:16) dim(LI ) − dim(cid:16) X#J=p, J6=I LJ ∩ LI(cid:17)(cid:17) = dim(L(p)). so that all the inequalities are equalities, as required. (2) Since P#J=p,J6=I LJ ∩ LI ⊂ VI it is enough to check dimensions coincide. By 3.1 and the last equation in the proof of (1) we deduce dim(WI ) = dim(LI ) − dim(cid:16)P#J=p LJ ∩ LI(cid:17) and so LJ ∩ LI(cid:17). dim(VI ) = dim(LI ) − dim(WI ) = dim(cid:16) X#J=p, J6=I (3) In view of (1) it suffices to show that LI =PJ⊃I WJ . We proceed by descending induction on #I, being the first step I = {1, . . . , s} obvious. Assume the result true for #I > p and let us check it for #I = p. By definition LI = VI ⊕ WI and by (2) we have J6=I VI = X#J=p, J6=I LJ ∩ LI For each J 6= I with #J = p we have #(J ∪ I) > p and by induction hypothesis LJ ∩ LI = LJ∪I =PK⊃J∪I WK and so LI = WI + VI = WI + X#J=p, J6=I LJ ∩ LI = WI + X#J=p, J6=I XK⊃J∪I WK = XJ ′⊃I WJ ′, as claimed. For the final assertion of the statement, note that L(1) =LJ WJ follows straightforward from (1), and hence we can indeed obtain B as explained. Furthermore, by (3), B ∩ LI is the union of all BJ with J ⊃ I and a basis of LI . (cid:3) 16 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ We will prove that extremal families are linear isomorphic to families of coordinate linear varieties. To that aim we introduce the following definition. Definition 3.7. Let L = {L1, . . . , Ls} be a family of linear varieties of Rm. We say that a basis B of Rm is adapted to L if for all i = 1, . . . , s the intersection B ∩ Li is a basis of Li. For example, for any family of coordinate linear varieties of Rm, the standard basis E of Rm is an adapted basis of the family. Moreover, note that if a family L = {L1, . . . , Ls} admits an adapted basis B then any bijection B ↔ E induces a linear isomorphism f : Rm → Rm such that f (Li) is a coordinate linear variety of Rm. Thus, the following is the result we were interested in: Proposition 3.8. Let L be a family of linear varieties of Rm. Then L is a extremal family of Rm if and only if it there is a basis of Rm adapted to L. Remark 3.9. If B is an adapted basis of L = {L1, . . . , Ls} then (i) B ∩ LI is a basis of LI for all I ⊂ {1, . . . , s}, and (ii) (Pℓ i=1(LJi ∩ LI ) for all J1, . . . , Jℓ, I ⊂ {1, . . . , s}. Indeed, it is enough to notice that if G1,G2 ⊂ B then L[G1]∩L[G2] = L[G1∩G2] and L[G1]+L[G2] = L[G1 ∪ G2]. Proof of Proposition 3.8. We proved one direction in Lemma 3.6. Now, assume that L admits an adapted basis B and denote Bi = B ∩ Li the basis of Li for i = 1, . . . , s. Let us check that L is extremal. Recall that L(p) = L(p+1) +P#I=p WI (see 3.1) and let us check that the previous sum is direct. It is enough to see that vector 0 only admits the trivial representation as a sum of vectors of the linear varieties L(p+1) and WI where #I = p. Indeed, write with up+1 ∈ L(p+1), wI ∈ WI . i=1 LJi) ∩ LI = Pℓ Then for any I with #I = p we have wI 0 = up+1 + X#I=p −wI = up+1 + X#J=p, J6=I WJ(cid:17)∩WI ⊂(cid:16)L(p+1)+ X#J=p, J6=I WJ(cid:17) ∩ WI . wJ ∈ (cid:16)L(p+1) + X#J=p, J6=I LJ(cid:17)∩LI = L(p+1)∩LI + X#J=p, J6=I By Remark 3.9, (cid:16)L(p+1)+ X#J=p, J6=I so that −wI ∈ (L(p+1) ∩ LI) ∩ WI = VI ∩ WI = {0} and so also up+1 = 0, as required. particular, we deduce that LJ∩LI ⊂ L(p+1)∩LI In Thus, since dim(WI ) = dim(LI ) − dim(VI ), to prove that L is extremal it is enough to show that dim(WI ). dim(L(p)) = dim(L(p+1)) + X#I=p dim(VI ) = dim(cid:16) X#J=p, J6=I LK(cid:17) ∩ LI = X#K=p+1 LJ ∩ LI(cid:17). But since L admits an adapted basis we have (see Remark 3.9) VI = L(p+1) ∩ LI =(cid:16) X#K=p+1 LK ∩ LI ⊂ X#J=p, J6=I LJ ∩ LI ⊂ L(p+1) ∩ LI = VI , as required. Remark 3.10. Let L be an extremal family of linear varieties (without immersions). Once we know it is, up to linear isomorphism, a family of coordinate linear varieties, we can bound the Indeed, by associating to every coordinate variety L ⊂ Rm the set number of varieties in L. {xi1, . . . , xir} of the variables appearing in the equations of L we define a bijection from the set of all coordinate linear varieties in Rm onto the set of all subsets of {x1, . . . , xm}. Clearly, this bijection reverses inclusions, hence transforms L in an Sperner family of a finite set of m elements. (cid:3) APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 17 Now, it is a beautiful result, the Sperner Theorem [L], that such a family has at most (cid:0) m [m/2](cid:1) elements. Thus #(L) ≤(cid:18) m [m/2](cid:19). This is behind the value q in Theorem 1.7 (see the final step in its proof 8.4). Now, we need to determine whether two extremal families are linearly isomorphic. We introduce the following general definition. Definition 3.11. Let L = {L1, . . . , Ls} be a family of linear varieties of Rm. As before, we denote for each subset I ⊂ {1, . . . , s} the intersection LI = Tj∈I Lj. Next, to each family of different nonempty subsets I1, . . . , Ir ⊂ {1, . . . , s}, r ≥ 1, we associate the number The collection of all the previous dimensions will be called the load of L. dim(LI1 + ··· + LIr ). 1, . . . , L′ Note that this notion of load does depend on the ordering the varieties are listed in the family L. Also note that the combinatorial information to determine if a family L is extremal is contained in the load of L. Proposition 3.12. Let L = {L1, . . . , Ls} and L′ = {L′ Then, there is a linear isomorphism f of Rm such that f (Li) = L′ of L and L′ coincide. If this is the case, we say that the families L and L′ are equivalent. Proof. Assume that the loads of L and L′ coincide. By Lemma 3.6 we have that L(1) =LJ WJ and that to construct a basis B of Rm adapted to L it is enough to choose a basis BI of WI for each I ⊂ {1, . . . , s} and then to extend the union of all BI 's to obtain B. Similarly, we construct a basis B′ of Rm adapted to L′ that extends a union of the bases B′ I 's. Since the loads of L and L′ coincide, BI and B′ I have the same number of elements for each I ⊂ {1, . . . , s}. I induce another B ↔ B′. Hence, there is a linear isomorphism f of Thus, bijections BI ↔ B′ I for all I. Since again by Lemma 3.6 we know that Li = Li∈I WI Rm that maps WI onto W ′ and L′ I , we are done. The converse implication is clear since isomorphisms preserve dimensions. (cid:3) s} be two extremal families of Rm. i for all i if and only if the loads i = Li∈I W ′ I 's of the W ′ Remarks 3.13. (1) We have a necessary condition for a union of non-singular germs to be a monomial singularity: its tangent cone must be extremal (for a full characterization, just involving arithmetic conditions, see Corollary 5.6). Note that given a collection of linear varieties, the mere list of their dimensions and the dimensions of their intersections is not enough to determine if they are a monomial singularity: for instance, three lines through the origin in R3 are a monomial singularity if and only if they generate R3, hence we cannot drop the dimension of the sum of the lines. (2) Since the equivalence of extremal families of linear varieties is determined by their loads, it is an arithmetic relation as mentioned in the Introduction. (3) Let Xx and X ′ x be two monomial singularities. Then, Xx and X ′ x are Nash isomorphic if and only if their types Λ and Λ′ are equivalent if and only if their tangent cones have up to reordering the same load. Thus, in the end, the type of a monomial singularity is characterized arithmetically by the load of the tangent cone. 4. Nash monomial singularity germs Here we will prove the semialgebraicity result stated in Proposition 1.2. But previously we must analyse the ideals of monomial singularities. 18 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ 4.A. Square-free monomial ideals. In this section D will either denote one of the domains: (i) N(Rm) of global Nash functions on Rm, (ii) Nm of Nash function germs at the origin in Rm and (iii) Om of analytic function germs at the origin in Rm. A square-free monomial ideal of D is an ideal generated by monomials xσ = xσ1 m with exponents σi = 0 or 1. Our aim here is to show that: the ideals of unions of coordinate linear varieties are exactly the square-free monomial ideals. To that end, it will be useful the following lemma. Lemma 4.A.1. Let I ⊂ D be a proper ideal that admits a system of generators f1, . . . , fℓ that do not depend on x1. Then, x1 is a non zero divisor mod I. 1 ··· xσm Proof. We have to show that if x1g ∈ I for some g ∈ D, then g ∈ I. But, if x1g = h1f1 + ··· + hℓfℓ for some hi ∈ D, we can write hi = x1qi + ri, where ri = hi(0, x2, . . . , xm) does not depend on x1, and qi = hi − ri x1 ∈ D. Therefore, x1g = x1(q1f1 + . . . + qℓfℓ) + (r1f1 + . . . + rℓfℓ). If we make x1 = 0, and since ri, fi do not depend on x1, we deduce that r1f1 + . . . + rℓfℓ = 0 and hence x1g = x1(q1f1 +. . .+qℓfℓ). As D is a domain we conclude that g = q1f1 +. . .+qℓfℓ ∈ I, as desired. Definition 4.A.2. Let X = L1 ∪ ··· ∪ Ls be a union of coordinate linear varieties of Rm. We say that a xi is a variable of Lj if xi = 0 is one of the equations of Lj. Now, we first consider the collection of all monomials xj1 ··· xjs such that xji is a variable of Li for each i ∈ {1, . . . , s}; then, if a variable of xj1 ··· xjs appears several times (because it comes from several Li's) we just take it once. The resulting monomials are the associated square-free monomials of X. Proposition 4.A.3. Let X = L1 ∪ ··· ∪ Ls be a union of coordinate linear varieties of Rm. Then its associated square-free monomials generate: (i) the ideal I(X) of global Nash functions vanishing in X, (ii) the ideal I(X0) of Nash function germs vanishing on the germ at the origin X0 and (iii) the ideal IO(X0) of analytic function germs vanishing on X0. (cid:3) Proof. All cases are the same, so we write down (i). We first show that X is the zero set of the system defined by its associated square-free monomials. Indeed, if Li has equations xi1 = ··· = xir = 0 we can use instead x2 i1 + ··· + x2 ir = 0, so that X is given by 0 =Yi (x2 i1 + ··· + x2 xiℓ)2, ir ) =Xℓ (Yi and so the equations Qixiℓ = 0 define X. Of course, we can eliminate repetitions of variables in Qi xiℓ to get the equations given by the associated square-free monomials. By the real Nullstellensatz it remains to show that every square-free monomial ideal I is real. To that end, it is enough to prove that: A square-free monomial ideal I is an intersection of prime ideals generated by subsets of vari- ables appearing in the given generators of I. We argue by induction on the total number of times the variables occur in the given generators following [D]. For instance, in (x1x2, x1x3) there are 4 occurrences. In the induction step we will use Lemma 4.A.1. For one occurrence there is nothing to prove, and the same happens for the more general case when each generator is one single variable. Suppose now some variable appears in all monomials, say I = (x1f1, . . . , x1fp) where the fk's are square-free monomials without x1. For the induction we only need to show I = (x1) ∩ (f1, . . . , fp). Here, the inclusion left to right is clear. For the other suppose x1h ∈ (f1, . . . , fp). By Lemma 4.A.1, the variable x1 is not a APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 19 zero-divisor mod (f1, . . . , fp), hence this ideal contains h and consequently x1h ∈ I. Thus we can assume that some generator has at least two variables and that no variable appears in all of them; say I = (x1f1, . . . , x1fp, g1, . . . , gq), f1 6= 1 and no gj contains x1. We claim that I = (x1, g1, . . . , gq) ∩ (f1, f2, . . . , fp, g1, . . . , gq), which gives way to induction again. The inclusion to prove is right to left, which reduces to: if x1h ∈ (f1, f2, . . . , fp, g1, . . . , gq) then x1h ∈ I. But, by Lemma 4.A.1, x1 is not a zero-divisor mod (f1, f2, . . . , fp, g1, . . . , gq), and consequently h ∈ (f1, f2, . . . , fp, g1, . . . , gq) so that x1h ∈ I, as claimed. (cid:3) From this result we directly obtain: Corollary 4.A.4. A Nash germ Xx is a monomial singularity if and only if the ideal I(Xx) ⊂ NM,x is generated by square-free monomials on some local coordinates at x. After this discussion of square-free monomials ideals we can turn to: 4.B. Semialgebraicity of monomial singularities loci. We know this to be true for normal crossings [FGR, 1.5], and we are to generalize the argument there. But we give full details because of its technical nature. Proof of Proposition 1.2. We will prove the semialgebraicity of T (Λ) for any fixed type Λ. Con- sider the Nash ideal I = I(X) of X, which is finitely generated by some f1, . . . , fp ∈ N(M ). By Corollary 4.A.4 there are square-free monomials m(Λ) (x) ∈ Z[x1, . . . , xm] depending only on Λ such that x ∈ T (Λ) if and only if (4.B.1) There is a regular system of parameters u = (u1, . . . , um) of the local regular ring NM,x such that Xx = {f1 = 0, . . . , fp = 0}x = {m(Λ) (u) = ··· = m(Λ) (u) = 0}x. 1 (x), . . . , m(Λ) r 1 r We must show that this condition is semialgebraic. Before proceeding, we apply the Artin- Mazur Theorem [BCR, 8.4.4] to assume that M is an open subset of a nonsingular algebraic set V ⊂ Rn and f1, . . . , fp are the restrictions to M of some polynomial functions that we denote by the same letters. Let J be the ideal of V in the polynomial ring R[x] = R[x1, . . . , xn], and let b1, . . . , bq be generators of J. By [BCR, 8.7.15] the stalk NM,x at a point x ∈ M is the henselization of the localization of R[x]/J at the ideal (x − x) = (x1 − x1, . . . , xn − xn). The henselization of the local ring R[x](x−x) is R[[x − x]]alg and so NM,x = R[[x − x]]alg/Jx where Jx = JR[[x − x]]alg. Therefore, the parameters ui are the classes modulo Jx of some hi ∈ R[[x − x]]alg; let Bkx, Hix stand for the derivatives at x of the bk, hi. Suppose that condition (4.B.1) holds for a point x ∈ M . We deduce that all fj's belong to the ideal of NM,x generated by m(Λ) (u), . . . , m(Λ) r (u); hence 1 (1) There are Nash function germs gej, ajk ∈ R[[x − x]]alg such that ajkbk. m(Λ) (h1, . . . , hm)gej +Xk fj =Xe e (4.1) On the other hand, that the ui's form a regular system of parameters of NM,x just means that (2) The hi's vanish at x, and the linear forms Hix are linearly independent over R modulo the linear forms Bkx. Let us now see how these new two conditions are semialgebraic. We look at equation (4.1) as a system of polynomial equations in the unknowns hi, gej, ajk. Then we recall M. Artin's approximation theorem with bounds [Ar, 6.1]: 20 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ (4.B.2) For any integer α there exists another integer β which only depends on n, α, the degrees of the fj's, the degrees of the m(Λ) 's, the degrees of the bk's and the number of variables hi, gej, aij, such that the polynomial equations e fj =Xe m(Λ) e (h1, . . . , hm)gej +Xk ajkbk have an exact solution in the local ring R[[x − x]]alg if they have an approximate solution modulo (x − x)β; furthermore that exact solution coincides with the approximate solution till order α. Now, fix α = 2, so that the exact solution coincides with the approximate one till order 2, and define S as the set of points x ∈ M such that: (1*) There are polynomials hi, gej, ajk ∈ R[x] of degree ≤ β such that fj ≡Xe m(Λ) e (h1, . . . , hm)gej +Xk ajkbk mod (x − x)β. (2*) The polynomials hi vanish at x and their derivatives Hi,x at x are linearly independent modulo the linear forms Bkx. Notice that if the approximate solution hi verifies (2*) then the exact one hi verifies (2). Thus, if the equation (4.2) has an approximate solution hi, gej, ajk ∈ R[x] of degree ≤ β modulo (x− x)β satisfying (2*), then the equation (4.1) has an exact solution hi, gej, ajk ∈ R[[x − x]]alg satisfying (2). Since the converse implication is trivial, both assertions are equivalent. Now, the existence of approximate solutions of fixed order β (described by conditions (1*) and (2*) above) is a first order sentence, and we conclude that the set S of points x ∈ M for which conditions (1) and (2) hold true (or equivalently conditions (1*) and (2*) hold) is a semialgebraic set. (4.2) Next we analyze the exact meaning of (1) and (2); let x ∈ S. From (1) we get that {m(Λ) 1 (u) = ··· = m(Λ) r (u) = 0}x ∩ Mx ⊂ {f1 = 0, . . . , fp = 0}x ∩ Mx and therefore Xx = {f1 = 0, . . . , fp = 0}x ∩ Mx = =(cid:16){m(Λ) 1 (u) = ··· = m(Λ) r (u) = 0}x ∩ Mx(cid:17) ∪ (Y1,x ∪ ··· ∪ Ys,x), (4.3) where the Yℓ,x's are the irreducible Nash components of Xx on which some m(Λ) (u) does not vanish identically. Thus we must get rid of those Yℓ,x's. To that end we use the topology of the germ Xx. i Let us denote X ′ m(Λ) 1 (x), . . . , m(Λ) r x = Sλ∈Λ{uλ = 0}x ∩ Mx. Recall that by definition of the monomials (x) we also have x = {m(Λ) X ′ 1 (u) = ··· = m(Λ) r (u) = 0}x ∩ Mx. For every integer d denote cd the number of connected components of dimension d of the smooth locus of X ′ x; this number only depends on Λ and d. Now consider the semialgebraic set Smoothd(X) = {x ∈ X : the germ Xx is smooth of dimension d}. We know that the set Cd = {x ∈ X : the germ Smoothd(X)x has cd connected components} is semialgebraic [FGR, 4.2], and so is the intersection C = Td Cd. To conclude, we see that S ∩ C = T (Λ). Clearly S∩C contains T (Λ). For the other inclusion, pick x ∈ S∩C. We must see that the Yℓ,x's in equation (4.3) are redundant. Otherwise, let d be the biggest dimension of the non-redundant ones. We can write Smoothd(X)x =(cid:16)Smoothd(X ′ x) \[ℓ Yℓ,x(cid:17) ∪(cid:16)Smoothd(cid:16)[ℓ x(cid:17), Yℓ,x(cid:17) \ X ′ APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 21 where only non-redundant Yℓ,x's are considered. As x ∈ Cd, in the left hand side we have exactly cd connected components. In the right hand side, we have at least cd coming from the first bracket. This is so because we have cd connected components from Smoothd(X ′ x) by definition of cd, and none of them can be lost insideSℓ Yℓ,x. Indeed, if one of these connected components, say E, was contained in the union SℓYℓ,x, it would be contained in some of the Yℓ,x of dimension d (dimension cannot be bigger by construction). The connected component E is an open subset of {uλ = 0}x ∩ Mx for some λ ∈ Λ and therefore {uλ = 0}x ∩ Mx = Yℓ,x because both are irreducible and have the same dimension. In particular Yℓ,x is redundant, a contradiction. Now, looking at the second bracket, if Yℓ,x is not redundant of dimension d, the germ is not empty, and adds some connected component, which is impossible. This contradiction completes the argument. (cid:3) Smoothd(Yℓ,x) \ X ′ x 4.C. Nash functions on Nash monomial singularity germs. In Definition 1.5 we intro- duced Nash and c-Nash functions. We now prove that both concepts coincide for Nash monomial singularity germs. This will allow in the next section to show the corresponding result for Nash sets with monomial singularities (see Theorem 1.6). Proposition 4.C.1. Let X = L1 ∪ ··· ∪ Ls be a union of coordinate linear varieties in Rm, and let h : X → R be a semialgebraic cSν (resp. cN) function. For every non-empty set of indices I we consider the intersection LI =Ti∈I Li and the orthogonal projection πI : Rm → LI . Then the function H =XI (−1)#I+1h ◦ πI . is a well defined Sν (resp. Nash) extension of h to Rm. In particular, Sν(X) ≡ cSν(X) and N(X) ≡ cN(X). Proof. We write the proof for the differentiable case, as the Nash one is a copy. First note that every LI is contained in some Li, hence h ◦ πI = (hLi) ◦ πI is a composition of semialgebraic Sν functions, hence a semialgebraic Sν function. Consequently, H is a sum of semialgebraic Sν functions, and so a semialgebraic Sν function. Thus the thing to check is that H extends h, which we prove by induction on s. For s = 1, we just have H = h ◦ π1, where π1 is the orthogonal projection onto X = L1. Thus, π1X = IdX and HX = h. Now suppose s > 1 and the result true for s − 1 coordinate linear varieties. Denote Y = L1 ∪ ··· ∪ Ls−1, and G =Xs /∈J (−1)#(J)+1h ◦ πJ , that is, J runs among the non-empty sets of indices that do not contain s. The induction hypothesis says that GY = hY , that is, h − G vanishes on Y . Let us deduce from this that HX = h. Denote πs : Rm → Ls the orthogonal projection onto Ls. Since the Li's are coordinate linear varieties, πs(Li) = Li ∩ Ls ⊂ Li, which implies that πs(Y ) ⊂ Y and that (h − G) ◦ πs vanishes on Y as h − G does. On the other hand, (h − G) ◦ πs and h − G coincide on Ls, because πs is the identity on Ls. Thus (h − G) − (h − G) ◦ πs vanishes on X, that is, h = G + (h − G) ◦ πs on X. But: (h − G) ◦ πs = h ◦ πs −Xs /∈J = h ◦ πs −Xs /∈J =Xs∈K (−1)#(K)+1h ◦ πK, (−1)#(J)+1h ◦ πJ ◦ πs (−1)#(J)+1h ◦ πJ∪{s} = h ◦ πs +Xs /∈J (−1)#(J)+2h ◦ πJ∪{s} 22 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ that is, K runs among the non-empty subsets of indices that do include s. Hence on X we have that h =Xs /∈J as required. (−1)#(J)+1h ◦ πJ +Xs∈K (−1)#(K)+1h ◦ πK =XI (−1)#(I)+1h ◦ πI = H, (cid:3) We remark that Proposition 4.C.1 defines a extension linear map h 7→ H Sν(X) ≡ cSν (X) → Sν (Rm), N(X) ≡ cN(X) → N(Rm) for the local models X = L1 ∪ ··· ∪ Ls. Note however that the above extensions disregard topologies and therefore we must return to this identification later (see Lemma 6.1). In any case, we deduce the following. Proposition 4.C.2. Let X ⊂ M be a set such that Xx is a Nash monomial singularity germ. Then the inclusions N(Xx) → cN(Xx) and S(Xx) → cS(Xx) are bijective. Proof. We prove the Nash case, the Sν one is similar. It is enough to prove that the inclusion is surjective. Fix a semialgebraic open neighborhood U of x ∈ X and u : U → Rm a Nash diffeomorphism such that u(X ∩ U ) is a union of coordinates linear varieties L1, . . . , Ls. Any fx ∈ cN(Xx) is represented by a semialgebraic map f : V ∩ X → R where V is an open semialgebraic subset of U such that u(V ) is a ball B = B(0, ε) centered at the origin. Consider the Nash x√ε2−kxk2 , which satisfies ψ(Li ∩ B) = Li. The map f ◦ u−1 ◦ diffeomorphism ψ : B → Rm, x 7→ ψ−1 : L1 ∪ ··· ∪ Ls → R is a c-Nash function and by Proposition 4.C.1 it has a Nash extension H : Rm → R. Hence H ◦ ψ ◦ u : V → R is a Nash extension of f . (cid:3) Even though we will not use it until Section 9 let us write down here the following consequence of Proposition 4.C.1. Corollary 4.C.3. Let X = {x1 ··· xs = 0} ⊂ Rm and Q = {x1 ≥ 0, . . . , xs ≥ 0} ⊂ Rm with 1 ≤ s ≤ m. Let f : Q → R and let g : X → R be Sν functions that coincide on Q∩ X. Then, they define a Sν function on Q ∪ X, that is, there is an Sν function ζ : Rm → R such that ζQ = f and ζX = g. Proof. Consider Sν extensions F, G : Rm → R of f, g. Then h = F − GX vanishes on Q∩ X. Let H : Rm → R be the Sν extension of h given by Proposition 4.C.1. If x ∈ Q and I ⊂ {1, . . . , s} is nonempty then πI (x) ∈ Q ∩ X. Thus, by definition HQ = 0 and ζ = F − H solves the problem. (cid:3) Remark 4.C.4. Observe that for the Nash case, the previous result is trivially true. Given f and g Nash, the unique Nash extension F to Rm of f coincides, by the Identity Principle, with g on X. 5. Nash sets with monomial singularities Recall from the Introduction that a Nash set X of M is a Nash set with monomial singularities if Xx is a monomial singularity for every x ∈ X (Definition 1.3). In this section we prove the finiteness and weak normality (Theorems 1.4 and 1.6). First let us state a fundamental fact concerning Nash sets with monomial singularities. Lemma 5.1. Let X ⊂ M be a closed semialgebraic set whose germs Xx, x ∈ X, are all monomial singularities. Then X is a coherent Nash set. In particular, its irreducible components X1, . . . , Xs are pure dimensional and any union of intersections of them is also a Nash set with monomial singularities. APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 23 Proof. By definition X is locally the zero set of analytic functions. It is coherent in the analytic sense because it is locally a finite union of analytic manifolds [C, Prop. 13]. Therefore by Fact 2.B.1 it is a global analytic set, hence a Nash set by Fact 2.A.1. Furthermore, by Proposition 2.B.3 it follows that X is also coherent in the Nash sense. Let X1, . . . , Xs be its irreducible components. By Facts 2.B.1 and Proposition 2.B.4 they are coherent and pure dimensional. Moreover, Xi,x is the union of some irreducible components of Xx for any x ∈ Xi ⊂ X (Proposition 2.B.4). Therefore each Xi is a Nash set with monomial singularities. Finally, that any union of intersections of irreducible components is a Nash set with monomial singularities is a local problem, hence reduces to observe that any intersection of coordinate linear varieties is again a coordinate linear variety and any union of coordinate linear varieties is a Nash set with monomial singularities. (cid:3) We now prove weak normality making use of the notation and concepts introduced in 2.B. The corresponding result for Sν and cSν functions is also true with the corresponding topologies (Proposition 6.2), but we must postpone it until finiteness has been proved. (5.2) Proof of Theorem 1.6. Let J be the sheaf of N-ideals given by Jx = I(X)Nx. Let X1, . . . , Xs be the irreducible components of X. For 1 ≤ i, j ≤ s denote Xij = Xi ∩ Xj and let Ji and Jij be respectively the sheaves of N-ideals given by Ji,x = I(Xi)Nx and Jij,x = I(Xij )Nx. By Lemma 5.1 all the Nash sets X, Xi and Xij are coherent and therefore we have that (N/J)x = Nx/I(Xx) = N(Xx), (N/Ji)x = Nx/I(Xi,x) = N(Xi,x) and (N/Jij )x = Nx/I(Xij,x) = N(Xij,x). In 2.B we showed that N(X) can be naturally identified with the global sections of N/J and cN(X) with the global sections of the kernel of the sheaf morphism φ : N/J1 ×···× N/Js →Qi<j N/Jij which by coherence at each stalk is φx : N(X1,x) × ··· × N(Xs,x) →Yi<j N(Xij,x) : (f1, . . . , fs) 7→ (fiXij − fjXij )i<j for every x ∈ M . In particular, ker(φ)x = ker(φx) = {(f1, . . . , fs) : fiXij − fjXij = 0, i < j}. Consider the multiple restriction monomorphism i : N/J → ker(φ) ⊂ N/J1 × ··· × N/Js given at the level of stalks by ix : N(Xx) → ker(φ)x ⊂ N(X1,x) × ··· × N(Xs,x) : f 7→ (fX1,x , . . . , fXs,x). We prove that each ix is actually surjective and therefore i : N/J → ker(φ) induces a surjection on global sections N(X) → cN(X), as required. To prove the surjectivity of ix consider the commutative diagram N(Xx) ix / / ker(φ)x s s s s s s s s s ys cN(Xx) where the vertical arrow is the inclusion and ker(φ)x → cN(Xx) is the injective morphism that maps (f1,··· , fs) to the function germ in cN(Xx) that equals fi on Xi,x. By Proposition 4.C.2 the vertical arrow is an isomorphism and therefore all maps are isomorphisms, and we are done.(cid:3) We will need the following notion in the proof of the finiteness property (Theorem 1.4). Definition 5.3. Let X ⊂ M be a Nash set with monomial singularities of pure dimension d. Then the type Λ of X at any point x ∈ X must consist of sets λ = {ℓ1, . . . , ℓr} of cardinal r = dim(M ) − d; the number of those sets is the multiplicity of X at x and will be denoted mult(X, x). Note this coincides with the usual multiplicity of the local ring O/IO(Xx). For, IO(Xx) is clearly the intersection of mult(X, x) prime ideals of height dim(M ) − d and multiplicity 1.   y 24 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ We start the proof of Theorem 1.4 with a geometric preamble. Proposition 5.4. Let X ⊂ M be a Nash set with monomial singularities. Then X is the union of finitely many connected Nash manifolds Σ on which the type of X is constant and each is contained in an open semialgebraic set U such that X ∩ U = Y1 ∪ ··· ∪ Ys, where the Yi's are closed Nash manifolds in U . Furthermore, Σ ⊂ Y1 ∩ ··· ∩ Ys and for each x ∈ Σ the germs Y1,x, . . . , Ys,x are the irreducible components of Xx. Proof. Let X1, . . . , Xs be the irreducible components of X, of dimensions d1 ≤ ··· ≤ ds = dim(X); the Xi's are also Nash sets with monomial singularities, and have pure dimension. Now, we choose a finite semialgebraic stratification G of M compatible with X and the Xi's. This can be done so that each stratum is locally connected at every adherent point in M of it (see Preliminaries). Furthermore, by Proposition 1.2 we can suppose that the types of X and all Xi's are constant on every stratum. For, G can be chosen compatible with the semialgebraic sets defined by all types Λ of the singularities of X and the Xi's. (5.4.1) Let Γ ∈ G be a stratum of dimension di contained in Xi. Then for each point x ∈ Γ ⊂ Xi an the Nash closure Γ x of the germ Γx is non-singular of dimension di; in fact, it is an irreducible component of Xi,x. Indeed, given x ∈ Γ and since Xi is a Nash set with monomial singularities of pure dimension di at x, there exist non-singular germs Y1,x, . . . , Yri,x, all of dimension di, such that Since all germs Y1,x, . . . , Yri,x, Γ k = 1, . . . , ri such that Γ an x = Yk,x, which implies (5.4.1). an x ⊂ Xi,x = Y1,x ∪ ··· ∪ Yri,x. Γ an x are irreducible of the same dimension, there exists an index Let us fix a stratum Σ ∈ G contained in X and let us fix an irreducible component Xi with Σ ⊂ Xi (note here that Xi contains Σ if Xi ∩ Σ 6= ∅). Since the type of each Xi is constant on Σ we have that mult(Xi, x) is a constant ri independent of x ∈ Σ . Consider We claim that GΣ,i = {Γ ∈ G : Σ ⊂ Γ ⊂ Xi}. (5.4.2) There exist strata Γi1, . . . , Γiri ∈ GΣ,i such that for every x ∈ Σ ⊂ Xi Xi,x = Γ an i1,x ∪ ··· ∪ Γ an iri,x is the decomposition of the analytic germ Xix into irreducible components. Γ is a neighborhood of Σ in Xi, so that ∆x = Xi,x for every Indeed, the union ∆ = SΓ ∈GΣ,i x ∈ Σ. Now, fix a point x0 ∈ Σ. We have [Γ ∈GΣ,i Γ an x0 = ∆ an x0 = Xi,x0 = A1,x0 ∪ ··· ∪ Ari,x0, an where A1,x0, . . . , Ari,x0 are distinct, non-singular germs at x0 of dimension di. By (5.4.1), Ak,x0 = Γ ik,x0 for some Γik ∈ GΣ,i of dimension di. As far, the strata Γi1, . . . , Γiri work only for the chosen point x0, but we see readily that they work for all points in Σ. Indeed, for any other point x ∈ Σ ⊂ Xi we have an Γ iℓ,x ⊂ Xi,x = Y1,x ∪ ··· ∪ Yri,x, where Y1,x, . . . , Yri,x are distinct, non-singular germs of dimension di at x. Thus, every Γ coincides with one of the Yk,x's, and what we must see is that Γ an an iℓ,x 6= Γ iℓ′,x for ℓ 6= ℓ′. Consider an iℓ,x APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 25 the set E defined as follows in two different ways: an an an an {x ∈ Σ : Γ iℓ,x 6= Γ iℓ′,x} = {x ∈ Σ : dim(Γ iℓ,x ∩ Γ iℓ′,x) < di}. Let us show that the first description implies that E is closed in Σ, while the second implies that E is open in Σ (there is a subtlety here since we first consider germs and then analytic closures). an Consider x ∈ Σ. Since Γ iℓ,y ⊂ Dy for y ∈ Σ∩ V x. But Γ an iℓ,x is non-singular of dimension di, it is the germ at x of a affine Nash manifold D of dimension di, hence the point x has a neighborhood V x such that Γiℓ∩V x ⊂ D, and an iℓ,y has dimension di, and Dy is irreducible of that dimension, so Γ an iℓ,y = Dy. Similarly, shrinking V x we find a affine Nash manifold D′ of dimension di such hence Γ an iℓ′,y = D′ that Γ y for y ∈ Σ ∩ V x. Thus E ∩ V x = {y ∈ Σ ∩ V x : Dy 6= D′ y} = {y ∈ Σ ∩ V x : dim(Dy ∩ D′ y) < di}. Now the fact that E is closed and open in Σ is clearer. Finally, since x0 ∈ E, the semialgebraic set E is not empty, and since Σ is connected, we conclude that E = Σ, as desired. The claim (5.4.2) is proved. To complete the argument, consider the locally compact semialgebraic set Tik = Γ ik ⊃ Σ ik,x for each x ∈ Tik, condition (5.4.1) implies that there for k = 1, . . . , ri. Since T exists a connected affine Nash manifold Sik ⊂ M of dimension di, containing Tik, hence Σ, with Sik,x = T an ik,x = Γ an an ik,x for x ∈ Σ (see Fact 2.E.2). Thus: and Xi,x = Γ an iri,x = T an i1,x ∪ ··· ∪ T an iri,x = Si1,x ∪ ··· ∪ Siri,x, an i1,x ∪ ··· ∪ Γ Xx =[i Xi,x =[i(cid:0)Si1,x ∪ ··· ∪ Siri,x(cid:1) for all x ∈ Σ. semialgebraic subset of X. Therefore, there exists an open semialgebraic neighborhood U of Σ in M such that It follows that Σ is contained in IntX(cid:0)Si(Si1 ∪ ··· ∪ Siri)(cid:1), which is an open (5.1) X ∩ U =[i(cid:0)Si1 ∪ ··· ∪ Siri(cid:1) ∩ U. Σ ⊂\i(cid:0)Si1 ∩ ··· ∩ Siri(cid:1). Moreover, since each Sik is locally closed we can assume that Sik is closed in U . Finally note that Thus, we end by renaming the Sik∩U 's as the Yj's of the statement. Concerning the last assertion there, we only need to remark that no irreducible component of a germ Xi,x is contained in one of Xj,x, because Xi and Xj are irreducible of pure dimension. (cid:3) (5.5) Proof of Theorem 1.4. By Proposition 5.4 we are reduced to the following claim. (5.5.1) Let X ⊂ M be a Nash set with monomial singularities that is a union of closed Nash manifolds Y1, . . . , Ys. Let Σ ⊂ Y1∩···∩ Ys be a connected Nash manifold such that: (i) the type of X is constant on Σ and (ii) Y1,x, . . . , Ys,x are the distinct irreducible components of the germ Xx for every x ∈ Σ. Then Σ can be covered by finitely many open semialgebraic sets U equipped with Nash diffeomorphisms u : U → Rm such that u(X ∩ U ) is a union of coordinate linear varieties. To start with, we refine the fact that the type Λ is constant on Σ. We show that the load of {TxY1, . . . , TxYs} is the same for all x ∈ Σ. Since Σ is connected it is enough to show that the load is locally constant. Fix x ∈ Σ and let u : U 7→ Rm be a local Nash diffeomorphism with u(x) = 0 and u(X ∩ U ) = Sλ∈Λ{uλ = 0}. By (ii), for U small enough, the irreducible components of X ∩ U are Y1 ∩ U, . . . , Ys ∩ U and we denote Li = u(Yi ∩ U ). For any point y ∈ Σ ∩ U ⊂ (Y1 ∩···∩ Ys)∩ U = u−1(L1 ∩···∩ Ls) we have dy(TyYi) = Li, and consequently the loads of {TyY1, . . . , TyYs} and {TxY1, . . . , TxYs} are the same. Thus, the load is locally constant in Σ and therefore constant. 26 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ After this, fix any family L = {L1, . . . , Ls} whose type is Λ and whose load is that of {TxY1, . . . , TxYs} for all x ∈ Σ. Now, X being a Nash set with monomial singularities, we can assume that every intersection Yi1 ∩ ··· ∩ Yir is an affine Nash manifold, and at every x ∈ Y1 ∩ ··· ∩ Ys we have Tx(Yi1 ∩ ··· ∩ Yir ) = TxYi1 ∩ ··· ∩ TxYir , or equivalently dimx(Yi1 ∩ ··· ∩ Yir ) = dim(Li1 ∩ ··· ∩ Lir ). Indeed, since Σ is connected we can assume that Y1, . . . , Ys are connected. Then, by the Identity Principle Y1, . . . , Ys are the irreducible components of X because by hypothesis their germs at any point x ∈ Σ are the distinct irreducible components of Xx. Moreover, again by the Identity Principle, for any x ∈ X the irreducible components of Xx are the Yi,x with x ∈ Yi. Thus, and since X has monomial singularities, for any x ∈ X the germ Yi1,x ∪ ··· ∪ Yir,x is Nash diffeomorphic to a union of germs In particular, Yi1,x ∩ ··· ∩ Yir,x is Nash diffeomorphic to a germ of a of coordinate varieties. coordinate variety and the equality of dimensions above is straightforward. After this preparation, the unionSi Li is the model to which X =Si Yi must be diffeomorphic near Σ in the local semialgebraic sense. To confirm this, we work up the levels and the corresponding levels L(p) =(cid:8)LI = \i∈I Y (p) =(cid:8)YI = \i∈I Li : #I = p(cid:9) Yi : #I = p(cid:9) for p = 1, . . . , s. For the moment being we do not look for Nash diffeomorphisms whose images are Rm, they will be just open semialgebraic subsets of Rm. We will fix this later. Step 1. At bottom level p = s we have one single intersection Y1 ∩ ··· ∩ Ys which is a Nash manifold of the same dimension that the linear coordinate variety L1∩···∩Ls. Thus, Y1∩···∩Ys, hence Σ, can be covered by finitely many open semialgebraic sets W such that the intersection Us = W ∩ Y1 ∩ ··· ∩ Ys is Nash diffeomorphic to an open semialgebraic set Ωs of L1 ∩ ··· ∩ Ls. Since there are finitely many W 's, we can suppose simply that Σ ⊂ Us and we have a Nash diffeomorphism ϕs : Ωs → Us. Denote S = ϕ−1 open semialgebraic subsets Ωp and Up of S#I=p LI and S#I=p YI respectively, with Σ ⊂ Up, and a continuous semialgebraic map ϕp : Ωp → Up which restricts to Nash diffeomorphisms LK ∩ Ωp → YK ∩ Up; here K stands for any subset of {1, . . . , s} with at least p indices. We also assume that Ωk ⊂ Ωp and ϕpΩk = ϕk for k = p + 1, . . . , s. We want an extension to a similar map ϕp−1 at level p − 1, after some suitable finite semialgebraic partition of Σ. We proceed as follows. Step 2. Now we formulate carefully the recursion procedure. Suppose we have, at level p, s (Σ). Step 3. Let YJ be an intersection of Yi's at level p−1 and consider the continuous semialgebraic restriction φJ of ϕp to LJ ∩ Ωp. Moreover, for any set I of p indices the restriction induces a Nash diffeomorphism LJ ∩ LI ∩ Ωp → YJ ∩ YI ∩ Up (note that every intersection of YJ with another YJ ′ at the same level p − 1 is included in some intersection YI at level p). By considering a finite open semialgebraic covering of YJ we can suppose it is Nash diffeomorphic to an affine space, and we pick any identification YJ ≡ Ra. Of course a = dim(LJ ). Now, for any subset I of p indices the tangent bundle of the affine Nash manifold YJ ∩ YI ⊂ Ra is generated by its global Nash sections [BCR, 12.7.10]. Since Us is contained in all those YJ ∩YI's, the sums X#I=p Tx(YJ ∩ YI ) ⊂ Ra ≡ LJ for x ∈ Us are the fibers of a Nash vector bundle T on Us also generated by its global Nash sections. The same is true for the orthogonal bundle T⊥, say it is generated by finitely many global Nash for x ∈ Σ. Next consider VJ = P#I=p LJ ∩ LI and notice that the orthogonal supplement WJ of VJ in LJ ⊂ Rm is a coordinate linear variety of dimension c generated by suitable vectors eℓ1, . . . , eℓc of the canonical basis. Let Ω′ = Ωs ⊕ WJ which is an open semialgebraic subset of the coordinate linear variety L′ = (L1 ∩ ··· ∩ Ls) ⊕ WJ . We extend ϕs as follows: ϕ∗ s(v + α1eℓ1 + ··· + αceℓc) = ϕs(v) + α1ζ1(ϕs(v)) + ··· + αcζc(ϕs(v)). This extension ϕ∗ s is defined on Ω′. Since Ω′ ∩ (LJ ∩ Ωp) = Ωs and φJΩs = ϕs, the maps ϕ∗ and φJ glue into a continuous semialgebraic map φ′ J : Ω′ ∪ (LJ ∩ Ωp) → Ra ≡ YJ . Clearly, the components of φ′ J are c-Nash functions on the Nash sets with monomial singularities Ω′∪(LJ∩Ωp). Consequently, they are Nash functions, and φ′ J has a Nash extension φJ : ΩJ → Ra ≡ YJ to some open semialgebraic neighborhood ΩJ of Ω′ ∪ (LJ ∩ Ωp) in LJ . We claim that φJ is a local diffeomorphism at every v ∈ S = ϕ−1 s (Σ). s Indeed, fix such a v ∈ S and put x = ϕs(v). Since φJ is a diffeomorphism on every LJ ∩ LI ∩ Ωp with #I = p we have dvφJ (LJ ∩ LI ) = Tx(YJ ∩ YI ) and so dvφJ (LJ ) ⊂ Ra contains the sum P#I=p Tx(YJ ∩ YI ). Moreover, for every eℓk we have φJ (v + teℓk ) = x + tζk(x), t ∈ R. Hence, dvφJ (eℓk ) = ζk(x). Consequently, Ra =X#I=p Tx(YJ ∩ YI ) + L[ζ1(x), . . . , ζc(x)] ⊂ dvφJ (LJ ), APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 27 sections ζi : Us → Ra. For every x ∈ Σ the codimension c of P#I=p Tx(YJ ∩ YI ) in TxYJ is that of P#I=p LJ ∩ LI in LJ (since the load of L = {L1, . . . , Ls} and {TxY1, . . . , TxYs} are the same). Therefore, for every x ∈ Σ, we have that c of the ζi(x)'s are independent. After considering once again a finite open covering of Σ, we can assume we have exactly c independent Nash maps ζ1, . . . , ζc : Us → Ra ≡ YJ such that ζ1(x), . . . , ζc(x) span T⊥ x for every x ∈ Σ. In other words, Ra =X#I=p Tx(YJ ∩ YI ) + L[ζ1(x), . . . , ζc(x)]. and dvφJ : LJ → Ra ≡ YJ is onto, hence a linear isomorphism (as a = dim(LJ )). Our claim is proved. Consequently, shrinking ΩJ , φJ is a local Nash diffeomorphism onto an open semialgebraic neighborhood UJ of Σ in YJ . In this situation there is a finite open semialgebraic covering {ΩJk} of ΩJ such that every restriction φJΩJ k is a Nash diffeomorphism onto an open semialgebraic set UJk of YJ (see [BCR, 9.3.9, p. 226]). As usual, this means that we can suppose φJ : ΩJ → UJ is a diffeomorphism. By construction, the semialgebraic setSJ ΩJ is a neighborhood of S inSJ LJ , hence it contains an open semialgebraic neighborhood Ωp−1. Notice that the diffeomorphisms φJ glue together to give a continuous semialgebraic extension ϕp−1 : Ωp−1 → Up−1 of ϕp that verifies all conditions required. Step 4. Thus climbing from level to level, we get a continuous semialgebraic map ϕ1 : Ω1 → U1 from an open semialgebraic neighborhood Ω1 of S in L1 ∪ ··· ∪ Ls into another U1 of Σ in X = Y1 ∪ ··· ∪ Ys, which induces Nash diffeomorphisms Li ∩ Ω1 → Yi ∩ W1. Then we apply again the same argument above to extend ϕ1 to a Nash diffeomorphism ϕ : Ω → U from an open semialgebraic neighborhood Ω of S in Rm onto one U of Σ in M . (Of course in climbing we have used many finite semialgebraic coverings and shrunk the neighborhood of Σ, so that what we really have is a finite collection of such maps whose images U cover Σ.) Final arrangement. In this situation, ϕ−1 : U → Ω is the diffeomorphism we were looking for, except that Ω need not be Rm. To amend this, notice that S = ϕ−1(Σ) ⊂ L1 ∪ ··· ∪ Ls is contained in the intersection L1∩···∩ Ls. Now this intersection can be written as the intersection 28 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ Hj1 ∩···∩ Hjq of the coordinate hyperplanes Hj that contain some Li (that is, we just collect the equations xjℓ = 0 of all Li's). This said, we know that Ω contains a smaller neighborhood of S, which is a finite union of open semialgebraic sets Nash diffeomorphic to Rm, by diffeomorphisms that preserve the coordinate hyperplanes Hjℓ, hence the coordinate linear varieties Li (see [FGR, 4.4.5]). Composing ϕ−1 with the latter we obtain the diffeomorphism u we sought. (cid:3) The preceding construction gives a characterization of monomial singularities as announced before. Corollary 5.6. Let Xx be a Nash germ whose irreducible components X1,x, . . . , Xs,x are non- singular. Then Xx is a monomial singularity if and only if the tangent cone {TxX1, . . . , TxXs} is a extremal family and dim(Xi1,x ∩ ··· ∩ Xir,x) = dim(TxXi1 ∩ ··· ∩ TxXir ) for 1 ≤ i1 < ··· < ir ≤ s. Proof. The only if part is clear. For the converse implication, we have to prove first that each intersection Xi1,x ∩ ··· ∩ Xir,x is non-singular. Once this is done, by hypothesis which enables us to repeat the preceding proof 5.5. Tx(Xi1,x ∩ ··· ∩ Xir,x) = TxXi1 ∩ ··· ∩ TxXir Using the obvious induction argument, to show that each intersection Xi1,x ∩ ··· ∩ Xir,x is x are two non-singular germs such non-singular, we are reduced to prove that: If X1,x, X2,x ⊂ Ra that dim(X1,x ∩ X2,x) = dim(TxX1,x ∩ TxX2,x), then X1,x ∩ X2,x is non-singular. Indeed, as it is well-known, the orthogonal projection πj : Ra → Lj induces a Nash diffeomor- phism between Xj,x and Lj,x. Consider next the orthogonal projections π : Ra → L = L1 + L2 and ρj : L → Lj. Since πj = ρj ◦ π it holds that X ′ j,x = π(Xj,x) ⊂ Lx is Nash diffeomorphic to Xj,x and in particular π(X1,x ∩ X2,x) is a Nash subset germ of each Xj,x Nash diffeomorphic to 1,x and X ′ X1,x ∩ X2,x; hence, dim(π(X1,x ∩ X2,x)) = dim(L1 ∩ L2). On the other hand, since X ′ in Lx are transversal, X ′ 2,x is the germ of a Nash manifold of dimension dim(L1 ∩ L2). As π(X1,x ∩ X2,x) is a Nash subset germ of X ′ 2,x of its same dimension, and the lat- ter is irreducible, we deduce that they are equal and therefore X1,x ∩ X2,x is non-singular, as required. 1,x ∩ X ′ 1,x ∩ X ′ 2,x (cid:3) We cannot drop the condition on the dimension of the intersections and the tangent spaces. In R4, let X1 = {x3 = x4 = 0}, X2 = {x2 = 0, x4 = x2 1} and let X be their union. The tangent cone of X at the origin is {x3 = 0, x4 = 0},{x2 = 0, x4 = 0} and therefore it is extremal. However, the intersection X1 ∩ X2 is the origin. 6. Extension linear maps for Nash sets with monomial singularities This section is preliminary for Nash approximation. As explained in 2.C, for any Nash set X the ring Sν (X) of Sν-functions is equipped with the topology as the quotient Sν (X) = Sν(M )/I ν (X). If X1, . . . , Xs are the irreducible components of X then cSν (X) is equipped with the topology induced by the inclusion cSν(X) → Sν(X1) × ··· × Sν(Xs) given by the multiple restriction f 7→ (fX1, . . . , fXs). Moreover, the inclusion γ : Sν(X) → cSν(X) is always continuous. Here our purpose is to prove that if X is a Nash set with monomial singularities then there exist a continuous extension linear map θ : Sν(X) → Sν(M ) and that γ is a homeomorphism (Proposition 6.2). Both statements will be deduced from the existence of a continuous extension linear map cθ : cSν(X) → Sν(M ). Recall that in Proposition 4.C.1 we already found such an extension for unions of coordinate linear varieties. However the continuity of the extension trick there may fail APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 29 because composition on the right with the orthogonal projections need not be continuous [Sh, II.1.5, p. 83]. Our purpose now is to amend this. Lemma 6.1. Let X = L1 ∪ ··· ∪ Ls be a union of coordinate linear varieties in Rm. Then there is a continuous extension linear map cθ : cSν (X) → Sν (Rm). Proof. Fix a non-empty set of indices I = {i1, . . . , ir} ⊂ {1, . . . , s} and set LI = Ti∈I Li and X I = Si∈I Li. By Proposition 4.C.1 every cSν function h : X I → R has the following Sν extension to Rm HI = X∅6=J⊂I (−1)#(I)+1h ◦ πJ . Now consider the open semialgebraic set ΩI = {x ∈ Rm : dist(x, LI ) < 1)} \[j /∈I Lj, which satisfies X I ∩ ΩI = X ∩ ΩI . We claim that the extension linear map cθI : cSν (X I ) → Sν (ΩI ) : h 7→ HIΩI is continuous. Note that this is our previous extension linear map followed by a restriction that makes it continuous. Indeed, it is enough to show that h 7→ h ◦ πJΩI is continuous. Here we consider the topology defined in cSν(X I ) as subset of Sν (Li1) × ··· × Sν (Lir ) (this is the reason to keep referring to semialgebraic cSν functions, although we already know they are all Sν functions). Clearly, it is enough to see that if all restrictions hLi, i ∈ I, are close enough to zero, then h ◦ πJΩI is arbitrarily close to zero. Thus, pick any positive continuous semialgebraic function ε : ΩI → R. We know from Lojasiewicz's inequality [BCR, 2.6.2] that there are a constant C > 0 and an integer p large enough so that 1 (C + kxk2)p < ε(x) for every x ∈ ΩI . Let x ∈ ΩI and J ⊂ I; in particular, LI ⊂ LJ . Then for the orthogonal projection πJ : Rm → LJ we have kxk2 = dist(x, LJ )2 + kπJ (x)k2 ≤ dist(x, LI )2 + kπJ (x)k2 < 1 + kπJ (x)k2 and so at a point x ∈ ΩI . Since composition with πJ is substituting zero for the coordinates in LJ , we see that h◦ πJ does not depend on those coordinates, which implies that the above partial derivative is zero whenever such a coordinate appears in the derivative. Thus we look at derivatives that do not include them. But for those, we have ∂α(h ◦ πJ ) 1 ··· ∂xαm ∂xα1 m (cid:12)(cid:12)(cid:12) (x)(cid:12)(cid:12)(cid:12) =(cid:12)(cid:12)(cid:12) ∂αh 1 ··· ∂xαm m ∂xα1 (πJ (x))(cid:12)(cid:12)(cid:12) < δ(πJ (x)) < ε(x) because πJ (x) ∈ LJ ⊂ Li for some i ∈ I. Hence cθI is continuous, as required. Finally, we glue the cθI's. Consider a semialgebraic Sν partition of unity {φ, φI : I} subordi- nated to {Rm \ X, ΩI : I}, which is an open semialgebraic covering of Rm. Define cθ : cSν(X) → Sν(Rm) : h 7→XI φI · cθI (hX I ) 1 1 (C + 1 + kπJ (x)k2)p < (C + kxk2)p < ε(x). Denote δ(x) = Sν topology. Let us check that h ◦ πJ is ε close to zero. Look at any partial derivative (C+1+kxk2)p and suppose that all restrictions hLi , i ∈ I, are δ close to zero in the 1 ∂α(h ◦ πJ ) ∂xα1 1 ··· ∂xαm m (x), where α = α1 + ··· + αm ≤ ν, 30 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ where each φI · cθI(hX I ) extends by zero off ΩI . Since φ vanishes on X,PI φI ≡ 1 on X; hence, cθ(h) is a semialgebraic Sν extension of h. Finally, cθ is continuous because so are all cθI's, as required. (cid:3) The preceding construction can be extended by finiteness to an arbitrary Nash set with mono- mial singularities. Proposition 6.2. Let X ⊂ M be a Nash set with monomial singularities. Then there are continuous linear maps θ : Sν (X) → Sν (M ) such that θ(h)X = h. Moreover, if X1, . . . , Xs are the irreducible components of X, then the multiple restriction homomorphism Sν(X) → Sν(X1)× ··· × Sν(Xs) is a closed embedding. Proof. Using a partition of unity and the preceding local result we obtain a continuous extension linear map cθ : cSν (X) → Sν (M ). Consider the commutative diagram γ−→ cSν (X) ֒→ Sν(X1) × ··· × Sν(Xs) Sν (X) ❅❅■ ρ ❅ ❅❅❘❅ θ cθ ❄ Sν (M ) where θ = cθ ◦ γ and ρ : Sν (M ) → Sν (X) is the continuous restriction epimorphism. All maps here are continuous and therefore θ = cθ ◦ γ is a continuous extension linear map. On the other hand, γ−1 = ρ◦ cθ and so it is continuous, that is, γ is a homeomorphism. In 2.C we proved that the inclusion of Sν (X) in Sν(X1) × ··· × Sν (Xs) is closed and therefore the multiple restriction is a closed embedding. (cid:3) 7. Approximation for functions We now discuss approximation for Nash sets with monomial singularities. By Proposition 2.C.5 absolute approximation is possible in a general situation. However, we are interested for the applications in a stronger relative version which in Proposition 7.6 we prove for Nash sets with monomial singularities. Let us see first some useful facts that relate the zero-ideal of a Nash set and this relative approximation. Lemma 7.1. Let X be a Nash set of a Nash manifold M ⊂ Ra of dimension m and let ν ≥ m. If I ν (X) ⊂ I(X)Sν−m(M ) then every semialgebraic Sν function F : M → R whose restriction to X is Nash can be Sν−m approximated by Nash functions H : M → R that coincide with F on X. Proof. Suppose that I ν(X) ⊂ I(X)Sν−m(M ). Since FX is Nash there is some Nash function G : M → R with GX = FX . Therefore F − G is a semialgebraic Sν function vanishing on X, so that F − G =P ψifi for some Nash functions fi ∈ I(X) and some Sν−m functions ψi. But by absolute approximation (Fact 2.C.1), we find Nash functions ϕi that are Sν−m close to ψi, and then H = G +P ϕifi is a Nash function Sν−m close to F such that HX = FX. (cid:3) Under coherence, it is enough to control ideals for a finite covering. Lemma 7.2. Let X ⊂ M be a coherent Nash set. Suppose there is a finite semialgebraic open covering X ⊂ U1 ∪ ··· ∪ Us such that I ν(X ∩ Ui) ⊂ I(X ∩ Ui)Sν−m(Ui) for 1 ≤ i ≤ s. Then I ν(X) ⊂ I(X)Sν−m(M ). Proof. Adding M \ X to the covering we can assume M =Si Ui. Let {ϕi}i be an Sν partition of unity subordinated to the Ui's; recall that {ϕi > 0} ⊂ Ui. Since X is coherent, I(X) generates I(Ui ∩ X) (see equation (2.2) after Fact 2.B.3). We also know that I(X) is finitely generated, say APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 31 by f1, . . . , fp. Now let f ∈ Sν(M ) vanish on X. It follows that fUi vanishes on Ui ∩ X and, by hypothesis, there are Nash functions gik ∈ Sν−m(Ui) such that fUi =Xk gik(fkUi). Now consider the functions ϕigik. Although defined on Ui, they vanish off {ϕi > 0} ⊂ Ui, hence can be extended by zero off Ui. Thus we have in fact ϕigik ∈ Sν−m(M ). Finally one readily checks that f =Xk(cid:0)Xi ϕigik(cid:1)fk ∈ I(X)Sν−m(M ), which concludes the proof. (cid:3) These lemmas reduce the relative approximation problem for Nash sets with monomial singu- larities to the study of the zero-ideal of a union of coordinate linear varieties. Proposition 7.3. Let X ⊂ Rm be a union of coordinate linear varieties L1, . . . , Ls and let I(X) be the ideal of Nash functions vanishing on X. Denote by xσ the square-free monomials associated to X (see Definition 4.A.2) and let #(xσ) be the number of variables in xσ. Let ν ≥ maxσ #(xσ) and let f : Rm → R be an Sν function vanishing on X. Then f =Xσ fσxσ, where fσ is an Sν−#(xσ) function for each σ. Remarks 7.4. Note that, since the monomials xσ are square-free, none has more than m vari- ables, that is #(xσ) ≤ m. Consequently, I ν (X) ⊂ I(X)Sν−m(M ). The proof of Proposition 7.3 consists of a double induction on the dimension m and the number of linear varieties involved. We will use the following lemma: Lemma 7.5. Let f be an Sν function on Rm. Write x = (x1, x′) the variables in Rm. Then there is an expansion f (x) = f1(x)x1 + f (0, x′), where f1 is an Sν−1 function. Proof. We know this from elementary Analysis with f1(x) =Z 1 0 ∂ ∂x1 f (tx1, x′)dt, but an integral can well not be semialgebraic; however, this particular one is semialgebraic. Indeed, the function g(x) = (f (x) − f (0, x′))/x1 is semialgebraic in x1 6= 0, and the graph of f1 is the closure of the graph of g. Hence f1 is in fact semialgebraic. (cid:3) If m = 1 then X = R or Proof of Proposition 7.3. We argue by induction on the dimension. X = {0}; in the first case it is obvious and in the second one it follows from Lemma 7.5 with m = 1 and f (0) = 0. Hence we assume m > 1 and the result proved for dimension m − 1. Let X ⊂ Rm be a union of coordinate linear varieties L1, . . . , Ls of Rm and let f be an Sν function vanishing on X. We argue again by induction, now on the number s of Li's. The case s = 1. We have X = L1 and can suppose the equations of L1 are x1 = ··· = xr = 0. Lemma 7.5 gives f = f1x1 + g1, where f1 is Sν−1 and g1 is Sν and does not depend on x1. Again by the lemma, g1 = f2x2 + g2, where f2 is Sν−1 and g2 is Sν and does not depend on either x1 or x2. And so on, till we have: where all fk's are Sν−1 and g does not depend on either of x1, . . . , xr. Thus, substituting 0 for these variables does not affect g, and since f vanishes on x1 = ··· = xr = 0 we get g ≡ 0. In other words, f = f1x1 + ··· + frxr and we are done. f = f1x1 + ··· + frxr + g, 32 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ Let now s > 1 and assume the result known for less than s linear varieties in Rm. We can suppose that x1 is a variable of L1 (see 4.A.2 for the precise terminology). Denote as usual x = (x1, x′) the variables in Rm. By Lemma 7.5 once again we have an Sν−1 function f1 such that f (x) = f1(x)x1 + g(x′). We claim that g vanishes on the Li's that do not contain the variable x1. Indeed for such a coordinate linear variety Li, z = (z1, z′) ∈ Li if and only if (0, z′) ∈ Li. Since f vanishes on Li we have 0 = f (0, z′) = f1(0, z′)0 + g(z′) = g(z′) = g(z), which implies that f (x) = f1(x)x1 on Li. Consequently, since fLi = 0, f1 vanishes on Li∩{x16= 0}, which is dense in Li, and so f1 vanishes on Li. Thus f1 vanishes on the union E of all Li that do not have the variable x1. This excludes L1, hence E is a union of less than s coordinate linear varieties, and we can apply the induction hypothesis on the number of coordinate varieties to the Sν−1 function f1 to get an expression f1(x) =Xτ f1τ xτ , for some Sν−1−#(xτ ) functions f1τ . Here we have ν − 1 ≥ #(xτ ). To check that, notice that in this sum, each monomial xτ contains one variable from each Li in E and all the others Li have the variable x1, so that xτ x1 is one of our generators of I(X). Since #(xτ x1) = 1 + #(xτ ), we have ν − 1 ≥ max σ #(xσ) − 1 ≥ #(xτ ). Next we turn to the other summand g(x′). If some Li is the hyperplane x1 = 0, then f vanishes on that hyperplane, and so 0 = f (0, x′) = f1(x)0 + g(x′). We conclude f (x) = f1(x)x1 =Xτ f1τ xτ x1 and the argument is complete. Hence, suppose that all Li have some variable other than x1. We consider x′ as coordinates in Rm−1 and denote L′ i ⊂ Rm−1 the coordinate linear variety with the same equations that Li, x1 excluded in case it is in Li, that is, L′ i corresponds to Li ∩ ({0} × Rm−1) after erasing the first coordinate, which is 0. Now, g(x′), as a function on i, then z = (0, z′) ∈ Li and since f vanishes on Rm−1, vanishes on L′ Li, 1 ∪ ··· ∪ L′ s. Indeed, if z′ ∈ L′ 0 = f (z) = f (0, z′) = f1(z)0 + g(z′), as claimed. Thus, we can apply to g(x′), which is Sν, the induction hypothesis on the dimension to get: g(x′) =Xσ gσ(x′)x′σ, where the gσ's are Sν−#(xσ). Here each monomial x′σ has one variable from each L′ each Li, and so it is in fact one generator of I(X). i, hence from All in all the expression f (x) = f1(x)x1 + g(x′) =Xτ f1τ xτ x1 +Xσ gσx′σ verifies all conditions required, and the proof is complete. (cid:3) As remarked before, this settles approximation for Nash sets with monomial singularities, with some restriction on differentiability classes. We state that solution in full below. APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 33 Proposition 7.6. Let X be a Nash set with monomial singularities in a Nash manifold M ⊂ Ra of dimension m. Let ν ≥ m and let F : M → R be an Sν function. Then every Nash function h : X → R which is Sν close enough to f = FX has a Nash extension H : M → R which is Sν−m close to F . In particular, an Sν function F whose restriction to X is Nash can be Sν−m approximated by Nash functions H that coincide with F on X. Proof. First we prove the particular case when FX is Nash. By Theorem 1.4, X can be covered by finitely many open semialgebraic sets Ui each equipped with a Nash diffeomorphism ui : Ui → Rm that maps X ∩ Ui onto a union of coordinate linear varieties. Hence by Lemmas 7.1 and 7.2 it is enough to prove I ν(X ∩ Ui) ⊂ I(X ∩ Ui)Sν−m(Ui) for all i, but this follows from Proposition 7.3. Now we deduce the general case. Let U ⊂ Sν−m(M ) be an open Sν−m neighborhood of F . Then U ′ = U ∩ Sν (M ) is open in Sν(M ), and since the restriction ρ : Sν (M ) → Sν(X) is an open homomorphism, ρ(U ′) is an open Sν neighborhood of f in Sν(X). Thus, if our Nash function h is in ρ(U ′), it has a semialgebraic Sν extension G ∈ U ′ ⊂ U . By the particular case, there are Nash functions H : M → R arbitrarily Sν−m close to G with HX = h. Since U is Sν−m open, we can choose H ∈ U . (cid:3) 8. Approximation for maps The aim of this section is to prove approximation for maps instead of functions (Theorem 1.7). In 2.D we defined and equipped with a topology the spaces of Sν and cSν maps from a Nash set into a semialgebraic set. We apply extension for functions (Proposition 6.2) to prove that Sν and cSν maps coincide for Nash sets with monomial singularities. Proposition 8.1. Let X ⊂ M be a Nash set with monomial singularities whose irreducible components are X1, . . . , Xs and let T ⊂ Rb be a semialgebraic set. The multiple restriction homomorphism Sν(X, T ) → Sν (X1, T ) × ··· × Sν(Xs, T ) is a closed embedding that provides a topological identification Sν(X, T ) ≡ cSν (X, T ). Proof. Indeed, since cSν(X, Rb) = cSν (X, R)b, Proposition 6.2 gives the result for T = Rb. Then, for arbitrary T ⊂ Rb, we have the commutative diagram Sν (X, T ) −→ cSν (X, T ) ֒→ Sν(X1, T ) × ··· × Sν (Xs, T ) ❄ ❄ ❄ Sν(X, Rb) −→ cSν(X, Rb) ֒→ Sν (X1, Rb) × ··· × Sν(Xs, Rb) As the result is true for the lower row and the vertical arrows are continuous, it follows for the upper one. (cid:3) Relative approximation extends straightforwardly to maps into Rb. Moreover, we can use Nash tubular neighborhoods to show the following. Proposition 8.2. Let M ⊂ Ra and N ⊂ Rb be Nash manifolds of dimensions m and n respectively and let X ⊂ M be a Nash set with monomial singularities. Let ν ≥ m and let F : M → N be an Sν map. Then every Nash map h : X → N which is Sν close enough to f = FX has a Nash extension H : M → N which is Sν−m close to F . Proof. Consider a Nash retraction η : W → N of N in Rb (see Fact 2.A.2). By Proposition 7.6 there is a Nash extension G : M → Rb of h which is Sν−m close to F as a map into Rb. Since F (X) ⊂ N ⊂ W we can choose the approximation close enough so that G(X) ⊂ W , and then H = η ◦ G : M → N is a well defined Nash map that extends h. But composition on the left is continuous (Proposition 2.D.1) and therefore the composite H = η ◦ G is close to the composite η ◦ F = F , provided G is close enough to F . (cid:3) 34 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ The preceding extension is the key fact to obtain an approximation result for maps into Nash sets with monomial singularities. What we want is to approximate differentiable semialgebraic maps f : X → Y of Nash sets. The first restriction is that f must preserve irreducible components, that is, it must map each irreducible component Xi of X into one Yk of Y . Indeed, suppose that for each k the component Xi contains a point xk with f (xk) /∈ Yk and pick ε > 0 smaller than If f can be approximated by Nash maps, pick one g : X → Y all distances dist(f (xk), Yk). that is ε close. Since g is Nash, the inverse images g−1(Yk) are Nash sets and therefore, since the irreducible Xi is contained in their union, we deduce that g(Xi) ⊂ Yk for some k. Thus g(xk) ∈ Yk and we get dist(f (xk), Yk) ≤ dist(f (xk), g(xk)) < ε, a contradiction. To progress further we pause to look at Nash sets with monomial singularities from a global viewpoint. We recall that among local normal crossings one distinguishes the so-called normal crossing divisors by asking their irreducible components to be non-singular [FGR, 1.8]. This is a global condition that can be applied to monomial singularities. We call a Nash set with monomial singularities X a Nash monomial crossings if its irreducible components X1, . . . , Xs are all non-singular; in other words, X1, . . . , Xs are Nash manifolds. Proposition 8.3. Let X be a Nash monomial crossings whose irreducible components we denote X1, . . . , Xs. Then all intersections Xi1 ∩ ··· ∩ Xir are Nash manifolds, and any union of them is a Nash monomial crossings. Proof. By Lemma 5.1 it is enough to show that every intersection XI = Xi1 ∩···∩ Xir is an affine Nash manifold. Fix a point x ∈ XI . The distinct irreducible components of the germ Xx are the germs Xi,x such that x ∈ Xi. Indeed, if, say, Xi,x ⊂ Xj,x, then dim(Xi) = dim(Xi ∩ Xj) and since Xi is irreducible, Xi ⊂ Xj, a contradiction. Now pick any Nash coordinates u of M at x such that u(x) = 0 and Xx =Sλ∈Λ{uλ = 0}x. Observe that u maps each irreducible component Xi,x of Xx onto a coordinate linear variety Li ⊂ Rm. In particular this happens for i = i1, . . . , ir, and we deduce u(XI,x) = u(cid:0)Xi1,x ∩ ··· ∩ Xir,x(cid:1) = Li1,0 ∩ ··· ∩ Lir,0. Since any intersection of linear varieties is a linear variety, the germ XI,x is non-singular, and we are done. (cid:3) After this remark we can prove approximation for Nash maps between Nash set with monomial singularities. (8.4) Proof of Theorem 1.7. By hypothesis, f maps each irreducible component Xi of X into some irreducible component of Y ; we choose one and denote it by Yk(i). As usual, we classify the intersections Yi's into levels: at level p we have the intersections YI = Ti∈I Yi with #I = p. At bottom level s we have the intersection Y1 ∩ ··· ∩ Ys of all irreducible components. In general, at level p we have the union Y (p) = S#I=p YI of intersections YI at level p. Now, we denote XI the intersection of all Xi's with k(i) ∈ I, and say this is an intersection of Xi's at level p. Quite naturally, we denote X (p) the union of all these XI . We keep in mind this leveled collection of XI 's as an inverse image of the leveled collection of YI's. Observe that in this construction there may be empty intersections. Thus, we stop at the last non-empty X (r) =SI XI , which implies that all XI 's at this level are disjoint. After this organization of data, note that every X (p) is a Nash set with monomial singularities and the corresponding Y (p) is a Nash monomial crossings (Lemma 5.1 and Proposition 8.3). To complete the setting, note that f restricts to an Sν map f (p) : X (p) → Y (p). We are to approximate every f (p) starting at the bottom level p = r and climbing to level p = 1. In each jump of level APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 35 we will use Proposition 8.2 and therefore there will be a loss of differentiability. As f = f (1) this will complete the proof. The proof runs in several steps. Step 1. We start at level p = r. Since X (r) 6= ∅, we have the map f (r) : X (r) → Y (r) and also Y (r) 6= ∅. Since the union X (r) = SI XI is disjoint and f (r) maps XI into the corresponding YI we only have to approximate the restriction f (r)XI : XI → YI for XI 6= ∅ (and so also YI 6= ∅). But YI is one single intersection, hence an affine Nash manifold (Proposition 8.3). By Proposition 2.D.3 there are Nash maps gI : XI → YI arbitrarily Sν close to f (r)XI . These gI 's glue into the desired approximation g(r) of f (r). The construction itself guarantees that g(r) maps each intersection XI at level r into the corresponding YI . Step 2. Now suppose we are at level p ≥ r (hence X (p) 6= ∅), and we have Nash maps g(p) : X (p) → Y (p) arbitrarily Sν−mj close to f (p) : X (p) → Y (p) where j(= r − p) is the number of levels already done. Moreover, we have that both f (p) and g(p) map each non-empty intersection XI at level p into YI . The argument that follows only needs these non-empty intersections. In view of our discussion of topologies and irreducible components (Proposition 6.2), the restrictions gI = g(p)XI : XI → YI (I with p indices and XI 6= ∅) are Nash approximations of the restrictions fI = f (p)XI : XI → YI . Let us now approximate f (p−1) : X (p−1) → Y (p−1). Step 3. Let XJ be an intersection of Xi's at level p− 1. Every intersection of XJ with another XJ ′ at the same level p − 1 is included in some intersection XI at level p, hence consider the restrictions to XJ ∩ X (p) of f (p) and of g(p); let ϕ stand for the first and ψ for the second one. Of course, we only care for the non-empty intersections of that type. The intersection XJ ∩ X (p) is a Nash set with monomial singularities, and the restrictions are maps into YJ , which is a Nash manifold because Y is a Nash monomial crossings (Proposition 8.3 again). By Proposition 8.2 if ψ is Sµ close enough to ϕ, it has an extension ψJ : XJ → YJ arbitrarily Sµ−m close to fJ (in fact, that proposition gives an extension to M that we restrict to XJ ). In our case we take µ = ν − mj, hence µ− m = ν − m(j + 1). The ψJ 's glue together into a c-Nash map g(p−1) : X (p−1) → Y (p−1) arbitrarily Sν−m(j+1) close to f (p−1) : X (p−1) → Y (p−1). Notice that g(p−1) maps each non-empty intersection XJ at level p − 1 into YJ . By Proposition 8.1 the c-Nash maps are Nash, also concerning topologies of maps, and we have obtained the desired approximation of f (p−1). Final arrangement. In the process above the differential class loses m units at every level jump. As we start at level r with an Sν approximation, in the end have an Sν−m(r−1) approximation. (cid:3) [n/2](cid:1). Thus pick any point x ∈ XI with XI 6= ∅ at level r. Now it remains to prove that r ≤ (cid:0) n Since f (XI ) ⊂ YI , the point y = f (x) is in YI = Yk1 ∩ ··· ∩ Ykr , and the r germs Yk1,y, . . . , Ykr,y are irreducible components of Yy. Since Y ⊂ N has a monomial singularity at y, its tangent cone [n/2](cid:1) elements. This gives the required is an extremal family, and by Remark 3.10 it has at most(cid:0) n bound for r. [n/2](cid:1) is sharp: we could have all the coordinates varieties Remark 8.5. Note that the bound (cid:0) n of dimension [n/2]. The loss of differentiability class in the approximation obtained above has been formulated to show that it only depends on dimensions. A different obvious bound for r is the number of irreducible components Yk(i) of Y we had chosen to start with, and that number [n/2](cid:1). For instance, if Y ⊂ N is a normal crossing divisor, i.e., can certainly be smaller than (cid:0) n codim(Y ) = 1, then r ≤ n and so Theorem 1.7 holds for q = m(n − 1). However, in general it is difficult to get a better estimation of its size if we do not know a priori the dimensions of the Yi's. (cid:3) 9. Classification of affine Nash manifolds with corners In this section we use our approximation results to deduce Theorem 1.8 concerning the clas- sification of affine Nash manifolds with corners. First, recall that a map h : Z → T is an Sν 36 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ diffeomorphism if it is a bijection and both h and h−1 are Sν maps. With respect to approxima- tion, diffeomorphisms of affine Nash manifolds behave well. Fact 9.1. [Sh, II.1.7, p. 86] Let h : M → N be an Sν diffeomorphism of affine Nash manifolds. If an Sν map g : M → N is Sν close enough to h, then g is also an Sν diffeomorphism, and g−1 is Sν close to h−1. In particular, from this and Proposition 2.D.3 we deduce that for all ν ≥ 1 every Sν diffeomor- phism f : M → N can be approximated by Nash diffeomorphisms, hence the Sν classification and the Nash classification coincide for affine Nash manifolds. Our Theorem 1.8 says this is true for manifolds with corners (for suitable ν). For the proof of Theorem 1.8 we need some results that can be of interest by themselves. Lemma 9.2. Let f : T → T ′ be a semialgebraic local homeomorphism of locally compact semial- gebraic sets which restricts to a homeomorphism from a closed semialgebraic subset S of T onto another S′ of T ′. Then f restricts to a homeomorphism from an open semialgebraic neighborhood W of S in T onto another W ′ of S′ in T ′. Proof. We are to prove first that we may assume f −1(S′) = S. Consider the semialgebraic set C = {x ∈ T \ S : there is some y ∈ S such that f (y) = f (x)}. We claim that no point in S is adherent to C. For, suppose there are sequences {xk}k off S converging to x ∈ S and {yk}k in S such that f (yk) = f (xk). Then limkf (yk) = limkf (xk) = f (x) ∈ S′ and since fS : S → S′ is a homeomorphism, {yk}k must converge to x. But f is injective near x, hence xk = yk ∈ S for k large, a contradiction. Thus replacing T by T \ C and T ′ by its open image (since f is a local homeomorphism it is an open map) we can assume that f −1(S′) = S. Now, since f is a local homeomorphism, there is a finite open semialgebraic covering U1, . . . , Ur of T such that all restrictions fUi : Ui → f (Ui) are homeomorphisms [BCR, 9.3.9]. Since f is open, we can also suppose that the f (Ui)'s form a cover of T ′. If r = 1 we are done, so let us see that when r > 1 we can modify the covering to reduce the number r of open sets, which will give the result. First of all we consider the open semialgebraic set ∆ = f (U1) ∪ f (U2) and find a semialgebraic shrinking of the covering {f (U1), f (U2)}, i.e., open semialgebraic sets ∆1, ∆2 that still cover ∆ and such that ∆∩ ∆i ⊂ f (Ui) (for instance, take a continuous semialgebraic function δ on ∆ with δ∆\f (U1) = −1 and δ∆\f (U2) = 1 using [BCR, 2.6.9] and set ∆1 = {δ > − 1 2}). 2} and ∆2 = {δ < 1 Next, consider the open semialgebraic sets Vi = Ui ∩ f −1(∆i). We claim that S ∩ (V1 ∪ V2) = S ∩ (U1 ∪ U2). (9.1) Indeed, for the relevant inclusion right to left, let us consider x ∈ S ∩ (U1 ∪ U2). Then f (x) ∈ S′ ∩ (f (U1) ∪ f (U2)) = S′ ∩ ∆. We can assume that f (x) ∈ ∆1 ⊂ f (U1), so that f (x) = f (y) for some y ∈ U1. Thus y ∈ f −1(S′) = S and, since f is injective on S, we deduce that x = y ∈ U1 ∩ f −1(∆1) = V1. Now we show that no x ∈ S ∩ (U1 ∪ U2) is adherent to the semialgebraic set C ′ = {x ∈ V1 \ V2 : there is some y ∈ V2 such that f (y) = f (x)}. Indeed, suppose there are sequences {xk}k in V1 \ V2 converging to x ∈ S ∩ (U1 ∪ U2) and {yk}k in V2 with f (yk) = f (xk). Since f (x) ∈ f (U1) ∪ f (U2) = ∆ we have f (x) = limkf (xk) = limkf (yk) ∈ ∆ ∩ f (V2) ⊂ ∆ ∩ ∆2 ⊂ f (U2), and since fU2 : U2 → f (U2) is a homeomorphism, there exists y ∈ U2 with y = limk yk. Con- sequently, f (y) = f (x) ∈ S′. As f −1(S′) = S, we have y ∈ S and since fS is injective, y = x. APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 37 Thus the two sequences {xk}k and {yk}k converge to x and f being locally injective, xk = yk for k large, a contradiction. We have so proved that S ∩ (U1 ∪ U2) ∩ C ′ = ∅. (9.2) Now set W2 = (V1 ∪ V2) \ C ′, which is an open semialgebraic neighborhood of the first equality by equation (9.2) and the second one by equation (9.1). S ∩ W2 = S ∩ (V1 ∪ V2) = S ∩ (U1 ∪ U2) (cid:3) Finally, by definition of C ′, f is injective on W2, and so f restricts to a homeomorphism W2 → f (W2). All in all we can replace the two open sets U1 and U2 of the initial covering U1, . . . , Ur by the single open set W2 so getting a new covering of S by r − 1 open sets, as desired. Lemma 9.3. Let Q ⊂ Ra be a Nash manifold with corners. Let M ⊂ Ra be a Nash envelope of Q such that the Nash closure X of ∂Q in M is a normal crossings. Let N ⊂ Rb be a Nash manifold and f : Q → N , g : X → N be Sν maps such that g∂Q = f∂Q. Then for M small enough there exists an Sν map H : M → N such that HQ = f and HX = g. Proof. Let m = dim(M ) = dim(Q). By Theorem [FGR, Thm. 1.11] there is a finite open covering U1, . . . , Uℓ of ∂Q equipped with Nash diffeomorphisms (ui1, . . . , uim) : Ui → Rm with Ui ∩ X = {ui1 ··· uiki = 0} and Ui ∩ Q = {ui1 ≥ 0, . . . , uiki ≥ 0} for some 1 ≤ ki ≤ m. By Corollary 4.C.3 for each Ui there is an Sν map ζi : Ui → Rb such that ζi = f on Ui ∩ Q and ζi = g on Ui ∩ X. Set U0 = Int(Q) and ζ0 = fU0. Let σ0, σ1, . . . , σℓ be an Sν partition of unity subordinated to {U0, U1, . . . , Uℓ}. Consider U = Sℓ i=0 σiζi : U → Rb which coincides with f on Q and with g on X. This map H solves the problem except that its image is not contained in N . To amend this, pick a Nash retraction ρ : V → N of N in Rb and replace M by H −1(V ) and H by ρ ◦ H. Proposition 9.4. Let Q ⊂ Ra be a Nash manifold with divisorial corners. Let M ⊂ Ra be a Nash envelope of Q such that the Nash closure X of ∂Q in M is a Nash normal crossing divisor. Let X1, . . . , Xs be the irreducible components of X and let Y be a Nash monomial crossings in a Nash affine manifold N ⊂ Rb. Let f : Q → N be an Sν map such that each f (∂Q ∩ Xi) is contained in an irreducible component Yk(i) of Y . Then for M small enough there exists an Sν extension F : M → N of f such that F (Xi) ⊂ Yk(i) for i = 1, . . . , s. Proof. We will use the notion of iterated faces introduced in 2.F. Recall that Q is the unique m- face and for d < m a d-face is a face of a (d + 1)-face. This construction ends at some d = m0 ≥ 0, so that the m0-faces are affine Nash manifolds. For M small enough we can assume the properties of Proposition 2.F.3. i=0 Ui and the Sν map H = Pℓ (cid:3) Now, to prove the statement it is enough to construct, for each Nash closure Z of an iterated face, an Sν map F Z : Z → N such that (i) F ZZ∩Q = fZ∩Q, (ii) if Z ′ is the Nash closure of an iterated face with Z ′ ⊂ Z then F Z ′ (iii) if Z ⊂ Xi then im(F Z ) ⊂ Yk(i). We proceed by ascending induction on the dimension d of the faces. For d = m0, every m0-face is an affine Nash manifold contained in Q and therefore it coincides with its Nash closure, so that we can just take the restriction of f to that m0-face. Now, fix the Nash closure Z of a d-face D, d > m0, for which we know Z ∩ Q = D. Let Z1, . . . , Zℓ be the irreducible components of the Nash closure of ∂D in Z. Note that each Zj is the Nash closure of a face Dj of D, which is a (d− 1)-face. By induction, properties (i)-(iii) hold for the Zj's. Let I = {i : Z ⊂ Xi} and consider YI =Ti∈I Yk(i), which is an affine Nash manifold by Proposition 8.3. = F ZZ ′, 38 E. BARO, JOS´E F. FERNANDO, AND JES ´US M. RUIZ Let F Zj : Zj → YI be the Sν maps provided by the induction hypothesis for j = 1, . . . , ℓ, which are well defined by property (iii). By Proposition 2.F.3(3), the connected components of Zj ∩ Zj ′ meeting Q are Nash closures of iterated faces (of smaller dimensions) and therefore, by (ii), F Zj and F Zj′ coincide on those connected components. Let C be the union of the connected components of all intersections Zj ∩ Zj ′ that do not intersect Q. Since C is a closed semialgebraic set disjoint from Q, we can shrink M to M \ C. In particular, we can assume that the map G : ℓ[j=1 Zj → YI where G = FZj on Zj Proposition 8.1. is a well-defined cSν function. But that union Sj Zj is a normal crossings in Z, hence G is Sν by Finally, by (i), G∂D = f∂D and we can apply Lemma 9.3 to find an Sν map F Z : Z → YI such that F ZD = fD and F Z∂D = G∂D. This as usual after shrinking Z to Z \ C for some closed semialgebraic set C ⊂ Z, that is, after shrinking M to M \ C. (cid:3) (9.5) Proof of Theorem 1.8. We can suppose Q1 and Q2, hence their interiors, connected. Let h : Q1 → Q2 be an Sν diffeomorphism. Let M ⊂ Ra and N ⊂ Rb be Nash envelopes of Q1 and Q2. Let X and Y be the Nash closures of ∂Q1 and ∂Q2 in M and N . By hypothesis both X and Y are Nash normal crossing divisors. Now, since h is an Sν diffeomorphism, we show that it maps the intersection of ∂Q with each irreducible component of X into some irreducible component of Y . For, let X ′ be an irreducible component of ∂X, that is, the Nash closure of a face D of ∂Q1. By definition, D is the topological closure of a connected component C of Smooth(∂Q1) and we can assume that X ′ ∩ ∂Q1 = D. Clearly, h(C) is an Sν manifold open in ∂Q2 and therefore h(C) ⊂ Smooth(∂Q2) (just note that in ∂Q2 being a smooth point in the Nash or Sν sense coincide). Then, h(C) is contained in a connected component of Smooth(∂Q2) and therefore h(C) ⊂ Y ′ where Y ′ is an irreducible component of Y . In particular, h(∂Q1 ∩ X ′) = h(D) = h(C) = h(C) ⊂ Y ′ = Y ′, as required. Now, by Proposition 9.4 we can assume there is an Sν extension H : M → N of h such that H(X) ⊂ Y . This extension is a local diffeomorphism at every point of Q1, hence shrinking M and N we may assume that H is a local diffeomorphism. But it is a homeomorphism from Q1 onto Q2 and hence by Lemma 9.2 after a new shrinking H is a diffeomorphism. After this preparation, we use Theorem 1.7 and Remark 8.5 to obtain a semialgebraic Sν−q approximation of g = HX : X → Y by a Nash map f : X → Y , where q = m(m − 1). Then, by Proposition 8.2, f has a Nash extension F : M → N which is Sν−q−m close to H. Note that ν − q − m = ν − m(m − 1) − m = ν − m2 ≥ 1. Since H is a diffeomorphism, a close enough approximation F is also a diffeomorphism (Fact 9.1). In particular, X is Nash equivalent to Y . Let us check that F (Q1) = Q2. Since the manifolds are the closures of their interiors, it is enough to see that F maps Int(Q1) onto Int(Q2). We know that F (X) = f (X) = Y , and conse- quently F (M \ X) = N \ Y . Since Q1 ∩ X = ∂Q1, the interior Int(Q1) is a connected component of M \ X, hence F maps it onto one of N \ Y . Among these latter is Int(Q2) (also Q2∩ Y = ∂Q2), and we know that H(Int(Q1)) = Int(Q2). Pick any point x ∈ Int(Q1). Since Int(Q2) is open in N , we have dist(H(x), N \ Int(R)) > 0 and for F close to H we conclude F (x) /∈ N \ Int(Q2). Consequently F (Int(Q1)) ∩ Int(Q2) 6= ∅ and so F (Int(Q1)) = Int(Q2). (cid:3) Remark 9.6. Let us review the preceding proof for two Nash manifolds Q1, Q2 with boundary without corners. Notice that their boundaries ∂Q1, ∂Q2 are Nash smooth hypersurfaces of the Nash envelopes. APPROXIMATION ON NASH SETS WITH MONOMIAL SINGULARITIES 39 First of all, the boundaries are their own Nash closures, that is, X = ∂Q1 and Y = ∂Q2. Hence, Proposition 9.4 is not needed. On the other hand, Theorem 1.7 can be replaced by Fact 2.D.3 and this approximation does not lower the differentiability class. Thus, we have a Nash map f : X → Y which is an Sν approximation of g = HX . Next, the Nash extension F of f provided by Proposition 8.2 is an Sν−1 approximation of H. Indeed, this proposition comes from: (i) Lemma 7.2 and Proposition 7.3, here needed for one coordinate hyperplane only, and there- fore giving I ν (X) ⊂ I(X)Sν−1(M ), and (ii) Proposition 7.6, which for a Nash smooth hypersurface X of M gives by (i) a Nash extension that is an Sν−1 approximation. This leads to an Sν−1 approximation F of H, as claimed. All in all, we conclude: Two m-dimensional affine Nash manifolds with boundary without corners which are S2 diffeomorphic are Nash diffeomorphic. References [Ar] M. Artin: Algebraic approximation of structures over complete local rings, Publ. Math. I.H.E.S. 36 (1969), 23 -- 58. [BCR] J. Bochnak, M. Coste, M.-F. Roy: Real algebraic geometry. Ergeb. Math. 36. Springer-Verlag, Berlin: [C] 1998. H. Cartan: Vari`et`es analytiques r`eelles et vari`et`es analytiques complexes, Bull. Soc. Math. France. 85 (1957), 77 -- 99. [CRS1] M. Coste, J.M. Ruiz, M. Shiota: Global Problems on Nash Functions. Rev. Mat. Complut. 17 (2004), no. 1, 83 -- 115. [CRS2] M. Coste, J.M. Ruiz, M. Shiota: Approximation in compact Nash manifolds. Amer. J. Math. 117 (1995), 905 -- 927. [CRS3] M. Coste, J.M. Ruiz, M. Shiota: Separation, factorization and finite sheaves on Nash manifolds. Compositio Math. 103 (1996), no. 1, 31 -- 62. [CS] M. Coste, M. Shiota: Nash functions on noncompact Nash manifolds. Ann. Sci. ´Ecole Norm. Sup. (4), 17 [D] [E] [FG] (2004), no. 1, 1 -- 33. H. Dao: Answer in Mathoverflow to Are squarefree monomial ideals on a regular system of parameters in a regular local ring regular?, http://mathoverflow.net/questions/55422 (version: 2011-02-16). G.A. Efroymson: The extension theorem for Nash functions. Real algebraic geometry and quadratic forms (Rennes, 1981), pp. 343 -- 357, Lecture Notes in Math., 959, Springer, Berlin-New York: 1982. J.F. Fernando, J.M. Gamboa: On the irreducible components of a semialgebraic set. Internat. J. Math. 23 (2012), no. 4, 1250031, 40 pp. [FGR] J.F. Fernando, J.M. Gamboa, J.M. Ruiz: Finiteness problems on Nash manifolds and Nash sets. To appear in J. Eur. Math. Soc. D. Lubell: A short proof of Sperner's lemma. J. Comb. Theory 1 (1966), 299. R. Narasimhan: Introduction to the theory of analytic spaces. Lecture Notes in Math., 25. Springer-Verlag, Berlin: 1966. R. Narasimhan: Analysis on real and complex manifolds. Second edition. Advanced Studies in Pure Math- ematics, 1. Masson & Cie, ´Editeur, Paris; North-Holland Publishing Co., Amsterdam-London; American Elsevier Publishing Co., Inc., New York: 1973. M. Shiota: Nash manifolds. Lecture Notes in Math., 1269. Springer-Verlag, Berlin: 1987. J. Stasica: Smooth points of a semialgebraic set. Ann. Polon. Math. 82 (2003), no. 2, 149 -- 153. [L] [N1] [N2] [Sh] [St] Departamento de ´Algebra, Facultad de Ciencias Matem´aticas, Universidad Complutense de Ma- drid, 28040 MADRID (SPAIN) E-mail address: [email protected],[email protected] Departamento de Geometr´ıa y Topolog´ıa, Facultad de Ciencias Matem´aticas, Universidad Com- plutense de Madrid, 28040 MADRID (SPAIN) E-mail address: [email protected]
1009.0178
2
1009
2011-09-11T20:28:18
Rational points over finite fields for regular models of algebraic varieties of Hodge type $\geq 1$
[ "math.AG", "math.AC", "math.NT" ]
Let $R$ be a discrete valuation ring of mixed characteristics $(0,p)$, with finite residue field $k$ and fraction field $K$, let $k'$ be a finite extension of $k$, and let $X$ be a regular, proper and flat $R$-scheme, with generic fibre $X_K$ and special fibre $X_k$. Assume that $X_K$ is geometrically connected and of Hodge type $\geq 1$ in positive degrees. Then we show that the number of $k'$-rational points of $X$ satisfies the congruence $|X(k')| \equiv 1$ mod $|k'|$. Thanks to \cite{BBE07}, we deduce such congruences from a vanishing theorem for the Witt cohomology groups $H^q(X_k, W\sO_{X_k,\Q})$, for $q > 0$. In our proof of this last result, a key step is the construction of a trace morphism between the Witt cohomologies of the special fibres of two flat regular $R$-schemes $X$ and $Y$ of the same dimension, defined by a surjective projective morphism $f : Y \to X$.
math.AG
math
RATIONAL POINTS OVER FINITE FIELDS FOR REGULAR MODELS OF ALGEBRAIC VARIETIES OF HODGE TYPE ≥ 1 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING Abstract. Let R be a discrete valuation ring of mixed characteristics (0, p), with finite residue field k and fraction field K, let k′ be a finite extension of k, and let X be a regular, proper and flat R-scheme, with generic fibre XK and special fibre Xk. Assume that XK is geometrically connected and of Hodge type ≥ 1 in positive degrees. Then we show that the number of k′-rational points of X satisfies the congruence X(k′) ≡ 1 mod k′. Thanks to [BBE07], we deduce such congruences from a vanishing theorem for the Witt cohomology groups H q(Xk, W OXk,Q), for q > 0. In our proof of this last result, a key step is the construction of a trace morphism between the Witt cohomologies of the special fibres of two flat regular R-schemes X and Y of the same dimension, defined by a surjective projective morphism f : Y → X. Contents Introduction and first reductions 1. 2. Application of p-adic Hodge theory 3. An injectivity theorem for coherent cohomology 4. Koszul resolutions and local description of the trace morphism τf 5. Preliminaries on the relative de Rham-Witt complex 6. The Hodge-Witt trace morphism for projective spaces 7. The Hodge-Witt fundamental class of a regularly embedded subscheme 8. Proof of the injectivity theorem for Witt vector cohomology 9. An example Appendix: Complete intersection morphisms of virtual relative dimension 0 A. The canonical section of the relative dualizing sheaf B. The trace morphism τf on Rf∗(OY ) References 2 7 21 22 26 38 42 51 59 65 74 83 Date: September 9, 2011. 2010 Mathematics Subject Classification. Primary: 11G25. Secondary: 13F35, 14F30, 14G05. Key words and phrases. Complete intersections, de Rham-Witt complex, fundamental class, Hodge type, p-adic cohomology, p-adic Hodge theory, rational points, regular models, slope filtra- tion, trace morphism, Witt vectors, zeta function. Partially supported by the DFG Leibnitz Preis, the SFB/TR45, the ERC Advanced Grant 226257. 2 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING 1. Introduction and first reductions Let R be a discrete valuation ring of mixed characteristics (0, p), with perfect residue field k, and fraction field K. The main goal of this article is to prove the following theorem. Theorem 1.1. Let X be a proper and flat R-scheme, with generic fibre XK , such that the following conditions hold: a) X is a regular scheme. b) XK is geometrically connected. c) H q(XK , OXK ) = 0 for all q ≥ 1. If k is finite, then, for any finite extension k′ of k, the number of k′-rational points of X satisfies the congruence (1.1.1) X(k′) ≡ 1 mod k′. Condition c) should be viewed as a Hodge theoretic property of XK , which can be stated by saying that XK has Hodge type ≥ 1 in positive degrees. From this point of view, this theorem fits in the general analogy between the vanishing of Hodge numbers for varieties over a field of characteristic 0, and congruences on the number of rational points with values in finite extensions for varieties over a finite field. This analogy came to light with the coincidence between the numerical values in Deligne's theorem on smooth complete intersections in a projective space [SGA 7 II, Expos´e XI, Th. 2.5], and in the Ax-Katz theorem on congruences on the number of solutions of systems of algebraic equations [Kz71, Th. 1.0]. It has been made effective by Katz's conjecture [Kz71, Conj. 2.9] relating the Newton and Hodge polygons associated to the cohomology of a proper and smooth variety (and generalizing earlier results of Dwork for hypersurfaces [Dw64]). For varieties in characteristic p, this conjecture was proved by Mazur ([Ma72], [Ma73]) and Ogus [BO78, Th. 8.39]. In the mixed characteristic case, where a stronger form can be given using the Hodge polygon of the generic fibre, it is a consequence of the fundamental results in p-adic Hodge theory. Our proof of Theorem 1.1 makes essential use of the unequality between these two polygons, but the setup of the theorem is actually more general, since the scheme X is not supposed to be semi-stable over R. Let us also recall that a result similar to Theorem 1.1 has been proved by the second author [Es06, Th. 1.1] by ℓ-adic methods, with condition c) replaced by a coniveau condition: for any q ≥ 1, any cohomology class in H q ´et(XK , Qℓ) vanishes in H q ´et(UK , Qℓ) for some non empty open subset U ⊂ XK . It is easy to see, using [De71], that this coniveau condition implies that the Hodge level of XK is ≥ 1 in degree q ≥ 1 (see [Il06, 4.4 (d)] for a more general discussion). It would actually follow from Grothendieck's generalized Hodge conjecture [Gr69] that the two conditions are equivalent. In this article, the use of p-adic methods, and in particular of p-adic Hodge theory, allows us to derive congruence (1.1.1) directly from Hodge theoretic hypotheses. 1.2. As explained by Ax [Ax64], congruences such as (1.1.1) can be expressed in terms of the zeta function of the special fibre Xk of X. We recall that the rationality RATIONAL POINTS OF REGULAR MODELS 3 of the zeta function Z(Xk, t) allows to define the slope < 1 part Z <1(Xk, t) of Z(Xk, t) as follows [BBE07, 6.1]. Let k = pa, and write Z(Xk, t) =Yi (1 − αit)/Yj (1 − βjt), with αi, βj ∈ Qp and αi 6= βj for all i, j. Normalizing the p-adic valuation v of Qp by v(pa) = 1, one sets Z <1(Xk, t) = Yv(αi)<1 (1 − αit)/ Yv(βj )<1 (1 − βjt). Then the congruences (1.1.1) are equivalent to (1.2.1) [BBE07, Prop. 6.3]. Z <1(Xk, t) = 1 1 − t On the other hand, let W (OXk ) be the sheaf of Witt vectors with coefficients in OXk , and W OXk,Q = W (OXk ) ⊗ Q. Then the identification of the slope < 1 part of rigid cohomology with Witt vector cohomology provides the cohomological interpretation (1.2.2) det(1 − tF aH i(Xk, W OXk,Q))(−1)i+1 , Z <1(Xk, t) =Yi where F is induced by the Frobenius endomorphism of W (OXk ) [BBE07, Cor. 1.3]. Therefore, Theorem 1.1 is a consequence of the following theorem, where k is only assumed to be perfect: Theorem 1.3. Let X be a regular, proper and flat R-scheme. Assume that H q(XK , OXK ) = 0 for some q ≥ 1. Then: (1.3.1) H q(Xk, W OXk,Q) = 0. Proof of Theorem 1.1, assuming Theorem 1.3. Let us prove here this implication, which is easy and does not use the regularity assumption on X. Let W = W (k), and K0 = Frac(W ). Thanks to (1.2.1) and (1.2.2), Theorem 1.3 implies that it suffices to prove that the homomorphism K0 → H 0(Xk, W OXk,Q) is an isomorphism. Since X is proper and flat over R, H 0(X, OX ) is a free finitely generated R- module. As the generic fibre XK is geometrically connected and geometrically re- duced, the rank of H 0(X, OX ) is 1. The homomorphism R → H 0(X, OX ) maps 1 to 1, hence Nakayama's lemma implies that it is an isomorphism. Applying Zariski's connectedness theorem, it follows that Xk is connected, and even geometrically con- nected, since the same argument can be applied after any base change from R to R′, where R′ is the ring of integers of a finite extension of K. On the other hand, let ¯k be an algebraic closure of k, and let k′ be a finite extension of k such that X¯k red is defined over k′. As k′ is separable over k, the homomor- phisms Wn(k) → Wn(k′) are finite ´etale liftings of k → k′, and the homomorphisms Wn(k′) ⊗Wn(k) Wn(OXk ) → Wn(OXk′ ) are isomorphisms [Il79, I, Prop. 1.5.8]. It fol- lows that the homomorphism W (k′) ⊗W (k) H 0(Xk, W (OXk )) → H 0(Xk′, W (OXk′ )) 4 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING is an isomorphism, and that it suffices to prove the claim for Xk′. Using the fact that ∼ H 0(Xk′, W OXk′ ,Q) −−→ H 0(Xk′ red, W OXk′ red,Q) by [BBE07, Prop. 2.1 (i)], it suffices to check that, if Z is a proper, geometri- cally connected and geometrically reduced k-scheme, the homomorphism W (k) → H 0(Z, W (OZ )) is an isomorphism. Under these assumptions, the homomorphism k → H 0(Z, OZ ) is an isomorphism. As the homomorphism R : Wn(OZ ) → Wn−1(OZ ) is the projection of a product onto one of its factors, the homomorphisms H 0(Z, Wn(OZ )) → H 0(Z, Wn−1(OZ )) are sur- jective, and one gets by induction that the homomorphism Wn(k) → H 0(Z, Wn(OZ )) is an isomorphism for all n. Taking inverse limits, the claim follows. (cid:3) 1.4. Theorem 1.3 is deeper, and most of our paper is devoted to developing the techniques used in its proof. We may observe though that, in the context of Theorem 1.1, there is a case where (1.3.1) is trivial: namely, if we replace the condition on the Hodge numbers of XK , which is equivalent to requiring that the modules H q(X, OX ) be p-torsion modules, by the stronger condition that H q(X, OX ) vanishes for all q ≥ 1. Indeed, the flatness of X over R allows to apply the derived base change formula for coherent cohomology and to conclude that H q(Xk, OXk ) = 0 for all q ≥ 1. By induction on n, one gets that H q(Xk, Wn(OXk )) = 0 for all n, q ≥ 1, and (1.3.1) follows for all q ≥ 1 (even before tensoring with Q). In the general case, where the H q(X, OX ) are p-torsion modules, we do not know any direct argument to derive the vanishing property stated in (1.3.1). Our strategy is then to use the results of p-adic Hodge theory relating the Hodge and Newton polygons of certain filtered F -isocrystals on k, which allow to study separately the cohomology groups for a given q as in Theorem 1.3. In particular, when X is semi- stable on R, a straightforward argument using the fundamental comparison theorems of p-adic Hodge theory allows to deduce (1.3.1) from the unequality between the two polygons defined by the log crystalline cohomology of Xk. We explain this argument in Theorem 2.1. In the rest of Section 2, we show that this argument can be modified to prove the vanishing of H q(Xk, W OXk,Q) in the general case. For any finite extension K ′ of K, with ring of integers R′, let XR′ be deduced from X by base change from R to R′. After reducing to the case where R is complete, the first step is to apply de Jong's alteration theorem to construct for any m an m-truncated simplicial scheme Y• over the ring of integers R′ of a suitable extension K ′ of K, endowed with an augmentation morphism Y0 → XR′, such that the Yi's are pullbacks of proper semi- stable schemes, and Y• → XR′ induces an m-truncated proper hypercovering of XK ′ (see Lemma 2.2 for a precise statement). Then, using Tsuji's extension of the comparison theorems to truncated simplicial schemes [Ts98], we show that, in this situation, the cohomology group H q(Y•k, W OY• k,Q) vanishes. However, due to the possible presence of vertical components in the coskeletons, the special fibre Y•k of the m-truncated simplicial scheme Y• may not be a proper hypercovering of Xk, and it is unclear how the groups H q(Y•k, W OY• k,Q) are related to the groups H q(Xk, W OXk,Q). Therefore another ingredient will be necessary to complete the RATIONAL POINTS OF REGULAR MODELS 5 proof. It will be provided by the following injectivity theorem, the proof of which will be given in section 8. Theorem 1.5. Let X, Y be two flat, regular R-schemes of finite type, of the same dimension, and let f : Y → X be a projective and surjective R-morphism, with reduction fk over Spec k. Then, for all q ≥ 0, the functoriality homomorphism (1.5.1) is injective. f ∗ k : H q(Xk, W OXk,Q) −→ H q(Yk, W OYk,Q) 1.6. We will deduce Theorem 1.5 from the existence of a trace morphism (1.6.1) τi,π : Rf∗(W OYk,Q) −→ W OXk,Q, defined by means of a factorization f = π ◦i, where π is the projection of a projective space P d X on X, and i is a closed immersion. The key fact used in the construction of this trace morphism is that, under the assumptions of Theorem 1.5, i is a regular immersion of codimension d, or, said otherwise, that f is a complete intersection morphism of virtual relative dimension 0, in the sense of [SGA 6, Expos´e VIII]. Sections 3 to 7 are devoted to the construction of τi,π. In section 3, we state a similar result for OX , providing a canonical trace morphism τf : Rf∗(OY ) → OX, whenever X is a noetherian scheme with a relative dualizing complex, and f : Y → X is a proper complete intersection morphism of virtual relative dimension 0 (see Theorem 3.1). The existence of τf has been observed by El Zein as a particular case of his construction of the relative fundamental class [El78, IV, Prop. 6]. However, there does not seem to be in the literature a complete proof of the properties listed in Theorem 3.1. Due to the many corrections and complements to [Ha66] made by Conrad in [Co00], we have included in an Appendix the details of a proof of Theorem 3.1 based on [Co00]. So we refer to B.7 for the definition of τf , and to B.9 for the proof of Theorem 3.1. When Y is finite locally free of rank r over X, the composition of the functoriality morphism OX → Rf∗(OY ) with τf is multiplication by r on OX . This has striking consequences for the functoriality maps induced by f on coherent cohomology (see Theorem 3.2). For example, if r is invertible on X, one obtains an injectivity theorem which may be of independent interest. An outline of the construction of τf is given in the introduction to the Appendix. To construct the trace morphism τi,π, we consider more generally a projective complete intersection morphism f : Y → X of virtual relative dimension 0 between two noetherian Fp-schemes with dualizing complexes. Under these assumptions, we construct a compatible family of morphisms τi,π,n : Rf∗(Wn(OY )) → Wn(OX ) for n ≥ 1, with τi,π,1 = τf . Our main tool here is the theory of the relative de Rham-Witt complex developped by Langer and Zink [LZ04]. In Section 5, we recall some basic facts about their construction, and we extend to the relative case some structure theorems proved by Illusie [Il79] when the base scheme is perfect (see in particular Proposition 5.7 and Theorem 5.13). Then we define τi,π,n by combining 6 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING two morphisms. On the one hand, we consider a projective space P := P d projection π on X, and we define in Section 6 a trace morphism X with Trpπ,n : Rπ∗(WnΩd P/X[d]) → Wn(OX ), using the d-th power of the Chern class of the canonical bundle OP (1). On the other hand, we consider a regularly embedded closed subscheme Y of a smooth X-scheme P , and we define in Section 7 a relative Hodge-Witt fundamental class for Y in P , which is a section of Hd P/X) and defines a morphism Y (WnΩd γi,π,n : i∗Wn(OY ) → WnΩd P/X[d], with i : Y ֒→ P and d = codimP (Y ). This allows to define the morphism τi,π,n as being the composition Trpπ,n ◦ Rπ∗(γi,π,n). The proof of Theorem 1.5 is then completed in Section 8 thanks to a theorem relating the morphisms τi,π,n defined by the reduction mod p of a factorization of the given morphism f : Y → X over R, and the morphism τf defined by f . It may be worth pointing out here that these results seem to indicate that Grothendieck's relative duality theory for coherent O-modules can be generalized to some extent to the Hodge-Witt sheaves, as was already apparent from [Ek84] when the base scheme is a perfect field. We do not try to develop such a generaliza- tion in this article, and we limit ourselves to the properties needed for the proof of Theorem 1.1. For example, it is very likely that the morphisms τi,π,n only depend on f , and not on the chosen factorization f = π ◦ i, but this is not needed here, and we do not prove it in this article. A natural context one might think of for developing our results is the theory of the trace map for projectively embeddable morphisms outlined in [Ha66, III, 10.5 and §11]. However, as discussed by Conrad in [Co00, p. 103-104], the foundational work needed for the definition of such a theory has not really been done even for coherent O-modules. We hope to return to these questions in another article. Finally, we conclude in Section 9 by giving a family of examples to which Theorem 1.1 can be applied, but which are not covered by earlier results, nor by cases where Theorem 1.3 can be proved directly, such as the trivial case where H i(X, OX ) = 0 for all i ≥ 1, or the semi-stable case. These examples are obtained for p ≥ 7, and are quotients of an hypersurface of degree p in a projective space P p−2 by a free (Z/pZ)-action. Their generic fibre is a smooth variety of general type, and their special fibre has isolated singularities, at least when p is not a Fermat number. R Acknowledgements The authors thank the referee for his careful reading of the manuscript, and for his very useful comments, questions and suggestions. Part of this work was done in March 2008, when the second and third authors enjoyed the hospitality of the Department of Mathematics at Rennes. General conventions 1) All schemes under consideration are supposed to be separated. By a pro- jective morphism f : Y → X, we always mean a morphism which can be factorized RATIONAL POINTS OF REGULAR MODELS 7 as f = π ◦ i, where i is a closed immersion in some projective space Pn natural projection Pn X → X. X, and π is the 2) In this paper, we use the terminology of [SGA 6] for complete intersection morphisms: a morphism of schemes f : Y → X is said to be a complete intersection morphism if, for any y ∈ Y , there exists an open neighbourhood U of y in Y such that the restriction of f to U can be factorized as f U = π ◦ i, where π is a smooth morphism and i a regular immersion [SGA 6, VIII, 1.1]. Note that this notion of complete intersection morphism is more general than the notion of "local complete intersection map" used in [Ha66] and [Co00], where "lci map" is only used for regular immersions. If d is the codimension of i at y, and n the relative dimension of π at i(y), the integer m = n − d does not depend upon the local factorization f U = π ◦ i, and is called the virtual relative dimension of f at y [SGA 6, VIII, 1.9]. One says that f has constant virtual relative dimension m if the integer m does not depend upon y. We will always assume in this paper that the virtual relative dimension of the morphisms under consideration is constant (however, the dimension of the fibres of such morphisms can vary). 3) Apart from the previous remark, we will use the definitions and sign con- ventions from Conrad's book [Co00]. In particular, when i : Y ֒→ P is a regular immersion of codimension d defined by an ideal I ⊂ OP , we define ωY /P by ωY /P = ∧d((I/I 2)∨) rather than (∧d(I/I 2))∨ as in [Ha66, III, p. 141] (see [Co00, p. 7]). The canonical identification between both definitions is given by [Bo70, III, §11, Prop. 7]. 4) If R, S are commutative rings, R → S a ring homomorphism, and X an R-scheme, we denote by XS the S-scheme Spec S ×Spec R X. 5) If E • is a complex, we denote by (σ≥iE •)i∈Z the naive filtration on E •, i.e., the filtration defined by σ≥iE n = 0 if n < i, σ≥iE n = E n if n ≥ i. 6) If X is a scheme (resp. locally noetherian scheme), we denote by Db qc(OX ) (resp. Db coh(OX )) the full subcategory of the derived category D(OX ) which has as objects the bounded complexes with OX -quasi-coherent (resp. OX -coherent) coho- mology sheaves. We denote by Db fTd(OX ) ⊂ D(OX ) the full subcategory of com- plexes which are isomorphic to a bounded complex of flat OX -modules. Adding several of the indices to Db(OX ), as in Db qc,fTd(OX ), means taking the intersection of the corresponding subcategories. When relevant, we will use similar notations for the analogous subcategories of D(OX ) and D+(OX ). 2. Application of p-adic Hodge theory We explain in this section how the fundamental results of p-adic Hodge theory can be used to prove Theorem 1.3. We begin with the semi-stable case, where p-adic Hodge theory suffices to conclude, and which will serve as a model for the general case. We use the notations R, K, k as in the introduction. 8 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING Theorem 2.1. Let X be a proper and semi-stable R-scheme, with generic fibre XK and special fibre Xk, and let q ≥ 0 be an integer. If H q(XK , OXK ) = 0, then H q(Xk, W OXk,Q) = 0. Proof. We may assume that R is a complete discrete valuation ring. Indeed, if bR is the completion of R, bK = Frac(bR) and eX = X bR, then eX is proper and semi-stable ) = bK ⊗K H q(XK , OXK ) = 0, and X and eX have isomorphic over bR, H q(eX bK , O eX bK special fibres. So the theorem for eX implies the theorem for X. We endow S = Spec R with the log structure defined by the divisor Spec k ⊂ S, S0 = Spec k with the induced log structure, and we denote by S, S0 the corre- sponding log schemes. Similarly, we endow X with the log structure defined by the special fibre Xk, Xk with the induced log structure, and we denote by X, X k the corresponding log schemes. Then X is smooth over S, and X k is smooth of Cartier type [Ka89, (4.8)] over S0. Let Wn = Wn(k) (resp. W = W (k)), and let Σn (resp. Σ) be the log scheme obtained by endowing Σn = Spec Wn (resp. Σ = Spec W ) with the log structure associated to the pre-log structure defined by the morphism MS0 → OS0 = OΣ1 → OΣn (resp. OΣ) provided by composition with the Teichmuller representative map. We can then consider the log crystalline cohomology groups H q crys(X/Σn), which are finitely generated Wn-modules endowed with a Frobenius action ϕ and a monodromy operator N . The log scheme X k also carries a logarithmic de Rham-Witt complex WΩ• , constructed by Hyodo [Hy91] in the semi-stable case, and generalized by Hyodo and Kato [HK94, (4.1)] to the case of smooth S0-log schemes of Cartier type. In degree 0, we have = lim←−n WnΩ• X k X k (2.1.1) by [HK94, Prop. (4.6)]. WnΩ0 X k = Wn(OXk ), It follows from [HK94, Th. (4.19)] that, for all q, there are canonical isomorphisms (2.1.2) H q crys(X k/Σn) ∼ −−→ H q(X k, WnΩ• X k ), which are compatible when n varies, and commute with the Frobenius actions. As Xk is proper over S0, these cohomology groups are artinian W -modules. Therefore one can apply the Mittag-Leffler criterium to get canonical isomorphisms (2.1.3) H q ←−− H q(X k, WΩ• H q(X k, WnΩ• crys(X k/Σn) crys(X k/Σ) H q ∼ ∼ X k ) X k ) −−→ lim←− n ∼ −−→ lim←− n compatible with the Frobenius actions. Using the naive filtration of WΩ• tensoring by K0, one obtains a spectral sequence Xk and (2.1.4) Ei,j 1 = H j(X k, WΩi X k ) ⊗ K0 =⇒ H i+j crys(X k/Σ) ⊗ K0 endowed by functoriality with a Frobenius action F ∗. The operators d, F and V on the logarithmic de Rham-Witt complex satisfy the same relations than on the usual de Rham-Witt complex [HK94, (4.1)], and the structure theorems of [Il79] remain valid in the logarithmic case [HK94, Th. (4.4) and Cor. (4.5)]. It follows that, for all i, j, the K0-vector space H j(X k, WΩi )⊗K0 is finite dimensional, and the action of X k RATIONAL POINTS OF REGULAR MODELS 9 F ∗ on this space has slopes in [i, i + 1[ [Lo02, 3.1]. Therefore, the spectral sequence (2.1.4) degenerates at E1, and yields in particular an isomorphism (H q crys(X k/Σ) ⊗ K0)<1 ∼ crys(X k/Σ) ⊗ K0 where Frobenius acts with slope < 1. Thanks to (2.1.1), we finally get a canonical isomorphism ) ⊗ K0, the source being the part of H q −→ H q(X k, WΩ0 X k (2.1.5) (H q crys(X k/Σ) ⊗ K0)<1 ∼ −−→ H q(Xk, W OXk,Q). On the other hand, the choice of an uniformizer of R determines a Hyodo-Kato isomorphism [HK94, Th. (5.1)] (2.1.6) ρ : H q crys(X k/Σ) ⊗W K ∼ −−→ H q(XK , Ω• XK /K ). This allows to endow H q crys(X k/Σ) ⊗W K with the filtration deduced via ρ from the Hodge filtration of H q(XK , Ω• XK /K ). Together with its Frobenius action and monodromy operator, H q crys(X k/Σ)⊗W K is then a filtered (ϕ, N )-module as defined by Fontaine [Fo94, 4.3.2 and 4.4.8]. As such, it has both a Newton polygon, built as usual from the slopes of the Frobenius action, and a Hodge polygon, built as usual from the Hodge numbers of H q(XK , Ω• XK /K ). Now, let K be an algebraic closure of K, and let Bst, BdR be the Fontaine p-adic period rings. Then Tsuji's comparison theorem [Ts99, Th. 0.2] provides a Bst-linear isomorphism (2.1.7) Bst ⊗K0 H q crys(X k/Σ) ∼ −−→ Bst ⊗K H q ´et(XK , Qp), compatible with the natural Galois, Frobenius and monodromy actions on both sides, and with the natural Hodge filtrations defined on both sides after scalar extension from Bst to BdR. Thus H q crys(X k/Σ) ⊗ K0 is an admissible filtered (ϕ, N )-module [Fo94, 5.3.3], and therefore is weakly admissible [Fo94, 5.4.2]. This implies that its Newton polygon lies above its Hodge polygon [Fo94, 4.4.6]. In particular, either H q crys(X k/Σ)⊗K0 = 0, or the smallest slope of its Newton polygon is bigger than the smallest slope of its Hodge polygon. By assumption, the latter is at least 1, which forces the part of slope < 1 of H q crys(X k/Σ) ⊗ K0 to vanish. Thanks to (2.1.5), this implies the theorem. (cid:3) In the general case, we will use truncated simplicial log schemes satisfying the conditions of the next lemma. We will assume that all the log schemes under con- sideration are fine log schemes [Ka89, (2.3)], and all constructions involving log schemes will be done in the category of fine log schemes. For any finite extension K ′ of K, with ring of integers R′, we will endow Spec R′ with the log structure defined by its closed point, and pullbacks of log schemes to Spec R′ will mean pullbacks in the category of log schemes. Note that, because of [Ka89, (4.4) (ii) and (4.3.1)], the underlying scheme of such a pullback is the usual pullback in the category of schemes. We will denote log schemes by underlined letters, and drop the underlining to denote the underlying schemes. Lemma 2.2. Assume that R is complete, and that X is an integral, flat R-scheme of finite type. Let m ≥ 0 be an integer. Then there exists a finite extension K ′ of K, with ring of integers R′, a split m-truncated simplicial R′-log scheme Y • = (Y•, MY•) 10 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING [SGA 4 II, Vbis, 5.1.1], and an augmentation morphism u : Y0 → XR′ over R′, such that the following conditions hold: a) For each r, Yr is projective over XR′, and Y r is a disjoint union of pullbacks to R′ of semi-stable schemes over the integers of sub-K-extensions of K ′ endowed with the log structure defined by their special fibre; b) Via the augmentation morphism induced by u, Y•,K ′ is an m-truncated proper hypercovering of XK ′; c) There exists a projective R-alteration f : Y → X, where Y is semi-stable over the ring of integers R1 of a sub-K-extension K1 of K ′, and there exists finitely many R-embeddings σi : R1 ֒→ R′, such that, if u1 : Y → XR1 denotes the R1-morphism defined by f , and if Yσi (resp. uσi : Yσi → XR′ ) denotes the R′-scheme (resp. R′- morphism) deduced by base change via σi from Y (resp. u1), then Y0 =`i Yσi and = uσi. uYσi Therefore, we obtain the following commutative diagram: (2.2.1) Y0 =`i Yσi u %KKKKKKKKKK Yσi uσi XR′ Y u1 f !CCCCCCCCC / XR1 / X. Proof. This is a well known consequence of de Jong's alteration theorem [dJ96, Th. 6.5]. For the sake of completeness, we briefly recall how to construct such a simplicial log scheme. For r ≥ 0, we denote by [r] the ordered set {0, . . . , r}, and by ∆ (resp. ∆[m]) the category which has the sets [r] (resp. with r ≤ m) as objects, the set of morphisms from [r] to [s] being the set of non-decreasing maps [r] → [s]. One proceeds by induction on m. Assume first that m = 0. De Jong's theorem provides a finite extension K1 of K, an integral semi-stable scheme Y over the ring of integers R1 of K1, and an R-morphism f : Y → X which is a projective alteration. Let u1 : Y → XR1 be the morphism defined by f . Let K ′ be a finite extension of K1 such that K ′/K is Galois, and let R′ be its ring of integers. For any g ∈ Gal(K ′/K), let σg be the composition K1 → K ′ g −→ K ′, and let Y g (resp. ug : Yg → XR′ ) be the R′-log scheme (resp. R′-morphism) deduced from Y (resp. u1) by base change via σg : R1 → R′. Then one defines Y 0 and u by setting Y 0 = ag∈Gal(K ′/K) Y g, uYg = ug. One easily checks by Galois descent that Y0,K ′ → XK ′ is surjective, and conditions a) - c) are then satisfied. Assume now that the lemma has been proved for m − 1. Over the ring of integers R′′ of some finite extension K ′′ of K, this provides a split (m−1)-truncated simplicial log scheme Y ′′ 0 → XR′′, so as to satisfy conditions a) - c). Note that these conditions remain satisfied after a base change to the ring of integers of any finite extension of K ′′. Let coskm−1(Y ′′ •) be • in the category of simplicial fine R′′-log schemes, and Z = the coskeleton of Y ′′ •, together with an augmentation morphism u′′ : Y ′′ % ? _ o o / /     ! / / RATIONAL POINTS OF REGULAR MODELS 11 coskm−1(Y ′′ •)m its component of index m. Denote by Z1, . . . , Zc those irreducible components of Z which are flat over R′′, and endow each Zj with the log structure induced by the log structure of Z. As a consequence of condition a), this log structure induces the trivial log structure on the generic fibre Z j,K ′′. Applying de Jong's theorem to Zj, one can find a finite extension K ′ j of K ′′, with ring of integers R′ j, an integral semi-stable scheme Tj over R′ j and a projective alteration fj : Tj → Zj. One endows Tj with the log structure defined by its special fibre. Because the log structure of the generic fibre Zj,K ′′ is trivial, the morphism fj extends uniquely to a log morphism fj : T j → Zj. Let K ′ be a Galois extension of K ′′ containing K ′ j for all j, 1 ≤ j ≤ c, and let R′ be its ring of integers. Arguing as in the case m = 0 above, one can deduce from the alterations fj an R′-morphism (2.2.2) T −→ Z j,R′ −→ Z R′ ∼ −−→ coskm−1(Y ′′ • ,R′)m caj=1 where T satisfies condition a), and TK ′ → coskm−1(Y ′′ •,K ′)m is projective and surjec- tive (note that, since all log structures are trivial on the generic fibres, the generic fibre of the coskeleton computed in the category of fine log schemes is the coskeleton of the generic fibres computed in the category of schemes). One can then follow the method of Saint-Donat [SGA 4 II, Vbis, 5.1.3] and Deligne [De74, (6.2.5)] to extend • ,R′ as a split m-truncated simplicial log scheme Y • over R′. The R′-log scheme Y ′′ Y m is defined by Y m = T ` a[m]։[l], l<m N (Y l), where N (Y l) is the complement of the union of the images of the degeneracy mor- phisms with target Y l. It satisfies condition a) because T does and Y • is split. Sim- ilarly, the morphism Ym,K ′ → coskm−1(Y•,K ′)m is proper and surjective because the morphism TK ′ → coskm−1(Y ′′ •,K ′)m is proper and surjective. Thus the m-truncated simplicial scheme Y•,K ′ is an m-truncated proper hypercovering of XK ′. Finally, condition c) is satisfied thanks to the induction hypothesis. (cid:3) 2.3. We recall how to associate cohomological invariants to simplicial schemes and truncated simplicial schemes (see [SGA 4 II, Vbis, 2.3], [De74, 5.2], [Ts98, (6.2)]). If T is a topos, we denote by T ∆ (resp. T ∆[m]) the topos of cosimplicial objects (resp. m-truncated cosimplicial objects) in T . Let A be a ring in T , and A• the constant cosimplicial ring defined by A. If E • is an A•-module of T ∆ (resp. T ∆[m]), one associates to E • the complex ε∗E • = E 0 → E 1 → · · · → E r (resp. εm ∗ E • = E 0 → E 1 → · · · → E m → 0 → · · · ). Pj(−1)j ∂j −−−−−−−→ E r+1 → · · · One views ε∗E • (resp. εm ∗ E •) as a filtered complex of A-modules using the naive fil- tration. The functors ε∗ and εm ∗ are exact functors from the category of A•-modules to the category of filtered complexes of A-modules (which means that they transform a short exact sequence of A•-modules into a short exact sequence of filtered com- plexes, i.e., such that the sequence of Fili's is exact for all i). Hence, they factorize 12 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING ∗ from D+(T ∆, A•) (resp. D+(T ∆[m], A•)) so as to define exact functors Rε∗ and Rεm to D+F (T , A). For any complex E •,• ∈ D+(A•), they provide functorial spectral sequences (2.3.1) Er,q 1 = Hq(E r,•) ⇒ Hr+q(Rε∗(E •,•)) and similarly for Rεm 1 = 0 for r > m (we use here the first index to denote the simplicial degree). Note that the truncation functor induces a functorial morphism ∗ with Er,q (2.3.2) Rε∗(E •,•) −→ Rεm ∗ (skm(E •,•)), and therefore a morphism between the corresponding spectral sequences (2.3.1). It follows that, if Hq(E r,•) = 0 for q < 0 and all r, then the morphism (2.3.2) is a quasi-isomorphism in degrees < m. Let Y• be a simplicial scheme (resp. m-truncated simplicial scheme), and Sets the topos of sets. If R is a commutative ring, and E • a (Zariski, ´etale, . . . ) sheaf of R- modules on Y•, one can associate to E • a cosimplicial R•-module Γ•(Y•, E •) ∈ Sets∆ (resp. Sets∆[m]) by setting for all r ≥ 0 Γr(Y•, E •) = Γ(Yr, E r). The functor Γ• can be derived, and its right derived functor RΓ• can be computed using resolutions by complexes I •,• such that, for each r, q, the sheaf I r,q is acyclic on Yr. The cohomology of Y• with coefficients in a complex E •,• is then by definition RΓ(Y•, E •,•) = Rε∗RΓ•(Y•, E •,•) (resp. Rεm ∗ ), H q(Y•, E •,•) = H q(RΓ(Y•, E •,•)). If Y• is a smooth simplicial (resp. m-truncated simplicial) R-scheme, this can be applied to the complex Ω• Y•/R, defining the naive filtration. This provides the definition of the de Rham cohomology of Y•, and of its Hodge filtration. Y•/R and to its sub-complexes σ≥iΩ• Proposition 2.4. Let K be a field of charactristic 0, X a proper and smooth K- scheme, Y• → X an m-truncated proper hypercovering of X over K such that Yr is proper and smooth for all r. Then, for all q < m, the canonical homomorphism (2.4.1) H q(X, Ω• X/K ) −→ H q(Y•, Ω• Y•/K ) is an isomorphism of filtered K-vector spaces for the Hodge filtrations. Proof. Since algebraic de Rham cohomology (endowed with the Hodge filtration) commutes with base field extensions, standard limit arguments allow to assume that K is of finite type over Q. Choosing an embedding ι : K ֒→ C, we are reduced to the case where K = C. Using resolution of singularities, we can find a proper and smooth hypercovering Z• of X such that skm(Z•) = Y•. As the morphism (2.3.2) for σ≥iΩ• Z•/C is a quasi-isomorphism in degrees < m for all i, it suffices to prove the proposition with Y• replaced by Z•. This now follows from [De74, Prop. (8.2.2)]. (cid:3) RATIONAL POINTS OF REGULAR MODELS 13 Corollary 2.5. Under the assumptions a) and b) of Lemma 2.2, assume in addition that XK is proper and smooth, and that H q(XK , OXK ) = 0 for some q < m. Then the smallest Hodge slope of H q(Y•K ′, Ω• Y•K′ ) is at least 1. Proof. Assumption a) and b) imply that the hypotheses of the proposition are sat- isfied by Y•K ′ → XK ′, and the corollary is then clear. (cid:3) Y • = lim←−n WnΩ• Y •. 2.6. Let Σn, Σ be as in the proof of Theorem 2.1. We now denote by Y • = (Y•, MY•) an m-truncated simplicial log scheme over Σ1. We assume that each Y r is smooth of Cartier type over Σ1, so that, for all n ≥ 1, its de Rham-Witt complex WnΩ• Y r is defined [HK94, (4.1)]. When r varies, the functoriality of the de Rham-Witt Y • on complex turns the family of complexes (WnΩ• Y•. One defines its cohomology as in 2.3, and one has similar definitions for the de Rham-Witt complex WΩ• Y r )0≤r≤m into a complex WnΩ• For a morphism α : [r] → [s] in ∆[m], let αcrys : (Y s/Σn)crys → (Y r/Σn)crys be the morphism between the log crystalline topos induced by the corresponding morphism Y s → Y r. One defines the log crystalline topos (Y •/Σn)crys as being the topos of families of sheaves (Er)0≤r≤m, where Er is a sheaf on the log crystalline site Crys(Y r/Σn), endowed with a transitive family of morphisms α−1 crysEr → Es for morphisms α in ∆[m]. In particular, the family of sheaves OY r/Σn defines the structural sheaf of (Y •/Σn)crys, denoted by OY •/Σn. There is a canonical morphism uY •/Σn : (Y •/Σn)crys → Y•Zar, such that uY •/Σn∗(E •)r = uY r/Σn∗(Er) for all r. If E •,• is a complex of abelian sheaves in (Y •/Σn)crys, one proceeds as in 2.3 to define its log crystalline cohomology RΓcrys(Y •/Σn, E •,•) and its projection on the Zariski topos RuY •/Σn∗(E •,•). One gives similar definitions for the log crystalline topos (Y •/Σ)crys relative to Σ. By construction, there are canonical isomorphisms (2.6.1) (2.6.2) RΓ(Y•, RuY •/Σn∗(E •,•)) RΓ(Y•, RuY •/Σ ∗(E •,•)) ∼ −−→ RΓcrys(Y •/Σn, E •,•), ∼ −−→ RΓcrys(Y •/Σ, E •,•). Y • (P •) on Y •, and one can form the de Rham complex P log If Y • ֒→ P • is a closed immersion of the m-truncated simplicial log scheme Y • into a smooth m-truncated simplicial Σn-log scheme P • (resp. Σ-formal log scheme), the family of PD-envelopes P log (P r) (resp. completed PD-envelopes) [Ka89, (5.4)] de- Y r fines a sheaf P log Y • (P •) ⊗OP• Ω• P •/Σ), which is supported in Y •. Because the lin- earization functor L used in the proof of the comparison theorem between crystalline and de Rham cohomologies [Ka89, (6.9)] makes sense simplicially, this theorem ex- tends to the simplicial case and there is a canonical isomorphism in D+(Y•, Wn) (resp. D+(Y•, W )) Y • (P •) ⊗OP• Ω• (resp. P log P •/Σn (2.6.3) (2.6.4) RuY •/Σn∗(OY •/Σn) (resp. RuY •/Σ ∗(OY •/Σ) ∼ −−→ P log ∼ −−→ P log Y • (P •) ⊗OP • Ω• Y • (P •) ⊗OP• Ω• P •/Σn P •/Σ ). Proposition 2.7. With the hypotheses of 2.6, assume that Y • is split. Then there exists in D+(Y•, Wn) (resp. D+(Y•, W )) canonical isomorphisms compatible with the 14 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING transition morphisms and the Frobenius actions (2.7.1) (2.7.2) RuY •/Σn∗(OY •/Σn) (resp. RuY •/Σ ∗(OY •/Σ) ∼ −−→ WnΩ• Y • ∼ −−→ WΩ• Y • ). The proof will use the following lemma, due to Nakkajima [Na09, Lemma 6.1]. Lemma 2.8. Under the assumptions of 2.7, there exists an m-truncated simplicial log scheme Z • and a morphism of m-truncated simplicial log schemes Z • → Y • such that, for 0 ≤ r ≤ m, Zr is a disjoint union of affine open subsets of Yr covering Yr, and the morphism Zr → Yr induces the natural inclusion on each of these subsets. Definition 2.9. Let X be a scheme on which p is locally nilpotent, and n ≥ 1 an integer. We denote by X the topological space underlying X, and by Wn(X) the ringed space (X, Wn(OX )), which is a scheme ([Il79, 0, 1.5] and [LZ04, 1.10]). The ideal V Wn−1(OX ) carries a canonical PD-structure ([Il79, 0, 1.4] and [LZ04, 1.1]), which turns the nilpotent immersion u : X ֒→ Wn(X) into a PD-thickening of X. If X = (X, MX ) is a log scheme, we denote by Wn(X) = (Wn(X), MWn(X)) the log scheme obtained by sending MX to Wn(OX ) by the Teichmuller representative map, and taking the associated log structure [HK94, Def. (3.1)]. The immersion u is then in a natural way an exact closed immersion u : X ֒→ Wn(X), functorial with respect to X. Lemma 2.10. Under the assumptions of 2.7, there exists a bisimplicial log scheme Z •,•, m-truncated with respect to the first index and augmented towards Y • with respect to the second index, a bisimplicial formal log scheme T •,• over Σ, m-truncated with respect to the first index, and a closed immersion of bisimplicial formal log schemes i•,• : Z •,• ֒→ T •,•, such that the following conditions are satisfied: a) For 0 ≤ r ≤ m, Zr,0 is a disjoint union of affine open subsets of Yr cov- ering Yr, the augmentation morphism Zr,0 → Yr induces the natural inclusion on each of these subsets, and the canonical morphism Z r,• → coskY r 0 (Z r,•)) is an isomorphism. 0 (skY r b) For 0 ≤ r ≤ m and t ≥ 0, the formal log scheme T r,t is smooth over Σ (i.e., its reduction mod pn is smooth over Σn for all n), and the canonical morphism T r,• → coskΣ 0 (T r,•)) is an isomorphism. 0 (skΣ c) Let i•,•;n : Z •,• ֒→ T •,•;n be the reduction mod pn of i•,•, and let u•,•;n : Z •,• ֒→ Wn(Z •,•) denote the morphism of bisimplicial log schemes defined by the canonical immersions. For variable n, there exists a compatible family of Σn-morphisms of bisimplicial schemes h•,•;n : Wn(Z •,•) → T •,•;n such that h•,•;n ◦ u•,•;n = i•,•;n. with Z α Proof. Let j• : Z • → Y • be a morphism of m-truncated simplicial log schemes r , satisfying the conclusions of Lemma 2.8. One chooses a decomposition Z r =`α Z α r ⊂ Yr open affine such that jrZ α r is affine and smooth over Σ1, and Σn−1 ֒→ Σn is a nilpotent exact closed immersion, there exists for each r, α and each n ≥ 2 a smooth ∼ log scheme Zα −→ Σn−1 ×Σn Zα r;n [Ka89, Prop. (3.14) (1)]. Taking limits when n → ∞, we obtain a smooth formal log r;n over Σn endowed with an isomorphism Zα r is the natural inclusion. r;n−1 Let Z α r;1 = Z α r . Since Zα RATIONAL POINTS OF REGULAR MODELS 15 scheme Z α of Z α Σn-morphisms gα Zα r over Σ and an isomorphism Zα r r . Moreover, the smoothness r;n over Σn for all n implies that we can find inductively a compatible family of r ) → r;n such that the composition Z α r ֒→ Wn(Z α r;n : Wn(Zα r ) → Zα ∼ −→ Σ1 ×Σ Z α r ֒→ Zα r ֒→ Z α r;n and Z α r;n is the chosen immersion Z α r;n, Z r = `α Z α Let Zr;n = `α Z α r;n. r , let vr;n : Zr ֒→ Z r;n, vr : Zr ֒→ Z r be defined by the immersions Zα r , and let gr;n : Wn(Z r) → Zr;n be defined by the morphisms gα r;n. We now use the method of Chiarellotto and Tsuzuki ([CT03, 11.2], [Tz04, 7.3]) to deduce from these data a closed immersion i• of Z • into an m-truncated simplicial formal log scheme T •, smooth over Σ, with reduction T •;n over Σn, and a compatible family of Σn-morphisms of m-truncated simplicial log schemes h•;n : Wn(Z •) → T •;n such that h•;n ◦ u•;n = i•;n, where u•;n : Z •;n ֒→ Wn(Z •;n) is the canonical morphism, and i•;n is the reduction mod pn of i•. First, we set for 0 ≤ s ≤ m r ֒→ Z α Γs(Z r) = Yγ:[r]→[s] Z r,γ, where the product is taken over Σ and indexed by the set of morphisms γ : [r] → [s] in ∆[m], and Z r,γ = Z r for all γ. Then any morphism η : [s′] → [s] in ∆[m] defines a morphism Γs(Z r) → Γs′(Z r) having as component of index γ′ the projection of Γs(Z r) to the factor of index η◦γ′. One obtains in this way an m-truncated simplicial formal log scheme Γ•(Z r) over Σ, the terms of which are smooth over Σ. For each γ : [r] → [s], there is a commutative diagram Wn(Z s) Wn(γ) / Wn(Z r) us;n Z s ur;n γ / Z r gr;n &LLLLLLLLLLL vr;n / Z r;n  / Z r. For fixed r and variable s, the family of morphisms Zs → Γs(Z r) having the compo- γ −→ Zr ֒→ Z r as component of index γ defines a morphism of m-truncated sition Z s simplicial formal log schemes Z • → Γ•(Z r). We set T • = Y0≤r≤m Γ•(Z r), and we define i• : Z • → T • as having the previous morphism as component of index r, for 0 ≤ r ≤ m. For each r, the morphism Z r → Γr(Z r) has the closed immersion vr : Zr ֒→ Z r as component of index Id[r]. It follows that Z r → T r is a closed immersion for all r. Wn(γ) −−−−→ Wn(Z r) Similarly, the family of morphisms Wn(Z s) → Γs(Z r) having the composition gr;n−−→ Zr;n ֒→ Z r as component of index γ defines a mor- Wn(Zs) phism of m-truncated simplicial log schemes Wn(Z •) → Γ•(Z r). We define h• : Wn(Z •) → T • as having the previous morphism as component of index r for 0 ≤ r ≤ m, and h•;n : Wn(Z •) → T •;n as being the reduction of h• mod pn. It is clear that h•;n ◦ u•;n = i•;n, and that the morphisms h•;n form a compatible family when n varies. / & ?  O O / ?  O O   /  / 16 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING We now set Z •,0 = Z •, T •,0 = T •, and we define Z •,• = coskY • 0 (Z •,0), T •,• = coskΣ 0 (T •,0), the coskeletons being taken respectively in the category of simplicial m-truncated simplicial log schemes over Y • and of simplicial m-truncated simplicial formal log schemes over Σ. The augmentation morphism Z •,0 → Y • is given by j•, and the morphism i•,• is defined by setting i•,0 = i• : Z •,0 ֒→ T •,0, and extending i•,0 by functoriality to the coskeletons. As seen above, i•,0 is a closed immersion, and it follows from the construction of coskeletons that i•,t is a closed immersion for all t. Since coskΣ 0 (T r,0)t = T r ×Σ × · · · ×Σ T r (t + 1 times), T r,t is smooth over Σ for all r, t. Finally, we define h•,•;n : Wn(Z •,•) → T •,•;n as being the composition Wn(coskY • 0 (Z •,0)) → coskWn(Y •) 0 (Wn(Z •,0)) → coskΣn 0 (T •,0;n) ≃ Σn ×Σ coskΣ 0 (T •,0), where the first map is defined by the universal property of the coskeleton (and is actually an isomorphism), the second one is defined by functoriality by the morphism h•;n : Wn(Z •,0) → T •;n = T •,0;n, and the last one is the base change isomorphism for coskeletons. The relations h•,•;n ◦ u•,•;n = i•,•;n and the compatibility for variable n follow from the similar properties for the morphisms h•;n. Properties a) - c) of the Lemma are then satisfied. (cid:3) 2.11. Proof of Proposition 2.7. Let  i•,• Z •,• T •,• j•,• Y • / Σ be a commutative diagram satisfying the properties of Lemma 2.10. Since, for all r ≤ m, the morphism jr,0 is locally an open immersion, the scheme underlying Z r,t is the usual fibred product Zr,0 ×Yr · · · ×Yr Zr,0 (t + 1 times). Keeping the notations of the proof of Lemma 2.10, let Ur = (Z α r )α be an affine covering of Yr such that r is the natural inclusion. Then, for any abelian sheaf E on r and jr,0Z α Zr,0 =`α Z α Yr, the complex εr ∗(jr,• ∗j−1 r,• E) =hjr,0 ∗j−1 r,0 E → · · · → jr,t ∗j−1 r,t E Pk(−1)k∂k −−−−−−−→ jr,t+1 ∗j−1 r,t+1E → · · ·i is the Cech resolution of E defined by the covering Ur. If E • is an abelian sheaf on Y•, the fact that j•,0 is an augmentation morphism in the category of m-truncated simplicial schemes implies that the complex εr ∗(jr,• ∗j−1 r,• E r) is functorial with respect to [r] ∈ ∆[m], and we obtain a resolution ε• ∗(j•,• ∗j−1 •,• E •) of E • in the category of abelian sheaves on Y•. In particular, taking into account that each jq,q′ is locally an open immersion, we obtain for all n a resolution of the de Rham-Witt complex of Y • given by (2.11.1) WnΩ• Y • qis −−→ ε• ∗(j•,•∗WnΩ• Z •,•).  / /     / RATIONAL POINTS OF REGULAR MODELS 17 On the other hand, one can also define for all r a complex on Crys(Y r/Σn) by setting εr ∗(jr,• crys ∗(OZ r,•/Σn)) = hjr,0 crys ∗(OZ r,0/Σn) → · · · → jr,t crys ∗(OZ r,t/Σn) Pk(−1)k ∂k −−−−−−−→ · · ·i. Since Zr,• → Yr is the Cech simplicial scheme defined by an affine open covering of Yr, this complex is a resolution of OY r/Σn [Be74, III, Prop. 3.1.2 and V, Prop. 3.1.2]. Since Z •,• is a bisimplicial scheme, these resolutions are functorial with respect to [r] and yield a resolution ε• ∗(j•,• crys ∗(OZ •,•/Σn)) of OY •/Σn. Let T •,•;n be the reduction mod pn of T •,•. The linearization functor L [Ka89, (6.9)] is functorial with ) on Crys(Z •,•/Σn). respect to embeddings, hence it provides a complex L(Ω• This complex is a resolution of OZ •,•/Σn thanks to the log Poincar´e lemma which follows from [Ka89, Prop. (6.5)]. For each (r, t) and each i, one checks easily that the term jr,t crys ∗(L(Ωi )) is acyclic with respect to uY r/Σn∗ (use [Be74, V, (2.2.3)] and the equality uY r/Σn∗ ◦ jr,t crys ∗ = jr,t ∗ ◦ uZr,t/Σn∗). Hence, the complex ))) is an uY •/Σn∗-acyclic resolution of OY •/Σn. Moreover, ε• ∗(j•,• crys ∗(L(Ω• the closed immersion of bisimplicial schemes i•,• defines a family of PD-envelopes P log Z •,•(T •,•;n), supported in Z•,•. They provide a de Rham complex P log Z •,•(T •,•;n) ⊗ , which can be viewed as a complex of abelian sheaves on Z•,•, and it Ω• follows from [Be74, V, (2.2.3)] that T •,•;n/Σn T •,•;n/Σn T •,•;n/Σn T r,t;n/Σn uY •/Σn∗(j•,• crys ∗(L(Ω• T •,•;n/Σn))) = j•,• ∗(P log Z •,•(T •,•;n) ⊗ Ω• T •,•;n/Σn). As the jr,t's are affine morphisms (as a consequence of 1) in our general conventions), we finally get in D+(Z•, Wn) an isomorphism (2.11.2) RuY •/Σn∗(OY •/Σn) ∼ −−→ ε• ∗(j•,• ∗(P log Z •,•(T •,•;n) ⊗ Ω• T •,•;n/Σn)). Z r,t T r,t;n/Σn (T r,t;n)⊗Ωi Zr,t and P log To prove Proposition 2.7, it suffices to define a quasi-isomorphism between the right hand sides of (2.11.1) and (2.11.2). Note that, for each r, t, i, the sheaves WnΩi are jr,t ∗-acyclic. Indeed, Zr,t is a disjoint union of affine open subsets of Yr, and on the one hand WnΩi Zr,t has a finite filtration with subquotients which are coherent over suitable Frobenius pullbacks of Zr,t [HK94, Th. (4.4)], on the other hand P log is a quasi-coherent OTr,t;n- Z r,t module with support in Zr,t, hence is a direct limit of submodules which have a finite filtration with subquotients which are coherent over Zr,t. Therefore, it suffices to construct a quasi-isomorphism (T r,t;n) ⊗ Ωi T r,t;n/Σn (2.11.3) P log Z •,•(T •,•;n) ⊗ Ω• T •,•;n/Σn −→ WnΩ• Z •,• in the category of complexes of Wn-modules over Z•,•. We can now argue as in the proof of [HK94, Th. (4.19)]. Since the PD-immersion ur,t;n : Z r,t ֒→ Wn(Z r,t) is an exact closed immersion for all r, t, the morphism h•,•;n : Wn(Z •,•) → T •,•;n defines uniquely a PD-morphism P log Z •,•(T •,•;n) → Wn(OZ •,•) in the category of sheaves of W -modules on the bisimplicial scheme T•,•;n. As h•,•;n is a morphism of bisimplicial log schemes, it defines by functoriality a morphism 18 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING of complexes Ω• morphism of complexes with support in Z•,• Wn(Z •,•)/Σn T •,•;n/Σn → Ω• on T•,•;n. This morphism extends as a P log Z •,•(T •,•;n) ⊗ Ω• T •,•;n/Σn −→ Ω• Wn(Z •,•)/Σn/N • •,•, Wn(Z •,•)/Σn •,• ⊂ Ω• where N • denotes the graded ideal generated by the sections d(a[i]) − a[i−1]da for all sections a of V Wn−1(OZ•,•) and all i ≥ 1. The differen- tial graded algebra WnΩ• [HK94, Prop. (4.7)], and the generators of N • Z •,• is p-torsion free), so we finally get the morphism (2.11.3). To check that it is a quasi-isomorphism, it suffice to do so on each Zr,t, and this follows from [HK94, Th. (4.19)]. We obtain in this way the isomorphism (2.7.1). Wn(Z •,•)/Σn Z •,• (because WΩ• Z •,• is a quotient of Ω• •,• vanish in WnΩ• •,•, h′′ Z ′′ •,• and h′′ To construct the isomorphism (2.7.2), it suffices to observe that the compatibility of the previous constructions when n varies implies that they make sense in the category of inverse systems indexed by n ∈ N. Then one can apply the functor R lim←−n to the isomorphism (2.7.1) viewed an an isomorphism in the derived category of inverse systems of sheaves of W -modules on Y•, and this provides the isomorphism (2.7.2), since the local structure of the WnΩi recalled above implies that they form a lim←−-acyclic inverse system. The isomorphisms (2.7.1) and (2.7.2) do not depend upon the choices made in their construction. If (Z •,•, T •,•, j•,•, i•,•, h•,•;n) and (Z ′ •,•;n) are two sets of data provided by Lemma 2.10, one can construct a third set of data (Z ′′ •,•, T ′ •,•, h′ •,•, j′ •,•, i′ •,•, T ′′ •,•, j′′ •,•, i′′ •,•;n) mapping to the two previous ones by setting •,• = Z •,• ×Y • Z′ •,• = T •,• ×Σ T ′ •,•, •,•, T ′′ •,•, i′′ •,•, T ′ •,•, T ′′ and defining j′′ •,•;n by functoriality. Then the independence property of (2.7.1) and (2.7.2) follows from the functoriality of the canonical isomorphisms used in their construction with respect to the projections from (Z ′′ •,•) to (Z •,•, T •,•) and (Z ′ •,•). Moreover, one can also prove the functoriality of (2.7.1) and (2.7.2) with respect to Y • by similar arguments using the graph construction: for a mor- phism ϕ• : Y ′ • → Y • between two m-truncated simplicial log schemes satisfying the assumptions of Lemma 2.7, one can find sets of data (Z •,•, T •,•, j•,•, i•,•, h•,•;n) •,•, j′ and (Z′ •,•, T ′ •,•;n) satisfying the conditions of Lemma 2.10 relatively to Y • and Y ′ •, and such that there exists morphisms of bisimplicial log schemes ψ•,• : Z ′ •,• → T •,• satisfying the obvious compatibilities. Then the functoriality of (2.7.1) and (2.7.2) with respect to ϕ• follows from the functo- riality of the canonical isomorphisms used in their construction with respect to ϕ•, ψ•,• and θ•,•. In particular, one obtains in this way that the isomorphisms (2.7.1) and (2.7.2) are compatible with the Frobenius actions. (cid:3) •,• → Z •,•, θ•,• : T ′ •,•, h′ •,•, i′ 2.12. Proof of Theorem 1.3, assuming Theorem 1.5. To conclude this section, we prove that Theorem 1.5 implies Theorem 1.3. We keep the notations of 1.1, and we first observe that if Theorem 1.3 holds when R is complete, then it holds in general. Indeed, let bR be the completion of R, and eX = X bR. Then eX is a regular scheme: on the one hand, its generic fibre is smooth over bK = Frac(bR); on the other hand, its special fibre is isomorphic to Xk, and the completions of the local rings of X and RATIONAL POINTS OF REGULAR MODELS 19 implies the theorem for X. eX are isomorphic at any corresponding points of their special fibres. It follows that eX satisfies the assumptions of Theorem 1.3 relatively to bR, and the theorem for eX Therefore, we assume in the rest of the proof that R is complete. We fix an integer m > q. Let K ′ be a finite extension of K, with ring of integers R′ and residue field k′, such that there exists an m-truncated simplicial log scheme Y • over R′, with an augmentation morphism u : Y0 → XR′ , such that properties a) - c) of Lemma 2.2 are satisfied. Let W ′ n, Σ′ be the log schemes defined by W ′ n = Wn(k′), W ′ = W (k′), K ′ 0 = Frac(W ′), and let Σ′ n, W ′ as in 2.1. Thanks to property a) of Lemma 2.2, the log schemes (Y r)k′ are smooth of Cartier type over Σ′ 1. Therefore, we can consider the log crystalline cohomology of Y • k′ RΓcrys(Y • k′/Σ′, OY • k′ /Σ′) := Rεm crys(Y • k′/Σ′, OY • k′ /Σ′), ∗ RΓ• as defined in 2.6. Using the naive filtration on the functor Rεm ∗ (see 2.3), its basic properties follow from those of the log crystalline cohomology of the proper and smooth log schemes (Yr)k′. In particular, since Yr is proper over Σ′ 1 for all r, the complex RΓcrys(Y • k′/Σ′, OY • k′ /Σ′) is a perfect complex of W ′-modules, and the cohomology space H q 0-vector space. By functoriality, it is endowed with the semi-linear Frobenius action defined by the absolute Frobenius endomorphism of Y • k′. crys(Y • k′/Σ′, OY • k′ /Σ′) ⊗ K ′ 0 is a finite dimensional K ′ From (2.6.2) and (2.7.2), we deduce an isomorphism H q crys(Y • k′/Σ′, OY • k′ /Σ′) ⊗ K ′ 0 ∼ −−→ H q(Y • k′, WΩ• Y • k′ ) ⊗ K ′ 0, which is compatible with the Frobenius actions thanks to Proposition 2.7. The fil- tration of the complex WΩ• Y • k′ provides a spectral sequence Y • k′ by the subcomplexes σ≥iWΩ• Ei,j 1 = H j(Y • k′, WΩi Y • k′ ) ⊗ K ′ 0 =⇒ H i+j crys(Y • k′/Σ′, OY • k′ /Σ′) ⊗ K ′ 0, which is endowed with a Frobenius action. Using the naive filtration on Rεm ∗ , we deduce from the case of a single log scheme that each term Ei,j is a finite dimensional 1 K0-vector space on which the Frobenius action is bijective with slopes in [i, i + 1[. Therefore the spectral sequence degenerates at E1, and, taking (2.1.1) into account, we get in particular an isomorphism (2.12.1) (H q crys(Y • k′/Σ′, OY • k′ /Σ′) ⊗ K ′ 0)<1 ∼ −−→ H q(Y• k′, W OY• k′ ,Q). Since Y • satisfies property a) of 2.2, the construction of the monodromy operator N on log crystalline cohomology can be extended to the case of Y • k′ [Ts98, (6.3)]. Moreover, the Hyodo-Kato isomorphism ρ can also be extended to the case of Y • k′ [Ts98, (6.3.2)], providing an isomorphism ρ : H q crys(Y • k′/Σ′, OY • k′ /Σ′) ⊗ K ′ ∼ −−→ H q(Y• K ′, Ω• Y• K′ ). (2.12.2) Thus, H q crys(Y • k′/Σ′, OY • k′ /Σ′) ⊗ K ′ inherits a filtered (ϕ, N )-module structure. It follows from [Ts98, Th. 7.1.1] (generalizing [Ts99, Th. 0.2]) that, endowed with 0 is an admissible filtered (ϕ, N )-module, ´et(Y• K, Qp). Therefore it is weakly 0 = 0, or its smallest this structure, H q corresponding to the Galois representation H q admissible. In particular, either H q crys(Y • k′/Σ′, OY • k′ /Σ′)⊗K ′ crys(Y • k′/Σ′, OY • k′ /Σ′) ⊗ K ′ 20 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING Newton slope is greater or equal to its smallest Hodge slope. Since H q(XK , OXK ) = 0, Corollary 2.5 implies that the smallest Hodge slope is at least 1. Therefore the part of Newton slope < 1 vanishes. By (2.12.1), we obtain (2.12.3) H q(Y• k′, W OY• k′ ,Q) = 0. As Y• → XR′ satisfies property 2.2 c), there exists a sub-K-extension K1 ⊂ K ′, with ring of integers R1 and residue field k1, a semi-stable scheme Y over R1, a projective R-alteration f : Y → X, and finitely many R-embeddings σi : R1 ֒→ K ′ such that, if u1 : Y → XR1 denotes the R-morphism defined by f , and if Yσi (resp. uσi : Yσi → XR′ ) denotes the R′-scheme (resp. R′-morphism) deduced by base change via σi from Y (resp. u1), then Y0 =`i Yσi, and the augmentation morphism = uσi. This provides a commutative diagram u : Y0 → XR′ is defined by uYσi (2.12.4) Y0,k′ =`i Yσi,k′ s0 4hhhhhhh *VVVVVVVV Y• k′ u• k′ / Xk′ Yk1 / Yk 'PPPPPP 7oooooo fk Xk in which we identify schemes with their Zariski topos, Yk := Spec k ×Spec R Y , and: • k′E is the family the morphism u• k′ is such that, for any sheaf E on Xk′, u−1 (i) of sheaves (ur)−1 k′ E, with ur : Yr → XR′ defined by the augmentation morphism, the morphism s0 is such that, for any sheaf F • on Y• k′, s−1 the morphism Yσi,k′ → Yk1 is the projection corresponding to σi. (ii) (iii) By functoriality, we obtain a commutative diagram for the corresponding Witt 0 F • = F 0, cohomology spaces (2.12.5) H q(Xk′, W OXk′ ,Q) / H q(Y• k′, W OY• k′ ,Q) H q(Yk, W OYk,Q) ∼ / H q(Yk1, W OYk1 ,Q) In this diagram, the lower horizontal arrow is an isomorphism because Yk1 ֒→ Yk is a nilpotent immersion [BBE07, Prop. 2.1 (i)]. The lower right arrow is injective on each summand, because each σi turns k′ into a finite separable extension of k1, hence it follows from [Il79, 0, Prop. 1.5.8] that ∼ W (k′) ⊗W (k1) Γ(U, W OYk1 ) −−→ Γ(Uσi , W OYσi,k′ ) for any affine open subset U ⊂ Yk1 with inverse image Uσi ⊂ Yσi,k′ ; as one can compute Witt cohomology using Cech cohomology, this implies that −−→ H q(Yσi,k′, W OYσi,k′ ). W (k′) ⊗W (k1) H q(Yk1, W OYk1 ∼ ) Finally, f : Y → X is a projective alteration between two flat regular schemes of finite type over R, so Theorem 1.5 implies that f ∗ k is injective. Therefore, the 5kkkkkkkkkk H q(Xk, W OXk,Q) )SSSSSSSSSS f ∗ k *UUUUUUUUUUU Li H q(Yσi,k′, W OYσi,k′ ,Q). 4iiiiiiiiiii / ' 4 *   / 7 / * 5 ) / '  4 RATIONAL POINTS OF REGULAR MODELS 21 functoriality map H q(Xk, W OXk,Q) → Li H q(Yσi,k′, W OYσi,k′ ,Q) is injective. But (2.12.3) implies that the composition of the upper path in the diagram is 0. follows that H q(Xk, W OXk,Q) = 0. It (cid:3) 3. An injectivity theorem for coherent cohomology We now begin our preliminary work in view of the proof of Theorem 1.5. One of the key ingredients in this proof is a theorem which bounds the order of elements in the kernel of the functoriality map induced on coherent cohomology by a proper surjective complete intersection morphism f : Y → X of virtual relative dimension 0. Such a result is a consequence of the existence of a "trace morphism" τf : Rf∗OY → OX which satisfies the properties stated in the following theorem: Theorem 3.1. Let X be a noetherian scheme with a dualizing complex, and let f : Y → X be a proper complete intersection morphism of virtual relative dimension 0. There exists a morphism τf : Rf∗OY → OX which satisfies the following properties: If g : Z → Y is a second proper complete intersection morphism of virtual (i) relative dimension 0, then the composed morphism (3.1.1) R(f ◦ g)∗OZ ∼= Rf∗Rg∗OZ Rf∗(τg ) −−−−−→ Rf∗OY τf−→ OX is equal to τf g. (ii) Let X ′ be another noetherian scheme with a dualizing complex, u : X ′ → X a morphism such that X ′ and Y are Tor-independent over X, and f ′ : Y ′ → X ′ the pull-back of f by u. If f is projective, or if either f is flat, or u is residually stable [Co00, p. 132], then the morphism (3.1.2) Rf ′ ∗OY ′ ∼= Lu∗Rf∗OY Lu∗(τf ) −−−−−→ OX ′, defined by the base change isomorphism (A.1.2), is equal to τf ′. If f is finite and flat, then, for any section b ∈ f∗OY , (iii) (3.1.3) τf (b) = tracef∗OY /OX (b). As explained in the introduction, we refer to B.7 for the definition of τf , and to B.9 for the proof of the theorem. It may be worth recalling a few examples of complete intersection morphisms of virtual relative dimension 0 (in short: ci0): 1) If X and Y are two regular schemes with the same Krull dimension, any morphism f : Y → X which is locally of finite type is ci0. This is the situation where we will use Theorem 3.1 in this article. 2) If X and Y are smooth over a third scheme S, with the same relative dimension, any S-morphism Y → X is ci0. 3) If X is a scheme, Z ֒→ X a regularly embedded closed subscheme, and f : Y → X the blowing up of X along Z, then f is ci0 [SGA 6, VII, Proposition 1.8]. The existence of τf has a remarkable consequence for the functoriality maps in- duced on coherent cohomology. 22 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING Theorem 3.2. Let X be a noetherian scheme with a dualizing complex, and f : Y → X a proper complete intersection morphism of virtual relative dimension 0. Assume that there exists a scheme-theoretically dense open subset U ⊂ X such that f −1(U ) → U is finite locally free of constant rank r ≥ 1. Then, for any complex E • ∈ Db qc(OX ) and any q ≥ 0, the kernel of the functoriality map (3.2.1) H q(X, E •) → H q(Y, Lf ∗E •) is annihilated by r. In particular, when r is invertible on X, the functoriality maps are injective. τf−→ OX is multiplication by r Proof. By 3.1 (iii), the composition OX → Rf∗OY over U . Since U is scheme-theoratically dense in X, it is multiplication by r over X. The complete intersection hypothesis implies that f has finite Tor-dimension, qc(OY ). Moreover, we can apply the projection formula hence Lf ∗E • belongs to Db [SGA 6, III, 3.7] to obtain a commutative diagram E • Rf∗OY &LLLLLLLLLLLLL × r L ⊗OX E • ∼ Rf∗Lf ∗E • τf ⊗Id vnnnnnnnnnnnnnn E • , in which the upper composed morphism is the adjunction morphism. Applying the functors H q(X, −) to the diagram, the theorem follows. (cid:3) 4. Koszul resolutions and local description of the trace morphism τf We recall here some well-known explicit constructions based on the Koszul com- plex which enter in the definition of the trace morphism τf . Later on, this will allow us to define generalizations of τf for sheaves of Witt vectors. As in the whole article, we follow Conrad's constructions and conventions [Co00]. in degree k by 4.1. Let P be a scheme, and let t = (t1, . . . , td) be a regular sequence of sections of OP , defining an ideal I ⊂ OP . We denote by Y ⊂ P the closed subscheme defined by I, and by i : Y ֒→ P the corresponding closed immersion. Classically, the Koszul complex K•(t) defined by the sequence (t1, . . . , td) is the chain complex concentrated in homological degrees [0, d], such that E := K1(t) is a free OP -module of rank d E for all k, and such that the differential is given with basis e1, . . . , ed, Kk(t) =Vk kXj=1 (−1)j−1tij ei1 ∧ . . . ∧ceij ∧ . . . ∧ eik . It is often more convenient to consider K•(t) as a cochain complex concentrated in cohomological degrees [−d, 0], by setting (K•(t))k = K−k(t) and leaving the differential unchanged [Co00, p. 17]. dk(ei1 ∧ . . . ∧ eik ) = / / & / /   v RATIONAL POINTS OF REGULAR MODELS 23 Since t is a regular sequence, K•(t) is a free resolution of OY over OP . For any OP -module M, this resolution provides an isomorphism (4.1.1) Ext d OP (OY , M) := H d(Hom • OP (K•(t), M)) ψt,M−−−→ ∼ Hom OP (Vd IHom OP (Vd E, M) E, M) , where ψt,M is the tautological isomorphism multiplied by (−1)d(d+1)/2 (see [Co00, definition of (1.3.28) and (2.5.2)]). For any section m of M, we will denote by (4.1.2) (cid:20) m t1, . . . , td (cid:21) ∈ Ext d OP (OY , M) the section corresponding by (4.1.1) to the class of the homomorphism ut,m which sends e1 ∧ . . . ∧ ed to (−1)dm (the (−1)d sign being needed to obtain relation (4.5.1) later). Note that this section is linear with respect to m, only depends on the class of m mod IM, and is functorial with respect to M. Its dependence on the regular sequence t is given by the following lemma. 1, . . . , t′ Lemma 4.2. Let t′ = (t′ d) be another regular sequence of sections of OP , generating an ideal I ′ such that I ′ ⊂ I. Let C = (ci,j)1≤i,j≤d be a matrix with entries in OP such that t′ (OP /I, M) → Ext d (OP /I ′, M) is the functoriality homomorphism, then If α : Ext d OP j=1 ci,jtj for all i. i = Pd d (cid:21) . α((cid:20) t1, . . . , td (cid:21)) =(cid:20) det(C)m t′ 1, . . . , t′ m OP (4.2.1) Proof. Let K•(t′) be the Koszul resolution of OP /I ′, and E ′ = K1(t′), with basis 1, . . . , e′ d. One defines a morphism of resolutions φ : K•(t′) → K•(t) by setting e′ φ1(e′ diagram i) =Pj ci,jej, and φk = ∧kφ1 for 0 ≤ k ≤ d. Then φ provides a commutative Hom OP (∧dE, M) Ext d OP (OP /I, M) φd=det(C) α Hom OP (∧dE ′, M) / Ext d OP (OP /I ′, M). The lemma follows. (cid:3) 4.3. Under the assumptions of 4.1, the morphism d1 : E ։ I defines an isomorphism E/IE ∼ −−→ I/I 2. Using the canonical isomorphisms, this provides (4.3.1) Hom OP (Vd IHomOP (Vd E, M) E, M) E)∨ ⊗OY M/IM E)∨/I(Vd ∼ −−→ (Vd −−→ Vd ∼ ∼ ((E/IE)∨) ⊗OY M/IM −−→ ωY /P ⊗OY i∗M. Note that, due to the commutation between dual and exterior power, the composi- 1 ) ⊗ i∗(m), where ¯tk denotes tion (4.3.1) maps the class of ut,m to (−1)d(¯t ∨ the class of tk mod I 2. d ∧ . . . ∧ ¯t ∨ / /     / 24 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING Composing (4.1.1) and (4.3.1), one obtains the fundamental local isomorphism [Ha66, III, 7.2] as defined by Conrad [Co00, (2.5.2)] in the local case: (4.3.2) ηY /P : Ext d OP (OY , M) ∼ −−→ ωY /P ⊗OY i∗M. Applying Lemma 4.2 to the case of two regular sequences of generators of the ideal I, one sees that the isomorphism ηY /P does not depend on the sequence t, so that local constructions can be glued to define ηY /P for any regular immersion i : Y ֒→ P , without assuming that I is defined globally by a regular sequence. One obtains in this way the fundamental local isomorphism in the general case [Co00, (2.5.1)]. Let us now recall from [Co00, 2.5] how the isomorphism (4.3.2) allows to define functorially for any M• ∈ D(OP ) the isomorphism [Co00, (2.5.3)] (4.3.3) ηi : RHom OP (OY , M•) ∼ −−→ ωY /P [−d] L ⊗OP Li∗(M•). Applying [Co00, Lemma 2.1.1] and using the isomorphism of functors defined by ηY /P , one gets the isomorphism RHom OP (OY , M•) ∼ −−→ (ωY /P L ⊗OP Li∗(M•))[−d]. The isomorphism ηi is then obtained by composition with the canonical isomorphism (ωY /P L ⊗OP Li∗(M•))[−d] ∼ −−→ ωY /P [−d] L ⊗OP Li∗(M•) defined by the general convention [Co00, p. 11]. From the discussion p. 53 of [Co00], it follows that ηi satisfies the following properties: a) If M• = M[0] for an OP -module M, then the homomorphism induced by ηi between the cohomology sheaves in degree d is the isomorphism ηY /P . b) The isomorphism ηi commutes with translations in D(OP ) (using the general convention [Co00, (1.3.6)] for the right hand side of (4.3.3)). 4.4. Let π : P → X be a smooth morphism of relative dimension d, i : Y ֒→ P a regular immersion of codimension d, and f = π ◦ i. Let ζ ′ i,π : ωY /X ∼ −−→ ωY /P [−d] L ⊗OY Li∗(ωP/X [d]) be the canonical isomorphism (A.2.6), which induces in degree 0 the tautological isomorphism ζ ′ i,π provided in (A.2.5) by the construction of ωY /X in A.2. Let δf be the canonical section of ωY /X (defined by (A.7.2)), and ϕf : OY → ωY /X the morphism sending 1 to δf . We define the morphism (4.4.1) γf : OY → ωP/X[d] as being the composition ϕf OY ωY /X ζ ′ i,π ∼ / / ωY /P [−d] L ⊗OY Li∗(ωP/X[d]) η−1 i ∼ / / RHom OP (OY , ωP/X[d]) γf can ωP/X[d]. / / / /   RATIONAL POINTS OF REGULAR MODELS 25 Proposition 4.5. Under the assumptions of 4.4, the isomorphism η−1 in the definition of γf induces in degree 0 an isomorphism i ◦ζ ′ i,π entering η−1 i ◦ ζ ′ i,π : ωY /X ∼ −−→ Ext d OP (OY , ωP/X), which is such that (4.5.1) η−1 i ◦ ζ ′ i,π(δf ) =(cid:20) dt1 ∧ · · · ∧ dtd t1, . . . , td (cid:21) . Proof. Let us consider first the isomorphism η−1 Y /P ◦ ζ ′ i,π : ωY /X ∼ −−→ Ext d OP (OY , ωP/X), where ηY /P is defined by (4.3.2). By definition, (cid:20) dt1 ∧ · · · ∧ dtd t1, . . . , td (cid:21) is mapped to ut,dt1∧...∧dtd by (4.1.1), and we observed in 4.3 that ut,dt1∧...∧dtd is mapped to (−1)d(¯t ∨ d ) ⊗ i∗(dtd ∧ . . . ∧ dt1) by construction, we get the relation 1 ) ⊗ i∗(dt1 ∧ . . . ∧ dtd) by (4.3.1). Since ζ ′ i,π(δf ) = (¯t ∨ d ∧ . . . ∧ ¯t ∨ 1 ∧ . . . ∧ ¯t ∨ (4.5.2) Y /P ◦ ζ ′ η−1 i,π(δf ) = (−1)d(cid:20) dt1 ∧ · · · ∧ dtd t1, . . . , td (cid:21) . Let ηi,ωP /X and ηi,ωP /X [d] be the isomorphisms (4.3.3) relative to the complexes ωP/X[0] and ωP/X[d]. By 4.3 a), ηi,ωP /X induces in degree d the isomorphism ηY /P . On the other hand, 4.3 b) shows that ηi,ωP /X [d] is identified with ηi,ωP /X [d] when ∼ using the canonical isomorphisms RHom(OY , ωP/X[d]) −−→ RHom(OY , ωP/X )[d] ⊗ Li∗(ωP/X ))[d]. The first one involves and ωY /P [−d] no sign, and, as ωY /P [−d] is concentrated in degree d, the second one is given by multiplication by (−1)d2 = (−1)d on ωY /P ⊗ i∗(ωP/X ). Thus relation (4.5.2) implies relation (4.5.1). (cid:3) ∼ −−→ (ωY /P [−d] ⊗ Li∗(ωP/X[d]) L L Proposition 4.6. Let X be a separated noetherian scheme with a dualizing complex, P = P d X a projective space over X, π : P → X the structural morphism, i : Y ֒→ P a regular immersion of codimension d, and f = π ◦ i. Then the trace morphism τf : Rf∗OY → OX of Theorem 3.1 is equal to the composition (4.6.1) Rf∗(OY ) Rπ∗(γf ) −−−−−→ Rπ∗(ωP/X[d]) Trpπ−−−→ OX , where Trpπ is the trace morphism for the projective space defined in [Co00, (2.3.1)- (2.3.5)]. 26 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING Proof. By construction (see B.7), τf is the composition Trf ◦ Rf∗(λf ) ◦ Rf∗(ϕf ) in the commutative diagram Rf∗(OY ) Rf∗(ϕf ) Rf∗(ωY /X) Rf∗(ζ ′ i,π ) ≀ Rf∗(λf ) ∼ / Rf∗(f !(OX )) Rf∗(c−1 i,π) ≀ Trf / OX Trπ Rf∗(ωY /P [−d] ⊗L Li∗(ωP/X [d])) Rf∗(i!π!(OX )) Rπ∗(Tri) / Rπ∗(π!(OX )) Rf∗(η−1 i ) ≀ Rf∗(i!(eπ)) ≀ Rπ∗(eπ) ≀ Rf∗(RHom OP (OY , ωP/X [d])) Rf∗(di) ∼ / Rf∗(i!(ωP/X[d])) Rπ∗(Tri) / Rπ∗(ωP/X[d]), in which the isomorphism λf is defined by the commutativity of the left rectangle before applying Rf∗ (cf. B.1), and di, eπ, ci,π are defined as follows: a) di is the canonical isomorphism of functors i♭ := RHom OP (OY , −) ∼ −−→ i!, defined by [Co00, (3.3.19)]; b) eπ is the canonical isomorphism of functors π♯ := ωP/X[d] ⊗L π∗(−) ∼ −−→ π!, defined by [Co00, (3.3.21)]. c) ci,π is the transitivity isomorphism f ! ∼ −−→ i!π!, defined by [Co00, (3.3.14)]. Moreover, the upper right square commutes because of the transitivity of the trace morphism [Co00, 3.4.3, (TRA1)], and the lower right square commutes by functori- ality of the trace morphism Tri with respect to eπ. In this diagram, the composition of the right vertical arrows is the projective trace morphism Trpπ [Co00, 3.4.3, (TRA3)], and the isomorphism di on the bottom row identifies Tri with the trace morphism Trf i for finite morphisms [Co00, 3.4.3, (TRA2)]. As the latter is the canonical morphism i∗RHom OP (OY , −) → Id defined by OP ։ OY , it follows that the composition of the left column and the bottom row of the diagram is equal to Rπ∗(γf ), which proves the proposition. (cid:3) 5. Preliminaries on the relative de Rham-Witt complex We extend here to the relative de Rham-Witt complex constructed by Langer and Zink [LZ04] structure theorems which are classical when the base is a perfect scheme of characteristic p ([Il79], [IR83]). We begin by recalling some basic facts from their construction. From now on, we fix a prime number p. We denote by Z(p) the localization of Z at the prime ideal (p). Although many results of [LZ04] are valid for Z(p)-schemes, we limit our exposition to the case of schemes on which p is locally nilpotent, which will suffice for our applications. 5.1. Let S be a scheme on which p is locally nilpotent, and let f : X → S be a morphism of schemes. An F -V -pro-complex of X/S as defined in [LZ04] is a pro- complex {R : E • n is a differential graded Wn(OX )/f −1Wn(OS)-algebra (i.e., E • n is a graded Wn(OX )-algebra together with an f −1Wn(OS )-linear map d : E • n(1) such that d2 = 0, satisfying ηω = n}n≥1 of sheaves on X, where E • n+1 → E • n → E •     / /   O O / O O / O O / O O RATIONAL POINTS OF REGULAR MODELS 27 (−1)deg ω deg ηωη and d(ωη) = (dω)η + (−1)deg ωωdη for any homogeneous sections ω, η ∈ E • n), which is equipped with a map of graded pro-rings called the Frobenius morphism, and with a map of graded abelian groups F : E • • +1 → E • •, V : E • • → E • • +1, called the Verschiebung morphism, such that the following properties hold: (i) The structure map W•(OX ) → E0 (ii) The following relations hold: • is compatible with F and V . (5.1.1) (5.1.2) (5.1.3) F V = p, F dV = d, V (ωF (η)) = V (ω)η, for all ω ∈ E • n, η ∈ E • n+1, n ≥ 1, F (d[a]) = [a]p−1d[a], for all a ∈ OX , where [a] denotes the Teichmuller lift of a to Wn(OX ), for any n. A morphism between two F -V -pro-complexes of X/S is a map of pro-differential graded W•(OX )/f −1W•(OS )-algebras compatible with F and V . By [LZ04, Prop. 1.6, Rem. 1.10] there exists an initial object in the category of F -V -pro-complexes of X/S, which is called the relative de Rham-Witt complex of X/S and is denoted by {R : Wn+1Ω• X/S is a quasi-coherent sheaf on the scheme Wn(X) := (X, Wn(OX )) defined in 2.9, and the transition morphisms R are epimorphisms. When S is a perfect scheme of characteristic p, the relative de Rham-Witt complex coincides with the one defined in [Il79]. Notice that we have the following properties: X/S}n≥1. Each sheaf WnΩq X/S → WnΩ• WnΩ0 X/S = Wn(OX ), W1Ω• X/S = Ω• X/S, and that, by [LZ04, (1.16), (1.17) and (1.19)], relations (5.1.1) and (5.1.2) imply that (5.1.4) (5.1.5) V (ωdη) = V (ω)dV (η), for all ω, η ∈ WnΩ• X/S, n ≥ 1, V d = pdV, dF = pF d. In addition, when S is an Fp-scheme, the operators F and V satisfy the relation V F = p. We also recall the behaviour of the de Rham-Witt complex with respect to ´etale pull-backs. Let X ′ S′ h g X / S be a commutative diagram in which h is ´etale and g unramified. Then, for all q ≥ 0 and r ≥ n ≥ 1, Wn(X ′) is ´etale over Wn(X) and we have the Wr(OX ′ )-linear / /     / 28 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING isomorphisms (5.1.6) Wr(OX ′) ⊗h−1Wr(OX ) h−1WnΩq X/S ∼ −−→ WnΩq X ′/S′, (5.1.7) Wr(OX ′ ) ⊗h−1Wr(OX ) h−1(F r−n where, for any Wn-module M , F r−n Wr → Wn [LZ04, Prop. 1.11, Prop. A.8 and Cor. A.11]. ∼ −−→ F r−n ∗ WnΩq X/S) ∗ WnΩq X ′/S′, a ⊗ ω 7→ F r−n(a)ω, ∗ M denotes M viewed as a Wr-module via F r−n : X/S := X/S are still epimorphisms. Finally, the completed relative de Rham-Witt complex is defined by WΩ• WnΩ• X/S; the canonical morphisms WΩ• lim←−n 5.2. Let S = Spec A be affine. We want to recall the calculation of WΩq Γ(Ad X/S → WnΩ• S, WΩq Ad A weight is a function k : [1, d] = {1, 2, . . . , d} → Z[ 1 ). We need some notations for this. S /S p ]≥0. We write ki := k(i), for i ∈ [1, d]. The support of k, supp k, consists of those i ∈ [1, d] with ki 6= 0. For any weight k we choose once and for all a total ordering on the elements of the support of k, A[x1,...,xd]/A := (5.2.1) such that: supp k = {i1, . . . , ir}, (i) ordp ki1 ≤ ordp ki2 ≤ · · · ≤ ordp kir . (ii) The ordering on supp k and on supp pak agree, for any a ∈ Z. We say k is integral if ki ∈ Z, for all i ∈ [1, d]. We say k is primitive if it is integral and not all ki are divisible by p. We set (5.2.2) t(ki) := −ordp ki and t(k) :=(max { t(ki) i ∈ supp k } if supp k 6= ∅, if k = 0. 0 If k 6= 0, t(k) is the smallest integer such that pt(k)k is primitive, and we have t(k) = t(ki1) ≥ t(ki2) ≥ · · · ≥ t(kir ). We denote by u(k) the smallest non-negative integer such that pu(k)k is integral, i.e., u(k) = max {0, t(k)}. Notice that k is integral iff u(k) = 0 iff t(k) ≤ 0, and k is primitive iff t(k) = 0. An interval of the support of k is by definition a subset I ⊂ supp k of the form I = {is, is+1, . . . , is+m}. We denote by kI the weight which equals k on I and is zero on [1, d] \ I. If k is fixed and I is an interval of the support of k, we write u(I) := u(kI ) and t(I) := t(kI ). An admissible partition P of length q of supp k (or just of k) is a tuple of intervals of supp k, P = (I0, I1, . . . , Iq), such that: supp k = I0 ⊔ I1 ⊔ . . . ⊔ Iq. (i) (ii) The elements in Ij are smaller than the elements in Ij+1 (with respect to the ordering (5.2.1)) for all j = 0, . . . , q − 1. (iii) The intervals I1, . . . , Iq are non-empty (but I0 may be). Notice that u(k) = u(I0) if I0 6= ∅ and u(k) = u(I1) if I0 = ∅. RATIONAL POINTS OF REGULAR MODELS 29 For any n ≤ ∞, we write Xi := [xi] ∈ Wn(A[x1, . . . , xd]). If k is an integral weight as above, we write X k = X ki1 i1 · · · X kir ir ∈ Wn(A[x1, . . . , xd]). Let k be any weight and η ∈ W (A). We define (5.2.3) and (5.2.4) e0(η, k) := V u(k)(ηX pu(k)k) ∈ W (A[x1, . . . xd]) e1(η, k) :=(dV u(k)(ηX pu(k)k) ηF −t(k)dX pt(k)k if k is not integral if k is integral ∈ WΩ1 A[x1,...,xd]/A. Definition 5.3 (Basic Witt differentials [LZ04, 2.2]). Let k be a weight, P = (I0, I1, . . . , Iq) an admissible partition of k, and ξ = V u(k)(η) ∈ W (A). The basic Witt differential e(ξ, k, P) ∈ WΩq A[x1,...,xd]/A is defined as follows: e(ξ, k, P) :=(e0(η, kI0 )e1(1, kI1 ) · · · e1(1, kIq ) e1(η, kI1 )e1(1, kI2 ) · · · e1(1, kIq ) if I0 6= ∅, if I0 = ∅. Rules 5.4 ([LZ04, Prop. 2.5, Prop. 2.6]). Let k be a weight, P = (I0, I1, . . . , Iq) a partition of k and ξ = V u(k)(η) ∈ W (A). Note that u(k) ≥ 1 when k is not integral, so that one can then define V −1ξ := V u(k)−1η. Then: (i) ρe(ξ, k, P) = e(ρξ, k, P) for all ρ ∈ W (A). e(V −1ξ, pk, P) (ii) F e(ξ, k, P) =(e(F ξ, pk, P) (iii) V e(ξ, k, P) =(e(V ξ, 1 (iv) de(ξ, k, P) = e(pV ξ, 1 p k, P) p k, P) 0 e(ξ, k, (∅, P)) p−t(k)e(ξ, k, (∅, P)) if I0 6= ∅ or k integral, if I0 = ∅ and k not integral if I0 6= ∅ or 1 if I0 = ∅ and 1 p k integral, p k not integral. if I0 = ∅, if I0 6= ∅ and k not integral, if I0 6= ∅ and k integral. Theorem 5.5 ([LZ04, Thm. 2.8]). Every ω ∈ WΩq as A[x1,...,xd]/A can uniquely be written ω =Xk,P e(ξk,P, k, P), where the sum is over all weights k with supp k ≥ q and over all admissible parti- tions of length q of k, and the sum converges in the sense that, for any m ≥ 0, we have ξk,P ∈ V mW (A) for all but finitely many ξk,P. n(η, k) ∈ WnΩ1 For a weight k, n ≥ 1 and η ∈ Wn−u(k)(A) we define e0 n(η, k) ∈ Wn(A[x1, . . . , xd]) and e1 A[x1,...,xd]/A by the same formulas as in (5.2.3) and (5.2.4). For P an admissible partition of length q of k and ξ = V u(k)(η) ∈ Wn(A), we then define en(ξ, k, P) ∈ WnΩq A[x1,...,xd]/A by the same formula as in Definition 5.3 but with ei replaced by ei n, i = 0, 1. 30 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING Corollary 5.6 ([LZ04, Prop. 2.17]). Every ω ∈ WnΩq written as a finite sum A[x1,...,xd]/A may uniquely be ω =Xk,P en(ξk,P , k, P), ξk,P ∈ V u(k)Wn−u(k)(A), where the sum is over all weights k with supp k ≥ q and such that pn−1k is integral and over all admissible partitions P of k of length q. We now assume that S is an Fp-scheme; the absolute Frobenius endomorphism of S will then be denoted FS, or F when no confusion can arise. The following proposition is known if S is perfect (see [IR83, II, (1.2.2)]); a similar result has been proved by M. Olsson when, ´etale locally, S has a flat lifting over Zp to which FS can be lifted [Ol07, Th. 4.2.15]. Proposition 5.7. Let S be a locally noetherian Fp-scheme and X a smooth S- scheme. Then the sequence ∗ OS ⊗Wn+1(OS ) Wn+1Ωq−1 F n X/S (1⊗F n,−1⊗F nd) −−−−−−−−−−→ F n ∗ Ωq−1 X/S ⊕ F n ∗ Ωq X/S dV n+V n −−−−−−→ Wn+1Ωq X/S −→ R∗WnΩq X/S −→ 0 is an exact sequence of Wn+1(OS)-modules. Proof. The question is local, we thus assume S = Spec A, X = Spec B and B is ´etale over B1 = A[x1, . . . , xd]. As WΩ• X/S is an epimorphism, [LZ04, Prop. 2.19] provides the exactness of the second line, and we only have to show that X/S → Wn+1Ω• (∗B/A) : F n ∗ A⊗Wn+1Ωq−1 B/A (1⊗F n,−1⊗F nd) −−−−−−−−−−→ F n ∗ Ωq−1 B/A ⊕F n ∗ Ωq B/A dV n+V n −−−−−−→ Wn+1Ωq B/A is exact. Notice that it is a complex, as for a ∈ A and ω ∈ Wn+1Ωq−1 B/A we have dV n(aF nω) − V n(aF ndω) = 0. Notice also that, if we let W2n+2(B) act through F n+1 : W2n+2(B) → Wn+1(B), the differentials of this complex are W2n+2(B)-linear, since dF n+1 = pn+1F n+1d = 0 in Wn+1. We claim (5.7.1) (∗B/A) = F n+1 ∗ (∗B1/A) ⊗W2n+2(B1) W2n+2(B). Indeed we have the following diagrams (where the tensor products with W2n+2(B) are taken over W2n+2(B1)): F n+1 ∗ (F n ∗ A ⊗ Wn+1Ωq−1 B/A) 1⊗F n / F 2n+1 ∗ Ωq−1 B/A F n+1 ∗ (F n ∗ A ⊗ Wn+1Ωq−1 B1/A) ⊗ W2n+2(B) (1⊗F n)⊗1 F n+1 ∗ (F n ∗ Ωq−1 B1/A) ⊗ W2n+2(B), / / / O O O O RATIONAL POINTS OF REGULAR MODELS 31 F n+1 ∗ (F n ∗ A ⊗ Wn+1Ωq−1 B/A) −1⊗F nd / F 2n+1 ∗ Ωq B/A F n+1 ∗ (F n ∗ A ⊗ Wn+1Ωq−1 B1/A) ⊗ W2n+2(B) (−1⊗F nd)⊗1 F n+1 ∗ (F n ∗ Ωq B1/A) ⊗ W2n+2(B), both with vertical maps (a ⊗ ω) ⊗ b 7→ a ⊗ F n+1(b)ω, η ⊗ b 7→ F 2n+1(b)η, and F n+1 ∗ F 2n+1 ∗ Ωq−1 B/A ⊕ F 2n+1 ∗ Ωq B/A dV n+V n / F n+1 ∗ Wn+1Ωq B/A (F n ∗ Ωq−1 B1/A ⊕ F n ∗ Ωq B1/A) ⊗ W2n+2(B) (dV n+V n)⊗1 ∗ Wn+1Ωq F n+1 B1/A ⊗ W2n+2(B), with vertical maps (η, ω) ⊗ b 7→ (F 2n+1(b)η, F 2n+1(b)ω), ω ⊗ b 7→ F n+1(b)ω. Using again the relation dF n+1 = pn+1F n+1d = 0 in Wn+1, one checks immediately that all three diagrams commute. Now the claim (5.7.1) follows, since the vertical maps are isomorphisms by (5.1.7). As W2n+2(B1) → W2n+2(B) is ´etale [LZ04, Prop. A.8], we are thus reduced to the case B = B1 = A[x1, . . . , xd]. Now take α ∈ Ωq B/A and β ∈ Ωq−1 that there exists an element γ ∈ F n B/A with V n(α) = −dV n(β). We have to show ∗ A ⊗ Wn+1Ωq−1 B/A with (5.7.2) −(1 ⊗ F nd)(γ) = α and (1 ⊗ F n)(γ) = β. By Corollary 5.6 (and keeping the notation used there), we can write α and β uniquely as finite sums (5.7.3) α =Xk,P e1(ξk,P, k, P), β =Xk,Q e1(ηk,Q, k, Q), with ξk,P, ηk,Q ∈ A, where the sums are over all integral weights k and all admissible partitions P = (I0, . . . , Iq) of length q (resp. over all admissible partitions Q = (J0, . . . , Jq−1) of length q − 1). Using the rules 5.4 (iii) and (iv), we obtain (5.7.4) V n(α) = n−1Xi=0 Xk pi primitive and I0=∅ en+1(pn−iV n(ξk,P), k pn , P) + Xk pn integral or I06=∅ en+1(V n(ξk,P ), k pn , P) / / / O O O O / / / O O O O PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING 32 and (5.7.5) − dV n(β) = Xk pn integral and J06=∅ −pt( k pn )en+1(V n(ηk,Q), k pn , (∅, Q)) + Xk pn not integral and J06=∅ −en+1(V n(ηk,Q), k pn , (∅, Q)), where t(k/pn) is defined as in (5.2.2). By the uniqueness of this presentation, and since V n : A → Wn+1(A) is injective, the equality V n(α) = −dV n(β) thus gives the following set of equations: (5.7.6) ξk,P = −p−t( k pn )ηk,Q, if k pn is integral, P = (∅, Q) and J0 6= ∅, ηk,Q = −pn−iξk,P, if ξk,P = 0, if I0 6= ∅. k pi is primitive, P = (∅, Q), J0 6= ∅ and 0 ≤ i ≤ n − 1, We claim that (5.7.2) holds for the following choice of γ ∈ F n ∗ A⊗Wn+1(A) Wn+1Ωq−1 B/A: γ := n−1Xi=0  Xk pi primitive and J06=∅ + Xk primitive pi and J0=∅ (cid:16)−ξk,(∅,Q) ⊗ en+1(V n−i(1), k pn , Q)(cid:17) pn , Q)(cid:17) (cid:16)ηk,Q ⊗ en+1(V n−i(1), k + Xk pn integral, Q ηkQ ⊗ en+1(1, k pn , Q). Indeed the rules 5.4 (ii) and (iv) yield the following formulas for k an integral weight, ξ ∈ V u( k pn )(A) and Q = (J0, . . . , Jq−1) a partition of length q − 1 of supp k : pn )Wn+1−u( k F nen+1(ξ, k pn , Q) = e1(F n(ξ), k, Q) e1(F iV −(n−i)(ξ), k, Q) if J0 6= ∅ or k if J0 = ∅ and k pn integral, pi is primitive, for 0 ≤ i ≤ n − 1 RATIONAL POINTS OF REGULAR MODELS 33 0 p−t( k pn )e1(F n(ξ), k, (∅, Q)) e1(F iV −(n−i)(ξ), k, (∅, Q)) if J0 = ∅, if J0 6= ∅ and k if J0 6= ∅ and k pn integral, pi is primitive, for 0 ≤ i ≤ n − 1. and F nden+1(ξ, k pn , Q) =  Using this, the rule 5.4 (i) and the relations (5.7.6) we obtain (−1 ⊗ F nd)(γ) = e1(ξk,(∅,Q), k, (∅, Q)) primitive pi and J06=∅ n−1Xi=0 Xk + Xk pn integral and J06=∅ = α e1(−p−t( k pn )ηk,Q, k, (∅, Q)) and (1 ⊗ F n)(γ) =  Xk n−1Xi=0 + Xk pn integral, Q pi primitive and J06=∅ e1(−pn−iξk,(∅,Q), k, Q) + Xk pi primitive and J0=∅ e1(ηkQ, k, Q) = β. This proves the proposition. e1(ηk,Q, k, Q) (cid:3) 5.8. We now recall some facts from [Il79, 0, 2] about the Cartier operator and its iterates. Let S be an Fp-scheme, X → S a smooth morphism, and set X (pn) := S ×S,F n X. X/S of the relative Frobenius S We have the usual diagram, which defines the iterates F n morphism (we write FX/S = F 1 X/S, W = W 1): F n X X X (pn) W n / X F n X/S "EEEEEEEEE F n S / S. S Notice that F n X/S = FX (pn−1 )/S ◦ . . . ◦ FX/S. For an S-morphism f : X ′ → X we denote by f (pn) the base-change morphism f (pn) = IdS × f : X ′(pn) → X (pn). / / "     /   / 34 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING The inverse Cartier operator is a homomorphism of graded OX (p)-algebras C −1 X/S : Ω• X (p)/S → H•(Ω• X/S), which is uniquely determined by (5.8.1) C −1 X/SO X(p) = F ∗ X/S and C −1 X/S(W ∗dx) = xp−1dx, for all x ∈ OX . The inverse Cartier operator is an isomorphism (since X/S is smooth). For n ≥ 0, one defines abelian subsheaves of Ωq X/S (5.8.2) via BnΩq X/S ⊂ ZnΩq X/S ⊂ Ωq X/S B1Ωq X/S = BΩq B0Ωq X/S = dΩq−1 X/S = 0, Z0Ωq X/S, Z1Ωq X/S = ZΩq X/S = Ωq X/S, X/S = Ker(d : Ωq X/S → Ωq+1 X/S), and, for n ≥ 1, (5.8.3) X/S : BnΩq C −1 X (p)/S (5.8.4) X/S : ZnΩq C −1 We obtain a chain of inclusions X (p)/S ≃ −→ Bn+1Ωq X/S/B1Ωq X/S, ≃ −→ Zn+1Ωq X/S/B1Ωq X/S. (5.8.5) 0 ⊂ B1Ωq X/S ⊂ . . . ⊂ BnΩq X/S ⊂ Bn+1Ωq X/S ⊂ . . . ⊂ Zn+1Ωq X/S ⊂ ZnΩq X/S ⊂ . . . ⊂ Z1Ωq X/S ⊂ Ωq X/S. Proposition 5.9 ([Il79, 0, (2.2.7), Prop. 2.2.8]). Let S be an Fp-scheme and X a smooth S-scheme. Then, for all q ≥ 0 and n ≥ 1, the sheaves ZnΩq satisfy the following properties. X/S and BnΩq X/S X/S are locally free OX (pn)-modules of finite type, and, for (i) ZnΩq X/S and BnΩq any h : S′ → S, we have h(pn)∗ X ZnΩq ∼ −−→ ZnΩq X ′/S′, h(pn)∗ X BnΩq X/S where hX : X ′ := S′ ×S X → X is the base-change map. X/S ∼ −−→ BnΩq X ′/S′, (ii) If f : X ′ → X is an ´etale S-morphism, then there are natural isomorphisms f (pn)∗ZnΩq X/S ∼ −−→ ZnΩq X ′/S, f (pn)∗BnΩq X/S ∼ −−→ BnΩq X ′/S. (iii) BnΩq X/S is the sub-OS-module of Ωq X/S locally generated by sections of the form apr−1 · · · apr−1 q 1 da1 · · · daq, with ai ∈ OX and 0 ≤ r ≤ n − 1. (iv) ZnΩq X/S is the sub-OS-module of Ωq sections of the form bapn−1 · · · apn−1 q 1 X/S locally generated by BnΩq da1 · · · daq, with ai ∈ OX and b ∈ OX (pn ). X/S and Proposition 5.10 (cf. [Il79, I, Prop. 3.3]). For X/S smooth as above, there is a unique map of Wn(OS)-modules n : FS∗Wn(OS ) ⊗Wn(OS ) WnΩq C −1 X/S −→ WnΩq X/S dV n−1Ωq X/S , RATIONAL POINTS OF REGULAR MODELS 35 which makes the following diagram commutative F∗Wn(OS) ⊗Wn+1(OS ) Wn+1Ωq X/S 1⊗F WnΩq X/S 1⊗R F∗Wn(OS) ⊗Wn(OS ) WnΩq X/S C−1 n WnΩq X/S dV n−1Ωq X/S . For n = 1 we have FS∗OS ⊗OS Ωq is the inverse Cartier operator. X/S = Ωq X (p)/S , and C −1 n = C −1 : Ωq X (p)/S −→ Ωq dΩq X/S X/S Proof. Since 1 ⊗ R is surjective, it is enough to see that the kernel of 1 ⊗ R is mapped to dV n−1Ωq X/S under 1 ⊗ F . But an element in the kernel of 1 ⊗ R is a sum of elements of the form a ⊗ V nω and a ⊗ dV nη, with a ∈ Wn(OS), ω ∈ Ωq X/S and η ∈ Ωq−1 X/S. We have in WnΩq X/S (1 ⊗ F )(a ⊗ V nω) = aV n−1(pω) = 0, (1 ⊗ F )(a ⊗ dV nη) = dV n−1(F n−1(a)η). This gives the existence and the uniqueness of C −1 n . The second statement follows from the fact that 1 ⊗ F is compatible with products, and from the formula 1 ⊗ F (a ⊗ d[x]) = axp−1dx, for a ∈ OS, x ∈ OX . (cid:3) Corollary 5.11 (cf. [Il79, I, Prop. 3.11]). Let X/S be as above. Then: X/S. X/S) = BnΩq Im(1 ⊗ F n : F n Im(1 ⊗ F n−1d : F n ∗ OS ⊗Wn+1(OS ) F∗WnΩq−1 ∗ OS ⊗Wn+1(OS) Wn+1Ωq X/S) = ZnΩq X/S → Ωq X/S → Ωq (ii) (i) X/S . Proof. We do induction on n. For n = 1, (i) follows from Proposition 5.10 and the relation d = F dV , and (ii) holds by definition. Now assume the statements are proven for n. To prove (i) for n + 1, we consider the following commutative diagram of abelian sheaves on X: ∗ OS ⊗Wn+2(OS ) Wn+2Ωq F n+1 X/S 1⊗F n F∗OS ⊗W2(OS ) W2Ωq X/S 1⊗F Ωq X/S 1⊗R 1⊗R ∗ OS ⊗Wn+1(OS ) Wn+1Ωq F n+1 X/S 1⊗F n / F∗OS ⊗OS Ωq X/S = Ωq X (p)/S C−1 Ωq X/S dΩq−1 X/S . By induction hypothesis we have Im(cid:16)(1⊗R)◦(1⊗F n)(cid:17) = Im(cid:16)(1⊗F n)◦(1⊗R)(cid:17) = FS∗OS ⊗OS ZnΩq where the last equality follows from the compatibility with base-change. Now, thanks to the relation d = F n+1dV n+1, (i) follows from the definition of Zn+1Ωq X/S. The proof of (ii) is similar. (cid:3) Lemma 5.12. Let X/S be as above. The sheaf BnΩq X/S is given by X/S = ZnΩq , X (p)/S Im(1 ⊗ F n−1d : F n = {(1 ⊗ F nd)(α) α ∈ F n ∗ OS ⊗Wn+1(OS ) F∗WnΩq−1 X/S → Ωq X/S) ∗ OS ⊗Wn+1(OS ) Wn+1Ωq−1 X/S with (1 ⊗ F n)(α) = 0}. / /     / / / /   / /     / / / 36 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING Proof. We call the left hand side A, and the right hand side B. We know from the previous corollary that BnΩq X/S = A, and we want now to show that A = B. In the following, all non-specified tensor products are over Wn+1(OS ). We have the commutative diagram ∗ OS ⊗ F∗WnΩq−1 F n X/S 'OOOOOOOOOOOO 1⊗F n−1d 1⊗V Ωq X/S ∗ OS ⊗ Wn+1Ωq−1 F n X/S wooooooooooooo 1⊗F nd . Since we also have (1 ⊗ F n) ◦ (1 ⊗ V ) = 0 it follows that A ⊂ B. It remains to show (5.12.1) Ker(cid:16)1 ⊗ F n : F n ⊂ Im(cid:18)F n ∗ OS ⊗ Wn+1Ωq−1 X/S → F n ∗ Ωq−1 X/S(cid:17) ∗ OS ⊗ (F∗WnΩq−1 X/S ⊕ F∗WnΩq−2 X/S) 1⊗(V +dV ) −−−−−−−→ F n ∗ OS ⊗ Wn+1Ωq−1 X/S(cid:19) . Indeed, if we take an element α in the kernel on the left hand side and we write it as an element in the right hand side α = (1 ⊗ V )(β) + (1 ⊗ dV )(γ), then (1 ⊗ F nd)(α) = (1 ⊗ F n−1d)(β), i.e., B ⊂ A. The question is local in X, we may thus assume X is ∗ M∗F s for M viewed as a left ´etale over Ad Wn+r(OS )-module via F r and as a right Wn+s(OX )-module via F s. Then we have the following commutative diagram, in which the most right tensor product in the upper line is over W2n+2(OAd S. For a Wn(OX )-module M we write F r ): S (cid:18)F n ∗ OS ⊗ F∗(WnΩq−2 Ad S/S (1⊗can)⊗1 )∗F n+2 1⊗dV ∗ OS ⊗ (Wn+1Ωq−1 F n Ad S /S )∗F n+1(cid:19) ⊗ W2n+2(OX ) (1⊗can)⊗1 ∗ OS ⊗ F∗(WnΩq−2 F n X/S)∗F n+2 1⊗dV / F n ∗ OS ⊗ (Wn+1Ωq−1 X/S)∗F n+1. If we write V instead of dV and q − 1 on the left hand side instead of q − 2, we obtain again a commutative diagram. Since X/Ad S is ´etale, the vertical maps are isomorphisms (in both diagrams). Thus if we denote the image in (5.12.1) by Im(X/S) we obtain Im(X/S) ∼= Im(Ad S/S)∗F n+1 ⊗W2n+2(O ) W2n+2(OX ). Ad S Similarly, denoting the kernel in (5.12.1) by Ker(X/S) one finds Ker(X/S) ∼= Ker(Ad S/S)∗F n+1 ⊗W2n+2(O ) W2n+2(OX ). Ad S And, since W2n+2(OX ) is ´etale over W2n+2(OAd ) [LZ04, Prop. A.8], it is thus enough to prove (5.12.1) in the case S = Spec A, with A an Fp-algebra, and X = Spec B, with B = A[x1, . . . , xd]. S / / ' w / /     / RATIONAL POINTS OF REGULAR MODELS 37 Now, using the notation of Corollary 5.6, any element α ∈ F n ∗ A ⊗ Wn+1Ωq−1 B/A can ai⊗en+1(V u(k)(ηk,P,i), k, P), ηk,P,i ∈ Wn+1−u(k)(A). be written as a finite sum (5.12.2) α =Xi Xpnk integral F nen+1(V u(k)(η), k, P) =(e1(F n−u(k)(η), pnk, P) By the rule 5.4, (ii) we have P=(I0,...,Iq−1) 0 if I0 = ∅ or (I0 6= ∅, k integral), if I0 6= ∅ and k not integral. It follows that an element α as in (5.12.2) lies in Ker(1 ⊗ F n) = Ker(B/A) iff it satisfies (5.12.3) aiF n−u(k)(ηk,P,i) = 0, for I0 = ∅ or (I0 6= ∅, k integral). Xi We consider the following three cases: ηen+1(1, k, P). By (5.12.3), we get 1) k is integral, i.e., u(k) = 0. Then, by Definition 5.3, en+1(η, k, P) = ai ⊗ en+1(ηi,k,P , k, P) = Xi Xi aiF n(ηi,k,P)! ⊗ en+1(1, k, P) = 0. 2) k is not integral and I0 = ∅. In this case en+1(η, k, P) ∈ Im(dV ) by Defini- tion 5.3. Thus Hence Xi Xi ai ⊗ en+1(ηi,k,P , k, P) ∈ Im(1 ⊗ dV ). ai ⊗ en+1(ηi,k,P, k, P) ∈ Im(1 ⊗ V ). 3) k is not integral and I0 6= ∅. Now en+1(η, k, P) ∈ Im(V ) by Definition 5.3. Putting the three cases together, we see that α ∈ Ker(1 ⊗ F n) implies α ∈ Im(1 ⊗ V + 1 ⊗ dV ) = Im(B/A). This gives the statement. (cid:3) Theorem 5.13 (cf. [Il79, I, Cor. 3.9], [Ol07, Th. 4.2.15]). Let S be an Fp-scheme and let X be a smooth S-scheme. For n, q ≥ 0, denote by grnWΩq X/S the n-th graded piece of the canonical filtration X/S = V nWΩq X/S = Ker(WΩq X/S + dV nWΩq X/S → WnΩq FilnWΩq X/S). Then we have an exact sequence of OX -modules (5.13.1) 0 −→ F n+1 X∗ Ωq X/S BnΩq X/S V n −−→ grnWΩq X/S Un−−→ F n+1 X∗ Ωq−1 X/S ZnΩq−1 X/S −→ 0, where the map Un is given by V n(α) + dV n(β) 7→ β and the OX -module structure on grnWΩq X/S is given via OX = WnOX V Wn−1OX F−−→ Wn+1OX pWnOX . 38 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING Furthermore F n X/S∗ Ωq X/S BnΩq X/S and F n X/S∗ Ωq−1 X/S ZnΩq−1 X/S are locally free OX (pn)-modules. Proof. The exactness of the sequence follows from Proposition 5.7, Corollary 5.11 and Lemma 5.12. The second statement is proven as in [Il79, I, Cor. 3.9]. By ´etale base change (Proposition 5.9, (ii)), we reduce the question of the local freeness of the two extreme OX (pn)-modules in the exact sequence to the case X = Ad S. Since everything is compatible with arbitrary base change in the base S (by Proposition 5.9, (i)), we may also assume S = Spec Fp, and even S = Spec k with k algebraically closed. But now the sheaves in question are coherent on (Ad k, hence locally free in some non-empty open subset, whose translates under certain closed points k)(pn). As they are invariant under translation, this gives the cover the whole of (Ad statement. k)(pn) ∼= Ad (cid:3) 6. The Hodge-Witt trace morphism for projective spaces Let X be a noetherian Fp-scheme with a dualizing complex, and let f : Y → X be a projective complete intersection morphism of virtual relative dimension 0. Our goal in the next two sections is to prove that, given a factorization f = π ◦ i, where π : P = P d X → X is the structural morphism of some projective space over X, and i : Y ֒→ P is a closed immersion, one can define for all n ≥ 1 a morphism τi,π,n : Rf∗WnOY −→ WnOX so as to satisfy the following properties: (i) For n = 1, τi,π,n is the morphism τf of Theorem 3.1; (ii) For variable n, τi,π,n commutes with R, F and V . Our construction of τi,π,n will be based on a generalization for arbitrary n of the description of τf given in Proposition 4.6: we will construct on the one hand a trace morphism Rπ∗WnΩd P/X[d] → WnOX , which will be a generalization of the trace morphism Trpπ for the projective space, and on the other hand a morphism i∗WnOY → WnΩd P/X[d] which will be a generalization of the morphism γf : OY → ωP/X[d] defined in (4.4.1). We begin with the trace morphism for projective spaces. 6.1. We recall first from [Il90, D´ef. 1.1] that a smooth proper Fp-morphism f : X → S is called ordinary, if it satisfies Rif∗BΩq X/S = 0, for all i, q ≥ 0. This notion is compatible with arbitrary base-change in the base S, and P d Fp is ordinary over Spec Fp [Il90, Prop. 1.2, Prop. 1.4]. Hence if E is a locally free OX- module of finite rank on some Fp-scheme X, then P(E) = Proj (SymOX E) is ordinary over X. Lemma 6.2. Let f : X → S be ordinary. Then, for all n ≥ 1 and q ≥ 0, V n : F n+1 S∗ Rf∗Ωq X/S ∼ −−→ Rf∗grnWΩq X/S RATIONAL POINTS OF REGULAR MODELS 39 is an isomorphism in the derived category of quasi-coherent OS-modules (where the OS-module structure on the right hand side comes from the OX -module structure defined in Theorem 5.13 ). Proof. This follows immediately from Theorem 5.13 and the following claim: (6.2.1) Rif∗ZnΩq ∼ −−→ Rif∗Ωq X/S, Rif∗BnΩq X/S = 0, for all i, q ≥ 0, n ≥ 1. X/S We prove this by induction on n. The statement for B1 holds by definition of ordinarity and for Z1 follows from the exact sequence 0 −→ ZΩq X/S −→ Ωq X/S d−→ BΩq+1 X/S −→ 0. Now for the general case consider the following commutative diagram (in which f∗ is viewed as a functor on the category of abelian sheaves for the Zariski topology on X = X (p)) Rif∗ZnΩq X (p)/S C−1 X/S Rif∗ Zn+1Ωq B1Ωq X/S X/S Rif∗Ωq X (p)/S C−1 X/S / Rif∗ Z1Ωq B1Ωq X/S X/S Rif∗Zn+1Ωq X/S Rif∗Z1Ωq X/S . The horizontal maps are isomorphisms as is the vertical map on the left by induction (notice that X (p)/S is also ordinary). Hence all maps in the diagram are isomor- phisms, which yields the claim for Zn+1. To prove the statement for Bn+1 it is enough to consider the upper line in the diagram, with Z replaced by B, and one immediately obtains the statement. (cid:3) 6.3. Let S be a scheme on which p is locally nilpotent, and X an S-scheme. As in the classical case [Il79, I, 3.23], we define for any n ≥ 1 the log derivation dlog n to be the morphism of abelian sheaves dlog n : O× X −→ WnΩ1 X/S, a 7→ dlog n(a) := d[a] [a] . We may write simply dlog if n is fixed. For variable n, the maps dlog n satisfy the following relations: (6.3.1) R(dlog n(a)) = dlog n−1(a), F (dlog n(a)) = dlog n−1(a). The maps dlog n allow to define Chern classes for line bundles, and to prove for relative Hodge-Witt cohomology the analog of the classical theorem on the coho- mology of projective bundles (cf. [SGA 7 II, XI, Thm. 1.1]). Theorem 6.4. Let X be an Fp-scheme, E a locally free OX -module of rank d + 1, P = P(E), and let π : P → X be the canonical projection. Denote by ηn ∈ H 0(X, R1π∗WnΩ1 P , and by ηq P/X) its q-fold cup product. Then, for all n ≥ 1 and all q such that 0 ≤ q ≤ d, we have P/X) the image under dlog n of the class of OP (1) in R1π∗O× n ∈ H 0(X, Rqπ∗WnΩq (6.4.1) Rjπ∗WnΩq P/X = 0 for j 6= q, / /       o o / o o 40 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING and multiplication with ηq modules n induces an isomorphism in the derived category of Wn(OX )- (6.4.2) Wn(OX )[−q] ∼ −−→ Rπ∗WnΩq P/X. Furthermore these isomorphisms are compatible with restriction, Frobenius and Ver- schiebung on both sides. Proof. To prove (6.4.1), we can argue by induction using the exact sequences 0 −→ grnWn+1Ωq P/X −→ Wn+1Ωq P/X −→ WnΩq P/X −→ 0. For n = 1, the claim follows from [SGA 7 II, XI, Thm. 1.1], and, since P(E) is ordinary over X, Lemma 6.2 implies similarly the claim for all n. Therefore, we obtain a canonical isomorphism (6.4.3) Rπ∗WnΩq P/X ∼ −−→ Rqπ∗WnΩq P/X[−q], and we can define the morphism (6.4.2) as corresponding via (6.4.3) and translation to the morphism (6.4.4) Wn(OX ) −→ Rqπ∗WnΩq P/X, w 7→ wηq n. This reduces the proof of the theorem to proving that (6.4.4) is an isomorphism, compatible with R, F and V . From (6.3.1), we get for all w ∈ Wn+1(OX ) the relations (6.4.5) in Rqπ∗WnΩq R(wηq n+1) = R(w)ηq n, F (wηq n+1) = F (w)ηq n P/X. From the second relation, we also get (6.4.6) V (wηq n−1) = V (wF (ηq n)) = V (w)ηq n for all w ∈ Wn−1(OX ). So the homomorphisms (6.4.4) satisfy the required compat- ibilities. To prove that the homomorphisms (6.4.4) are isomorphisms, we may now again argue by induction on n, using the compatibility with R and V . Then Lemma 6.2 reduces the proof to the case n = 1, which is known by [SGA 7 II, Exp. XI, Thm. 1.1]. (cid:3) Definition 6.5. Under the assumptions of Theorem 6.4, we define the Hodge-Witt trace morphism for the projective space P(E) to be the WnOX-linear map (6.5.1) Trpπ,n : Rπ∗WnΩd P(E)/X [d] ∼ −−→ WnOX obtained by inverting the isomorphism (6.4.2), shifting by d and multiplying by (−1)d(d−1)/2. Theorem 6.4 implies that Trpπ,n is compatible with restriction, Frobe- nius and Verschiebung. Proposition 6.6. With the hypotheses of Theorem 6.4, assume in addition that X is locally noetherian. Then the morphism Trpπ,1 : Rπ∗Ωd −−→ OX (6.6.1) P(E)/X [d] ∼ defined by (6.5.1) for n = 1 is equal to the morphism Trpπ defined by [Co00, (2.3.5)] for OX . RATIONAL POINTS OF REGULAR MODELS 41 Proof. By (6.4.2), it suffices to prove the proposition locally on X. So we may assume that P(E) = P d X. Let X0, . . . , Xd be the standard homogeneous coordinates on P d X, xi = Xi/X0, Ui = D+(Xi), and let U = (Ui)i=0,...,d be the corresponding X. Using Cech cohomology relative to U, η1 is defined by the 1-cocycle covering of P d (dlog (Xj/Xi))i<j = (d(Xj /Xi)/(Xj /Xi))i<j , and ηd 1 by the d-cocycle given by dlog (X1/X0) ∧ · · · ∧ dlog (Xd/Xd−1) = dx1/x1 ∧ (dx2/x2 − dx1/x1) ∧ . . . ∧ (dxd/xd − dxd−1/xd−1) = dx1 ∧ · · · ∧ dxd/x1 · · · xd on U0 ∩ . . . ∩ Ud. Thus Trpπ,1 is the only morphism which induces on degree 0 coho- mology the isomorphism mapping the class dx1 ∧ · · · ∧ dxd/x1 · · · xd to (−1)d(d−1)/2. To prove the proposition, it suffices to check that, with Conrad's definitions, the map induced by Trpπ : Rf∗(f ♯(OX )) = Rf∗(ωP/X[d]) → OX on degree 0 cohomology is such that (6.6.2) Trpπ(dx1 ∧ · · · ∧ dxd/x1 · · · xd) = (−1)d(d−1)/2. As (−1)d(−1)d(d−1)/2 = (−1)d(d+1)/2, this follows from the definition of the isomor- phism [Co00, (2.3.1)] (6.6.3) γ : Rdπ∗(ωP/X) ∼ −−→ OX , which sends dx1 ∧ · · · ∧ dxd/x1 · · · xd to (−1)d(d+1)/2 [Co00, (2.3.3)], and from the discussion on pp. 35-36 of [Co00], which explains that an additional (−1)d sign is required to recover (6.6.3) from the map induced in degree 0 by Trpπ (note that by "induced", we mean that we use here as we always do the standard identifications [Co00, (1.3.1), (1.3.4)] to compute the cohomology objects of a translated complex). This ends the proof of the proposition, but, as formula (6.6.2) is only implicit in the discussion [Co00, p. 35-36], it may be worth adding a few lines to give a proof explaining where this extra (−1)d sign comes from. Conrad's construction of the projective trace Trpπ is the same as Hartshorne's in [Ha66, III, 4.3], but using [Co00, Lemma 2.1.1] instead of [Ha66, I, Proposition 7.4]. Because π∗ has cohomological dimension d on the category of quasi-coherent OP -modules, and any quasi-coherent OP -module can be written as a quotient of modules for which the functors Riπ∗ vanish for i 6= d [Ha66, III, Lemmas 4.1 and 4.2], Lemma 2.1.1 of [Co00] provides an isomorphism of functors on D(Qcoh(OP )) ψ : Rπ∗ ∼ −−→ L(Rdπ∗)[−d]. For complexes of the form F • = F[0], where F is a quasi-coherent OP -module, ψF • induces in degree d the identity of Rdπ∗(F) [Co00, Cor. 2.1.2]. Moreover, the compatiblity of ψ with translations, given by [Co00, (2.1.1)], implies that, for any m ∈ Z, we have ψF •[m] = (−1)mdψF •[m]. 42 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING In particular, ψωP /X [d] induces in degree 0 multiplication by (−1)d2 Rdπ∗(ωP/X). Now, for G • ∈ D+ = (−1)d on qc(OX ), the trace morphism for G • is the composition Rπ∗(π♯G •) ψπ♯G· Rπ∗(ωP/X[d] ⊗OP π∗G •) ∼ / L(Rdπ∗)(ωP/X[d] ⊗OP π∗G •)[−d] Trpπ,G· G • γ⊗Id ∼ (Rdπ∗)(ωP/X ) ⊗OX G • ∼ (Rdπ∗)(ωP/X ⊗OP π∗G •) ∼ (see [Ha66, III, 4.3] for details). Taking G • = OX[0], and applying the previous remark to π♯OX = ωP/X[d], we obtain that Trpπ,OX induces (−1)dγ in degree 0, which gives (6.6.2). (cid:3) Remark. Because of the differences in sign conventions between Hartshorne [Ha66, III, Th 3.4] and Conrad [Co00, 2.3], our trace morphism Trpπ,n differs by (−1)d(d−1)/2 from the trace morphism defined by Ekedahl [Ek84, I, Lemma 3.2] when X = Spec k, k being a perfect field. 7. The Hodge-Witt fundamental class of a regularly embedded subscheme In this section, we assume that X is a locally noetherian scheme of characteristic p, and we consider a regular immersion i : Y ֒→ P of codimension d, where P is a smooth X-scheme. Under these assumptions, we want to associate to Y a canonical class γY ∈ Γ(P, Hd P/X)), for each n ≥ 1. Y (WnΩd Proposition 7.1. Under the previous assumptions: (i) If t1, . . . , td is a regular sequence of sections of OP , then, for all n ≥ 1 and all r ≥ 1, [t1]r, . . . , [td]r is a regular sequence of sections of Wn(OP ). (ii) For all n ≥ 1 and all q, Hj Y (WnΩq P/X) = 0 for j 6= d. Proof. We proceed by induction on n. In the exact sequence of Wn+1(OP )-modules 0 −→ F n V n −−→ Wn+1(OP ) R−−→ Wn(OP ) −→ 0, ∗ OP ∗ OP is given by multiplication by trpn the action of [ti]r on F n locally noetherian scheme, the sequence trpn claim follows easily. 1 , . . . , trpn d on OP . As P is a is regular in OP , and the first i For n = 1, the second one is a well known consequence of the regularity of the sequence t1, . . . , td. As OP is locally free of finite rank over OP (pn), we also have Hj Y (OP (pn)) = 0 for j 6= d. In the exact sequence 0 −→ grnWn+1Ωq P/X −→ Wn+1Ωq P/X R−−→ WnΩq P/X −→ 0, Theorem 5.13 allows to endow the kernel grnWn+1Ωq P/X with an OP -module struc- ture for which it is an extension of two OP -modules which are locally free over OP (pn). Therefore, Hj P/X) = 0 for j 6= d. The second claim follows by induction. (cid:3) Y (grnWn+1Ωq   /   o o o o RATIONAL POINTS OF REGULAR MODELS 43 1, . . . , t′ Theorem 7.2. Under the assumptions of this section, let t = (t1, . . . , td) and t′ = (t′ d) be two regular sequences of sections of OP generating the ideal I of Y in P . Let n ≥ 1 be an integer, and let J = ([t1], . . . , [td]), J ′ = ([t′ d]) be the ideals of Wn(OP ) generated by the Teichmuller representatives of these generators. If 1], . . . , [t′ βJ : Ext d Wn(OP )(Wn(OP )/J , WnΩd P/X) −→ Hd Y (WnΩd P/X) is the canonical homomorphism (and similarly for βJ ′), then, with the notations of 4.1, (7.2.1) βJ ((cid:20) d[t1] · · · d[td] [t1], . . . , [td] (cid:21)) = βJ ′((cid:20) d[t′ 1] · · · d[t′ d] 1], . . . , [t′ [t′ d] (cid:21)). Proof. It suffices to prove (7.2.1) in a neighbourhood of each point y ∈ Y . Localizing, one can reduce the proof of Theorem 7.2 to the case of a very simple change of generators in I, thanks to the following remarks (see also [SGA 4 1 2 , Cycle, Lemme 2.2.3]). a) If the sequence (t′ 1, . . . , t′ d) is deduced from (t1, . . . , td) by permutation, then J = J ′, and formula (4.2.1) implies the theorem. b) If there exists invertible sections a1, . . . , ad ∈ O× i = aiti for all i] = [ai][ti] for all i. So J = J ′, we can apply Lemma 4.2, and we can i, then [t′ choose the matrix C to be the diagonal matrix with entries [ai]. Then the theorem follows from formula (4.2.1), because an element such as (4.1.2) only depends upon the class of m mod (t1, . . . , td)M , and here we have the congruence P such that t′ d[t′ 1] · · · d[t′ d] ≡ ( [ai]) d[t1] · · · d[td] mod J WnΩd P/X. dYi=1 σ(1), . . . , t′ c) Given y ∈ Y , there exists a permutation σ ∈ Sd such that, for any i, 1 ≤ i ≤ d, the sequence t(i) = (t′ σ(i), ti+1, . . . , td) is a regular sequence of generators of I around y. Indeed, a sequence of elements of Iy is a regular sequence of generators if and only if it gives a basis of Iy/myIy, and this reduces the claim to an elementary result in linear algebra over a field. If we set t(0) = (t1, . . . , td), then t(0) = t, and t(d) is deduced from t′ by permutation. So, using remark a), it suffices to prove the theorem for the couple of sequences t(i−1) and t(i), for all i, 1 ≤ i ≤ d. This reduces the proof to the case where there exists an integer i0 ∈ {1, . . . , d} such that t′ i = ti for i 6= i0, t′ i0 = ci0,jtj. dXj=1 Using remark a), we may assume that i0 = 1. Moreover, the fact that t and t′ induce bases of the vector space Iy/myIy implies that the coefficient c1,1 is invertible around y. d) In this last case, we define inductively elements t(j) 1 for 0 ≤ j ≤ d by setting t(0) 1 = t1, t(1) 1 = c1,1t(0) 1 , 1 = t(j−1) t(j) 1 + c1,jtj for 1 < j. If, for 0 ≤ j ≤ d, we define t(j) = (t(j) 1 , t2, . . . , td), then t(0) = t, t(d) = t′, and it suffices to prove the theorem for each of the couples t(j−1), t(j), for 1 ≤ j ≤ d. The 44 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING theorem is true for t(0), t(1), thanks to remark b), and, applying again remark a), we can write all the remaining couples as changes of generators of the form (7.2.2) t′ 1 = t1 + ct2, for some c ∈ OP , t′ i = ti for i ≥ 2. Thus it suffices to prove the theorem for the change of generators of I given by (7.2.2). Let h ∈ V Wn−1(OP ) be defined by setting (7.2.3) in Wn(OP ). Since [t′ (7.2.4) [t1] + [c][t2] = [t1 + ct2] + h = [t′ 1] + h 2] = [t2], this can be rewritten as 2] + h. 1] − [c][t′ [t1] = [t′ The binomial formula gives (7.2.5) [t1]pn−1 = ([t′ 1] − [c][t′ 2])pn−1 + pn−1! (pn−1 − i)!i! hi([t′ 1] − [c][t′ 2])pn−1−i. pn−1Xi=1 Because the ideal V Wn−1(OP ) ⊂ Wn(OP ) is a PD-ideal, we can write hi = i!h[i], with h[i] ∈ V Wn−1(OP ) when i ≥ 1. Therefore the numerical coefficient of h[i] in the i-th term of the sum is divisible by pn−1 for all i ≥ 1. Since pn−1 kills V Wn−1(OP ), equation (7.2.5) reduces to (7.2.6) [t1]pn−1 = ([t′ 1] − [c][t′ 2])pn−1 . If, for all k ≥ 1, we denote by J (k) the ideal ([t1]k, . . . , [td]k), this shows that d]) and ), which are regular by Lemma 7.1. Moreover, we can write J (pn−1) ⊂ J ′. So we can apply Lemma 4.2 to the sequences ([t′ , . . . , [td]pn−1 ([t1]pn−1 equation (7.2.6) as 1], . . . , [t′ [t1]pn−1 = [t′ 1]pn−1−1 · [t′ 1] + c1,2 · [t′ 2], so that we can use as matrix C in Lemma 4.2 an upper triangular matrix with diagonal entries [t′ i] for i ≥ 2). d]pn−1−1. Thus, formula (4.2.1) provides the In particular, det(C) = [t′ equality d]pn−1−1 (since [ti]pn−1 1]pn−1−1, . . . , [t′ 1]pn−1−1 · · · [t′ i]pn−1−1 · [t′ = [t′ (7.2.7) α′((cid:20) d[t′ 1] · · · d[t′ d] 1], . . . , [t′ [t′ d] (cid:21)) =" [t′ 1]pn−1−1 · · · [t′ [t1]pn−1 d]pn−1−1 d[t′ , . . . , [td]pn−1 1] · · · d[t′ d] # , where α′ is the canonical homomorphism Ext d Wn(OP )(Wn(OP )/J ′, WnΩd P/X) −→ Ext d Wn(OP )(Wn(OP )/J (pn−1), WnΩd P/X). On the other hand, we also have J (pn−1) ⊂ J . So we can also apply Lemma 4.2 ), using now for C to the regular sequences ([t1], . . . , [td]) and ([t1]pn−1 the diagonal matrix with entries [t1]pn−1−1, . . . , [td]pn−1−1. If we denote by α : Ext d Wn(OP )(Wn(OP )/J (pn−1), WnΩd Wn(OP )(Wn(OP )/J , WnΩd P/X) −→ Ext d , . . . , [td]pn−1 P/X) the canonical homomorphism, formula (4.2.1) provides the second equality α((cid:20) d[t1] · · · d[td] [t1], . . . , [td] (cid:21)) =" [t1]pn−1−1 · · · [td]pn−1−1 d[t1] · · · d[td] , . . . , [td]pn−1 [t1]pn−1 # . (7.2.8) RATIONAL POINTS OF REGULAR MODELS 45 As βJ = βJ (pn−1) ◦ α and βJ ′ = βJ (pn−1) ◦ α′, relation (7.2.1) will follow if we prove the equality (7.2.9) # =" [t1]pn−1−1 · · · [td]pn−1−1 d[t1] · · · d[td] , . . . , [td]pn−1 [t1]pn−1 # " [t′ 1]pn−1−1 · · · [t′ [t1]pn−1 d]pn−1−1 d[t′ , . . . , [td]pn−1 1] · · · d[t′ d] Wn(OP )(Wn(OP )/J (pn−1), WnΩd in Ext d the congruence P/X). To prove it, it suffices to prove in WnΩd P/X (7.2.10) 1]pn−1−1[t′ [t′ 2]pn−1−1 · · · [t′ d]pn−1−1 d[t′ 1] d[t′ 2] · · · d[t′ d] ≡ [t1]pn−1−1[t2]pn−1−1 · · · [td]pn−1−1 d[t1] d[t2] · · · d[td] , . . . , [td]pn−1 mod ([t1]pn−1 multiplicativity to prove in WnΩ2 , [t2]pn−1 P/X. As ti = t′ )WnΩd P/X the congruence i for i > 2, it suffices by 1]pn−1−1[t′ [t′ , [t2]pn−1 mod ([t1]pn−1 applying F n−1 if we prove the congruence 2]pn−1−1 d[t′ )WnΩ2 1] d[t′ 2] ≡ [t1]pn−1−1[t2]pn−1−1 d[t1] d[t2] P/X, and, thanks to (5.1.3), the latter will follow by (7.2.11) d[t′ 1] d[t′ 2] ≡ d[t1] d[t2] mod ([t1], [t2])W2n−1Ω2 P/X. So let us prove (7.2.11). We still denote by h ∈ V W2n−2OP the difference h = 2 = t2, it 1] = [t1] + [ct2] − [t1 + ct2] computed in W2n−1OP . Since t′ [t1] + [c][t2] − [t′ suffices to prove the congruence (7.2.12) For all i, let dh d[t2] ≡ 0 mod ([t1], [t2])W2n−1Ω2 P/X. Si(X0, . . . , Xi, Y0, . . . , Yi) ∈ Z[X0, . . . , Xi, Y0, . . . , Yi] be the universal polynomial defining the i-th component of the sum of two Witt vectors, and (7.2.13) si(X0, Y0) = Si(X0, 0, . . . , 0, Y0, 0, . . . 0) ∈ Z[X0, Y0]. Note that, for i ≥ 1, the polynomial si(X0, Y0) is divisible by X0Y0, since (0, . . . , 0) is the zero element in a Witt vector ring. By definition, we have [t1] + [ct2] = (t1 + ct2, s1(t1, ct2), . . . , s2n−2(t1, ct2)), and h = (0, s1(t1, ct2), . . . , s2n−2(t1, ct2)). Since si(X0, Y0) is divisible by Y0, we can write si(t1, ct2) = zit2 for some section zi ∈ OP . We obtain which we can write as h = (0, z1t2, . . . , z2n−2t2), h = 2n−2Xi=1 V i([zi][t2]). 46 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING For each i, 1 ≤ i ≤ 2n − 2, we now obtain the relations dV i([zi][t2]) d[t2] = dV i([zi][t2] F i(d[t2])) = dV i([zi][t2]pi d[t2]) = dV i([zi]F i([t2])d[t2]) = d([t2]V i([zi]d[t2])), d([t2]V i([zi]d[t2])) ≡ d[t2] V i([zi]d[t2]) mod [t2]W2n−1Ω2 P/X, d[t2] V i([zi]d[t2]) = V i(F i(d[t2])[zi]d[t2]) = V i([t2]pi−1d[t2][zi]d[t2]) = 0, which imply (7.2.12). (cid:3) Definition 7.3. Under the assumptions of this section, we define the n-th Hodge- Witt fundamental class γY,n of Y in P relatively to X as being the section of P/X) obtained by glueing the sections βJ ((cid:20) d[t1] · · · d[td] [t1], . . . , [td] (cid:21)) defined locally by regular sequences of generators of the ideal I of Y in P . Hd Y (WnΩd Proposition 7.4. For n ≥ 1, let R : Hd F : Hd V : Hd Y (Wn+1Ωd Y (Wn+1Ωd Y (WnΩd P/X) −→ Hd P/X) −→ Hd P/X) −→ Hd Y (WnΩd Y (WnΩd Y (Wn+1Ωd P/X), P/X), P/X) be the homomorphisms defined by functoriality. Then (7.4.1) R(γY,n+1) = γY,n, F (γY,n+1) = γY,n, V (γY,n) = pγY,n+1. Proof. We may assume that there exists a regular sequence t1, . . . , td such that I = (t1, . . . , td). For each n ≥ 1, let Jn be the ideal of Wn(OP ) generated by the Teichmuller representatives [ti] of the ti's, and let K•([t]n) be the Koszul complex defined by the [ti]'s over Wn(OP ). Since R([ti]) = [ti], scalar extension through R yields an isomorphism Wn(OP ) ⊗Wn+1(OP ) K•([t]n+1) ∼ −−→ K•([t]n). Using the fact that the [ti]'s form a regular sequence both in Wn+1(OP ) and in Wn(OP ), it can be seen in the derived category of Wn(OP )-modules as an isomor- phism (7.4.2) Wn(OP ) L ⊗Wn+1(OP ) Wn+1(OP )/Jn+1 ∼ −−→ Wn(OP )/Jn. By adjunction, (7.4.2) defines for any Wn(OP )-module M and any q ≥ 0 an isomor- phism (7.4.3) Ext q Wn(OP )(Wn(OP )/Jn, M) ∼ −−→ Ext q Wn+1(OP )(Wn+1(OP )/Jn+1, M), RATIONAL POINTS OF REGULAR MODELS 47 and we obtain the diagram (7.4.4) Hd(Hom • Wn+1(OP )(K•([t]n+1), Wn+1Ωd P/X)) ∼ +WWWWWWWWWWWWWW R Ext d Wn+1(OP )(Wn+1(OP )/Jn+1, Wn+1Ωd P/X) βJn+1 Hd Y (Wn+1Ωd P/X) Hd(Hom • Wn+1(OP )(K•([t]n+1), WnΩd +WWWWWWWWWWWWWW ∼ P/X)) ≀ Ext d Wn+1(OP )(Wn+1(OP )/Jn+1, WnΩd P/X) R R Hd(Hom • P/X)) Wn(OP )(K•([t]n), WnΩd +WWWWWWWWWWWWWW Wn(OP )(Wn(OP )/Jn, WnΩd Ext d ∼ ≀ (7.4.3) P/X) βJn+1 βJn / Hd Y (WnΩd P/X) / Hd Y (WnΩd P/X), in which the lower left hand square commutes by construction. On the other hand, (7.4.3) implies that injective Wn(OP )-modules are acyclic for the functor Hom Wn+1(OP )(Wn+1(OP )/Jn+1, −). Replacing WnΩd P/X by an injective resolution over Wn(OP ), it is then easy to check that the lower right square commutes. As the upper part of the diagram commutes by functoriality, and R(d[t1] · · · d[td]) = d[t1] · · · d[td], the first relation of (7.4.1) follows. Viewing now Wn(OP ) as a Wn+1(OP )-algebra via F , one proceeds similarly to prove the second one. Since F ([ti]) = [tp i ] = [ti]p, and the sequence [ti]p, . . . , [td]p is a regular sequence in Wn(OP ), we obtain for any Wn(OP )-module M and any q ≥ 0 isomorphisms Wn(OP ) ⊗Wn+1(OP ) K•([t]n+1) ∼ −−→ K•([t]p n), (7.4.5) (7.4.6) Wn(OP ) Wn(OP )(Wn(OP )/J (p) L ⊗Wn+1(OP ) Wn+1(OP )/Jn+1 ∼ −−→ Ext q n , M) Ext q ∼ −−→ Wn(OP )/J (p) n , Wn+1(OP )(Wn+1(OP )/Jn+1, M).   + / /     + / O O + O O / 48 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING They provide a commutative diagram similar to (7.4.4): (7.4.7) Hd(Hom • Wn+1(OP )(K•([t]n+1), Wn+1Ωd P/X)) ∼ +WWWWWWWWWWWWWW F Ext d Wn+1(OP )(Wn+1(OP )/Jn+1, Wn+1Ωd P/X) βJn+1 Hd Y (Wn+1Ωd P/X) Hd(Hom • Wn+1(OP )(K•([t]n+1), WnΩd +WWWWWWWWWWWWWW ∼ P/X)) ≀ Ext d Wn+1(OP )(Wn+1(OP )/Jn+1, WnΩd P/X) F F Hd(Hom • Wn(OP )(K•([t]p ∼ P/X)) n), WnΩd +WWWWWWWWWWWWWW Wn(OP )(Wn(OP )/J (p) Ext d ≀ (7.4.6) n , WnΩd P/X) βJn+1 β J (p) n / Hd Y (WnΩd P/X) / Hd Y (WnΩd P/X). Since F (d[t1] · · · d[td]) = [t1]p−1 · · · [td]p−1d[t1] · · · d[td], it follows that F (βJn+1((cid:20) d[t1] · · · d[td] [t1], . . . , [td] (cid:21))) = βJ (p) n ((cid:20) [t1]p−1 · · · [td]p−1 d[t1] · · · d[td] [t1]p, . . . , [td]p (cid:21)). On the other hand, if α denotes the canonical homomorphism α : Ext d Wn(OP )(Wn(OP )/Jn, WnΩd P/X) −→ Ext d Wn(OP )(Wn(OP )/J (p) n , WnΩd P/X), we have by (4.2.1) α((cid:20) d[t1] · · · d[td] [t1], . . . , [td] (cid:21)) =(cid:20) [t1]p−1 · · · [td]p−1 d[t1] · · · d[td] [t1]p, . . . , [td]p (cid:21) . ◦ α = βJn, it follows that F (γY,n+1) = γY,n. As βJ (p) n pγY,n+1. The last relation of (7.4.1) follows formally, because V (γY,n) = V (F (γY,n+1)) = (cid:3) Proposition 7.5. Let n ≥ 1 be an integer, and let γY,n ∈ Hd Hodge-Witt fundamental class of Y in P relatively to X, as defined in 7.3. Y (WnΩd P/X) be the (i) The linear homomorphism Wn(OP ) → Hd Y (WnΩd P/X) sending 1 to γY,n vanishes on Wn(I) := Ker(Wn(OP ) ։ i∗Wn(OY )). (ii) Let γi,π,n be the composition (7.5.1) γi,π,n : i∗Wn(OY ) −→ Hd Y (WnΩd P/X) ∼ −−→ RΓY (WnΩd P/X[d]) −→ WnΩd P/X[d], where the first morphism is defined thanks to the previous assertion. Then γi,π,n commutes with R, F and V . (iii) For n = 1, we have γi,π,1 = γf , where γf is the morphism defined by (4.4.1).   + / /     + / O O + O O / RATIONAL POINTS OF REGULAR MODELS 49 Proof. To prove assertion (i), we may again assume that I is generated by a regular sequence t1, . . . , td. Any section w of Wn(I) can then be written as a sum w = V i([ai,1][t1] + · · · + [ai,d][td]), n−1Xi=0 with ai,j ∈ I and [ai,j], [tj] ∈ Wn−i(OP ). By functoriality, we have V (a)ω = V (aF (ω)) for any a ∈ Wi(OP ), ω ∈ Hd P/X), i ≥ 1. Using (7.4.1), we obtain Y (Wi+1Ωd V i([ai,j][tj])γY,n = V i([ai,j][tj]F i(γY,n)) = V i([ai,j][tj]γY,n−i). The symbol (4.1.2) is linear with respect to m, therefore we have [ai,j][tj]γY,n−i = βJ ((cid:20) [ai,j][tj] d[t1] · · · d[td] [t1], . . . , [td] (cid:21)) = 0 since the upper entry in the symbol belongs to ([t1], . . . , [td])Wn−iΩd P/X. In the definition of γi,π,n, the last two arrows commute with R, F and V by functoriality. Relations (7.4.1) imply that the first one also commutes with R, F and V , since R(1) = F (1) = 1, and V (1) = p. Let us assume that n = 1, and check assertion (iii). By construction, γi,π,1 is P/X) sending 1 to γY,1 with the the composition of the morphism i∗OY → Hd canonical morphism Y (Ωd Hd Y (Ωd P/X) ∼ −−→ RΓY (Ωd P/X[d]) −→ Ωd P/X[d]. Comparing with the definition of γf in 4.4, and using the same notations, it suffices to show that the composed morphism OY ϕf−→ ωY /X ◦ ζ ′ η−1 i −−−−−−→ Ext d i,π OP (OY , Ωd P/X ) βI−→ Hd Y (Ωd P/X) sends 1 to γY,1. Since this is a morphism of sheaves (rather than complexes in the derived category), it is a local verification, which is provided by Proposition 4.5. (cid:3) Definition 7.6. Let X be a noetherian Fp-scheme with a dualizing complex, E a locally free OX -module of rank d+1, P = P(E), π : P → X the canonical projection, i : Y ֒→ P a regular closed immersion of codimension d. For each integer n ≥ 1, we define a trace morphism τi,π,n by (7.6.1) τi,π,n : Rf∗(Wn(OY )) Rπ∗(γi,π,n) −−−−−−−→ Rπ∗(WnΩd P/X[d]) Trpπ,n−−−−→ Wn(OX ), where γi,π,n is the morphism (7.5.1), and Trpπ,n is the Hodge-Witt trace morphism defined in (6.5.1). Remark. As mentioned in the introduction, we expect that τi,π,n depends only on f , and not on the factorization f = π ◦ i. We also expect that the analog of Theorem 3.1 holds for the trace morphisms τf,n that would be thus defined. More generally, one can hope that these constructions are part of a theory of canonical classes for relative de Rham-Witt cohomology (see [El78], [Ek84], [Gs85] for such results over a field). In order to develop this program, generalizations and non-trivial properties of our constructions are needed (even for the independence statement), which would 50 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING lead to expand too much this article. As most of them are not needed for the proof of our main results, we do not include them here, and we hope to return to these questions elsewhere. However, we will give in the next section a partial generalization of Theorem 3.1, (iii), which is the key to the injectivity property of Theorem 1.5. Proposition 7.7. Under the assumptions of 7.6, the morphisms τi,π,n satisfy the following properties. (i) For variable n, τi,π,n commutes with R, F and V . (ii) For n = 1, τi,π,1 = τf . Proof. Taking into account Proposition 4.6, both assertions follow from the similar properties of γi,π,n and Trpπ,n proved in 7.5 and 6.6. (cid:3) Definition 7.8. Under the assumptions of 7.6, we can use the previous constructions to define a morphism τi,π : Rf∗(W (OY )) −→ W (OX ) which commutes with F and V , and is such that Rn ◦ τi,π = τi,π,n ◦ Rn for all n, Rn denoting both restriction maps W (OX) → Wn(OX ) and W (OY ) → Wn(OY ). To construct τi,π, we first recall that, for any scheme X, the inverse system (Wn(OX ))n≥0 is lim←−-acyclic, as the cohomology of each term vanishes on affine open subsets, and the inverse system of sections on such a subset has surjective transition maps. So, if f• ∗ denotes the obvious extension of the direct image functor to the category of inverse systems, it suffices to define a morphism (7.8.1) τi,π,• : Rf• ∗(W•(OY )) −→ W•(OX ) in the derived category of inverse systems on X, and to apply the functor R lim←− and the canonical isomorphism Rf∗ ◦ R lim←− ◦ Rf• ∗. On the one hand, the relations R(γY,n+1) = γY,n imply that, for variable n, the fundamental classes define a morphism of inverse systems i• ∗(W•(OY )) → Hd P/X). As the canonical morphisms ≃ R lim←− Y (W•Ωd Hd Y (W•Ωd P/X) ∼ −−→ RΓY (W•Ωd P/X[d]) −→ W•Ωd P/X[d] make sense in the derived category of inverse systems, we can define in this derived category a morphism γi,π,• : i• ∗(W•(OY )) → W•Ωd P/X[d] which has the morphisms γi,π,n defined in (7.5.1) as components. On the other hand, the homomorphisms dlog n used to define Chern classes for invertible bundles form an inverse system of homomorphisms, hence, for variable n, the powers of the Chern classes of OP (1) define a morphism W•(OP )[−d] → Rπ• ∗(W•Ωd P/X), which is an isomorphism of the derived category of inverse systems. Composing its inverse with the projection by Rπ• ∗ of γi,π,• provides τi,π,•. It is clear that τi,π,• has the morphisms τi,π,n as components, and commutes with F and V . Then the morphism (7.8.2) τi,π : Rf∗(W (OY )) has the required properties. ∼ −−→ R lim←− Rf• ∗(W•(OY )) R lim←−(τi,π,•) −−−−−−−→ W (OX ) Finally, as f is a morphism of noetherian schemes, f∗ and Rf∗ commute with ten- sorisation with Q. So we can define a morphism again denoted τi,π : Rf∗(W OY,Q) −→ RATIONAL POINTS OF REGULAR MODELS 51 W OX,Q by (7.8.3) τi,π : Rf∗(W OY,Q) ∼ −−→ Rf∗(W OY ) ⊗ Q τi,π⊗Q −−−−→ W OX,Q. This morphism also commutes with F and V . 8. Proof of the injectivity theorem for Witt vector cohomology The main result of this section is Theorem 8.1 below, which gives an injectivity property for the functoriality morphisms induced on Witt vector cohomology by some complete intersection morphisms of virtual relative dimension 0. As explained in Remark 8.2, Theorem 1.5 is a particular case of this result. Theorem 8.1. Let f : Y → X be a projective morphism between two flat noether- ian Z(p)-schemes with dualizing complexes, which is complete intersection of virtual relative dimension 0. We assume that there exists a scheme-theoretically dense open subscheme U ⊂ X such that f −1(U ) → U is finite locally free of constant rank r ≥ 1. Let fn : Yn → Xn be the reduction of f mod pn+1. (i) For all q ≥ 0, the kernels of the functoriality homomorphisms (8.1.1) (8.1.2) (8.1.3) (8.1.4) f ∗ : H q(X, OX ) −→ H q(Y, OY ), f ∗ n : H q(Xn, OXn) −→ H q(Yn, OYn), f ∗ 0 : H q(X0, Wn(OX0)) −→ H q(Y0, Wn(OY0)), f ∗ 0 : H q(X0, W (OX0)) −→ H q(Y0, W (OY0 )), are annihilated by r. (ii) For all q ≥ 0, the functoriality homomorphism (8.1.5) is injective. f ∗ 0 : H q(X0, W OX0,Q) −→ H q(Y0, W OY0,Q) Remark 8.2. Theorem 8.1 implies Theorem 1.5. Indeed, let f : Y → X be as in 1.5. The morphisms Xk ֒→ X0 and Yk ֒→ Y0 are nilpotent immersions, hence the canonical homomorphisms H q(X0, W OX0,Q) −→ H q(Xk, W OXk,Q), H q(Y0, W OY0,Q) −→ H q(Yk, W OYk,Q) are isomorphisms [BBE07, Prop. 2.1]. Therefore it suffices to check that f satisfies the hypotheses of Theorem 8.1. We may assume that X is connected, and replace Y by one of its connected components mapping surjectively to X, so that X and Y are integral schemes. At any closed point y ∈ Y , with image x = f (y), we may choose a closed immersion Y ֒→ P around y, with P smooth over X. If dim OX,x = n, then OP,y is a regular local ring of dimension n + d for d = dim(P/X), and OY,y is a regular quotient of OP,y of dimension n. Therefore, the ideal I of Y in P is regular of codimension d around y, and it follows that f is complete intersection of virtual relative dimension 0. Moreover, the function field extension K(X) ֒→ K(Y ) is finite, hence f is finite and locally free of constant rank ≥ 1 above a non empty open subset U . As X is integral, U is scheme-theoretically dense and the hypotheses of Theorem 8.1 are satisfied. 52 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING In order to prove Theorem 8.1, we will choose a factorization f = π ◦ i, where i : Y ֒→ P = P d X is a closed immersion, and π : P → X the structural morphism. Let i0, π0 be the reductions mod p of i, π. The key point will be to relate the trace morphisms τi0,π0,n constructed in 7.6 to the trace morphism τf given by Theorem 3.1, and this is made possible by the following constructions. Lemma 8.3. Let X be a scheme on which p is locally nilpotent, P a smooth X- scheme, a ⊂ OX a quasi-coherent ideal, X ′ ֒→ X the closed subscheme defined by a, P ′ = X ′ ×X P . For each n ≥ 1, let N • P/X be the additive subgroup generated by sections of the form n ⊂ WnΩ• (8.3.1) V r([a]ω), dV r([a]ω), with a ∈ a, ω ∈ Wn−rΩ• Then, for variable n, the canonical homomorphisms WnΩ• a transitive family of isomorphisms P/X, 0 ≤ r ≤ n − 1. P/X → WnΩ• P ′/X ′ induce (8.3.2) WnΩ• P/X/N • n ∼ −−→ WnΩ• P ′/X ′. Proof. Thanks to (5.1.2), one first notices that N • n is a differential graded ideal of P/X. Using (5.1.5), one sees that, for all n ≥ 1, V (N • WnΩ• n+1. Using (5.1.1) (and a direct computation for r = 0), one sees that F (N • n. Therefore, the projective system {WnΩ• n} is an F -V -procomplex over P/X. In degree 0, it is easy to see by induction on n that the ideal N 0 n ⊂ Wn(OP ) is the kernel of Wn(OP ) → Wn(OP ′). It follows that {WnΩ• n} is actually an F -V -procomplex over P ′/X ′. It is then clear that it satisfies the universal property which defines P ′/X ′ }, which implies that (8.3.2) is an isomorphism of F -V -procomplexes. (cid:3) {WnΩ• n+1) ⊂ N • P/X/N • P/X/N • n ) ⊂ N • Proposition 8.4 (see also [Ol07, Th. 4.2.3]). Let X be a Z(p)-scheme and denote Xn = X ⊗Z(p) Z(p)/pn+1. (i) For all n ≥ 1, there exists a unique homomorphism of sheaves of rings making the following diagram commute eF n : Wn(OX0 ) −→ OXn−1 Wn+1(OXn−1 ) F n 7ooooooooooo eF n OXn−1 , Wn(OX0) where the vertical map is the natural reduction map. Furthermore, if we assume X to be flat over Z(p) and denote by Rn : W (OX0) → Wn(OX0) the natural reduction map, then (8.4.1) Ker(F − Id : W (OX0) → W (OX0)) ∩ \n≥1 Ker(eF n ◦ Rn) = 0. (ii) Let P be a smooth X-scheme and denote Pn = P ×X Xn. For all n ≥ 1, there exists a unique homomorphism of sheaves of graded algebras eF n : WnΩ• P0/X0 −→ H•(Ω• Pn−1/Xn−1 ) / /   7 RATIONAL POINTS OF REGULAR MODELS 53 making the following diagram commute Wn+1Ω• Pn−1/Xn−1 F n ZΩ• Pn−1/Xn−1 WnΩ• P0/X0 eF n / H•(Ω• Pn−1/Xn−1 ). Furthermore, for all a ∈ O× P0 and all a ∈ O× Pn−1 lifting a, we have (8.4.2) eF n(dlog ([a])) = cl(da/a). When X0 is a perfect scheme and Xn−1 = Wn(X0), eF n is the isomorphism −−→ H•(Ω• θn : WnΩ• Pn−1/Xn−1 (8.4.3) P0/X0 ∼ ) defined by Illusie-Raynaud [IR83, III, (1.5)]. Note that, in formula (8.4.2), the class of da/a does not depend upon the choice of the liftng a: if b = a + pw, then db/b = da/a + d(log(1 + p w a )), where log(1 + pw/a) is defined thanks to the canonical divided powers of p. Proof. (i) We may assume X is affine. The kernel of the vertical map in the diagram is locally generated (as an abelian group) by elements of the form V n([a]) and V r([pb]) for some a, b ∈ OPn−1 and 0 ≤ r ≤ n. As these elements are clearly mapped to 0 under F n, this gives the unique existence of eF n. To prove (8.4.1), let w ∈ Ker(F − Id) ∩Tn Ker(eF n ◦ Rn). If w 6= 0, we can write V i([ai]), with ai ∈ OX0 and as 6= 0. w = ∞Xi≥s Then Rs+1(w) = V s([as]) ∈ Ws+1(OX0). If as ∈ OXs is any lifting of as, and if [as] is the Teichmuller representative of as in W2(OXs), so that V s([as]) is a lifting of V s([as]) in Ws+2(OXs), we have by construction eF s+1(V s([as])) = F s+1(V s([as])) = psF ([as]) = psap in OXs. Thus eF s+1(Rs+1(w)) = 0 if and only if psap over Z/ps+1Z, we obtain ap we have s ∈ pOXs, in particular ap s s = 0 in OXs. Since Xs is flat s = 0 ∈ OX0. But by assumption F (w) =Xi≥s V i([ap i ]) =Xi≥s V i([ai]) = w. Hence as = ap s = 0, a contradiction. Pn−1/Xn−1 is clearly contained in ZΩ• (ii) First of all, since dF n = pnF nd, the image of F n : Wn+1Ω• → . Thus the diagram makes sense. Ω• Now, Lemma 8.3 and [LZ04, Prop. 2.19] imply that, in degree q, the kernel of the vertical map on the left hand side is locally generated (as an abelian group) by sections of the following form Pn−1/Xn−1 Pn−1/Xn−1 (8.4.4) V n(α), dV n(β), V r([p]ω), dV r([p]η), / /     / 54 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING Pn−1/Xn−1 Pn−1/Xn−1 , β ∈ Ωq−1 with α ∈ Ωq η ∈ Wn+1−rΩq−1 are mapped to 0 in Hq(Ω• representative [a]m in Wm(OPn−1 ), we have F ([a]m) = [a]p obtain for the last two sections and . One immediately sees that, via F n, the first two sections ). Since, for any m ≥ 1 and any Teichmuller m−1 ∈ Wm−1(OPn−1), we , 0 ≤ r ≤ n, ω ∈ Wn+1−rΩq Pn−1/Xn−1 Pn−1/Xn−1 Pn−1/Xn−1 F nV r([p]ω) = prF n−r([p]ω) = ppn−r+rF n−r(ω) = 0, F ndV r([p]η) = F n−rd([p]η) = F n−r([p]dη) = ppn−r−(n−r)d(F n−r(η)) = 0 in Hq(Ω• 0 in Hq(Ω• If a ∈ O× Pn−1/Xn−1 Pn−1/Xn−1 lifts a, we get by construction Pn−1 ). Thus F n maps all elements in the kernel of the vertical map to ). Since the vertical map is surjective, this yields the statement. which gives (8.4.2). eF n(d[a]/[a]) = cl(F n(d[a]/[a])) = cl([a]pn−1d[a]/[a]pn ), Finally, let us assume that X0 is perfect and Xn−1 = Wn(X0). By [IR83, III, ) has the structure of a differential graded algebra (dga) with ) given by the boundary of (1.5)], H•(Ω• the differential d : Hi(Ω• the long exact cohomology sequence coming from the short exact sequence ) → Hi+1(Ω• Pn−1/Xn−1 Pn−1/Xn−1 Pn−1/Xn−1 0 −→ Ω• Pn−1/Xn−1 pn −−→ Ω• P2n−1/X2n−1 −→ Ω• Pn−1/Xn−1 −→ 0. The isomorphism θn is compatible with the differential and the product, and in- duces thus an isomorphism of dga's θn : WnΩ• ). On the −−→ H•(Ω• ∼ P0/X0 Pn−1/Xn−1 other hand, it follows from the relation dF n = pnF nd that the morphism eF n is compatible with the differentials. Therefore eF n also induces a morphism of dga's eF n : WnΩ• ). In degree 0, θn is defined by 0 + papn−1 θn(a0, . . . , an−1) = apn + · · · + pn−1ap −−→ H•(Ω• Pn−1/Xn−1 n−1, P0/X0 ∼ 1 i=0 piapn−i where a0, . . . , an−1 are liftings to OPn−1 of a0, . . . , an−1 [IR83, p. 142, l. 8]. This definition shows that, in degree 0, θn is the factorization of the n-th ghost component , with pnan = 0 in OPn−1. From the definition of the morphism of functors F n : Wn+1 → W1, we wn : Wn+1(OPn−1 ) → OPn−1, given by wn(a0, . . . , an) =Pn also get that, in degree 0, eF n is the factorization of the n-th ghost component. Since eF n = θn in degree 0 and WnΩ• 0, eF n and θn have to be equal. Lemma 8.5. Let S be Spec Z(p), X an S-scheme, π : P := P d X → X the struc- tural morphism of a projective space over X. For n ≥ 0, denote by Sn, Xn, Pn, πn the reductions modulo pn+1, and let BΩd be the subsheaf of exact differential forms. is generated as dga by its sections in degree Pn−1/Xn−1 ⊂ Ωd Pn/Xn Pn/Xn (cid:3) i (i) For all n ≥ 0, the canonical homomorphism Pn/Xn) −→ Rdπn ∗(Ωd bd n : Rdπn ∗(Ωd (8.5.1) Pn/Xn/BΩd Pn/Xn) is an isomorphism. RATIONAL POINTS OF REGULAR MODELS 55 (ii) Assume that X is flat over S, and let Y0 ֒→ P0 be a regular closed immersion of codimension m. Then, (8.5.2) ∀ j 6= m, ∀ n ≥ 0, Hj Y0 (Ωd Pn/Xn/BΩd Pn/Xn) = 0. Proof. Let Q = P d S , and let T0, . . . , Td be homogeneous coordinates on Q. We define an S-endomorphism φ : Q → Q by sending Ti to T p i , 0 ≤ i ≤ d. By base change by u : X → S, we obtain an X-endomorphism of P , for which we will keep the notation φ, as well as for its reduction mod pn+1. Let us fix n ≥ 0. We can use the morphism φn+1 and view φn+1 as a complex of quasi-coherent OPn-modules, the differential of which is then OPn-linear. But Pn has an open covering by d + 1 open subsets which are relatively affine with respect to Xn, and therefore Rdπn ∗ is a right exact functor on the category of quasi-coherent OPn-modules. As Rdπn ∗(Ωd−1 ∗ Ω• Pn/Xn To prove assertion (ii), we use φn+2 to view φn+2 as a complex of quasi- coherent OPn-modules with an OPn-linear differential, and we claim that the sheaf of OPn-modules Pn/Xn Pn/Xn ) = 0, assertion (i) follows. ∗ Ω• Hd(φn+2 ∗ Ω• Pn/Xn) = φ∗(φn+1 ∗ Ωd Pn/Xn/Bφn+1 ∗ Ωd Pn/Xn) has a filtration by sub-OPn-modules, the graded of which is locally free over OP0. As Y0 is locally defined in P0 by a regular sequence of m sections, the claim clearly implies assertion (ii). To prove the existence of this filtration, we may replace X, P by S, Q, because the projection v : P → Q is flat, and v∗(φn+2 ∗ Ω• Qn/Sn) ∼ −−→ φn+2 ∗ Ω• Pn/Xn. Now S0 is a perfect scheme, and Sn = Wn+1(S0). Thanks to the last assertion of Proposition 8.4 (ii), F n+1 defines an isomorphism of graded algebras ∼ −−→ H•(Ω• Qn/Sn). Q0/S0 eF n+1 : Wn+1Ω• Q0/S0 OQn-module structure provided by the homomorphism OQn → H0(Ω• by φn+2, and Wn+1Ω• ) with the ) defined with the structure corresponding to the previous one via ∼ −−→ Wn+1(OQ0). The canonical filtration of Wn+1Ωd is then a filtration by sub-OQn-modules, which can be transported to Hd(Ω• We may view eF n+1 as an OQn-linear isomorphism by endowing H•(Ω• (eF n+1)−1 : H0(Ω• via eF n+1. As we know by [Il79, I, Cor. 3.9] that the corresponding graded pieces are locally free OQ0-modules for the structure defined by the homomorphism Q0/S0 ) Qn/Sn Qn/Sn Qn/Sn Qn/Sn ) (8.5.3) F : OQ0 −→ Wn+1(OQ0)/pWn+1(OQ0) 56 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING factorizing F : Wn+1(OQ0) → Wn+1(OQ0), the proof will be complete if we check the commutativity of the diagram (8.5.4) OQn φn+2 ∗ H0(Ω• Qn/Sn ) ( eF n+1)−1 ∼ / Wn+1(OQ0) OQ0 F / Wn+1(OQ0)/pWn+1(OQ0). It is enough to check that the diagram induced on sections over D+(Ti) ⊂ Qn commutes, for 0 ≤ i ≤ d. So we may replace OQn by A = (Z/pn+1Z)[x], with x = (x1, . . . , xd) and φ∗(xj) = xp aI ∈ Z/pn+1Z. Then j , 1 ≤ j ≤ d. Take f = PI aI xI ∈ A, with (eF n+1)−1 ◦ φn+2 ∗(f ) = XI aI (eF n+1)−1(xpn+2I ). As eF n+1 is the factorization of the (n+2)-th ghost component wn+1 : Wn+2(A) → A, we see that (eF n+1)−1(xpn+1 ) = [xj], 1 ≤ j ≤ d. Therefore, we obtain aI [x]pI . j (eF n+1)−1 ◦ φn+2 ∗(f ) = XI Since F is given by lifting an element of A0 to Wn+1(A0), applying Frobenius and reducing modulo p, this gives the commutativity of (8.5.4). (cid:3) Proposition 8.6. Under the assumptions of Theorem 8.1, let f = π ◦ i be a factor- ization of f as the composition of a regular closed immersion i : Y ֒→ P = P d X of Y into a projective space on X, followed by the canonical projection π : P → X. For all n ≥ 1, let fn, in, πn be the reductions of f, i, π modulo pn+1. Then the compositions (8.6.1) (8.6.2) (8.6.3) (8.6.4) (8.6.5) OX f ∗ −→ Rf∗(OY ) f ∗ n−→ Rfn ∗(OYn) OXn f ∗ 0−→ Rf0 ∗(Wn(OY0)) Wn(OX0) W (OX0) W OX0,Q f ∗ 0−→ Rf0 ∗(W (OY0)) f ∗ 0−→ Rf0 ∗(W OY0,Q) τf−→ OX , τfn−−→ OXn, τi0,π0,n −−−−−→ Wn(OX0 ), τi0,π0−−−→ W (OX0), τi0,π0−−−→ W OX0,Q, are given by multiplication by r. Proof. Since the restriction of f above U is finite locally free of rank r, it follows from (3.1.3) that the endomorphism of OU induced by τf ◦ f ∗ is mutiplication by r. But U is scheme-theoretically dense in X, therefore the same relation holds on X itself. So (8.6.1) is multiplication by r. Thanks to the flatness of X and Y over Z(p), the spectral sequence for the com- position of Tor's implies that, for all n ≥ 1, Xn and Y are Tor-independent over X. Therefore, by Theorem 3.1, (ii), the morphism τfn ◦ f ∗ n is deduced from τf ◦ f ∗ by base change from X to Xn, and (8.6.2) is also multiplication by r. / /     /     / RATIONAL POINTS OF REGULAR MODELS 57 We want to deduce from this result that (8.6.3) is also multiplication by r. We observe first that the homomorphisms eF n defined by Lemma 8.4 provide morphisms X : Wn(OX0) −→ OXn−1, Y ) : f0 ∗(Wn(OY0)) −→ fn−1 ∗(OYn−1 ), P0/X0 ) −→ Rdπn−1 ∗(Ωd Pn−1/Xn−1 /BΩd Pn−1/Xn−1 ). eF n P ) : Rdπ0 ∗(WnΩd f∗(eF n n)−1 ◦ Rdπ∗(eF n Rdπ∗(eF n eGn P := (bd We consider the diagram (8.6.6) Moreover, we can use the isomorphism (8.5.1) and define P ) : Rdπ0 ∗(WnΩd P0/X0 ) −→ Rdπn−1 ∗(Ωd Pn−1/Xn−1 ). Wn(OX0 ) f ∗ 0 / f0 ∗(Wn(OY0)) π0 ∗(γi0 ,π0,n) / Rdπ0 ∗(WnΩd P0/X0 Trpπ0,n ∼ / ) / Wn(OX0 ) eF n X OXn−1 f∗( eF n Y ) eGn P eF n X f ∗ n−1 / fn−1 ∗(OYn−1 ) πn−1 ∗(γfn−1 ) / Rdπn−1 ∗(Ωd Pn−1/Xn−1 Trpπn−1 ∼ / / OXn−1 , ) where the compositions of the upper and lower rows are respectively the maps in- duced by (8.6.3) and (8.6.2) on degree 0 cohomology. Let us prove that this diagram X is functorial with respect to X. To prove that the right square commutes, it suffices to show that, if ξdRW and ξdR are the de Rham-Witt and de Rham Chern classes of OP (1), then ξd ). As is commutative. The left square commutes because the morphism eF n R•πn−1 ∗(eF n n are compatible with cup-products, it suffices to show that the dR have same image in Rdπn−1 ∗(Ωd dRW and ξd P ) and b• diagram Pn−1/Xn−1 Pn−1/Xn−1 /BΩd dlog R1π0 ∗(O× P0 ) / R1π0 ∗(WnΩ1 ) P0/X0 R1πn−1 ∗(O× Pn−1 ) dlog / R1πn−1 ∗(H1(Ω• Pn−1/Xn−1 )) R1π∗( eF n P ) is commutative, which follows from (8.4.2). To simplify notations, we drop the base scheme from the indices, and denote . To prove the commutativity of the central square of (8.6.6), C d it suffices to prove the commutativity of the diagram /BΩd = Ωd Pn−1 Pn−1 Pn−1 i0 ∗(Wn(OY0)) Hd Y0 (WnΩd P0 ) ∼ RΓY0(WnΩd P0 )[d] / WnΩd P0 [d] i∗( eF n Y ) Hd Y ( eF n P ) Hd Yn−1 (C d Pn−1 RΓY ( eF n P )[d] eF n P [d] ∼ ) RΓYn−1(C d Pn−1 )[d] / C d Pn−1 [d] in−1 ∗(OYn−1 ) / Hd Yn−1 (Ωd Pn−1 ) ∼ RΓYn−1(Ωd Yn−1 )[d] / Ωd Yn−1 [d] , to apply the functor Rπn−1 ∗, and to pass to cohomology sheaves in degree 0. In this diagram, the upper left (resp. lower left) horizontal arrow maps 1 to γY0,n (resp.   /   /     / / /   O O / / /     o o   /   o o / / O O o o O O / O O 58 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING γYn−1,1), and the middle horizontal arrow is an isomorphism thanks to Lemma 8.5 (ii). The middle and right squares commute by functoriality, and it suffices to prove that the left rectangle commutes. This part of the diagram comes from a diagram of morphisms of sheaves, therefore the verification is local on P . Thus we may assume that Y is defined by a regular sequence t1, . . . , td in P . Then, since Y and P are flat over Z(p), the images of this sequence in OPn−1 and OP0 (still denoted t1, . . . , td) are regular sequences defining Yn−1 and Y0. It is enough to show that the symbols (cid:20) d[t1] · · · d[td] [t1], . . . , [td] (cid:21) ∈ Ext d Wn(OP0 )(Wn(OY0), WnΩd P0) and OPn−1 Pn−1) (OYn−1 , Ωd (cid:20) dt1 · · · dtd t1, . . . , td (cid:21) ∈ Ext d ). By functoriality, the image of (cid:20) dt1 · · · dtd t1, . . . , td (cid:21) in (cid:21). On the other hand, it follows from the ) is(cid:20) cl(dt1 · · · dtd) t1, . . . , td have same image in Hd Y (C d Pn−1 Ext d OPn−1 (OYn−1, C d Pn−1 Pn−1 cl(tpn−1 i P ([ti]) = tpn dti) ∈ H1(Ω• ). Since the tpn i construction of eF n in Proposition 8.4 that eF n argue as in the proof of Proposition 7.4 to show that the symbols(cid:20) d[t1] · · · d[td] [t1], . . . , [td] (cid:21) and" cl(tpn−1 P (d[ti]) = 's form a regular sequence in OPn−1, we may # have same image in Hd i ∈ OPn−1, and eF n · · · tpn−1 tpn 1 , . . . , tpn ). The wanted dt1 · · · dtd) d (C d Pn−1 Yn−1 d 1 equality is then a consequence of Lemma 4.2, and the commutativity of (8.6.6) follows. Returning to the homomorphism (8.6.3), we observe that it is defined by multipli- cation by a section κn of Wn(OX0 ). Proposition 7.7 (i) implies that, for variable n, the sections κn form a compatible family under restriction, and satisfy F (κn) = κn−1. κn ∈ Γ(X0, W (OX0)), then F (κ−r) = κ−r. On the other hand, the com- If κ = lim←−n X (κn − r) = 0. So, if Rn : W (OX0) → Wn(OX0) mutativity of (8.6.6) implies that eF n is the restriction homomorphism, we obtain that κ − r ∈ Ker(F − Id) ∩ \n≥1 X ◦ Rn), Ker(eF n which is zero by (8.4.1). Thus κ = r, hence κn = r for all n. If we now consider in the derived category of inverse systems the composition W•(OX0 ) f ∗ 0 •−−→ Rf0 • ∗(W•(OY0 )) τi,π,•−−−→ W•(OX0 ), we obtain a morphism which has (8.6.3) as component of degree n. Therefore, this composition is multiplication by r on the inverse system W•(OX0 ). It follows that the composition W (OX0) R lim←− f ∗ −−−−−→ R lim←− 0 • Rf0 • ∗(W•(OY0)) R lim←− τi,π,• −−−−−−→ W (OX0) RATIONAL POINTS OF REGULAR MODELS 59 is multiplication by r. Using the isomorphism R lim←− ◦ Rf0 • ∗ ≃ Rf0 ∗ ◦ R lim←−, we obtain that (8.6.4) is multiplication by r. Tensoring by Q and using the commutation of Rf0 ∗ with tensorisation by Q, we obtain that (8.6.5) is multiplication by r. (cid:3) 8.7. Proof of Theorem 8.1. The first assertion is a particular case of Theorem 3.2. To prove the other ones, we choose a factorization f = π◦i, where i is a closed immersion of Y into a projective space P = P d X over X, and π is the structural morphism, and we keep the notations of the previous subsections. Applying the functor H q(Xn, −) (resp. H i(X0, −)), the morphisms τfn, τi,π,n and τi,π define homomorphisms H q(Yn, OYn) H q(Y0, Wn(OY0 )) H q(Y0, W (OY0)) H q(Y0, W OY0,Q) τfn−−→ H q(Xn, OXn ), τi,π,n−−−→ H q(X0, Wn(OX0)), τi,π−−→ H q(X0, W (OX0)), τi,π−−→ H q(X0, W OX0,Q). Proposition 8.6 implies that the composition of these homomorphisms with the func- toriality homomorphisms defined by fn (resp. f0) is multiplication by r, and this implies Theorem 8.1. (cid:3) This also completes the proof of Theorems 1.5, 1.3 and 1.1. 9. An example Because Theorem 1.1 was previously known in some cases, and can be proved in some other cases without using the most difficult results of this paper, it may be worth giving an example for which we would not know how to prove congruence (1.1.1) without using them. We give here such an example for each p ≥ 7, except perhaps when p is a Fermat number. 9.1. We begin with a list of conditions that we want our example to satisfy. In these conditions, R, K and k are as in Theorem 1.1, and X is an R-scheme. (1) X is a regular scheme, projective and flat over R. (2) H 0(XK , OXK ) = K, and H q(XK , OXK ) = 0 for all q ≥ 1. (3) There exists q ≥ 1 such that H q(Xk, OXk ) 6= 0. (4) X is not a semi-stable R-scheme (in particular, not smooth). (5) dim XK ≥ 3. (6) XK is a variety of general type. Conditions (1) and (2) will ensure that X satisfies the hypotheses of Theorem 1.1. Condition (3) will ensure that we are not in the trivial situation described in the first paragraph of subsection 1.4. Condition (4) will ensure that Theorem 2.1 does not suffice to conclude. Condition (5) will rule out the case of surfaces, for which Theorem 1.1 is already known by [Es06, Th. 1.3]. Condition (6) rules out rationally connected varieties, for which Theorem 1.1 is also known because they satisfy the coniveau condition of [Es06, Th. 1.1]. It also grants that, if X can be embedded as a global complete intersection in some projective space over R, then congruence (1.1.1) cannot be proved by applying Katz's theorem [Kz71, Th. 1.0] to Xk, since a 60 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING smooth complete intersection in a K-projective space for which Katz's µ invariant is ≥ 1 is a Fano variety. Remarks 9.2. We begin with a few remarks that make it easier to find an example satisfying the previous conditions. (i) Examples such that dimk H 1(Xk, OXk ) > dimK H 1(XK , OXK ) = 0 have been known since Serre's construction of a counter-example to Hodge symmetry in characteristic p [Se58, Prop. 16]. The general principle behind such examples, which goes back to Grothendieck (see [SGA 1, XI, 6.11, (*)] over an algebraically closed field, and [Ra70, Prop. 6.2.1] for a general statement), is that the datum of a torsor Y on X under a finite group G defines a morphism G′ → PicX/R, where G′ is the Cartier dual of G. Then, under certain conditions, the Lie algebra of G′ k can have a non-zero image in the tangent space H 1(Xk, OXk ) to PicXk/k. The simplest case (which was the one considered by Serre) is when G is the ´etale group Z/pZ. Then the Artin-Schreier exact sequence shows that, when the torsor Yk remains non-trivial after extension to an algebraic closure k of k, its class gives a non-zero element in H 1(Xk, OXk ), and therefore H 1(Xk, OXk ) 6= 0. This happens in particular when Yk is a complete intersection in some projective space, since we then have dimk H 0(Yk, OYk ) = 1. To simplify our quest, we will therefore replace condition (3) (and condition (5)) by the more restrictive condition: (3') X is the quotient of an hypersurface Y in a projective space Pn R of relative dimension n ≥ 4 over R by a free action of the group Z/pZ. (ii) Assume that X satisfies condition (3'). Then H 0(YK , OYK ) = K, and H q(YK, OYK ) = 0 for q 6= 0, n − 1. Because char(K) = 0, we have H q(XK , OXK ) = H q(YK, OYK )G. Hence, H 0(XK , OXK ) = K, and condition (2) is satisfied if and only if χ(OXK ) = 1. As YK is an ´etale cover of XK of degree p, the Riemann-Roch- Hirzebruch formula implies that (9.2.1) χ(OYK ) = pχ(OXK ). Then condition (2) is satisfied if and only if χ(OYK ) = p. If d is the degree of the hypersurface Y , we obtain (−1)n−1(p − 1) = dimK H n−1(YK , OYK ) = dimK H n(Pn = dimK H 0(Pn K, OPn K, OPn K K (−d)) (d − n − 1)). The simplest choice for checking this equation is d − n − 1 = 1, so that we get dimK H 0(Pn (d − n − 1)) = n + 1. Then we have to satisfy the conditions K, OPn K (9.2.2) p > 2, n = p − 2, d = p. Therefore, we will simplify even further our quest by replacing condition (3') by the following more precise condition, which implies (2), (3) and (5): (3") X is the quotient of an hypersurface Y of degree p in the projective space Pn R of relative dimension n = p − 2 over R by a free action of the group Z/pZ, with p ≥ 7. RATIONAL POINTS OF REGULAR MODELS 61 (iii) Assuming that X satisfies conditions (1) and (3"), then condition (6) follows automatically. Indeed, YK is smooth over K since char(K) = 0, and its canonical sheaf is then OYK (−n − 1 + d) = OYK (1). Since YK is an ´etale covering of XK, it is the inverse image of the canonical sheaf on X, which therefore is ample too. So it suffices for our purpose to construct an example satisfying conditions (1), (3") and (4). 9.3. We now begin the construction of our example. Assume that p ≥ 5, and let E be the free Z(p)-module (Z(p))p. We denote by e0, . . . , ep−1 its canonical basis. Let σ be a generator of G := Z/pZ. We let σ act on E by cyclic permutation of the basis: (9.3.1) σ : e0 7→ e1 7→ · · · 7→ ep−1(7→ e0). Let H ⊂ E be the hyperplane consisting of elements for which the sum of coordinates is 0. It is stable under the action of G, and we endow it with the basis v1, . . . , vp−1 defined by vi = ei − ei−1. We take as projective space the space P(H) ≃ Pp−2 , with Z(p) the induced G-action, and we denote by X1, . . . , Xp−1 the homogeneous coordinates on P(H) defined by the dual basis to the basis v1, . . . , vp−1 of H. Letting G act by composition on functions on H, one checks easily that the orbit of X1 is described by (9.3.2) X1 7→ −Xp−1 7→ Xp−1 − Xp−2 7→ Xp−2 − Xp−3 7→ · · · 7→ X2 − X1 (7→ X1). Let g0(X1, . . . , Xp−1) be the sum of the elements of the orbit of X p 1 , i.e., (9.3.3) g0(X1, . . . , Xp−1) = X p 1 + (−Xp−1)p + (Xi − Xi−1)p. p−1Xi=2 Then g0 ∈ pZ[X1, . . . , Zp−1], and we can define a polynomial g(X1, . . . , Xp−1) ∈ Z[X1, . . . , Zp−1] by (9.3.4) g(X1, . . . , Xp−1) = 1 p g0(X1, . . . , Xp−1). Let Z ⊂ P(H) be the hypersurface defined by g. Since g is G-invariant, the action of G on P(H) induces an action on Z. We denote by g the reduction of g in Fp[X1, . . . , Xp−1]. We first study the singular points of ZFp. They are solutions of the system of homogeneous equations ∂g/∂Xi = 0, 1 ≤ i ≤ p − 1, which can be written as (9.3.5) X p−1 1 = (X2 − X1)p−1 (X2 − X1)p−1 = (X3 − X2)p−1 ... = ... (Xp−1 − Xp−2)p−1 = (−Xp−1)p−1 .  Lemma 9.4. Let Fp be an algebraic closure of Fp. (i) The solutions of (9.3.5) in Pn(Fp) belong to Pn(Fp), and they correspond bijectively to the families (u1, . . . , up−1) ∈ (F× p )p−1 such that (9.4.1) 1 + u1 + · · · + up−1 = 0. 62 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING (ii) For u ∈ F× p , let u = [u] ∈ µp−1(Zp) be its Teichmuller representative. Then a point x ∈ Pn(Fp) which is a solution of (9.3.5) belongs to ZFp if and only if (9.4.2) where (u1, . . . , up−1) ∈ (F× 1 + u1 + · · · + up−1 ∈ p2Zp, p )p−1 corresponds to x by (i). p )p−1 satisfying (9.4.1), the corresponding solution Proof. Given (u1, . . . , up−1) ∈ (F× x = (ξ1 : . . . : ξp−1) ∈ Pn(Fp) of the system (9.3.5) is obtained by choosing ξ1 ∈ F× p , setting (9.4.3) ξi − ξi−1 = ui−1ξ1 for 2 ≤ i ≤ p − 1, and observing that (9.4.1) implies that −ξp−1 = up−1ξ1. Assertion (i) is then straightforward. Let η1 ∈ Zp be a lifting of ξ1, and let ηi be defined inductively for 2 ≤ i ≤ p − 1 by (9.4.4) Define α ∈ Zp by (9.4.5) ηi − ηi−1 = ui−1η1. 1 + u1 + · · · + up−1 = pα. Then we get by adding the equations in (9.4.4) (9.4.6) ηp−1 = (1 + · · · + up−2)η1 = (pα − up−1)η1. We can now substitute (9.4.4) and (9.4.6) in g0, and we obtain the relation g0(η1, . . . , ηp−1) = ηp 1(1 + up 1 + . . . + up p−2 + (up−1 − pα)p) (9.4.7) = ηp 1(1 + u1 + . . . + up−2 + up−1 + ≡ pαηp 1 mod p2Zp. pXj=1(cid:18)p j(cid:19)up−j p−1(−pα)j) Hence we get (9.4.8) g(η1, . . . , ηp−1) ≡ αηp 1 mod pZp, and assertion (ii) follows. (cid:3) Lemma 9.5. (i) The action of G on ZFp is free. (ii) If p is not a Fermat number, then ZFp is singular, and is not the special fibre of a semi-stable scheme. Let us recall that the Fermat numbers are the integers of the form 22n + 1 with n ≥ 0, that any prime number of the form 2n + 1 with n > 0 is a Fermat number, and that the only known prime Fermat numbers are 3, 5, 17, 257 and 65537. Proof. Over Fp, the matrix of the action of σ on (Fp)p has 1 as unique eigen- value, with a corresponding eigenspace of dimension 1, generated by the eigen- vector (1, . . . , 1). This eigenvector belongs to H/pH, where it has coordinates (p − 1, p − 2, . . . , 1) = −(1, . . . , p − 1) in the basis v1, . . . , vp−1. Therefore, the only fixed point of σ in Pn(Fp) is the point x0 = (1 : 2 : . . . : p − 1). This point is the RATIONAL POINTS OF REGULAR MODELS 63 solution of (9.3.5) corresponding to u1 = . . . = up−1 = 1. Lemma 9.4 (ii) implies that it does not belong to ZFp, which proves assertion (i). As the system (9.3.5) has only a finite number of solutions, the singular points of ZFp are isolated. In particular, since dim ZFp ≥ 4, ZFp cannot be the special fibre of a semi-stable scheme if it has a singular point. To find a singular point on ZFp, Lemma 9.4 shows that it suffices to construct a family (ui)1≤i≤p−1 of (p − 1)-th roots of unity in Zp such that 1 +Pi ui ∈ p2Zp. Since p is not a Fermat number, p − 1 has an odd prime factor q. We can choose a primitive q-th root of unity ζ, and set ui = ζ i for 1 ≤ i ≤ q − 1, ui = 1 for q ≤ i ≤ q + (p − q)/2 − 1, ui = −1 for q + (p − q)/2 ≤ i ≤ p − 1. So ZFp is singular. (cid:3) 9.6. We now address the regularity condition in 9.1 (1). We replace Z by another equivariant lifting of ZFp defined as follows. Let R be the ring of integers of a finite extension K of Qp, of degree > 1, with residue field k. If K/Qp is unramified, we set π = p, otherwise we choose a uniformizer π of R. Let λ ∈ R be an element satisfying the following condition: a) b) If K/Qp is unramified, then the reduction of λ mod p does not belong to Fp; If K/Qp is ramified, then λ ∈ R×. Let h ∈ Z[X1, . . . , Xp−1] be the product of the elements of the orbit of X1, i.e., (9.6.1) h(X1, . . . , Xp−1) = X1(−Xp−1) and let f ∈ R[X1, . . . , Xp−1] be defined by (9.6.2) f = g + πλh. (Xi − Xi−1), p−1Yi=2 We define Y ⊂ Pn R to be the hypersurface with equation f . Since f is invariant under G, the action of G on Pn Its special fibre Yk is equal to Zk, on which G acts freely by Lemma 9.5. Then the fixed locus of σ is a closed subscheme of Y , and its projection on Spec R is a closed subset which does not contain the closed point. Therefore it is empty, and the action of G on Y is free. We define X to be the quotient scheme X = Y /G. R induces an action on Y . Proposition 9.7. Assume that p is an odd prime which is not a Fermat number. Then the scheme X defined above satisfies conditions (1) - (6) of 9.1. Proof. As observed in 9.2 (iii), it suffices to check that X satisfies conditions (1), (3") and (4), and condition (3") holds by construction. The hypersurface Y is projective and flat over R, since g is not divisible by π. So X is also projective and flat. As Yk = Zk, Lemma 9.5 (ii) implies that Y is not semi-stable. Since Y → X is ´etale and semi-stability is a local property for the ´etale topology, X is not semi-stable either. So we only have to prove that X is regular. This is again a local property for the ´etale topology, hence it suffices to prove that Y is regular. Because Y is excellent, its singular locus is closed, and the same holds for its projection to Spec R. So it is enough to check the regularity of Y at the points of its special fibre. The regularity is clear at the smooth points of Yk, and we need to prove it at the singular points. 64 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING Let x = (ξ1 : . . . : ξp−1) ∈ Pn(k) be a singular point of Yk. As Yk = Zk, x corresponds by Lemma 9.4 to a family (u1, . . . , up−1) ∈ (F× p )p−1 such that 1 + u1 + · · · + up−1 = p2β (9.7.1) for some β ∈ Zp. We have seen in the proof of Lemma 9.4 that ξ1 ∈ F× p , so we may assume that ξ1 = 1. We set η1 = 1, and we define inductively ηi for 2 ≤ i ≤ p − 1 by (9.4.4). This allows to work on the affine space An R, and we will denote R = D+(X1) ⊂ Pn a∗(X2, . . . , Xp−1) := a(1, X2, . . . , Xp−1) for any homogeneous polynomial a(X1, . . . , Xp−1) ∈ R[X1, . . . , Xp−1]. For 2 ≤ i ≤ p − 1, we set Xi = ηi + Yi, so that (π, Y2, . . . , Yp−1) is a regular sequence of generators of the maximal ideal mx of the regular local ring OAn R,x. We want to prove that OAn R,x/(f∗) is regular, i.e., that f∗ /∈ m2 x. We first claim that (9.7.2) g∗ ≡ pβ mod m2 x. Indeed, applying (9.4.7) with α = pβ, we obtain the congruence g0 ∗(η2, . . . , ηp−1) ≡ p2β mod p3Zp, hence (9.7.3) g∗(η2, . . . , ηp−1) ≡ pβ mod p2Zp ⊂ m2 x. On the other hand, equations (9.4.4) show that, for 2 ≤ i ≤ p − 2, (9.7.4) ∂g∗ ∂Xi (η2, . . . , ηp−1) = 0. Finally, equations (9.4.4) and (9.4.6) show that ∂g∗ ∂Xp−1 (η2, . . . , ηp−1) = (ηp−1 − ηp−2)p−1 − ηp−1 p−1 (9.7.5) = 1 − (p2β − up−1)p−1 ≡ 0 mod p2Zp ⊂ m2 x. Applying (9.7.3), (9.7.4) and (9.7.5) to the Taylor development of g∗ proves (9.7.2). From the definition of h, we obtain (9.7.6) h∗(η2, . . . , ηp−1) = −(p2β − up−1) ui ≡ ui mod mx. As h∗ ≡ h∗(η2, . . . , ηp−1) mod mx, f∗ satisfies the congruence (9.7.7) f∗ = g∗ + πλh∗ ≡ π( p π β + λ ui) mod m2 x. Let w = p is a unit. If K/Qp is unramified, then π = p, and condition 9.6 a) implies that the π β + λQi ui. If K/Qp is ramified, then condition 9.6 b) implies that w p−1Yi=1 p−2Yi=1 p−1Yi=1 RATIONAL POINTS OF REGULAR MODELS 65 reduction mod p of w is non-zero, hence w is again a unit. In each case, f∗ /∈ m2 x, and OY,x is regular. (cid:3) Appendix: Complete intersection morphisms of virtual relative dimension 0 As mentioned in the introduction, we explain here the construction of the mor- phism τf : Rf∗OY → OX for a proper complete intersection morphism f : Y → X of virtual dimension 0, and we give a proof of Theorem 3.1. The Appendix consists of two sections. In section A, we recall the construction of the invertible sheaf ωY /X associated to a complete intersection morphism f : Y → X, and we prove some of its properties. We do not use duality theory here, even if we keep for convenience the terminology "relative dualizing sheaf". Instead, we use the complete intersection assumption to deduce our constructions from the elementary properties of smooth morphisms and regular immersions, thanks to the canonical isomorphisms defined by Conrad [Co00, 2.2]. It is then easy to define the canonical section δf of ωY /X when f has virtual relative dimension 0, and to prove its basic properties. In section B, we assume that X is noetherian and has a dualizing complex. We ∼ −−→ f !OX to deduce τf from the then use duality theory and the identification ωY /X canonical section δf . To translate the properties of δf into the properties of τf listed in Theorem 3.1, we need to use the fundamental identifications of duality theory, as well as the various compatibilities between these identifications. Our proofs rely in an essential way on Conrad's exposition [Co00]. It may be worth pointing out that we need in this article the compatibility of τf with base change in a context which is not covered by the base change results of [Co00]. Indeed, we consider morphisms f which are not flat in general (such as in Theorem 1.5), and base change morphisms which are not flat either (such as reduction mod pn in the proof of Proposition 8.6). The key property we use here, which is familiar to the experts, but not so well documented in the literature, is the Tor-independence of f and the base change morphism. A. The canonical section of the relative dualizing sheaf We recall now the construction of the invertible sheaf ωY /X for a complete inter- section morphism, and we explain some of its properties. As often, the main work is to prove that the constructions are well-defined, and in particular to check the sign conventions. As the details are easy but tedious, we leave most of them as exercises, and only sketch the main steps of the verifications. We first recall a standard base change result for complete intersection morphisms (see [SGA 6, 3.7.1] for the finite Tor-dimension of Rf∗E •). Proposition A.1. Let f : Y → X be a complete intersection morphism of virtual relative dimension m, and let (A.1.1) Y ′ f ′ X ′ v u Y f / X / /     / 66 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING be a cartesian square such that X ′ and Y are Tor-independent over X. (i) The morphism f ′ is a complete intersection morphism of virtual relative dimension m. (ii) Assume that X is quasi-compact, and that f is separated of finite type. qc(OY ) is of finite Tor-dimension over OY , then Rf∗E • is of finite Tor- If E • ∈ Db dimension over OX , and the base change morphism (A.1.2) Lu∗Rf∗E • −→ Rf ′ ∗Lv∗E • is an isomorphism. Proof. The first claim is local on Y ′, so we may assume that there exists a factor- ization f = π ◦ i such that π : P → X is a smooth morphism of relative dimension n, and i : Y ֒→ P is a closed immersion of codimension d = n − m. Then i is a regular immersion, defined by an ideal I ⊂ OP , and, since the claim is local, we may assume that I is generated by a regular sequence t1, . . . , td of sections of OP . Then the Koszul complex K•(t1, . . . , td) is a resolution of OY by OP -modules which are flat relatively to X. Let P ′ = X ′ ×X P , and let t′ d be the images of t1, . . . , td in OP ′. Since X ′ and Y are Tor-independent over X, the Koszul complex d) is a resolution of OY ′ over OP ′, which shows that f ′ is a complete K(t′ intersection morphism of virtual relative dimension m. 1, . . . , t′ 1, . . . , t′ Assume now that the hypotheses of (ii) are satisfied. Since X is quasi-compact, it suffices to check that Rf∗E • is of finite Tor-dimension when X is affine. We can then choose a finite covering U of Y by affine open subsets Uα, and we may assume that the Uα are small enough so that the restriction fα of f to Uα can be factorized as fα = πα ◦ iα, where πα : Pα → X is smooth and iα : Uα ֒→ Pα is a closed immersion defined by a regular sequence of sections of OPα. For each sequence α0 < · · · < αr, denote Uα = Uα0 ∩ · · · ∩ Uαr , jα : Uα ֒→ Y , and let fα be the restriction of f If I • is an injective resolution of E •, then the alternating Cech complex to Uα. C•(U, I •) is a resolution of E •. Since jα is an affine open immersion, the complex jα ∗j∗ qc,fTd(OY ) for each α. Therefore it suffices to prove that Rf∗E • ∈ Db qc,fTd(OX ) for complexes E • of the form Rj∗F •, where j is the inclusion of an affine open subscheme U , and F • ∈ Db qc,fTd(OU ). This reduces the proof to the case where Y is affine. Then there exists a bounded complex of OY -modules P • with flat quasi-coherent terms, and a quasi-isomorphism P • → E •. Since OY has finite Tor-dimension over OX, so does any flat OY -module, and the first assertion of (ii) follows. αE • belongs to Db αI • = Rjα ∗j∗ The complex Lv∗E • belongs to Db qc,fTd(OY ′), and the base change morphism (A.1.2) can be defined by adjunction as usual. Arguing as before, it suffices to prove that it is an isomorphism when X is affine and E • is of the form Rj∗F •, where j is the inclusion of an affine open subscheme U ⊂ Y , and F • ∈ Db qc,fTd(OU ). Let U ′ = X ′ ×X U , and let w : U ′ → U be the projection, j′ : U ′ ֒→ Y ′ the pull- back of j. Since j is an affine morphism and F • ∈ Db qc,fTd(OU ), the base change ∗Lw∗F • is an isomorphism. This implies that the base morphism Lv∗Rj∗F • → Rj′ change morphism for f and E • is an isomorphism if and only if the base change RATIONAL POINTS OF REGULAR MODELS 67 morphism for f ◦ j and F • is an isomorphism. If one chooses a bounded, flat, quasi- coherent resolution P • of F •, the Tor-independence assumption implies that, for each n, (f ◦ j)∗P n is u∗-acyclic. It follows easily that the base change morphism for P • is an isomorphism, which ends the proof. (cid:3) Remark. Assertion (ii) holds more generally if one replaces the complete intersection hypothesis on f by the assumption that E • has finite Tor-dimension over OX . It is also standard to extend the assertion to the case where f is only assumed to be coherent, i.e., quasi-compact and quasi-separated. A.2. Let f : Y → X be a complete intersection morphism of relative dimension m. We now recall how one can associate to f an invertible OY -module ωY /X, called the relative dualizing sheaf of Y /X (or f ). We will use here the direct construction based on elementary algebra1, which makes explicit the existence of the canonical section when m = 0, and is a natural extension of Conrad's constructions for the canonical isomorphisms ζ ′ f,g [Co00, 2.2]. If f = π ◦ i is a factorization of f where π : P → X is a smooth morphism of relative dimension n and i : Y ֒→ P a closed immersion of codimension d = n − m, defined by a regular ideal I ⊂ OP , then one defines ωY /X by setting (A.2.1) ωY /X = ωY /P ⊗OY i∗ωP/X = ∧d((I/I 2)∨) ⊗OY i∗Ωn P/X. Up to canonical isomorphism, this construction is made independent of the choice of the factorization as follows. Let f = π′ ◦ i′ be another factorization of f through a smooth morphism π′ : P ′ → X, and let ωP Y /X be the invertible OY -modules defined by (A.2.1) using the two factorizations. Assume first that there exists an X- morphism u : P ′ → P such that u ◦ i′ = i, and which is either a smooth morphism or a regular immersion. Then, one defines an isomorphism εP ′,P (u) : ωP by the commutative diagram Y /X and ωP ′ ∼ −−→ ωP ′ Y /X Y /X (A.2.2) Y /X = ωY /P ⊗ i∗ωP/X ωP ωY /P ′ ⊗ i′∗ωP ′/P ⊗ i′∗u∗ωP/X ζ ′ i′,u ⊗ Id ∼ +VVVVVVVVVVVVVVVVVVV ∼ εP ′,P (u) ∼ Id ⊗ i′∗(ζ ′ u,π) ωY /P ′ ⊗ i′∗ωP ′/X = ωP ′ Y /X. The definitions of ζ ′ u,π depend upon whether u is a smooth morphism or a regular immersion (the two definitions agree when u is an open and closed immersion): i′,u and ζ ′ a) If u is smooth, then ζ ′ i′,u is defined by [Co00, p. 29, (d)], and ζ ′ u,π is defined by [Co00, p. 29, (a)]. 1 For a more intrinsic construction, one can use the general properties of the cotangent complex LY /X [Il71]. Here, LY /X is a perfect complex, of perfect amplitude in [−1, 0], and of rank m [Il71, 3.2.6]. Taking its (graded) determinant in the sense of Knudsen-Mumford [KM76], one obtains the complex ωY /X [m]. Special attention should be paid in this construction to sign compatibilities, as, for historical reasons, the sign conventions used in [Il71] and [KM76] conflict with those of [Co00]. / / + O O 68 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING b) If u is a regular immersion, then ζ ′ i′,u is defined by [Co00, p. 29, (b)], and ζ ′ u,π is defined by [Co00, p. 29, (c)]. Let f = π′′ ◦ i′′ be a third factorization of f through a smooth morphism π′′ : P ′′ → X, let ωP ′′ Y /X be defined by (A.2.1) using this factorization, and assume that there exists an X-morphism v : P ′′ → P ′ such that v ◦ i′′ = i′ and such that each of the morphisms v and u ◦ v is either a smooth morphism or a regular immersion. Then it follows readily from Conrad's general transitivity relation for compositions of smooth morphisms and regular immersions [Co00, (2.2.4)] that (A.2.3) εP ′′,P ′ (v) ◦ εP ′,P (u) = εP ′′,P (u ◦ v). If f = π ◦ i = π′ ◦ i′ are any factorizations as above, let now P ′′ = P ′ ×X P , and let i′′ : Y ֒→ P ′′ be the diagonal immersion, and q : P ′′ → P, q′ : P ′′ → P ′ the two projections. One defines the canonical isomorphism εP ′,P : ωP Y /X by setting ∼ −→ ωP ′ Y /X (A.2.4) εP ′,P := εP ′′,P ′ (q′)−1 ◦ εP ′′,P (q). Whenever there exists a smooth morphism or a regular immersion u : P ′ → P as above, it follows from (A.2.3) that εP ′,P (u) = εP ′,P . One checks similarly that the isomorphisms εP ′,P satisfy the usual cocycle condition for three factorizations. Thanks to this cocycle condition, one can then define the invertible sheaf ωY /X even when there does not exist a global factorization f = π ◦ i as above, by choos- ing local factorizations and glueing the invertible sheaves obtained locally by the previous construction. By construction, the sheaf ωY /X commutes with Zariski lo- calization, and is equipped with a canonical isomorphism for which we keep the notation ζ ′: (A.2.5) ζ ′ π,i : ωY /X ∼ −−→ ωY /P ⊗OY i∗ωP/X, for any factorization f = π ◦ i where π is a smooth morphism and i is a regular immersion. If m is the virtual relative dimension of Y over X, we will need to work with the complex ωY /X[m] which is the single OY -module ωY /X sitting in degree −m. If f = π ◦ i as above, we define in Db(OY ) the isomorphism of complexes (A.2.6) ζ ′ i,π : ωY /X[m] ∼ −−→ ωY /P [−d] L ⊗OY Li∗(ωP/X[n]) by (A.2.5) in degree −m, without any sign modification. If f is a smooth morphism or a regular immersion, this definition is consistent with [Co00, (2.2.6)]. By [Co00, (1.3.6)], the isomorphism (A.2.6) is equal to the composed isomorphism ωY /X[m] ∼ −−−−−−→ (A.2.5)[m] (ωY /P L ⊗OY Li∗(ωP/X))[m] ∼ −−→ (ωY /P L ⊗OY Li∗(ωP/X[n]))[−d] ∼ −−→ ωY /P [−d] L ⊗OY Li∗(ωP/X[n]) RATIONAL POINTS OF REGULAR MODELS 69 and differs from the composed isomorphism ωY /X[m] ∼ −−−−−−→ (A.2.5)[m] (ωY /P L ⊗OY Li∗(ωP/X))[m] ∼ −−→ (ωY /P [−d] L ⊗OY Li∗(ωP/X))[n] ∼ −−→ ωY /P [−d] L ⊗OY Li∗(ωP/X [n]) by multiplication by (−1)dn. Lemma A.3. Under the assumptions of Proposition A.1, there exists a canonical isomorphism (A.3.1) Lv∗(ωY /X) ∼= v∗(ωY /X ) ∼ −−→ ωY ′/X ′. Moreover, if the assumptions of Proposition A.1 (ii) are satisfied, the canonical base change morphism (A.3.2) Lu∗Rf∗(ωY /X) → Rf ′ ∗(ωY ′/X ′). is an isomorphism. ∼ Proof. Since ωY /X is invertible, Lv∗(ωY /X) −−→ v∗(ωY /X). To prove the isomor- phism (A.3.1), assume first that there exists a factorization f = π ◦ i where π is smooth and i is a regular immersion. Let f ′ = π′ ◦ i′ be the factorization de- duced from f = π ◦ i by base change. Then, if I and I ′ are the ideals defining i and i′, the Tor-independence assumption implies that the canonical homomorphism u∗(I/I 2) → I ′/I ′2 is an isomorphism, which defines (A.3.1). It is easy to check that, for two factorizations of f , the corresponding isomorphisms are compatible with the identifications (A.2.4). This provides the isomorphism (A.3.1) in the general case. When the assumptions of A.1 (ii) are satisfied, the isomorphism (A.3.2) follows (cid:3) from (A.3.1) and (A.1.2). A.4. Let Y ′ f ′ X ′ be a cartesian square, and assume that: v u Y f / X a) f and u are complete intersection morphisms of relative dimensions m and n; b) X ′ and Y are Tor-independent over X. Then Lemma A.3 provides canonical isomorphisms v∗(ωY /X ) ∼ −−→ ωY ′/X ′ , f ′∗(ωX ′/X ) ∼ −−→ ωY ′/Y . One defines the canonical isomorphism (A.4.1) χf,u : ωY ′/Y ⊗OY ′ v∗(ωY /X ) ∼ −−→ ωY ′/X ′ ⊗OY ′ f ′∗(ωX ′/X ) as being the product by (−1)mn of the composite ωY ′/Y ⊗OY ′ v∗(ωY /X) ∼ −−→ f ′∗(ωX ′/X ) ⊗OY ′ ωY ′/X ′ ∼ −−→ ωY ′/X ′ ⊗OY ′ f ′∗(ωX ′/X ), / /     / 70 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING where the first isomorphism is the product of the previous base change isomorphisms, and the second one is the usual commutativity isomorphism of the tensor product (see [De83, Appendix, (a)] and [Co00, p. 215-216]). The following relations follow easily from the local description of the isomorphisms ζ ′ f,g given in [Co00, p. 30, (a) - (d)]: (i) In the above cartesian square, assume that each of the three morphisms u, f and u ◦ f ′ = f ◦ v is either a smooth morphism or a regular immersion. Then the following isomorphisms ωY ′/X −−→ ωY ′/X ′ ⊗OY ′ f ′∗(ωX ′/X ) are equal: ∼ (A.4.2) (ii) Let f ′,u = χf,u ◦ ζ ′ ζ ′ v,f . v Y ′  Y f j = i f ′ X ′′   u X ′  / X be a commutative diagram in which the square is cartesian, f is smooth, i and u are regular immersions. Then the following isomorphisms ωX ′′/X ∼ −−→ ωX ′′/Y ′ ⊗OX′′ j∗(ωY ′/X ′) ⊗OX′′ i∗(ωX ′/X ) are equal: (A.4.3) (iii) Let (ζ ′ j,f ′ ⊗ Id) ◦ ζ ′ i,u = (Id ⊗ j∗(χf,u)) ◦ (ζ ′ j,v ⊗ Id) ◦ ζ ′ vj,f . Y ′′ f ′′ X ′′ v′ u′ Y ′ f ′ / X ′ v u Y f / X be a commutative diagram in which both squares are cartesian, each of the mor- phisms f , u, u′ and u ◦ u′ is either a smooth morphism or a regular immersion, X ′ and Y are Tor-independent over X, and X ′′ and Y ′ are Tor-independent over X ′ (so that X ′′ and Y are Tor-independent over X, and all immersions are regular). Then the following isomorphisms ωY ′′/Y ⊗OY ′′ (vv′)∗(ωY /X) ∼ −−→ ωY ′′/X ′′ ⊗OY ′′ f ′′∗(ωX ′′/X ′ ⊗OX′′ u′∗(ωX ′/X )) are equal: (A.4.4) (Id ⊗ f ′′∗(ζ ′ u′,u)) ◦ χf,uu′ = (χf ′,u′ ⊗ Id) ◦ (Id ⊗ v′∗(χf,u)) ◦ (ζ ′ v′,v ⊗ Id). We will also need to extend the isomorphism χf,u to the derived category. We define (A.4.5) χf,u : ωY ′/Y [n] L ⊗OY ′ Lv∗(ωY /X[m]) ∼ −−→ ωY ′/X ′ [m] L ⊗OY ′ Lf ′∗(ωX ′/X[n])  / /      / / .  = / / /   / /     / / RATIONAL POINTS OF REGULAR MODELS 71 by (A.4.1) in degree −(m+n), without any further sign modification. Because of the sign convention in the commutativity isomorphism for the derived tensor product [Co00, p. 11], χf,u can also be described as the composite ωY ′/Y [n] L ⊗OY ′ Lv∗(ωY /X [m]) ∼ −−→ Lf ′∗(ωX ′/X [n]) ∼ −−→ ωY ′/X ′[m] L ⊗OY ′ ωY ′/X ′[m] ⊗OY ′ Lf ′∗(ωX ′/X[n]), L where the first isomorphism is the tensor product of the base change isomorphisms, and the second one is the commutativity isomorphism. With this definition, the previous relations (A.4.2) to (A.4.4) remain valid in Db(OY ′). A.5. We now consider the composition of two complete intersection morphisms f : Y → X, g : Z → Y , and define a canonical isomorphism −−→ ωZ/Y ⊗OZ g∗(ωY /X ) ζ ′ g,f : ωZ/X (A.5.1) ∼ extending the isomorphism (A.2.5). Assume first that there exists a factorization f = π ◦ i, where π : P → X is a smooth morphism, and a factorization i ◦ g = π′′ ◦ j, where π′′ : P ′′ → P is a smooth morphism (such factorizations always exist when X, Y and Z are affine). Let π′ : P ′ → Y be the pull-back of π′′, so that we get a commutative diagram (A.5.2) j i′ >~~~~~~~~ P ′ g Z P ′′ (cid:3) i′′ > !BBBBBBBB π′ / Y π′′ AAAAAAAA >}}}}}}}} i P f ψ π @@@@@@@@ / X where the middle square is cartesian. Using (A.2.5) for (j, ψ), and the isomorphisms i′,i′′ ⊗ j∗(ζ ′ ζ ′ π′′,π), we obtain an isomorphism ωZ/X ∼= ωZ/P ′′ ⊗ j∗(ωP ′′/X ) ∼ −−→ (ωZ/P ′ ⊗ i′∗(ωP ′/P ′′)) ⊗ j∗(ωP ′′/P ⊗ π′′∗(ωP/X )) ∼ −−→ ωZ/P ′ ⊗ i′∗(ωP ′/P ′′ ⊗ i′′∗(ωP ′′/P )) ⊗ g∗i∗(ωP/X). Using the isomorphism χπ′′,i : ωP ′/P ′′ ⊗ i′′∗(ωP ′′/P ) ∼ −−→ ωP ′/Y ⊗ π′∗(ωY /P ) defined in A.4, and (ζ ′ i′,π′ ⊗ g∗(ζ ′ i,π))−1, we then obtain the composed isomorphism ωZ/X ∼ −−→ ωZ/P ′ ⊗ i′∗(ωP ′/Y ⊗ π′∗(ωY /P )) ⊗ g∗i∗(ωP/X) ∼ −−→ (ωZ/P ′ ⊗ i′∗(ωP ′/Y )) ⊗ g∗(ωY /P ⊗ i∗(ωP/X)) ∼ −−→ ωZ/Y ⊗ g∗(ωY /X), which defines (A.5.1). To prove that this isomorphism is well defined, and to glue the local constructions to obtain a global one when a diagram (A.5.2) does not exist globally, we must check that it does not depend on the chosen factorizations. If we have two diagrams   .  > ! ;  0 0 /  > / .  > / 72 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING (A.5.2), with factorizations f = πk ◦ ik, ik ◦ g = π′′ k ◦ jk, for k = 1, 2, we can embed Y diagonally in P1 ×X P2, and Z in P ′′ 1 ×X P ′′ 2 . This allows to reduce the verification to the case where there exists a smooth X-morphism u : P2 → P1 such that u ◦ i2 = i1, and a smooth morphism u′′ : P ′′ 1 ◦ u′′ = u ◦ π′′ 2 , and j1 = u′′ ◦ j2. Morever, the same argument shows that we may assume that the 2 → P ′′ morphism P ′′ 1 ×P1 P2 is smooth. The verification can then be reduced to the following two cases: 1 such that π′′ 2 → P ′′ (i) The morphism P ′′ (ii) The morphism P2 → P1 is an isomorphism. 1 ×P1 P2 is an isomorphism; 2 → P ′′ In each of these cases, the equality of the two definitions of (A.5.1) breaks down to a succession of elementary commutative diagrams involving isomorphisms of the form ζ ′ f,g and χf,u. We omit details here, and only point out that, in addition to [Co00, (2.2.4)], the first case uses relation (A.4.2), and the second one uses relation (A.4.3). In particular, the sign convention introduced in the definition of χf,u in A.4 is necessary for this independence result. If m and m′ are the virtual relative dimensions of f and g, we define as in A.2 the derived category variant of (A.5.1) as being the morphism (A.5.3) ζ ′ g,f : ωZ/X[m + m′] ∼ −−→ ωZ/Y [m′] ⊗L OZ Lf ∗(ωY /X [m]) defined by applying (A.5.1) to the underlying modules (sitting in degree −m − m′), without any sign modification. With the definition of ζ ′ g,f provided by (A.5.1) (resp. (A.5.3)), we now extend to complete intersection morphisms Conrad's transitivity relation [Co00, (2.2.4)]. Proposition A.6. Let h−−→ Z g −−→ Y T f −−→ X be three complete intersection morphisms. Then h,f g = (ζ ′ (Id ⊗ h∗(ζ ′ g,f )) ◦ ζ ′ (A.6.1) h,g ⊗ Id) ◦ ζ ′ gh,f . Proof. As the verification is local on T , we may assume that there exists a commu- tative diagram Q′′ Q′ k′′ ϕ′′ > BBBBBBBB ={{{{{{{{ i′ R′′ P ′ g j′ ψ′′ AAAAAAAA >}}}}}}}} !BBBBBBBB π′ / Z / Y j′′ > !CCCCCCCC π′′ i′′ >}}}}}}}} P ′′ h T ϕ′ ???????? >}}}}}}}} i P f π @@@@@@@@ / X in which the three squares are cartesian, the morphisms π, ϕ′, ψ′′ are smooth, and the immersions i, i′, i′′ are regular. Using [Co00, (2.2.4)] and the relation (A.4.4), the .  >  .  > ! .  > ! .  > / .  = / .  > / RATIONAL POINTS OF REGULAR MODELS 73 proof of (A.6.1) again breaks into a succession of elementary commutative diagrams, which we do not detail here. (cid:3) A.7. We now assume that f : Y → X is a complete intersection morphism of (virtual) relative dimension 0, and, under this hypothesis, we define a section δf ∈ Γ(Y, ωY /X), which we call the canonical section. We first assume that there is a factorization f = π ◦ i such that π : P → X is a smooth morphism of relative dimension n, and i : Y ֒→ P is a regular closed immersion, necessarily of codimension n since f has relative dimension 0. Let I ⊂ OP be the ideal defining i. The canonical derivation d : OP → Ω1 P/X induces an OY -linear homomorphism ¯d : I/I 2 → i∗Ω1 P/X. Taking its n-th exterior power, we obtain a linear homomorphism (A.7.1) ∧n ¯d : ∧n(I/I 2) −→ i∗Ωn P/X. Through the canonical isomorphisms Hom OY (∧n(I/I 2), i∗Ωn P/X) ∼= (∧n(I/I 2))∨ ⊗OY i∗Ωn ∼= ∧n((I/I 2)∨) ⊗OY i∗Ωn = ωY /X, P/X P/X it can be seen as a section of ωY /X, which is the section δf . If (t1, . . . , tn) is a regular sequence of generators of I on a neighbourhood U of some point y ∈ Y , then δf = (¯t ∨ 1 ∧ . . . ∧ ¯t ∨ n ) ⊗ i∗(dtn ∧ . . . ∧ dt1) ∈ Γ(U, ωY /X), (A.7.2) since the canonical isomorphism (∧n(I/I 2))∨ ∼= ∧n((I/I 2)∨) maps (¯tn ∧ . . . ∧ ¯t1)∨ to ¯t ∨ 1 ∧ . . . ∧ ¯t ∨ n . To end the construction of δf , it suffices to check that the section obtained in this way does not depend on the chosen factorization. Using the diagonal embedding, it suffices as usual to compare the sections δf and δ′ f defined by two factorizations f = π ◦ i = π′ ◦ i′ when there exists a smooth X-morphism u : P ′ → P such that u ◦ i′ = i. Let I ′ be the ideal of Y in P ′, and Y /X = ∧n′ ω′ ((I ′/I ′2)∨) ⊗OY i′∗Ωn′ P ′/X, where n′ is the codimension of Y in P ′. Then the canonical identification ωY /X Y /X is given by (A.2.2), case a), and, thanks to (A.7.2), the equality δf = δ′ ω′ from [Co00, p. 30, (a) and (d)]. ∼= f follows Proposition A.8. Let f : Y → X be a complete intersection morphism of virtual relative dimension 0. (i) Let g : Z → Y be a second complete intersection morphism of virtual g,f defined in (A.5.1) relative dimension 0. The image of δf g under the isomorphism ζ ′ is given by (A.8.1) g,f (δf g) = δg ⊗ g∗(δf ). ζ ′ (ii) For any cartesian square (A.1.1), the isomorphism (A.3.1) v∗(ωY /X ) ∼ −−→ ωY ′/X ′ 74 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING maps v∗(δf ) to δf ′. Proof. As the first claim is local on Z, we may assume that there exists a diagram (A.5.2) in which the immersion i is defined by a regular sequence (t1, . . . , tn), and the i = π′′∗(ti). immersion j = i′′ ◦ i′ by a regular sequence (t′ If we set ¯t′ n′). By construction, δf g corresponds by ζ ′ i), then i′ is defined by the regular sequence (¯t′ n), with t′′ i = i′′∗(t′ 1, . . . , t′′ 1, . . . , ¯t′ 1, . . . , t′ j,ψ to the section n′, t′′ (t′′ n ∨ ∧ . . . ∧ t′′ 1 ∨ ∧ t′ n′ ∨ ∧ . . . ∧ t′ 1 ∨) ⊗ j∗(dt′ 1 ∧ . . . ∧ dt′ n′ ∧ dt′′ 1 ∧ . . . ∧ dt′′ n) ∨∧. . .∧t′′ 1 (¯t′ n′ ∨∧. . .∧¯t′ 1 ∨)⊗i′∗(t′′ n i′,i′′ ⊗ j∗(ζ ′ π′′,π) to the section of ωZ/P ′′ ⊗ j∗(ωP ′′/X ), which is mapped by ζ ′ ((−1)nn′ n′)⊗π′′∗(dt1∧. . .∧dtn)) of (ωZ/P ′ ⊗ i′∗(ωP ′/P ′′)) ⊗ j∗(ωP ′′/P ⊗ π′′∗(ωP/X)). We then get via χπ′′,i the section ∨) ⊗ j∗π′′∗(dt1 ∧ . . . ∧ dtn)), (¯t′ n′ of ωZ/P ′ ⊗ i′∗(ωP ′/Y ) ⊗ i′∗π′∗(ωY /P ) ⊗ j∗π′′∗(ωP/X ), which, by construction, corre- sponds by (ζ ′ i,π))−1 to the section δg ⊗ g∗(δf ) of ωZ/Y ⊗ g∗(ωY /X). n′) ⊗ i′∗π′∗(tn 1 ∧ . . . ∧ d¯t′ ∨ ∧ . . . ∧ ¯t′ 1 ∨))⊗j∗((dt′ 1 ∧. . .∧dt′ i′,π′ ⊗ g∗(ζ ′ ∨ ∧ . . . ∧ t1 ∨) ⊗ i′∗(d¯t′ The second claim follows from (A.7.2). (cid:3) B. The trace morphism τf on Rf∗(OY ) Let f : Y → X be a proper complete intersection morphism of virtual relative dimension 0. This section is devoted to the construction of the "trace morphism" τf : Rf∗OY → OX , derived from the canonical section of ωY /X defined in A.7. The key step is to define an identification λf between ωY /X as defined in A.2, and f !OX. The construction is then a straightforward application of the relative duality theorem, and the properties of τf listed in Theorem 3.1 follow from corresponding properties of δf and λf . B.1. For the whole section, we assume that X is a noetherian scheme with a dualiz- ing complex. If f : Y → X is a morphism of finite type, and K • a residual complex on X, let f ∆K • be its inverse image on Y in the sense of residual complexes, which is a residual complex on Y . Then K • and f ∆K • define respectively duality δ-functors DX on Dcoh(OX ) and DY on Dcoh(OY ). We recall that, following [Co00, 3.3], the functor f ! : D+ coh(OY ) is defined by f ! = DY ◦ Lf ∗ ◦ DX . We also recall that, when f is smooth of relative dimension m, f ♯ : D+ coh(OY ) denotes the functor defined by coh(OX ) → D+ coh(OX ) → D+ (B.1.1) f ♯(E •) := ωY /X[m] coh(OX ) → D+ L ⊗OY Lf ∗(E •), while, when f is finite, f ♭ : D+ coh(OY ) denotes the functor defined by (B.1.2) f ♭(E •) := f ∗ RHom OX (f∗OY , E •), where f is the (flat) morphism of ringed spaces (Y, OY ) → (X, f∗OY ). Assume now that f : Y → X is a complete intersection morphism of virtual relative dimension m. We first explain the relation between the relative dualizing module ωY /X defined in the previous section, and the extraordinary inverse image functor f !. We will consider complexes of the form E • = L[r] ∈ Db coh(OX ), where RATIONAL POINTS OF REGULAR MODELS 75 r ∈ Z is some integer, and L is an invertible OX-module. For such a complex, we generalize the above notation f ♯, and define again (B.1.3) f ♯(E •) := ωY /X[m] L ⊗OY Lf ∗(E •). We observe that f ♯(E •) is another complex concentrated in a single degree, with an invertible cohomology sheaf. We can then construct a canonical isomorphism (B.1.4) as follows. λf,E • : f ♯(E •) ∼ −−→ f !(E •) a) If f is smooth, then definitions (B.1.1) and (B.1.3) coincide, and we set (B.1.5) λf,E • = ef : f ♯(E •) ∼ −−→ f !(E •), where ef is the isomorphism defined by [Co00, (3.3.21)]. b) If f is a regular immersion, then we define λf,E • to be the composition (B.1.6) λf,E • : f ♯(E •) η−1 f−−→ ∼ f ♭(E •) df−→ ∼ f !(E •), where ηf is defined by [Co00, (2.5.3)] and df by [Co00, (3.3.19)]. c) In the general case, let us assume first that there exists a factorization f = π◦i, where π : P → X is a smooth morphism of relative dimension n, and i is a regular immersion of codimension d = n − m. Then we define λf,E • by the commutative diagram (B.1.7) ωY /X[m] L ⊗OY Lf ∗E • ζ ′ i,π⊗Id ∼ λf,E· ∼ / f !E • ∼ ci,π ωY /P [−d] L ⊗OY Li∗π♯E • λi,π♯E· ∼ / i!π♯E • i!(λπ,E·) ∼ / i!π!E • where ci,π is the transitivity isomorphism [Co00, (3.3.14)]. This isomorphism is actually independent of the chosen factorization. To check it, one can argue as in A.2 to reduce the comparison between the isomorphisms (B.1.4) defined by two factorizations f = π ◦ i = π′ ◦ i′ to the case where there is a smooth X-morphism u : P ′ → P such that u ◦ i′ = i. It is then a long but straightforward verification, using various functorialities, the compatibility between i′,u and the isomorphism i♭ ≃ i′♭u♯ [Co00, (2.7.4)], the compatibility between ζ ′ ζ ′ u,π and the isomorphism π′♯ ≃ u♯π♯ [Co00, (2.2.7)], and the properties (VAR1), (VAR3) and (VAR5) of the functor f ! (see [Ha66, III, Th. 8.7] and [Co00, p. 139]). Since f !E • is acyclic outside degree −m − r, a morphism ωY /X[m] ⊗L Lf ∗E • → f !E • in D(OY ) is simply a module homomorphism ωY /X ⊗ f ∗L → H−m−r(f !E •). Therefore, the previous construction provides in the general case local isomorphisms which can be glued to define a global isomorphism even if there does not exist a global factorization f = π ◦ i as above. When E • = OX [0], the isomorphism (B.1.4) will simply be denoted (B.1.8) λf : ωY /X[m] ∼ −−→ f !OX .   /   / / 76 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING If f is flat, hence is a CM map, it provides the identification between the construction of ωY /X used in this article, and the construction of Conrad for CM maps [Co00, 3.5, p. 157]. We now give for the isomorphisms λf,E • a transitivity property which generalizes (B.1.7). Proposition B.2. Let f : Y → X, g : Z → Y be two complete intersection mor- phisms, with virtual relative dimensions m, m′, and let E • = L[r], for some invertible OX-module L and some integer r. Then the diagram (B.2.1) ωZ/X[m + m′] ζ ′ g,f ⊗Id L ⊗OZ L(f g)∗E • ∼ λf g,E· ∼ ωZ/Y [m′] L ⊗OZ Lg∗f ♯E • λg,f ♯E· ∼ / g!f ♯E • g!(λf,E·) ∼ / (f g)!E • ∼ cg,f / g!f !E • ωZ/Y [m′] L ⊗OZ Lg∗f ♯E • g♯(λf,E·) ∼ / ωZ/Y [m′] L ⊗OZ Lg∗f !E • λg,f !E· ∼ / g!f !E • commutes. Proof. The commutativity of the lower part of the diagram is due to the functo- riality of the isomorphism λg with respect to morphisms between two complexes concentrated in the same degree. We first observe that the commutativity of (B.2.1) is clear in the following cases: If f is smooth and g is a closed immersion, the diagram is (B.1.7), which a) commutes by construction. b) If f and g are smooth, the isomorphism (f g)♯ ∼= g♯f ♯ is defined by ζ ′ g,f , hence the commutativity of (B.2.1) is the compatibility of the isomorphisms ef with composition, i.e., property (VAR3) of the functor f ! [Co00, p. 139]. c) If f and g are regular immersions, then isomorphisms such as ηf g commute with ζ ′ g,f and cg,f [Co00, Th. 2.5.1], and the commutativity of (B.2.1) follows from the compatibility of the isomorphisms df with composition, i.e., property (VAR2) of the functor f ! [Co00, p. 139]. We will also use the following remark. Let h : T → Z be a third complete intersection morphism, yielding the four couples of composable complete intersection morphisms (h, g), (g, f ), (gh, f ) and (h, f g). Then, if the diagrams (B.2.1) for the couples (h, g) and (g, f ) are commutative, the commutativity of (B.2.1) for (gh, f ) is equivalent to the commutativity of (B.2.1) for (h, f g): this is a consequence of (A.6.1) and of the compatibility of the isomorphisms cg,f with triple composites (i.e., property (VAR1) of the functor f ! [Co00, p. 139]). In the general case, the complexes entering in (B.2.1) are concentrated in the same degree, hence its commutativity can be checked locally. So we may assume that there exists a diagram (A.5.2). Thanks to the three particular cases listed above, one can then deduce the commutativity of (B.2.1) for (f, g) from the commutativity of (B.2.1) for (π′, i), by applying the previous remark successively to the triples (i′, i′′, π′′), (i′′i′, π′′, π), (i′, π′, i) and (g, i, π).   /   / / / / RATIONAL POINTS OF REGULAR MODELS 77 To prove the commutativity of (B.2.1) for (π′, i), we use the factorization i ◦ π′ = π′′ ◦ i′′ to define λiπ′,E . Let d be the codimension of Y in P , and n′ the relative dimension of P ′′ over P . Then, if E is a complex on P as in B.1, (B.2.1) for (π′, i) is made of the exterior composites in the diagram (B.2.2) ζ ′ π′,i ⊗Id ωP ′/P [n′ − d] wnnnnnnnn L ⊗ Li∗E •) ∼ χπ′′,i⊗Id ∼ / L ⊗ L(iπ′)∗E • ζ ′ i′′,π′′ ⊗Id (RRRRRRRRR ∼ ωP ′/P ′′[−d] ωP ′/Y [n′] L ⊗ π′∗(ωY /P [−d] ∼ π′♯(η−1 i ) π′♯(λi,E·) π′♯i♭E • ∼ L ⊗ Li′′∗(ωP ′′/P [n′] L ⊗ π′′∗E •) η−1 i′′ ∼ i′′♭π′′♯E • λi′′,π′′♯E· di′′ ∼ i′′!π′′♯E • ∼ π′♯(di) π′♯i!E • ∼ eπ′ =λπ′,i!E· π′!i!E • ∼ cπ′,i i′′!(λπ′′ ,E·)=i′′!(eπ′′ ) ∼ (iπ′)!E • = (π′′i′′)!E • ∼ ci′′,π′′ / i′′!π′′!E •. Here, the middle horizontal arrow is the standard isomorphism [Co00, Lemma 2.7.3], and the lower rectangle commutes thanks to property (VAR4) of the functor f ! [Co00, Theorem 3.3.1]. The upper triangle commutes thanks to (A.4.2). To check the comutativity of the middle rectangle, one observes on the one hand that ηi commutes with the flat base change π′′ and that ηi′′ commutes with tensorisation by the invertible sheaf ωP ′′/P (see [Co00, last paragraph of p. 54]). On the other hand, ηi′′ commutes also with the translation by n′, provided that the convention [Co00, (1.3.6)] is used for the commutation of the tensor product with translations applied to the second argument (see the discussion on [Co00, p. 53]). This requires here multiplication by (−1)dn′ on ωP ′/P ′′ ⊗ i′′∗ωP ′′/P , since ωP ′/P ′′ sits in degree d. As this is the sign entering in the definition of χπ′′,i, this ends the proof. (cid:3) B.3. Assume now that f is proper. As in B.1, let E • = L[r] ∈ Db coh(OX ), L being an invertible OX -module and r an integer. Using (B.1.4), we can define the trace morphism Tr ♯ Lf ∗E •) as the composite f,E • on Rf∗f ♯E • = Rf∗(ωY /X[m] ⊗L OY Rf∗f !E • Trf−−→ E •, Rf∗(λf,E·) −−−−−−→ f,E • : Rf∗f ♯E • Tr ♯ ∼ (B.3.1) where Trf denotes the classical trace morphism defined in [Ha66, VII, Cor. 3.4] and [Co00, 3.4]. When E • = OX [0], we will use the shorter notation (B.3.2) Tr ♯ f : Rf∗(ωY /X[m]) → OX . We first give some basic properties of the morphism Tr ♯ f . Lemma B.4. With the previous hypotheses, let (B.4.1) µf,E • : Rf∗(ωY /X[m]) L ⊗OX E • ∼ −−→ Rf∗(ωY /X[m] L ⊗OY Lf ∗E •) = Rf∗f ♯E • w ( /   ! !   } } / /         o o / 78 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING be the isomorphism given by the projection formula [Ha66, II, Prop. 5.6]. Then the diagram (B.4.2) Rf∗(ωY /X [m]) L ⊗OX E • ∼ (QQQQQQQQQQQQQQ Tr ♯ f ⊗Id µf,E· ∼ E • Rf∗f ♯E • ∼ zvvvvvvvvv Tr ♯ f,E· commutes. Proof. When f is flat, it suffices to invoke [Co00, Th. 4.4.1]. Since we make no such assumption on f , we give a direct argument, which is made a lot simpler by the very special nature of the complex E •. Using the fact that E • = L[r], with L invertible, one easily sees that there is a canonical isomorphism which commutes with translations acting on E • (B.4.3) f !OX L ⊗OY Lf ∗E • ∼ −−→ f !E •. On the other hand, we have by definition a canonical isomorphism (B.4.4) f ♯OX L ⊗OY Lf ∗E • ∼ −−→ f ♯E •, which also commutes with translations. A first observation is that the diagram (B.4.5) f ♯OX L ⊗OY Lf ∗E • (B.4.4) ∼ / f ♯E • λf ⊗Id ∼ ∼ λf,E· f !OX L ⊗OY Lf ∗E • (B.4.3) ∼ / f !E • commutes. Indeed, all complexes are concentrated in the same degree m − r, hence the verification can be done locally. This allows to assume that L = OX , which reduces the verification to the commutation of the vertical arrows with translations acting on E •. This now follows from the fact that the isomorphisms ef , ηf and df used in the construction of λf commute with translations. Applying Rf∗ to this diagram, and using the functoriality of the projection formula isomorphism, the proof is reduced to proving the commutativity of the diagram (B.4.6) Rf∗f !OX L ⊗OX E • νf ∼ / / Rf∗(f !OX ⊗OY Lf ∗E •) L (B.4.3) ∼ / / Rf∗f !E • *UUUUUUUUUUUUUUUUUUU Trf ⊗Id ∼ ∼ ukkkkkkkkkkkkkkkkk Trf E • , where νf is the projection formula isomorphism. As all morphisms of the diagram commute with translations, we may assume that r = 0. We recall that Trf is defined as the morphism of functors defined by the composite Rf∗f !(·) ∼ −−→ RHom OX (DX (·), f∗f ∆K) Trf,K−−−→ RHom OX (DX (·), K) ∼ ←−− Id, where the first isomorphism follows from the definition of f ! and the adjunction between Lf ∗ and Rf∗, the second morphism is defined by the trace morphism for / / ( z   /   / * u RATIONAL POINTS OF REGULAR MODELS 79 residual complexes Trf,K and the last isomorphism is the local biduality isomor- phism (see [Co00, p. 146]). Each of these morphisms has a natural compatibility with respect to the tensor product of the argument by an invertible sheaf. Putting together these compatibilities yields the commutativity of (B.4.6). (cid:3) Proposition B.5. Let g : Z → Y be a second proper complete intersection mor- phism, with virtual relative dimension m′. Then the diagram (B.5.1) Rf∗Rg∗(ωZ/X[m′ + m]) Rf∗Rg∗(ζ ′ g,f ) ∼ / Rf∗Rg∗(ωZ/Y [m′] L ⊗OZ Lg∗(ωY /X[m])) Tr ♯ f g OX ∼ Rf∗(Rg∗(ωZ/Y [m′]) L ⊗OY ωY /X[m]) Tr ♯ f Rf∗(Tr ♯ g ⊗Id) Rf∗(ωY /X[m]) (where the second isomorphism is given by the projection formula) is commutative. Proof. It follows from Lemma B.4 that the right vertical arrow is equal to the mor- phism Rf∗Rg∗(ωZ/Y [m′] ⊗ g∗(ωY /X[m])) Tr ♯ −−−−−−−→ Rf∗(ωY /X[m]). g,ωY /X [m] Then, using adjunction between Rf∗ and f !, and adjunction between Rg∗ and g!, one sees that the commutativity of (B.5.1) is equivalent to the commutativity of (B.2.1). (cid:3) Proposition B.6. With the hypotheses of Proposition A.1, assume in addition that X and X ′ are noetherian schemes with dualizing complexes, and that one of the following conditions is satisfied: a) f is projective; b) f is proper and u is residually stable [Co00, p. 132]; c) f is proper and flat. Then the triangle (B.6.1) is commutative. Lu∗Rf∗(ωY /X[m]) (A.3.2) ∼ OX ′ Lu∗(Tr ♯ f ) 'OOOOOOOOOOOO 7oooooooooooo Tr ♯ f ′ Rf ′ ∗(ωY ′/X ′[m]) Proof of Case a). We can choose a factorization f = π ◦ i, where π : P → X is the structural morphism of some projective space P = Pn X over X, and i is a regular immersion of codimension d = n − m. Let f ′ = π′ ◦ i′ be the factorisation of f ′   /     o o   ' 7 80 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING defined by base change, with π′ : P ′ = Pn projection. X ′ → X ′, and let w : P ′ → P be the i,π and ζ ′ The isomorphisms ζ ′ i′,π′ are clearly compatible with the base change isomorphisms (A.3.1) relative to f and u, and the same holds for the projection formula isomorphisms µi,ωP /X [n] and µi′,ωP ′/X′ [n], and the base change isomorphisms (A.3.1) relative to i and w. Then, using Proposition B.5, one sees that it suffices to prove the proposition for f = i and for f = π. When f = π : Pn X → X, let X0, . . . , Xn be the canonical coordinates on Pn X, and xi = Xi/X0, 1 ≤ i ≤ n. X defined by X0, . . . , Xn, the corresponding alternating Cech resolution provides a canonical isomorphism If U is the relatively affine covering of Pn f∗( C•(U, ωP/X)[n]) ∼ −−→ Rf∗(ωP/X[n]). (B.6.2) Recall that eπ : π♯ ∼= π! identifies the trace morphism for projective spaces Trpπ with the general trace morphism Trπ [Co00, Lemma 3.4.3, (TRA3)]. Then the commutativity of (B.6.1) for π follows from the fact that, thanks to (6.6.2), Trpπ can be characterized as the only morphism which, via (B.6.2), induces on H0 the map sending the cohomology class dx1 ∧ . . . ∧ dxn/x1 · · · xn to (−1)n(n−1)/2. When f = i : Y ֒→ P , recall that di : i♭ ∼= i! identifies the trace morphism for finite morphisms Trfi with the general trace morphism Tri [Co00, Lemma 3.4.3, (TRA2)], and that Trf i : RHom OP (OY , OP ) → OP is the canonical morphism induced by OP ։ OY . Using local cohomology with supports in Y , it can be factorized as (B.6.3) Trfi : RHom OP (OY , OP ) → RΓY (OP ) → OP . On the other hand, there exists a canonical morphism (B.6.4) Lw∗RΓY (OP ) −→ RΓY ′(OP ′), which is an isomorphism: to check this, it suffices to choose a finite affine covering V of V = P \ Y , and to identify RΓY (OP ) with its flat resolution provided by the total complex OP → j∗ C(V, OV ), where j denotes the inclusion of V in P and OP sits in degree 0. Moreover, this shows that the diagram Lw∗RΓY (OP ) / Lw∗(OP ) ∼ RΓY ′(OP ′) ∼ / OP ′ commutes. Therefore, it suffices to prove the commutativity of the diagram (B.6.5) Lw∗i∗(ωY /P [−d]) / Lw∗RΓY (OP ) ∼ ∼ ωY ′/P ′[−d] / RΓY ′(OP ′) .   /   /   /   / RATIONAL POINTS OF REGULAR MODELS 81 Since Y ′ ֒→ P ′ is a regular immersion of codimension d, all complexes in this diagram are acyclic except in degree d, so that, up to translation by −d, the diagram is actually a diagram of morphisms of OP ′-modules. It follows that its commutativity can be checked locally on P ′. Thus we may assume that P is affine, and that the ideal I of Y in P is generated by a regular sequence t1, . . . , td. Then the ideal I ′ of Y ′ in P ′ is generated by the images t′ d of t1, . . . , td in OP ′, which form a regular sequence. Let V = (V1, . . . , Vd) be the open covering of P \ Y defined by the sequence (t1, . . . , td). For any section a ∈ Γ(P, OP ), let us still denote by a/t1 · · · td the image of a/t1 · · · td ∈ Γ(V1 ∩ . . . ∩ Vd, OP ) under the canonical homomorphisms 1, . . . , t′ Γ(V1 ∩ . . . ∩ Vd, OP ) → H d−1(P \ Y, OP ) → H d Y (P, OP ) = Γ(P, Hd Y (OP )). Then the canonical morphism ωY /P ∼ −−→ Ext d OP (OY , OP ) → Hd Y (OP ) maps (¯t∨ (see [Co00, p. 252-254]). The commutativity of (B.6.5) follows. d ) ⊗ a to ε(d)a/t1 · · · td, where ε(d) ∈ {±1} only depends upon d (cid:3) 1 ∧ . . . ∧ ¯t∨ Proof of Case b). When u is residually stable, the diagram analogous to (B.6.1) commutes, thanks to [Co00, 3.4.3, (TRA4)]. Moreover, the isomorphisms eπ and di entering in the local definition of λf in B.1.4 c) also commute with base change by u, thanks to [Co00, p. 139, (VAR6)]. Then it suffice to observe that ηi commutes with flat base change, which is clear. (cid:3) Proof of Case c). When f is flat, f is a CM map, and the results of [Co00, 3.5 - 3.6] can be applied. Then the commutativity of (B.6.1) follows from [Co00, Theorem 3.6.5], provided one checks that λf identifies the base change isomorphism (A.3.1) for ωY /X with the more subtle base change isomorphism βf,u for ωf defined in [Co00, Theorem 3.6.1]. As we will not use Case c) in this article, we leave the details to the reader. (cid:3) B.7. Let X be a noetherian scheme with a dualizing complex, and f : Y → X a proper complete intersection morphism of virtual relative dimension 0. One can define in Db coh(X) a "trace morphism" (B.7.1) τf : Rf∗(OY ) −→ OX as follows. Thanks to the relative duality theorem (see [Ha66, VII, 3.4] or [Co00, Th. 3.4.4]), defining τf is equivalent to defining a morphism OY → f !OX . Using the isomorphism λf , this is also equivalent to defining a morphism (B.7.2) ϕf : OY −→ ωY /X, i.e., a section of the invertible sheaf ωY /X. We define ϕf as being the morphism which maps 1 to the canonical section δf of ωY /X, defined in A.7. From this construction, it follows that the morphism τf can be described equiva- lently either as the composition (B.7.3) τf : Rf∗(OY ) Rf∗(λf ◦ϕf ) −−−−−−−→ Rf∗(f !OX ) Trf−−→ OX , 82 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING or as the composition (B.7.4) τf : Rf∗(OY ) Rf∗(ϕf ) −−−−−→ Rf∗(ωY /X) Tr ♯ f−−→ OX , where Tr ♯ f is the trace map defined in (B.3.1). Before proving Theorem 3.1, we relate τf to the residue symbol defined in [Co00, (A.1.4)] (which differs by a sign from Hartshorne's definition in [Ha66]). Proposition B.8. With the hypotheses of B.7, assume in addition that f is finite and flat, and that f = π ◦ i, where π is smooth of relative dimension d, and i is a closed immersion, globally defined by a regular sequence (t1, . . . , td) of sections of OP . Then, for any section a of OP , with reduction ¯a on Y , we have (B.8.1) (B.8.2) t1, . . . , td τf (¯a) = ResP/X(cid:20) a dt1 ∧ . . . ∧ dtd ResP/X(cid:20) (cid:21) . t1, . . . , td (cid:21) = (−1)d(d−1)/2 ϕω(1), ω Proof. Let ω = a dt1 ∧ . . . . . . ∧ dtd. By construction, the residue symbol is given by 1 ∧ . . . ∧ t∨ where ϕω : f∗OY → OX is the image of (t∨ of complexes concentrated in degree 0 [Co00, (A.1.3)] d ) ⊗ i∗(ω) by the isomorphism (B.8.3) ωY /P [−d] L ⊗OY Li∗(ωP/X[d]) η−1 i−−→ ∼ i♭π♯OX ψ−1 i,π−−→ ∼ f ♭OX ; here f ♭OX = Hom OX (f∗OY , OX ) viewed as a OY -module, and ψi,π is the canonical the morphism isomorphism of f∗Hom OX (f∗OY , OX ) → OX given by evaluation at 1, we can use the isomorphism df : f ♭ ∼ −−→ f ! and the equality Trf ◦ f∗(df ) = Trff [Co00, 3.4.3, (TRA2)] to write (B.8.4) ∼ −−→ i♭π♯. functors f ♭ Since Trf f is ResP/X(cid:20) ω t1, . . . , td (cid:21) = (−1)d(d−1)/2 Trf (f∗(df ◦ ψ−1 i,π ◦ η−1 i )(t∨ 1 ∧ . . . ∧ t∨ d ⊗ i∗(ω))). On the other hand, we have by definition i,π(δf ) = (−1)d(d−1)/2 t∨ ζ ′ 1 ∧ . . . ∧ t∨ d ⊗ i∗(dt1 ∧ . . . . . . ∧ dtd), so we deduce from (B.7.3) the equality τf (¯a) = Trf (f∗(λf ◦ ϕf )(¯a)) = (−1)d(d−1)/2 Trf (f∗(c−1 i,π ◦ i!(eπ) ◦ di ◦ η−1 i )(t∨ 1 ∧ . . . ∧ t∨ d ⊗ i∗(ω))). Therefore, it suffices to check that df ◦ ψ−1 i,π = c−1 i,π ◦ i!(eπ) ◦ di, and this results from (VAR5) [Co00, (3.3.26)]. (cid:3) RATIONAL POINTS OF REGULAR MODELS 83 B.9. Proof of Theorem 3.1. (i) The transitivity formula (3.1.1) is the equality of the exterior composites in the diagram Rf∗Rg∗OZ Rf∗Rg∗(ϕg) Rf∗Rg∗ωZ/Y Rf∗Rg∗(ϕf g ) Rf∗Rg∗(Id⊗Lg∗(ϕf )) Rf∗Rg∗ωZ/X Rf∗Rg∗(ζ ′ g,f ) Rf∗Rg∗(ωZ/Y L ⊗OZ Lg∗ωY /X) Rf∗(Tr ♯ g ) Tr ♯ f g OX Tr ♯ f ∼ Rf∗((Rg∗ωZ/Y ) L ⊗OY ωY /X) Rf∗(Tr ♯ g ⊗Id) Rf∗ωY /X Rf∗(ϕf ) Rf∗OY , where the upper left square commutes thanks to (A.8.1), the lower left square is the commutative square (B.5.1), and the right triangle commutes by functoriality. (ii) Thanks to Proposition B.6 and to the description (B.7.4) of τf , the assertion follows from the compatibility of the canonical section δf with Tor-independent pull-backs (proved in Proposition A.8 (ii)) and the functoriality of the base change morphism. X,x of the local ring of X at each point x. As the morphism Spec Oh (iii) To prove (3.1.3), it suffices to prove that the equality holds in the henseliza- tion Oh X,x → X is residually stable [Co00, p. 132], Proposition B.6 and the commutation with base change of the classical trace map for the finite locally free algebra f∗OY allow to assume that X = Spec A, where A is a henselian noetherian local ring. Then Y is a disjoint union of open subschemes Yi = Spec Bi, where Bi is a finite local algebra over A. Each of the morphisms Yi → X is a complete intersection morphism of virtual relative dimension 0 (since this is a local condition on Y ), and the additivity of the trace (valid both for Trf , hence for τf , and for tracef∗OY /OX ) shows that it suffices to prove (3.1.3) for each morphism Yi → X. So we may assume that B is local. We can choose a presentation B ∼= C/I, where C is a smooth A-algebra, and I is an ideal in C. Let P = Spec C, I = IOP , and let y ∈ Y ⊂ P be the closed point. Then Iy is generated by a regular sequence (t1, . . . , td). Shrinking P if necessary, we may assume that t1, . . . , td generate I globally on P , so that the hypotheses of B.8 are satisfied. Then (3.1.3) follows from (B.8.1) and from property (R6) of the residue symbol [Co00, p. 240]. (cid:3) References [Ax64] J. Ax, Zeroes of polynomials over finite fields, Amer. Journal of Math. 86 (1964), 255 -- 261. [Be74] P. Berthelot, Cohomologie cristalline des sch´emas de caract´eristique p, Lecture Notes in Math. 407, Springer-Verlag (1974). [BBE07] P. Berthelot, S. Bloch, H. Esnault, On Witt vector cohomology for singular varieties, Compositio Math. 143 (2007), 363 -- 392. / /       / /       o o o o 84 PIERRE BERTHELOT, H´EL`ENE ESNAULT, AND KAY R ULLING [BO78] P. Berthelot, A. Ogus, Notes on crystalline cohomology, Mathematical Notes 21, Princeton University Press (1978). [Bo70] N. Bourbaki, Alg`ebre, Chapitres 1 -- 3, Springer-Verlag (1970). [Co00] B. Conrad, Grothendieck Duality and Base Change, Lecture Notes in Math. 1750, Springer- Verlag (2000). [CT03] B. Chiarellotto, N. Tsuzuki, Cohomological descent of rigid cohomology for ´etale coverings, Rend. Sem. Mat. Univ. Padova 109 (2003), 63 -- 215. [De71] P. Deligne, Th´eorie de Hodge : II, Publ. Math. Inst. Hautes ´Etudes Sci. 40 (1971), 5 -- 57. [De74] P. Deligne, Th´eorie de Hodge : III, Publ. Math. Inst. Hautes ´Etudes Sci. 44 (1974), 5 -- 77. [De83] P. Deligne, Int´egration sur un cycle ´evanescent, Inv. Math. 76 (1983), 129 -- 143. [dJ96] A. J. de Jong, Smoothness, semi-stability and alterations, Publ. Math. Inst. Hautes ´Etudes Sci. 83 (1996), 51 -- 93. [Dw64] B. Dwork, On the Zeta Function of a Hypersurface: II, Annals of Math. 80 (1964), 227 -- 299. [Es06] H. Esnault, Deligne's integrality theorem in unequal characteristic and rational points over finite fields, Annals of Math. 164 (2006), 715 -- 730. [Ek84] T. Ekedahl, On the multiplicative properties of the de Rham-Witt complex I, Arkiv For Matematik 22 (1984), 185 -- 239. [El78] F. El Zein, Complexe dualisant et applications `a la classe fondamentale d'un cycle, M´em. Soc. Math. France 58 (1978), 5 -- 66. [Fo94] J.-M. Fontaine, Repr´esentations p-adiques semi-stables, in P´eriodes p-adiques, S´eminaire de Bures 1988, Ast´erisque 223 (1994), 113 -- 184. [Gr69] A. Grothendieck, Hodge's general conjecture is false for trivial reasons, Topology 8 (1969), 299 -- 303. [Gs85] M. Gros, Classes de Chern et classes de cycles en cohomologie de Hodge-Witt logarithmique, M´em. Soc. Math. France 21 (1985). [Ha66] R. Hartshorne, Residues and duality, Lecture Notes in Math. 20, Springer-Verlag (1966). [Hy91] O. Hyodo, On the de Rham-Witt complex attached to a semi-stable family, Compositio Math. 78 (1991), 241 -- 260. [HK94] O. Hyodo, K. Kato, Semi-stable reduction and crystalline cohomology with logarithmic poles, [Il71] [Il79] [Il90] [Il06] in P´eriodes p-adiques, S´eminaire de Bures 1988, Ast´erisque 223 (1994), 221 -- 268. L. Illusie, Complexe cotangent et d´eformations I, Lecture Notes in Math. 239, Springer- Verlag (1971). L. Illusie, Complexe de de Rham-Witt et cohomologie cristalline, Ann. Scient. ´Ec. Norm. Sup. 12 (1979), 501 -- 661. L. Illusie, Ordinarit´e des intersections compl`etes g´en´erales, in The Grothendieck Festschrift, vol. II, Birkhauser (1990), 375 -- 405. L. Illusie, Miscellany on traces in ℓ-adic cohomology : a survey, Japan. J. of Math. 1 (2006), 107-136. [IR83] L. Illusie, M. Raynaud, Les suites spectrales associ´ees au complexe de de Rham-Witt, Publ. Math. Inst. Hautes ´Etudes Sci. 57 (1983), 73 -- 212. [Ka89] K. Kato, Logarithmic structures of Fontaine-Illusie, in Algebraic analysis, geometry and number theory, John Hopkins Univ. Press (1989), 191 -- 224. [KM76] F. Knudsen, D. Mumford, The projectivity of the moduli space of stable curves I - Prelim- inaries on "det" and"Div", Math. Scand. 39 (1976), 19 -- 55. [Kz71] N. Katz, On a theorem of Ax, Amer. Journ. Math. 93 (1971), 485 -- 499. [Lo02] P. Lorenzon, Logarithmic Hodge-Witt Forms and Hyodo-Kato Cohomology, J. of Algebra 249 (2002), 247-265. [LZ04] A. Langer, T. Zink, De Rham-Witt cohomology for a proper and smooth morphism, J. Inst. Math. Jussieu 3 (2004), 231 -- 314. [Ma72] B. Mazur, Frobenius and the Hodge filtration, Bull. Amer. Math. Soc. 78 (1972), 653 -- 667. [Ma73] B. Mazur, Frobenius and the Hodge filtration (estimates), Annals of Math. 98 (1973), 58 -- 95. RATIONAL POINTS OF REGULAR MODELS 85 [Na09] Y. Nakkajima, Weight filtration and slope filtration on the rigid cohomology of a variety, preprint (2009). [Ol07] M. Olsson, Crystalline cohomology of algebraic stacks and Hyodo-Kato cohomology, Ast´erisque 316 (2008). [Ra70] M. Raynaud, Sp´ecialisation du foncteur de Picard, Publ. Math. Inst. Hautes ´Etudes Sci. 38 (1970), 27 -- 76. [Se58] J.-P. Serre, Sur la topologie des vari´et´es alg´ebriques en caract´eristique p, Symp. Int. Top. Alg. Mexico (1958), 24 -- 53. [SGA 1] Revetements ´etales et groupe fondamental, S´eminaire de G´eom´etrie Alg´ebrique du Bois- Marie 1960 -- 1961 (SGA 1), dirig´e par A. Grothendieck. Lecture Notes in Math. 224, Springer-Verlag (1971). [SGA 4 II] Th´eorie des topos et cohomologie ´etale des sch´emas. Tome 2, S´eminaire de G´eom´etrie Alg´ebrique du Bois-Marie 1963 -- 1964 (SGA 4), dirig´e par M. Artin, A. Grothendieck et J.-L. Verdier, avec la collaboration de N. Bourbaki, P. Deligne et B. Saint-Donat. Lecture Notes in Math. 270, Springer-Verlag (1972). 2 ] P. Deligne, Cohomologie ´etale, Lecture Notes in Math. 569, Springer-Verlag (1977). [SGA 4 1 [SGA 6] Th´eorie des intersections et th´eor`eme de Riemann-Roch, S´eminaire de G´eom´etrie Alg´ebrique du Bois-Marie 1966 -- 1967 (SGA 6), dirig´e par P. Berthelot, A. Grothendieck et L. Illusie, avec la collaboration de D. Ferrand, J. P. Jouanolou, O. Jussila, S. Kleiman, M. Raynaud et J. P. Serre. Lecture Notes in Math. 225, Springer-Verlag (1971). [SGA 7 II] Groupes de monodromie en g´eom´etrie alg´ebrique. II, S´eminaire de G´eom´etrie Alg´ebrique du Bois-Marie 1967 -- 1969 (SGA 7), dirig´e par P. Deligne et N. Katz. Lecture Notes in Math. 340, Springer-Verlag (1973). [Ts98] T. Tsuji, p-adic Hodge Theory in the Semi-Stable Reduction Case, Proceedings of the Inter- national Congress of Mathematicians, Vol. II (Berlin, 1998). Doc. Math. 1998, Extra Vol. II, 207 -- 216. [Ts99] T. Tsuji, p-adic ´etale cohomology and crystalline cohomology in the semi-stable reduction case, Inventiones Math. 137 (1999), 233 -- 411. [Tz04] N. Tsuzuki, Cohomological descent in rigid cohomology, in Geometric aspects of Dwork theory, edited by A. Adolphson et alii, vol. 2, De Gruyter (2004), 931 -- 981. IRMAR, Universit´e de Rennes 1, Campus de Beaulieu, 35042 Rennes cedex, France E-mail address: [email protected] Mathematik, Universitat Duisburg-Essen, FB6, Mathematik, 45117 Essen, Germany E-mail address: [email protected] Mathematik, Universitat Duisburg-Essen, FB6, Mathematik, 45117 Essen, Germany E-mail address: [email protected]
1512.03285
1
1512
2015-12-10T15:20:07
Universal cohomological expressions for singularities in families of genus 0 stable maps
[ "math.AG" ]
We consider families of curve-to-curve maps that have no singularities except those of genus 0 stable maps and that satisfy a versality condition at each singularity. We provide a universal expression for the cohomology class Poincar\'e dual to the locus of any given singularity. Our expressions hold for any family of curve-to-curve maps satisfying the above properties.
math.AG
math
Universal cohomological expressions for singularities in families of genus 0 stable maps Maxim Kazarian∗, Sergey Lando†, Dimitri Zvonkine‡ Abstract We consider families of curve-to-curve maps that have no singular- ities except those of genus 0 stable maps and that satisfy a versality condition at each singularity. We provide a universal expression for the cohomology class Poincar´e dual to the locus of any given singu- larity. Our expressions hold for any family of curve-to-curve maps satisfying the above properties. 1 Introduction 1.1 A family of curve-to-curve maps A family of curve-to-curve maps is a triple of smooth complex manifolds X, Y and B and a commutative diagram X J p J^ f - Y q  B (1) satisfying the following properties. 1. The map p is a flat family of reduced not necessarily compact nodal curves. 5 1 0 2 c e D 0 1 ] . G A h t a m [ 1 v 5 8 2 3 0 . 2 1 5 1 : v i X r a Economics ∗Steklov Mathematical Institute RAS, National Research University Higher School of †National Research University Higher School of Economics, Independent University of ‡Institut math´ematique de Jussieu, CNRS, 4 place Jussieu, 75005 Paris, France. E- Moscow mail: [email protected]. Partly supported by the ANR project "Geometry and Integrability in Mathematical Physics" ANR-05-BLAN-0029-01. 1 2. The map q is a flat family of reduced not necessarily compact smooth curves. 3. The map f is proper. In Section 1.4 we will introduce a list of allowed singularities for the map f . These will be precisely the singularities of genus 0 stable maps. For each singularity we will also require a versal deformation property: infor- mally, the map f should locally present all possible deformations of each singularity. Assuming that f has allowed singularities only and satisfies the versal deformation property, we will compute the cohomology class of every singularity locus. The forthcoming paper [4] extends the results of the present one to the study of multisingularities. 1.2 Thom polynomials: a motivation Let f : X → Y be a sufficiently generic map between two complex compact manifolds. Denote by c(f ) = c(f∗(T∗Y ))/c(T∗X) = 1 + c1(f ) + c2(f ) + . . . , the full Chern class of the map f , ci(f ) ∈ H 2i(X). Given a singularity type α, one can consider the locus of points x ∈ X such that f has a singularity of type α at x. This locus is a submanifold, and we will denote the Poincar´e dual cohomology class of its closure by [α] ∈ H∗(X, Z). In [8] Thom proved that for every singularity type α and for every generic map f we have [α] = Pα(c1(f ), c2(f ), . . . ), where Pα(c1, c2, . . . ) is a polyno- mial independent of f . For instance, we have [A1] = c1(f ), thus PA1 = c1. The polynomials Pα are called Thom polynomials. It is, in general, quite hard to compute them (Thom's proof is not constructive); however comput- ing even a single Thom polynomial may provide a solution for many seemingly unrelated problems of enumerative geometry. This paper grew out of an attempt to "apply" Thom's theorem to the map f of Eq. (2), a case to which it is not applicable since the genericity condition for f brakes down due to existence of nonisolated singularities. In Section 1.4 we describe all singularity types that can appear in a family of genus 0 stable maps. In Section 1.7 we introduce a family of cohomology classes (larger than just the classes ci(f )) that we will call basic classes. Then we prove that the class determined by every singularity type has a 2 universal polynomial expression in terms of the basic classes, independent of the family f . Moreover, we give a simple way to compute these expressions. 1.3 Local models Consider two families of curve-to-curve maps. X J p J^ f - Y q  B f(cid:48) - X(cid:48) p(cid:48) J J^ B(cid:48) Y (cid:48) q(cid:48)  (2) Let y ∈ Y and b = q(y) ∈ B; y(cid:48) ∈ Y (cid:48) and b(cid:48) = q(cid:48)(y(cid:48)) ∈ B(cid:48). Let X0 be a closed subset of f−1(y) and X(cid:48) 0 a closed subset of (f(cid:48))−1(y(cid:48)). In practice, X0 will be either a point or a connected component of f−1(y) contracted by f . Definition 1.1 We say that the family (X(cid:48), Y (cid:48), B(cid:48)) is a local model for the family (X, Y, B) if there exist three holomorphic submersions ϕX : X → X(cid:48), ϕY : Y → Y (cid:48), ϕB : B → B(cid:48) defined in a neighborhood of X0, y, and b respectively, identifying X0 with X(cid:48) 0 and commuting with the maps p, q, f . 1.4 Singularity types Let x ∈ X be any point, b = p(x) ∈ B and y = f (x) ∈ Y its images in B and Y , respectively, and Xb the fiber of p through x. We will describe several types of singularities of f at x, namely, ramification points, nodes, and contracted components. These are the only singularities that can arise in families of genus 0 stable maps. A map C → CP1 of a genus 0 nodal curve C to CP1 is said to be stable if its automorphism group is finite. In other words, a map is stable if each irreducible component of the domain C that is contracted to a point has at least three points of intersection with other irreducible components of C. For each singularity type we will define a versality condition using local models. Both points and nodes are isolated singularities, while contracted 3 ybX0fpq components are nonisolated ones. Universal polynomials for isolated singu- larities in the absence of non-isolated ones were found in [2, 3]. They de- termine the singularity classes modulo cohomology classes supported on the subvariety in X consisting of contracted components. Including contracted components completes the computation of universal polynomials for genus 0 curves. 1.4.1 Ramification points Definition 1.2 We say that f presents a singularity of type Ak at x if x is a smooth point of Xb and fXb has a ramification of order k + 1 at x. The local model for the Ak singularity is given by the family v(u, γ1, . . . , γk−1) = uk+1 + γ1uk−1 + ··· + γk−1u. Here: (γ1, . . . , γk−1) is a system of coordinates in B(cid:48), (γ1, . . . , γk−1, u) a system of coordinates in X(cid:48), (γ1, . . . , γk−1, v) a system of coordinates in Y (cid:48). Definition 1.3 Let x ∈ X be a point at which f presents an Ak singularity. We say that the family (X, Y, B) is versal at x if it is locally modeled at the neighborhood of x by the family (X(cid:48), Y (cid:48), B(cid:48)) above. 1.4.2 Nodes Definition 1.4 We say that f presents a singularity of type Ik1,k2 at x if x is a node of Xb and f has ramifications of orders k1 and k2 on the branches of Xb at x. A local model for the singularity Ik1,k2 is given by the family v = uk1 1 + γ(1) 1 uk1−1 1 + ··· + γ(1) β = u1u2, k1−1u1 + uk2 2 + γ(2) 1 uk2−1 2 + ··· + γ(2) k2−1u2. k1−1, γ(2) k1−1, γ(2) k1−1, γ(2) 1 , . . . , γ(2) 1 , . . . , γ(2) 1 , . . . , γ(2) Here: (γ(1) 1 , . . . , γ(1) (γ(1) 1 , . . . , γ(1) (γ(1) 1 , . . . , γ(1) Definition 1.5 Let x ∈ X be a point at which f presents an Ik1,k2 singular- ity. We say that the family (X, Y, B) is versal at x if it is locally modeled at the neighborhood of x by the family (X(cid:48), Y (cid:48), B(cid:48)) above. k2−1, β) is a system of coordinates in B(cid:48), k2−1, u1, u2) a system of coordinates in X(cid:48), k2−1, β, v) a system of coordinates in Y (cid:48). 4 branches C k(cid:96) k2 x k1 fb y Figure 1: The contracted part C of a point x. Branches are indicated, to- gether with the orders k1, . . . , k(cid:96) of the restriction of the function fb to the corresponding branch 1.4.3 Contracted components Stable maps of genus 0 can also present non-isolated singularities. Suppose that x happens to lie on an irreducible component of Xb contracted by f . Then we denote by C the connected component of f−1 b (y) ⊂ Xb containing x. Note that it necessarily includes the irreducible component of Xb contain- ing x, but may include other irreducible components as well. We call C the contracted part of x. Finally we call branches the connected components of the neighborhood of C in Xb \ C. If, on C, we mark its intersection points x1, . . . , x(cid:96) with the branches, it becomes a stable curve. Definition 1.6 For (cid:96) ≥ 3, we say that f presents a singularity of type Ik1,k2,...,k(cid:96) at x if • f is constant on the irreducible component of Xb containing x; • the contracted part C has genus 0; • there are (cid:96) branches and they can be numbered from 1 to (cid:96) in such a way that f has ramification orders k1, . . . , k(cid:96) at their intersection points with C. To define a local model for this singularity we need to introduce the space of relative stable maps; however, the local model may actually be only a subspace of this space, see Definition 1.7. 5 Let M0;k1,...,k(cid:96)(CP1;∞) be the space of stable maps of genus 0 to CP1, defined up to an additive constant, and relative to ∞ with ramification profile (CP1;∞) (k1, . . . , k(cid:96)). This space has a distinguished sub-orbifold MZ that we will call the zero locus. It parametrizes maps from curves with (cid:96) + 1 components: one contracted rational component and (cid:96) more rational components that intersect the contracted one and on which the map equals zki, 1 ≤ i ≤ (cid:96). As an orbifold, the zero locus is a gerb with base M0;(cid:96) and 0;k1,...,k(cid:96) group(cid:81)(cid:96) Z/kiZ. Let i=1 p : C0;k1,...,k(cid:96)(CP1;∞) → M0;k1,...,k(cid:96)(CP1;∞) be the universal curve, q : CP1 × M0;k1,...,k(cid:96)(CP1;∞) → M0;k1,...,k(cid:96)(CP1;∞) the universal target f : C0;k1,...,k(cid:96)(CP1;∞) → CP1 × M0;k1,...,k(cid:96)(CP1;∞) the universal map. curve, and The union of contracted parts of fibers over the zero locus MZ forms the zero locus CZ (CP1;∞) in the universal curve. 0;k1,...,k(cid:96) Let x ∈ X be a point at which f presents an Ik1,...,k(cid:96) singularity and C (CP1;∞) be the the corresponding contracted component. Let b ∈ MZ point of the zero locus such that the contracted component is isomorphic to C. Consider a local chart of M0;k1,...,k(cid:96)(CP1;∞) at b. Now, let B(cid:48) be any sub-manifold of this local chart that is transversal to the zero locus and of dimension at least 0;k1,...,k(cid:96) (CP1;∞) 0;k1,...,k(cid:96) (cid:88) dimM0;k1,...,k(cid:96)(CP1;∞) − dimMZ 0;k1,...,k(cid:96) (CP1;∞) = ki. Let X(cid:48) be a neighborhood of C in the preimage (p(cid:48))−1(B(cid:48)) and Y (cid:48) a neigh- borhood of y = f (x) in the preimage (q(cid:48))−1(B(cid:48)). Definition 1.7 Let x ∈ X be a point at which f presents an Ik1,...,k(cid:96) sin- gularity and C the corresponding contracted component. We say that f is versal at x if there exists a family (X(cid:48), Y (cid:48), B(cid:48)) as above such that (X, Y, B) is locally modeled on (X(cid:48), Y (cid:48), B(cid:48)) at a neighborhood of C. 1.5 Versal families Definition 1.8 A genus 0 versal family of maps is a family of maps that only has singularities of types Ak and Ik1,...,k(cid:96) for (cid:96) ≥ 2 and satisfies the versality condition for each singularity. Genus 0 versal families are precisely the object of our study. Before proceeding we give several examples of genus 0 versal families. 6 Example 1.9 Consider the space M0;d(CP1) of stable genus 0 degree d maps to CP1 without marked points. The universal map over this space is a genus 0 versal family. Example 1.10 Given a line bundle L → B over a smooth base B, consider the space (L \ {zero section}) ×C∗ M0;d(CP1) → B. Here C∗ acts on M0;d(CP1) via the natural action on the target CP1. The universal map over this space is a genus 0 versal family. Example 1.11 Given a principal PSL(2, C) bundle G → B over a smooth base B, consider the space G ×PSL(2,C) M0;d(CP1) → B. Here PSL(2, C) acts on M0;d(CP1) via the natural action on the target CP1. The universal map over this space is a genus 0 versal family. Example 1.12 Consider the space M0;k1,...,k(cid:96)(CP1;∞) of stable genus 0 maps to CP1 relative to ∞ with ramification profile (k1, . . . , k(cid:96)). Consider the universal map over this space. Remove the ∞ section in the target curve and its preimages in the source curve. The family thus obtained is a genus 0 versal family. Example 1.13 Given a line bundle L → B over a smooth base B, consider the space (L \ {zero section}) ×C∗ M0;k1,...,k(cid:96)(CP1;∞) → B. Here C∗ acts on M0;k1,...,k(cid:96)(CP1;∞) via the natural action on the target CP1. Consider the universal map over this space. Remove the ∞ section in the target curve and its preimages in the source curve. The family thus obtained is a genus 0 versal family. Example 1.14 Consider the space Mg;d(C) of degree d genus g stable maps to a given smooth curve C. Consider the universal map over this space. Remove from the universal map all the source curves that contain contracted components of genus greater than 0. Also remove the images of these curves in the target. The family thus obtained is a genus 0 versal family. 7 Example 1.15 Let z = z1, . . . , z = z(cid:96) be a given set of points on CP1 with affine coordinate z. Consider the family of maps (cid:34)(cid:18) ui z − zi (cid:96)(cid:88) i=1 (cid:19)ki (cid:19)ki−1 (cid:18) ui z − zi f (z) = + γi,1 + ··· + γi,ki−1 (cid:18) ui (cid:19)(cid:35) z − zi . This family is extended to ui = 0 in the following way. The source CP1 acquires a bubble with global coordinate wi. The point wi = 0 is attached to z = zi. The function on the bubble is given by i + γi,1wki−1 wki i + ··· + γi,ki−1wi. This family is a genus 0 versal family. 1, . . . , z(cid:48) = z(cid:48) Example 1.16 This is an example of a family that is NOT versal. Let z(cid:48) = z(cid:48) (cid:96)1+1 be a given set of points on the projective line with global coordinate z(cid:48) and z(cid:48)(cid:48) = z(cid:48)(cid:48) (cid:96)2+1 be a given set of points on another projective line with global coordinate z(cid:48)(cid:48). The point z(cid:48) (cid:96)1+1 is identified with the point z(cid:48)(cid:48) (cid:96)2+1. Consider the following family of maps on the nodal curve thus obtained: 1 , . . . , z(cid:48)(cid:48) = z(cid:48)(cid:48) (cid:34)(cid:18) u(cid:48) (cid:96)1(cid:88) (cid:34)(cid:18) u(cid:48)(cid:48) (cid:96)2(cid:88) i=1 z(cid:48) − z(cid:48) i i i i (cid:19)k(cid:48) (cid:19)k(cid:48)(cid:48) i z(cid:48)(cid:48) − z(cid:48)(cid:48) i i=1 f (z(cid:48)) = f (z(cid:48)(cid:48)) = (cid:18) u(cid:48) (cid:18) u(cid:48)(cid:48) z(cid:48) − z(cid:48) i i i−1 (cid:19)k(cid:48) (cid:19)k(cid:48)(cid:48) i z(cid:48)(cid:48) − z(cid:48)(cid:48) i i −1 + γi,1 + γi,1 (cid:19)(cid:35) (cid:18) u(cid:48) (cid:18) u(cid:48)(cid:48) z(cid:48) − z(cid:48) i i i z(cid:48)(cid:48) − z(cid:48)(cid:48) i +C(cid:48); (cid:19)(cid:35) +C(cid:48)(cid:48); + ··· + γi,k(cid:48) i−1 + ··· + γi,k(cid:48)(cid:48) i −1 (cid:96)1+1) = f (z(cid:48)(cid:48) where the constants C(cid:48) and C(cid:48)(cid:48) are such that f (z(cid:48) family is extended to u(cid:48) i = 0 and u(cid:48)(cid:48) i = 0 as in the previous example. (cid:96)2+1) = 0. This This family is not versal. Indeed, even though it satisfies the versal prop- erty for the singularity Ik(cid:48) , a generic map of this family also has an I1,1 singularity that does not satisfy the versal property because the node is never smoothened. ,k(cid:48)(cid:48) 1 ,...,k(cid:48)(cid:48) 1,...,k(cid:48) (cid:96)2 (cid:96)1 1.6 Singularity loci Every singularity from the list in Sec. 1.4 determines a cycle in X, namely the closure of the set of points x ∈ X where f presents a given singularity. By abuse of notation we will usually denote in the same way the singularity type and the corresponding cycle in X. For instance, the cycle I1,1 is the set of nodes of the fibers Xb. 8 The versal deformation property insures that the codimension of the Ak locus equals k, that of the Ik1,k2 locus equals k1 + k2, and that of the Ik1,...,k(cid:96) locus equals(cid:80) ki. In the notation Ik1,...,k(cid:96) the indices form a multiset, that is, the order of the indices is immaterial. Since the cycle Ik1,...,k(cid:96) is the closure of the set where f presents the corresponding singularity, the ramification orders of f at the attachment points of the branches can actually be greater than (but not smaller than) ki. Thus, for instance, the cycle I1,2 presents a self-intersection along the cycle I2,2. Therefore the normalization of I1,2 contains two copies of I2,2. Let Aut{k1, . . . , k(cid:96)} be the number of permutations σ ∈ S(cid:96) such that Fix an order of the indices k1, . . . , k(cid:96). Let (cid:98)Ik1,...,k(cid:96) be the space of couples kσ(i) = ki for every i. (x ∈ Ik1,...,k(cid:96), numbering of the branches), such that the ramification order at Then (cid:98)Ik1,...,k(cid:96) is a finite nonramified covering of the normalization of Ik1,...,k(cid:96). the branch number i is equal to ki (greater than or equal to ki in the closure). Its degree equals Aut{k1, . . . , k(cid:96)}. Poincar´e dual to the cycle Ak. We denote by ik1,...,k(cid:96) ∈ H 2(cid:80) ki(X, Q) Notation 1.17 We denote by ak ∈ H 2k(X, Q) the cohomology class the cohomology class Poincar´e dual to the cycle Ik1,...,k(cid:96) multiplied by Aut{k1, . . . , k(cid:96)}. Remark 1.18 Including the factor Aut{k1, . . . , k(cid:96)} in the definition of co- homology classes will greatly simplify the formulas in the sequel. Notation 1.19 We denote by U : (cid:98)Ik1,...,k(cid:96) → Ik1,...,k(cid:96) the covering map and by V : (cid:98)Ik1,...,k(cid:96) → M0,(cid:96)+1 the natural map from the covering (cid:98)I to the moduli space of curves: Given a class α ∈ H∗(M0,(cid:96)+1), we denote by αik1,...,k(cid:96) the class U∗V ∗(α). Note that the notation is coherent: if α = 1 we have U∗V ∗(1) = ik1,...,k(cid:96) as defined previously, because the map U has degree Aut{k1, . . . , k(cid:96)}. Definition 1.20 The cohomology classes ak and αik1,...,k(cid:96) in H∗(X, Q) are called singularity classes. 1.7 Basic classes Now we are going to introduce a different set of cohomology classes in X. We will call them basic classes. 9 U←−(cid:98)Ik1,...,k(cid:96) Ik1,...,k(cid:96) V−→ M0,(cid:96)+1. Definition 1.21 Let LX → X be the line bundle whose fiber over x ∈ X is the cotangent line to Xb at x. This line bundle is well-defined outside I1,1 and can be uniquely extended to I1,1 (in the standard way). We denote by ψ = c1(LX) ∈ H 2(X, Q) its first Chern class. Let LY → X be the line bundle whose fiber over x ∈ X is the cotangent line to f (Xb) at y = f (x). This line bundle is well-defined everywhere and we denote by ξ = c1(LY ) ∈ H 2(X, Q) its first Chern class. (cid:98)I1,...,1 consider the (cid:96) line bundles whose fibers are the cotangent lines to the Consider the cycle I1,...,1 with (cid:96) subscripts. Over its (cid:96)!-sheeted covering branches at their intersection points with the contracted part. Denote their In a more traditional notation, we have first Chern classes by ν1, . . . , ν(cid:96). νi = V ∗ψi, i = 1 . . . , (cid:96). Let α ∈ H∗(M0,(cid:96)+1) be a cohomology class. We denote by αδm1,...,m(cid:96) the cohomology class U∗(νm1 1 ··· νm(cid:96) (cid:96) V ∗α). Definition 1.22 The classes ψk, and αδk1,...,k(cid:96) are called basic cohomology classes. Our aim is now to express singularity classes in terms of the basic classes and vice versa. 1.8 Main results We start with an explicit equality. Theorem 1 For every m ≥ 1 we have (rψ − ξ) = am + Pm, where Pm is a linear combination of terms of the form ξqψpik1,...,k(cid:96). Moreover P1 = 0 and r=1 m(cid:81) k1 ··· k(cid:96) ik1,...,k(cid:96) + (mψ − ξ) Pm−1 (cid:32)(cid:88) 1 (cid:96)! (cid:88) (cid:96)≥2 k1+···+k(cid:96)=m (cid:33) Pm = for m ≥ 2. Sample calculations for 1 ≤ m ≤ 6 are given in the appendix. The equality of Theorem 1 does not express a singularity class via basic classes or a basic class via singularity classes. However it is the key result in the proof of the following theorem. Theorem 2 Every singularity class can be expressed as a linear combination of basic classes multiplied by powers of ξ. Every basic class can be expressed as a linear combination of singularity classes multiplied by powers of ξ. 10 The proof of this theorem is constructive, that is, the expressions can be effectively computed. In the appendix we write out the expressions for all classes up to codimension 5. Finally, in the following theorem, we give an explicit formula for certain coefficients in the expressions of basic classes in terms of singularity classes. Introduce the following polynomials in variables xk: (cid:88) (cid:96) 1 (cid:96)! (cid:88) (cid:80) ki=m−(cid:96)+2 k1,...,k(cid:96) Xm = (cid:96)(cid:89) i=1 m! (m − (cid:96) + 2)! kixi. Denote by α(cid:96) ∈ H top(M0;(cid:96)+1) the cohomology class Poincar´e dual to a point. Theorem 3 Choose m and k1, . . . , k(cid:96) so that m + 2 = (cid:96) + ki. The co- efficient of α(cid:96)ik1,...,k(cid:96) in the expression for ψm is equal to the coefficient of xk1 ··· xk(cid:96) in Xm, that is, to i=1 1 Aut{k1, . . . , k(cid:96)} m! (m − (cid:96) + 2)! ki. (cid:96)(cid:80) (cid:96)(cid:80) i=1 ki. The co- (cid:89) s(cid:80) j=1 Choose m1, . . . , ms and k1, . . . , k(cid:96) so that 2s + mj = (cid:96) + efficient of α(cid:96)ik1,...,k(cid:96) in the expression for αsδm1,...,ms is equal to the coefficient of xk1 ··· xk(cid:96) in Xm1 ··· Xms. In Section 3 we relate these coefficients with Okounkov and Pandhari- pande's completed cycles. 2 Computations with singularity and basic classes 2.1 Proof of Theorem 1 The proof goes by induction. The equality ψ − ξ = A1 follows from the fact that df is a section of the line bundle LX ⊗L∗ Y (see Definition 1.21) that has a simple zero precisely on the A1 locus. Assume that the equality is true for m − 1 and prove it for m. By the induction assumption we have m(cid:89) (rψ − ξ) = (mψ − ξ)(am−1 + Pm−1). r=1 11 To do that, note that dmf is a section of the line bundle L⊗m The term (mψ − ξ)Pm−1 is a part of the expression for Pm. Thus our main task is to compute the product (mψ − ξ)am−1. X ⊗ L∗ Y restricted to the locus Am−1. Thus we must describe the components of the zero locus of dmf and their multiplicities. The obvious component of the zero locus is Am ⊂ Am−1. On this compo- nent the section dmf has a simple zero as follows from the versality condition Indeed, in the local model of Definition 1.3 the stratum Am−1 is for Am. parametrized by one variable a via um+1 + γ1um−1 + . . . γm−1u = (u − a)m(u + ma) + mam; the section dmfu=a in this parametrization is just (m+1)a, so it has a simple zero at a = 0. The other components are the cycles Ik1,...,k(cid:96) that lie in the closure of Am−1 and are of codimension 1 in Am−1. We are going to show that all cycles Ik1,...,k(cid:96) of codimension m, that is, (cid:80) ki = m, lie in the closure of Am−1 and that the vanishing multiplicity of dmf on the cycle Ik1,...,k(cid:96) equals k1 ··· k(cid:96). To do that, we study the local models for each of these singularities. We will assume that (cid:96) ≥ 3 leaving the simpler and very similar case (cid:96) = 2 to the reader (see also [3]). Let M0;k1,...,k(cid:96)(CP1;∞) be the space of stable maps of genus 0 to CP1, defined up to an additive constant, and relative to ∞ with ramification pro- file (k1, . . . , k(cid:96)). Denote by C0;k1,...,k(cid:96)(CP1;∞) the universal curve and let CZ (CP1;∞) ⊂ C0;k1,...,k(cid:96)(CP1;∞) be the zero locus in the universal curve (see the discussion before Definition 1.7). Lemma 2.1 The natural section dmf : Am−1 → L⊗m Y on the stra- tum Am−1 ⊂ C0;k1,...,k(cid:96)(CP1;∞) has a vanishing of order k1 ··· k(cid:96) along CZ X ⊗ L∗ 0;k1,...,k(cid:96) 0;k1,...,k(cid:96) From the lemma we conclude that (mψ − ξ)am−1 = am + (CP1;∞), whenever(cid:80) ki = m. (cid:88) (cid:88) k1,...,k(cid:96)(cid:80) ki=m (cid:88) = am + 1 (cid:96)! (cid:96) (cid:96) k1 ··· k(cid:96) [Ik1,...,k(cid:96)] (cid:88) (cid:80) ki=m k1 ··· k(cid:96) ik1,...,k(cid:96). k1≤···≤k(cid:96) This holds for any family that satisfies the versality condition of Defini- ♦ tion 1.7. Theorem 1 is proved. 12 Proof of the lemma. We say that a point of C0;k1,...,k(cid:96)(CP1;∞) is generic if the contracted part is smooth. We will study the vanishing order at a generic point. Introduce a coordinate z on the contracted part and denote by x, z1, . . . , z(cid:96) the marked point and the intersection points with the branches. It is proved in [1] that M0;k1,...,k(cid:96)(CP1;∞) is a cone over its zero locus and that the fiber of this cone is parametrized by coordinates ui, 1 ≤ i ≤ (cid:96) and aij, 1 ≤ j ≤ ki − 1 if we write the stable map in the following form: (cid:19)(cid:35) (cid:18) ui z − zi . (3) + ··· + ai,ki−1 (cid:34)(cid:18) ui (cid:96)(cid:88) (cid:19)ki z − zi + ai,1 z − zi (cid:19)ki−1 (cid:18) ui (cid:81)(cid:96) (z − x)m i=1(z − zi)ki . f (z) = f (z) = i=1 Let This is the unique, up to transformations f (cid:55)→ cf + b, rational function with poles of orders ki at the points zi and such that f(cid:48)(x) = f(cid:48)(cid:48)(x) = ··· = f (m−1)(x) = 0. Expand f in the form (3) and denote by ¯ui and ¯aij the coefficients thus obtained. Further, let K = LCM(k1, . . . , k(cid:96)) and ri = for 1 ≤ i ≤ (cid:96). K ki Then the stratum Am−1 in M0;k1,...,k(cid:96)(CP1;∞) consists of d = k1 ··· k(cid:96)/L irreducible components parametrized by a complex parameter c in the fol- lowing way: aij = αj ui = αicri ¯ui, i cjri¯aij. Here (α1, . . . , α(cid:96)) is a collection of roots of unity, αki i = 1. We should choose one such collection in each of the d orbits of Z/KZ in Z/k1Z × ··· × Z/k(cid:96)Z and this choice determines the irreducible component of the stratum Am−1. In each component of the stratum, the mth derivative at x of the function corresponding to parameter c is equal to cK · f (m)(x). Thus the vanishing order of the mth derivative at c = 0 equals K for each component. Since there are k1 ··· k(cid:96)/K components, we get the total vanishing order equal to k1 ··· k(cid:96). ♦ 13 2.2 Proof of Theorem 2 2.2.1 Classes in M0;(cid:96)+1 The definition of both basic and singularity classes involves a class α ∈ H∗(M0;(cid:96)+1). Before proving the theorem we must introduce notation for these classes. They will be described using marked trees. Definition 2.2 An (cid:96)-tree or just a tree is a rooted tree with (cid:96) leaves; the valencies of all vertices except the leaves and the root are at least 3; the valency of the root is 1 (but we don't call it a leaf). A marked (cid:96)-tree or just a marked tree is an (cid:96)-tree whose all vertices except the root are marked with nonnegative integers. The picture below shows a marked tree. If T is a marked tree we will denote by t the underlying tree obtained by forgetting the markings. Take a marked tree T and number its leaves in an arbitrary way. This tree with numbered leaves determines a cohomology class in M0;(cid:96)+1 as fol- lows. Consider the set of curves in M0;(cid:96)+1 whose dual graph1 is isomorphic to t, the marked point x being the root of the tree. Every interior vertex of t corresponds to an irreducible component of the curve. On this irre- ducible component there is exactly one marked point that leads to x (in other words, from every vertex v of t exactly one edge leads to the root). The first Chern class of the cotangent line bundle to this point is denoted by ψv. The cohomology class that we assign to T with numbered leaves is supported on the boundary stratum correspond- ing to t, where the product runs over all internal vertices v of t and mv is the corresponding marking. Note that this class does not depend on the markings of the leaves. given by [T ] =(cid:81) v∈V (T ) ψmv v Using the class [T ] we can now assign to a marked tree T both a singularity class and a basic class. These classes will not depend on the numbering of the leaves of T , but will depend on their markings. Denote by m1, . . . , m(cid:96) the integer markings on the leaves. 1The dual graph of a curve is obtained by replacing every irreducible component by a vertex, every node by an edge, every marked point by a leaf, and the point x by the root. 14 2011001102 Basic class. The basic class that we assign to T is given by [T ]basic = U∗ V ∗[T ] where U : (cid:98)I1,...,1 → I1,...,1 and V : (cid:98)I1,...,1 → M0;(cid:96)+1 are as in Definition 1.22. l (cid:32) (cid:89) (cid:33) νml l , The singularity class. The singularity class assigned to T is where U : (cid:98)Im1+1,...,m(cid:96)+1 → Im1+1,...,m(cid:96)+1 and V : (cid:98)Im1+1,...,m(cid:96)+1 → M0;(cid:96)+1 are as [T ]sing = U∗(V ∗[T ]), in Notation 1.19 and Definition 1.20. Example 2.3 If T is the tree then [T ]sing = ψpim1+1,...,m(cid:96)+1, [T ]basic = ψpδm1,...,m(cid:96). Notation 2.4 If T is the tree we let, by convention, [ [ ]sing = 1. ]basic = ψm, [ ]sing = am, [ ]basic = Thus every singularity class and every basic class can be represented as a linear combination of trees. Remark 2.5 Keel [5] proved that the cohomology classes Poincar´e dual to the boundary strata of M0;(cid:96)+1 span the whole cohomology ring of M0;(cid:96)+1. Therefore our tree notation is sufficient to express any class, but is strongly redundant. For instance, whenever the integer dv assigned to an interior vertex of T is greater than the valency of v minus 3, the corresponding class vanishes. As another example of redundancy, we have [T1]sing = 3[T2]sing, where 15 ...m1m2mnpmmm0000000T1=T=20001 However the classes in our expressions naturally appear in the form of marked (cid:96)-trees, and we chose not to simplify them further, since we do not know any simple nonredundant expression for the result of such simplification. We finish this section with a definition that we will need soon. Definition 2.6 Suppose (cid:96) marked trees T1, . . . , T(cid:96) are assigned to the leaves of a marked (cid:96)-tree T . The substitution of T1, . . . , T(cid:96) into T is the tree obtained by erasing the leaves of T and gluing the roots of T1, . . . , T(cid:96) into the vertices on which the leaves grew. Note that every vertex of the substitution (except the root) inherits an integer from either T or one of the Ti's, but not both. If instead of the trees T1, . . . , T(cid:96) we have (cid:96) linear combinations of trees we can, of course, extend the operation of substitution by multi-linearity. 2.2.2 Basic classes via singularity classes Recall the list of basic classes: ψk and [T ]basic, where T is a marked (cid:96)-tree, (cid:96) ≥ 2. We must express every class like that as a linear combination of singularity classes multiplied by powers of ξ. Expressions for ψm. Theorem 1 gives an expression for (rψ − ξ) as a linear combination of singularity classes multiplied by powers of ξ. Writing ψ = (ψ − ξ) + ξ, ψ2 = ξ · (ψ − ξ) + ξ2, m(cid:81) r=1 · (2ψ − ξ)(ψ − ξ) + · (3ψ − ξ)(2ψ − ξ)(ψ − ξ) + 3 2 1 2 1 6 ψ3 = ξ · (2ψ − ξ)(ψ − ξ) + 11 12 ξ2 · (ψ − ξ) + ξ3, 7 4 and so on, we obtain similar expressions for powers of ψ. Example 2.7 We have ψ = a1 + ξ, ψ2 = 1 2 a2 + 1 4 i1,1 + 3 2 ξa1 + ξ2. More computations are given in the appendix. To sum up, for every m there exists a linear combination Lm of trees with coefficients in Q[ξ] such that ψm = [Lm]sing. 16 20110011021110200100212 The general basic class. Consider class [T ]basic and denote by m1, . . . , m(cid:96) the integers on the leaves of T . Proposition 2.8 Let (cid:98)T be the substitution of Lm1, . . . , Lm(cid:96) into the tree T . Then the class [T ]basic is equal to [(cid:98)T ]sing. Proof. This proposition is almost obvious. It suffices to note that every class νi plays the role of the class ψ on the corresponding branch, and there- ♦ fore we can use the expressions for ψm to express νmi for each i. i Example 2.9 Let us compute the expression of δ0,1,2 in singularity classes. We have (cid:20) (cid:21) δ0,1,2 = ]sing , (cid:20) ψ = [ (cid:21) , basic ]sing + ξ [ ]sing , 1 = [ ψ2 = 1 2 [ ]sing + 1 4 + 3 2 ξ [ sing ]sing + ξ2 [ ]sing . Substituting the last three expressions into the first tree, we obtain a linear combination of 8 trees that simplifies to 1 2 i1,2,3 + 1 2  3 2 ξi1,2,2 +  δ0,1,2 = + 1 4 + 1 4 ξ  5 2  . ξi1,1,3 + ξ2i1,1,2 + ξ3i1,1,1+ sing sing 2.2.3 Singularity classes via basic classes Let T be a marked (cid:96)-tree with the markings of the leaves equal to m1, . . . , m(cid:96). We call m1 + ··· + m(cid:96) the weight of T . It follows from the previous section that the expression of [T ]basic via singularity classes has the form [T ]basic = 1 m1!··· m(cid:96)! [T ]sing + lower weight terms. Therefore these expressions form a triangular change of basis and can be inverted. Theorem 2 is proved. The expressions for the simplest singularity classes are computed in the appendix. 17 0012010200010001000000000 2.3 Proof of Theorem 3 The expression for ψm. The first claim of Theorem 3 gives an explicit in the expression of ψm, formula for the coefficient of the class α(cid:96)ik1,...,k(cid:96) where α(cid:96) is the class of a point in M0;(cid:96)+1. We will prove this formula using Theorem 1; however, since the terms we are interested in do not contain the class ξ, we can reduce the formula of Theorem 1 modulo ξ. We obtain: where (cid:101)P1 = 0, and (cid:32)(cid:88) (cid:101)Pm = m!ψm = am + (cid:101)Pm (mod ξ), (cid:33) (cid:88) k1 ··· k(cid:96) ik1,...,k(cid:96) 1 (cid:96)! n≥2 k1+···+k(cid:96)=m + mψ (cid:101)Pm−1 for m ≥ 2. Thus we see that the term ik1,...,k(cid:96) first "appears" in (cid:101)Pm−(cid:96)+2 with coefficient k1 ··· k(cid:96) Aut{k1, . . . , k(cid:96)} and then gets multiplied by (m − (cid:96) + 3)ψ, . . . , mψ, until it becomes k1 ··· k(cid:96) Aut{k1, . . . , k(cid:96)} m! (m − (cid:96) + 2)! ψ(cid:96)−2ik1,...,k(cid:96). Since the class ψ(cid:96)−2 is precisely the class of a point in M0;(cid:96)+1 (see, for in- stance, [1]), we obtain precisely the same coefficient as in the formulation of the theorem. The expression for αsδm1,...,ms. Recall that α(cid:96) ∈ H (cid:96)−2(M0;(cid:96)+1) is the class of a point. According to Proposition 2.8, to obtain the expression of αsδm1,...,ms we must substitute the expressions for ψmi into the marked (cid:96)-tree T : It is easy to determine when a substitution of (cid:96) marked trees T1, . . . , T(cid:96) into T contributes to the class α(cid:96)ik1,...,k(cid:96). First of all, the marked trees Tj must have (cid:96) leaves altogether with markings k1 − 1, . . . , k(cid:96) − 1. Second, if a marked tree 18 ...m1m2mss−2 1 − 1, . . . , k(j) 1 ,...,k(j) nj nj − 1, then it must describe a . The coefficient of the singu- Tj has (cid:96)j leaves with markings k(j) singularity class proportional to αnj ik(j) larity class αnj ik(j) is equal to the coefficient of the monomial xk(j) in the polynomial Xmj . Therefore, as claimed in the theorem, the coefficient of the singularity class α(cid:96)ik1,...,k(cid:96) in the expres- sion for αsδm1,...,ms is equal to the coefficient of the monomial xk1 ··· xk(cid:96) in the polynomial Xm1 ··· Xms. ··· xk(j) nj in the expression for νmj j 1 ,...,k(j) nj 1 3 Completed cycles 3.1 Completed cycles and the classes ψm Let CSN be the group algebra of the symmetric group SN . If f ∈ CSN is a central element of the group algebra, and λ a partition of N , we denote by f (λ) the scalar by which f acts in the irreducible representation λ. (By Schur's lemma every central element of CSN acts by a scalar in every irre- ducible representation of SN .) This identifies the center of CSN with the algebra of functions on the set of partitions of N . Let Ck1,...,k(cid:96) be the sum of all permutations in SN with (cid:96) numbered cycles of lengths k1, . . . , k(cid:96) respec- of all transpositions, C1,1 is N (N − 1) times the identity permutation, etc. Note that Ck1,...,k(cid:96) defines a central element in CSN simultaneously for all N . For instance, the sum of transpositions C2 is a well-defined element of CSN tively, and N −(cid:80) ki non-numbered fixed points. For instance, C2 is the sum for any N . If N < (cid:80) ki, then the corresponding central element vanishes. Thus Ck1,...,k(cid:96) is actually a family of central elements in the group algebras CSN for all N . Definition 3.1 Given the positive integers k1, . . . , k(cid:96), the family of central elements Ck1,...,k(cid:96) ∈ CSN for N ≥ 1 is called a stable2 central element. Theorem 4 (Kerov, Olshanski [6]) For every m ≥ 0 there exists a unique linear combination C m+1 of stable central elements Ck1,...,k(cid:96) such that (cid:2)(λi − i + 1/2)m+1 − (−i + 1/2)m+1(cid:3) C m+1(λ) = 1 (m + 1)! for every partition λ = (λ1 ≥ λ2 ≥ . . . ) of every integer N . 2This has nothing to do with stable maps. 19 Example 3.2 We have C 1 = C1, C 2 = C2, C3 + C1,1 + C1, C4 + C1,2 + C2, C 3 = C 4 = C 5 = 1 2 1 6 1 24 1 4 1 3 1 8 1 24 5 24 1 12 C5 + C1,3 + C2,2 + 1 36 C1,1,1 + 11 48 C3 + 1 32 C1,1 + 1 1920 C1. Definition 3.3 The linear combination C m+1 is called the completed (m+1)- cycle. Remark 3.4 The set Sλ = {(λi − i + 1/2)i≥1} ⊂ Z + 1 2 is a uniform way to encode a partition of any size. The empty partition corresponds to the set S∅ of negative half-integers. The set Sλ differs from S∅ by a finite number of elements. The function C m+1 is basically the renormalized sum of (m + 1)-st powers of the elements of Sλ divided by (m + 1)!. Denote by S(z) the power series S(z) = sinh(z/2) z/2 . Proposition 3.5 (Okounkov, Pandharipande, [7], Proposition 3.2) We have C m+1 = ρg;k1,...,k(cid:96)Ck1,...,k(cid:96), where the rational constant ρg;k1,...,k(cid:96) is the coefficient of z2g in the power series (cid:88) g≥0, n≥1 1 (cid:96)! (cid:88) k1,...,k(cid:96) (cid:80) ki+(cid:96)+2g−2=m (cid:96)(cid:89) S(kiz), (cid:81) ki K! S(z)K−1 (cid:96)(cid:88) K = ki. i=1 i=1 Remark 3.6 Our normalization of completed cycles C m+1 differs from that of [7] in three ways. First, by a factor of m!; second, by the absence of a constant term corresponding to (cid:96) = 0; third, by the fact that, contrary to [7], for us Ck1,...,k(cid:96) is the sum of permutations with (cid:96) numbered cycles, which changes the corresponding coefficient in the completed cycle by a factor Aut{k1, . . . , k(cid:96)}. 20 Every constant ρg;k1,...,k(cid:96) is associated to a singularity of a higher genus stable map. Namely, by analogy with Definition 1.6, we can say that f presents a singularity of type Ig;k1,...,k(cid:96) at x ∈ C if x lies on a contracted part of genus g meeting (cid:96) branches of C at ramification points of orders k1, . . . , k(cid:96). It is well-known that if a stable map f presents a singularity of type Ig;k1,...,k(cid:96) at x, then the image f (x) must be considered a branch point of multiplicity m = (cid:80) ki + (cid:96) + 2g − 2. In particular, if the stable map can be deformed into a generic smooth map, the branch point will split into m simple branch points. Since in this paper we only study stable maps of genus 0, we will be only interested in the genus 0 part of the completed cycles, that is, the part that corresponds to the g = 0 terms in the sum of Proposition 3.5, or in other words, to the terms satisfying(cid:80) ki + (cid:96) = m + 2. Proposition 3.7 If (cid:80) ki + (cid:96) = m + 2, then the coefficient of Ck1,...,k(cid:96) in C m+1 is the same as the coefficient of α(cid:96)ik1,...,k(cid:96) in the expression of ψm via the singularity classes, where α(cid:96) is the class of a point in M0;(cid:96)+1. Proof. According to Theorem 3 and Proposition 3.5 both are equal to (cid:81) ki ((cid:80) ki)! 1 Aut{k1, . . . , k(cid:96)} · . ♦ Remark 3.8 In [7] Okounkov and Pandharipande established a relation be- tween Gromov-Witten invariants of CP1 and the completed cycles. There- fore the result of Proposition 3.7 was to be expected. Note, however, that Gromov-Witten invariants are only intersection numbers, while here we are working with cohomology classes. Therefore we can expect to get more in- formation than what is contained in the completed cycles. And indeed, our expressions for ψm involve terms that do not appear in the completed cycles and do not contribute to the computation of Gromov-Witten invariants. 3.2 Products of completed cycles and the classes αsδm1,...,ms Proposition 3.9 ([6]) The product of two stable central elements is a finite linear combination of stable central elements. Example 3.10 We have C 2 of all transpositions, so C 2 2 = C2,2 + 3C3 + 1 2C1,1. Indeed, C2 is the sum 2 is the sum of products of all possible pairs of 21 (cid:88) (cid:96) (cid:88) (cid:80) ki=m−(cid:96)+2 k1,...,k(cid:96) 1 (cid:96)! Xm = (cid:96)(cid:89) i=1 m! (m − (cid:96) + 2)! kixi. transpositions. A product of two transpositions can give either a permutation with two disjoint 2-cycles (in a unique way if we number the 2-cycles in the same order as the transpositions), or a 3-cycle (and every 3-cycle can be decomposed into a product of two transpositions in three possible ways), or an identity permutation with two marked elements (but we can number these marked element in two different ways). The proof in the general case is a simple generalization of the above example and is left to the reader. Proposition 3.11 Let α(cid:96) ∈ H top(M0;(cid:96)+1) be the class Poincar´e dual to a point. Choose m1, . . . , ms and k1, . . . , k(cid:96) such that 2s +(cid:80) mj = (cid:96) +(cid:80) ki. Then the coefficient of α(cid:96)ik1,...,k(cid:96) in the expression of αsδm1,...,ms in terms of singularity classes equals the coefficient of the stable central element Ck1,...,k(cid:96) in the product C m1+1 ··· C ms+1. Proof. Accroding to Theorem 3, the coefficient of α(cid:96)ik1,...,k(cid:96) in the expres- sion for αsδm1,...,ms is equal to the coefficient of the monomial xk1 ··· xk(cid:96) in the product of polynomials Xm1 ··· Xms. Recall that the polynomials Xm are defined as Let us call (cid:96) +(cid:80) ki the order of the stable central element Ck1,...,k(cid:96). Then They are transformed into the genus 0 part of the completed cycles C m if we replace every monomial xk1 ··· xk(cid:96) by the stable central element Ck1,...,k(cid:96). the genus g terms of a completed cycle C m have order m + 2 − 2g. The genus 0 elements have the biggest possible order m + 2. Consider two stable central elements Ck1,...,k(cid:96) and Cr1,...,rs. Denote their orders by d1 and d2. Then we have Ck1,...,k(cid:96) · Cr1,...,rs = Ck1,...,k(cid:96),r1,...,rs + lower order terms, where the order of the first term equals d1 + d2. We conclude that if we only keep the highest order terms, then stable central elements multiply like monomials: (xk1 ··· xk(cid:96)) · (xr1 ··· xrs) = xk1 ··· xk(cid:96)xr1 ··· xrs. This is enough to prove the second assertion of Theorem 3. Indeed, we have already identified the highest order terms of a completed cycle C m+1 with the coefficients of the polynomial Xm. Therefore the highest order terms of the product C m1+1 ··· C ms+1 are identified with the coefficients of the product of polynomials Xm1 ··· Xms. 22 4 Appendix: sample computations 4.1 Expressions for (cid:81)(rψ − ξ) Theorem 1 leads to the following expressions for(cid:81)(rψ − ξ): • ψ − ξ = a1 • (ψ − ξ)(2ψ − ξ) = a2 + 1 2i1,1 • (ψ − ξ)(2ψ − ξ)(3ψ − ξ) = 2ξi1,1. 6i1,1,1 − 1 a3 + 2i1,2 + 1 • (ψ − ξ)(2ψ − ξ)(3ψ − ξ)(4ψ − ξ) = 24i1,1,1,1 + 2 a4 + 3i1,3 + 2i2,2 + i1,1,2 + 1 • (ψ − ξ)(2ψ − ξ)(3ψ − ξ)(4ψ − ξ)(5ψ − ξ) = 3ψi1,1,1 − ξ(2i1,2 + 1 6i1,1,1) + 1 2ξ2i1,1 2i1,1,3 + 2i1,2,2 + 1 a5 + 4i1,4 + 6i2,3 + 3 24ψi1,1,1,1−ξ(3i1,3 +2i2,2 +i1,1,2 + 1 + 5 − 1 2ξ3i1,1 3i1,1,1,2 + 1 24i1,1,1,1 + 3 120i1,1,1,1,1 + 5ψi1,1,2 2ψi1,1,1)+ξ2(2i1,2 + 1 6i1,1,1) 4.2 Expressions for powers of ψ By taking linear combinations of the equalities of Section 4.1, we obtain the following expressions for the powers of ψ (the underlined terms appear in Okounkov and Pandharipande's completed cycles): 12i2,2 + 1 8i1,3 + 1 72ξ2a2 + 7 30i1,4 + 1 576ψi1,1,1,1 + 137 16ξ2i1,1 + 15 20i2,3 + 1 + 11 • ψ5 = 1 24a4 + 1 216ξi1,1,1 + 85 120a5 + 1 24ψi1,1,2 + 1 6912ξi1,1,1,1 + 11 32ξ3i1,1 + 31 + 1 + 25 + 15 16ξ4a1 + ξ5 24i1,1,2 + 1 576i1,1,1,1 + 1 36ψi1,1,1 + 25 72ξa3 + 11 18ξi1,2 8 ξ3a1 + ξ4 80i1,1,3 + 1 60i1,2,2 + 1 1440ξa4 + 25 96ξi1,3 + 25 360i1,1,1,2 + 1 288ξi1,1,2 144ξi2,2 + 25 14400i1,1,1,1,1 216ξψi1,1,1 + 415 864ξ2a3 + 85 108ξ2i1,2 + 85 1296ξ2i1,1,1 + 575 432ξ3a2 23 • ψ = a1 + ξ • ψ2 = 1 • ψ3 = 1 • ψ4 = 1 2a2 + 1 4i1,1 + 3 2ξa1 + ξ2 6a3 + 1 3i1,2 + 1 36i1,1,1 + 11 12ξa2 + 3 8ξi1,1 + 7 4ξ2a1 + ξ3. 4.3 Basic via singularity classes Using the expressions for the powers of ψ we obtain the expressions for the other basic classes: Codimension 2: • δ0,0 = i1,1 Codimension 3: • δ0,0,0 = i1,1,1 • δ0,1 = i1,2 + ξi1,1 Codimension 4: • δ0,0,0,0 = i1,1,1,1 • δ0,0,1 = i1,1,2 + ξi1,1,1 • ψδ0,0,0 = ψi1,1,1 • δ1,1 = i2,2 + 2ξi1,2 + ξ2i1,1 • δ0,2 = 1 4ψi1,1,1 + 3 2i1,3 + 1 2ξi1,2 + ξ2i1,1 Codimension 5: • δ0,0,0,0,0 = i1,1,1,1,1 • δ0,0,0,1 = i1,1,1,2 + ξi1,1,1,1 • αδ0,0,0,0 = αi1,1,1,1 for any class α ∈ H 2(M0;5) • δ0,1,1 = i1,2,2 + 2ξi1,1,2 + ξ2i1,1,1   • δ0,0,2 = 1 2i1,1,3 + 1 4 + 3 2ξi1,1,2 + ξ2i1,1,1 • ψδ0,0,1 = ψi1,1,2 + ξψi1,1,1 • δ1,2 = 1 4ψi1,1,2 + 1 2i2,3 + 1 2ξi1,3 + 3 2ξi2,2 + 1 4ξψi1,1,1 + 5 2ξ2i1,2 + ξ3i1,1 sing 24 000000   sing + 11 12ξi1,3+ 3 8ξψi1,1,1+ 7 4ξ2i1,2+ • δ0,3 = 1 6i1,4+ 1 3ψi1,1,2+ 1 36 ξ3i1,1 4.4 Singularity via basic classes Inverting the relations of Sections 4.2 and 4.3 we obtain the following expres- sions for the singularity classes in terms of basic classes: Codimension 1: • a1 = ψ − ξ Codimension 2: • i1,1 = δ0,0 • a2 = 2ψ2 − 1 2δ0,0 − 3ξψ + ξ2 Codimension 3: • i1,1,1 = δ0,0,0 • i1,2 = δ0,1 − ξδ0,0 • a3 = 6ψ3 − 2δ0,1 − 1 6δ0,0,0 − 11ξψ2 + 5 2ξδ0,0 + 6ξ2ψ − ξ3 Codimension 4: • i1,1,1,1 = δ0,0,0,0 • i1,1,2 = δ0,0,1 − ξδ0,0,0 • ψi1,1,1 = ψδ0,0,0 • i2,2 = δ1,1 − 2ξδ0,1 + ξ2δ0,0 • i1,3 = 2δ0,2 − 1 2ψδ0,0,0 − 3ξδ0,1 + ξ2δ0,0 • a4 = 24ψ4 − 6δ0,2 − δ1,1 − δ0,0,1 + 13 7 6ξδ0,0,0 + 35ξ2ψ2 − 15 2 ξ2δ0,0 − 10ξ3ψ + ξ4 6 ψδ0,0,0 − 1 24δ0,0,0,0 − 50ξψ3 + 1 5ξδ0,1+ 25 000000 References [1] T. Ekedahl, S. Lando, M. Shapiro, A. Vainshtein. Hurwitz num- bers and intersections on moduli spaces of curves. -- Invent. Math. 146 (2001), no. 2, 297 -- 327. [2] M. Kazaryan, S. Lando. On intersection theory on Hurwitz spaces. (Russian) Izv. Ross. Akad. Nauk Ser. Mat. 68 (2004), no. 5, 91 -- 122; translation in Izv. Math. 68 (2004), no. 5, 935 -- 964. [3] M. Kazarian, S. Lando. Thom polynomials for maps of curves with isolated singularities. -- Tr. Mat. Inst. Steklova, 258 (2007), Anal. i Osob. Ch. 1, 93 -- 106 (Russian) ; translation in Proc. Steklov Inst. Math., 258 (2007), no. 1, arXiv: 0706:1523. [4] M. Kazarian, S. Lando, D. Zvonkine. In preparation. [5] S. Keel. Intersection theory of moduli space of stable n-pointed curves of genus zero. -- Trans. Amer. Math. Soc., 330 (1992), no. 2, 545 -- 574. [6] S. Kerov, G. Olshanski. Polynomial functions on the set of Young diagrams. -- C. R. Acad. Sci. Paris, S´er. I, Math. 319 (1994), no. 2, 121 -- 126. [7] A. Okounkov, R. Pandharipande. Gromov-Witten theory, Hurwitz theory, and completed cycles. Ann. of Math.163 (2006), no. 2, 517 -- 560. [8] R. Thom. Quelques propri´et´es globales des vari´et´es diff´erentiables. -- Comment. Math. Helv., 28 (1954), 17 -- 86. 26
1311.0743
3
1311
2016-06-11T22:46:30
Rational maps from punctual Hilbert schemes of K3 surfaces
[ "math.AG" ]
The purpose of this short note is to study dominant rational maps from punctual Hilbert schemes of length $k>1$ of projective K3 surfaces $S$ containing infinitely many rational curves. Precisely, we prove that their image is necessarily rationally connected if this rational map is not generically finite. As an application, we simplify the proof of C. Voisin's of the fact that symplectic involutions of any projective K3 surface $S$ act trivially on $\mathrm{CH}_0(S)$.
math.AG
math
RATIONAL MAPS FROM PUNCTUAL HILBERT SCHEMES OF K3 SURFACES by Hsueh-Yung Lin Résumé. -- The purpose of this short note is to study dominant rational maps from punctual Hilbert schemes of length k ≥ 2 of projective K3 surfaces. Precisely, we prove that their image is necessarily rationally connected if the rational map is not generically finite. As an application, we simplify C. Voisin's proof of the fact that symplectic involutions of any projective K3 surface S act trivially on CH0(S). 1 Introduction In this note, we will work throughout over the field of complex numbers C. Recall that a K3 surface S is by definition a smooth projective surface with trivial canonical bundle KS = Ω2 S and vanishing H1(S, OS). The Hilbert scheme of zero-dimensional subschemes of length k ≥ 2 on the K3 surface S will be denoted by S[k]. Recall that a proper variety X is said to be uniruled (resp. rationally connected) if a general point x ∈ X (resp. two general points x, y ∈ X) is contained in the image of a non-constant map P1 → X. These are obviously birationally invariant properties. It is also clear that CH0(X) = Z for any rationally connected variety X. When X is smooth, rational connectedness is equivalent to the a priori weaker condition that two general points can be joined by a chain of rational curves [KMM92, Theorem 2.1]. The following is the main result we obtain in this article : Theorem 1.1. -- If f : S[k] d B is a dominant rational map to a variety B with dim B < dim S[k], then either B is a point or rationally connected. One can compare this theorem with a result of Matsushita [Mat99], showing that for any surjective morphism from an irreducible symplectic manifold to a projective base B of positive dimension, if the general fiber is of positive dimension, then B is Fano. As an application of Theorem 1.1, we will give an alternative proof of C. Voisin's main result in [Voi12]. Theorem 1.2 (Voisin). -- Suppose S is a projective K3 surface and ı is a symplectic involution acting on S, then ı acts as the identity on CH0(S). The motivation for the statement of Theorem 1.2 comes from the following conjecture, which is a consequence of the generalized Bloch conjecture for surfaces [Voi02, Partie VII] : RATIONAL MAPS FROM PUNCTUAL HILBERT SCHEMES OF K3 SURFACES 2 Conjecture 1.3. -- If S is a surface with q = h0,1 = 0 and f : S → S is an automorphism of finite order acting trivially on H0(S, Ω2 S), then the induced map f∗ acts as the identity on CH0(S). A series of examples of surfaces with q = 0 is provided by K3 surfaces S. Such surfaces have one- S) generated by a non-degenerated holomorphic two-form η. An automorphism f : dimensional H0(S, Ω2 S → S such that f ∗η = η is called a symplectic automorphism. A recent new advance of Conjecture 1.3 for K3 surfaces was made by D. Huybrechts and M. Kemeny in [HK13]. They worked with invariant elliptic curves and solved Conjecture 1.3 for K3 surfaces with symplectic involutions f in one of the three series in the classification introduced by van Geemen and Sarti [BvG07]. In [Voi12], C. Voisin showed in general that symplectic involutions act trivially on CH0(S) for any projective K3 surface S. The general statement of Conjecture 1.3 for K3 surfaces is proved soon after in [Huy12] by D. Huybrechts. Theorem 1.4 (Huybrechts, Voisin). -- Let S be a projective K3 surface, η be a non-zero holomorphic two-form on S, and f : S → S be a symplectic automorphism of finite order on S, then f acts trivially on CH0(S). As was shown by Nikulin in [Nik80], the only possible orders of f range from one to eight. In Huybrechts' proof, he studied case by case according to these finitely many possible orders using derived technique and Garbagnati and Sarti's classification results [GS07] on lattices of the invariant part H2(X, Z) f of the action of symplectic automorphism f with prime order. The main construction in Voisin's proof [Voi12] of Theorem 1.2 is the factorization S[] γS Γ∗ CH0(S) P(S,H)( eC /C ) (1.1) where P(S,H)( eC /C ) is the universal Prym variety associated to a complete linear system of curves of genus  in the quotient surface S/ı. Details of the above construction will be given in Section 4. Our main application of Theorem 1.1 is the following Theorem 1.5. -- Any smooth projective compactification of P(S,H)( eC /C ) is rationally connected. The organization of this paper is as follows. In Section 2, we will recall some well-known facts concerning rationally connected varieties and use them to reformulate Theorem 1.1. Then we will prove Theorem 1.1 in Section 3. In Section 4, we will explain how Theorem 1.5 gives an alternative proof of Theorem 1.2. Acknowledgement I would like to thank my supervisor Claire Voisin for helpful discussions, her kindness in sharing her ideas, and her patience during this work. I also thank De-Qi Zhang for the simplification of the original proof of Theorem 1.1. 2 Remarks on rationally connected varieties In this section, we first recall the definition of MRC-fibrations and introduce the main theorem of Graber-Harris-Starr in [GHS03]. Then we will use them to give a reformulation of Theorem 1.1. RATIONAL MAPS FROM PUNCTUAL HILBERT SCHEMES OF K3 SURFACES 3 Recall from [KMM92] that for any variety X, there exists a rational map φ : X d B unique up to birational equivalence characterized by the following properties : (i) a general fiber of φ is rationally connected ; (ii) for a general point b ∈ B, any rational curve passing through Xb := φ−1(b) is actually contained in Xb. The map φ : X d B is called the maximal rationally connected (or MRC for short) fibration of X. The funda- mental question whether the base B of an MRC-fibration is not uniruled remained open for a while [Kol96, Conjecture IV.5.6]. It was finally answered by T. Graber, J. Harris, and J. Starr as a direct corollary of their main theorem in their paper [GHS03] : Theorem 2.1 (Graber-Harris-Starr). -- Let  : X → C be a proper morphism between complex varieties where C is a smooth curve. If the general fiber of  is rationally connected, then  has a section. Remark 2.2. -- This theorem was later generalized by Starr and de Jong to varieties defined over an arbitrary algebraically closed field : any proper morphism from a smooth variety to a smooth curve whose general fibers are smooth and separably rationally connected has a section [?]. In particular, Theorem 2.1 implies : Corollary 2.3 (Graber-Harris-Starr [GHS03]). -- Let  : X d Z be a maximal rationally connected fibration where X is an irreducible variety, then Z is not uniruled. Let us recall for completeness the proof of the above corollary. Proof. -- After resolving the rational map  and singularities of X, we can assume that  is a morphism and X is smooth. Suppose that Z were uniruled, then there exists a rational curve C passing through a general point of Z and we can suppose that  is dominant on C. Denote by eC the normalization of C. Up to replacing X ×Z C by its desingularization, the map C : X ×Z C → C has a section D by Theorem 2.1, which is also a rational curve. Thus if y ∈ D, the rational curve D passes through y and is not contracted by , which is absurd because  : X → Z is an MRC-fibration. (cid:3) Remark 2.4. -- An equivalent formulation of Corollary 2.3 is the following : if f : X d B is a dominant map such that both the general fibers of f and the base B are rationally connected, then X is rationally connected as well [Kol96, Proposition IV.5.6.3]. Thanks to Corollary 2.3, we can easily show that Corollary 2.5. -- In Theorem 1.1, it is equivalent to show that B is either a point or uniruled. Proof. -- Only one direction needs to be proved. Assume that Theorem 1.1 is true when replacing "rationally connected" with "uniruled". Suppose B is not a point ; let B d B′ be an MRC fibration of B. Since B′ is not uniruled by Corollary 2.3, we deduce that B′ is a point by assumption. Hence B is rationally connected. (cid:3) Let us first prove the following general result. 3 Proof of Theorem 1.1 Lemma 3.1. -- Let f : X d B be a rational dominant map between smooth projective varieties such that dim X > dim B. If D is an ample divisor on X, then the restriction map fD is still dominant provided B is not uniruled. RATIONAL MAPS FROM PUNCTUAL HILBERT SCHEMES OF K3 SURFACES 4 Remark 3.2. -- The following example shows that the hypothesis of B not being uniruled is essential in the above lemma. Let l be a line in P2 and p a point in P2 which is not contained in l. If f : P2 d l is the projection from p, then f is dominant and every line in P2 containing p is contracted to a point under f . Remark 3.3. -- Before we start the proof, we remark that any birational modification of the rational map f : S[k] d B and the base B will not affect the hypotheses and the conclusion in Theorem 1.1. For instance up to desingularization of B, we can always suppose that B is smooth and complete. This kind of modification will be repeatedly used in the proof of Theorem 1.1. Proof of Lemma 3.1. -- Let f : eX → B be a resolution of f after a sequence of blow-ups π : eX → X of X. Denote by eD the proper transform of D and D′ := π−1(D). Since D is ample and [D′] = π∗[D] in Pic(eX), one deduces that D′ is a big divisor. Suppose k = dim X − dim B and denoted by H ∈ NS(eX) the class of an ample divisor on eX. As f is dominant, the class f ∗[x] · Hk−1 ∈ H2(X, R), where x is a point of B, is a class of a movable curve. So [D′] · f ∗[x] · Hk−1 > 0 [BDPP13, Corollary 2.5], hence the restriction fD′ is dominant. Now let y be a very general point of B such that there is no rational curve passing through y. We know hand, this fiber is connected and rationally connected. As there is no rational curve passing through y, that there exists x ∈ D′ such that y = f (x). As π(x) ∈ D, the fiber Ex := π−1(π(x)) intersects eD. On the other f −1(y) ∩ eD , ∅, so feD (hence fD) is the fiber Ex is contracted to y by f . As Ex meets eD, we conclude that dominant. (cid:3) Proof of Theorem 1.1. -- Assuming B is not uniruled, by Corollary 2.5 it suffices to show that B is a point. Using the natural map Sk d S[k], instead of dealing with S[k] d B it is equivalent to look at the maps Sk d B which are invariant under the action of Sk on Sk by permutation. Let D be a rational curve lying in an ample linear system of S (whose existence is due to Bogomolov- Mumford [SM83]). Since O(D)⊠k is ample on Sk, we deduce by Lemma 3.1 that the restriction of f to ∪k i=1Si−1 × D × Sk−i (with S0 × D × Sk−1 = D × Sk−1 and Sk−1 × D × S0 = Sk−1 × D) is dominant. As this union is finite, we can suppose without loss of generality that the restriction to D × Sk−1 of f is dominant. (In fact, since f is symmetric, this is always the case.) Since D × Sk−1 d B is dominant, for a general point z ∈ Sk−1 the rational curve D × z is contracted to a point by f (whenever defined). Up to birational equivalence of B, we can suppose that D × z is contracted to a point for every point z ∈ Sk−1 by Remark 3.3. Since D × z is ample in S × z, the fibers of the restriction map fz := fS×z have positive dimension (which is a consequence of Hodge index theorem). So either S × z is contracted to a curve Cz or to a point. Next, suppose that S × z is contracted to a curve Cz. Since 0 = q(S) ≥ q(Cz), Cz is necessarily a rational curve. Set U := [ S×z contracted to a curve S × z ⊂ S × Sk−1. Since by assumption B is not uniruled, the restriction of f to U is not dominant. Therefore up to birational modification of B, we can suppose that f contracts S × z to a point for any z ∈ Sk−1. As f is symmetric, we deduce that for all 0 < i ≤ k, the map f contracts z × S × z′ to a point for any z ∈ Si−1 and z′ ∈ Sk−i. Therefore the image of f is a point, and we are done. (cid:3) RATIONAL MAPS FROM PUNCTUAL HILBERT SCHEMES OF K3 SURFACES 5 4 Triviality of symplectic involution actions on CH0(S) The aim of this section is to give an alternative proof of Theorem 1.2 that symplectic involutions of a K3 surface S act as the identity map on CH0(S) using Theorem 1.1. Let ı be a symplectic involution of S and let Γ = ∆S − Γı ∈ CH2(S × S) where Γı ∈ S × S is the graph of ı. Before we start the proof, let us recall the factorization of Γ∗ : S[] → CH0(S) constructed by Voisin in [Voi12], which is used in an essential way both in Voisin's proof and ours. 4.1 Prym varieties and a factorization of Γ∗ : S[] → CH0(S) Let π : eC → C be an étale double cover of a smooth curve C and consider the involution i : eC → eC that interchanges the preimages of any point p ∈ C. This involution i induces an endomorphism on the Jacobian of eC denoted i∗ : J(eC) → J(eC), and the Prym variety of eC → C is defined as PrymeC/C = Im (Id − i∗) = Ker (Id + i∗)◦ , where for any algebraic group G, G◦ denotes the connected component containing the identity of G. It is an abelian variety carrying a principal polarization and it is easy to see that PrymeC/C is also isomorphic to Ker (π∗)◦, where π∗ is the norm map π∗ : J(eC) → J(C). Using that π∗ is surjective and the Riemann-Hurwitz formula, we deduce that dim PrymeC/C Now let Σ := S/ı be the (singular) quotient surface of S by the involution ı. Choose a very ample line bundle H ∈ Pic(Σ) and assume that c1(H)2 = 2 − 2. Since the canonical line bundle KΣ is trivial, the genus of smooth curves in H is  and h0(H) =  + 1. =  − 1 where  is the genus of C. Let U ⊂ S[] be a Zariski open subset parametrizing reduced subschemes s = s1 + . . . + s ∈ S[] such that there exists a unique smooth curve Cs in H passing through the image of each si in Σ. Since Cs is a smooth curve, its inverse image eCs ⊂ S is smooth, connected, and is an étale cover of Cs, which contains s1, . . . , s. One notices that Γ∗(s) = Pi ([si] − ı∗[si]) ∈ CH0(S) does not depend on albeCs (Pi ([si] − ı∗[si])) ∈ PrymeCs/Cs thus we obtain the following factorization of ΓU∗ : U → CH0(S) : , ΓU ∗ CH0(S) p U γ P(S,H)( eC /C ) (4.1) where C → U′ ⊂ H (resp. eC → U′) is the universal smooth curve over the Zariski open set U′ of H parametrizing smooth curves (resp. universal family of double coverings over U′), P(S,H)( eC /C ) is the corresponding universal Prym varieties over U′, and γ is defined as γ(s) = albeCs  X i ([si] − ı∗[si]) . 4.2 Proof of Theorem 1.2 With the same notations introduced in the last paragraph, let us first make the factorization 4.2 more precise. Let I ∈ CH2(S[] × S) be the class of the incidence correspondence. Notice that Γ∗ : S[] → CH0(S) factors through Γ∗ : CH0(S[]) → CH0(S), RATIONAL MAPS FROM PUNCTUAL HILBERT SCHEMES OF K3 SURFACES 6 where Γ := I − (IdS[], ı)(I) ∈ CH2(S[] × S). Let P be a smooth compactification of P(S,H)( eC /C ). Let p S[] V γ q P (4.2) be a resolution of γ defined in the previous paragraph. Since Chow groups of zero-cycles are invariant under birational modifications, the rational map γ defines canonically the pushfoward map by γ∗ := q∗p∗ : CH0(S[]) → CH0(P). Lemma 4.1. -- There exists a codimension 2 correspondence Γ′ ∈ CH2(P × S) such that ∗ ◦ γ∗ = Γ∗ : CH0(S[]) → CH0(S). Γ′ Proof. -- From the definition of γ, it suffices to show that there exists Γ′ ∈ CH2(P × S) such that the morphism p : P(S,H)( eC /C ) → CH0(S) introduced in 4.2 factors through Γ′ ∗ : CH0(P) → CH0(S). Let D be the restriction to P(S,H)( eC /C ) ×U eC of the universal Poincaré divisor (unique up to linear equivalence) in J ac(cid:16) eC /U(cid:17). The inclusions of each fiber of eC → U in S define a map φ : P(S,H)( eC /C )×U eC → P(S,H)( eC /C )×S. factorization of p : P(S,H)( eC /C ) → CH0(S) by construction. Let Γ′ be the closure of φ(D) in P × S, then the induced correspondence Γ′ ∗ : CH0(P) → CH0(S) gives a (cid:3) From the existence of the rational map γ : S[] d P and Theorem 1.1, one now deduces Corollary 4.2. -- CH0(cid:16)P(cid:17) is isomorphic to Z. Proof. -- One has the dominant rational map γ : S[] d P with dim S[] = 2 > 2 − 1 = dim P. Hence P is rationally connected by Theorem 1.1, so CH0(P) ≃ Z. (cid:3) Corollary 4.3. -- The morphism Γ∗ : CH0(S[]) → CH0(S) is identically zero. Proof. -- Using Lemma 4.1, Γ∗ factors through CH0(cid:16)P(cid:17). As CH0(cid:16)P(cid:17) ≃ Z by Corollary 4.2, Γ∗[z] is inde- pendent of z ∈ S[] where [z] again denotes the class of z in CH0(cid:16)S[](cid:17). By choosing for z an ı-invariant zero-cycle, we conclude that Γ∗[z] = 0 for any z. (cid:3) Proof of Theorem 1.2. -- Corollary 4.3 implies Theorem 1.2 by the following factorization Γ∗ S[] CH0(S) CH0(S[]) γS ∗ CH0(P) using the fact that for a point z of S[] corresponding to a subscheme Z of S, the classes [z] ∈ CH0(S[]) and [Z] ∈ CH0(S) satisfy Γ∗[z] = [Z] − ı∗[Z]. (cid:3) Remark 4.4. -- It is tempting to ask whether one could apply this method to symplectic automorphisms σ of arbitrary finite order d > 2 instead of symplectic involutions. Unfortunately, this method fails to RATIONAL MAPS FROM PUNCTUAL HILBERT SCHEMES OF K3 SURFACES 7 generalize for dimension reasons. Indeed, choosing a very ample line bundle H on the quotient K3 surface S/( f ) such that c1(H)2 = 2 − 2, then exactly as above, for a general point s = (s1, . . . , s) ∈ S there exists a unique smooth curve Cs ∈ H of genus  such that its inverse image eCs in S contains s1, . . . , s. Taking one could again construct the factorization S[] γS Γ∗(s) = X (cid:0)[si] − f∗[si](cid:1) , i Γ∗ P(S,H)( eC /C ) CH0(S) p with this time, the fiber of the universal Prym variety P(S,H)( eC /C ) → U over a smooth curve C ∈ U is the Prym variety of the étale cyclic covering π : eC → C induced by the quotient map S → S/( f ) and is defined by = Im (Id − σ∗) = Ker(cid:16)Id + σ∗ + · · · + σd−1 ∗ (cid:17)◦ , PrymeC/C where σ∗ : J(eC) → J(eC) is the induced map on the Jacobian. Hence dim P(S,H)( eC /C ) = (d − 1)( − 1) +  ≥ 2 = dim S[], so we cannot apply Theorem 1.1. References [BDPP13] S. Boucksom, J.-P. Demailly, M. Paun, and T. Peternell. The pseudo-effective cone of a compact Kähler manifold and varieties of negative Kodaira dimension. J. of algebraic geom., 22 :201 -- 248, 2013. [BvG07] A. Sarti B. van Geemen. Nikulin involutions on K3 surfaces. Math. Z., 255(4) :731 -- 753, 2007. [GHS03] T. Graber, J. Harris, and J. Starr. Families of rationally connected varieties. J. Amer. Math. Soc., 16 :57 -- 67, 2003. [GS07] A. Garbagnati and A. Sarti. Symplectic auomorphisms of prime order. J. Algebra, 218 :323 -- 350, 2007. [HK13] D. Huybrechts and M. Kemeny. Stable maps and chow groups. Documenta mathematica, 18 :507 -- 517, 2013. [Huy12] D. Huybrechts. Symplectic automorphisms of K3 surfaces of arbitrary order. arXiv :1205.3433, 2012. [KMM92] J. Kollár, Y. Miyaoka, and S. Mori. Rationally connected varieties. J. of Algebraic geom., 1 :429 -- 448, 1992. [Kol96] J. Kollár. Rational curves on algebraic varieties, volume 32 of Ergebnisse der Math. Springer-Verlag, Berlin, 1996. [Mat99] D. Matsushita. On fibre space structures of projective irreducible symplectic manifold. Topology, 38 :79 -- 83, 1999. [Nik80] V. Nikulin. Finite groups of automorphisms of Kähler K3 surfaces. Proc. Moscow Math. Society, 38 :71 -- 135, 1980. [SM83] S. Mukai S. Mori. Mumford's theorem on curves on K3 surfaces. Ann. Math., 61 :197 -- 278, 1983. [Voi02] C. Voisin. Théorie de Hodge et géométrie algébrique complexe, volume 10 of Cours spécialisés. Société Mathématique de France, 2002. [Voi12] C. Voisin. Symplectic involutions of K3 surfaces act trivially on CH0. Documenta Math., 17 :851 -- 860, 2012. Hsueh-Yung Lin, Centre de Mathématiques Laurent Schwartz, 91128 Palaiseau Cédex, France
1706.09862
4
1706
2018-11-26T21:07:59
Topologically singular points in the moduli space of Riemann surfaces
[ "math.AG" ]
In 1962 E. H. Rauch established the existence of points in the moduli space of Riemann surfaces not having a neighbourhood homeomorphic to a ball. These points are called here topologically singular. We give a different proof of the results of Rauch and also determine the topologically singular and non-singular points in the branch locus of some equisymmetric families of Riemann surfaces.
math.AG
math
Noname manuscript No. (will be inserted by the editor) Topologically singular points in the moduli space of Riemann surfaces Antonio F. Costa · Ana M. Porto To our friend Mar´ıa Teresa Lozano Received: date / Accepted: date Abstract In 1962 E. H. Rauch found the points in the moduli space of Rie- mann surfaces not having a neighbourhood homeomorphic to a ball. These points are called here topologically singular. We give a different proof of some of the results of Rauch and also determine the topologically singular points in the branch locus of some equisymmetric families of Riemann surfaces. Keywords Riemann surface · Moduli space · Orbifold · Teichmuller space Mathematics Subject Classification (2010) 32G15 · 14H15 · 30F10 · 30F60 1 Introduction Let M be a manifold and p : M → N be a regular branched covering; N has then a structure of (good) orbifold. The set of singular values of p is called the branch locus and it is the image by p of the fixed points of automorphisms of the covering p; it consists of both ordinary manifold points and of points we call topologically singular points, meaning that they do not admit a neighbourhood Authors partially supported by the project MTM2014-55812-P of Ministerio de Economia y Competitividad (Spain) Antonio F. Costa Departamento de Matem´aticas Fundamentales, Facultad de Ciencias, UNED, Senda del rey, 9, 28040 Madrid, Spain Tel.: +34 91 3987224, E-mail: [email protected] Ana M. Porto Departamento de Matem´aticas Fundamentales, Facultad de Ciencias, UNED, Senda del rey, 9, 28040 Madrid, Spain Tel.: +34 91 3987233, E-mail: [email protected] 8 1 0 2 v o N 6 2 ] . G A h t a m [ 4 v 2 6 8 9 0 . 6 0 7 1 : v i X r a 2 Antonio F. Costa, Ana M. Porto homeomorphic to a ball. Note that all the points outside the branch locus are manifold points. We shall assume, all over this paper, that g is an integer ≥ 2. The moduli space Mg of surfaces of genus g is endowed with the structure of an orbifold given by the Teichmuller space Tg and the action of the mapping class group that produces a covering Tg → Mg. In [9] Rauch proves, that for g > 3, every point in the branch locus Bg of Mg is topologically singular, the branch loci B2 and B3 containing topologically non-singular and singular points. In this article, we present a topological proof of these results. Finally, we study some equisymmetric families of dimension 4, showing how both topologically singular and non-singular points appear in the branch loci of moduli spaces of these families. Acknowledgements. We wish to thank the referees for comments and sug- gestions. 2 Preliminaries 2.1 Uniformization of Riemann surfaces and automorphisms using Fuchsian groups A Fuchsian group ∆ is a discrete subgroup of PSL(2, R), i.e. the group Isom+(H2) of direct isometries of H2. If H2/∆ is compact, the algebraic structure of ∆ is given by the signature s = (h; m1, ..., mr), where h is the genus of the quotient surface H2/∆ and the mi are the branched indices of the covering H2 → H2/∆ (the order of the isotropy groups of the conic points of the orbifold H2/∆). The group ∆ admits a canonical presentation: Dai, bi; i = 1, ..., h; xj; j = 1, ..., r : x1...xrQh[aibi] = xmj j = 1E We shall consider only compact Riemann surfaces. A Riemann surface X of genus g > 1, may be uniformized by a surface Fuchsian group, i.e. X = H2/Γ , where Γ is a Fuchsian group with signature (g; ), surface group of genus g. The group Γ is isomorphic to the fundamental group of X. When g > 1, the group of automorphisms of the Riemann surface X is a finite group Aut(X). If G ≤ Aut(X) the quotient orbifold X/G is isomorphic to H2/∆, where ∆ is a Fuchsian group containing Γ and such that ∆/Γ ∼= G. If we have an n-fold covering X = H2/Γ → H2/∆, where Γ is a surface genus g Fuchsian group and ∆ has signature (h; m1, ..., mr), the following Riemann-Hurwitz formula holds: 2g − 2 = n(2h − 2 + Xr (1 − 1 mj )) Topologically singular points in the moduli space of Riemann surfaces 3 2.2 Teichmuller and moduli spaces Let G be an abstract group isomorphic to a Fuchsian group with signature s. Two representations α1 and α2 of G in P SL(2, R) are equivalent if there is γ ∈ PSL(2, R) such that α1(ζ) = γα2(ζ)γ−1, for all ζ ∈ G. The Teichmuller space Ts is the space of equivalence classes of representations ρ of G in PSL(2, R) such that ρ(G) is a Fuchsian group with signature s. This space with the topology induced by PSL(2, R) is homeomorphic to a ball of dimension dim T(h;m1,...,mr) = 6h − 6 + 2r Note that the group G is isomorphic to the orbifold fundamental group of H2/ρ(G), where [ρ] ∈ Ts. If s = (g; ) the corresponding Teichmuller space is noted Tg. Let Modg be the mapping class group of surfaces of genus g. The group Modg acts by composition on Tg and the quotient Tg/Modg = Mg is the moduli space of Riemann surfaces of genus g. Note that Mg is, by construction, an orbifold and its universal covering is Π : Tg → Mg. The set of branch values of the covering Π is the branch locus Bg of the orbifold Mg. The branch locus Bg is the image by Π of the fixed points by finite subgroups of Modg and represents in Mg the surfaces with non-trivial automorphism group (up to the exception of M2, since B2 consists of surfaces having non-trivial automorphisms different from the hyperelliptic involution). Let θ : G → G be an epimorphism from the abstract group G isomorphic to a Fuchsian group with signature s = (h; m1, ..., mr) and such that ker θ is isomorphic to a surface group of genus g. There is a natural embedding iθ : Ts → Tg. The image Π(iθ(Ts)) ⊂ Mg consists of the surfaces of genus g having a subgroup of their automorphism groups isomorphic to G with a specific action determined by θ. We say that Π(iθ(Ts)) is the moduli space of an equisymmetric family given by θ. If we consider the subset Ss,θ of Π(iθ(Ts)) consisting of the surfaces whose full automorphism group is G, we obtain a stratification of Bg by the sets Ss,θ which are called the equisymmetric strata (see [2]). 3 Topologically singular points in moduli space Definition 1 (Topological singular point) A point X in Bg is topologically singular if X has not a neighbourhood in Mg homeomorphic to a (6g − 6) ball. In other words: Definition 2 (Rauch definition [9]) A point X in Bg is singular if X is not a manifold (or uniformizable) point in Mg. Theorem 1 The group Aut(X), for X in the branch loci of Bg, acts as a subgroup of O(6g − 6) in S6g−7. The point X is topologically non-singular if and only if S6g−7/Aut(X) is homeomorphic to S6g−7. 4 Antonio F. Costa, Ana M. Porto Proof Let X ∈ Bg and Y ∈ Π −1(X) ⊂ Tg. Since Modg acts discontinuously on Tg, there is a (6g − 6) ball U ⊂ Tg, with center Y , such that h ∈ Modg satisfies the condition h(U ) ∩ U 6= ∅ if and only if h fixes Y . An element h of Modg such that h(U )∩U 6= ∅ is given by an automorphism of the Riemann surface represented by Y in Tg, consequently by X in Mg, and we may identify h with an element (still called h) of Aut(X). Under such identification, Aut(X) acts on the ball U and it follows, since h is an isometry of Tg (with the Teichmuller metric), that h acts as an isometry on both U and ∂U (∼= S6g−7); therefore, Aut(X) acts as a subgroup G of O(6g − 6). Now Π(U ) = U/Aut(X) = U/G is a neighbourhood of X, so X is non- singular whenever U/G is homeomorphic to the (6g − 6) ball or, equivalently, whenever Π(∂U ) ∼= S6g−7/G is the sphere S6g−7. The following theorem will be used to produce our proof of the main The- orem of [9]. Theorem 2 Let X ∈ Bg. If for each equisymmetric stratum S such that X ∈ S, the codimension of S is greater than 2, then X is topologically singular. Proof Let Π : Tg → Mg be the covering given by the action of Modg. Let S the set of equisymmetric strata S such that X ∈ S. Let U be a ball containing a point of Π −1(X) and such that Π(U ) does not cut Bg but on the strata in S. We shall show that if codim(S) > 2 for all S ∈ S then U/Aut(X) is not a ball. We consider the covering p = Π Π−1(∂U): ∂U → Π(∂U ) with branch locus Π(∂U ) ∩ Bg = Π(∂U ) ∩ (∪S S). There is a finite triangulation of U in such a way that Fix(Aut(X)) is a subpolyhedron and the action of Aut(X) preserves the triangulation (note that Aut(X) acts as a finite order rotation group of O(6g − 6)). Since codim(S) > 2, for all S ∈ S, the codimension of the polyhedron Π(∂U ) ∩ Bg = Π(∂U ) ∩ (∪S S) is greater than 2 in Π(∂U ). If Π(∂U ) is a manifold then π1(Π(∂U ) − Π(∂U ) ∩ Bg) ∼= π1(Π(∂U )) (see [6] Theorem 2.3, page 146). If g > 2, the covering p has Aut(X) 6= 1 sheets (in case g = 2 the covering p has Aut(X)/ hhi 6= 1 sheets, where h is the hyperelliptic involution) and p is branched on Π(∂U ) ∩ Bg. In both cases π1(Π(∂U ) − Π(∂U ) ∩ Bg) must be not trivial. Hence either π1(Π(∂U )) 6= 1 or Π(∂U ) is not a manifold and, in both cases, Π(∂U ) cannot be homeomorphic to the sphere S6g−7; therefore, X is topologically singular. Corollary 1 Let X ∈ Bg and suppose that X is isolated in Bg (see [8]) then X is topologically singular. Proof If X is isolated then {X} is the unique equisymmetric stratum S such that X ∈ S and {X} has dimension 0 (codimension greater than 2). Remark 1 Note that if X is an isolated point of Bg and U is a ball in Tg containing a point of Π −1(X) such that Π(U ) ∩ Bg = {X}, the covering ∂U → Π(∂U ) is a regular unbranched covering with deck transformation group Aut(X), so Π(∂U )) is a manifold and π1(Π(∂U )) ∼= Aut(X). Topologically singular points in the moduli space of Riemann surfaces 5 Theorem 3 For g ≥ 4, every point in Bg is topologically singular. Proof Note that every stratum of Bg is contained in the closure Sk of some equisymmetric stratum defined by the action of a prime order automorphism (for instance see [3], [4], [5]). We shall then study the dimension of such strata. Note that in some cases the full group of automorphisms of the surfaces in Sk may contain strictly the cyclic group generated by the prime order automor- phism. Assume that Sk is the closure of an equisymmetric stratum of Bg defined by a cyclic subgroup Ck of Modg of prime order k such that the corresponding action on surfaces has r fixed points. Let h be the genus of all the quotients of the surfaces in Sk by the action of the group Ck. The Riemann-Hurwitz formula gives: Since k ≥ 2 then 2g − 2 = k(2h − 2 + (Xr (1 − 1 k )) 2g − 2 ≥ 2(2h − 2 + r/2) and r ≤ 2g − 4h + 2. Hence the dimension of Sk is smaller than 4g − 2h − 2. Supposing the codimension of Sk is two, we have 6g − 8 = 4g − 2h − 2 yielding g = 2 or 3, which contradicts our hypothesis. We thus conclude that the codimension of Sk is greater than 2 and Theorem 2 completes the proof. Remark 2 The proof in [9] of the Theorem 3 is based on a Theorem of Zariski [12]. Let us now study the topologically singular points of the moduli space of surfaces genus 2 and 3. We shall say that an equisymmetric stratum S is maximal if S 6⊂ S ′ for every equisymmetric stratum S ′ with S 6= S ′. We need the following corollary of Theorem 2: Corollary 2 Let X ∈ Bg and suppose that X ∈ S where S is a maximal equisymmetric stratum of codimension greater than 2. Then X is topologically singular. Proof Let X ∈ S where S is a maximal equisymmetric stratum of codimension greater than 2. By Theorem 2, each point Y ∈ S is singular since Y is the unique stratum containing Y . Since X ∈ S we have X is singular. First we consider the case g = 3. Theorem 4 The points of B3 corresponding to surfaces having an automor- phism different from the hyperelliptic involution are topologically singular, while the points of B3 corresponding to surfaces having only the hyperellip- tic involution are non-singular. 6 Antonio F. Costa, Ana M. Porto Proof The orders of prime order automorphisms of surfaces of genus 3 are 2, 3 and 7 (see [1]). Let S(i) k be the equisymmetric strata corresponding to genus 3 surfaces where Ck acts in a determined topological way, k = 2, 3, 7. Let us denote S the hyperelliptic locus. Every point in B3 is in the closure of some S(i) k . (1) 2 Order 2: there are three topological types of automorphisms. Type A: the hyperelliptic involution with 8 fixed points and quotient of genus 0. The stratum S(1) corresponding to this topological type of action has dimension 6 × 0 − 6 + 2 × 8 = 10 (codimension two in M3). Each point in S(1) has a neighbourhood that is homeomorphic to the quotient space of a ball by the action of an order two rotation with fixed point set of codimension 2, then the points in S(1) are not topologically singular. 2 2 2 Type B: the involution has 4 fixed points and the quotient has genus 1. In this case the stratum S(2) 2 has dimension 6 × 1 − 6 + 2 × 4 = 8 (codimension > 2). Or the signature (1; 2, 2, 2, 2) is maximal (see for instance [5]), then this stratum is maximal. By Corollary 2 all the points in S (2) 2 are singular. Type C: for this type of automorphisms there are no fixed points and the quotient surface has genus 2. This stratum is not maximal since all Riemann surfaces of genus three with a fixed point free involution are hyperelliptic and as well admit an involution with four fixed points (the signature of Fuchsian groups determining this stratum is not maximal, see [5]). Hence S(3) is con- tained in the closure of S(2) and all its points are singular. 2 2 Order 3: there are two topological types of automorphisms of order three. Type A: two fixed points and genus of quotient 1. This stratum S(1) is not (2) 2 , so all the points corresponding to surfaces 3 maximal and it is contained in S with this type of action are singular. Type B: five fixed points and genus of quotient 0. The dimension of this is is 6 × 0 − 6 + 10 = 4 (codimension > 2). The stratum S(2) stratum S(2) maximal and all the points in the closure of S(2) are singular. 3 3 3 Order 7: There are two points: K and W . The point K corresponds to the Klein quartic and the point W to the Wiman's curve of type I in genus 3 (i) (see [11]). The surface K admits order three automorphisms, then it is in S 3 (1) and K is a singular point. The point W is in S 2 with Aut(W ) = C14 and it corresponds to the curve of equation y2 = x7 − 1. If U is a neighbourhood of W ′, where W ′ is in the preimage of W in Teichmuller space, and Aut(X) = C14 = (cid:10)t : t14 = 1(cid:11), we have that ∂U/(cid:10)t2(cid:11) is the sphere or ∂U/(cid:10)t2(cid:11) → ∂U/ hti is an unbranched cyclic 7-fold covering; then ∂U/ hti is not simply connected and W is topologically singular. Finally, we must consider the points in S k . These points correspond to hyperelliptic surfaces having an automorphism of order (1) 4 that is a square root of the hyperellipticity. Let S4 ⊂ S 2 be the stratum corresponding to surfaces with full automorphism group C4. The codimension but not in any S(i) (1) 2 Topologically singular points in the moduli space of Riemann surfaces 7 of S4 is greater than two (dim S4 = 6×0−6+2×5 = 4). Let X be a point in S4 and Y ∈ Π −1(X). Let U be a neighbourhood of Y in Tg where Aut(X) = C4 = (cid:10)t : t4 = 1(cid:11) acts. We have that ∂U/(cid:10)t2(cid:11) is the sphere but ∂U/(cid:10)t2(cid:11) → ∂U/ hti is a 2-fold covering branched on a subpolyhedra of codimension > 2; then, by a similar argument to the one in the proof of Theorem 2, we prove that ∂U/ hti is not simply connected and X is thus topologically singular. Hence all the points in S4 are singular. Finally we consider the case g = 2. In this case, using our approach we cannot give a complete description of the topological singular points in B2. Theorem 5 The points in the stratum S2 of B2 corresponding to surfaces having full automorphism group C2 × C2 are not topologically singular. The isolated point of B2 corresponding to the Kulkarni surface y2 = x5 − 1 is singular. The Kulkarni surface y2 = x5 − 1 is also known as Wiman's curve of type I in genus 2 (see [11]). Proof First note that dim M2 = 6 × 2 − 6 = 6. By Theorem 1, the points of S2 have a neighbourhood U that is homeo- morphic to the quotient of a ball B by a rotation of order two having as fixed point set a linear subspace of codimension two (intersection with B). Hence, U is a ball and the points in S2 are not singular. The stratum S5 (surfaces with an order 5 automorphism) has a single point and it is an isolated point of B2 (see [8]). By Corollary 1 this point is singular. Remark 3 The points in B2 different from Kulkarni isolated surface and the surfaces in S2 are in strata completely included in S2 (then these strata are non-maximal) and our methods do not provide information on the singularity of such points. 4 Singular points of equisymmetric families In the case of (real) dimension two equisymmetric families, as for all two dimensional orbifolds, the points in the branch locus are not topologically singular. We shall show that in families of greater dimensions the points in the singular locus may be either topologically singular or non-singular. In that direction, let us study the singular points of some equisymetric families of (real) dimension four. Example 1 The families Wq,w consist of Riemann surfaces that are q × w−fold cyclic coverings of the sphere branched on five points, where q, w > 5 are prime integers q 6= w. The type of coverings defining the families Wq,w will be described now. 8 Antonio F. Costa, Ana M. Porto Let T(0;q,q,q,qw,w) be the Teichmuller space of groups ∆ with signature (0; q, q, q, qw, w), and hxi, i = 1, ..., 5 : xq i = 1, i = 1, 2, 3, xqw 4 = xw 5 = 1 and x1...x5 = 1i be a canonical presentation for these Fuchsian groups. The surfaces of the family Wq,w are uniformized by the surface groups in the kernel of the epi- morphism θ : ∆ → Cqw = hl : lqw = 1i, given by θ(xi) = lw, i = 1, 2, 3 and θ(x4) = l−3wlq, θ(x5) = l−q. The inclusion ker θ ⊂ ∆ induces an embedding e : T(0;q,q,q,qw,w) → Tg and the moduli space of Wq,s is Π(e(T(0;q,q,q,qw,w))). The branch locus of the family consists of one point X with isotropy group C3. This point corresponds to the case where the group ∆ is inside the tri- angular group (0; q, 3qw, 3w). The point X is singular and the boundary of a neighbourhood of X is homeomorphic to the lens space L(3, 1) (see a survey on lens spaces in [10], there, precisely, the lens spaces are studied as quotient singularities). Example 2 The family Q consists of Riemann surfaces that are q-fold (where q > 5 is a prime) cyclic coverings of the sphere branched on five points. The q-cyclic coverings defining the family have some special types which we shall describe in terms of Fuchsian groups. Let T (0;q, 5...,q) be the Teichmuller space of groups ∆ with signature (0; q, 5..., q), and hxi, i = 1, ..., 5 : xq i = 1, i = 1, ..., 5 and x1...x5 = 1i q−3 be a canonical presentation for these Fuchsian groups. The family Q has di- mension four, the surfaces of the family are uniformized by the surface groups in the kernel of the epimorphism θ : ∆ → Cq = hl : lq = 1i, given by θ(xi) = l, i = 1, ..., 3 and θ(x4) = θ(x5) = l 2 . The inclusion ker θ ⊂ ∆ induces an em- bedding e : T(0;q,q,q,q,q) → Tg and the moduli space of Q is Π(e(T (0;q, 5...,q))). The branch locus of the family consists of a dimension two subset L corre- sponding to cone points with isotropy group of order 2 and one point Y with isotropy group D3. The points in L correspond to Fuchsian groups ∆ in (0;q, 5...,q) contained in Fuchsian groups Λ with signature (0; 2, q, q, 2q) and the T point Y corresponds to the case where the group ∆ is inside the triangular group (0; 2, 2q, 3q). The points in L have a neighbourhood U such that the boundary is the tridimensional sphere, since the covering Π(e(T (0;q, 5...,q)) ∩ Π −1(∂U )) → ∂U is given by the quotient of a rotation around a trivial knot. The singular point Y has also a neighbourhood V whose boundary is homeomorphic to S3, because the intersection ∂V ∩ L is the trefoil knot and the covering Π(e(T (0;q, 5...,q)) ∩ Π −1(∂U )) → ∂U is the universal covering of the orbifold defined on S3 with singular orbifold locus the trefoil knot with isotropy group C2 for the points in the branch locus. Topologically singular points in the moduli space of Riemann surfaces 9 The covering is equivalent to the composition of coverings S3 3:1→ L(3, 1) 2:1→ S3 (the first covering is unbranched and the second one is the given by the Montesinos involution). Note that in L(3, 1) there are two types of involutions (see [7]): one of them being represented by the Montesinos involution and the other one given by circle action but in this case, the lifts of this involution together with the order three automorphism produce no a dihedral action on S3. We feel very happy to conclude this article in honour of Professor Maite Lozano with this application of the theory of branched coverings of 3−manifolds. References 1. Bartolini, G.; Costa, A. F.; Izquierdo, M.; Porto, A. M. On the connectedness of the branch locus of the moduli space of Riemann surfaces. Rev. R. Acad. Cienc. Exactas F´ıs. Nat. Ser. A Math. RACSAM 104 (2010), no. 1, 81 -- 86. 2. Broughton, S. A. The equisymmetric stratification of the moduli space and the Krull dimension of mapping class groups. Topology Appl. 37 (1990), no. 2, 101 -- 113. 3. Cornalba, M. On the locus of curves with automorphisms. Ann. Mat. Pura Appl. (4) 149 (1987), 135 -- 151. 4. Costa, A. F.; Izquierdo, M. On the connectedness of the branch locus of the moduli space of Riemann surfaces of genus 4. Glasg. Math. J. 52 (2010), no. 2, 401-408. 5. Costa, A. F.; Izquierdo, M.; Porto, A. M. Maximal and Non-maximal NEC and Fuchsian groups uniformizing Klein and Riemann surfaces. In Riemann and Klein Surfaces, Automorphisms, Symmetries and Moduli Spaces, Contemporary Mathe- matics 629, American Mathematical Society, Providence (RI) USA, 2014. DOI: http://dx.doi.org/10.1090/conm/629. 6. Godbillon, C. ´Elements de topologie alg´ebrique, Hermann, Paris 1971. 7. Hodgson, C.; Rubinstein J.H. Involutions and isotopies of Lens Spaces. In Knot Theory and Manifolds: Proceedings of a Conference held in Vancouver, Canada, June 2 -- 4, 1983, Springer Berlin Heidelberg, 1985, 60 -- 96. 8. Kulkarni, R. S. Isolated points in the branch locus of the moduli space of compact Riemann surfaces. Ann. Acad. Sci. Fenn. Ser. A I Math. 16 (1991), no. 1, 71 -- 81. 9. Rauch, H. E. The singularities of the modulus space. Bull. Amer. Math. Soc. 68 (1962) 390 -- 394. 10. Weber, C. Lens spaces among 3-manifolds and quotient surface singularities, Preprint 2017, to appear in RACSAM in this volume. 11. Wiman, A. ber die hyperelliptischen Curven und diejenigen von Geschlechte p=3 Jwelche eindeutige Transformationen in sich zulassen. - Bihang till K. Svenska Vet.-Akad. Handlingar, Stockholm 1895 -- 6, bd. 21, 1 -- 28. 12. Zariski, O. On the purity of the branch locus of algebraic functions. Proc. Nat. Acad. Sci. U.S.A. 44 (1958) 791 -- 796.
1710.01418
1
1710
2017-10-03T23:06:30
Kernels from Compactifications
[ "math.AG" ]
To any affine scheme with a $\mathbb{G}_m$-action, we provide a Bousfield colocalization on the equivariant derived category of modules by constructing, via homotopical methods, an idempotent integral kernel. This endows the equivariant derived category with a canonical semi-orthogonal decomposition. As a special case, we demonstrate that grade-restriction windows appear as a consequence of this construction, giving a new proof of wall-crossing equivalences which works over an arbitrary base. The construction globalizes to yield interesting integral transforms associated to $D$-flips.
math.AG
math
KERNELS FROM COMPACTIFICATIONS MATTHEW R BALLARD, COLIN DIEMER, AND DAVID FAVERO Abstract. Associated to any affine scheme with a Gm-action, we provide a Bousfield colocalization on the equivariant derived cat- egory D(ModGm R) by constructing an idempotent integral kernel using homotopical methods. This endows the equivariant derived category with a canonical semi-orthogonal decomposition. As a special case, we demonstrate that grade-restriction windows ap- pear as a consequence of this construction, giving a new proof of wall-crossing equivalences which works over an arbitrary base. The construction globalizes to yield interesting integral transforms associated to D-flips. 1. Introduction A central question in the study of derived categories of coherent sheaves of algebraic varieties is their relationship with birational ge- ometry. Historically, such investigations originated with Orlov's con- struction of a semi-orthogonal decomposition associated to a blow-up [Orl92], as well as Bondal and Orlov's derived equivalences induced by certain elementary flops [BO95]. Much recent effort has since centered around Bondal and Orlov's conjecture that flops in general induce de- rived equivalences as well as Kawamata's related conjecture [Kaw04] that K-equivalent varieties have equivalent derived categories. Perhaps the most striking result in this direction is Bridgeland's construction of a derived equivalence for any threefold flop [Bri02]. A more recent line of inquiry is the description of the equivariant derived categories of geometric invariant theory (GIT) quotients via so-called grade restric- tion windows, see e.g. [Seg11, BFK12, HHP09, HL15, Bal17, SvdB17, HS16]. These methods sometimes give equivalences or semi-orthogonal decompositions associated to birational maps by viewing the maps as GIT wall-crossings. Nevertheless, a quick survey of the subject will convince an ob- server that there is not an agreed upon uniform approach to producing Key words and phrases. derived geometry, derived categories, equivariant geom- etry, birational geometry. 1 2 BALLARD, DIEMER, AND FAVERO the functors expected by the K-equivalence conjecture. For exam- ple, Bridgeland's techniques require that the flop come from a small contraction over a base of relative dimension one, which limits the applicability to higher dimensional flops. Various families of explicit flops have been considered; notably Namikawa and Kawamata's study of Mukai flops [Nam03, Nam04, Kaw06] and Cautis, Kamnitzer, and Licata's study of the the stratified Mukai flop [CKL12]. In these strati- fied examples a derived equivalence has indeed been observed, but only via fine-tuned choices of explicit Fourier-Mukai kernels. The grade re- striction window techniques mentioned above have so far been most effective only for so-called elementary wall-crossings, or when the ac- tion is specialized to be quasi-symmetric in the language of [SvdB17]. In particular, there does not presently appear to be a consensus in the literature for approaching the following problem: given an arbitrary birational map X ⇢ Y of Mori theoretic origin, provide a uniform method of producing a homologically well-behaved functor between Db(coh X) and Db(coh Y ). In other words, how to systematically pro- duce a Fourier-Mukai kernel object P ∈ Db(coh X × Y ) consistent with the expectations of the Bondal-Orlov and Kawamata conjectures? We summarize the main construction of this paper as follows. Construction. Let Y be a scheme with a trivial Gm-action, A be a quasi-coherent sheaf of Z-graded OY -algebras, and set Z ∶= Spec A. Y ● Form a Z2-graded sheaf Qder(A) of A ⊗OY A-algebras by deriving a certain partial compactification [Dri13] of the action groupoid. ● Realize Qder(A) as an object of D(QcohG2 ● Restrict Qder(A) to open sets to get an equivariant Fourier-Mukai m Z × Z). transform ΦQwc der(A) ∶ D(QcohGm U +) → D(QcohGm U −). where U ± are the corresponding semi-stable loci. A central case of interest is when X ⇢ Y is a flip relative to a divisor D on X, i.e. a D-flip. An observation of Reid, see e.g. [Tha96], allows one to repackage the data of the D-flip as a scheme Z affine over the contraction and carrying an action of Gm, with X and Y the respective GIT quotients, so that the above construction applies. We ask in Ques- tion 5.5.10 if ΦQwc der(A) induces an equivalence for D-flops, a potential solution to [Kaw04, Conjecture 5.1]. However, we presently content ourselves with the following results (see Proposition 5.4.7, Proposi- tion 5.5.9, Theorem 4.2.9, and Corollary 4.2.11 for more precise state- ments). KERNELS FROM COMPACTIFICATIONS 3 Theorem. There is an object Sder ∈ D(QcohG2 orthogonal decomposition m Z × Z) and a semi- D(QcohGm Z) =⟨Im ΦQder(A), Im ΦSder(A)⟩. When Z is smooth and affine then the image of Im ΦQder(A) over U + × Z is equal to the grade restriction window defined in [Seg11, BFK12, HL15]. Hence, when [U +~Gm] and [U −~Gm] are K-equivalent, ΦQwc der(A) is an equivalence. In the smooth case, we need not derive our construction and sim- ply denote this object by Q. Here, the semi-orthogonal decomposition above comes from a certain idempotent property enjoyed by Q which we call Property P, see Definition 3.3.7, which shows that Q induces a Bousfield localization. We remark that when Z is smooth, the proof given here is quite different than those articles as here we produce an explicit geometric kernel, prove functorial identities of that kernel, and deduce these results as corollaries. The proof also works over an arbi- trary base. The essential observation here is that the construction of Q behaves well under strongly ´etale base change (see Proposition 4.2.3) which allows us to reduce to the case of affine space using the Luna Slice Theorem. When Z is singular, it is not the case that the object Q literally enjoys the idempotent Property P mentioned above. This is problem- atic as the case of singular affine varieties equipped with a Gm-action is quite important for the demands of birational geometry (even for the elementary Mukai flop, the corresponding space Z is singular, for example). This is the reason that we must derive Q, i.e. promote Q to a object in derived algebraic geometry. In Section 5 we observe that the functor Q extends to a left Quillen functor on the category of graded simplicial rings, i.e. on derived affine schemes equipped with an action of Gm. Theorem 5.2.7 shows that this derived variant Qder does indeed satisfy an analogue of the idempotent property, which we call Property Pder, and so we still obtain a semi- orthogonal decomposition in analogy with the smooth case. At an intu- itive level, the failure of Q to be well-behaved for singular spaces arises from the non-vanishing of some higher Tor's (see e.g. Lemma 3.3.6), and the homotopical methods mentioned above allow us to bypass this obstruction by encoding the higher Tor's in an intrinsic derived affine scheme. Our actual construction of Q comes from the following geometric consideration: if an algebraic group G acts on a scheme Z, we consider a space Z which equivariantly extends the action and projection maps; 4 BALLARD, DIEMER, AND FAVERO see Definition 3.1.1. Such a construction for Gm-actions was already considered by Drinfeld [Dri13]. To such data, one can always exhibit an associated faithful functor, see e.g. Proposition 3.2.6. Given Drin- feld's construction and the central role of Gm-actions via birational cobordisms, we will likewise focus primarily on this case in this article, although the reader will find a discussion of more general group actions in Section 3. We also remark that at a technical level most of the results in the paper are formulated for affine varieties only, but in Section 5.5 we discuss how to associate sheaves to these constructions in order to extend our results to essentially arbitrary D-flips, as mentioned above. The paper is organized as follows. ● In Section 2 we introduce our main object of study, Q(R), and study its basic properties. Subsection 2.2 goes on to show by example how this object arises in the study of flips and flops. ● In Section 3 we discuss the geometric interpretation of Q in terms of compactifications of group actions; here we also introduce the basic criteria for fully-faithfulness and its relation to Bousfield localizations. ● Section 4 treats the case where Spec R is smooth and exhibits the essential image as a grade restriction window. ● Section 5 then treats the general case by deriving the construc- tion of Q itself. Subsection 5.3 shows that when Spec R is smooth, these derived replacements effectively trivialize down to the meth- ods of Section 4. Section 5.5 discusses the globalization process for attaching a kernel object to a general D-flip. ● In a brief appendix we comment on a small, but important, dis- tinction between our Q(R) and the related constructions in Drin- feld's article [Dri13]. Acknowledgements: This paper began during a meeting with Ludmil Katzarkov and Maxim Kontsevich at the Institut des Hautes ´Etudes Scientifiques. We thank them for their continual support and numerous insights as the project evolved. We also benefited from an illuminating conversation with Dan Halpern-Leistner, which in particular drew our attention to the relevant work of Drinfeld. We thank him for his inter- est. In addition, D. Favero would like to thank Bumsig Kim for helpful conversations exploring non-abelian cases, and M. Ballard would like to thank L. Borisov and D. Orlov for discussions during his visit at the Institute for Advanced Study. M. Ballard was partially supported by NSF Standard Grant DMS-1501813 and is appreciative of the Insti- tute for Advanced Study for furnishing a wonderful work environment KERNELS FROM COMPACTIFICATIONS 5 during his membership. Finally, he would like to express apprecia- tion to both Bob Ballards in his life. D. Favero was partially funded by the Natural Sciences and Engineering Research Council of Canada and Canada Research Chair Program under NSERC RGPIN 04596 and CRC TIER2 229953 and also thanks the Korean Institute for Advanced Study for their hospitality. C. Diemer is supported by a Simons Col- laborative Grant postdoctoral fellowship and acknowledges support by the Laboratory of Mirror Symmetry NRU HSE, RF Government grant, ag. No. 14.641.31.0001, and would like to thank everyone at the IHES for an ideal working environment. 2. Affine constructions 2.1. The functor. Let k be a fixed commutative ring. If R is a Z- graded k-algebra, we consider the two maps π ∶ R → R ⊗k k[u, u−1] ≅ R[u, u−1] σ ∶ R → R ⊗k k[u, u−1] ≅ R[u, u−1] (2.1) (2.2) where π(r) = r is the identity and σ is the (co-)action map determined by σ(r) = rudeg(r) when r is homogenous. In terms of affine schemes these maps respectively correspond to the projection and action maps Gm ×k SpecR π σ SpecR. (2.3) Throughout this article, we will let CRGm k denote the category of finitely generated Z-graded k-algebras, i.e. the opposite category of the cate- gory of affine k-schemes which are equipped with Gm-actions. We will often refer to objects of CRGm as rings with a Gm-action. Likewise, if an algebraic group G acts on Spec R we will say that R has a G-action. When R is an object of CRGm the product space Gm ×k SpecR admits m-action whose structure is given in the following lemma, the proof a G2 of which is elementary and which we omit. k k Lemma 2.1.1. The ring R[u, u−1] carries a natural G2 m-action σ ∶ R → R[u, u−1, v1, v−1 1 , v2, v−1 2 ] uniquely determined by σ(π(r)) = σ1(r) ∈ R[u, u−1, v1, v−1 σ(σ(r)) = σ2(r) ∈ R[u, u−1, v1, v−1 1 , v2, v−1 1 , v2, v−1 2 ] 2 ] where σi ∶ R → R[vi, v−1 Equation (2.2). i ] for i = 1, 2 are both identified with σ from 6 BALLARD, DIEMER, AND FAVERO Moreover, the map π ∶ Gm × Spec R → Spec R becomes equivariant m → Gm onto the first factor, and the m → Gm with respect to the projection π1 ∶ G2 map σ becomes equivariant with respect to the projection π2 ∶ G2 onto the second factor. Remark 2.1.2. The G2 m-action above is equivalent to saying that R[u, u−1] has a Z2-grading with the degree of r ∈ R ⊂ R[u, u−1] with r homogenous being (deg r, 0), the degree of σ(r) being (0, deg r), and the degree of u being (−1, 1). R[u, u−1] → S[u, u−1]. Here we regard R[u, u−1] and S[u, u−1] as ob- jects of CRG2 , one obtains the obvious induced map Given a map R → S in CRGm k m k[u]. m Remark 2.1.3. To stay consistent with Remark 2.1.2, when we write k[u] we require that k[u] is equipped with a Gm-action such that CRG2 deg u = (−1, 1). Thus objects S in CRG2 k[u] → S which respects this grading, and likewise morphisms respect k[u] are equipped with a map m In other words, CRG2 this structure. affine schemes equipped with G2 A1 k. m k[u] is opposite to the category of m-actions and an equivariant map to k[u] denote the functor deter- Definition 2.1.4. Let ∆ ∶ CRGm mined by k m → CRG2 ∆(R) ∶= R[u, u−1], justified by the following. m-action from Lemma 2.1.1. m-action and its two R-module structures where ∆(R) is equipped with the G2 We may view ∆(R) with its G2 m R ⊗k R) via π and σ as an object of the derived category Db(modG2 Z2-graded modules by regarding ∆(R) as the complex concentrated in homological degree zero. The notation ∆(R) introduced above is Lemma 2.1.5. The object ∆(R) ∈ Db(modG2 m R ⊗k R) is the Fourier- Mukai kernel of the identity functor on Db modGm(R). Proof. The Fourier-Mukai transform associated to ∆(R) is given by Φ∆(R)(M) ∶= π∗(∆(R) ⊗L σ∗M)Gm where M ∈ Db modGm(R) and the functors π∗ and σ∗ are derived on the left and on the right respectively. In equation (2.4) we recall that taking a derived-push forward in the equivariant setting takes invariants, see [BKR01, Section 6], which is well-defined since Gm is reductive e.g. (2.4) KERNELS FROM COMPACTIFICATIONS 7 and so its functor of invariants is exact. By Remark 2.1.2, the Gm- invariants referred to in Equation (2.4) mean taking invariants on the left factor in the Z2-grading, i.e. taking terms of degree (0, ∗). Since ∆(R) is flat (via either module structure), we then compute (∆(R) ⊗L σ∗M)Gm =(∆(R) ⊗ σ∗M)Gm =(R[u, u−1] ⊗ σ∗M)(0,∗) ≅ σ∗M so that Φ∆(R)(M) = π∗σ∗M. Likewise, for the inverse Fourier-Mukai transform, we have M ↦ σ∗(∆(R) ⊗L π∗M)Gm ≅ σ∗π∗M. To prove the lemma, we thus need to exhibit the two natural isomor- phisms M ≅ σ∗π∗M M ≅ π∗σ∗M. for any M ∈ Db(modGm R). For M ∈ modGm(R) these isomorphisms are respectively given by composing the maps M → σ∗π∗M m ↦ σM(m) ∈ M[u, u−1] ≅ ∆(R) ⊗R M and M → π∗σ∗M m ↦ m ∈ M[u, u−1] ≅ ∆(R) ⊗R M. and where (∆(R) ⊗ M)Gm ≅ M. The result follows for any M ∈ Db(modGm R) since ∆(R) is again flat via either module structure. (cid:3) The following definition, though simple, is crucial for this article. Definition 2.1.6. Given an object R of CRGm , we define k Q(R) ∶=⟨π(R), σ(R), u⟩ ⊆ R[u, u−1] to be the k-subalgebra of R[u, u−1] generated by u and the images of the co-action and projection maps. We regard Q(R) as an object of CRG2 By construction, the maps π and σ both have images in Q(R). We k[u]. m thus have maps (2.5) R p s Q(R) (2.6) which equal π and σ respectively (in general Q(R) ≠ R[u, u−1], hence we reserve different symbols for these maps). The same construction 8 BALLARD, DIEMER, AND FAVERO the Z-grading on R, we may equivalently write as in Lemma 2.1.1 then shows that Q(R) has G2 Riui, R[u]⟩ Q(R) =⟨ࣷ i<0 m-action. In terms of where R = ࣷi∈Z Ri. Notice that when r ∈ R is homogeneous with non- negative degree, s(r) = rudeg(r) ∈ R[u], and thus, in addition to R[u], only the additional summands Riui with i negative are required to generate. Example 2.1.7. Let R be a polynomial ring equipped with a Gm- action (i.e. a Z-grading). Relabelling variables as necessary, write R = k[x+ 1 , . . . , x+ k , x− 1 , . . . , x− l ] where each x+ i has non-negative (possibly zero) degree and each x− strictly negative degree. Write the degrees as ai = deg(x+ bj = deg(x− i) ≤ 0 and j) < 0. As in Equation (2.7), Q(R) is generated (over k) by j has u, x+ 1 , . . . x+ k , x− 1 , . . . x− l , x− 1 ub1, . . . , x− l ubl. To compress notation, write x+ for the set of x+ Setting y− i 's and similarly for x−. j ubj for j = 1, . . . , l and inspecting the relations, one has j = x− Q(R) = k[u, x+, x−, y− l ). l u−bl − x− i = x+ It is convenient to write this more symmetrically by setting y+ for i = 1, . . . , k as well so that, in compressed notation, l ]~(y− 1 u−b1 − x− 1 , . . . , y− 1 , . . . , y− i uai Q(R) = k[u, x+, x−, y+, y−]~(x+ua − y+, y−u−b − x−) In these variables, the maps s, p ∶ R → Q(R) are given by: (2.7) (2.8) (2.9) (2.10) (2.11) (2.12) = k[u, x+, y−]. p(x+ i) = x+ i) = x+ s(x+ i i uai p(x− s(x− j) = y− j) = y− j u−bj j . We will interpret these maps geometrically in Example 3.1.3. Also, in Subsection 2.2 we will revisit this example (when k = l) while studying the Atiyah flop. Example 2.1.8. Suppose that R is any non-negatively graded ring, i.e. Gm acts on R with non-negative weights. Then Q(R) is actually generated by π(R) and u as a k-algebra. As a module over R via p, we thus have Q(R) ≅ R[u] or, geometrically, Spec Q(R) ≅ A1 If R is instead non-positively graded, then of course Q(R) is not generated by just π(R) and u. However, we may still construct an iso- morphism Q(R) ≅ R[u] where now Q(R) has the R-module structure via s. Indeed, one always has the map R[u] → Q(R) given by r ↦ σ(r) k × Spec R. KERNELS FROM COMPACTIFICATIONS 9 and u ↦ u. This map is always injective, and when the weights are all non-positive, it is easily checked to be surjective as well. For later use we record an elementary property that Q enjoys with respect to polynomial extensions; in particular, the next lemma shows that Spec Q(R[x]) ≅ Spec A1 k × Spec Q(R). Lemma 2.1.9. Given R ∈ CRGm giving x degree a. Then k , endow R[x] with a Gm-action by Q(R)[y] ≅ Q(R[x]) where y has degree (a, 0) if a ≥ 0 and degree (0, a) if a ≤ 0. Proof. Assume the degree of x is a ≤ 0. The case a ≥ 0 is analogous. We have a map (2.13) Q(R)[y] → Q(R[x]) y ↦ uax This map is clearly surjective, and has kernel if and only if uax is algebraic over Q(R). If so, then uax is algebraic over R[u, u−1] in R[x, u, u−1] which is impossible, since u is a unit. The statement about the weight of y is just the weight of uax via the grading from Remark 2.1.2. (cid:3) Remark 2.1.10. The reader desiring a more geometric understanding of Q(R) will find such a discussion in Section 3. Remark 2.1.11. Our definition of Q(R) is actually equivalent to Namely, if R =ࣷ Ri is Z-graded, Drinfeld considers the the k[t]-algebra the affine case of a construction of Drinfeld, see [Dri13, Section 2.3]. R generated by all symbols r where r ∈ R, and which are required to be k-linear and subject to the relations r1 r2 = tµ(n1,n1) r1 r2 where each ri ∈ Rni and where µ(m, n) ∶= min{m,n} to check that the assignment u ↦ t, ri ↦ tniri for i < 0, and ri ↦ ri if m and n have opposite signs, and µ(m, n) = 0 otherwise. It is easy for i ≥ 0 induces an isomorphism Q(R) ≅ R. In the Appendix we will discuss in more detail the role of our Q(R) in the context of Drinfeld's The study of Q(R) as a Fourier-Mukai kernel is, in a sense, the fun- of Q(R) is functorial. damental goal of this article. To this end, we first note that formation article. 10 BALLARD, DIEMER, AND FAVERO Lemma 2.1.12. The assignment of Q(R) to R defines a functor Q ∶ CRGm k → CRG2 m k[u] which preserves surjective morphisms. Proof. Given φ ∶ S → R, the corresponding map S[u, u−1] → R[u, u−1] Q(S) under this map is Q(R). is surjective when φ is. Furthermore, one checks that the image of (cid:3) We can relate the functors Q and ∆ by a natural transformation by considering the inclusions ηR ∶ Q(R) ↪ R[u, u−1] = ∆(R) (2.14) which come from the definition of Q(R) as a subalgebra of ∆(R). These inclusions behave predictably; in particular, we have the follow- ing commutative diagrams: Q(R) ηR p R[u, u−1] π R Q(R) and ηR s R[u, u−1] σ R. (2.15) Definition 2.1.13. Let η ∶ Q → ∆ be the natural transformation of functors induced by the inclusions ηR ∶ Q(R) ↪ ∆(R). Here Q and ∆ m k[u] from Lemma 2.1.12 and Definition 2.1.4 are the functors CRGm respectively. k → CRG2 We will sometimes abuse notation and denote a map ηR by η when it is clear from context that we are working with rings and not the natural transformation. Another way of understanding the relation between Q and ∆ is the following result, which shows in particular that they become identified after localizing by elements of non-zero degree. k Lemma 2.1.14. Let R be an object of CRGm R-module via σ and a left R-module via π. Let r ∈ R be a homogeneous element, and let R → Rr be the corresponding homogeneous localization. . View Q(R) as a right R-module via s and a left R-module via p, and view ∆(R) as a right If deg(r) > 0, then is an isomorphism. If deg(r) < 0, then 1 ⊗s η ∶ Rr ⊗s Q(R) → Rr ⊗σ ∆(R) η p⊗ 1 ∶ Q(R) p⊗ Rr → ∆(R) π⊗ Rr KERNELS FROM COMPACTIFICATIONS 11 is an isomorphism. If deg(r) = 0, then Q(f) p⊗ 1 ∶ Q(R) p⊗ Rr → Q(Rr) ⊗ Rr ≅ Q(Rr) is an isomorphism. Proof. In all three cases, injectivity of the maps holds because localiza- verifying surjectivity in the cases of non-zero degree is to demonstrate that u−1 lies in the image. Indeed, in the case where positive degree tion is flat and η and Q(f) are inclusions. The only non-trivial part of one has that (1 ⊗s η)( 1 r ⊗ udeg(r)−1r) = u−1 and in the case of negative r) = u−1. Surjectivity in degree one has that (η p⊗ 1)(s(r)u−1−deg(r)] ⊗ 1 the degree zero case is clear. (cid:3) We conclude this section with some natural loci coming from a Gm- action. k Definition 2.1.15. If R is an object of CRGm let I ± ⊆ R be the ideals generated by elements of positive or respectively negative degree. Set R+ ∶= R~I −, and R− ∶= R~I + (note the swap in signs), and set R0 ∶= R~(I +, I −). The reason for defining R0 and R± in the above fashion is to match notation with Definition 2.1.16 below, which gives their geometric de- scription. action. Equip Spec(k) with the trivial Gm-action. Let A1 Definition 2.1.16. Let X be any k-scheme equipped with a Gm- + denote − denote A1 A1 equipped with its usual Gm-action by scaling, and let A1 equipped with the inverse action t ⋅ x = t−1x. Then is the fixed point locus for the Gm-action, X 0 ∶= HomGm(Spec(k), X) ={x ∈ X t ⋅ x = x for all t ∈ A1} t ⋅ x exists in X} X + ∶= HomGm(A1 t−1 ⋅ x exists in X} X − ∶= HomGm(A1 +, X) ={x ∈ X lim −, X) ={x ∈ X lim t→∞ t→∞ is the attracting locus, and is the repelling locus. We refer to [Dri13, Section 1] for basic properties of these loci. As alluded to above, when X ∶= Spec R is affine and equipped with a Gm- action, we have and X 0 = Spec R0 X ± = Spec R± = V(I ±) 12 BALLARD, DIEMER, AND FAVERO using the notation of Definition 2.1.15. By construction there are in- clusions R0 ⊂ R± which give maps q± ∶ X ± → X 0. (2.16) stood as being induced by the Gm-equivariant inclusions Spec(k) ↪ A1 In terms of Definition 2.1.16 the maps q± can equivalently be under- ±. Geometrically, this means that the maps q± correspond to taking a point to its limit along either Gm-action or the inverse Gm-action. Remark 2.1.17. The ideals I ± can be recovered directly from Q(R) equipped with its maps p and s. Namely, I + = s−1(uQ(R)) ⊆ R I − = p−1(uQ(R)) ⊆ R. 2.2. Flop equivalences via Q(R). We now observe that Q(R) pro- vides the derived equivalence constructed by Bondal and Orlov for the standard Atiyah flop [BO95, Section 3]. For n ≥ 2, let R = k[x+ 1 , . . . , x+ n, x− 1 , . . . , x− n] i) = 1 and deg(x− with the Gm-action given by taking deg(x+ each i. The ideals I ± ⊂ R from Definition 2.1.15 are I + =(x+ (x+) and I − =(x− 1 , . . . , x− n) =(x−). Let X = Spec R = A2n and set U ± ∶= X ∖ V(I ±) X~~± ∶= U ±~Gm X~~0 ∶= Spec RGm. (2.17) i) = −1 for n) = 1 , . . . , x+ as the fixed locus X 0. Remark 2.2.1. The invariant subring RGm should not be confused with R0 from Definition 2.1.15 where the defining ideal is (I +, I −); in other words, the invariant theory quotient X~~0 is not the same thing For the standard Atiyah flop, X~~±Gm are both isomorphic to the total space of O(−1)⊕n on Pn−1 and X~~0Gm is a singular affine quadric. Let p± ∶ X~~+ Ð→ X~~0 be the corresponding birational contractions. The diagram X~~+ X~~− X~~0 KERNELS FROM COMPACTIFICATIONS 13 p− is the prototypical example of a flop. [BO95, Theorem 3.9] proves that the functor ∗p+∗ ∶ Db(coh X~~+) → Db(coh X~~−). (2.18) is an equivalence of derived categories of coherent sheaves (where the ∗ and p+∗ are, of course, derived on the right and left, respec- functors p− tively). For this example, the actions of Gm on U ± are free, and so we may identity (2.19) and recognize the Bondal-Orlov flop equivalence as an equivalence of equivariant derived categories. Db(coh X~~+) = Db(cohGm U ±) ∗p+∗ is equivalent to taking Remark 2.2.2. Taking the functor to be p− the fiber product along p± as the Fourier-Mukai kernel of the functor. Instead of the fiber product, one could also take as a Fourier-Mukai kernel a common blow-up resolving the birational map. For the Atiyah flop, though, the fiber product and the resolution are isomorphic, see [Kaw04, Proposition 5.5]. We now observe that the fiber product agrees with an appropriate restriction of Q(R). Proposition 2.2.3. With Q(R) as in Equation (2.17), let denote the restriction of Q(R) to the open subset U − ×U + ⊂ X ×X. (The Y wc ∶= Spec Q(R) ×X×X (U + × U −) "wc" stands for "wall-crossing".) Then there is a natural isomorphism (2.20) Y wc ≅ U + ×X~~0 U −. In other words, if we regard Qwc = Γ(Y wc, O) as an object of Db(cohG2 U +), then the Fourier-Mukai functor ΦQwc ∶ Db(cohGm U −) → Db(cohGm U +) is an equivalence and ΦQwc is naturally isomorphic to to p− the identification in Equation (2.19). ∗p+∗ under (2.21) m U −× Proof. The universal property of the fiber product gives a map φ ∶ R ⊗RGm R → Q(R) which is, explicitly, given by r1 ⊗ r2 ↦ r1s(r2), i.e. φ = p ⊗RGm s. Taking k = n and l = n in Example 2.1.7 to describe Q(R), we have: n]. k[x+, x−, y+, y−] (x+y− − y+x−) → k[u, x+, y−, u−1y− We can work on a cover of U + × U − given by inverting the monomials in x1 and y2. One can check that inverting any such pair reduces φ to an 1 , . . . u−1y− φ ∶ (2.22) 14 BALLARD, DIEMER, AND FAVERO isomorphism. The statement about ΦQwc being an equivalence is then just a rephrasing of the Bondal-Orlov equivalence, i.e. Equation (2.18). (cid:3) Example 2.2.4. Let R = k[x1, . . . , xn, y1, . . . , yn]~(x1y1 + . . . + xnyn) Spec R is singular. In Example 3.2.11 we will see that such singularities with n ≥ 2 and with each deg(xi) = 1 and each deg(yj) = −1. Here Spec R~~+ ⇢ Spec R~~− is the elementary Mukai flop. Note that here lead to a poorly behaved Fourier-Mukai functor associated to Q(R), and we will correct this behavior in Example 5.2.3 by replacing Q(R) with a suitable affine derived scheme Qder(R). It is not difficult to show that a suitable generalization of Proposition 2.2.3 holds once this correction is made. 3. Functors from compactifications 3.1. Partial compactifications of group actions. We now address the geometric interpretation of Q(R). In this section, in order to pro- vide context for the connection between Q(R) and geometric invariant theory (GIT), we will give definitions and results for varieties which are not necessarily affine and actions by algebraic groups other than Gm. The reader only concerned with the level of generality required for the later portions of this paper may well assume that G = Gm and X = Spec R is affine throughout this section as well; under these assumptions Proposition 3.1.2 and Proposition 3.1.6 summarize the relevant statements needed from this subsection. Let G be an algebraic group acting on a variety X. Letσ ∶ G×X → X be the action map and π ∶ G × X → X the projection. The space G × X itself admits a G × G action given by (g1, g2) ⋅(g, x) =(g2gg−1 1 ,σ(g1, x)) which makes π and σ equivariant. With this action, the map G × X → X × X given by (g, x) ↦(x,σ(g, x)) becomes equivariant, where X × X has the obvious G × G action. Definition 3.1.1. Let X be an algebraic variety together with an ac- tion of G×G which is equipped with a G×G-equivariant open immersion (3.1) and a G × G-equivariant morphism i ∶ G × X ↪ X, (p,s) ∶ X → X × X such that the following diagram commutes: KERNELS FROM COMPACTIFICATIONS 15 X p s X i σ π G × X of G on X. In the above situation, we say that X i.e. p ○ i = π and s ○ i = σ. equipped with the maps p,s, i, is a partial compactification of the action The following result shows that Q(R) encodes exactly this kind of structure when G = Gm and X is affine. Proposition 3.1.2. Let Spec R be an affine variety with a Gm-action. Consider the maps Spec Q(R) p s Spec R. corresponding to the ring maps p, s ∶ R → Q(R) and let η ∶ Spec Q(R) → Spec R[u, u−1] = Gm × Spec R be the open immersion corresponding to η ∶ Q(R) → R[u, u−1] . Then Spec Q(R) together with the maps p,s,η form a partial compactification of the action of Gm on Spec R. Proof. This follows easily from previous observations: ant structure on Q(R) is determined by the Z2-grading given in Re- mark 2.1.2. The fact that η is an open immersion follows from the fact that it is the localization along u. That Figure 3.1.1 commutes is exactly dual to the commutativity of the diagrams in (2.15). (cid:3) the equivari- Example 3.1.3. Let us revisit Example 2.1.7 where we considered the case of a Gm-action on a polynomial ring, i.e. we now consider Spec(R) = Ak+l where the first k-coordinates are acted on with non- negative weights and the last l-coordinates with strictly negative weights. Following Example 2.1.7, write Ak+l+1 = Spec R[u, x+, y−] = Spec Q(R). The p and s module structures computed in Equations (2.11) and (2.12) determine the maps p and s in the commutative diagram 16 BALLARD, DIEMER, AND FAVERO Ak+l+1 i p s σ π Ak+l Gm × Ak+l where σ and s are the action and projection maps, and i is an inclusion chosen to make the diagram commute. Explicitly: i(u, x+ p(u, x+ s(u, x+ 1 , . . . , x+ 1 , . . . , x+ 1 , . . . , x+ k , x− k , y− k , y− 1 , . . . , x− 1 , . . . , y− 1 , . . . , y− l ) =(u, x+ l ) =(x+ l ) =(ua1x+ 1 , . . . , x+ k , ub1x− 1 , . . . x+ k , u−b1y− 1 , . . . uakx+ 1 , . . . , ublx− 1 , . . . , u−bly− 1 , . . . , y− k , y− l) l ) l ). Note that the map i is not the "naive" inclusion in these coordinates. The data of the partial compactification of an action of G on X auto- matically encodes a notion of a boundary on X as well as distinguished "unstable" and "semistable" loci in X, as in the following definition. Definition 3.1.4. If X is a partial compactification of an action σ ∶ G × X → X, we define the boundary of X to be ∂ X ∶= Xƒi(G × X), X us =s(∂ X), the s-unstable locus to be and the s-semistable locus to be X ss = X ∖ X us. Note that X ss itself admits a G-action as the G × G action on X extends the action on G × X and s was assumed equivariant. In this article we are primarily concerned with the affine group scheme Gm and the case where X = Spec R is affine, in which case ∂ X is a automatically a divisor in X (although X us =s(∂ X) will rarely be a divisor in X). Remark 3.1.5. The terminology in Definition 3.1.4 is chosen to em- phasize the connection with geometric invariant theory (GIT). The next lemma demonstrates the precise relationship between our notion of s-semistability and semistability in GIT. We will also explain the re- lationship for actions by higher rank tori in Example 3.1.7 and Propo- sition 3.1.8. KERNELS FROM COMPACTIFICATIONS 17 Proposition 3.1.6. Let Gm act on X = Spec R and let Spec Q(R) partially compactify the action as in Proposition 3.1.2. Then the s- unstable locus is the repelling locus X − as defined in Definition 2.1.16. In particular, X ss = X ∖ X −. Proof. Since Q(R) =⟨u, π(R), σ(R)⟩, the ideal which defines the bound- ary ∂ ⊆ Spec Q(R) is simply ⟨u⟩ ⊆ Q(R). Hence, s(∂) is given by the ideal ⟨u⟩ ∩ s(R). By the explicit gradings in Remark 2.1.2, it is easy to see that ⟨u⟩ ∩ s(R) = I + is the ideal generated by all homogeneous elements of positive degree, which in turn is the ideal defining X −. (cid:3) It follows immediately from the above proposition that, in the case of a Gm-action on X = Spec R equipped with the partial compactification of the action given by Spec Q(R), one has X ss = X ∖ X − = Spec R ∖ Spec(R~I +) = r∈I + Spec Rr. (3.2) We conclude this subsection with two brief but instructive examples of partial compactifications of actions which go beyond the case of a Gm-action (and thus are not logically necessary for the remainder of this article). In the first example, G is a higher rank torus acting on any affine space. In the second example, G is non-abelian. Example 3.1.7. Let T = Gn denote its character lattice. Suppose that T acts on a ring R with co-action map m be a torus and let χ∗(T) = Hom(T, Gm) n]. Now, let C ⊂ χ∗(T) be any finitely generated submonoid. Let σ ∶ R → R ⊗k k[T] ≅ R ⊗k k[χ∗(T)] ≅ R[x± T(R) =⟨R[C], σ(R)⟩ ⊆ R[χ∗(T)] 1 , . . . , x± QC be the subalgebra generated by the image of the action and the monoid ring. A trivial generalization of Proposition 3.1.2 shows that Spec QC is a partial compactification of the action of T on X = Spec R. T(R) Now, suppose Pic(Spec R) ⊗ Q is trivial. Then, following [Tha96, divisors on X = Spec R is a cone in χ∗(T) ⊗ Q. This cone admits a Section 2] or [DH98, Section 3], the cone of ample T -linearized Q- chamber decomposition such that the relative interiors of the respective chambers correspond exactly to GIT quotients. That is, two characters χ and χ′ lie in the relative interior of the same chamber exactly when the GIT semistable loci X ss(χ) and X ss(χ′) are equal Of particular interest is the case where the monoid C is itself the (integral points of) of a GIT chamber. Although not required for the remainder of the article, it seems worthwhile to observe that a gen- eralization of Proposition 3.1.6 holds in this setting which relates the 18 BALLARD, DIEMER, AND FAVERO notion of semistability given in Definition 3.1.4 with the usual notion of semistability in Geometric Invariant Theory (GIT). Proposition 3.1.8. As above, let a torus T act on a ring R and sup- pose Pic(Spec R) ⊗ Q = 0. Let C be the monoid of integral points of the closure of a GIT chamber, and let L be any line bundle which lies in the relative interior of the same GIT chamber. Then the s-semistable locus in the sense of Definition 3.1.4 equals the semistable locus for the action of T on X equipped with the linearization L in the sense of GIT. monoid C. Thus, Proof. The proof is essentially the same as that of Proposition 3.1.6. In- deed, the ideal defining the boundary of Spec QC T(R) is ⟨χ1 ⊗1, . . . , χk ⊗ 1⟩ where the χi are any set of characters minimally generating the ⟨χ1 ⊗ 1, . . . , χk ⊗ 1⟩ ∩ s(R). However, localizing R at I is the same interior of C, since we have χ =∑ aiχi with each ai > 0. as localizing R at χ ⊗ 1 where χ ∈ C is any character in the relative (cid:3) ideal defining the s-unstable locus in X is I ∶= Remark 3.1.9. As the reader may have already guessed, Example 3.1.7 in particular shows how to realize toric varieties via partial compacti- fications of torus actions on affine space. Namely, if one specializes to the case where R = k[x1, . . . , xn] and so Spec R = An, the above discussion reduces to torus actions on affine space. Here the GIT chambers agree with the cones in the GKZ-fan (see e.g. [CLS11, Section 14.4]), and the GIT quotients are the real- ization of toric varieties via the Cox construction. It is not difficult to give an explicit combinatorial description of a rational polyhedral cone describing the affine toric variety QC T where C is any chamber of the GKZ-fan, although doing so would be too much of a digression, so we leave this as an exercise to the combinatorially minded reader. Example 3.1.10. Let W be a vector space of dimension k and V be a vector space of dimension n with k ≤ n. Set G = Gl(W) and X = Hom(W, V) so that G acts on X by right multiplication. Then X = End(W) ×X is a partial compactification of G ×X. The projection X ss is identified with the set of matrices of full rank, so that X ss~G is the Grassmannian Gr(n, k). and multiplication maps extend naturally to maps X → X, and here 3.2. Functors from partial compactifications. We now show how to associate functors between derived categories given the data of a partial compactification of an action. This is essentially done by taking the partial compactification itself as a Fourier-Mukai kernel. KERNELS FROM COMPACTIFICATIONS 19 Definition 3.2.1. Given X a partial compactification of a G-action on X in the sense of Definition 3.1.1, define QX,G ∶=(p ×s)∗O X ∈ Db(QcohG×GX × X), (3.3) which is simply the derived push-forward of the structure sheaf under the extended action and projection maps. Remark 3.2.2. If X is affine and G = Gm then, since the functor (p × s)∗ is exact, Proposition 3.1.2 shows that Q(R) with its p-s- bimodule structure corresponds to the equivariant sheaf QSpec R,Gm . However, as suggested by the examples of flops from Section 2.2, we as would obtained by taking QX,G as a kernel object. Rather, we are are not literally interested in endofunctors D(QcohGX) → D(QcohGX) interested in functors D(QcohGX ss) → D(QcohGX) where X ss is the semistable locus from Definition 3.1.4. Definition 3.2.3. Let X be a partial compactification of a G-action on X. Then Qss X,G denotes the quasi-coherent sheaf on X ss ×X obtained by restricting QX,G from X × X. That is, Qss X,G =(j × Id)∗Q (3.4) where j ∶ X ss → X is the inclusion. For the purposes of the paper, we are primarily concerned with the case of a Gm-action on an affine variety, for which we reserve different notation as follows. the quasi-coherent sheaf obtained by restricting the sheaf associated to Definition 3.2.4. If X = Spec R has a Gm-action, then Q+(R) denotes Q(R) to the quasi-affine variety X ss × X = X ∖ X − × X. We will frequently abuse notation and drop the implicit reference to R and just write Q+, and will also use Q+ ∈ Db(QcohGm×Gm X ss × X) to denote the corresponding object of the derived category. Taking Q+ as a Fourier-Mukai kernel, we have the functor ΦQ+ ∶ D(QcohGm X ss) → D(QcohGm X). Remark 3.2.5. For an arbitrary ring R, Q is a module, hence a bounded complex. Since it may not be perfect in general, tensor prod- uct with Q may not preserve boundedness unless R has finite Tor di- mension. When X = Spec R is smooth, we will establish a "cohomolog- ical properness" result in Proposition 4.1.5 which in particular implies that the essential image lands in the derived category of complexes (3.5) (3.6) 20 BALLARD, DIEMER, AND FAVERO with bounded and coherent coherent sheaves. We leave cohomological properness of ΦQ beyond the smooth case for a fuller discussion. We now show that this functor is automatically faithful. Proposition 3.2.6. Let Gm act on X = Spec R. Let Q+ be as in Equa- tion (3.5). Then the Fourier-Mukai functor ΦQ+ from Equation (3.6) is faithful. Proof. This essentially follows from Lemma 2.1.14. In more detail, by the open affine cover of X ss given in Equation (3.2), one obtains the obvious cover of X ss × X. Lemma 2.1.14 says exactly that Q restricts to ∆ on each open affine subset of this cover, which by Lemma 2.1.5 is the Fourier-Mukai kernel of the identity functor. If denotes the inclusion and j∗ ∶ D(Qcoh X) → D(Qcoh X ss) the restric- tion functor, we thus have that j∗ ○ΦQ+ = Id, and the result follows. (cid:3) j ∶ X ss → X (3.7) Remark 3.2.7. It is not much more difficult to prove a strengthened version of Proposition 3.2.6 valid for any kernel Qss X,G from Defini- tion 3.2.3. Indeed, the other main result from this Section, Propo- sition 3.3.9, also admits such a generalization as well. Since we do not require such generality for the remainder of the article, though, we omit these generalizations for the sake of brevity. Fullness of the functor ΦQ+ is more subtle and addressing this is, in a sense, the main technical content of Section 3.3 (when Spec R is smooth) and Section 5.1 (in general). We will see in Lemma 3.3.6 that fullness of the functor ΦQ+ is intimately related to properties of the tensor product Q(R) s⊗p Q(R), which we now study. The tensor product Q(R) s⊗p Q(R) inherits a G3 m-action on Q(R) ⊗k Q(R). Explicitly, let (a, b) ∈ Z2 denote the weight of a homogenous element of r ∈ Q(R). Then we have the fol- lowing weights on homogenous elements of Q(R) s⊗p Q(R): m-action from the G4 The G3 m-action is such that the map Spec Q(R) ×Spec R Spec Q(R) → Spec R × Spec R corresponding to p ⊗ s is equivariant for the projection G3 the first and third factor. The next lemma relates Q(R) s⊗p Q(R) to Q(R) directly after taking suitable invariants. In what follows, let m → G2 m onto deg(r ⊗ 1) =(a, b, 0) deg(1 ⊗ r) =(0, a, b). (3.8) (3.9) KERNELS FROM COMPACTIFICATIONS 21 Z ⊂ Z3 be the inclusion into the middle factor. If M is a Z3-graded module, we let (M)0 denote the Z2-graded sub-module obtained by map of Z3-graded modules, (f0) ∶ (M)0 → (N)0 is the corresponding taking degree zero in the middle factor. Likewise, if f ∶ M → N is a restriction. Lemma 3.2.8. We have a commutative diagram (1⊗η)0 (Q(R) s⊗π ∆(R))0 (Q(R) s⊗p Q(R))0 ∼ ∼ Q(R) (η⊗1)0 (∆(R) σ⊗p Q(R))0 inside ∆(R) s⊗π ∆(R) which happens to land in Q(R). Proof. Both the top and bottom map take a ⊗b to the degree zero piece (cid:3) Definition 3.2.9. Following Lemma 3.2.8, we set ρR to be the map ρR ∶(Q(R) s⊗p Q(R))0 → Q(R). given by either (1 ⊗ η)0 or equivalently (η ⊗ 1)0. (3.10) In very specific situations, the map ρR may be an isomorphism. Lemma 3.2.10. Let R be an object of CRGm and assume that either the weights of R are all non-positive or all non-negative. Then ρR is an isomorphism. k Proof. Assume for simplicity that all the weights are non-negative; the proof for non-positive weights is similar. Using Lemma 3.2.8, it suffices is exact. To demonstrate surjectivity, we have to exhibit elements of to show that (η ⊗ 1)0 is an isomorphism. By Example 2.1.8, Q(R) ≅ R[u] so that Q(R) is flat over R via p. We first note that injectivity of (η ⊗1)0 then follows from flatness and because the functor of invariants Q(R) s⊗p Q(R) of middle degree zero that map to σ(r), π(r), and u. They are, respectively, 1 ⊗ σ(r), π(r) ⊗ 1, and u ⊗ u. Note that in the above lemma, one has Q(R) s⊗p Q(R) ≅ R[u, v], and the gradings from Remark 2.1.2 immediately imply that (Q(R) s⊗p Q(R))0 ≅ Q(R) as R-bimodules. The above lemma shows that ρR indeed implements this isomorphism. (cid:3) Example 3.2.11. However, the map ρR is not an isomorphism in gen- eral. An elementary counterexample is R = k[x, y]~(xy) where [x] = 1 and [y] = −1. Letting z = yu−1, one has Q(R) ≅ k[x, z, u]~(xz) 22 and BALLARD, DIEMER, AND FAVERO Q(R) ⊗R Q(R) ≅ k[x, z′, u, u′]~(xz′uu′). The element x ⊗ z′ has middle degree 0 in Q(R) ⊗R Q(R) and is sent to 0 in Q(R) under ρR. In Section 5.2 we will remedy such a fail- ure by deriving Q, and will revisit this particular example again in Example 5.2.3. Remark 3.2.12. Note that in the above example Spec R is singular. We will see in Lemma 4.2.5 that ρR is an isomorphism whenever Spec R is smooth. Remark 3.2.13. Intuitively, the property of ρR being an isomorphism is similar to the characterization of derived open immersions as being ⊗A B ≅ B, see e.g. In particular, this suggests we should derive finitely-presented ring maps f ∶ A → B such that B [TV08, Lemma 2.1.6]. the tensor product in Q(R) s⊗p Q(R) to obtain a fully-faithful functor. When Spec R is smooth, we will see in Proposition 4.2.6 that this tensor product is automatically derived (i.e. its higher Tor's vanish). For singular cases we will implicitly derive this tensor product during the course of Section 5.1 by deriving the Q functor itself. L 3.3. Bousfield localizations. In this section we address the fullness of the functor ΦQ+ from Equation (3.6). Indeed, we will show more and in Proposition 3.3.9 exhibit a semi-orthogonal decomposition of D(QcohGm Spec R) such that ΦQ+ gives the inclusion of one of the fac- tors. This semi-orthogonal decomposition will come from a Bousfield localization. Definition 3.3.1. Let T be a triangulated category. A Bousfield lo- calization is an exact endofunctor L ∶ T → T equipped with a natural transformation δ ∶ IdT → L such that: a) Lδ = δL and b) Lδ ∶ L → L2 is invertible. If instead we have an endofunctor C ∶ T → T equipped with a natural transformation ǫ ∶ C → 1 such that a) Cǫ = ǫC and b) Cǫ ∶ C 2 → C is invertible, then one calls C a Bousfield colocalization. We refer to [Kra12, Section 4] for background on Bousfield localizations and colocalizations for triangulated categories. KERNELS FROM COMPACTIFICATIONS 23 m R ⊗k R) and δ ∶ P → P ′ is any Remark 3.3.2. If P, P ′ ∈ Db(modG2 map, then one can easily check that ΦP(δ)(A) = δ(ΦP(A)) for any A ∈ D(ModGm R). This means that the first condition for being a Bousfield localization or colocalization is automatically satisfied in the setting of Fourier-Mukai functors, provided has a morphism δ ∶ ∆ → P or ǫ ∶ P → ∆ (recall ∆ is the kernel of the equivariant identity functor). Definition 3.3.3. Suppose we have maps of endofunctors C ǫ Ð→ IdT δ Ð→ L of a triangulated category T such that Cx ǫCx ÐÐ→ x δx Ð→ Lx is an exact triangle for any object x. Then C → IdT → L is called a a Bousfield triangle for T if any of the following equivalent conditions are satisfied: (i) L is Bousfield localization and C(ǫx) = ǫCx, (ii) C is a Bousfield colocalization and L(δx) = δLx, (iii) L is Bousfield localization and C is a Bousfield colocalization. Let us prove the equivalence of the three conditions in the definition. Proof. i) ⇒ ii): Suppose L is Bousfield localization. Set x = Ly. Then we get a triangle CLy ǫLy Ð→ Ly δLy ÐÐ→ L2y and the map Ly → L2y is an isomorphism. Therefore, CLy = 0. Now, consider the triangle C 2y C(ǫy) ÐÐÐ→ Cy C(δy) ÐÐÐ→ CLy. Since CLy = 0 the first map is an isomorphism as desired. ii) ⇒ i) is by symmetry. As i) is equivalent to ii), it is obvious that they are both equivalent to iii). (cid:3) Lemma 3.3.4. Let C → IdT → L be a Bousfield triangle for a triangu- lated category T . Then there is a weak semi-orthogonal decomposition T =⟨Im L, Im C⟩. (3.11) Here Im denotes the essential image. Proof. By definition, for any object x of T , we have a triangle Cx → x → Lx in T . Let f ∶ Cx → Ly be any map. We have a commutative diagram 24 BALLARD, DIEMER, AND FAVERO Cf CLy f Ly C 2x ∼ Cx with the left vertical map an isomorphism since C is a Bousfield colo- calization. Since Ly → L2y is also an isomorphism, its (co)cone CLy must be 0. Thus, up to composition with an isomorphism, f = 0. (cid:3) Lemma 3.3.5. Let C1 triangles for a triangulated category T such that L1C2 isomorphism. Then there is a weak semi-orthogonal decomposition δ2 Ð→ L2 be Bousfield L1(ǫ2) ÐÐÐ→ L1 is an ǫ2 Ð→ 1 ǫ1 Ð→ 1 δ1 Ð→ L1 and C2 T =⟨Im C2 ○ L1, Im C2 ○ C1, Im L2⟩. This induces a fully-faithful functor F ∶ T~ Im C1 → T . Proof. For the first statement, by Lemma 3.3.4 we only need to fur- ther decompose Im Q in Equation (3.11). From Lemma 3.3.4 and Lemma 3.3.8, we know that for any object M ∈ T there is exact triangle C2(C1(M)) → C2(M) → C2(L1(M)) so we only need to check semi-orthogonality. Let N ∈ T be another object and be any morphism. Using the semi-orthogonality of Im C2 and Im L2, f corresponds uniquely to a map f ∶ C2(C1(M)) → C2(L1(N)) g ∶ C2(C1(M)) → L1(N). We have a commutative diagram C2(C1(M)) g L1(C2(C1(M))) L1g L1(N) ∼ L2 1(N) Next, we note that L1(C2(C1(M))) ≅ L1(C1(M)) ≅ 0 by assumption since L1C1 = 0. KERNELS FROM COMPACTIFICATIONS 25 Hence g = 0 and therefore f = 0. For the second part of the statement, apply [Orl09, Lemma 1.4] (see also Proposition 4.9.1 of [Kra12]) to the case D = Im C2, N = Im C2C1. This gives an equivalence Now apply [Orl09, Lemma 1.1] with D = Im C2, D′ = T , N = Im C2C1, and N ′ = Im C1. This gives an equivalence Im C2L1 ≅ Im C2~ Im C2C1. Im C2~ Im C2C1 ≅ T~ Im C1. T~ ImC1 ≅ Im C2L1. Tracing through these equivalences, we have an equivalence (cid:3) (3.12) which induces the fully-faithful functor F . We now show that, under certain conditions, the exact triangle Q ηR Ð→ ∆ → cone ηR → Q[1] in Db(modGm R ⊗k R) yields a Bousfield triangle for the associated Fourier-Mukai transforms. Lemma 3.3.6. Let R be an object of CRGm functors k ΦQ η Ð→ Id → Φcone ηR . Then the triangle of is a Bousfield triangle if b) TorR a) The map ρR ∶(Q(R) s⊗p Q(R))0 → Q(R) is an isomorphism, and i (Q(R)p, Q(R))s) = 0 for all i > 0, where the subscripts on Q(R) denote the R-module structures given by p or s respectively. Proof. By Remark 3.3.2, the functors form a Bousfield triangle if and only if (Q L Q(η) ÐÐ→ Q s⊗p Q)0 i (Q(R)p, Q(R))s) = 0 for all i > 0, (Q is an isomorphism. Since TorR Q)0 = (Q s⊗p Q)0. Therefore, the map Q(η) is just ρR, which is an isomorphism by assumption. (cid:3) L s⊗p Definition 3.3.7. If R is an object of CRGm is such that conditions a) and b) appearing in Lemma 3.3.6 are both satisfied, we say that R has Property P. k In particular, Lemma 3.3.4 says that if R has Property P, then there is a semi-orthogonal decomposition D(QcohGm Spec R) =⟨Im ΦQ, Im Φcone η⟩. (3.13) 26 BALLARD, DIEMER, AND FAVERO We can further refine this semi-orthogonal decomposition using Lemma 3.3.5 by considering Bousfield (co)localizations coming from local cohomology. Let Γ+ denote the local cohomology of X = Spec R along V(I +) = X −. Lemma 3.3.8. There is a Bousfield triangle Γ+ → Id → J+ for Db(QcohGm Spec R) where and j ∶ Spec R ∖ V(I +) → Spec R is the inclusion. J+ ∶= j∗ ○ j∗ (3.14) (3.15) Proof. This is standard. See Example 1.2 of [HR17] which applies to algebraic stacks and, in particular, our situation. (cid:3) We then refine our semi-orthogonal decomposition from Equation (3.13) as follows. Proposition 3.3.9. Let R be an object of CRGm Then there is a semi-orthogonal decomposition k which has Property P. Furthermore, the functor D(QcohGm Spec R) =⟨Im ΦQ ○ J+, Im ΦQ ○ Γ+, Im Φcone η⟩. ∶ D(QcohGm Spec RƒV(I +)) → D(QcohGm Spec R) ΦQ+ is fully-faithful. Proof. We note that J+ ○ ΦQ = J+ since inverting any r ∈ R of posi- tive weight trivializes Q by Lemma 2.1.14. Therefore, we may apply Lemma 3.3.5 to obtain the result, noting that the map F in that lemma is exactly ΦQ+ in this case. (cid:3) 4. The smooth case We now study the faithful functor ΦQ+ ∶ Db(cohGm X −) → Db(QcohGm X) (4.1) from Equation (3.6) in the case where X = Spec R is a smooth affine scheme with Gm-action. As mentioned in the introduction, the smooth- ness hypothesis is unreasonably restrictive for the demands of birational geometry, although it does subsume many simple examples and, not surprisingly, permits some dramatic simplifications compared to the general case that we will pursue in Section 5.2. KERNELS FROM COMPACTIFICATIONS 27 4.1. Affine space. We first consider the case where X = An; here we study ΦQ+ via direct calculations. We begin by showing in two lemmas that X = An (equipped with any weights) has Property P, i.e. satisfies the hypotheses of Lemma 3.3.6. Lemma 4.1.1. Let R = k[x1 . . . , xn] where the xi are equipped with any degrees. Then the map ρR from Definition 3.2.9 is an isomorphism. Proof. This can be verified by an explicit calculation using the gradings on Q(R) s⊗p Q(R) from Equations (3.8) and (3.9). However, let us We induct on n. First, if n = 1, i.e. R = k[x1], then the statement instead give a quick proof more in the spirit of arguments we will use later in Section 5.2. of the Lemma follows directly from Lemma 3.2.10. (Alternately, note that the statement for n = 0 is trivial). We then need to show that for any object S of CRGm , if ρS is an isomorphism then ρS[x] is also an isomorphism (where x is given any weight). Indeed, by Lemma 2.1.9 k and it's easily verified that Q(S[x]) ⊗ Q(S[x]) ≅ Q(S)[y] ⊗ Q(S)[y] (Q(S)[y] ⊗ Q(S)[y])0 ≅(Q(S) ⊗ Q(S))0[y]. Then, using Lemma 2.1.9 again, (Q(S) ⊗ Q(S))0[y] = Q(S)[y] = Q(S[x]). (cid:3) Lemma 4.1.2. Let R = k[x1, . . . , xn] where the xi are equipped with any degrees. Then TorR i (Q(R)p, Q(R)s) = 0 for i > 0. Proof. In the notation of Example 2.1.7, we have where Q(R) = k[x+, x−, y+, y−, u]~⟨y+ − uax+, x− − u−by−⟩ which exhibits Q(R) as a complete intersection ring inside R ⊗k R[u] Since R ⊗k R[u] is a flat R ⊗k R-module, the Koszul resolution of Q(R) as R ⊗k R[u]-module gives a flat resolution of Q(R)s as an R ⊗k R = k[x+, x−, y+, y−]. R⊗kR[u](y+ − uax+, x− − u−by−). K● R-module. Hence, Q(R)p ⊗ L R Q(R)s = Q(R)p ⊗R K ● R⊗kR[u](y+ − uax+, x− − u−by−) = K ● Q(R)⊗kR[u](y+ − uavas+, s− − u−by−). 28 BALLARD, DIEMER, AND FAVERO where we identify Q(R) ⊗k R[u] = k[s+, s−, y+, y−, u, v]. Since y+ i certain variables), this complex has no higher homology. is a regular sequence (it just solves out (cid:3) − uaivais+ − u−biy− i i , s− i Combining Lemmas 4.1.1 and 4.1.2 with Proposition 3.3.9 immediately gives the following. Proposition 4.1.3. Let Gm act on An = Spec R. Then ΦQ+ ∶ Db(cohGm An ∖ V(I +)) → Db(QcohGm An) is fully-faithful. We now study the essential image of ΦQ+ in the affine space case and show that it coincides with the subcategory generated by weights in the interval (µ, 0] where −µ is the sum of the positive weights of the Gm-action. To this end, we first establish a useful set of related generating objects. Lemma 4.1.4. Equip R = k[x1, . . . , xn] with any Gm-action and set µ = − Qdeg xj >0 deg xj. (4.2) i with respect to the Gm-action. Then Db(cohGm An ∖ V(I +)) is generated by objects of the form j∗ O(i) where i ∈ N is in (µ, 0]. Here j ∶ An ∖ V(I +)) → An is the inclusion and R(i) denotes structure sheaf of An = Spec R equipped with weight Proof. The Koszul complex on{xj deg xj > 0} is acyclic on An ∖V(I +), so the object j∗ O(µ) is generated by such objects. Similarly, tensoring the Koszul complex by some j∗ O(t), we get that j∗ O(µ+t) is generated by those O(i) with µ + t < i ≤ t. By induction, we can thus generate any j∗ O(k) with k ≤ 0 by the objects j∗ O(i) with i ∈(µ, 0]. Similarly, we can generate an object of the form j∗ O(i) with i > 0 by twisting the Koszul complex by j∗ O(i) and peeling off the top term. (cid:3) Proposition 4.1.5. With notation as above, the essential image of the equivariant Fourier-Mukai functor ΦQ+ ∶ Db(cohGm An ∖ V(I +)) → Db(QcohGm An) is the full subcategory generated by those R(i) such that i ∈(µ, 0]. KERNELS FROM COMPACTIFICATIONS 29 objects, thus proving the Proposition. Proof. Since ΦQ+ is fully-faithful by Proposition 4.1.3, Lemma 4.1.4 tells us that the essential image of ΦQ+ is equivalent to the full sub- category of Db Qcoh(An) generated by the objects ΦQ+(j∗ O(i)) with Indeed, we will show that ΦQ+(j∗ O(i)) = R(i) for such i ∈ (µ, 0]. With notation as in Example 2.1.7, write R = k[x+, x−] and Q(R) = R ⊗k R[u]~⟨y+ − uax+, x− − u−by−⟩ = k[x+, y−, u]. To push forward an object from AnƒV(x+) × An to An, we let C ⋅ be the Cech resolution of AnƒV(x+) × An given by inverting the x+ i 's. We then compute: ΦQ+(O(i)) =(C ⋅ ⊗R⊗kR Q(R)s ⊗R R(i, 0))(0,∗) ⊗R⊗k R(Q(R)s)(i,∗) ⊗R⊗k R R ⊗k R[u]~⟨y + − uaix+ i = C ⋅ = C ⋅ i , x− i − u−biy − i ⟩. to taking Gm-invariants in the equivariant Fourier-Mukai transform). We thus have reduced the problem to computing the Cech cohomology Here, the notation (i, ∗) refers to the Z2 grading C ⊗R M ⊗R Q(R) inherits as a R ⊗k R-module, and (i, ∗) means the sum over all degree (i, d) pieces for all d ∈ Z (this is the degree restriction which corresponds i =(0, bi), and deg u = of Q(R) = k[x+, y−, u] with deg x+ (−1, 1). This is a well known computation. Namely, one can check that the cohomology vanishes when the cohomological degree is in (µ, 0), and in degree zero is given by u−ik[uax+, y−] = R(i). i =(ai, 0), deg y − (cid:3) Remark 4.1.6. The above proof also exhibits an isomorphism of func- tors ΦQ+ ○ j ∗ = Id when restricted to the full subcategory generated by those R(i) such that i ∈(µ, 0]. Another consequence is that the essential image of ΦQ+ actually lies in Db coh(An), as promised in Remark 3.2.5. k 4.2. The smooth affine case in general. We now consider the case where a ring R in CRGm is such that Spec R is smooth. To make reductions to affine space case, we will repeatedly make use of the Luna Slice Theorem, to which we refer to [Dre04] for expository background or [Lun73] for the original result. Accordingly, from this point on we assume k is a field. Let us state a version at the level of generality we require; in particular we will only require the theorem near points on the fixed locus, which simplifies the statement. 30 BALLARD, DIEMER, AND FAVERO Proposition 4.2.1. Let Gm act on a smooth affine variety X = Spec R, and let x ∈ X be a fixed point for the action. Then there exists a Gm- invariant affine subvariety V ⊆ X containing x, called the slice at x, and a diagram V g f TxX X where the maps f, g are strongly ´etale (see Remark 4.2.2 below for a reminder of this definition), and TxX denotes the tangent space to X at x. Moreover, it can be arranged so that V = Spec Rr where deg r = 0 and that the image of V under g is Spec Tt where deg t = 0, where we have written TxX ∶= Spec T . Let us prove that our version indeed does follow from the version in [Dre04, Theorems 5.3 and 5.4]. Proof. The only non-trivial differences between 4.2.1 and [Dre04, The- orems 5.3 and 5.4] are that we require g to be strongly ´etale and not just strongly ´etale onto its image, and that we may take V = Spec Rr as claimed. Taking V from [Dre04, Theorems 5.3 and 5.4], we may cover V by open subsets of the form Spec Ra. If some Spec Ra contains the fixed point x, we must have deg r = 0. Since strongly ´etale morphisms base change under localization by degree zero elements we may replace V by Spec Ra with deg a = 0 and we still know that g is strongly ´etale onto its image. Now, similarly cover im g by open subsets of the form Spec Tt. Once again, in order to contain the origin (the image of the fixed point), we must have deg t = 0. Hence, if we replace Spec Ra by Spec Rag(t), then g is a strongly ´etale map to Spec Tt. Furthermore as deg t = 0, the inclusion into Spec T is strongly ´etale as well. Therefore, g, as a composition, is strongly ´etale. (cid:3) Remark 4.2.2. Let G be any reductive group. Recall that if φ ∶ R → S is a map of rings, such that φ ∶ Spec S → Spec R is equivariant, one says that φ is strongly ´etale if the induced map on invariant subrings φG ∶ RG → SG is ´etale and there is an isomorphism S ≅ R ⊗RG SG in CRG tensor product is taken with respect to φ and the inclusion RG ⊆ R. k where the KERNELS FROM COMPACTIFICATIONS 31 We now show that the functor Q satisfies base change for strongly ´etale ring maps. Proposition 4.2.3. Let f ∶ R → S be a strongly ´etale map of rings with Gm-actions. Then, a) there is an isomorphism of bimodules S f ⊗ s Q(R) ≅ Q(R) p⊗ f S ≅ Q(S), b) and an isomorphism of functors, f ∗ ○ ΦQ(R) ≅ ΦQ(S) ○ f ∗. Proof. We first prove a). We have the obvious map Q(R) ⊗R S → Q(S) given by Q(f) ⊗ 1; we claim this map is an isomorphism. This map sits as the top arrow in the commutative diagram f S Q(R) p⊗ R[u, u−1] π⊗f S Q(S) S[u, u−1]. = and the left vertical arrow is injective since f ∶ R → S is ´etale and thus, To demonstrate surjectivity, we use the isomorphism S ≅ R ⊗RG SG afforded by the strongly ´etale condition. Namely, given s ∈ S, we write in particular, is flat. It follows that Q(f) ⊗ 1 is also injective. s = ∑i ri ⊗ si where each ri ∈ R and si ∈ S Gm. Let σ ∶ S → S[u, u−1] be the co-action ring map for the Gm-action on S. Then σ(s) = σ(Q ri ⊗ si) =Q σ(ri) ⊗ si. Now, write each ri ∈ R as a sum of homogenous elements: ri = ∑ rj where deg(rj i) ∶= nij. Then i and so σ(ri) ⊗ si =Qj rj i unij σ(ri) =Qj i unij ⊗ si =Qj rj unij ⊗ f(rj i)si since the tensor product is over R via f on the right. To show that σ(s) is in the image of Q(f) ⊗1, it suffices to show that if uk ∈ Q(R) for some k ∈ Z, then uk ∈ Q(S). But this is clear as Q(f) ∶ Q(R) → Q(S) is obtained, by definition, by restricting the map R[u, u−1] → S[u, u−1] induced by f . It is trivial that any s ∈ S is in the image of Q(f) ⊗ 1, and so we have shown that any element of the form s, σ(s), and also the element u are all in the image, but such elements generate Q(S) by definition. 32 BALLARD, DIEMER, AND FAVERO Statement b) is the following chain of equalities: f ∗(ΦQ(R)(M)) = S ⊗R(Q(R) ⊗R M)(∗,0) = S0 ⊗R0 (Q(R) ⊗R M)(∗,0) =(S0 ⊗R0 Q(R) ⊗R M)(∗,0) =(S ⊗R Q(R) ⊗R M)(∗,0) =(Q(S) ⊗R M)(∗,0) = ΦQ(S)(f ∗M). since f is strongly ´etale since f is strongly ´etale by part a) (cid:3) Remark 4.2.4. In the above proposition, it is necessary that the map f ∶ R → S is strongly ´etale and not just ´etale. For example, k[x] → k[x, x−1] with deg x > 0 is an open immersion, but it is easy to verify that base change for Q does not hold. Lemma 4.2.5. Let R be an object of CRGm smooth. Then the map ρR from Definition 3.2.9 is an isomorphism. such that X = Spec R is k Proof. It suffices to prove the map is locally an isomorphism. We know that Q(R) ≅ ∆(R) away from the contracting locus, hence ρR is an iso- morphism over Spec RƒV(I +). Now, for each point in the fixed locus one obtains Gm-invariant affine open neighborhood produced by the Luna Slice theorem 4.2.1. These neighborhoods cover the contracting locus so it remains to check that ρR is an isomorphism upon restriction to each such neighborhood. But ρR respects base change for strongly ´etale morphisms by Proposition 4.2.3, so the Luna Slice Theorem re- duces us to the case of affine space, which was Lemma 4.1.1. (cid:3) Proposition 4.2.6. Let R be an object of CRGm smooth. Then Q(R) with its R-module structure induced by p is Tor independent of Q(R) with its R-module structure induced by s. That is, such that Spec R is k for all p > 0. TorR p(Q(R)p, Q(R)s) = 0 Proof. The proof is similar to that of the previous lemma. Namely, by Lemma 2.1.14, ΦQ(R) = Id away from X −, and since Q base changes under all maps in the Luna Slice Theorem by Proposition 4.2.3, we are reduced to the case where Spec R = An, where the statement was proved in Lemma 4.1.2. (cid:3) KERNELS FROM COMPACTIFICATIONS 33 and set µ to be the sum Definition 4.2.7. Let R be an object of CRGm of the weights of the conormal bundle of Spec R~I + = X − in X = Spec R. (Note that X − is smooth since X = Spec R was assumed smooth, see e.g. [Dri13, Proposition 1.4.20].) Assume that the fixed locus is con- nected. The grade restriction window, denoted by =W=, is the full sub- category of Db(cohGm Spec R) consisting of objects A such that for any fixed point x and some affine ´etale slice V = Spec S at x, the restriction k A ⊗R S ∈ Db(cohGm Spec S) is generated by S(i) for i ∈(µ, 0]. Lemma 4.2.8. Let R be an object of CRGm such that X = Spec R is smooth. Assume that the fixed locus is connected. Then the essential image of Fourier-Mukai functor k ΦQ+ ∶ Db(cohGm X ∖ V(I +)) → Db(QcohGm X) phism of functors lies in =W= ⊆ Db(cohGm X). Furthermore, on =W= there is an isomor- where j ∶ X ∖ V(I +) → X is the inclusion. ○ j ∗)=W= = Id=W= (ΦQ+ Proof. We know that ΦQ satisfies base change for strongly ´etale mor- phisms by Proposition 4.2.3. Furthermore, since semi-stable loci are preserved under strongly ´etale morphisms ([SvdB16, Lemma 3.2.1]), ΦQ+ also satisfies base change. Furthermore, since =W= is defined lo- cally, to show that Im ΦQ+ ⊆ =W=, it suffices to verify that the essential image of ΦQ+ lands in =W= locally. (4.3) Cover Spec R by open affine Gm-invariant neighborhoods V = Spec S of the fixed locus produced by the Luna Slice Theorem, and let g ∶ V → TxX ∶= Spec T be the strongly ´etale map. If Spec L is any Gm-invariant subvariety of Spec R, let =W= L be the grade restriction window on Spec L and jL ∶ Spec L∖ Spec L− → Spec L be the inclusion. We know that ΦQ+(T ) lands in =W= more, Db(cohGm V ∖ V −) is generated by j ∗ Hence, ΦQ+(S) lies in =W= T by Proposition 4.1.5. Further- SS(i) = g ∗j ∗ T T(i) for all i ∈ Z. S. Thus ΦQ+ lands in =W= locally, as desired. To prove the latter statement of the lemma, suppose M ∈ =W=, then we can cover Spec R by open affine Gm-invariant neighborhoods V = Spec S of the fixed locus produced by the Luna Slice Theorem where MV is generated by S(i) for µ < i ≤ 0, as above. We have that g ∗=W= T = g ∗ ○(ΦQ+(T ) ○ j ∗ S)=W= =(ΦQ+(S) ○ j ∗ S T)=W= T ○ g ∗ by Proposition 4.1.5 by strongly ´etale base change. 34 BALLARD, DIEMER, AND FAVERO Since the generators of =W= S lie in the essential image of g ∗, this implies (ΦQ+(S) ○ j ∗)=W= S = Id=W= S Hence, we have shown this isomorphism locally on Spec R. Now, we know that j ∗, j∗ and ΦQ satisfy base change for strongly ´etale morphisms by Proposition 4.2.3. Furthermore the projection for- mula gives an isomorphism of functors ΦQ+ ○ j ∗ ≅ ΦQ ○ j∗ ○ j ∗. Hence ΦQ+ are two natural morphisms ○j ∗ satisfies base change for strongly ´etale morphisms. There ΦQ Id ΦQ ○ j∗ ○ j ∗ The vertical arrow is the unit of the adjunction and the horizontal arrow comes from the map Q → ∆. We have checked that both maps are (locally) isomorphisms on =W= and the result follows. (cid:3) Theorem 4.2.9. Let Spec R be a smooth affine variety with a Gm- action and connected fixed locus. The functor ΦQ+ ∶ Db(cohGm Spec R ∖ V(I +)) → =W= is an equivalence of categories. Proof. Combining Lemmas 4.2.5 and 4.2.6 with Proposition 3.3.9 gives that ΦQ+ is fully-faithful. Essential surjectivity follows immediately from the isomorphism (ΦQ+ ○ j ∗)=W= = Id=W= which is part of Lemma 4.2.8. (cid:3) Remark 4.2.10. The assumption of a connected fixed locus can be removed by putting more care into the definition of =W=. Namely, one needs to keep track of the parameter µ for each connected component of the fixed locus. Corollary 4.2.11. Let Spec R be a smooth affine variety with a Gm- action and connected fixed locus. Let µ± be the sum of the weights of the conormal bundle of Spec R~I ± in X = Spec R. Define where j− ∶ Spec R ∖ V(I −) → Spec R is the inclusion. If µ+ + µ− = 0 then ○(− ⊗ O(−µ+ − 1)) ○ ΦQ+ Φwc ∶ Db(cohGm Spec R ∖ V(I +)) → Db(cohGm Spec R ∖ V(I −)) Φwc ∶= j ∗ − KERNELS FROM COMPACTIFICATIONS 35 is an equivalence of categories. Proof. By the previous theorem, ΦQ+ gives an equivalence Db(cohGm Spec R ∖ V(I +)) ≅ =W= +. On the other hand, we may regard Spec R with the inverse Gm-action to produce an isomorphic stack. This exchanges I + with I − so we get an equivalence Under these identifications, the assumption that µ+ = µ− ensures that Db(cohGm Spec R ∖ V(I −)) ≅ =W= −. =W= + ⊗ O(−µ+ − 1) = =W= −. The result follows as j ∗ − is the inverse to ΦQ− on =W=. (cid:3) 5. The general case: homotopical methods 5.1. Deriving Q. We let sCRGm k denote the category of simplicial com- mutative rings with Gm-actions over k, i.e. the category of simplicial objects in the category of Z-graded commutative rings. We may refer to a ring R in CRGm as an ordinary ring when it becomes necessary to emphasize that it is not a simplicial ring. For simplicial sets, we will denote n-simplices by ∆[n], the union of their faces by ∂∆[n], and let Λi[n] denote the the simplicial horn, which we recall is the is the union of all of the faces of ∆[n] except for the i-th face. k k Remark 5.1.1. We will denote elements of sCRGm by R●, and for n ∈ Z we denote the ordinary ring structure at level n with respect to the underlying simplicial set by Rn. The reader is allowed to be concerned about a potential clash in notation with the grading on a Z-graded ring (for example, each Rn as above is itself Z-graded). We have made efforts to ensure that no explicit Z-gradings are referred to in this section, and so the reader should henceforth assume that all subscripts refer to a simplicial level, unless told otherwise. Recall that sCRGm k has distinguished objects which play the role of free objects. Namely, let F ∶ sSet → sCRk be the left adjoint to the forgetful functor from simplicial rings to sim- plicial sets (here F stands for "free" and not "forget"). Given a sim- plicial set X and a weight a ∈ Z, we likewise have an object of sCRGm denoted F(X)a, where we declare the degrees of the generators (with respect to the Z-grading) of F(Xn)a to all be a. k 36 BALLARD, DIEMER, AND FAVERO Proposition 5.1.2. The category sCRGm k brantly generated model structure where: possesses a simplicial cofi- ● the generating cofibrations are the maps F(∂∆[n])a → F(∆[n])a ● The generating trivial cofibrations are F(Λr[n])a → F(∆[n])a for for a ∈ Z and n ≥ 0. a ∈ Z, n ≥ 0, and 0 ≤ r ≤ n. ● The weak equivalences are those of the underlying simplicial sets, i.e. a map is a weak equivalence if and only if it is a weak equiv- alence after applying the forgetful functor to simplicial sets. Proof. This seems to be well-known. For example, it is a special case of [DHK97, Theorem 9.8] (see the discussion above the theorem for how their result specializes to sCRGm ). It is also [GJ99, Example 5.10]. (cid:3) Remark 5.1.3. A similar statement holds for sCRG2 k m k[u]. The main property we will need is the following. Corollary 5.1.4. Any cofibrant object in sCRGm quential colimit of pushouts along the generating cofibrations F(∂∆[n])a → F(∆[n])a. is a retract of a se- k Proof. For cofibrantly generated model categories in general, this fact [Hov01, Theorem, is known as the small object argument, see e.g. 2.1.14 and Corollary 2.1.15]. (cid:3) By applying the functor Q from Section 2.1 level-wise and to all face and degeneracy maps, we obtain a functor which we also denote by Q. More precisely, by viewing sCRGm as the functor category from the opposite of the simplex category to CRGm k[u]), we obtain an induced functor (and likewise for CRG2 m k k Q ∶ sCRGm k → sCRG2 m k[u]. (5.1) which by abuse of notation we also denote Q. For any R● an object of sCRGm simplicial ring maps , the object Q(R●) comes equipped with action and projection k (5.2) which agree level-wise with the ordinary action and projection ring maps. p, s ∶ R● → Q(R●) Lemma 5.1.5. The functor Q ∶ sCRGm k → sCRG2 m k[u] is left Quillen. Proof. We first show that Q preserves cofibrations and trivial cofibra- tions. Let fCRa consisting of free commutative k-algebras whose generating set of elements has weight k denote the full subcategory of CRGm k KERNELS FROM COMPACTIFICATIONS 37 a. By Example 2.1.8, Q(Rn) ≅ Rn[u] for each n with respect to one of the two module structures depending on the sign of the weights, i.e. Q(R●) ≅ R●[u] (5.3) for any object R● ∈ fCRa a < 0. Given this, we see that k where [u] = (a, 0) if a ≥ 0 and [u] = (0, a) of QFk(X)a ≅Fk[u](X)(a,0) Fk[u](X)(0,a) if a ≥ 0 if a < 0 were Fk, Fk[u] denote the free functors for simplicial commutative k and k[u]-algebras respectively. In otherwords there is an isomorphism of functors QFk ≅ Fk[u]. It follows that Q preserves the generating cofibrations and the gener- ating trivial cofibrations from 5.1.2. By [Hov01, Lemma 2.1.20 ], this implies that that Q preserves all cofibrations and trivial cofibrations. It remains to show that Q admits a right adjoint. In the Appendix in Equation (6.2) we will show that Q ∶ CRGm k → CRG2 m k[u] has a right adjoint. As taking simplicial objects is a 2-functor, we have a corresponding adjoint for Q regarded as the induced functor on simplicial objects. (cid:3) The above result allows us to define our promised derived replace- ment of the functor Q. Definition 5.1.6. Let LQ ∶ HosCRGm k  → HosCRG2 k[u] m (5.4) be the total left derived functor of Q. Here Ho denotes the homotopy category, i.e. the localization of the category sCRGm k[u]) at weak equivalences. (resp. sCRG2 m k In other words, if R● is an object of sCRGm , we have k LQ(R●) = Q(S●) (5.5) where S● → R● is any cofibrant replacement, which is well-defined since Q is a left Quillen functor. That is, if Cofib denotes a cofibrant re- placement functor in sCRGm then k LQ ∶= Q ○ Cofib . (5.6) 38 BALLARD, DIEMER, AND FAVERO 5.2. Property Pder. We now introduce the analogue of the map ρR from Definition 3.2.9 and the analogue of Property P from Defini- tion 3.3.7. The main result of this subsection will be Theorem 5.2.7, which will show that cofibrant objects have Property Pder. This result has the effect of bypassing the Tor vanishing assumptions in Lemma 3.3.6, which was our original criterion for fully-faithfulness of the functor on derived categories associated to Q+. Definition 5.2.1. We say that an object R● of sCRGm Pder if the map k has Property βR● ∶(p ⊗k s)∗Q(R●) L s⊗p Q(R●) → Q(R●) (5.7) given by the composition s⊗p Q(R●) →(p ⊗k s)∗Q(R●) s⊗p Q(R●) ρR● → Q(R●) (p ⊗k s)∗Q(R●) L truncation of the derived tensor product after application of (p ⊗ s)∗ is a weak equivalence. Here the first map in the composition is the (with takes middle invariants with respect to the G3 m-action on the tensor product) and ρR● is the map from Definition 3.2.9 applied level- wise. We use the notation β (with no subscript) to denote the natural transformation of the two functors sCRGm k[u] determined by the left and right side of Equation (5.7). In particular, R● has Property → sCRG2 Pder exactly when β(R●) = βR● is a weak equivalence. m k Example 5.2.2. Let R● be an object of sCRGm such that, at each level n, Rn has only non-negative weights (resp. at each level has only non- positive weights) Then Q(R●) is level-wise flat over R● via one of either k s or p, and so the map Q(R●) L s⊗p Q(R●) → Q(R●) s⊗p Q(R●) [Qui67, Corollary II.6.10]. Also ρR● is is a weak equivalence, see e.g. a weak equivalence, indeed it is actually an isomorphism level-wise by Lemma 3.2.10. It follows that R● has Property Pder. In particular, the objects F(∂∆[n])a and F(∆[n])a of CRGm For more explicit examples of Q(R●) and βR●, it is typically more Property Pder for any a ∈ Z and n ≥ 0. convenient to work with dg-algebras instead of simplicial rings (which one may do via the Dold-Kan correspondence, at least in characteristic zero). have k KERNELS FROM COMPACTIFICATIONS 39 Example 5.2.3. Consider the example R = k[x, y]~xy with deg x = 1 and deg y = −1. Assume that k is a field of characteristic zero. Exam- ple 3.2.11 showed that without deriving this example, βR (i.e. ρR from Definition 3.2.9) is not an isomorphism. However, R has a cofibrant replacement by the dg-algebra S = k[x, y, e] with d(e) = xy where e has homological degree 1 and weight 0. To compute βS we take the degree zero part of Q(S) ⊗S Q(S) = k[x, yu−1, e, u] ⊗k[x,y,e] k[x′, y ′u′−1, e′, u′] = k[x, y ′u′−1, e, u, u′]. The degree zero part is k[x, y ′u′−1, e, uu′] which is the realization of the isomorphism Q(S) = (Q(S) ⊗S S[u, u−1])0. Hence, we have corrected the failure of βR to be an isomorphism. That is, Property Pder holds. We now prove a general lemma which gives conditions for a natural transformation between model categories to assign cofibrant objects to weak equivalences. Lemma 5.2.4. Let F, G ∶ C → D be functors between model categories and η ∶ F → G be a natural transformation. Assume that a) C is cofibrantly generated; b) D is a combinatorial model category in the sense of [Dug01]; c) there is an initial cofibrant object c0 ∈ C and η(c0) is a weak- d) if η(c) is a weak equivalence for some object c ∈ C, then any pushout of η(c) along a generating cofibration is a weak equiva- equivalence; lence; e) η commutes with sequential colimits. Then, η(c) is a weak equivalence for any cofibrant object c ∈ C. Proof. By assumption, any cofibrant object is a retract of a sequential colimit of pushouts along generating cofibrations. Since any natural transformation respects retracts, it suffices to prove that any sequential colimit of pushouts along generating cofibrations is a weak equivalence. This follows from transfinite induction and the assumptions since in a combinatorial model category being a filtered colimit of weak equiva- lences, is itself a weak equivalence by [Dug01, Proposition 7.3]. (cid:3) We now begin the process of verifying that the hypotheses of Lemma 5.2.4 are satisfied by the natural transformation β. Only the hypotheses d) and e) are non-trivial to verify. We first show that β satisfies condition d). 40 BALLARD, DIEMER, AND FAVERO Lemma 5.2.5. Assume that R● has Property Pder and that we have a map f ∶ F(∂∆[n])a → R●. Then the pushout along the natural map F(∂∆[n])a → F(∆[n])a has Property Pder. Proof. For notational simplicity let S ∶= F(∂∆[n])a and T ∶= F(∆[n])a. We want to check that βT ⊗S R● ∶(p ⊗ s)∗(Q(T ⊗S R●) L s⊗p Q(T ⊗S R●)) → Q(T ⊗S R●) is a weak equivalence. The map S → T is a cofibration so level-wise it is a retract of a free commutative extension. By Lemma 5.1.5, the map Q(S) → Q(T) is then also a cofibration. In particular, it is level-wise flat. Therefore, the natural map Q(T) L ⊗Q(S) Q(R●) → Q(T) ⊗Q(S) Q(R●) is a weak equivalence. Now, we have the following chain of weak equivalences (5.8) ⊗ Q(S) ⊗T Q(T) L Q(T) L ≅(Q(T) L ⊗Q(S) Q(R●)) L ≅(Q(T) ⊗Q(S) Q(R●)) L ≅ Q(T ⊗S R●) L s⊗p Q(T ⊗S R●) L ⊗R Q(R●) s⊗pQ(S)Q(R●) L s⊗p(Q(T) L ⊗Q(S) Q(R●)) s⊗p(Q(T) ⊗Q(S) Q(R●)) The first weak equivalence above holds because it is an isomorphism for tensor products or ordinary rings, and a well chosen cofibrant re- placement functor commutes with taking tensor products/coproducts (see Proposition 2.3 of [Dug01] which we may apply due to Proposi- tion 5.1.2). The second weak equivalence follows from Equation (5.8). For the last equivalence above, we will show in the appendix in Corol- lary 6.0.2 that Q preserves arbitrary colimits; in particular, it preserves tensor products. Denote the above chain of weak equivalences by φ. Since Gm is linearly reductive, (p ⊗k s)∗ preserves weak equivalences, and so Since βT , βS are weak equivalences by Example 5.2.2 and βR● is a weak equivalence by assumption, the right hand side is a colimit of weak equivalences. Hence, by the 2 out of 3 condition for weak equivalences, βT ⊗S R● is a weak equivalence, as desired. (cid:3) βT ⊗S R● ○(p ⊗ s)∗φ = βT L ⊗βS βR●. (5.9) KERNELS FROM COMPACTIFICATIONS 41 We now verify that β satisfies condition e). Lemma 5.2.6. The natural transformation β from Definition 5.2.1 commutes with sequential colimits. Proof. The functor Q preserves colimits by Corollary 6.0.2 (this applies to simplicial objects since taking simplicial objects is a 2-functor). Fur- thermore, colimits commute with coproducts. The result follows. (cid:3) Theorem 5.2.7. Let R● be a cofibrant object of sCRGm Property Pder. k . Then R● has k , D = sCRG2 Proof. This is a direct application of Lemma 5.2.4. Namely, we set C = sCRGm k[u] and η = β. The initial object is k which is cofibrant and trivially satisfies Property Pder. The remaining conditions are verified by Proposition 5.1.2, Corollary 5.1.4, Lemma 5.2.5, and Lemma 5.2.6. (cid:3) m k 5.3. Base change and recovery of the smooth case. In this sec- tion we address the difference between Q(R) and LQ(R) when R is an object of CRGm , i.e. an ordinary Z-graded commutative ring. In par- ticular, in Proposition 5.3.4 we will exhibit a weak equivalence between them when Spec R is smooth, so that the general approach of Section 5 effectively reduces to the results of Section 4 under this assumption. We first prove a strongly ´etale base change result which is a derived version of Proposition 4.2.3. Proposition 5.3.1. Let R and S be objects of CRGm and let f ∶ R → S be a strongly ´etale graded homomorphism of ordinary rings. Regard R and S as objects of sCRGm by viewing them as constant simplicial rings. Then the base change map k k LQ(R) L s⊗f S → LQ(S) is a weak equivalence. Proof. By definition of strongly ´etale, there is an isomorphism of rings S = R ⊗R0 S0 where the subscripts here refer to degree with respect to the Z-gradings and not the level (as these rings are not simplicial). Now, let R● → S● be the image of R → S under the cofibrant replacement functor on sCRGm . Since taking the degree 0 piece at each level preserves weak equivalences, takes generating cofibrations to generating cofibrations, k 42 BALLARD, DIEMER, AND FAVERO and commutes with sequential colimits, we have a cofibrant replace- ment (R●)0 → (S●)0 of R0 → S0 in sCRk. Since R0 → S0 is flat, this gives a weak equivalence of coproducts Now apply Q to get weak equivalences S = R ⊗R0 S0 = R● ⊗(R●)0 (S●)0 LQ(S) = Q(R● ⊗(R●)0 (S●)0) = Q(R●) ⊗(R●)0 (S●)0 = LQ(R) ⊗(R●)0 (S●)0 = LQ(R) L = LQ(R) L ⊗R0 S0 ⊗R S. (cid:3) Corollary 5.3.2. If R → S is strongly ´etale, then there is an isomor- phism of S-modules πi(LQ(S)) = πi(LQ(R)) s⊗R S. Proof. This follows from a Quillen spectral sequence ([Qui67, Theorem 6 (c), Section 6.8] ). Namely, we have an E2 page E2 pq = πp(πq(LQ(R)) ⊗ L R S) ⇒ πp+q(LQ(R) ⊗ L R S) = πp+q(LQ(S)). Since R, S are constant simplicial rings and S is in particular flat over R, this forces p = 0 and the spectral sequence degenerates. (cid:3) Recall from Lemma 2.1.14 that, for ordinary rings with a Gm-action, Q(R) and ∆(R) become identified away from the contracting locus, i.e. after localizing by elements of positive degree. This intuitively suggests that Q does not require deriving after taking such a localization. We formulate this intuition more precisely as follows. Lemma 5.3.3. If R is an object of CRGm then there is a weak equivalence and r ∈ R has positive degree, k LQ(R) s⊗R Rr = Q(R) s⊗R Rr = ∆(R) σ⊗R Rr. Proof. The second equality is an isomorphism and is just Lemma 2.1.14. For the first weak equivalence, notice that the inclusion Q(R) → ∆(R) gives to a short exact sequence of simplicial R[u]-modules: 0 → LQ(R) → L∆(R) → L(∆(R)~Q(R)) → 0. Applying (− ⊗k[u] k[u, u−1]) annihilates L(∆(R)~Q(R)) giving an iso- morphism LQ(R) ⊗k[u] k[u, u−1] → L∆(R) ⊗k[u] k[u, u−1]. KERNELS FROM COMPACTIFICATIONS 43 In particular, we have an isomorphism of R[u]-modules πi(LQ(R) ⊗k[u] k[u, u−1]) = πi(L∆(R) ⊗k[u] k[u, u−1]) = πi(∆(R)) =R[u, u−1] 0 if i = 0 if i > 0. Now, under the isomorphism π0(LQ(R) L s⊗R Rr) = π0(LQ(R)) s⊗R Rr, the element u ⊗1 becomes a unit since r has positive degree (recall that s(r) = rudeg r). Hence, πi(LQ(R)) s⊗R Rr =Rr[u, u−1] 0 if i = 0 if i > 0. (cid:3) such that Proposition 5.3.4. Suppose that R is an object of CRGm Spec R is smooth. Then there is a weak equivalence k LQ(R) = Q(R). map Proof. We have a map LQ(R) → Q(R); we must show that the induced of R-modules is an isomorphism, i.e. that πi(LQ(R)) = 0 for i > 0 and π0(LQ(R)) = Q(R). This can be done locally on R. In particular, by π∗LQ(R) → Q(R) taking open sets coming from ´etale slices, we must exhibit isomorphisms πi(LQ(R)) ⊗R Rr = Q(R) ⊗R Rr when r has degree zero (to cover the fixed locus) and when r has strictly positive degree (to cover the contracting locus). On the fixed locus, Proposition 4.2.1 gives in particular a strongly ´etale map f ∶ T → Rr where T is a free object of CRGm and r has degree zero. Now LQ(T) = Q(T) since T is cofibrant when regarded as a constant simplicial ring. Proposition 5.3.1 and Proposition 4.2.3 then give k In particular, LQ(Rr) = Q(T) ⊗T Rr = Q(Rr). πi(LQ(Rr)) =Q(Rr) πi(LQ(Rr)) = πi(LQ(R)) ⊗R Rr. if i = 0 if i > 0 0 On the other hand, Corollary 5.3.2 says that 44 BALLARD, DIEMER, AND FAVERO It follows that πi(LQ(R)) ⊗R Rr =Q(Rr) = Q(R) ⊗R Rr 0 if i = 0 if i > 0 Similarly, if deg r > 0, Lemma 5.3.3. gives πi(LQ(R)) ⊗R Rr =Q(R) ⊗R Rr 0 if i = 0 if i > 0. (cid:3) 5.4. Localizations and Semi-orthogonal Decompostions in the General Case. This section develops localizations and semi-orthogonal decompositions associated to objects of sCRGm in analogy with the case of smooth commutative rings which we considered in Section 3.3. An important step in this direction is understanding how to inter- pret LQ(R●) as a Fourier-Mukai kernel object. That is, associated to LQ(R●) we wish to construct a corresponding object in the homotopy category of simplicial modules k HosModG2 m(R● L ⊗k R●), which we will do in Definition 5.4.4. This requires some attention to the model structure on the category of simplicial modules. A reader unconcerned with these details may wish to simply inspect Defini- tion 5.4.4 and the corresponding semi-orthogonal decomposition in Proposition 5.4.7 and bypass the remainder of the section. For the model structure on the category of simplicial modules, we refer directly to [Qui67, Chapter II.6]. To endow a triangulated struc- ture on the respective homotopy categories we use categories of spectra which are a model for the derived category, see e.g. [Jar15, Section 8.2] or [Sch97, Sections 2 and 3]; we will follow the exposition of the latter article very closely. In particular, the next definition is just [Sch97, Definition 2.1.1 and Definition 2.1.2] applied to simplicial rings with a Gm-action. Definition 5.4.1. Let R● be an object of sModGmR●. A spectrum M in sModGmR● is a collection of objects Mn and maps ΣMn → Mn+1 for each degree n ∈ N where ΣMn is the suspension of Mn. Maps of spectra M → N are defined to be collections of maps Mn → Nn such that the obvious squares commute. The category of spectra in sModGmR● is denoted by (sModGmR●)∞. In the above Σ denotes the suspension endofunctor on sModGmR●, which is defined since sModGmR● is a simplicial model category with KERNELS FROM COMPACTIFICATIONS 45 a zero object. For later use, we recall that the right adjoint of Σ is the loop functor Ω. [Sch97, Corollary 3.1.4] describes the model category structure on the category of spectra; we do not reproduce the details here for the sake of brevity, and because we only require certain structural properties. However, we remind the reader that a map of spectra is said to be a strict fibration if it is degree-wise a fibration with respect to the model structure on sModGmR●. Likewise, a map of spectra is said to be a a strict weak equivalence if it is degree-wise a weak equivalence. The fibrations (sometimes called stable fibrations) and weak equivalences (sometimes called stable weak equivalences) require additional properties that we will delegate to the proofs below. Given any object X of sModGmR● we obtain a suspension spectrum, denoted Σ∞X, by setting ΣXn = ΣnX and taking the identity maps in each degree n. This induces a suspension functor Σ∞ ∶ sModGmR● →(sModGmR●)∞. (5.10) One defines an Ω-spectrum to be a spectrum M such that each map Mn → ΩMn+1 is a weak homotopy equivalence, and a spectrum is said to be connective if, for each n ≥ 1, the higher homotopies of ΩMn vanish. The following result, which is [Sch97, Lemma 2.2.2], allows us to use spectra to understand the homotopy category of the category of simplicial modules. Proposition 5.4.2. The total left derived functor of the suspension functor Σ∞ gives an equivalence between the homotopy category of graded simplicial modules and of the homotopy category of connective spectra. Remark 5.4.3. The triangulated category Ho(sModGmR●) is actually equivalent to a perhaps more familiar triangulated category via the Dold-Kan correspondence. For simplicity, assume R = R● is discrete, i.e. all higher homotopies vanish. Then the Dold-Kan correspondence gives a Quillen equivalence between sModGm R and ModGm ≤0 R, the cat- egory of chain complexes of R-modules vanishing in positive degrees. This induces a Quillen equivalence of spectra (sModGm R)∞ ≅(ModGm ≤0 R)∞. One can easily identify Ho(ModGm ≤0 R)∞ with connective spectra becom- ing complexes with homology concentrated in non-positive degrees, see e.g. [Jar15, Section 8.2]. Hence, if one prefers, one can always in practice work with differential graded modules over a graded commu- tative dg-algebra for the purposes of this section. We have not done so, though, because our definition of LQ via simplicial rings (instead 46 BALLARD, DIEMER, AND FAVERO of connective chain complexes) sits more naturally in the category of spectra of modules. Up to one important detail we will discuss immediately below, we now have enough structure in place to give a repackaging of LQ(R●) as a kernel object. Definition 5.4.4. Let R● be an object of sCRGm brant replacement. Then LQ(R●) = Q(S●) is naturally a Z2-graded simplicial S● ⊗k S●-module, and hence gives an object and S● → R● a cofi- k of Ho(sModG2 mR● ⊗ gories in Proposition 5.4.2. Qder(R●) ∶= Σ∞LQ(R●) = Σ∞Q(S●) k R●)∞ under the identification of homotopy cate- (5.11) L As hinted above, there is a subtlety in the above Definition, in that Qder(R●) is not actually well-defined unless the weak equivalence S● → R● induces an equivalence (sModG2 mS● L ⊗k S●)∞ →(sModG2 mR● L ⊗k R●)∞. Indeed, we will show in Proposition 5.4.6 that this is always the case. To this end, we first record a simple lemma. Lemma 5.4.5. Any strict fibration M → N of Ω-spectra in(sModGmR●)∞ is a fibration. Proof. Let S denote the endofunctor of sModGmR● which is defined before [Sch97, Definition 2.1.4] (and note that Schwede uses the nota- tion Q for this functor, which is totally different than our use of Q). Roughly, S(M) is a weakly equivalent Ω-spectrum associated to M. By [Sch97, Proposition 2.1.5], M → N being a fibration is equivalent to it being a strict fibration and the natural map f ∶ M → S(M) L ×S(N ) N being a weak equivalence. Since S(A) and S(N) are Ω-spectra, the maps M → S(M) and N → S(N) are strict weak equivalences. So we have a diagram S(M) ×S(N ) N g f M h S(M) = S(M) ×S(B) S(N) where g, h are weak equivalences. Hence, so is f , as desired. (cid:3) KERNELS FROM COMPACTIFICATIONS 47 Proposition 5.4.6. If f ∶ R′ then restriction of scalars induces a Quillen equivalence → R● is a weak equivalence in sCRGm k ● , Res ∶(sModGmR●)∞ →(sModGmR′ ●)∞. Likewise, restriction of scalars induces a Quillen equivalence (sModG2 mR● L ⊗k R●)∞ →(sModG2 mR′ ● L ⊗k R′ ●)∞. Proof. First, we show that Res gives a Quillen adjunction. Exten- sion of scalars is a left adjoint, so we must show that Res preserves (stable) weak equivalences and (stable) fibrations. The forgetful func- tor sModGmR● → sSet satisfies the remark given after [Sch97, Corollary 2.1.6]. Namely, the weak equivalences are those inducing isomorphisms on homotopy groups of the underlying spectra. Since the underlying spectra do not change under restriction of scalars, we see that Res preserves weak equivalences. We now show that Res preserves fibrations. Restriction of scalars preserves strict fibrations of spectra and Ω-spectra, as all objects of sModGmR● are fibrant, and so Res M → Res N is a strict fibration if M → N is a fibration. Again using [Sch97, Proposition 2.1.5], it remains to show that Res M → S ′ Res M L ×S′ Res N Res N is a weak equivalence, where S and S ′ denote the denote the endofunc- tors of sModGmR● and sModGmR′ ● used in the proof of Lemma 5.4.5. Now consider the following diagram. Res M Res SM S ′ Res SM Res N Res SN S ′ Res SN The left square is homotopy cartesian by assumption. Lemma 5.4.5 shows that the right square is homotopy cartesian. Hence the full rectangle is homotopy cartesian. We likewise have a diagram Res M S ′ Res M S ′ Res SM Res N S ′ Res N S ′ Res SN From above the full rectangle is homotopy cartesian and the right square is homotopy cartesian by [Sch97, Proposition 2.1.3(e)]. Hence, the left square is homotopy cartesian, so Res M → Res N is a fibration, as claimed, and so Res gives a Quillen adjunction. 48 BALLARD, DIEMER, AND FAVERO To see Res is a Quillen equivalence, it remains to check that we have an equivalence on the homotopy categories. By [SS03, Theorem A.1.1], the objects ΣlR●(a) for l, a ∈ Z form a compact set of generators for the homotopy category Ho(sModGmR●)∞ . Hence, it suffices to show that, in the homotopy category, restriction of scalars is fully-faithful on these objects. But we assumed R● → R′ ● is a weak equivalence, and restriction of scalars commutes with suspension and shifts, so we see that restriction of scalars is indeed fully-faithful on this category. We thus have that Res gives a Quillen equivalence, as claimed. For the second part of the statement regarding bimodules, the proof is entirely the same once one observes that R′ ● L ⊗k R′ ● L → R● ⊗k R● is also a weak equivalence since derived tensor products preserve weak equivalences. (cid:3) Thus, Qder(R●) as given in Definition 5.4.4 is indeed well-defined. Let ∆der(R●) be the object of Ho((sModG2 ●)∞) determined by the diagonal object ∆(R●), and let Sder(R●) denote the cone in Ho((sModG2 ⊗k R′ ⊗k R′ mR′ mR′ L L ● ● ●)∞) which fits into the exact triangle Qder(R●) → ∆(R●) → Sder(R●). ∶ Ho(sModGmR●)∞ → Ho(sModGmR●)∞ M ↦(M L ⊗s Qder)0 So we get ΦQder the corresponding equivariant integral transform on the homotopy cat- egory of spectra. We define ΦSder similarly. Proposition 5.4.7. For any object R● of sCRGm orthogonal decomposition k , there is a semi- Ho(sModGmR●)∞ =⟨Im ΦSder, Im ΦQder⟩ which preserves connective spectra. Proof. By Lemma 3.3.4, it is enough to show that ΦQder η Ð→ Id cone(η) ÐÐÐÐ→ ΦSder is a Bousfield triangle, where we recall that η is map on functors induced from Qder → ∆. Furthermore, by Remark 3.3.2, it is enough to show that the map Q(η) ∶(Qder ⊗R● Qder)0 → Qder KERNELS FROM COMPACTIFICATIONS 49 is an isomorphism. This is equivalent to showing that the map βR ∶(p ⊗k s)∗LQ L s⊗p LQ → LQ is an isomorphism. This follows immediately from Theorem 5.2.7. (cid:3) Example 5.4.8. This is a continuation of Examples 3.2.11 and 5.2.3. Recall that in these examples, R = k[x, y]~(xy) where deg x = 1 and deg y = −1. Assume that k is a field of characteristic zero. We saw that Q(R) = k[x, y, u−1y, u]~(xy) ≅ k[x, z, u]~(xz). and, as a dg-algebra, with d(e) = xzu. Regarding this as an R-module, we have Qder(R) = k[x, z, u, e] Qder(R) = k[x, z, u]~xzu where y acts by zu. It follows that Sder =(k[x, y, u, u−1]~xy)~(k[x, z, u]~xzu). Now, we can regard Qder(R) as a quotient of k[x, y, z, u]~xy by the regular element uz − y. This allows one to compute In other words (Qder(R) L s⊗p Qder(R))0 = Qder(R). ΦQder(R) ○ ΦQder(R) = ΦQder(R), which is exactly the property that grants the existence of a semi- orthogonal decomposition. By checking on the the set of generators R(i), R~y(i), R~x(i) for all i, it follows that Im ΦQder is equal to the full subcategory of Db(modGm k[x, y]~(xy)) generated by R(i) for i ≤ 0, which we denote by Perf ≤0. Hence, we get a semi-orthogonal decom- position, Db(modGm k[x, y]~(xy)) =⟨Perf ⊥ ≤0, Perf ≤0⟩. It is worthwhile to observe here that Im ΦQder lies in perfect complexes. 5.5. The global case and D-flips. We now undertake the the task of globalizing Qder beyond the affine case and applying this to diagrams coming from flips. The aim of this subsection is to show that, with this globalization, one can use Qder to give a "wall-crossing" functor associated to any D-flip. As this will require synthesizing many of the previous constructions in a new context of sheaves of modules, let us give a quick overview of this subsection to guide the reader. ● We first recall a well-known construction of Reid and Thaddeus which associates a sheaf of graded O-modules A to a D-flip. 50 BALLARD, DIEMER, AND FAVERO ● As a first step to building a wall-crossing functor associated to the flip, we consider Q(A), essentially a sheafy version of Q(R) from the affine case. Without difficulties we can likewise consider a functor Q on simplicial sheaves of graded O- modules. ● Via Blander's model structure on the category of simplicial sheaves, we can likewise consider derived variants LQ and Qder just as in the affine case. We now recall the definition of a D-flip. This is a small variation on the definition of a flip in the sense of Mori theory, but which is well adapted to the setting of Gm-actions. Definition 5.5.1. Let X −, X0 and X + be varieties over a field k such that: a) There is a small contraction f ∶ X − → X0, i.e. f is a proper birational morphism whose exceptional locus has codimension at least two (in particular, X0 is not Q-factorial). b) X − is equipped with a Q-Cartier divisor D such that O(−D) is ⇢ X + is such that O(h∗D) is Q-Cartier c) X + admits a small contraction g ∶ X → X0, and the induced birational map h ∶ X − and g-ample. f -ample. Then X + is the D-flip of X − over X0. It is easy to check that a D-flip is unique if it exists. The following construction is recalled from [Tha96, Proposition 1.7] and relates D- flips to graded rings. Proposition 5.5.2. Let X + be the D-flip of X − over X0. Fix n ∈ N and set i.e. these quotient stacks are represented by X −, X + respectively. is represented by Proj Proof. We choose n such that A≥0 is generated in degree one. We will show in Lemmas 5.5.3 and 5.5.4 below that it follows that[Spec A≥0 = ࣷk∈N OX0(kD) = ࣷk∈N OX−(kD). But Proj projectivization of the relatively ample line bundle OX −(D), and is → X0 is a small contraction, A≥0 is the relative A≥0. Since X − A+~Gm] X0 X0 X0 For n sufficiently large there are isomorphisms A ∶=ࣷk∈Z X − =[Spec X + =[Spec OX0(knD). A−~Gm] A+~Gm], X0 X0 KERNELS FROM COMPACTIFICATIONS 51 thus isomorphic to X −, as claimed. The other equality holds by sym- metry. (cid:3) We now formulate the two lemmas promised in the above proof. Lemma 5.5.3. Let R be an object of CRGm and suppose that I + is generated in degree one, i.e by elements in R1. Then the global quotient stack k [Spec R ∖ V(I +)~Gm] in R1. is represented by the scheme Y that is obtained by gluing the open affine varieties Spec(Rfα)0 along Spec(Rfαfβ)0 for a set of generators {fα} Proof. This is the same as showing that Spec R ∖ V(I +) is a Gm-torsor over Y . Now, by assumption, Spec R ∖ V(I +) is covered by Spec Rf over Spec(Rf)0. Indeed, it is the trivial torsor. Namely, there is an with f ∈ R1, so it is enough to check that Spec Rf is itself a Gm-torsor isomorphism of rings Rf →(Rf)0[u, u−1] g ↦ gf −iui (cid:3) for g ∈(Rf)i with the inverse map determined by u ↦ f and h ↦ h for h ∈(Rf)0. The space Y as constructed in Lemma 5.5.3 is a quotient of Spec RƒV(I +) by Gm, hence is equipped with a line bundle OY(1) coming from the pullback of the map to [pt~Gm]. Notice that if V(I +) has codimension at least 2 then there is an isomorphism Γ(Y, OY(i)) = Ri. in degree one and that V(I +) has codimension at least two. Then the line bundle OY(1) defined above is ample. Lemma 5.5.4. Let R be an object of CRGm such that I + is generated Proof. Ampleness is equivalent to the complements of sections form- ing a cover Y such that the natural map to the coordinate ring is an open immersion. By assumption, the complements of global sections of OY(1) given by elements f ∈ R1 cover Y . Hence, it suffices to show k that the map Y → Proj(ࣷi∈N Γ(Y, OY(i))) = Proj R≥0 is an open immersion. This is so because an open subset Spec(Rf)0 ⊆ Y is homeomorphic to its image Spec((R≥0)f)0, as the natural inclusion of rings(R≥0)f → Rf is an isomorphism since f is a unit in Rf . Namely, the inclusion is surjective since any element gf i is the image of gf N f i−N for N >> 0. (cid:3) 52 BALLARD, DIEMER, AND FAVERO We now want to use Qder from Equation (5.11) to define a wall- crossing functor for D-flips. Of course, Qder was only defined in the affine case, and so we must sheafify. Let us begin this process first with our original functor Q ∶ CRGm k → CRG2 m k[u] from Definition 2.1.6. This Q automatically gives a presheaf of OY - algebras on the affine site over Spec k. Denote the category of Z-graded OY -algebras by CRGm OY. Sheafifying gives a functor Q ∶ CRGm OY → CRG2 m OY [u] (5.12) which, again, by abuse we also denote Q. Remark 5.5.5. An object Q(A) with A an object of CRGm is not necessarily quasi-coherent. This is a minor complication for our main goal, which is to define Fourier-Mukai functor between suitable homo- topy categories of quasi-coherent sheaves, and not sheaves of arbitrary O-modules. The following lemma will help us alleviate these concerns. OY Lemma 5.5.6. Let Y be a scheme with trivial Gm-action and A an object of CRGm OY . Then Q(A) is quasi-coherent. In particular, if Y is affine, then Q(A) is the sheaf associated to Q(A(Y)). Q(A) is the same as having Proof. Let U ⊆ V ⊆ Y be affine open subsets. Then quasi-coherence of Q(A(U)) = Q(A(V) ⊗OY (V ) OY(U)) = Q(A(V)) ⊗OY (V ) OY(U). Indeed, the first equality holds because of quasi-coherence of A. The second equality follows since the Gm-action on Y is trivial. (cid:3) Since the space X0 in a D-flip diagram is in practice usually singular, we must derive Q ∶ CRGm OY → CRG2 m OY [u] in this sheaf-theoretic setting, analogous to how we derived Q in the affine setting during the course of Section 5. Fortunately, we will be able to directly transfer results from Section 5 with only minor diffi- culties, once we understand a suitable model structure on simplicial sheaves. To this end, let sCRGm denote the category of simplicial ob- OY jects in sCRGm . The following result is essentially [Bla01, Theorem OY 2.1]. KERNELS FROM COMPACTIFICATIONS 53 Proposition 5.5.7. The category sCRGm admits a cofibrantly generated OY simplicial model structure where the weak equivalences are the maps that induce isomorphisms on the homotopy sheaves, and a set of gen- erating cofibrations are SymOY i!OU ⊗ ∂∆[n]a → SymOY i!OU ⊗ ∆[n]a, where U → Y is any open immersion and a ∈ Z is any weight. (Here i!OU denotes the extension by zero sheaf.) Proof. From [Bla01, Theorem 2.1], and in particular the final remark in the proof, we know that the category sSh(Y) of simplicial sheaves of sets on Y has a model structure with the weak equivalences as above and a set of generating cofibrations given by i!OU ⊗ ∂∆[n] → i!OU ⊗ ∆[n] We then have a family of free, forgetful adjunctions between sCRGm OY and sSh(Y) which induce the claimed model category structure (see e.g., [DHK97, Lemma 9.1] or [GJ99, Theorem 5.8]). (cid:3) The functor Q from equation 5.12 gives a functor (which, as usual, we also denote by Q) on simplicial categories Q ∶ sCRGm OY → sCRG2 m OY [u] (5.13) in the obvious way. Using the model category structure in Proposi- tion 5.5.7, it is easy to verify that this functor Q is still left Quillen, so we can define a derived functor LQ in total analogy with the affine case from Definition 5.1.6. Definition 5.5.8. Let LQ ∶ HosCRGm OY → HosCRG2 OY [u] m (5.14) be the total left derived functor of the functor Q in Equation 5.13 In particular, any LQ(A) is an object of ⊗OY A). m(A sModG2 L , we may consider the category of Likewise, given A an object of sCRGm OY spectra of sheaves of simplicial A-modules (sModGmA)∞, and define an ⊗OY A)) and a corresponding Fourier- object Qder(A) of Ho(sModG2 the kernel object Qder(A) preserves quasi-coherence, and is itself quasi- Mukai functor just as in Definition 5.4.4. Let us verify that this process does indeed result in desirable properties; in particular, we want that coherent (at least, up to weak equivalence). m(A L 54 BALLARD, DIEMER, AND FAVERO Proposition 5.5.9. Let Y be a scheme with a trivial Gm-action and A an object of CRGm which is quasi-coherent as a OY -module. Then OY the object LQ(A) of sModG2 m(A L ⊗OY A) is locally weakly equivalent to a simplicial quasi-coherent sheaf. Hence the functor ΦQder ∶ Ho(sModGm A)∞ → Ho(sModGm A)∞ preserves the full subcategories with quasi-coherent homotopy sheaves. U Proof. Let U ⊆ Y be an affine open subset and write Spec Spec R. Let AU be the sheaf associated to the k-algebra R. We need A(U) = to verify that he sheaf associated to LQ(R) is weakly equivalent to LQ(A)U . We first verify this for the case where U = Y (so that Y itself is affine). Inspecting Proposition 5.5.7, we see that the sheaves associated to the generating cofibrations in sCRGm are a subset of generating cofibrations in sCRGm R sheafifies to a weak OY equivalence in sCRGm . Thus, if we take a cofibrant replacement S● → R OY and sheafify, we still get a cofibrant replacement. Using this resolution . Similarly, a weak equivalence in sCRGm to compute LQ(A) yields the conclusion by Proposition 5.5.6. In general, if U ⊂ Y is an affine open subset, let S● be a cofibrant k . Then (S●)U is a cofibrant replacement of replacement of A in sCRGm OY AU . Hence consisting of complexes with quasi-coherent cohomology. Let Qwc der be the induced object on X − × X + via restriction and the isomorphism in Proposition 5.5.2. This is a generalization of the kernel object Qwc that was introduced in Equation (2.20) for the Bondal-Orlov flop equivalence, or the functor j ∗ ○ ΦQ+ from Corollary 4.2.11. − Question 5.5.10. Suppose X − between two smooth projective k-varieties, i.e. a flop. Is the functor ΦQwc ⇢ X + is a D-flip which is a K-equivalence der an equivalence? LQ(A)U = Q(S●)U = Q((S●)U) = LQ(AU), which we have already argued is weakly equivalent to the sheaf associ- (cid:3) ated to LQ(R). subcategory of Ho(sModGm A)∞ with quasi-coherent homotopy sheaves The Dold-Kan correspondence gives an equivalence between the full and where the latter is defined to be the full subcategory of D(QcohGm SpecY A), D(ModGm OSpecY A) KERNELS FROM COMPACTIFICATIONS 55 der preserves quasi-coherence, so Remark 5.5.11. We know that ΦQwc that we can view this as a question about quasi-coherent unbounded complexes. Moreover, the compact objects of this category are precisely the perfect objects [Nee96, Corollary 2.3]. Since X +, X − are smooth by assumption, these coincide with bounded coherent complexes so a positive answer to this question should resolve [Kaw04, Conjecture 5.1] (see [Kaw08]). Remark 5.5.12. To answer the question affirmatively (at least in the Gorenstein case), it should suffice to prove it in the case where Y is affine (see [RMdS07, Theorem 1.22]). Furthermore, note that when Y is affine, one only requires the machinery up through Section 5.2 to study ΦQwc der. 6. Appendix: Relations with Drinfeld's space We expand on Remark 2.1.11. There we observed that Q(R) agrees with the affine case of the main construction of [Dri13], although we will soon record one subtle distinction. Let k be a field and Z a (not necessarily affine) k-scheme of finite type over a field k and possessing an action by Gm and an open cover which is Gm-stable, i.e. Gm acts locally linearly. Drinfeld defines a fpqc sheaf on the category of k-schemes over A1 k as follows: for an arbitrary scheme T over A1 assign the set HomGm A1 (X ×A1 T, Z) where X ∶= A1 × A1 is equipped with the product map X → A1 and X has the "hyperbolic" Gm-action t ⋅(x, y) = (tx, t−1y). Amongst other results, [Dri13, Section 2.4] proves that this functor is representable, and so there exists a scheme Z over A1 k such that (6.1) Equivalently, this means that we have an adjunction HomA1(T, Z) = HomGm F ∶{k-schemes over A1} Ð⇀ A1 (X ×A1 T, Z). ↽Ð{Gm-schemes} ∶ G on the category on the right. As in Remark 2.1.11, if Z = Spec R where F(T) ∶= X ×A1 T , G(Z) ∶= Z and the actions are locally linear is affine then so is Z, and Z = Spec Q(R). Therefore both F above and its adjoint G preserve affine schemes. Restricting the adjunction above to affine schemes and then taking opposite categories, we get an adjunction where Qad Drin(S) ∶= k[x, y] ⊗k[u] S. Q∶ CRGm k Ð⇀ ↽Ð CRk[u] ∶ Qad Drin. 56 BALLARD, DIEMER, AND FAVERO However, the reader should be careful here. Recall that in Lemma 2.1.12, we defined Q with the following target Q ∶ CRGm k → CRG2 m k[u] not with target CRk[u] as above. This is no inherent contradiction, as m-action on any Z. However, this [Dri13, Section 2.1.17] constructs a G2 does mean that Qad Drin as defined above is not the correct adjoint of our functor Q. We will prove below that Q has the following adjoint. Given a Z2-graded k-algebra S over k[u] where deg u = (−1, 1), let Qad(S) denote the subalgebra of S ⊗k[u] k[x, y]~(xy) generated by S(i,0)xi for i ≥ 0 and S(0,−i)y −i for i < 0 Here, u maps to xy (so it is zero), deg x = (0, 1), and deg y = (−1, 0). Since Qad(S) has non-zero degree only in the (i, i) summands, we may regard it as a Z-graded k-algebra. Proposition 6.0.1. There is an adjunction Q∶ CRGm k Ð⇀ ↽Ð CRG2 m k[u] ∶ Qad. (6.2) Proof. Let S, T ∈ CRG2 k with maps m k[u] and let R ∈ CRGm p ∶ R → Q(R) s ∶ R → Q(R). Given any map f ∶ Q(R) → S in CRG2 m k[u] , define φf ∶ R → Qad(S) ⊆ S ⊗k[u] k[x, y]~(xy) by r ↦f(p(r)) ⊗ xdeg r when deg r ≥ 0 f(s(r)) ⊗ y − deg r when deg r ≤ 0. Conversely, given a map g ∶ R → Qad(S) in CRGm by defining its values on the generating elements of Q(R) as follows: define k ψg ∶ Q(R) → S r ↦ g(r)x if deg r ≥ 0 s(r) ↦ g(r)y if deg r < 0 KERNELS FROM COMPACTIFICATIONS 57 where g(r)x (resp. g(r)y) is the x-component (resp. y-component) of g(r) under the decomposition as abelian groups S ⊗k[u] k[x, y]~(xy) ≅ S~u[x] ⊕ S~u[y]. The maps f ↦ φf and g ↦ ψg are inverse isomorphisms which give the adjunction. (cid:3) Since any functor with a right adjoint preserves colimits, we have the following. Corollary 6.0.2. The functor Q ∶ CRGm k → CRG2 m k[u] preserves colimits. References [Bal17] Ballard, Matthew Robert. Wall crossing for derived categories of moduli spaces of sheaves on rational surfaces. Algebr. Geom. 4 (2017), no. 3, 263–280. [BFK12] Ballard, Matthew; Favero, David; Katzarkov, Ludmil. Variation of geo- metric invariant theory quotients and derived categories (2012). To appear in Crelle's Journal. [Bla01] Blander, Benjamin A. Local projective model structures on simplicial presheaves. K-Theory 24 (2001), no. 3, 283–301. [BO95] Bondal, Alexey; Orlov, Dmitri. Semiorthogonal decomposition for algebraic varieties. arXiv:alg-geom/9506012 (1995). [Bri02] Bridgeland, Tom. Flops and derived categories. Invent. Math. 147 (2002), no. 3, 613–632. [BKR01] Bridgeland, Tom; King, Alastair, Reid, Miles. The McKay correspondence as an equivalence of derived categories. J. Amer. Math. Soc. 14 (2001), 535–554. [CKL12] Cautis, Sabin; Kamnitzer, Joel; Licata, Anthony. Derived equivalences for cotangent bundles of Grassmannians via categorical sl2 actions. J. Reine Angew. Math. 675 (2013), 53–99. [CLS11] Cox, David A.; Little, John B.; Schenck, Henry K. Toric varieties. Gradu- ate Studies in Mathematics, 124. American Mathematical Society, Providence, RI (2011), xxiv+841 pp. [DH98] Dolgachev, Igor; Hu, Yi. Variation of geometric invariant theory quotients. With an appendix by Nicolas Ressayre. Inst. Hautes ´Etudes Sci. Publ. Math. No. 87 (1998), 5–56. [Dre04] Dr´ezet, Jean-Marc. Luna's slice theorem and applications. Algebraic group actions and quotients, Hindawi Publ. Corp., Cairo (2004), 39– 89. [Dri13] Drinfeld, Vladimir. On algebraic spaces with an action of Gm. arXiv:1308.2604 (2013). [Dug01] Dugger, Daniel. Combinatorial model categories have presentations. Ad- vances in Mathematics 164.1 (2001), 177–201. [DHK97] Dwyer, W., P. Hirschhorn, and D. Kan. Model categories and more gen- eral abstract homotopy theory, Book in preparation. [GJ99] Goerss, Paul G.; Jardine, John F. Simplicial homotopy theory. Reprint of the 1999 edition. Modern Birkhauser Classics. Birkhauser Verlag, Basel, (2009), xvi+510 pp. 58 BALLARD, DIEMER, AND FAVERO [HR17] Hall, Jack, and David Rydh. The telescope conjecture for algebraic stacks. Journal of Topology 10.3 (2017), 776–794. [HL15] Halpern-Leistner, Daniel. The derived category of a GIT quotient. J. Amer. Math. Soc. 28 (2015), no. 3, 871–912. [HS16] Halpern-Leistner, Daniel; Sam, Steven. Combinatorial constructions of de- rived equivalences. arXiv:1601.02030 (2016). [HHP09] Herbst, Manfred; Hori, Kentaro; Page, David. B-type D-branes in toric Calabi-Yau varieties. Homological mirror symmetry, Lecture Notes in Phys., 757, Springer, Berlin, (2009), 27–44. [Hov01] Hovey, Mark. Model categories. Mathematical Surveys and Monographs, 63. American Mathematical Society, Providence, RI (1999), xii+209 pp. [Jar15] Jardine, John F. Local homotopy theory. Springer Monographs in Mathe- matics. Springer, New York (2015), x+508 pp. [Kaw04] Kawamata, Yujiro. D-equivalence and K-equivalence. J . Differential Geom. 61 (2002), no. 1, 147–171. [Kaw06] Kawamata, Yujiro. Derived equivalence for stratified Mukai flop on G(2,4). Mirror symmetry. V, AMS/IP Stud. Adv. Math., 38, Amer. Math. Soc., Prov- idence, RI, (2006), 285–294. [Kaw08] Kawamata, Yujiro. Flops Connect Minimal Models. Publications of the Research Institute for Mathematical Sciences 44 (2008), 419–423. [Kra12] Krause, Henning. Localization theory for triangulated categories. Triangu- lated categories, London Math. Soc. Lecture Note Ser., 375, Cambridge Univ. Press, Cambridge (2010), 161–235. [Lun73] Luna, Domingo. Slices ´etales. Sur les groupes alg´ebriques, Bull. Soc. Math. France, Paris, Mmoire 33 Soc. Math. France, Paris (1973), 81–105. [RMdS07] Ruip´erez, Daniel Hern´andez, Ana Cristina L´opez Martn, and Fernando Sancho de Salas. Fourier-Mukai transforms for Gorenstein schemes." Ad- vances in Mathematics 211.2 (2007), 594–620. [Nam03] Namikawa, Yoshinori. Mukai flops and derived categories. J. Reine Angew. Math. 560 (2003), 65–76. [Nam04] Namikawa, Yoshinori. Mukai flops and derived categories. II. Algebraic structures and moduli spaces, CRM Proc. Lecture Notes, 38, Amer. Math. Soc., Providence, RI, (2004), 149–175. [Nee96] A. Neeman. The Grothendieck duality theorem via Bousfield's techniques and Brown representability. Journal of the American Mathematical Society 9.1 (1996), 205–236. [Orl92] Orlov, Dmitry. Projective bundles, monoidal transformations, and derived categories of coherent sheaves. Izv. Ross. Akad. Nauk Ser. Mat. 56 (1992), no. 4, 852–862; translation in Russian Acad. Sci. Izv. Math. 41 (1993), no. 1, 133–141 [Orl09] D. Orlov. Derived categories of coherent sheaves and triangulated categories of singularities. Algebra, arithmetic, and geometry: in honor of Yu. I. Manin. Vol. II, Progr. Math., 270, Birkhauser Boston, Inc., Boston, MA (2009), 503– 531. [Qui67] Quillen, Daniel G. Homotopical algebra. Lecture Notes in Mathematics, No. 43 Springer-Verlag, Berlin-New York (1967), iv+156 pp. [Sch97] Schwede, Stefan. Spectra in model categories and applications to the alge- braic cotangent complex. J. Pure Appl. Algebra 120 (1997), no. 1, 77–104. KERNELS FROM COMPACTIFICATIONS 59 [SS03] Schwede, Stefan; Shipley, Brooke.Stable model categories are categories of modules. Topology 42 (2003), no. 1, 103–153. [Seg11] Segal, Ed. Equivalence between GIT quotients of Landau-Ginzburg B- models. Comm. Math. Phys. 304 (2011), no. 2, 411–432. [SvdB17] Spenko, Spela; Van den Bergh, Michel. Non-commutative resolutions of quotient singularities for reductive groups. Invent. Math. 210 (2017), no. 1, 3–67. [SvdB16] Spenko, Spela; Van den Bergh, Michel. Semi-orthogonal decomposition of GIT quotient stacks. arXiv:1603.02858 (2016). [Tha96] Thaddeus, Michael. Geometric invariant theory and flips. J. Amer. Math. Soc. 9 (1996), no. 3, 691–723. [TV08] Toen, Bertrand; Vezzosi, Gabrielle. Brave new algebraic geometry and global derived moduli spaces of ring spectra. Elliptic cohomology, London Math. Soc. Lecture Note Ser., 342, Cambridge Univ. Press, Cambridge (2007), 325–359. Matthew Robert Ballard University of South Carolina Department of Mathematics Columbia, SC, USA Email: [email protected] Colin Diemer Institut des Haute ´Etudes Scientifiques Bures sur Yvette, France Laboratory of Mirror Symmetry NRU HSE Email: [email protected] David Favero University of Alberta Department of Mathematical and Statistical Sciences Central Academic Building 632, Edmonton, AB, Canada T6G 2C7 Korean Institute for Advanced Study 85 Hoegiro, Dongdaemun-gu, Seoul, Republic of Korea 02455 Email: [email protected]
1509.06510
2
1509
2016-02-11T21:30:38
Hochschild cohomology of projective hypersurfaces
[ "math.AG" ]
We compute Hochschild cohomology of projective hypersurfaces starting from the Gerstenhaber-Schack complex of the (restricted) structure sheaf. We are particularly interested in the second cohomology group and its relation with deformations. We show that a projective hypersurface is smooth if and only if the classical HKR decomposition holds for this group. In general, the first Hodge component describing scheme deformations has an interesting inner structure corresponding to the various ways in which first order deformations can be realized: deforming local multiplications, deforming restriction maps, or deforming both. We make our computations precise in the case of quartic hypersurfaces, and compute explicit dimensions in many examples.
math.AG
math
HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES LIYU LIU AND WENDY LOWEN Abstract. We compute Hochschild cohomology of projective hypersurfaces starting from the Gerstenhaber-Schack complex of the (restricted) structure sheaf. We are particularly interested in the second cohomology group and its relation with deformations. We show that a projective hypersurface is smooth if and only if the classical HKR decomposition holds for this group. In general, the first Hodge component describing scheme deformations has an interesting inner structure corresponding to the various ways in which first order deformations can be realized: deforming local multiplications, deforming restriction maps, or deforming both. We make our computations precise in the case of quartic hypersurfaces, and compute explicit dimensions in many examples. 1. Introduction Hochschild cohomology originated as a cohomology theory for associative algebras, which is known to be closely related to deformation theory since the work of Gerstenhaber. Meanwhile, both the cohomology and the deformation side of the picture have been developed for a variety of mathematical objects, ranging from schemes [23] [16] to abelian [19], [18] and differential graded [14], [17] categories. One of the first generalizations considered after the algebra case was the case of presheaves of algebras, as thoroughly investigated by Gerstenhaber and Schack [7], [9], [10]. For a presheaf A, Hochschild cohomology is defined as an Ext of bimodules ExtA-A(A,A) in analogy with the algebra case. An important tool in the study of this cohomology is the (normalized, reduced) Gerstenhaber-Schack double complex C(A). We denote its associated total complex by CGS(A), and the cohomology of this complex by H n GS(A) = H nCGS(A). We have GS(A) ∼= Extn H n A-A(A,A). Unlike what the parallel result for associative algebras may lead one to expect, in general H 2 GS(A) is not identified with the family of first order deformations of the presheaf A. A correct interpretation of H 2 GS(A) is as the family of first order deformations of A as a twisted presheaf, and an explicit isomorphism H 2 GS(A) −→ Def tw(A) (1.1) is given in [6, Thm. 2.21]. Moreover, in loc. cit., if A is quasi-compact semi-separated, the existence of a bijective correspondence between the first order deformations of A as a twisted presheaf and the abelian deformations of the category Qch(A) of quasi-coherent sheaves is proven. Hence in this case there are isomorphisms H 2 GS(A) ∼= Def tw(A) ∼= Def ab(Qch(A)). 6 1 0 2 b e F 1 1 ] . G A h t a m [ 2 v 0 1 5 6 0 . 9 0 5 1 : v i X r a Throughout, let k be an algebraically closed field of characteristic zero. The situation when A is a presheaf of commutative k-algebras over V is very interesting. As discussed in [7], in this case the complex CGS(A) admits the Hodge decomposition of complexes (1.2) CGS(A) =Mr∈N CGS(A)r, GS(A) =Mr∈N H n GS(A)r. which induces the Hodge decomposition of the cohomology groups H n mology groups H n GS(A) in terms of the coho- (1.3) GS(A)r = H nCGS(A)r: H n The authors acknowledge the support of the European Union for the ERC grant No 257004-HHNcdMir and the support of the Research Foundation Flanders (FWO) under Grant No. G.0112.13N. 1 2 LIYU LIU AND WENDY LOWEN The zero-th Hodge complex CGS(A)0 is nothing but the simplicial cohomology complex of A, and the first Hodge complex CGS(A)1, which is called the asimplicial Harrison complex in [7], classifies first order deformations of A as a commutative presheaf. Hence, in this case the map (1.1) naturally restricts to (1.4) H 2 GS(A)1 −→ Def cpre(A). Recall that a GS n-cochain has n + 1 components coming from the double complex C(A), in particular (1.5) C2 GS(A) = C0,2(A) ⊕ C1,1(A) ⊕ C2,0(A). Following [6], we usually write a GS 2-cochain as (m, f, c) corresponding to the decomposition (1.5), and we call a 2-cocycle (m, f, c) untwined (we used the terminology decomposable in [6]) if (m, 0, 0), (0, f, 0) and (0, 0, c) are 2-cocycles as well. A GS 2-class γ ∈ H 2 GS(A) is called intertwined if there is no untwined representative (m, f, c) with γ = [(m, f, c)]. Since A is a presheaf of commutative algebras, for a 2-cocycle (m, f, c) we automatically have that (0, 0, c) is a cocycle so (m, f, c) is untwined if and only if (m, 0, 0) is a cocycle if and only if (0, f, 0) is a cocycle. Under the Hodge decomposition C2 GS(A) = C2 any 2-cocycle (m, f, c) factors as the sum GS(A)2 ⊕ C2 GS(A)1 ⊕ C2 GS(A)0, (1.6) of 2-cocycles. Locally, mab is symmetric, sending (a, b) to m(a, b)/2 + m(b, a)/2, and m − mab is anti-symmetric. Hence, (m, f, c) is untwined if and only if mab is a 2-cocycle. (m, f, c) = (m − mab, 0, 0) + (mab, f, 0) + (0, 0, c) There exist various presheaves A such that every GS 2-class admits a representative (m, f, c) with mab = 0, which is thus untwined. This happens if A(V ) is smooth for all V ∈ V (see [6, §3.3]). In this case, let T be the associated tangent presheaf, and thus m, f , c represent classes in H 0 simp(V,∧2T ), H 1 simp(V,A), respectively. The Hodge decomposition gives rise to isomorphisms simp(V,T ), H 2 H n GS(A) ∼= Mp+q=n H p simp(V,∧qT ) in terms of the simplicial cohomology for all n. A typical example of such a presheaf is obtained from a quasi-compact separated, smooth scheme. Let (X,OX ) be a quasi-compact separated scheme with an affine open covering V which is closed under intersection, and let A = OXV be the restriction of OX to the covering V. The cohomology H • GS(A) turns out to be isomorphic to the Hochschild cohomology HH •(X) := Ext• X×X (∆∗OX , ∆∗OX ) of the scheme X where ∆ : X → X × X is the diagonal map [18]. If furthermore, X is smooth, then the Hodge decomposition corresponds to the HKR decomposition and we obtain the familiar formula (1.7) HH n(X) ∼= Mp+q=n H p(X,∧qTX ) where TX is the tangent sheaf of X. This formula has been proved in various different contexts and ways [9], [16], [23], [25], [6]. GS(A)0 ∼= H 2 If A(V ) is not smooth for some V ∈ V (for instance, X has singularities), then whereas we still simp(V,∧2T ), the situation for the first Hodge have H 2 component H 2 GS(A)1 now becomes more interesting. In the decomposition (1.6) of a 2-cocycle (m, f, c), the contribution of mab is not always zero, and [(mab, f, 0)] is intertwined in general. simp(V,A) and H 2 GS(A)2 ∼= H 2 HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 3 Although cocycles of the form (mab, f, 0) are not as nice as (0, f, 0), in some situations we can simplify them in the alternative manner, that is by getting rid of f rather than mab. Consider the following two subgroups of H 2 GS(A)1: • the subgroup Eres of 2-classes of the form [(0, f, 0)]; • the subgroup Emult of 2-classes of the form [(m, 0, 0)]. We are interested in computing H 2 GS(A)1, Emult, Eres, as well as understanding the relations between those three groups, possibly depending on the scheme X (with A = OXV). Note that Eres ∼= H 1 simp(V,T ). The decomposition (1.7) has been generalized to the not necessarily smooth case by Buchweitz and Flenner in [4], using the Atyiah-Chern character. In terms of the relative cotangent complex LX/k, the generalization is given by (1.8) HH n(X) ∼= Mp+q=n Extp X (∧q LX/k,OX ) where ∧q should be understood as derived exterior product. Their arguments are mostly estab- lished in the derived category D(X), and an interpretation of cohomology classes in terms of GS-representatives is not immediate. Since we need GS-representatives in order to use the deformation interpretation from (1.1), in §4.1 we construct a smaller complex H• than CGS(A) and we give an explicit quasi-isomorphism H• → CGS(A). Our construction of H• builds on [2] and [20], in both of which the Hochschild (co)homology of affine hypersurfaces is computed. Following their methods, in §3 we describe the Hodge components of the affine Hochschild cohomology groups in terms of the cotangent complex. The other key ingredient in our approach to the projective case is the use of a mixed complex associated to a pair of orthogonal sequences in a commutative ring, which is developed in the self-contained section §2. In §4.2 we present the cotangent complex LX/k in terms of twisted structure sheaves OX (l), and we verify that the cohomology of H• agrees with (1.8), and H• can be considered to be a natural enhancement of (1.8). In §4.3, we compute the cohomology groups of H• in terms of two easier complexes C•(uuu; S) and K•(vvv; R) of graded modules. Our main theorem is the following: Theorem 1.1. Let X ⊂ Pn be a projective hypersurface of degree d. Denote by P i the i-th cohomology group of C•(uuu; S) and by Qi the i-th cocycle group. Denote by Z i the i-th cocycle group of K•(vvv; R). Then the cohomology of H• is given by (1) when d > n + 1, (2) when d = n + 1, (3) when d < n + 1, P i−2r r+(i−r)(d−1) ⊕ Q−i i ⊕ S (Z −i+n−1 d−i−2 ); , i P i−2r r+n(i−r) ⊕ Q−i P i−2r r+n(i−r) ⊕ Q−i P i−2r r+n(i−r), i ⊕ kn, i 6= n − 1, n, i = n − 1, i = n; H i(H•) ∼=Mr<i Mr<i Mr<i Mr≤i H i(H•) ∼=Mr<i  H i(H•) ∼= P i−2r r+(i−r)(d−1) ⊕ Q−i i . 4 LIYU LIU AND WENDY LOWEN In the above formulas, S is a linear map defined in (4.7), and the subscripts of P •, Q•, Z • stand for the degrees of homogeneous elements in P •, Q•, Z •. As an application of Theorem 1.1, we give a cohomological characterization of smoothness for projective hypersurfaces in §4.4. Recall that an affine hypersurface Spec(A) is smooth if and only if the first Hodge component H 2 (1)(A, A) vanishes (Remark 3.2). In deformation theoretic terms, this corresponds to the fact that A has only trivial commutative deformations. For a projective hypersurface X with restricted structure sheaf A = OXV, the parallel statement is that X is smooth if and only if the first Hodge component H 2 GS(A)1 coincides with its subgroup Eres ∼= H 1(X,TX ) which describes locally trivial scheme deformations of X. In other words, X is smooth if and only if the classical HKR decomposition (1.7) holds for the second Hochschild cohomology group of X (Theorem 4.14). In §4.4, we show that for A = OXV with X a projective hypersurface of dimension ≥ 2, we have (1.9) H 2 GS(A)1 = Emult + Eres, whence we can choose a complement E of Eres inside H 2 we visualize the situation with the aid of the following diagram: GS(A)1 such that E ⊆ Emult. Intuitively, Hodge components: H 2 GS(A)2 H 2 GS(A)1 H 2 GS(A)0 HKR components: H 0 simp(V,∧2T ) s s s s s s E H 1 simp(V,T ) H 2 simp(V,A) (0, f, 0) (0, 0, c) s s s s ys (m, 0, 0) representatives: Remarkably, based upon the results from §4.3, an intertwined 2-class (that is, a class in GS(A)1 \ Emult + Eres) can only exist for a non-smooth projective curve in P2 of degree ≥ 5, H 2 and we give concrete examples of such curves of degree ≥ 6 in §4.5. We also leave the existence of intertwined 2-classes for degree 5 curves as an open question. In §4.6, we study the case when X is a quartic surface in P3 in some detail. We show that the dimension of H 2 GS(A)1 lies between 20 and 32, reaching all possible values except 30 and 31. The minimal value H 2 GS(A)1 = 20 is reached in the smooth case, in which X is a K3 surface GS(A)1 ∼= H 1(X,TX ), as well as in some non-smooth examples like the Kummer surfaces. and H 2 From our general results, we know the dimension of Emult to be one less than the dimension of H 2 In the smooth case we have Emult ⊆ Eres and for the Fermat quartic, we give a concrete description of the (19 dimensional) Emult both by representatives of the form (m, 0, 0) (commutatively deforming the affine pieces) and by equivalent representatives of the form (0, f, 0) GS(A)1. (classical picture of a smooth scheme deformation arising from glueing trivial affine deformations). For non-smooth schemes, we encounter examples in which Eres is one-dimensional (and hence H 2 GS(A)1 = Emult ⊕ Eres) as well as examples in which Emult ∩ Eres 6= 0. Finally, let us mention that the zero-th Hodge component H 2 sional, and we know that the dimension of the second Hodge component H 2 Although our results allow us to compute the dimension of H 2 we have not determined the precise range of this dimension. GS(A)0 is invariably one dimen- GS(A)2 is at least one. GS(A)2 in concrete examples, so far  O  O  O  O  O  O z : z : z : z : z : z :  O  O  O   y     HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 5 Acknowledgement : The authors are grateful to Pieter Belmans for his interesting comments and questions concerning an earlier version of the paper, which led to the discovery of an error in §4.5 that is corrected in the current version. 2. Mixed complexes associated to orthogonal sequences This section is self-contained. In order to make preparations for future computations, we construct several complexes which are related to Koszul complexes, as well as quasi-isomorphisms between them. Let R be a commutative ring, and let uuu = (u0 . . . , un), vvv = (v0, . . . , vn) be two sequences in R. We call (uuu, vvv) a pair of orthogonal sequences of length n (an n-POS) if uivi = 0 nXi=0 holds in R. Let (K•(uuu; R), ∂uuu) be the Koszul cochain complex determined by uuu, namely, K•(uuu; R) is the DG R-algebra ∧•(Re0 ⊕ ··· ⊕ Ren) with ei = −1 and ∂uuu(ei) = ui. Similarly, let (K•(vvv; R), ∂vvv) = ∧•(Rf0 ⊕ ··· ⊕ Rfn) be the Koszul chain complex determined by vvv. Apply- ing HomR(−, R) to K•(vvv; R), we obtain a cochain complex Hom• R(K•(vvv; R), R) whose terms are R (K•(vvv; R), R) = HomR(Kp(vvv; R), R) = M0≤i1<···<ip≤n R(fi1 ∧ ··· ∧ fip )∗ Hom−p and whose differentials are (∂vvv)∗ : Hom−p R (K•(vvv; R), R) −→ Hom−p−1 (fi1 ∧ ··· ∧ fip )∗ 7−→ R nXj=0 (K•(vvv; R), R) vj(fj ∧ fi1 ∧ ··· ∧ fip )∗. Since for each p, the correspondence ei1∧···∧eip ←→ (fi1∧···∧fip )∗ establishes an isomorphism R (K•(vvv; R), R) in a natural way. The differentials (∂vvv)∗ induce between K−p(uuu; R) and Hom−p another complex structure on K•(uuu; R) given by ∂vvv : K−p(uuu; R) −→ K−p−1(uuu; R) ei1 ∧ ··· ∧ eip 7−→ vjej ∧ ei1 ∧ ··· ∧ eip . nXj=0 Remark 2.1. (K•(uuu; R), ∂vvv) is isomorphic to the Koszul complex determined by the sequence vvv⋆ = (v0,−v1, . . . , (−1)nvn). Lemma 2.1. K•(uuu, vvv; R) = (K•(uuu; R), ∂uuu, ∂vvv) is a mixed complex. Proof. Let us verify the equality ∂uuu∂vvv + ∂vvv∂uuu = 0. On one hand, ∂uuu(vj ej ∧ ei1 ∧ ··· ∧ eip ) ∂uuu∂vvv(ei1 ∧ ··· ∧ eip ) = = = = nXj=0 nXj=0 nXj=0 nXj=0 pXk=1 (−1)kuik ej ∧ ei1 ∧ ··· ∧ceik ∧ ··· ∧ eip(cid:19) vj(cid:18)ujei1 ∧ ··· ∧ eip + pXk=1 pXk=1 nXj=0 (−1)kuik vjej ∧ ei1 ∧ ··· ∧ceik ∧ ··· ∧ eip ujvj ei1 ∧ ··· ∧ eip + (−1)kuik vjej ∧ ei1 ∧ ··· ∧ceik ∧ ··· ∧ ··· ∧ eip . 6 LIYU LIU AND WENDY LOWEN On the other hand, ∂vvv∂uuu(ei1 ∧ ··· ∧ eip ) = ∂vvv(cid:18) pXk=1 pXk=1 nXj=0 = = (−1)k−1uik ei1 ∧ ··· ∧ceik ∧ ··· ∧ eip(cid:19) vj ej ∧ ei1 ∧ ··· ∧ceik ∧ ··· ∧ eip (−1)k−1uik vjej ∧ ei1 ∧ ··· ∧ceik ∧ ··· ∧ eip . (−1)k−1uik pXk=1 nXj=0 Thus ∂uuu∂vvv + ∂vvv∂uuu = 0 is true. (cid:3) This mixed complex gives rise to a double complex K•,•(uuu, vvv; R) in the first quadrant as in Figure 1. For r ∈ N, define τ rK•,•(uuu, vvv; R) to be the quotient double complex of K•,•(uuu, vvv; R) consisting of all entries whose coordinates satisfy 0 ≤ q ≤ r. ... ∂vvv ... ∂vvv ... ∂vvv . . . ... ∂vvv K−3(uuu, vvv; R) ∂uuu / K−2(uuu, vvv; R) ∂uuu / K−1(uuu, vvv; R) ∂uuu / K0(uuu, vvv; R) ∂vvv ∂vvv ∂vvv K−2(uuu, vvv; R) ∂uuu / K−1(uuu, vvv; R) ∂uuu / K0(uuu, vvv; R) ∂vvv ∂vvv K−1(uuu, vvv; R) ∂uuu / K0(uuu, vvv; R) ∂vvv K0(uuu, vvv; R) Figure 1. Double complex K•,•(uuu, vvv; R) Suppose that vt is invertible for some t ∈ {0, 1, . . . , n}. Let www = (u0, . . . ,but, . . . , un), and (K•(www; R), ∂www) be the corresponding Koszul complex. Define ι : K•(www; R) → K•(uuu; R) to be the canonical embedding morphism, and define π : K•(uuu; R) → K•(www; R) by ei1 ∧ ··· ∧ eip , if none of ij is t, −Xk6=t vkv−1 t ei1 ∧ ··· ∧ eij−1 ∧ ek ∧ eij+1 ∧ ··· ∧ eip , if t = ij for some j. π(ei1 ∧ ··· ∧ eip ) = for each p. Lemma 2.2. π : K•(uuu; R) → K•(www; R) is a morphism of complexes. Proof. It suffices to prove ∂wwwπ(ei1 ∧ ··· ∧ eip ) = π∂uuu(ei1 ∧ ··· ∧ eip ) when t = ij for some j. We have ∂wwwπ(ei1 ∧ ··· ∧ eip ) = −∂www(cid:18)Xk6=t = −Xk6=t vkv−1 vkv−1 t ei1 ∧ ··· ∧ eij−1 ∧ ek ∧ eij+1 ∧ ··· ∧ eip(cid:19) t ∂www(ei1 ∧ ··· ∧ eij−1 ∧ ek ∧ eij+1 ∧ ··· ∧ eip ) / O O / O O / O O O O / O O / O O O O / O O O O O O HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 7 t (cid:18)Xl6=j vkv−1 (−1)l−1uilei1 ∧ ··· ∧ ek ∧ ··· ∧ceil ∧ ··· ∧ eip = −Xk6=t + (−1)j−1ukei1 ∧ ··· ∧ bek ∧ ··· ∧ eip(cid:19) =Xk6=tXl6=j t ei1 ∧ ··· ∧ ek ∧ ··· ∧ceil ∧ ··· ∧ eip +Xk6=t t ei1 ∧ ··· ∧ bek ∧ ··· ∧ eip =Xk6=tXl6=j t ei1 ∧ ··· ∧ ek ∧ ··· ∧ceil ∧ ··· ∧ eip − (−1)jutei1 ∧ ··· ∧bet ∧ ··· ∧ eip . (−1)luilvkv−1 (−1)jukvkv−1 (−1)luilvkv−1 On the other hand, we have π∂uuu(ei1 ∧ ··· ∧ eip ) = π(cid:18) pXl=1 (−1)l−1uilei1 ∧ ··· ∧ceil ∧ ··· ∧ eip(cid:19) =Xl6=j (−1)luilXk6=t + (−1)j−1utei1 ∧ ··· ∧ceij ∧ ··· ∧ eip . vkv−1 t ei1 ∧ ··· ∧ ek ∧ ··· ∧ceil ∧ ··· ∧ eip This finishes the proof that ∂wwwπ = π∂uuu. (cid:3) Lemma 2.3. For all p, the sequence 0 −→ K−p+1(www; R) ∂vvv ι−−→ K−p(uuu; R) π−−→ K−p(www; R) −→ 0 is split exact. Proof. First of all, let us check that this is indeed a complex, namely, π∂vvvι = 0. Consider the base element ei1 ∧ ··· ∧ eip−1 in K−p+1(www; R). Suppose ij−1 < t < ij. We have vkek ∧ ei1 ∧ ··· ∧ eip−1 + vtet ∧ ei1 ∧ ··· ∧ eip−1(cid:19) vkek ∧ ei1 ∧ ··· ∧ eip−1 −Xk6=t = π(cid:18)Xk6=t =Xk6=t t ek ∧ ei1 ∧ ··· ∧ eip−1 π∂vvvι(ei1 ∧ ··· ∧ eip−1) vtvkv−1 = 0. Next, we consider the map id−ιπ. By the definition of π, if none of ij is t, then (id−ιπ)(ei1 ∧ ··· ∧ eip ) = 0; if t = ij, then (id−ιπ)(ei1 ∧ ··· ∧ eip ) = ei1 ∧ ··· ∧ eip +Xk6=t vkv−1 t ei1 ∧ ··· ∧ eij−1 ∧ ek ∧ eij+1 ∧ ··· ∧ eip t ei1 ∧ ··· ∧ eij−1 ∧ ek ∧ eij+1 ∧ ··· ∧ eip = (−1)j−1vkv−1 t ek ∧ ei1 ∧ ··· ∧ceij ∧ ··· ∧ eip t ei1 ∧ ··· ∧ceij ∧ ··· ∧ eip ). It follows that there exists a map ζ : K−p(uuu; R) → K−p+1(www; R) given by t ei1 ∧ ··· ∧ceij ∧ ··· ∧ eip , ζ(ei1 ∧ ··· ∧ eip ) =(0, (−1)j−1v−1 if none of ij is t, if t = ij for some j, = vkv−1 nXk=0 nXk=0 = ∂vvv((−1)j−1v−1 8 LIYU LIU AND WENDY LOWEN which satisfies ∂vvvιζ + ιπ = id. Moreover, πι = id, ζ∂vvvι = id. These facts indicate that the complex is split exact. (cid:3) Let τ ≥r be the stupid truncation functor. Since the top row of τ rK•,•(uuu, vvv; R) is the same as τ ≥0(K•(uuu; R)[−r]), we define the morphism ιt,(r) associated to t as the composition of τ ≥r(K•(www; R)[−2r]) ι−−→ τ ≥r(K•(uuu; R)[−2r]) ֒→ Tot(τ rK•,•(uuu, vvv; R)). Sometimes we suppress the subscript t in ιt,(r) if no confusion arises. Proposition 2.4. For any r ≥ 0, ι(r) : τ ≥r(K•(www; R)[−2r]) → Tot(τ rK•,•(uuu, vvv; R)) is a quasi- isomorphism with a quasi-inverse π(r) induced by π. Proof. By Lemmas 2.2, 2.3, the sequence 0 −→ K•(www; R)[1 − 2r] (−1)•∂vvv ι −−−−−−→ K•(uuu; R)[−2r] π−−→ K•(www; R)[−2r] −→ 0 of cochain complexes is exact. After shifting degrees, we have another exact sequence 0 −→ K•(www; R)[2 − 2r] (−1)•−1∂vvv ι −−−−−−−→ K•(uuu; R)[1 − 2r] π−−→ K•(www; R)[1 − 2r] −→ 0. Since ∂vvvιζ + ιπ = id (see the proof of Lemma 2.3), we have (−1)•∂vvvιπ = (−1)•∂vvv(id −∂vvvιζ) = (−1)•∂vvv. So the above two exact sequences are combined into a new one 0 −→ K•(www; R)[2 − 2r] Continuing the procedure, we obtain a long exact sequence −−−−−→ K•(uuu; R) π−−→ K•(www; R)[−2r] −→ 0. (−1)•∂vvv (−1)•−1∂vvv ι −−−−−−−→ K•(uuu; R)[1 − 2r] ··· −→ K•(uuu; R)[2 − 2r] (−1)•−1∂vvv −−−−−−−→ K•(uuu; R)[1 − 2r] (−1)•∂vvv −−−−−→ K•(uuu; R)[−2r] π ։ K•(www; R)[−2r]. Let the functor τ ≥r act on the long sequence, and then by using the sign trick, we make all the terms except the last one (i.e. τ ≥r(K•(www; R)[−2r])) into a double complex. It is obvious that the resulting double complex is nothing but τ rK•,•(uuu, vvv; R). Therefore, π induces a quasi-isomorphism π(r) : Tot(τ rK•,•(uuu, vvv; R)) −→ τ ≥r(K•(www; R)[−2r]) which is quasi-inverse to ι(r). (cid:3) Definition 2.1. An n-POS (uuu, vvv) is said to be proportional to another one (uuu′, vvv′) if there exist invertible λ, µ ∈ R such that (uuu′, vvv′) = (λuuu, µvvv). Notice that the (p, q)-entry of τ rK•,•(uuu, vvv; R) (resp. τ rK•,•(uuu′, vvv′; R)) is Kp−q(uuu, vvv; R) (resp. Kp−q(uuu′, vvv′; R)), and that Kp−q(uuu, vvv; R) and Kp−q(uuu′, vvv′; R) share the same rank as free R-modules. There are isomorphisms given by the multiplication by λpµq for all p, q, and they constitute an isomorphism λpµq : Kp−q(uuu, vvv; R) −→ Kp−q(uuu′, vvv′; R) (2.1) ξ(r) : τ rK•,•(uuu, vvv; R) −→ τ rK•,•(uuu′, vvv′; R) of double complexes. The induced isomorphism between their total complexes is denoted by ξTot (r) . HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 9 3. Hochschild cohomology of affine hypersurfaces Let A = k[y1, . . . , yn]/(G) be the quotient of the polynomial algebra k[y1, . . . , yn] by a unique relation G. There are several papers concerning the Hochschild and cyclic (co)homology of A, the treatment of the topic dating back to Wolffhardt's work on Hochschild homology of (analytic) complete intersections [24]. We base our exposition on the more recent papers [2], [20]. In [20], Michler describes the Hochschild homology groups of A as well as their Hodge decompositions when G is reduced, based on the cotangent complex of A. The Hochschild cohomology groups are not treated in [20]. In [2], the authors from BACH construct a nice finitely generated free resolution R•(A) of A over Ae under an additional condition on G. For the normalized bar resolution ¯Cbar (A), the authors give comparison maps • (3.1) and α′ : R•(A) → ¯Cbar authors compute the Hochschild homology and cohomology of A. (A) → R•(A) • • α : ¯Cbar (A) satisfying αα′ = id. By virtue of the smaller resolution R•(A), the From now on, we assume that G = G(y1, . . . , yn) has leading term yd 1 with respect to the lexicographic ordering y1 > ··· > yn. Under this assumption, we are able to use the resolution R•(A) from [2] and obtain the Hochschild cohomology groups as H p(A, A) = H p(L•(A)) where L•(A) = HomAe(R•(A), A). In this section, we first make the complex L•(A) explicit according to [2]. Next we restate L•(A) in terms of the cotangent complex, inspired by [20]. Finally, Hochschild cohomology of localizations of A is considered. By the construction of [2], L• (A) = ∧•(Ae1 ⊕ ··· ⊕ Aen) and L•(A) is the algebra of divided (A) in one variable s. Put ei = 1 and s(j) = 2j, L•(A) is made into a DG powers over L• A-algebra whose differential is given by ei 7→ (∂G/∂yi)s(1) and s(1) 7→ 0. By writing ei1...il instead of the product ei1 ∧ ··· ∧ eil , we have Lp(A) = M0≤j≤p/2 1≤i1<···<ip−2j ≤n Aei1...ip−2j s(j), and the differential Lp(A) → Lp+1(A) is given by ei1...ip−2j s(j) 7−→ (−1)l−1 ∂G ∂yil ei1... bil...ip−2j s(j+1). p−2jXl=1 It immediately follows that the A-module complex L•(A) admits a decomposition L•(A) = ⊕r∈NL•(A)r with (3.2) L•(A)r = τ ≥r(K•((∂G/∂yi)1≤i≤n; A)[−2r]). As stated in [20] (also see [13, Ch. III, Prop. 3.3.6]), the cotangent complex LA/k of A, which is unique up to homotopy equivalence, is given by 0 −→ Adz δ−−→ Adyi −→ 0 nMi=1 where the two nonzero terms sit in degrees −1 and 0 respectively, dz and dyi are base elements and nXi=1 Note that by [8], H p(A, A) has the Hodge decomposition ⊕r∈NH p A (∧r LA/k, A). (r)(A, A) ∼= Extp−r H p ∂G ∂yi δ(dz) = dyi. (r)(A, A), and the component 10 LIYU LIU AND WENDY LOWEN By [12, Ch. VIII, Cor. 2.1.2.2], ∧r LA/k is isomorphic to a complex determined by δ in the derived category Db(A), more explicitly, ∧r LA/k ∼= Mi+j=r ∧i(Ady1 ⊕ ··· ⊕ Adyn) ⊗A Γ j(Adz) where Γ j(−) is the degree j component of the divided power functor over A.1 It follows that Extp−r A (∧r LA/k, A) is the (p − r)-th cohomology group of (3.3) HomA(∧r LA/k, A) ∼= Mi+j=r ∧i(A(dy1)∗ ⊕ ··· ⊕ A(dyn)∗) ⊗A Γ j(A(dz)∗) Notice that the j-th term of the right-hand side of (3.3) is free of rank(cid:0) n r−j(cid:1) which is the same as τ ≥0(K•((∂G/∂yi)1≤i≤n; A)[−r]) for all 0 ≤ j ≤ r. By taking into account the differentials, we can construct an isomorphism HomA(∧r LA/k, A) ∼= τ ≥0(K•((∂G/∂yi)1≤i≤n; A)[−r]). Equivalently, HomA(∧r LA/k, A)[−r] ∼= L•(A)r by (3.2) and consequently the isomorphism L•(A) ∼=Mr∈N (r)(A, A) ∼= H p(HomA(∧r LA/k, A)[−r]) ∼= H p(L•(A)r), and the holds true in Db(A). Therefore, H p decomposition of H p(L•(A)) deduced from [2] actually corresponds to the Hodge decomposition of H p(A, A). HomA(∧r LA/k, A)[−r] Observe that the comparison map α from (3.1) gives rise to a quasi-isomorphism β : L•(A) → ¯C•(A, A) landing in the normalized Hochschild complex of A. The morphism β, whose explicit (r)(A, A) ∼= H p(L•(A)r). For our expression will be given later on, induces the isomorphism H p purpose, we first introduce some cochains. Note that the algebra A has a basis BA = {yp1 1 yp2 2 ··· ypn n 0 ≤ p1 ≤ d − 1, p2, . . . , pn ∈ N}. We define for 1 ≤ l ≤ n a normalized 1-cochain ◦∂/∂yl by = plyp1 (3.4) n = f 7−→ ◦∂f ∂yl 1 ··· ypl−1 l−1 ypl−1 l ypl+1 l+1 ··· ypn n 1 yp2 BA ∋ yp1 2 ··· ypn and a normalized 2-cochain ◦µ by ◦µ(f, g) =(0, (3.5) for an additional g = yq1 yp2+q2 2 yp1+q1−d 1 n ∈ BA. One can easily check that ◦µ is a 2-cocycle. ··· ypn+qn n 1 yq2 2 ··· yqn p1 + q1 < d, , p1 + q1 ≥ d. Now we give the expression of β =Pr β(r) : L•(A) → ¯C•(A, A): 2 ) β(p−j)(ei1...ip−2j s(j)) = (−1)(p−2j ◦∂ ∂yi1 ∪ ··· ∪ (3.6) ◦∂ ∂yip−2j ∪ ◦µ∪j. The notation ∪, not to be confused with the well-known cup product, is defined as P1 ∪ P2 ∪ ··· ∪ Pm = 1 m! Xσ∈Sm (−1)cµ ◦(cid:0)Pσ−1(1) ⊗ Pσ−1(2) ⊗ ··· ⊗ Pσ−1(m)(cid:1) where Pi ∈ ¯C•(A, A), µ is the multiplication map (or rather its unique extension by associativity to an m-ary multiplication map) and c = #{(i, j) i < j, σ−1(i) > σ−1(j), Pi, Pj have odd degrees}. Thus, the operation ∪ becomes supercommutative. For example, ◦∂ ∂yi ∪ ◦µ = 1 2 µ ◦(cid:18) ◦∂ ◦∂ ∂yi ∪ ∂yi ⊗ ◦µ + ◦µ ⊗ ◦∂ ◦∂ ∂yj ∪ ∂yj = − ◦∂ ∂yi . ◦∂ ∂yi(cid:19) = ◦µ ∪ ◦∂ ∂yi , 1Upright Γ(X, −) will denote the global section functor on a scheme X in §4. HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 11 Remark 3.1. Since β(r)(L•(A)r) ⊆ ¯C•(A, A)r, we also call L•(A) = ⊕r∈NL•(A)r the Hodge de- composition. Remark 3.2. Recall that the vanishing of the groups H 2 (1)(A, M ) for all A-modules M characterizes smoothness of A. By [15, Thm. 5.3], the condition is in turn equivalent to the vanishing of the (1)(A, A), i.e. H 2(L•(A)1) = 0. It follows that A is smooth if and only if the ideal single group H 2 (∂G/∂y1, . . . , ∂G/∂yn) is equal to A itself. Let ¯A be the localization of A at a multiplicatively closed set generated by yt1, . . . , yth where 2 ≤ t1 < ··· < th ≤ n. Let σ : ¯A → B be a morphism of commutative algebras such that B is a flat ¯A-module via σ. Then ¯A has a basis B ¯A = {yp1 1 yp2 2 ··· ypn n 0 ≤ p1 ≤ d − 1, pt1, . . . , pth ∈ Z, other pi ∈ N}. As above, cochains ◦∂/∂yl ∈ ¯C1( ¯A, ¯A) and ◦µ ∈ ¯C2( ¯A, ¯A) can be defined similarly. After compos- ing them with σ, we obtain cochains in ¯C1( ¯A, B), ¯C2( ¯A, B). Furthermore, one can easily check that there is a quasi-isomorphism β : B ⊗A L•(A) → ¯C•( ¯A, B) whose expression is similar to the one shown in (3.6). 4. Hochschild cohomology of projective hypersurfaces For any morphism X → Y of schemes or analytic spaces, Buchweitz and Flenner introduce the Hochschild complex HX/Y of X over Y [5], and they deduce an isomorphism HX/Y ∼= S(LX/Y [1]) in the derived category D(X) where LX/Y denotes the cotangent complex of X over Y and S(LX/Y [1]) is the derived symmetric algebra [4]. As a consequence, there is a decomposition of Hochschild cohomology in terms of the derived exterior powers of the cotangent complex (4.1) HH i(X/Y ) ∼= Mp+q=i Extp X (∧q LX/Y ,OX ) which generalizes the HKR decomposition in the smooth case. Around the same time, Schuh- macher also deduced the decomposition (4.1) using a different method [22]. In general, it may be hard to compute the right hand side of (4.1), but in some special situations, LX/Y has a very nice expression. For example, in [1, Expose VIII] Berthelot defines LX/Y as a complex concentrated in two degrees when X → Y factors as a closed immersion X → X ′ followed by a smooth morphism X ′ → Y . In particular, Berthelot's definition can be applied to the case when Y = Spec k and X is a projective hypersurface over k. Although LX/k admits a very simple expression in this case, we do not use it for our computation. As a sequel to [6], [18], we compute HH i(X) starting from the Gerstenhaber-Schack complex, since a deformation interpretation of Gerstenhaber-Schack 2-cocycles is at hand [6]. In §4.1 we construct a series of complexes of OX - modules, as well as morphisms from their associated Cech complexes to the respective components of the normalized reduced Gerstenhaber-Schack complex ¯C′ GS(OXV) (for a chosen covering V). Using the technique from §2, we prove that these maps are quasi-isomorphisms. Due to the theoretical significance of the cotangent complex, we give an expression of LX/k in terms of twisted structure sheaves OX (l) in §4.2 when X is a projective hypersurface. This allows us to explain directly how our results agree with Buchweitz and Flenner's. In §4.3, we prove our main theorem Theorem 1.1, providing a computation of the Hochschild cohomology groups of a projective hypersurface of degree d in Pn in terms of easier complexes. The result makes a basic distinction between the case d > n + 1, the harder case d = n + 1 and the easier case d ≤ n. 12 LIYU LIU AND WENDY LOWEN Based upon our computations in §4.3, we prove in §4.4 that a projective hypersurface is smooth if and only if the HKR decomposition of the second Hochschild cohomology group (1.7) holds (Theorem 4.14). This can be seen as an analogue of the characterization of smoothness of affine hypersurfaces (Remark 3.2). Recall that by definition of the GS complex, we have C2 GS(A) = C0,2(A) ⊕ C1,1(A) ⊕ C2,0(A). We call a 2-cocycle (m, f, c) ∈ C2 GS(A) untwined (decomposable in [6]) if (m, 0, 0), (0, f, 0) and (0, 0, c) are all 2-cocycles. A GS 2-class is called intertwined if it has no untwined representative (m, f, c). In §4.5, based upon the results from §4.3 we show that for a projective hypersurface as above if either n 6= 2 or n = 2 and d ≤ 4, no intertwined 2-class exists. We give a family of concrete examples of intertwined 2-class for n = 2 and d ≥ 5. Finally, in §4.6 we pay special attention to the case of quartic surfaces. We show that the GS(A)1 lies between 20 and 32, reaching all possible values except 30 and 31. The GS(A)1 = 20 is reached in the smooth K3 case. We also present an analysis of GS(A)1 is built up from 2-classes of type [(m′, 0, 0)] and 2-classes of type [(0, f ′, 0)], giving dimension of H 2 minimal value H 2 how H 2 explicit computations in concrete examples. 4.1. Construction of quasi-isomorphisms. Let R = k[x0, . . . , xn] and F = F (x0, . . . , xn) be a homogeneous polynomial of degree d ≥ 2. Let S = R/(F ) and X = Proj S ⊆ Pn. Suppose that F has a summand xd 0 when F is uniquely expressed as a sum of nonzero monomials. In this way, X can be covered by U = {Ui = X ∩ {xi 6= 0} 1 ≤ i ≤ n}. Let V = {Vi1...is = Ui1 ∩ ··· ∩ Uis 1 ≤ i1 < ··· < is ≤ n} be the associated covering closed under intersections. For any a p-simplex σ ∈ Np(V), denote its domain and codomain by ⋄σ and σ⋄ respectively. Let C•,•(A) be the Gerstenhaber-Schack double complex where A = OXV, namely, Cp,q(A) = Yσ∈Np(V) Homk(A(σ⋄)⊗q,A(⋄σ)) endowed with the (vertical) product Hochschild differential dHoch and the (horizontal) simplicial differential dsimp. Recall that a cochain f = (fσ) ∈ Cp,q(A) is called normalized if for any p-simplex σ, fσ is normalized, and it is called reduced if fσ = 0 whenever σ is degenerate. Let ¯C′•,•(A) be the normalized reduced sub-double complex of C•,•(A) and ¯C′• GS(A) be the associated total complex. Observe that for 1 ≤ i ≤ n, Ai = A(Ui) = k[y0, . . . ,byi, . . . , yn]/(Gi) where is monic. Here we assign an ordering y0 > ··· > yi−1 > yi+1 > ··· > yn. So we have complexes L•(Ai) as given in §3. Denote by wwwi the sequence , Gi = F (y0, . . . , yi−1, 1, yi+1, . . . , yn) = yd 0 + ··· , . . . , , . . . , ∂Gi ∂yi−1 ∂Gi ∂yi+1 ∂Gi ∂yn(cid:19). (cid:18) ∂Gi Then L•(Ai)r = τ ≥r(K•(wwwi; Ai)[−2r]). ∂y0 For any V ∈ V let Φ(V ) = {t ∈ {1, . . . , n} V ⊆ Ut}. If t ∈ Φ(V ) = {t1, . . . , tm}, we express A(V ) in term of generators and relations as A(V, t) = k[y0, . . . ,byt, . . . , yn, y−1 t1 , . . . ,dy−1 t Since A(V, t) is a localization of At, there is a quasi-isomorphism β : B ⊗At L•(At) −→ ¯C•(A(V, t), B) , . . . , y−1 tm ]/(Gt, yt1y−1 t1 − 1, . . . , ytmy−1 tm − 1). HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 13 for any flat morphism A(V, t) → B by the last paragraph of §3. If s also belongs to Φ(V ), the canonical isomorphism A(V, t) → A(V, s) is denoted by ζt,s. Unfortunately, ζt,s is not compatible with the differentials of L•(At) and L•(As), namely, the square B ⊗At L•(At) ζt,s ¯C•(A(V, t), B) ζ ∗ t,s B ⊗As L•(As) / ¯C•(A(V, s), B) β β fails to be commutative. So one does not expect that the complexes L•(A(V )) for all affine pieces V can be made into a complex L• of sheaves on X equipped with nice restriction maps. The reason is that the L•(A(V ))'s are too small. In order to study A globally, we have to put on their weight. Their "food" should be convenient for computation in principle. It follows from Euler's formula ∂F ∂xi · xi = d · F nXi=0 make up an n-POS in S. Also, there is an n-POS (uuui, vvvi) in Ai: that uuui =(cid:18) ∂Gi ∂y0 where uuu =(cid:18) ∂F ∂x0 , ∂F ∂x1 , . . . , , . . . , ∂Gi ∂yi−1 , Hi, ∂Gi ∂yi+1 , . . . , ∂F ∂xn(cid:19) and vvv = (x0, x1, . . . , xn) ∂yn(cid:19) and vvvi = (y0, . . . , yi−1, 1, yi+1, . . . , yn) ∂Gi Hi = ∂F ∂xi (y0, y1, . . . , yi−1, 1, yi+1, . . . , yn). Since wwwi is the subsequence of uuui by deleting Hi, the results from §2 apply. As before we get the mixed complex K•(uuu, vvv; S) and the double complex K•,•(uuu, vvv; S). Let r ≥ 0 and let us consider τ rK•,•(uuu, vvv; S). We twist the degrees of its entries as in Figure 2 so that it is made into a double complex of graded S-modules. The associated total complex S(r)(n+1 r ) ∂uuu / S(r + d − 1)(n+1 r−1) ∂uuu / / ··· ∂uuu / / S(rd − d + 1)(n+1 1 ) ∂uuu / / S(rd) ∂vvv ∂vvv ∂vvv S(r − 1)(n+1 r−1) ∂uuu / S(r + d − 2)(n+1 r−2) ∂uuu / ··· ∂uuu / S(rd − d) ... ∂vvv ∂vvv ... ∂vvv ∂vvv . . . S(1)(n+1 1 ) ∂uuu S(d) ∂vvv S Figure 2. Double complex τ rK•,•(uuu, vvv; S) gives rise to a complex of sheaves F • r : OX −→ OX (1)n+1 −→ ··· −→ OX (rd − d + 1)n+1 −→ OX (rd). We in turn have double complexes E •,• r and H•,• r r = C′p E p,q simp(V,F q r = C′p(V,F q r = C′p(U,F q r ). r , G•,• rV), Gp,q as follows: r ), Hp,q / /   / O O / / O O / O O / O O O O O O / / O O O O O O 14 LIYU LIU AND WENDY LOWEN Their associated complexes are denoted by E • r respectively. Since V is a refinement of U, we fix a map λ : V → U such that V ⊆ λ(V ) for all V ∈ V. There is a quasi-isomorphism ¯λ : H• r induced by λ, defined by for all f ∈ Hp,q r , r and H• r → G• r , G• Composing ¯λ with the explicit quasi-isomorphism G• isomorphism ¯¯λ : H• r → E • r given by ¯λ(f )Vi0 ...Vip = fλ(Vi0 )...λ(Vip ). r → E • r given in [6], we obtain a quasi- r ¯¯λ(f )Vj0 ⊆···⊆Vjp = fλ(Vj0 )...λ(Vjp ). (4.2) Let ¯C′•,•(A) = ⊕r∈N ¯C′•,• (A) be the Hodge decomposition. Our goal is to construct a family of r → ¯C′•,• morphisms E •,• (A) of double complexes for all r that give rise to quasi-isomorphisms r → ¯C′• E • GS(A) turns out to be isomorphic to the Hochschild cohomology of X (see [18, Thm. 7.8.1]), the cohomology HH •(X) can be computed by H• := ⊕r∈NH• GS(A)r. Since the cohomology of ¯C′• r Let σ ∈ Np(V) be a p-simplex and consider t, s ∈ Φ(σ⋄). We have quasi-isomorphisms τ ≥r(K•(wwwt;A(⋄σ, t))[−2r]) ∼= A(⋄σ, t) ⊗At L•(At) −→ ¯C•(A(σ⋄, t),A(⋄σ, t)) r, namely, HH i(X) ∼= H i(H•). βt : Mr∈N and βs, which is defined similarly. Let ◦∂t/∂yi, ◦µt and ◦∂s/∂yi, ◦µs be the resulting Hochschild cochains as defined in (3.4) and (3.5). According to the generators and relations of A(σ⋄, t) and A(⋄σ, t), we can regard ◦∂t/∂yi, ◦µt to be cochains in ¯C•(A(σ⋄, t),A(⋄σ, t)) by abuse of notation, and similarly for ◦∂s/∂yi, ◦µs. Lemma 4.1. Let ζ′ by ζt,s. Then t,s : ¯C•(A(σ⋄, t),A(⋄σ, t)) → ¯C•(A(σ⋄, s),A(⋄σ, s)) be the isomorphism induced (1) ζ′ (2) ζ′ (3) ζ′ t,s(◦µt) = yd t,s(◦∂t/∂yi) = yt · ◦∂s/∂yi if i 6= t, s. t,s(◦∂t/∂ys) = −Pi6=s ytyi · ◦∂s/∂yi. 0 ··· yps−1 t · ◦µs. Proof. (1) (2) Choose any f = yp0 have ζ′ t,s(cid:18) ◦∂t ∂yi(cid:19)(f ) = ζt,s ◦ s−1 yps+1 s+1 ··· ypn n ∈ BA(σ⋄,s) and let f = Pi6=t,s pi. We ◦∂t ∂yi ◦ ζs,t(f ) ◦∂t 0 ··· ypt−1 (yp0 = ζt,s ◦ ∂yi 0 ··· ypi−1 = ζt,s(piyp0 0 ··· ypi−1 = piyp0 ◦∂s ∂yi = yt ··· ypt+1 (f ) i i s t−1 ypt+1 ··· ypt−1 t+1 ··· y−f t−1 ypt+1 ··· yps−1 ··· ypn n ) t+1 ··· y−f ··· ypn n ) s−1 yps+1 s+1 ··· ypn n s t for all i 6= t, s, and ζ′ s 0 ··· ypt−1 (yp0 0 ··· ypt−1 t−1 ypt+1 t+1 ··· y−f t+1 ··· y−f −1 s−1 yps+1 s+1 ··· ypn t−1 ypt+1 ··· yps−1 n s 0 ··· ypt+1 t ··· ypn n ) ··· ypn n ) ◦∂t ∂ys t,s(cid:18) ◦∂t ∂ys(cid:19)(f ) = ζt,s ◦ = ζt,s(−fyp0 = −fyp0 = −ytff = −Xi6=s ytyi ◦∂s ∂yi (f ). HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 15 0 ··· yqs−1 s−1 yqs+1 (3) Let g = yq0 t,s(◦µt)(f, g) = ζt,s ◦ ◦µt(yp0 ζ′ = ζt,s(yp0+q0−d = yp0+q0−d 0 = yd 0 s+1 ··· yqn n ∈ BA(σ⋄,s) and g =Pi6=t,s qi. Assume p0 + q0 ≥ d. Then t−1 ypt+1 ··· yqn n ) ··· ypt−1+qt−1 n , yq0 ··· ypn ··· y−f −g t−1 yqt+1 ··· ypn+qn t+1 ··· y−g 0 ··· yqt−1 0 ··· ypt−1 t−1 ) n s s s t+1 ··· y−f ypt+1+qt+1 t+1 ··· yps−1+qs−1 s−1 yps+1+qs+1 s+1 ··· ypn+qn n ··· ypt+qt+d t t · ◦µs(f, g). t,s(◦µt)(f, g) = 0 = yd On the other hand, ζ′ t · ◦µs(f, g) trivially holds if p0 + q0 < d. There are proportional n-POS (ζt,s(uuut), ζt,s(vvvt)), (uuus, vvvs) in A(⋄σ, s) with uuus = yd−1 t vvvs = ytζt,s(vvvt). There is an isomorphism (cid:3) ζt,s(uuut) and t,s,(r) : Tot(τ rK•,•(ζt,s(uuut), ζt,s(vvvt);A(⋄σ, s))) −→ Tot(τ rK•,•(uuus, vvvs;A(⋄σ, s))) ξTot as given in (2.1). Since the t-th, s-th components of ζt,s(vvvt) and vvvs are invertible, we have the diagram (4.3) Tot(τ rK•,•(ζt,s(uuut), ζt,s(vvvt);A(⋄σ, s))) πt,(r) / τ ≥r(K•(ζt,s(wwwt);A(⋄σ, s))[−2r]) ξTot t,s,(r) β ′ t,(r) ¯C•(A(σ⋄, s),A(⋄σ, s)) βs,(r) Tot(τ rK•,•(uuus, vvvs;A(⋄σ, s))) πs,(r) / τ ≥r(K•(wwws;A(⋄σ, s))[−2r]) where β′ t,(r) is induced by βt,(r) and ζt,s. Lemma 4.2. The diagram (4.3) is commutative. Proof. Choose any base element E = ei1 ∧··· ∧ eip ∈ Kp(ζt,s(uuut), ζt,s(vvvt);A(⋄σ, s)). When viewed as a cochain in τ rK•,•(ζt,s(uuut), ζt,s(vvvt);A(⋄σ, s)), E locates in position (r − p, r). So ξTot t,s,(r)(E) = (yd−1 E. Let us prove the lemma by a case-by-case argument. )r−pyr t t E = yr+(d−1)(r−p) If t, s /∈ {i1, . . . , ip}, then t t,(r) ◦ πt,(r)(E) = β′ β′ t,(r)(ei1...ip s(r−p)) 2)ζ′ = ζt,s ◦ (−1)(p = (−1)(p = (−1)(p = yr+(d−1)(r−p) 2)yt t ∂yi1 ∪ ··· ∪ 2)(cid:18) ◦∂t t,s(cid:18) ◦∂t ∂yi1(cid:19) ∪ ··· ∪ ζ′ t ◦∂t ∂yip ∪ ◦µ∪(r−p) ∂yip(cid:19) ∪(cid:0)ζ′ t,s(cid:18) ◦∂t ◦∂s ∂yip ∪ (yd ◦∂s ∂yi1 ∪ ··· ∪ yt ◦∂s 2) (−1)(p ∂yi1 ∪ ··· ∪ βs,(r)(ei1...ip s(r−p)) t · ◦µs)∪(r−p) ◦∂s ∂yip ∪ ◦µ∪(r−p) s (cid:19) ◦ (ζs,t)⊗(2r−p) t,s(◦µt)(cid:1)∪(r−p) = yr+(d−1)(r−p) = βs,(r) ◦ πs,(r) ◦ ξTot If s /∈ {i1, . . . , ip} and t = ij for some j, then t t,s,(r)(E). β′ t,(r) ◦ πt,(r)(E) = −Xm6=t ymy−1 t ζt,s ◦ (−1)(p ◦∂t ∂yip ∪ ◦µ∪(r−p) 2)ζ′ ymy−1 (−1)(p ∂yi1 ∪ ··· ∪ 2)(cid:18) ◦∂t (cid:19) ◦ (ζs,t)⊗(2r−p) ∂yi1(cid:19) ∪ ··· ∪ ζ′ t,s(cid:18) ◦∂t t t ∪ ··· ∪ = −Xm6=t ◦∂t ∂yij−1 ∪ ◦∂t ∂ym ∪ ◦∂t ∂yij+1 t,s(cid:18) ◦∂t ∂ym(cid:19) ∪ ···   /   / O O 16 LIYU LIU AND WENDY LOWEN ◦∂s ∂ym ∪ ··· ∪ ◦∂s ∂yip t ymyr+(d−1)(r−p)−1 t,s(◦µt)(cid:1)∪(r−p) 2) (−1)(p t,s(cid:18) ◦∂t ∂yip(cid:19) ∪(cid:0)ζ′ ∪ ζ′ = − Xm6=t,s ∪ (◦µs)∪(r−p) − yr+(d−1)(r−p)−2 ∪(cid:18)−Xi6=s ◦∂s ∂yip ∪ (◦µs)∪(r−p) ◦∂s ∂yi1 ∪ ··· ∪ (−1)(p 2) ytyi ◦∂s t ◦∂s ∂yi1 ∪ ··· ∂yi(cid:19) ∪ ··· ∪ ◦∂s (−1)(p 2) ∂yi1 ∪ ··· ∪ βs,(r)(ei1...ip s(r−p)) = yr+(d−1)(r−p) t t = yr+(d−1)(r−p) = βs,(r) ◦ πs,(r) ◦ ξTot t,s,(r)(E). ◦∂s ∂yt ∪ ··· ∪ ◦∂s ∂yip ∪ (◦µs)∧(r−p) If t /∈ {i1, . . . , ip} and s = il for some l, then t,(r) ◦ πt,(r)(E) = (−1)(p β′ 2)ζ′ t,s(cid:18) ◦∂t ∂yi1(cid:19) ∪ ··· ∪ ζ′ 2) (−1)(p ∂yip(cid:19) ∪(cid:0)ζ′ t,s(cid:18) ◦∂t ∂yi1 ∪ ··· ∪(cid:18)−Xm6=s ◦∂s t,s(◦µt)(cid:1)∪(r−p) ∂ym(cid:19) ∪ ··· ytym ◦∂s = yr+(d−1)(r−p)−1 t t s ∪ = −yr+(d−1)(r−p) ◦∂s ∂yip ∪ ◦µ∧(r−p) Xm6=s Xm6=s = −yr+(d−1)(r−p) = βs,(r) ◦ πs,(r) ◦ ξTot ∪ ◦µ∪(r−p) s t t,s,(r)(E). ym(−1)(p 2) ◦∂s ∂yi1 ∪ ··· ∪ ◦∂s ∂ym ∪ ··· ∪ ◦∂s ∂yip ymβs,(r)(ei1...m...ip s(r−p)) If t = ij and s = il for some j, l then β′ t,(r) ◦ πt,(r)(E) = −Xm6=t ◦∂t ∂yij−1 ∪ ◦∂t ∂ym ∪ ◦∂t ∂yij+1 t t ymy−1 ymy−1 ∪ ··· ∪ ∂yi1 ∪ ··· ∪ 2)(cid:18) ◦∂t t ζt,s ◦ (−1)(p ◦∂t ◦∂t ∂yip ∪ ◦µ∪(r−p) ∂ys ∪ ··· ∪ ∂yi1(cid:19) ∪ ··· ∪ ζ′ t,s(cid:18) ◦∂t = − Xm6=t,s 2)ζ′ (−1)(p ∂yip(cid:19) ∪(cid:0)ζ′ t,s(cid:18) ◦∂t ∂ys(cid:19) ∪ ··· ∪ ζ′ t,s(cid:18) ◦∂t ∪ ζ′ = − Xm6=t,s (−1)(p 2) ∪(cid:18)−Xi6=s ◦∂s ∂yip ∪ (◦µs)∪(r−p) ◦∂s ∂yi1 ∪ ··· ∪ ymyr+(d−1)(r−p)−2 ytyi t (cid:19) ◦ (ζs,t)⊗(2r−p) t,s(cid:18) ◦∂t ∂ym(cid:19) ∪ ··· t,s(◦µt)(cid:1)∪(r−p) ◦∂s ∂ym ∪ ··· ymyi(−1)(p 2) ◦∂s ∂yi1 ∪ ··· ∪ ◦∂s ∂ym ∪ ··· ◦∂s ∂yi(cid:19) ∪ ··· ∪ Xm6=t,sXi6=s ◦∂s ∂yip ∪ (◦µs)∪(r−p) Xm6=t,s ymyt(−1)(p 2) = yr+(d−1)(r−p)−1 t ∪ ◦∂s ∂yi ∪ ··· ∪ = yr+(d−1)(r−p)−1 t ◦∂s ∂yi1 ∪ ··· ∪ ◦∂s ∂ym ∪ ··· HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 17 ∪ ◦∂s ∂yt ∪ ··· ∪ = −yr+(d−1)(r−p) t ◦∂s ∂yip ∪ (◦µs)∪(r−p) Xm6=s ◦∂s ∂yi1 ∪ ··· ∪ ym(−1)(p 2) ··· ∪ ◦∂s ∂yip ∪ (◦µs)∪(r−p) t Xm6=s = −yr+(d−1)(r−p) = βs,(r) ◦ πs,(r) ◦ ξTot t,s,(r)(E). ymβs,(r)(ei1...m...ip s(r−p)) ◦∂s ∂yt ∪ ··· ∪ ◦∂s ∂ym∪ (cid:3) Therefore we obtain a commutative diagram Mr∈N Tot(τ rK•,•(uuut, vvvt;A(⋄σ, t))) βt◦πt / ¯C•(A(σ⋄, t),A(⋄σ, t)) ζt,s ζ ′ t,s Tot(τ rK•,•(ζt,s(uuut), ζt,s(vvvt);A(⋄σ, s))) β ′ t◦πt / / ¯C•(A(σ⋄, s),A(⋄σ, s)) Mr∈N ξTot t,s Mr∈N Tot(τ rK•,•(uuus, vvvs;A(⋄σ, s))) βs◦πs / ¯C•(A(σ⋄, s),A(⋄σ, s)) t,s = ξTot where the vertical morphisms are isomorphisms and the horizontal ones are quasi-isomorphisms. Let ξ′ t,s ◦ ζt,s. The twisting number r + (d − 1)(r − d) of the (r − p, r)-entry in Figure 2 coincides with the exponent of yt in the proof of Lemma 4.2. This is equivalent to say that ξ′ is the canonical automorphism of F •(⋄σ) if we write A(⋄σ) in terms of different generators and relations. Moreover, it is easy to check the coherence conditions t,s ξ′ s,u ◦ ξ′ t,s = ξ′ t,u, ζ′ s,u ◦ ζ′ t,s = ζ′ t,u hold true for any additional u ∈ Φ(σ⋄). This gives rise to well-defined morphisms γσ = β ◦ π : F •(⋄σ) → ¯C•(A(σ⋄),A(⋄σ)) for all simplices σ ∈ N•(V) which commute with simplicial differentials. Remember that β and π preserve the Hodge decomposition. These facts are summarized as . The morphisms γσ : F •(⋄σ) → ¯C•(A(σ⋄),A(⋄σ)) for all Theorem 4.3. Let E •,• = ⊕r∈NE •,• simplices σ on V constitute a morphism γ : E •,• → ¯C′•,•(A) of double complexes that gives rise to a quasi-isomorphism E • → ¯C′• GS(A). Moreover, γ preserves the Hodge decomposition. r Recall that F is required to contain xd 0 as a summand. We claim that this condition is not too restrictive. In fact, a homogeneous polynomial F ∈ R of degree d always has the expression F (x0, x1, . . . , xn) = Xi1,...,in µi1,...,in xd−i1−···−in 0 xi1 1 ··· xin n where µi1,...,in ∈ k. Let Σ : R → R be the automorphism of graded algebras determined by x0 7→ x0, xj 7→ xj + λjx0 (1 ≤ j ≤ n) where λj are undetermined coefficients. So Σ(F ) = Xi1,...,in = Xi1,...,in µi1,...,in xd−i1−···−in 0 (x1 + λ1x0)i1 ··· (xn + λnx0)in µi1,...,in λi1 1 ··· λin n xd 0 + ··· .   /     / 18 LIYU LIU AND WENDY LOWEN If µ0,...,0 = 0, there exist at least an array (i1, . . . , in) such that µi1,...,in 6= 0 and one of il > 0 for 1 ≤ l ≤ n. So one can choose proper λj making Xi1,...,in µi1,...,in λi1 1 ··· λin n = 1. Therefore, Proj R/(F ) ∼= Proj R/(Σ(F )) with Σ(F ) containing xd 0 as a summand. With this isomorphism, most conclusions in §4 remain true for an arbitrary homogeneous polynomial F . Throughout the paper, to facilitate the computations, we maintain the condition that F contains xd 0 as a summand. 4.2. The cotangent complex of a hypersurface. As stated in the beginning of §4, an explicit expression of LX/Y is given by Berthelot, when X → Y factors as a closed immersion X → X ′ followed by a smooth morphism X ′ → Y . Let us recall it in the special case when Y = Spec k and X = Proj S. Obviously, X ′ can be chosen to be Proj R = Pn and so the factorization X ı−→ Pn → Spec k satisfies the condition. Let O = OPn, and let I ⊂ O be the sheaf of ideals determined by the closed immersion X → Pn. By definition, L0 X/k are all zero, the differential I/I 2 = ı∗I → ı∗ΩPn is induced by I ֒→ O d−→ ΩPn. X/k = I/I 2, and other Lj X/k = ı∗ΩPn , L−1 Note that there is a complex of graded modules 0 −→ S(−d) ∂uuu−−→ S(−1)n+1 ∂vvv−−→ S −→ 0 concentrated in degrees −1, 0 and 1. It gives rise to a complex 0 −→ OX (−d) ∂uuu−−→ OX (−1)n+1 ∂vvv−−→ OX −→ 0 It is easy to check that the complex is the same as F •∨ of OX -modules. HomOX (−,OX ). Proposition 4.4. There is a quasi-isomorphism LX/k → F •∨ 1 [−1]. 1 [−1] where (−)∨ = Proof. First of all, the sheaf of ideals I corresponds to the principal ideal (F ). So as OX -modules, I/I 2 is isomorphic to OX (−d) since deg F = d. Next, there is an exact sequence (4.4) 0 −→ ΩPn −→ O(−1)n+1 ∂vvv−−→ O −→ 0 by [11, Thm. 8.13]. Since ı∗ is right exact, we have an exact sequence ı∗ΩPn −→ OX (−1)n+1 ∂vvv−−→ OX −→ 0. We claim that the map ı∗ΩPn → OX (−1)n+1 is injective. localizing (4.4) at the point ı(x) we obtain a split exact sequence In fact, for any point x ∈ X, by 0 −→ ΩPn,ı(x) −→ O(−1)n+1 ı(x) −→ Oı(x) −→ 0 since Oı(x) is a free module over itself. After tensoring it with OX,x over Oı(x), we in turn have that 0 −→ ı∗ΩPn,x −→ OX (−1)n+1 x −→ OX,x −→ 0 is split exact. It follows that (4.5) is actually exact. 0 −→ ı∗ΩPn −→ OX (−1)n+1 ∂vvv−−→ OX −→ 0 HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 19 Thus there exists a morphism LX/k → F ∨• ∂uuu / / 0 / OX (−d) 1 [−1] given by / OX (−1)n+1 ∂vvv ··· ··· / 0 ∼= I/I 2 d ı∗ΩPn / OX 0 / 0 0 / ··· ··· which is a quasi-isomorphism since (4.5) is exact. (cid:3) Corollary 4.5. In the derived category D(X), ∧r LX/k ∼= F •∨ Proof. The derived exterior product ∧rK • has been described by T. Saito in [21, §4] when K • is a complex of locally free OX -modules of finite rank and K j = 0 for all j 6= −1, 0 on any scheme X. In particular, if K −1 is invertible, ∧rK • is given by r [−r] for any r ∈ N. (K −1)⊗r −→ K 0 ⊗ (K −1)⊗r−1 −→ ··· −→ ∧r−sK 0 ⊗ (K −1)⊗s −→ ··· −→ ∧rK 0 with the differentials d∧r defined by d−s ∧r = (∧r−s idK 0) ∧ d−1 K ⊗ (id⊗s−1 K−1 ). In our situation, I/I 2 ∼= OX (−d) is invertible, and ı∗ΩPn is locally free of rank n. So the above form can be applied to K • = LX/k. The (−s)-th term of ∧r LX/k is Pn (−sd). ∧r−sı∗ΩPn ⊗ (I/I 2)⊗s ∼= ı∗Ωr−s Recall the exact sequence (4.4). It can be generalized to the long exact sequence 0 −→ Ωl Pn −→ O(−l)(n+1 l ) ∂vvv−−→ O(−l + 1)(n+1 l−1) ∂vvv−−→ ··· ∂vvv−−→ O(−1)n+1 ∂vvv−−→ O −→ 0 for any l ∈ N.2 Just like the proof of Proposition 4.4, we use the localization and then deduce l−1) ∂vvv−−→ ··· ∂vvv−−→ OX (−1)n+1 ∂vvv−−→ OX −→ 0 l ) ∂vvv−−→ OX (−l + 1)(n+1 Pn −→ OX (−l)(n+1 0 −→ ı∗Ωl is also exact. These sequences constitute the diagram as follows, . . . OX (−d) ∂vvv ... ∂vvv OX ∂vvv ∂uuu / OX (−1)n+1 ... ∂vvv ∂vvv OX (d − rd) ∂uuu / ··· ∂uuu / / OX (2 − r − d)(n+1 r−2) ∂uuu / OX (−r + 1)(n+1 r−1) ∂vvv ∂vvv ∂vvv OX (−rd) ∂uuu / / OX (d − rd − 1)n+1 ∂uuu / ··· ∂uuu / / OX (1 − r − d)(n+1 r−1) ∂uuu OX (−r)(n+1 r ) OX (−rd) d−r ∧r ı∗ΩPn(d − rd) d−r+1 ∧r ··· d−2 ∧r / ı∗Ωr−1 Pn (−d) d−1 ∧r ı∗Ωr Pn where each column is exact, and the maps ∂uuu lift the differentials d∧r since d∧r are induced by the sequence uuu (we adapt the Koszul sign rule here). Note that the associated total complex of the double complex by deleting the bottom row is exactly F •∨ r [−r]. Hence the diagram gives rise to a quasi-isomorphisms ∧r LX/k → F •∨ (cid:3) r [−r]. 2The proof is parallel to the one of [11, Thm. 8.13]. / / / / / / / / O O / / O O / / O O / / O O / / O O / O O O O O O / / O O O O / O O / / O O O O / / O O / / O O / / / O O O O 20 LIYU LIU AND WENDY LOWEN Before closing this section, let us compare Buchweitz and Flenner's formula (4.1) (Y = Spec k) and ours (i.e. HH i(X) ∼= H i(H•)) via the isomorphisms ∧r LX/k → F •∨ r [−r]. Choose an injective resolution 0 → OX → I•. Recall that Hom(−,I•), −⊗I• : C(X) → C(X) q ,I•) ∼= q is are the total Hom functor and the total tensor functor respectively. Then Hom(F •∨ q ⊗ I• ∼= F • F •∨∨ locally free of finite rank. There is a quasi-isomorphism F • q ⊗ I• is a complex of injective sheaves since F • q is bounded and for each i, F i q ,I•). Thus Extp q → Hom(F •∨ X (∧q LX/k,OX ) = HomD(X)(∧q LX/k,OX [p]) q [−q],I•[p])) ∼= H 0(Hom(F •∨ ∼= H p+q(Γ(X,Hom(F •∨ = Hp+q(X,Hom(F •∨ ∼= Hp+q(X,F • q ) q q ,I•)) ,I•))) where the hypercohomology Hp+q(X,F • e.g. [3, Ch. 1]), namely, Hp+q(X,F • q ) ∼= H p+q(H• q ). So q ) can also be computed by the (total) Cech complex (see Mp+q=i Extp X (∧q LX/k,OX ) ∼= Mp+q=i H p+q(H• q ) = H i(H•). Both results agree. 4.3. Proof of the main theorem. Let us associate some graded modules to X = Proj S. Note that the ∂vvv constitute a morphism ··· ··· / K−3(uuu; S) ∂uuu / / K−2(uuu; S) ∂uuu / / K−1(uuu; S) ∂uuu / / K0(uuu; S) ∂vvv ∂vvv ∂vvv / K−2(uuu; S) ∂uuu / K−1(uuu; S) ∂uuu K0(uuu; S) 0 from which we obtain the cokernel complex C•(uuu; S): (4.6) ··· −→ K−3(uuu; S)/ im ∂vvv ∂uuu−−→ K−2(uuu; S)/ im ∂vvv ∂uuu−−→ K−1(uuu; S)/ im ∂vvv ∂uuu−−→ K0(uuu; S). The i-th cohomology group of C•(uuu; S) is denoted by P i and the i-th cocycle group by Qi. Clearly, the S-modules P i, Qi are graded modules. Denote by Z i the i-th cocycle group of K•(vvv; R), which is a graded R-module. Recall that we have quasi-isomorphisms H• → G• → E • → ¯C′• GS(A). From now on, let us compute H • GS(A) := H • ¯C′• GS(A) by using H•,•. We need some lemmas. Lemma 4.6. The cohomology groups of K•(vvv⋆; S) are H 0 = H 0 uuu⋆ = (∂F/∂x0,−∂F/∂x1, . . . , (−1)n∂F/∂xn), and H i = 0 for all i 6= 0, −1. Proof. Recall Remark 2.1 and note that vvv⋆ So K•(vvv⋆; S), which is the mapping cone of K•(vvv⋆ ··· −→ 0 −→ S/(vvv⋆ 0) 0; S) x0−−→ S/(vvv⋆ x0−→ K•(vvv⋆ 0) −→ 0. 0 = (−x1, x2, . . . , (−1)nxn) is a regular sequence in S. 0; S), is quasi-isomorphic to 0 = k, H −1 = H −1 d−1 = kuuu⋆ where 0) ∼= k[x0]/(xd 0), we have H 0 = k and H −1 ∼= k(1 − d) as graded modules. To show Since S/(vvv⋆ uuu⋆ is a base element in H −1, we will check that uuu⋆ never belongs to im ∂vvv⋆ . This is clear since ∂F/∂x0 contains dxd−1 (cid:3) as a summand. 0 The following is well known: Lemma 4.7. The cohomology groups of K•(vvv; R) are H 0 = H 0 0 = k, and H i = 0 for all i 6= 0. / / / O O / / O O / / O O O O HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 21 Lemma 4.8. Let τ r 0 be the zeroth graded component of τ rK•,•(uuu, vvv; S). (1) If 0 ≤ r ≤ n, then H i(Tot τ r 0, Q−r r , P i−2r r+(i−r)(d−1), 0 ≤ i < r, i = r, r < i ≤ 2r. 0 ) ∼= 0 ) ∼= 0 ) ∼=(0, (2) If r ≥ n + 1 and d = n + 1, then 0, k, P i−2r H i(Tot τ r r+(i−r)(d−1), (3) If r ≥ n + 1 and d 6= n + 1, then 0 ≤ i ≤ r, i 6= n, i = n, r < i ≤ 2r. H i(Tot τ r P i−2r r+(i−r)(d−1), 0 ≤ i ≤ r, r < i ≤ 2r. Proof. We prove the statements by computing the spectral sequence I Ep,q a determined by τ r 0 . (1) Let 0 ≤ r ≤ n. The p-th column of τ rK•,•(uuu, vvv; S) is the truncation τ ≤−(n+1−r+p)K•(vvv⋆; S) up to twist. Notice that −(n + 1 − r + p) ≤ −1. By Lemma 4.6, H i(τ ≤−(n+1−r+p)K•(vvv⋆; S)) = 0 if i 6= −(n + 1 − r + p). It follows that the p-th column of τ rK•,•(uuu, vvv; S) is exact except in spot (p, r). By considering the zeroth graded component, we have I Ep,q 1 = 0 if q 6= r, and I Ep,r 1 =(cid:0)S(r + p(d − 1))(n+1 To compute I Ep,r 2 , it suffices to consider the complex r−p)/ im ∂vvv(cid:1)0 =(cid:0)K−(r−p)(uuu; S)/ im ∂vvv(cid:1)r+p(d−1). ∂uuu−−→(cid:0)K0(uuu; S)(cid:1)rd. 2 = Q−r r , I Ep,r 2 = P −(r−p) r+(i−r)(d−1) when r < i ≤ 2r, and H r(Tot τ r r+p(d−1) if p ≥ 1. Hence ∞ = 0 ) ∼= I E0,r (cid:0)K−r(uuu; S)/ im ∂vvv(cid:1)r −→ ··· −→(cid:0)K−1(uuu; S)/ im ∂vvv(cid:1)r+(r−1)(d−1) Comparing this complex with (4.6), we have I E0,r H i(Tot τ r I E0,r 0 ) ∼= I Ei−r,r 2 = Q−r r . = I Ei−r,r = P i−2r ∞ 2 (2) Let r ≥ n + 1 and d = n + 1. Just like in the situation in (1), we have I Ep,r 1 =(cid:0)K−(r−p)(uuu; S)/ im ∂vvv(cid:1)r+p(d−1). 1 ∼= k. For I Ep,q By Lemma 4.6 and taking into consideration the degrees, we find one more nonzero I Ep,q I E0,n Note that Q−(n+1) I E0,r 1 , namely, 2 , as shown in (1), I E0,r r+p(d−1) for all 1 ≤ p ≤ 2r. is a k-submodule of (S/ im ∂vvv)n+1 = kn+1 = 0 and Q−s = 0 if s ≥ n + 2. Hence being all 2 = P −(r−p) r = 0 since r ≥ n + 1. We also have I E0,n 1 ∼= k and the rest I Ep,q 2 = I E0,n 2 = Q−r r 2 = Q−r and I Ep,r n+1 2 zero. Assertion (2) follows. (3) The proof is completely similar to (2). The only difference is that I E0,n 1 is zero since d 6= n + 1. (cid:3) The double complex H•,• r to calculate it. leads to a spectral sequence II Ep,q r,a by filtration by rows. We begin 22 LIYU LIU AND WENDY LOWEN 4.3.1. Case 1: d > n + 1. Suppose m ≥ 0 and O = OPn. By the exact sequence 0 −→ O(m − d) F−−→ O(m) −→ OX (m) −→ 0, we immediately conclude that H i(X,OX (m)) = 0 if i 6= 0, n − 1, and H n−1(X,OX (m)) ∼= H n(Pn,O(m− d)) ∼= H 0(Pn,O(d− n− 1− m))∗. Obviously, H 0(Pn,O(d− n− 1− m)) has a basis 0 xi1 {xi0 1 ··· xin n ∈ R i0 + i1 + ··· + in = d − n − 1 − m, i0, i1, . . . , in ≥ 0}. On the other hand, the Cech cohomology group H n−1(U,OX (m)) has a basis 0 xj1 {xj0 1 ··· xjn n ∈ Sx1···xn j0 + j1 + ··· + jn = m, 0 ≤ j0 ≤ d − 1, j1, . . . , jn ≤ −1} where Sx1···xn is the localization of S at x1 ··· xn. Since both groups have finite dimension over k, the duality gives rise to the bijection (4.7) S : H 0(Pn,O(d − n − 1 − m)) −→ H n−1(U,OX (m)), xi0 0 xi1 1 ··· xin n 7−→ xd−1−i0 0 x−1−i1 1 ··· x−1−in n . The map S induces H 0(Pn,O(d − n − 1 − m)r) −→ H n−1(U,OX (m)r) for any r ∈ N which is also denoted by S . Since H n−1(U,OX (m)) = 0 if m ≥ d, by the definition of H•,• r , we have II Ep,q H n−1(cid:0)U,OX(p)(n+1 r ) = r,1 = H q(U,F p d−n−1−p(cid:1)∗ p )(cid:1) =(cid:0)R r,1 −→ ··· −→ II Er−1,n−1 r,1 −→ II E1,n−1 = K−p(vvv; R)∗ (Tot τ r 0, II E0,n−1 0 )p, (n+1 p ) Since H n−1(cid:0)U,OX(p)(n+1 p )(cid:1), 0 ≤ p ≤ r, q = n − 1, 0 ≤ p ≤ 2r, q = 0, otherwise. d−n−1−p, the complex −→ II Er,n−1 r,1 r,1 is dual to (4.8) K0(vvv; R)d−n−1 ←− K−1(vvv; R)d−n−2 ←− ··· ←− K−r+1(vvv; R)d−n−r ←− K−r(vvv; R)d−n−1−r. By Lemma 4.7, the only non trivial cohomology of the complex K•(vvv; R) is H 0(K•(vvv; R)) = k. The zero-th cohomology group of (4.8) is zero since the (d − n − 1)-st graded component in k is zero. The unique possible nonzero cohomology of (4.8) is H −r = Z −r r,2 = d−n−1−r). Combining this with Lemma 4.8, we obtain that II Ep,q S (Z −r p = r, q = n − 1, r < p ≤ 2r, q = 0, p = r, q = 0, otherwise. r,n = 0 when r ≤ n − 1, since r + n > 2r; r,n = 0 since R has only r,n for any pair (r, n), it is sufficient to Immediately, II Ep,q on the other hand, in case r ≥ n, we have d− n− 1− r ≤ −1 and so II Er,n−1 non-negative grading. So in order to show II Ep,q prove the differential II En,n−1 S (Z −r d−n−1−r), P p−2r r+(p−r)(d−1), Q−r r , 0, d−n−1−r, yielding II Er,n−1 r,2 is given by r,2 = r,2 . On one hand, II Er+n,0 r,n = ··· = II Ep,q n,n (i.e. the case r = n) is zero. r,n+1 = II Ep,q II Ep,q n,n → II E2n,0 n,n n,n Since II En,n−1 is a sub-quotient of C′n−1(U,F n n ), we choose a cocycle cn−1,n ∈ C′n−1(U,F n n ) for any class in II En,n−1 . Performing a diagram chase, a cochain (c0,2n−1, c1,2n−2, . . . , cn−1,n) in H• can be given. Notice that c0,2n−1 ∈ H0,2n−1 = C′0(U,F 2n−1 ) = K−1(uuu; S)nd−d+1, and so dv,H(c0,2n−1) = ∂uuu(c0,2n−1) is a coboudary in K0(uuu; S)nd, i.e. dv,H(c0,2n−1) represents the zero class in P 0 n,n is a zero map. Therefore, II Ep,q n,n . It follows that the differential II En,n−1 n,n → II E2n,0 nd = II E2n,0 r,2 , and r,∞ = II Ep,q n H i(H•) ∼=Mr∈N Mp+q=i II Ep,q r,∞ =Mr<i P i−2r r+(i−r)(d−1) ⊕ Q−i i ⊕ S (Z −i+n−1 d−i−2 ). HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 23 4.3.2. Case 2: d = n + 1. The formula remains valid in this case. Note that the complex (4.8) has only one nonzero term K0(vvv; R)d−n−1 = R0 = k. By applying Lemma 4.8 again, we conclude that for 0 ≤ r ≤ n, p )(cid:1), 0 ≤ p ≤ r, q = n − 1, 0 ≤ p ≤ 2r, q = 0, otherwise. II Ep,q 0 )p, II Ep,q (Tot τ r 0, r,1 = H n−1(cid:0)U,OX (p)(n+1 r,2 = r,2 = k, k, P p−2r r+n(p−r), 0. k, P p−2r r+n(p−r), Q−r r , 0, II Ep,q p = 0, q = n − 1, r < p ≤ 2r, q = 0, p = r, q = 0, otherwise, p = 0, q = n − 1, p = n, q = 0, r < p ≤ 2r, q = 0, otherwise. and for r ≥ n + 1, It follows that II Ep,q r,n = ··· = II Ep,q r,2 . Yi1 r ) for any r is the sub-complex of Sxi1 xi2 −→ ··· −→ Yi1<···<in−1 Since for any Vi1...is ∈ V, the algebra Ai1...is = OX (Vi1...is) is identified with the zero-th graded component of Sxi1 ···xis , the localization of S with respect to the element xi1 ··· xis , we conclude that the Cech complex C′•(U,F 0 Sxi1 −→ Yi1<i2 ··· x−1 consisting of all cochains of degree zero. Since II E0,n−1 r ), it seems apt to choose xn . However, for the sake 0 x−1 of easy computation, we use x−1 n . Similar to 1 the argument in the case d > n + 1, one finds a cochain (c0,n−1, c1,n−2, . . . , cn−1,0) in H• with cn−1,0 = x−1 r,n sends the class represented by cn−1,0 to the one represented by dv,H(c0,n−1). r ) as a base element of II E0,n−1 n · ∂F/∂x0 flexibly rather than xn n · ∂F/∂x0. The differential II E0,n is a sub-quotient of C′n−1(U,F 0 Sxi1 ···xin−1 −→ Sx1···xn n ∈ C′n−1(U,F 0 r,n−1 → II En,0 ··· x−1 ··· x−1 ··· x−1 0 x−1 1 r,n r,n 1 1 r,n−1 → II En,0 r,n . (1) If 0 ≤ r ≤ n − 1, dv,H(c0,n−1) belongs to P n−2r r+n(n−r). Recall the shape and size of the triangle τ rK•,•(uuu, vvv; S). The element dv,H(c0,n−1) is zero itself if r is very small, or is a sum ∂uuu(?) + ∂vvv(?′) if r is larger. According to the construction of (4.8), ∂uuu(?) + ∂vvv(?′) In both cases, cn−1,0 is killed by the differential necessarily represents the zero class. II E0,n (2) If r = n, the diagram chase shows dv,H(c0,n−1) = uuu⋆ + im ∂vvv ∈ Q−n / im ∂vvv → im ∂vvv}. By the definition of C•(uuu; S), uuu⋆ + im ∂vvv happens to be a base element is injective and its n = ker{Sn+1 / im ∂vvv}. So II E0,n r,n−1 = k → II En,0 Sn(n+1)/2 2n of im{S0/ im ∂vvv → Sn+1 cokernel is given by Q−n n n /(kuuu⋆ + im ∂vvv) = P −n n . (3) If r ≥ n + 1, we claim that the differential II E0,n r,n−1 = k → II En,0 r,n = k is an isomorphism. r,n = Q−n n n The assertion follows from Lemma 4.9 which will be proven later on. Summarizing, the spectral sequence II Ep,q r,∞ = II Ep,q r,n+1 = k, P p−2r r+n(p−r), Q−r r , 0, II Ep,q r,∞ = II Ep,q r,n+1 =(P p−2r 0, r+n(p−r), p = 0, q = n − 1, r < p ≤ 2r, q = 0, p = r, q = 0, otherwise, r ≤ p ≤ 2r, q = 0, otherwise, if 0 ≤ r ≤ n − 1, if r = n, 24 LIYU LIU AND WENDY LOWEN II Ep,q r,∞ = II Ep,q r+n(p−r), r < p ≤ 2r, q = 0, otherwise, if r ≥ n + 1. Therefore, , i P i−2r r+n(i−r) ⊕ Q−i P i−2r r+n(i−r) ⊕ Q−i P i−2r r+n(i−r), i ⊕ kn, i 6= n − 1, n, i = n − 1, i = n. 0, r,n+1 =(P p−2r Mr<i Mr<i Mr≤i H i(H•) ∼=  Note that F q decomposition r is a direct sum of some terms as given in Figure 2, and hence Hp,q r admits a C′p(U,OX (q)(n+1 q )) ⊕ C′p(U,OX (q + d − 2)(n+1 q−2)) ⊕ C′p(U,OX (q + 2d − 4)(n+1 q−4)) ⊕ ··· q ) appearing in the first component corresponds to a graded left preferred if when q ≤ r. Intuitively, OX (q)(n+1 module located at the leftmost edge in Figure 2. We hence call a cochain in Hp,q it has possible nonzero component only in C′p(U,OX (q)(n+1 Lemma 4.9. Suppose d = n + 1 and r ≥ n. There exists a cochain (c0,n−1, c1,n−2, . . . , cn−1,0) in Hn−1 such that each cn−1−q,q is left preferred in Hn−1−q,q cn−1,0 = x−1 dF (c0,n−1, c1,n−2, . . . , cn−1,0) = ((−1)n−1uuu⋆, 0, . . . , 0). ··· x−1 q )). ∂F ∂x0 and n 1 , r r r Proof. During the proof, we will frequently meet elements in Sxi1 ···xim . To avoid confusion, we underline denominators to distinguish between similar looking elements. For example, x−1 1 ∈ Sx1 , x−1 1 x0 3 ∈ Sx1x2x3. The notations fj1...js stand for formal bases elements. When the Cech indices (i1, . . . , is) appear, the complements are denote by (j1, . . . , jn−s), 2 ∈ Sx1x2, x−1 1 x0 2x0 namely, the latter are obtained by deleting i1, . . . , is from (1, 2, . . . , n). The permutation (cid:18) 1 i1 . . . . . . s is s + 1 . . . . . . j1 n jn−s(cid:19) is a shuffle, whose parity (n2 − s2 + n − s)/2 − (j1 + ··· + jn−s) is denoted by ℘(i1, . . . , is) or even by ℘(ı) if no confusion arises. j ··· x−1 n ∂F ∂x0 fj nXj=1 ∂F ∂xj x−1 1 ··· x0 f0(cid:19). Starting with cn−1,0 = x−1 1 ··· x−1 n ∂F/∂x0, we have 1 dF (cn−1,0) = (−1)n−1x−1 nXj=1 = (−1)n−1 Choose cn−2,1 = (cn−2,1 i1,...,in−1 ) as ··· x−1 n x0 ∂F ∂x0 x−1 1 ··· x0 j ··· x−1 cn−2,1 i1,...,in−1 = (−1)℘(ı)+1x−1 i1 ··· x−1 fj − ∂x0 f0 + (−1)n−1 n (cid:18) ∂F in−1(cid:18) ∂F ∂x0 One can easily show that ∂uuu(cn−2,1) = 0. Thus dF (cn−2,1) = (−1)n−2∂vvv(cn−2,1) whose compo- nents are dF (cn−2,1 i1,...,in−1) = (−1)j1+1x−1 i1 ··· x−1 n−1Xl=1 n−1Xl=1 = (−1)j1+1 + (−1)j1+1 in−1 x0 ∂F ∂x0 f0j1 + (−1)j1+1 il ··· x−1 in−1 ∂F ∂x0 filj1 x−1 i1 ··· x0 ∂F ∂xj1 in−1 il ··· x−1 in−1(cid:18) ∂F ∂x0 x−1 i1 ··· x0 il ··· x−1 filj1 + ∂F ∂xj1 f0il − i1 ··· x−1 f0j1(cid:19). ∂F ∂xil in−1 xj1 ∂F ∂xj1 f0j1 fj1 − ∂F ∂xj1 f0(cid:19). x−1 i1 ··· x0 n−1Xl=1 f0il + (−1)j1+1x−1 HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 25 Choose cn−3,2 = (cn−3,2 i1,...,in−2 ) as f0j1(cid:19) which is again in ker ∂uuu. Thus dF (cn−3,2) = (−1)n−3∂vvv(cn−3,2) whose components are ∂F ∂x0 in−2(cid:18) ∂F i1 ··· x−1 i1 ··· x−1 x−1 i1 ··· x0 il ··· x−1 cn−3,2 i1,...,in−2 = (−1)℘(ı)x−1 i1,...,in−2) = (−1)n−j1−j2(cid:18)x−1 x−1 i1 ··· x0 n−2Xl=1 i1 ··· x−1 il ··· x−1 f0ilj2 + x−1 dF (cn−3,2 fj1j2 − ∂F ∂xj1 ∂F ∂xj2 ∂F ∂x0 in−2 xj1 f0j1j2 + in−2 x0 f0j2 + ∂x0 in−2 in−2 + filj1j2 ∂F ∂xj1 ∂F ∂xj2 f0j1il + x−1 i1 ··· x−1 in−2 xj2 il ··· x−1 in−2(cid:18) ∂F ∂x0 filj1j2 + ∂F ∂xj1 f0ilj2 ∂F ∂xj1 ∂F ∂xj2 f0j1j2 f0j1j2 n−2Xl=1 n−2Xl=1 + in−2 x−1 il ··· x−1 i1 ··· x0 n−2Xl=1 = (−1)n−j1−j2 + ∂F ∂xj2 f0j1il − x−1 i1 ··· x0 f0j1j2(cid:19). ∂F ∂xil in−3(cid:18) ∂F ∂x0 Choose cn−4,3 = (cn−4,3 i1,...,in−3 ) as = (−1)℘(ı)+1x−1 i1 ··· x−1 cn−4,3 i1,...,in−3 fj1j2j3 − ∂F ∂xj1 f0j2j3 + ∂F ∂xj2 f0j1j3 − ∂F ∂xj3 f0j1j2(cid:19). Set j0 = 0 by convention and continue the above procedure. We obtain (4.9) cs−1,n−s i1,...,is = (−1)℘(ı)+n−sx−1 i1 ··· x−1 is fj0... cjm...jn−s successively, which is obviously left preferred. In particular, when s = 1, ∂xjm n−sXm=0 (−1)m ∂F n−1Xm=0 (−1)m ∂F ∂xjm fj0... cjm...jn−1 c0,n−1 i1 = (−1)n−i1x−1 i1 and hence dF (c0,n−1 i1 ) = (−1)n−i1(cid:18)x0 = (−1)n−i1(cid:18)x0 = (−1)n−1x0 + (−1)i1 x0 i1 i1 ∂xjm ∂xjm n−1Xm=0 (−1)m ∂F n−1Xm=0 (−1)m ∂F i1(cid:18) Xjm<i1 (−1)m ∂F fj0... bi1...jn−1(cid:19) nXm=0 (−1)m ∂F ∂F ∂xi1 ∂xjm i1 i1 fi1j0... cjm...jn−1 + x−1 i1 fi1j0... cjm...jn−1 − x0 i1 fj0...jn−1(cid:19) xjm ∂F ∂xjm n−1Xm=0 fj0...jn−1(cid:19) (−1)m+1 ∂F ∂F ∂xi1 ∂xjm + Xjm>i1 fj0... cjm...i1...jn−1 fj0...i1... cjm...jn−1 = (−1)n−1x0 ) is actually the restriction of the global section (−1)n−1uuu⋆ to affine Vi1 . Hence the fj0... cjm...jn ∂xjm . (cid:3) So dF (c0,n−1 result follows. i1 With minor modification, the proof of Lemma 4.9 is valid if the hypothesis r ≥ n is changed to r < n. Thus we obtain one more lemma as follows. Lemma 4.10. Suppose d = n + 1 and 0 ≤ r ≤ n − 1. There exists a cocycle (0, . . . , 0, cn−1−r,r, cn−r,r−1, . . . , cn−1,0) r in Hn−1 Hn−1−q,q r where the components cn−1−q,q are given in (4.9). Each cn−1−q,q is left preferred in . 26 LIYU LIU AND WENDY LOWEN Note that there are n copies of k in the expression of H n−1(H•). They respectively come from C′n−1(U,F 0 r ) for 0 ≤ r ≤ n − 1. The class represented by the cocycle given in Lemma 4.10 is nontrivial since cn−1,0 represents a nontrivial class. Consider the quasi-isomorphisms ¯¯λ given in (4.2) and γ given in Theorem 4.3. The quasi-isomorphic image by γ¯¯λ : H• r → ¯C′• GS(A)r is a collection of local sections of the sheaf ∧rTX . More precisely, we summarize the fact as Proposition 4.11. Suppose d = n + 1. For every 0 ≤ r ≤ n − 1, there is a one-dimensional k-submodule of H n−1−r(X,∧rTX ), and consequently H n−1−r(X,∧rTX ) 6= 0. 4.3.3. Case 3: d < n + 1. This is an easy case, since the complex (4.8) is zero. The results are r,2 = H i(H•) ∼=Mr<i r+(i−r)(d−1) ⊕ Q−i P i−2r i . II Ep,q r,∞ = II Ep,q P p−2r r+(p−r)(d−1), Q−r r , 0, r < p ≤ 2r, q = 0, p = r, q = 0, otherwise, and 4.4. Characterization of smoothness. In this section, we give a necessary and sufficient con- dition under which a hypersurface is smooth. of H 2 In the proof (not in the statement) of Theorem 4.14, we make use of the following subgroups GS(A)1: • the subgroup Eres of 2-classes of the form [(0, f, 0)]; • the subgroup Emult of 2-classes of the form [(m, 0, 0)]. First of all, based upon the expression of H 2(H• P 0 d as a summand for any n and d. Every element t ∈ P 0 consider when t also belongs to Eres. 1) from §4.3, we obtain that H 2 GS(A)1 contains d corresponds to a class in Emult. Let us Since t ∈ P 0 d = (S/(im ∂uuu))d, t lifts to an element ¯t in Sd. We then identify ¯t to a global section of OX (d). For any V ∈ V, ¯tV ∈ A(V ) determines the left multiplication by ¯tV on A(V ), and so ¯tV ◦ ◦µ represents a class in H 2 (1)(A(V ),A(V )) which is independent of the choice of ¯t. GS(A)1 is represented by the GS 2-cocycle (¯t◦ ◦µ, 0, 0) := ((¯tV ◦ ◦µ)V , 0, 0) which only Hence t ∈ H 2 deforms the local multiplications of A. If ¯tV ◦ ◦µ happens to be a coboundary for all V , we have cochains sV ∈ C1(A(V ),A(V )) such that dHoch(sV ) = ¯tV ◦ ◦µ. Let s = (sV )V ∈ ¯C′0,1(A) and so (¯t ◦ ◦µ, 0, 0) − (0,−dsimp(s), 0) = dGS(s, 0). Thus t = [(¯t ◦ ◦µ, 0, 0)] = [(0,−dsimp(s), 0)] belongs to Emult ∩ Eres. In the other direction, if t ∈ Emult is also in Eres, then we assume its representation is (0, f, 0). The difference (¯t ◦ ◦µ, 0, 0) − (0, f, 0) has to be a GS coboundary, say dGS(s, 0). It follows that ¯tV ◦ ◦µ = dHoch(sV ) for all V ∈ V. Summarizing, we have t ∈ Emult ∩ Eres if and only if ¯tV ◦ ◦µ is a Hochschild 2-coboundary for every V ∈ V. Note that A(V ) is a localization of A(U ) if V ⊆ U . It follows that ¯tV ◦ ◦µ is a coboundary of A(V ) provided that ¯tU ◦ ◦µ is a coboundary of A(U ). So this condition is again equivalent to the fact that ¯tUi ◦ ◦µ is a coboundary of Ai for all 1 ≤ i ≤ n. By §3, H 2 (1)(Ai, Ai) = Ai(cid:30)(cid:18) ∂Gi ∂y0 , . . . , ∂Gi ∂yi−1 , ∂Gi ∂yi+1 , . . . , ∂Gi ∂y3(cid:19) and ¯tUi ◦ ◦µ is a coboundary if and only if ¯tUi is sent to zero by the projection Ai → H 2 Since Ai = k[y0 . . . , yi−1, yi+1, . . . , yn]/(Gi) and (1)(Ai, Ai). yj ∂Gi ∂yj Xj6=i + Hi = d · Gi, HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 27 we have H 2 (1)(Ai, Ai) = k[y0, . . . , yi−1, yi+1, . . . , yn](cid:30)(cid:18) ∂Gi ∂Gi ∂yi+1 Recall the definition of Hi given in §4.1. There is an algebra map P 0 → H 2 (1)(Ai, Ai) defined by xj 7→ yj if j 6= i and xi 7→ 1, whose kernel is (xi − 1)P 0. Thus t ∈ Eres if and only if t ∈ ∩n If t = (1 − xi)Ti for some Ti ∈ P 0, by comparing the homogeneous components, we conclude that t is annihilated by a power of xi and i which is actually a finite sum. In the opposite direction, if t is annihilated by a i=1(xi − 1)P 0. Notice that t is homogeneous. ∂yn(cid:19). ∂Gi ∂yi−1 , . . . , , . . . , , Hi, ∂Gi ∂y0 m=0 txm Ti =P∞ power of xi, then t = (1 − xi)P∞ Lemma 4.12. Let t ∈ P 0 m=0 txm i ∈ (xi − 1)P 0. Consequently, we have proven d . Then t ∈ Eres if and only if xi ∈pannP 0(t) for all 1 ≤ i ≤ n. Next let us recall the work [9] by Gerstenhaber and Schack. Starting from their Hodge decom- position for presheaves of commutative algebras (4.10) they prove the existence of the HKR type decomposition H i GS(A) =Mr∈N GS(A) ∼= Mp+q=i H i H i GS(A)r, H p simp(V,∧qT ) for any smooth complex projective variety X, where A = OXV (resp. T = TXV) is the restriction of the structure sheaf (resp. tangent sheaf) to an affine open covering V closed under intersection. In particular, GS(A) ∼= H 0 H 2 simp(V,∧2T ) ⊕ H 1 simp(V,T ) ⊕ H 2 simp(V,A). The roles played by the three summands in the deformation of A (viewed as a twisted presheaf) are explained in [6]. More concretely, elements in the three summands respectively deform the (local) multiplications, the restriction maps, and the twisting elements of A. If X is not necessarily smooth, Gerstenhaber and Schack's result remains partially correct: H i simp(V,∧rT ) simp(V,∧i−1T ) as a k-submodule. For if r = 0 or r = i, and in general H i i = 2, we more precisely have GS(A)i−1 contains H 1 GS(A)r ∼= H i−r (4.11) simp(V,T ) ∼= Eres ⊆ H 2 H 1 GS(A)1. In particular, (4.10) now yields GS(A) ∼= H 0 H 2 (4.12) simp(V,∧2T ) ⊕ H 1 simp(V,T ) ⊕ H 2 simp(V,A) ⊕ E. where E is a complement of Eres in H 2 GS(A)1. When X is a projective hypersurface, the isomorphism H p(X,∧qTX ) ∼= H p simp(V,∧qT ) holds for all p, q. The decomposition (4.12) is equivalent to HH 2(X) ∼= H 0(X,∧2TX ) ⊕ H 1(X,TX ) ⊕ H 2(X,OX ) ⊕ E. We have thus proven: Proposition 4.13. Let X be a projective hypersurface. The following are equivalent: (1) The HKR decomposition holds for the second cohomology, i.e. HH 2(X) ∼= H 0(X,∧2TX ) ⊕ H 1(X,TX ) ⊕ H 2(X,OX ). (2) We have H 1(X,TX ) ∼= Eres = H 2 GS(A)1. 28 LIYU LIU AND WENDY LOWEN Remark 4.1. In deformation theoretic terms, Proposition 4.13 states that for a projective hyper- surface X, the HKR decomposition holds for HH 2(X) if and only if every (commutative) scheme deformation of X can be realized by only deforming restriction maps while trivially deforming in- dividual algebras on an affine cover. This is the classical deformation picture for smooth schemes. We have the following converse of the HKR theorem for projective hypersurfaces: Theorem 4.14. Let X be a projective hypersurface. The following are equivalent: (1) X is smooth. (2) The HKR decomposition holds for all cohomology groups, i.e. HH i(X) ∼= Mp+q=i H p(X,∧qTX ), ∀ i ∈ N. (3) The HKR decomposition holds for the second cohomology, i.e. HH 2(X) ∼= H 0(X,∧2TX ) ⊕ H 1(X,TX ) ⊕ H 2(X,OX ). Proof. It remains to prove (3)⇒ (1). Assume X is a hypersurface of degree d in Pn which is not smooth. According to Proposition 4.13, it suffices to produce a class in H 2 GS(A)1 \ Eres. At least one of the algebras Ai is not smooth, say An. By Remark 3.2, it follows that H 2 (1)(An, An) 6= 0. As before, we know H 2 (1)(An, An) = k[y0, . . . , yn−1](cid:30)(cid:18) ∂Gn ∂y0 ∼= R(cid:30)(cid:18)xn − 1, = P 0/(xn − 1). Since P 0/(xn − 1) 6= 0 this implies that 0 6= xm GS(A)1, and xn /∈qannP 0(xd presents a non-trivial class in H 2 finishes the proof. ,··· , ∂F ∂x0 , Hn(cid:19) ,··· , ∂F ∂xn−1 ∂Gn ∂yn−1 ∂F ∂xn(cid:19) , n ∈ P 0 for any m ∈ N. In particular, xd n). By Lemma 4.12, xd n ∈ P 0 n /∈ Eres, which d (cid:3) 4.5. Examples of intertwined classes. We are particularly interested in HH 2(X) since it parameterizes the equivalence classes of first order deformations of X. We retain the notations used before. On one hand, we have the decomposition (4.12). On the other hand, any GS 2-cocycle (m, f, c) ∈ ¯C′0,2(A) ⊕ ¯C′1,1(A) ⊕ ¯C′2,0(A) factors as (m−mab, 0, 0)+(mab, f, 0)+(0, 0, c) under the Hodge decomposition where mab depends only on m. Since E ⊆ H 2 GS(A)1, the elements in E admit representatives of the form (m, f, 0). Normally, neither (m, 0, 0) nor (0, f, 0) is a cocycle. The cocycle is called untwined if (m, 0, 0) or, equivalently (0, f, 0) is a cocycle. A 2-class is called intertwined if it has no untwined representative. In this section, we will given examples of such intertwined 2-classes. By the decomposition of 2) have untwined representatives of the form 1). Moreover, we exclude the H• and by Theorem 4.3, classes in H 2(H• (0, 0, c) and (m, 0, 0) respectively. It is sufficient to consider H 2(H• case n = 1 since this is affine case. 0) and H 2(H• First of all, by the discussion in §4.3, H 2(H• Via the quasi-isomorphisms H• → G• → E • → ¯C′• GS 2-class of the form [(m, 0, 0)] ∈ H 2 2 , H 2(H• GS(A). So intertwined 2-class never exists if d < n + 1. 1) contains k as a direct summand if d = n+1. By Proposition 4.11, any nonzero element in k corresponds to a nonzero class in H 1(X,TX ) which clearly admits a representative of the form (0, f, 0). 1) is the direct sum of P 0 GS(A), any element in P 0 if d < n + 1. 2 2 gives rise to a d and Q−2 d or Q−2 Next, besides P 0 d and Q−2 HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 29 Thus an intertwined class exists only possibly in S (Z n−3 n ≤ 3 since Z n−3 = 0 for all n > 3. Since n = 3 implies S (Z 0 choice, and so d > 3. Moreover, by the definition of Z −1 d−4 ) in the case d > n + 1. Necessarily, 0), n = 2 is the unique d−4) ⊆ H 2(H• d−4, the short sequence ∂vvv−−→ Rd−3 −→ 0 (4.13) is exact. It follows that Z −1 0 −→ Z −1 d−4 −→ R3 d−4 6= 0 only if d > 4. d−4 We have proven: Proposition 4.15. Suppose either n 6= 2 or n = 2 and d ≤ 4. Then H 2 intertwined cohomology class. GS(A) does not contain an Now let d ≥ 6 and F = xd 1 → R2 in (4.13) sends (r0, r1, r2) to r0x0 + r1x1 + r2x2, whose kernel is 3-dimensional with a basis {(−x1, x0, 0), (−x2, 0, x0), (0,−x2, x1)}. Since S (Z −1 1 x2. The map ∂vvv : R3 1, we consider the double complex 0 +xd−1 1 ) arises from H• Sx1 ⊕ Sx2 ∂uuu x1 ⊕ S3 S3 x2 ∂vvv Sx1 ⊕ Sx2 / Sx1x2 ∂uuu S3 x1x2 ∂vvv Sx1x2 / 0 with three entries corresponding to H2 so 1 underlined. We choose the basis element (0,−x2, x1), and S (0,−x2, x1) = (0,−x4 , (d − 1)xd−2 1 1 x2, xd−1 0x−1 1 x−2 (d − 1)xd−2 1 x2 · (−x4 0xd−3 1 x−1 0x−1 0x−2 1 x−2 1 x−1 2 , x4 2 ) ∈ S3 ), ∂uuu(S (0,−x2, x1)) is equal to 2 ) + xd−1 1 x−1 0x−2 1 · x4 x1x2. 2 = −(d − 2)x4 1 x−1 0xd−3 0xd−3 1 x−1 2 . 2 ) ∈ Sx1 ⊕ Sx2 , and thus ((0, (d − 2)x4 2 ), S (0,−x2, x1), 0) is a Since uuu = (dxd−1 0 Choose (0, (d − 2)x4 2-cocycle in H• 1. 0xd−3 1 x−1 Let us prove that the class c := [((0, (d − 2)x4 2 ), S (0,−x2, x1), 0)] is intertwined. As- sume it can be written as [(m′, 0, 0)] + [(0, f ′, 0)], then m′ := (m′ 2) ∈ ker{Sx1 ⊕ Sx2 → Sx1x2}. Note that S, Sx1 and Sx2 can be regarded as k-submodules of Sx1x1 since S is a domain, and that Sx1 ∩ Sx2 = S. We then have m′ 2 ∈ im{∂uuu : S3 (4.14) 2 ∈ S. It follows that m′ x2 → Sx2}, say 2 + (d − 2)x4 2 and so m′ 1 x−1 1 = m′ 0xd−3 1, m′ 1 x2a2 + xd−1 a3 0 1 m′ 2 + (d − 2)x4 a1 + (d − 1)xd−2 for some a1, a2, a3 ∈ Sx2 . By considering their degrees, we have 1 x1−i0−i1 2 = dxd−1 λi0i1 1 xi0 1 x−1 0xd−3 0 xi1 2 a1 = X0≤i0<d i1≥0 − X1≤i0<d i1≥0 0 xd+i1−2 1 dλ0i1 Xi1≥0 + X0≤i0<d i1≥0 and similarly for a2, a3. The right-hand side of (4.14) is 1 xd−1 0 xi1 1 x1−i1 2 dλi0i1 1 xi0−1 0 xd+i1−1 1 x2−i0−i1 2 (d − 1)λi0i1 2 xi0 x2−i0−i1 2 + X0≤i0<d i1≥0 λi0i1 3 xi0 0 xd+i1−1 1 x1−i0−i1 2 . / / / O O O O / / O O O O / 30 LIYU LIU AND WENDY LOWEN Observe that the basis element x4 d ≥ 6 and i1 ≥ 0. Together with the fact m′ intertwined class. 1 x−1 0xd−3 2 never appears in any term of the right-hand side, since 2 ∈ S, we get a contradiction. Thus c is indeed an We remind the reader that the projective curve xd 1 x2 has a unique singularity (0 : 0 : 1). Next let us describe how the class deforms A in the case d = 6. We have U = {U1, U2} and 0 + xd−1 V = {V1, V2, V12}, and define λ : V → U by V2 7→ U2, The algebras A1, A2, A12 are expressed as k[y0, y2]/(y6 0 +y5 1) respectively. By the formula (4.2), we obtain a 2-cocycle (e0, e1, 0) in E • y5 0 +y2), k[y0, y1]/(y6 V12 7→ U2. V1 7→ U1, 1), k[y0, y1, y−1 1 given by 1 ]/(y6 0 + e0 V1 = 0, V2 = −4x5 0x3 e0 V12 = −4y5 0y3 e0 V12⊂V1 = −(0,−x4 e1 e1 V12⊂V2 = 0. 1x−1 2 V2 = −4y5 1 ∈ A12, 0x−1 1 x−2 0y3 1 ∈ A2, 2 , x4 0x−2 1 x−1 2 )V12 = (0, y4 0y−1 1 , y4 0y−2 1 ) ∈ A3 12, So by Theorem 4.3, the intertwined cocycle (m, f, 0) is given by ◦µA2 , ◦µA12, mV2 = −4y5 0y3 1 mV12 = −4y5 0y3 1 fV12⊂V1 =(cid:18)−y4 0y−2 1 ◦∂ ◦∂y0 + (y4 0y−1 1 − y4 0y−2 1 ) ◦∂ ◦∂y1(cid:19) ◦ ρV1 V12 and other components equal to zero, where ρV1 V12 : A1 → A12 is the restriction map. Unfortunately, the authors have not found any intertwined class in the case d = 5. So we pose the following open question: Question: Does an intertwined 2-class exist for a degree 5 curve in P2? 4.6. The second cohomology groups of quartic surfaces. As we exhibited in §4.5, inter- twined 2-classes exist for some non-smooth curves. In contrast, by Proposition 4.15 such classes do not exist for higher dimensional hypersurfaces, whence for these it suffices to study 2-cocycles of the form (m, 0, 0), (0, f, 0) and (0, 0, c) separately. Among projective hypersurfaces, we are particularly interested in quartic surfaces in P3. From now on, let X be a projective quartic surface in P3, i.e. n = 3 and d = 4. By the discussion in §4.3, GS(A)0 ∼= k; H 2 GS(A)1 ∼= k ⊕ P 0 H 2 4 ; GS(A)2 ∼= k ⊕ Q−2 H 2 2 . Now let us make the three deformations arising from the three components "k" above explicit, following Lemma 4.10 and formula (4.9). A direct computation shows that c2,0 123 = x−1 1 x−1 2 x−1 3 ∂F ∂x0 , c1,1 12 = x−1 1 x−1 c1,1 13 = x−1 1 x−1 ∂x3 2 (cid:18) ∂F 3 (cid:18)− f0 − ∂F ∂x2 ∂F ∂x0 f3(cid:19), f2(cid:19), ∂F ∂x0 f0 + HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 31 c1,1 23 = x−1 c0,2 1 = x−1 c0,2 2 = x−1 c0,2 3 = x−1 ∂x3 ∂x1 2 x−1 3 (cid:18) ∂F 1 (cid:18) ∂F 2 (cid:18)− 3 (cid:18) ∂F f02 − ∂F ∂x3 f01 − ∂x2 ∂F ∂x1 f3(cid:19), f03 + f0 − ∂F ∂x2 ∂F ∂x0 f01 + ∂F ∂x2 f03 − ∂F ∂x0 ∂F ∂x1 f02 + ∂F ∂x0 f23(cid:19), f23(cid:19), f12(cid:19). We choose a map λ : V → U by λ(Vj1...jr ) = Ujr if j1 < ··· < jr, and the algebra A(Vj1...jr ) is expressed as k[y0, . . . , yjr −1, yjr+1, . . . , y3, y−1 jr−1 ]/(Gjr ). By (4.2), c2,0 gives rise to a 2-cocycle (0, 0, e2) in E0 by j1 , . . . , y−1 V123⊂V12⊂V1 = −x−1 e2 1 x−1 2 x−1 3 This in turn gives rise to the GS cocycle (0, 0, c) by ∂F ∂x0(cid:12)(cid:12)(cid:12)(cid:12)V123 ∂G3 ∂y0 . = −y−1 1 y−1 2 ∂G3 ∂y0 . cV123⊂V12⊂V1 = −y−1 1 y−1 2 Using (4.2) again, we obtain a 2-cocycle (0, e1, e2) in E1 from (0, c1,1, c2,0) with e2 as above and e1 given by V12⊂V1 = y−1 e1 V13⊂V1 = y−1 e1 V23⊂V2 = y−1 e1 1 (cid:18)− 1 (cid:18) ∂G3 2 (cid:18)− ∂y2 ∂G2 ∂y3 f0 + ∂G2 ∂y0 f0 − ∂G3 ∂y0 ∂G3 ∂y1 f0 + ∂G3 ∂y1 f3(cid:19), f2(cid:19), f3(cid:19), V123⊂V1 = y−1 e1 V123⊂V2 = y−1 e1 V123⊂V12 = y−1 e1 1 (cid:18)− 2 (cid:18)− 2 (cid:18)− ∂G3 ∂y2 ∂G3 ∂y1 ∂G3 ∂y1 f0 + f0 + f0 + ∂G3 ∂y0 ∂G3 ∂y1 ∂G3 ∂y1 f2(cid:19), f3(cid:19), f3(cid:19). Then we can deduce a GS cocycle (0, f, 0) from (0, e1, e2). Notice that the expression of m is independent of e2. To have the expression explicitly, by the discussion in §2, we only have to replace the formal base element fi by ◦∂/◦∂yi, then compose with the restriction map. For example, mV12⊂V1 = y−1 1 (cid:18)− ∂G2 ∂y3 ◦∂ ◦∂y0 + ∂G2 ∂y0 ◦∂ ◦∂y3(cid:19) ◦ ρV1 V12 , and so on. Likewise, we conclude that the cocycle (e0, e1, e2) in E2 induced by (c0,2, c1,1, c2,0) has the form e0 V1 = ∂G1 ∂y3 f02 − ∂G1 ∂y2 f03 + ∂G1 ∂y0 f23, e0 V2 = − e0 V3 = ∂G2 ∂y3 f01 + ∂G2 ∂y2 f03 − ∂G2 ∂y0 f13, ∂G3 ∂y2 f01 − ∂G3 ∂y1 f02 + ∂G3 ∂y0 f12. (4.15) mV1 = Thus (e0, e1, e2) induces the GS cocycle (m, 0, 0) given by ◦∂ ◦∂ ∂G1 ∂y0 ∪ ∂y2 · ∂y3 ◦∂ ◦∂ ∂G2 ∂y0 ∪ ∂y2 · ∂y3 ◦∂ ◦∂ ∂G3 ∂y0 ∪ ∂y1 · ∂y2 ◦∂ ◦∂ ∂G1 ∂y2 − ∂y0 ∪ ∂y3 · ◦∂ ◦∂ ∂G2 ∂y0 ∪ ∂y3 · ∂y1 ◦∂ ◦∂ ∂G3 ∂y1 − ∂y0 ∪ ∂y2 · ◦∂ ◦∂ ∂G1 ∂y2 ∪ ∂y0 · ∂y3 ◦∂ ◦∂ ∂G2 ∂y1 ∪ ∂y2 · ∂y3 ◦∂ ◦∂ ∂G3 ∂y1 ∪ ∂y0 · ∂y2 GS(A)r for r = 0, 1, 2. Obviously, dim H 2 Let us look into the dimensions of H 2 mV2 = − mV3 = + + + + . , , P 0 4 = (S/(im ∂uuu))4 = (R/(im ∂uuu))4 = R4/P3 dim P 0 4 = dim R4 − dim GS(A)0 = 1. Since i,j=0 kxi · ∂F/∂xj, we have the following inequality 3Xi,j=0 xi ∂F ∂xj = 35 − dim 3Xi,j=0 kxi ∂F ∂xj ≥ 35 − 16 = 19. 32 LIYU LIU AND WENDY LOWEN Next we investigate the upper bound of dim P 0 independent provided that ∂F/∂xj 6= 0. dim P 0 4 . Obviously, {xi · ∂F/∂xj}0≤i≤3 is k-linearly i=0 kxi · ∂F/∂x0 = 4 and hence 4 ≤ 31. Interestingly, there is a gap between 31 and other possible dimensions. Let us prove In particular, dimP3 Lemma 4.16. If dim P 0 4 6= 31, then 19 ≤ dim P 0 4 ≤ 28. Proof. Suppose F = x4 degree t. 0 + f1x3 0 + f2x2 0 + f3x0 + f4 where ft ∈ k[x1, x2, x3] are homogeneous of First of all, let us reduce the lemma to the case f1 = 0. In fact, dim P 0 GS(A)1 − 1 is invariant under isomorphism of surfaces. By an argument similar to the argument presented in 4 = dim H 2 the paragraph after Theorem 4.3, f1 can be annihilated via the isomorphism x0 7→ x0 − 1 4 f1, xj 7→ xj ( j = 1, 2, 3 ). Now we safely assume f1 = 0. Since dim P 0 4 6= 31, one of ∂F/∂x1, ∂F/∂x2, ∂F/∂x3 is nonzero, say ∂F/∂x1 6= 0. By comparing the degrees of ∂F/∂x0 and ∂F/∂x1 with respect to x0, we obtain (λ1x1 + λ2x2 + λ3x3) ∂F ∂x1 ∈ kxl ∂F ∂x0 3Xl=0 3Mi=0 kxi for some λ1, λ2, λ3 ∈ k only when λ1 = λ2 = λ3 = 0. Hence kxi ∂F ∂xj ⊇ kxi ∂F ∂x0 + kxi ∂F ∂x1 = 3Xi=1 ∂F ∂x0 ⊕ 3Mi=1 kxi ∂F ∂x1 ∼= k7. (cid:3) 3Xi,j=0 3Xi=0 It follows that dim P 0 4 ≤ 35 − 7 = 28. Therefore, dim H 2 GS(A)1 ∈ {20, . . . , 29}∪ {32}. The dimension indeed reaches every number in the set. We list some examples in Table 1 showing this fact. By Lemma 4.12, we are able to check 4 also corresponds to a class in H 1(X,TX ). Accordingly, the dimensions of H 1(X,TX ) for if t ∈ P 0 these examples can be computed, as listed in the third column. F x4 0 + x4 0 + x2 (x2 0 + x2 0 + x2 (x2 1 + x4 1)2 + x4 1)2 + (x2 1 + x2 2 + x4 3 2 + x4 3 2 + x2 3)2 2)2 + x4 3 (x2 0 + x4 x4 (x2 0 + x2 0 + x2 (x2 1 + x4 2 1)2 + x4 2 2 + x2 3)2 2)2 1 + x2 0 + x2 1 + x2 (x2 0 + x4 x4 1 (x2 0 + x2 1)2 x4 0 dim H 2 GS(A)1 dim H 1(X,TX ) dim H 2 GS(A)2 1 20 20 21 22 23 24 25 26 27 28 29 32 4 2 2 1 1 1 1 1 1 1 1 2 5 1 5 17 17 11 11 31 Table 1. dimensions of several groups For r = 2, the group Q−2 2 comes from the complex S6 2 / im ∂vvv ∂uuu−−→ S4 5 / im ∂vvv ∂uuu−−→ S8 HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 33 by (4.6). It fits into a projection R6 2 S6 2 / im ∂vvv ∂uuu ∂uuu R4 5 ∂uuu R8 / S4 5 / im ∂vvv ∂uuu / S8 of complexes. By Euler's formula, the projection turns out to be a quasi-isomorphism. Hence 2 ∼= ker{∂uuu : R6 Q−2 5}. The dimension of the latter is easier to compute than that of Q−2 2 . Let elements in R6 2 → R4 2 be expressed by (a01, a02, a03, a12, a13, a13). If F = x4 0 + (x2 1 + x2 2)2, then and hence Q−2 2 ker{∂uuu : R6 2 → R4 5} = {(0, 0, 0, 0, x2u,−x1u) u ∈ R1} is equal to whose dimension is 4; if F = (x2 0 + x2 1 + x2 2 + x2 3)2, then Q−2 2 is equal to the (direct) sum of {(0, 0, 0, 0, x2u,−x1u) + im ∂vvv u ∈ S1} {(0, x3u,−x2u, 0, 0, x0u) + im ∂vvv u ∈ S1}, {(x3v, 0,−x1v, 0, x0v, 0) + im ∂vvv v ∈ S1}, {(x2p,−x1p, 0, x0p, 0, 0) + im ∂vvv p ∈ S1}, {(0, 0, 0, x3q,−x2q, x0q) + im ∂vvv q ∈ S1}, and so dim Q−2 2 = 16. We omit the computational details and list the dimensions of H 2 these examples in the right column of Table 1. It is obvious that the lower bound of dim H 2 1. However, in the general case, the authors do not know either the upper bound of dim H 2 or any gaps between the bound and 1. GS(A)2 of GS(A)2 is GS(A)2, Recall that when X is smooth, the Hodge numbers of X are defined to be hp,q = dim H q(X, Ωp X ). Let ωX = Ω2 X be the canonical sheaf of X. Then ωX ∼= OX and by [4, Cor. 3.1.4], X ) ⊕ H 1(X, ΩX ) ⊕ H 0(X,OX ). GS(A) ∼= HH 2(ωX ) ∼= H 2(X, Ω2 H 2 The dimensions of the three summands are h2,2 = 1, h1,1 = 20, h0,0 = 1 respectively. So dim H 2 GS(A)r reaches its smallest possible values for r = 0, 1, 2 if X is smooth. The converse is not true, as there indeed exist non-smooth surfaces with dim H 2 and dim H 2 GS(A)0 = dim H 2 0 + x4 GS(A)2 = 1. Let us give two examples here. 3. We know uuu = (4x3 3 + 3x4 1 + x4 2 − 4x2x3 Example 4.1. Let F = x4 3(x2 − x3)). A direct computation shows that dim P 0 2 = 0. Note that the surface has three isolated singularities (0 : 0 : 1 : ζr) for r = 0, 1, 2 where ζ is a primitive third root of 1. Furthermore, we have dim H 1(X,TX ) = 11, in accordance with Theorem 4.14. Example 4.2. The Kummer surfaces Kµ are a family of quartic surfaces given by 4 = 19 and dim Q−2 3),−12x2 2 − x3 1, 4(x3 0, 4x3 GS(A)1 = 20 F = (x2 0 + x2 1 + x2 2 − µ2x2 3)2 − λpqrs where and p, q, r, s are the tetrahedral coordinates λ = 3µ2 − 1 3 − µ2 q = x3 − x2 + √2x0, √2x1. s = x3 + x2 − √2x0, p = x3 − x2 − r = x3 + x2 + √2x1, / /     / /         / / 34 LIYU LIU AND WENDY LOWEN When µ2 6= 1/3, 1, or 3, Kµ has 16 isolated singularities which are ordinary double points. In this case, one can check that uuu is a regular sequence in R. Thus dim P 0 2 = 0. We also have dim H 1(X,TX ) = 1, in accordance with Theorem 4.14. The examples given above with dim H 0(X,∧2TX ) = dim H 2 GS(A)2 = 1 are all integral, and vice versa. We will give two examples to show this condition is neither necessary nor sufficient for 4 = 19 and dim Q−2 integrality of X. Example 4.3. Let F = (x2 1 + 2x2 dim H 0(X,∧2TX ) = 1. However, this is not integral. Example 4.4. Let F = x4 1 + 2x2 0 + x2 0 + x2 2)(x2 0 + x3 (0, 0, 0, 0, x1u,−x2u) + im ∂vvv, which is 4-dimensional. 1x2. This gives rise to an integral scheme. But Q−2 2 is spanned by u ∈ {x0, x1, x2, x3} 3). We can easily prove Q−2 2 = 0 and hence According to our general results, for a smooth K3 surface, we have P 0 Eres and dim P 0 interpretations of Hochschild 2-classes in P 0 3), P 0 in Table 1. Since uuu = (4x3 GS(A)1 = 4 = 19. To end this section, let us present the resulting two different deformation 4 for the Fermat quartic surface, i.e. the first example 4 has a basis 4 = Emult ⊆ H 2 2, 4x3 We fix the generators and relations of A(V ) for all V ∈ V as follows: 3 i0 + i1 + i2 + i3 = 4, 0 ≤ i0, i1, i2, i3 ≤ 2(cid:9). 0, 4x3 1 xi2 1, 4x3 2 xi3 0 xi1 (cid:8)xi0 0 + y4 2 + y4 1 + y4 0 + y4 0 + y4 1 ]/(y4 1 , y−1 2 ]/(y4 0 xi1 1 xi2 2 xi3 mV1 = yi0 0 yi2 2 yi3 mV12 = yi0 0 yi1 1 yi3 3 3 3 + 1), 2 + 1), 1 + y4 0 + y4 3 ∈ P 0 ◦µ, mV2 = yi0 ◦µ, mV13 = yi0 mV123 = yi0 A2 = k[y0, y1, y3]/(y4 A12 = k[y0, y1, y3, y−1 A23 = k[y0, y1, y2, y−1 0 + y4 1 ]/(y4 2 ]/(y4 1 + y4 0 + y4 0 + y4 3 + 1), 1 + y4 1 + y4 3 + 1), 2 + 1), 2 + 1), 1 + y4 4 , there is a deformation (m, 0, 0) of A given by 2 + 1). 3 0 yi1 0 yi1 0 yi1 1 yi3 1 yi2 1 yi2 2 2 ◦µ, mV3 = yi0 ◦µ, mV23 = yi0 ◦µ. 2 0 yi1 0 yi1 1 yi2 1 yi2 2 ◦µ, ◦µ, A1 = k[y0, y2, y3]/(y4 A3 = k[y0, y1, y2]/(y4 A13 = k[y0, y1, y2, y−1 A123 = k[y0, y1, y2, y−1 For any basis element xi0 We remark that although the same notation ◦µ is used, it stands for Hochschild 2-cocycles of individual algebras. Since in A1 one has it follows that 1 4 1 = 4y3 0(cid:18)− ◦µ = dHoch(cid:18)− ◦µ = dHoch(cid:18)− ◦µ = dHoch(cid:18)− 2(cid:18)− y0(cid:19) + 4y3 1 4 1 4 y0 ◦∂ ◦∂y0 − y3 ◦∂ 1 4 1 4 ◦∂ ◦∂y2 − 3(cid:18)− y3(cid:19), y2(cid:19) + 4y3 ◦∂y3(cid:19). ◦∂y3(cid:19), ◦∂y2(cid:19). ◦∂ ◦∂y1 − ◦∂ ◦∂y1 − 1 4 1 4 ◦∂ ◦∂ y3 y2 y2 y1 y1 1 4 1 4 1 4 Similarly, for A2 and A3, we respectively have ◦∂ ◦∂y0 − ◦∂ ◦∂y0 − 1 4 1 4 y0 y0 The three preimages are denoted by s1, s2, s3. By abuse of notation, they also denote 1-cochains of the algebras A12, A13 and so on. Then we have mV1 = dHoch(yi0 0 yi2 2 yi3 3 s1), mV2 = dHoch(yi0 0 yi1 1 yi3 3 s2), HOCHSCHILD COHOMOLOGY OF PROJECTIVE HYPERSURFACES 35 mV3 = dHoch(yi0 mV13 = dHoch(yi0 mV123 = dHoch(yi0 0 yi1 0 yi1 0 yi1 1 yi2 1 yi2 1 yi2 2 s3), 2 s3), 2 s3). mV12 = dHoch(yi0 mV23 = dHoch(yi0 0 yi1 0 yi1 1 yi3 1 yi2 3 s2), 2 s3), We choose a map λ : V → U by λ(Vj1...jr ) = Ujr if j1 < ··· < jr. We thus obtain an equivalent deformation (0, f, 0) whose nonzero components of f are V12 − ρV1 V13 − ρV1 V23 − ρV2 V123 − ρV1 V123 − ρV2 V123 − ρV12 2 yi3 0 yi2 V12 ◦ yi0 V13 ◦ yi0 0 yi2 2 yi3 V23 ◦ yi0 0 yi1 1 yi3 3 s2, V123 ◦ yi0 0 yi2 2 yi3 V123 ◦ yi0 0 yi1 1 yi3 V123 ◦ yi0 0 yi1 1 yi3 fV12⊆V1 = yi0 fV13⊆V1 = yi0 fV23⊆V2 = yi0 fV123⊆V1 = yi0 fV123⊆V2 = yi0 fV123⊆V12 = yi0 0 yi1 0 yi1 0 yi1 0 yi1 0 yi1 0 yi1 1 yi3 1 yi2 1 yi2 1 yi2 1 yi2 1 yi2 3 s2 ◦ ρV1 2 s3 ◦ ρV1 2 s3 ◦ ρV2 2 s3 ◦ ρV1 2 s3 ◦ ρV2 2 s3 ◦ ρV12 3 s1, 3 s2, 3 s1, 3 s2. 3 s1, References [1] Th´eorie des intersections et th´eor`eme de Riemann-Roch, Lecture Notes in Mathematics, vol. 225, Springer- Verlag, Berlin-New York, 1971, S´eminaire de G´eom´etrie Alg´ebrique du Bois-Marie 1966 -- 1967 (SGA 6), Dirig´e par P. Berthelot, A. Grothendieck et L. Illusie. Avec la collaboration de D. Ferrand, J. P. Jouanolou, O. Jussila, S. Kleiman, M. Raynaud et J. P. Serre. [2] Buenos Aires Cyclic Homology Group, Hochschild and cyclic homology of hypersurfaces, Adv. Math. 95 (1992), 18 -- 60. [3] J.-L. Brylinski, Loop spaces, characteristic classes and geometric quantization, Progress in Mathematics, vol. 107, Birkhauser Boston, Inc., Boston, MA, 1993. [4] R.-O. Buchweitz and H. Flenner, The global decomposition theorem for Hochschild (co-)homology of singular spaces via the Atiyah-Chern character, Adv. Math. 217 (2008), 243 -- 281. [5] [6] H. Dinh Van, L. Liu, and W. Lowen, Non-commutative deformations and quasi-coherent modules, , Global Hochschild (co-)homology of singular spaces, Adv. Math. 217 (2008), 205 -- 242. arXiv:1411.0331, 2014. [7] M. Gerstenhaber and S. D. Schack, On the deformation of algebra morphisms and diagrams, Trans. Amer. Math. Soc. 279 (1983), 1 -- 50. [8] [9] , A Hodge-type decomposition for commutative algebra cohomology, J. Pure Appl. Algebra 48 (1987), 229 -- 247. , Algebraic cohomology and deformation theory, Deformation theory of algebras and structures and applications (Il Ciocco, 1986), NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., vol. 247, Kluwer Acad. Publ., Dordrecht, 1988, pp. 11 -- 264. [10] , The cohomology of presheaves of algebras. I. Presheaves over a partially ordered set, Trans. Amer. Math. Soc. 310 (1988), 135 -- 165. [11] R. Hartshorne, Algebraic geometry, Graduate Texts in Mathematics, vol. 52, Springer-Verlag, New York- Heidelberg, 1977. [12] L. Illusie, Complexe cotangent et d´eformations. II, Lecture Notes in Mathematics, vol. 283, Springer-Verlag, Berlin-New York, 1972. [13] Luc Illusie, Complexe cotangent et d´eformations. I, Lecture Notes in Mathematics, vol. 239, Springer-Verlag, Berlin-New York, 1971. [14] B. Keller, Derived invariance of higher structures on the Hochschild complex, preprint http://www.math.jussieu.fr/~keller/publ/dih.dvi. [15] D. W. Knudson, On the deformation of commutative algebras, Trans. Amer. Math. Soc. 140 (1969), 55 -- 70. [16] M. Kontsevich, Homological algebra of mirror symmetry, Proceedings of the International Congress of Math- ematicians, Vol. 1, 2 (Zurich, 1994) (Basel), Birkhauser, 1995, pp. 120 -- 139. [17] W. Lowen and M. Van den Bergh, The curvature problem for formal and infinitesimal deformations, Preprint arXiv:1505.03698. [18] [19] [20] R. I. Michler, Hodge-components of cyclic homology for affine quasi-homogeneous hypersurfaces, Ast´erisque , Hochschild cohomology of abelian categories and ringed spaces, Adv. Math. 198 (2005), 172 -- 221. , Deformation theory of abelian categories, Trans. Amer. Math. Soc. 358 (2006), 5441 -- 5483. (1994), 321 -- 333. [21] T. Saito, Parity in Bloch's conductor formula in even dimension, J. Th´eor. Nombres Bordeaux 16 (2004), 403 -- 421. [22] F. Schuhmacher, Hochschild cohomology of complex spaces and Noetherian schemes, Homology Homotopy Appl. 6 (2004), 299 -- 340. [23] R. G. Swan, Hochschild cohomology of quasiprojective schemes, J. Pure Appl. Algebra 110 (1996), 57 -- 80. [24] K. Wolffhardt, The Hochschild homology of complete intersections, Trans. Amer. Math. Soc. 171 (1972), 51 -- 66. MR 0306192 (46 #5319) [25] A. Yekutieli, The continuous Hochschild cochain complex of a scheme, Canad. J. Math. 54 (2002), 1319 -- 1337. 36 LIYU LIU AND WENDY LOWEN (Liyu Liu) Departement Wiskunde-Informatica, Universiteit Antwerpen, Middelheimcampus, Middel- heimlaan 1, 2020 Antwerp, Belgium Current address: School of Mathematical Science, Yangzhou University, No. 180 Siwangting Road, 225002 Yangzhou, Jiangsu, China E-mail address: [email protected] (Wendy Lowen) Departement Wiskunde-Informatica, Universiteit Antwerpen, Middelheimcampus, Mid- delheimlaan 1, 2020 Antwerp, Belgium E-mail address: [email protected]
0812.0169
3
0812
2013-12-28T16:53:57
Quantum Field Theories on Algebraic Curves. I. Additive bosons
[ "math.AG", "hep-th", "math.QA", "math.RT" ]
Using Serre's adelic interpretation of cohomology, we develop a `differential and integral calculus' on an algebraic curve X over an algebraically closed filed k of constants of characteristic zero, define algebraic analogs of additive multi-valued functions on X and prove corresponding generalized residue theorem. Using the representation theory of the global Heisenberg and lattice Lie algebras, we formulate quantum field theories of additive and charged bosons on an algebraic curve X. These theories are naturally connected with the algebraic de Rham theorem. We prove that an extension of global symmetries (Witten's additive Ward identities) from the k-vector space of rational functions on X to the vector space of additive multi-valued functions uniquely determines these quantum theories of additive and charged bosons.
math.AG
math
QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES I. ADDITIVE BOSONS LEON A. TAKHTAJAN Abstract. Using Serre's adelic interpretation of cohomology, we de- velop a 'differential and integral calculus' on an algebraic curve X over an algebraically closed filed k of constants of characteristic zero, define algebraic analogs of additive multi-valued functions on X and prove cor- responding generalized residue theorem. Using the representation theory of the global Heisenberg and lattice Lie algebras, we formulate quantum field theories of additive and charged bosons on an algebraic curve X. These theories are naturally connected with the algebraic de Rham theo- rem. We prove that an extension of global symmetries (Witten's additive Ward identities) from the k-vector space of rational functions on X to the vector space of additive multi-valued functions uniquely determines these quantum theories of additive and charged bosons. Bibliography: 18 titles. 1. Introduction The classical theory of compact Riemann surfaces has an algebraic coun- terpart, the theory of algebraic functions of one variable over an arbitrary field of constants, as developed by Dedekind and Weber. The introduction of differentials into the algebraic theory by Artin and Hasse and the definition of id`eles and ad`eles adeles by Chevalley and Weil opened the way for ap- plication of infinite-dimensional methods to the theory of algebraic curves. Classical examples of using such methods are given by Serre's adelic inter- pretation of cohomology and the Riemann-Roch theorem [13] and Tate's proof of the general residue theorem [15]. In 1987, Arbarello, de Concini and Kac [1] interpreted Tate's approach in terms of central extensions of infinite-dimensional Lie algebras and gave a new proof of Weil's celebrated reciprocity law using the infinite-wedge representation. In 1987, Kazhdan [10] and Witten [17] proposed an adelic formulation of the quantum field theory of one-component free fermions on an alge- braic curve, and Witten [18] outlined an approach to other quantum field theories. Let X be an algebraic curve over an algebraically closed field k Key words and phrases. Algebraic curves and algebraic functions, ad`eles, additive multi-valued functions, additive Ward identities, Heisenberg algebra, current algebra on algebraic curve, generalized residue theorem, Fock spaces, quantum theories of additive and charged bosons, expectation value functional. This work was partially supported by the NSF grants DMS-0204628, DMS-0705263 and DMS-1005769. 1 2 LEON A. TAKHTAJAN of constants, and let L be a spin structure on X. We write M(L) for the infinite-dimensional k-vector space of meromorphic sections of L over X, and MP for the completions of M(L) at all points P ∈ X. In outline, the approach of [10, 17] can be described in terms of the following objects. • The global Clifford algebra ClX on X: a restricted direct product over all points P ∈ X of the local Clifford algebras ClP , which are related to the k-vector spaces MP by the residue maps ResP (f g). • The adelic Clifford module FX (the global fermion Fock space): a restricted Z/2Z-graded tensor product of the local Clifford modules FP over all P ∈ X. • The 'expectation value' functional: a linear map h · i : FX → k, satisfying the condition hf · ui = 0 for all f ∈ M(L) ⊂ ClX, u ∈ FX, (1.1) where the vector space M(L) is embedded diagonally into the global Clifford algebra ClX. In this pure algebraic formulation of one-component free fermions on an algebraic curve, the products of field operators at points P ∈ X are replaced by the vectors u = ⊗P ∈XuP ∈ FX, and the linear map h · i is a mathematical way of defining the correlation functions of quantum fields. At the physical level of rigor, these functions are introduced by the Feynman path integral. The vector space M(L) acts on FX by global symmetries, and the invariance of the quantum theory of free fermions with respect to these symmetries is expressed by the quantum conservation laws (1.1), also known as the additive Ward identities. It is proved in [17, 18] that if the spin structure L has no global holomorphic sections, then the additive Ward identities uniquely determine the expectation value functional h · i. The relations (1.1) are compatible with the global residue theorem on X: ResP (f dg) = 0, f, g ∈ M(L). XP ∈X Witten [18] developed the basics of quantum field theories associated with current algebras on an algebraic curves and mentioned the theories associ- ated with loop groups on algebraic curves. The global symmetries of these theories are respectively given by the rational maps of the algebraic curve X to a finite-dimensional semi-simple Lie algebra over k and the rational maps of X to the corresponding Lie group. In the latter case, the analogues of quantum conservation laws (1.1) were called multiplicative Ward identi- ties in [18]. It was emphasized in [18, Sect. IV] that if the genus of X is greater than zero, then the Ward identities do not uniquely determine the expectation value functional h · i, even in the Lie-algebraic case. Thus the main problem in the construction of quantum field theories on an algebraic curve is to find additional conditions which would uniquely determine the linear functional h · i. QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 3 When X is a Riemann surface (that is, an algebraic curve over the field C of complex numbers, equipped with the complex topology), the usual physicist's representation of correlation functions is given by the Feynman path integral, which uses the Lagrangian formulation of the theory. This approach is not applicable in the case when X is an algebraic curve over an arbitrary field of constants. Hence one needs other ways to define the correlation functions. The basic example is given by the 'integrable' case, when the expectation value functional is uniquely determined by the global symmetries (and hence so are all correlation functions). In [14] we we gave a solution of the problem of the unique determina- tion of the expectation value functional for the simplest scalar theories, when the finite-dimensional Lie algebra is the abelian Lie algebra k, and the corresponding Lie group is the multiplicative group k∗ = k \ {0}. We call these quantum field theories the theories of additive and multiplicative bosons respectively. The solution suggested in [14] involves enlarging the global symmetries by considering algebraic analogue of the vector space ad- ditive multi-valued functions on a Riemann surface (analogues of the clas- sical abelian integrals of the second kind with zero a-periods). Although the classical theory of abelian integrals was already developed by Riemann (see, for example, [7] and [11] for a modern exposition), the corresponding algebraic theory (integral calculus on algebraic curves) has not been fully developed. In this paper we partially fill this gap in the case when the field k of constants has characteristic zero and give an explicit construction of the quantum field theories of additive bosons on an algebraic curve. These theo- ries are naturally connected with the algebraic de Rham theorem, and their global symmetries form a vector space of additive multi-valued functions; see Theorems 6 and 7 for precise statements. Our construction of quantum field theories on algebraic curves may be regarded as an algebraic analogue of the geometric realization of conformal field theories on Riemann surfaces in [9]. The quantum field theory of multiplicative bosons requires an algebraic analogue of the group of multiplicative multi-valued functions on a Riemann surface (analogues of the exponentials of abelian integrals of the third kind with zero a-periods). We plan to discuss the analogue of this group and the corresponding multiplicative Ward identities in a separate publication. Here is the more detailed description of the contents of the paper. In Section 2 we recall the necessary basic facts from the theory of algebraic curves. Namely, let X be an algebraic curve of genus g over an algebraically closed field k of constants, F = k(X) the field of rational functions on X, and FP the corresponding local fields (the completions of F with respect to the regular discrete valuations vP corresponding to the discrete valuation rings at points P ∈ X). In Section 2.1 we introduce the ring of ad`eles AX = aP ∈X FP 4 LEON A. TAKHTAJAN as a restricted direct product of the local fields FP and describe Serre's adelic interpretation of cohomology. In Section 2.2 we recall the definitions of the F -module Ω1 F/k of Kahler differentials on X, the corresponding AX- module of differential ad`eles ΩX, the differential map d : AX → ΩX and the residue map Res : ΩX → k. In Section 2.3 we describe Serre duality and the Riemann-Roch theorem. In Section 3, assuming that the field k of constants is of characteristic zero, we recall the differential calculus on an algebraic curve X (the structural theory of the k-vector space of Kahler differentials Ω1 F/k on X) and develop a corresponding integral calculus. Namely, in Section 3.1 we follow [2] and [3] and endow the k-vector space Ω(2nd) of differentials of the second kind (that is, differentials on X with zero residues) with a skew-symmetric bilinear form ResP (d−1ω1 ω2), ω1, ω2 ∈ Ω(2nd). (ω1, ω2)X = XP ∈X The main result of the differential calculus is Theorem 4, an algebraic version of de Rham theorem. Theorem 4 goes back to Chevalley and Eichler and, for an algebraic curve X of genus g ≥ 1, states1 that the 2g-dimensional k-vector space Ω(2nd)/dF is a symplectic vector space with symplectic form ( , )X . Moreover, for every choice of a non-special effective divisor D = P1 + · · · + Pg of a degree g on X and uniformizers ti at Pi, there is an isomorphism Ω(2nd)/dF ≃ Ω(2nd) ∩ Ω1 F/k(2D). F/k(2D) has a natural symplectic basis {θi, ωi}g The space Ω(2nd)∩Ω1 i=1, where θi (resp. ωi) are differentials of the first (reps. second) kind with the following properties. The θi vanish at all points Pj for j 6= i, and θi = (1 + O(ti))dti at Pi, while the ωi are regular at all points Pj for j 6= i, and ωi = (t−2 i +O(ti))dti at Pi. Hence θi (reap. ωi) are algebraic analogues of the differentials of the first kind with normalized a-periods (resp. differentials of the second kind with second order poles, zero a-periods and normalized b-periods) on a com- pact Riemann surface. The a-periods of ω ∈ Ω(2nd) are algebraically defined by (ω, ωi)X , i = 1, . . . , g, and we write Ω(2nd) for the isotropic subspace of Ω(2nd) consisting of all differentials of the second kind with zero a-periods. By Proposition 1 we have Ω(2nd) (1.2) 0 = k · ω1 ⊕ · · · ⊕ k · ωg ⊕ dF. 0 In Section 3.1 we also introduce an algebraic notion of additive multi- valued functions on X. By definition, the k-vector space of additive multi- valued functions is a subspace A(X) of the ad`ele ring AX satisfying F ⊂ A(X) and dA(X) ⊂ Ω1 F/k and the additional condition that if a ∈ A(X) and da = 0, then a = c ∈ k. The main result of the integral calculus for dif- ferentials of the second kind with zero a-periods is the explicit construction 1The case g = 0 is trivial. QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 5 of the vector space A(X, D) in Example 1. This space plays a fundamental role in the theory of additive bosons. It is parametrized by the choices of a non-special divisor D = P1 + · · · + Pg of degree g on X, local uniformizers ti at the points Pi and solutions of the equations dηi = ωi in AX (with any fixed choice of the local additive constants). It is defined as A(X; D) = k · η1 ⊕ · · · ⊕ k · ηg ⊕ F ⊂ AX and possesses the property d(A(X; D)) = Ω(2nd) . We finally introduce the additive multi-valued functions η(n) P ∈ A(X; D) with a single pole of order n at P ∈ X and prove (see Lemma 3.1) that every rational function f ∈ F has a unique partial fraction expansion, the partial fractions being these η(n) P . We also explain the difficulties arising in an attempt to define algebraic analogues of multiplicative multi-valued functions. 0 In Section 4 we formulate the local quantum field theories of additive and charged bosons. The local theory of additive bosons is associated with the representation theory of the local Heisenberg algebra gP , a one-dimensional central extension of the abelian Lie algebra FP , P ∈ X, by the 2-cocycle cP (f, g) = − ResP (f dg). In Section 4.1 we introduce the highest-weight rep- resentation ρ of gP on the local Fock space FP and define the corresponding contragradient representation ρ∨ of gP on the dual local Fock space F ∨ P . In Section 4.2 we define a local lattice algebra lP as a semi-direct sum of the local Heisenberg algebra gP and the abelian Lie algebra k[Z], the group al- gebra of Z. The corresponding irreducible highest-weight lP -module is the local Fock space BP of 'charged bosons' (a tensor product of k[Z] and FP ). The material in Sections 4.1 and 4.2 is essentially standard (see [8, 4]). In Section 5 we finally state the global quantum field theories, starting in Section 5.1 with the theory of additive bosons on an algebraic curve X. This theory is naturally connected with the global Heisenberg algebra gX, a one-dimensional central extension of the abelian Lie algebra gl1(AX ) = AX is isotropic with by the 2-cocycle cX =PP ∈X cP . Since the subspace Ω(2nd) respect to the bilinear form ( , )X , we have 0 cX(a1, a2) = 0 ∀ a1, a2 ∈ A(X, D) ⊂ AX. (1.3) This may be regarded as a generalized residue theorem for additive multi- valued functions. The irreducible highest-weight module of the global Heisen- berg algebra gX is the global Fock space FX , a restricted tensor product of the local Fock spaces FP over all points P ∈ X. It may be regarded as the space of observables of the quantum theory of additive bosons on X. In Theorem 6 we prove that there is a unique normalized expectation value functional h · i : FX → k, uniquely characterized by the global symmetries hρ(a)vi = 0 ∀ a ∈ A(X; D), v ∈ FX . (1.4) Here A(X; D) ⊂ AX is the vector space of additive multi-valued functions on X, defined in Section 3.1, and ρ : gX → End FX is the corresponding 6 LEON A. TAKHTAJAN representation of the global Heisenberg algebra. Specifically, we show in Theorem 6 that hvi = (ΩX , v) for all v ∈ FX , where ΩX ∈ F ∨ system of equations X is a vector in the dual space to FX , satisfying an infinite ΩX · ρ∨(a) = 0 ∀ a ∈ A(X, D). (1.5) The vector ΩX is given by an explicit formula (see Theorem 6), which en- codes the analogues of all correlation functions of quantum additive bosons on X. The compatibility of the system (1.5) is based on the reciprocity law (proved in Lemma 3.1) for the differentials of the second kind with zero a-periods. The additive Ward identities (1.4) are also compatible with the general- ized residue theorem. Namely, since [ρ(x), ρ(y)] = cX(x, y)I for x, y ∈ AX, where I is the identity operator on FX , we get from (1.4) that for a1, a2 ∈ A(X, D), 0 = h(ρ(a1)ρ(a2) − ρ(a2)ρ(a1))vi = cX (a1, a2)hvi ∀ v ∈ FX , which yields (1.3). In Section 5.2 we define a global lattice algebra lX as a semi-direct sum of the global Heisenberg algebra gX and the abelian Lie algebra k[Div0(X)] with generators eD, where D ∈ Div0(X) is the group algebra of the additive group Div0(X) of divisors of degree 0 on X. Its irreducible highest weight module is the global Fock space BX of charged bosons, which is the tensor product of the group algebra k[Div0(X)] and the Fock space FX of addi- tive bosons. The main result of Section 5.2 is Theorem 7 on the existence and uniqueness of an expectation value functional h · i : BX → k which is normalized with respect to the action of the group algebra k[Div0(X)] and satisfies the additive Ward identities (1.4) with respect to the action of the global symmetries (additive multi-valued functions in A(X, D)) on the global Fock space BX. This functional is of the form hvi = ( ΩX, v) where the vector ΩX ∈ B∨ X in the space dual to BX is given by an explicit for- mula (see Theorem 7), which encodes all correlation functions of quantum charged additive bosons on X. In Section 5.3, following a suggestion of the referee, in Section 5.3 we give a more invariant formulation of Theorem 6. I am grateful to the referee for his careful reading of the manuscript, con- structive criticism, remarks, and valuable suggestions. 2. Basic Facts Here we recall necessary facts from the theory of algebraic curves. This material is essentially standard (see [2, 13, 7]). QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 7 2.1. Definitions. An algebraic curve X over an algebraically closed field k is an irreducible non-singular one-dimensional projective variety over k. It is equipped with the Zariski topology. The field F = k(X) of rational functions on X is a finitely generated extension of k of transcendence degree 1. Conversely, every finitely generated extension of k of transcendence degree 1 corresponds to a unique (up to isomorphism) algebraic curve over k. Closed points P on X correspond to discrete valuation rings OP (subrings of F ). The rings OP for all P ∈ X form a sheaf of rings on X: the structure sheaf OX, a subsheaf of the constant sheaf F . For every point P ∈ X let vP be the regular discrete valuation of F over k, corresponding to the discrete valuation ring OP . The completion of F with respect to vP is a complete closed field FP with valuation ring OP , which is the completed local ring at P with prime ideal p and residue class field k = OP /p. The ring AX of ad`eles of X is AX = aP ∈X FP , D = XP ∈X (f ) = XP ∈X deg D = XP ∈X the restricted direct product over all points P ∈ X of the local fields FP with respect to the local rings OP . By definition, x = {xP }P ∈X ∈ AX if xP ∈ OP for all but finitely many P ∈ X. The field F is contained in all local fields FP and is diagonally embedded in AX: F ∋ f 7→ {f P }P ∈X ∈ AX. The divisor group Div(X) of X is the free abelian group generated by the points P ∈ X. By definition, nP · P ∈ Div(X) if nP = vP (D) ∈ Z and nP = 0 for all but finitely many P ∈ X. Divisors of the form vP (f ) · P ∈ Div(X), where f ∈ F ∗ = F \ {0}, are called principal divisors. They form a subgroup PDiv(X) ≃ F ∗/k∗ of Div(X). The degree of a divisor D is nP = XP ∈X vP (D) ∈ Z, and we have deg(f ) = 0 for f ∈ F ∗. A divisor D is said to be effective, if vP (D) ≥ 0 for all P ∈ X. By definition, D1 and D2 are linear equivalent (D1 ∼ D2) if D1 − D2 = (f ) for some f ∈ F ∗. The equivalence classes of divisors form the divisor class group Cl(X) = Div(X)/PDiv(X). For every divisor D we define a subspace AX(D) of the k-vector space AX by putting AX(D) = {x ∈ AX : vP (xP ) ≥ −vP (D) ∀ P ∈ X} . 8 LEON A. TAKHTAJAN The ring AX of ad`eles is a topological ring with the product topology. A base of neighborhoods of 0 is given by the subspaces AX(D), D ∈ Div(X), and AX is a k-vector space with linear topology in the sense of Lefschetz [12, Ch. II, §6]. Since subspaces AX(D) is linear compact, AX is locally linear compact. The k-vector space F = k(X) is discrete in AX and the quotient space AX/F is linear compact [7, App., §3]. For every divisor D we have an algebraic coherent sheaf OX (D) on X whose stalk at any point P ∈ X is OX (D)P = {f ∈ F : vP (f ) ≥ −vP (D)}. Linear equivalent divisors correspond to isomorphic sheaves. We denote the Cech cohomology groups of the sheaf by OX (D) H i(X, OX (D)) (these are finite-dimensional vector spaces over k, trivial for i > 1) and put hi(D) = dimk H i(X, OX (D)). The zero divisor D = 0 corresponds to the structure sheaf OX . In this case, h0(0) = 1 and h1(0) = g is the arithmetic genus of X. We have H 0(X, OX (D)) = AX (D) ∩ F, H 1(X, OX (D)) ≃ AX/(AX (D) + F ), which is Serre's adelic interpretation of cohomology [13, Ch. II, §5]. 2.2. Differentials and residues. Let R be a ring over k. The module Ω1 R/k of Kahler differentials of R is the universal R-module with the property that there is a k-linear map d : R → Ω1 R/k satisfying the Leibniz rule d(f g) = f dg + gdf, f, g ∈ R. When R = F is the field k(X) of rational functions on an algebraic curve X, Ω1 F/k is a one-dimensional vector space over F . Let t ∈ F be a local coordinate at P in the Zariski topology, that is, a rational function on X with vP (t) = 1. Then dt is a basis of the F -vector space Ω1 F/k, that is, every Kahler differential can be written as ω = f dt for some f ∈ F . The order of ω ∈ Ω1 F/k at P is defined by vP (ω) = vP (f ). It is independent of the choice of the local coordinate at P and determines a valuation on Ω1 F/k. The family of OP -modules Ω1 OP /k for all points P ∈ X forms an algebraic coherent sheaf Ω, a subsheaf of the constant sheaf Ω1 F/k. Moreover, Ω1 F/k = Ω1 OP /k ⊗ OP F. When k has characteristic 0, the FP -module Ω1 FP /k is an infinite-dimensional FP -vector space for every point P ∈ X (the map d is not continuous with respect to the p-adic topology on FP ). Following [13, Ch. II, §11], we define Ω1 FP /k = Ω1 FP /k/Q, QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 9 where Q = ∩n≥0 pnd(OP ) and, therefore, dimFP FP -module Ω1 the valuation vP . The completion of the OP -module Ω1 Ω1 FP /k is the completion of the F -module Ω1 FP /k = 1. The resulting F/k with respect to OP /k is the OP -module Ω1 OP /k and FP /k = Ω1 Ω1 OP /k ⊗ OP FP . We define the AX-module ΩX of differential ad`eles of the sheaf Ω by the formula ΩX = aP ∈X Ω1 FP /k. This is the restricted direct product over all points P ∈ X of the FP - modules Ω1 is contained in all FP -modules Ω1 F/k FP /k and is diagonally embedded into ΩX: FP /k with respect to the OP -modules Ω1 OP /k. The F -module Ω1 Ω1 F/k ∋ ω 7→ {ωP }P ∈X ∈ ΩX. The k-vector space ΩX has a linear topology with a base of neighborhoods of zero given by the subspaces ΩX(D) for all D ∈ Div(X): ΩX(D) = {ω = {ωP }P ∈X ∈ ΩX : vP (ωP ) ≥ −vP (D) ∀ P ∈ X}. This topological space is locally linear compact. The maps d : FP → Ω1 FP /k for all P ∈ X determine a continuous map d : AX → ΩX satisfying the Leibniz rule. Remark 1. The AX-module ΩX is essentially the set of 'principal part sys- tems of degree 1' on X in the sense of Eichler (see [3, Ch. III, §5.2]). Take ω ∈ Ω1 is a basis of the OP -module Ω1 defined as FP /k, and let t be a local parameter of the field FP , so that dt FP /k → k is OP /k. The residue map ResP : Ω1 ResP (ω) = c−1, where ω = cntndt, ∞Xn≫−∞ and the symbol n ≫ −∞ means that the summation is taken over only finitely many negative values of n. The definition of the residue is indepen- dent of the choice of the local parameter. The residue map is continuous with respect to the p-adic topology on Ω1 FP /k and the discrete topology on k. The local residue maps ResP give rise to the global residue map Res : ΩX → k, Res ω = XP ∈X ResP (ωP ), ω = {ωP }P ∈X ∈ ΩX. The global residue map is well-defined, continuous, and possesses the fol- lowing fundamental property. 10 LEON A. TAKHTAJAN Theorem 1 (The residue formula). For every ω ∈ Ω1 F/k, Res ω = XP ∈X ResP (ωP ) = 0. 2.3. Serre's duality and the Riemann-Roch theorem. We put Ω1 F/k(D) = Ω1 F/k ∩ ΩX(D) = {ω ∈ Ω1 F/k : vP (ω) ≥ −vP (D) ∀ P ∈ X} and define the residue pairing ( , ) : ΩX ⊗K AX → k by the formula ResP (xP ωP ), where ω ∈ ΩX, x ∈ AX. (ω, x) = XP ∈X The residue pairing has the following properties: P1) (ω, x) = 0 if ω ∈ Ω1 P2) (ω, x) = if ω ∈ ΩX(−D) and x ∈ AX(D). F/k and x ∈ F , It follows from P1), P2) that for every D ∈ Div(X) the formula ı(ω)(x) = (ω, x) determines a k-linear map ı : Ω1 F/k(−D) → (AX/(AX (D) + F ))∨ , where V ∨ = Hom(V, k) is the topological dual of a k-vector space V with linear topology. Theorem 2 (Serre's duality). For every D ∈ Div(X) the map ı is an isomorphism. Hence the finite-dimensional k-vector spaces Ω1 F/k(−D) and AX/(AX(D) + F ) are dual with respect to the residue pairing. Corollary 1 (The strong residue theorem). (i) An ad`ele x ∈ AX corresponds to a rational function on X under the embedding F ֒→ AX if and only if (ω, x) = 0 for all ω ∈ Ω1 F/k. (ii) A differential ad`ele ω ∈ ΩX corresponds to a Kahler differential on F/k ֒→ ΩX if and only if (ω, f ) = 0 for all X under the embedding Ω1 f ∈ F . Proof. Suppose that the condition (i) holds. It follows from Serre's duality that x ∈ AX(D) + F for every D ∈ Div(X) and, since F ∩ AX(D) = 0 for D < 0, we have x ∈ F . To prove (ii), take ω0 ∈ Ω1 F/k, ω0 6= 0. Putting x = ω/ω0 ∈ AX, we have 0 = (ω, f ) = (f ω0, x) for all f ∈ F , whence x ∈ F by part (i). (cid:3) Remark 2. The strong residue theorem is stated in a slightly different form in [3, Ch. III, §5.3]. For ω ∈ Ω1 F/k we put (ω) = XP ∈X vP (ω) · P ∈ Div(X). QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 11 Since dimF Ω1 F/k = 1, all divisors (ω) are linear equivalent and determine a divisor class K ∈ Cl(X), the canonical class of X. The following result is ob- tained by combining the Riemann-Roch formula for the Euler characteristic of a divisor D: χ(D) = h0(D) − h1(D) = deg D + 1 − g, with Serre duality and the adelic interpretation of cohomology. Theorem 3 (Riemann-Roch theorem). For every D ∈ Div(X) we have h0(D) − h0(K − D) = deg D + 1 − g. An effective divisor D on X is called non-special if h0(K − D) = 0. It follows from the Riemann-Roch theorem that an effective divisor D of degree g is non-special if and only if h0(D) = 1. In other words, the only rational functions whose poles are contained in an effective non-special divisor of degree g are constant functions. 3. Differential and Integral Calculus From now on we assume that the algebraically closed field k has charac- teristic 0 and the algebraic curve X has genus g ≥ 1. 3.1. Differentials of the second kind and 'additive functions'. Fol- lowing the classical terminology, we call a Kahler differential ω ∈ Ω1 F/k a differential of the second kind if ResP ω = 0 for all P ∈ X. The k-vector space Ω(2nd) of differentials of the second kind on X carries a canonical skew- symmetric bilinear form ( , )X defined as follows. For every ω ∈ Ω(2nd) let x = {xP }P ∈X ∈ AX be an ad`ele satisfying the equality d xP = ωP ∀ P ∈ X. For every P ∈ X there is a unique (up to an additive constant in k) element xP ∈ FP with this property, and we have xP ∈ OP for all but finitely many P ∈ X. We define x = d−1ω and put The bilinear form ( , )X is independent of the choice of the additive con- stants in the definition of d−1 and is skew-symmetric. When X is a Riemann surface, the bilinear form ( , )X corresponds to the standard pairing in the cohomology under the isomorphism Ω(2nd)/dF ≃ H 1 dR(X) (see [6, Ch. III, §5]). The infinite-dimensional k-vector space Ω(2nd) has a g-dimensional sub- space Ω(1st) = Ω1 F/k(0) of the differentials of the first kind. The infinite- dimensional subspace Ω(1st) ⊕ dF of Ω(2nd) is isotropic with respect to the bilinear form ( , )X. Since there is no canonical choice of the isotropic com- plementary subspace to Ω(1st) ⊕ dF in Ω(2nd), the exact sequence 0 → Ω(1st) ⊕ dF → Ω(2nd) → Ω(2nd)/(Ω(1st) ⊕ dF ) → 0 (ω1, ω2)X = XP ∈X ResP (d−1ω1 ω2), ω1, ω2 ∈ Ω(2nd). 12 LEON A. TAKHTAJAN does not split canonically. Nevertheless we have the following fundamental result (see [2, Ch. VI, §8] and [3, Ch. III, §§5.3-5.4]), which may be regarded as an algebraic de Rham theorem. Theorem 4. (i) The restriction of the bilinear form ( , )X to Ω(2nd)/dF is non- degenerate and dimk Ω(2nd)/dF = 2g. (ii) For every effective non-special effective divisor D on X of degree g there is an isomorphism Ω(2nd)/dF ≃ Ω(2nd) ∩ Ω1 F/k(2D). (iii) Let D = P1 + · · · + Pg be a non-special divisor with distinct points. Then every choice of local uniformizers ti at Pi determines a sym- plectic basis {θi, ωi}g F/k(2D) with respect to the symplectic from ( , )X : i=1 of the k-vector space Ω(2nd) ∩ Ω1 (θi, θj)X = (ωi, ωj)X = 0, (θi, ωj)X = δij, i, j = 1, . . . , g. This basis consists of differentials θi of the first kind and differentials ωi of the second kind which are uniquely determined by the conditions vPi (θj − δijdti) > 0 and vPi(cid:0)ωj − δijt−2 i dti(cid:1) > 0, where i, j = 1, . . . , g. (iv) The subspace k · ω1 ⊕ · · · ⊕ k · ωg is an isotropic complement to Ω(1st) ⊕ dF in Ω(2nd). Proof. Let (ω)∞ = n1Q1 + · · · + nlQl be the polar divisor of ω ∈ Ω(2nd). Since char k = 0, for every Qi there is an fi ∈ F such that vQi(ω − dfi) ≥ 0. We define x = {xP }P ∈X ∈ AX by the formulae xP =( fiQi , P = Qi, P 6= Qi, 0, i = 1, . . . , l, i = 1, . . . , l. Since D is a non-special divisor of degree g, we have Ω1 F/k(D) = {0} and, by Serre duality, AX(D) + F = AX. Thus there in an f ∈ F with the property vP (f − x) ≥ −vP (D) for all P ∈ X, whence (ω − df ) ≥ −2D. Since D is non-special such an f is unique. This proves part (ii). To prove (i), we observe that by the Riemann-Roch theorem, dimk Ω1 F/k(2D) = 3g − 1, dimk Ω1 F/k(D) = 2g − 1. Let Ω(3rd) be the k-vector space of differentials of the third kind. This sub- space of Ω1 F/k is formed by the differentials with only simple poles. Since QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 13 Ω(2nd) ∩ Ω(3rd) = Ω(1st) and Ω(3rd) ∩ Ω1 F/k(D), we conclude that dimk Ω(2nd) ∩ Ω1 F/k(2D) = Ω1 F/k(2D) + dimk Ω1 F/k(D) = dimk Ω1 F/k(2D) + dimk Ω(1st). Using (ii), we have dimk Ω(2nd)/dF = (3g − 1) − (2g − 1) + g = 2g. To complete the proof, we define a k-linear map F/k(2D) → k2g L : Ω(2nd) ∩ Ω1 by the formula L(ω) = (α1(ω), . . . , αg(ω), β1(ω), . . . , βg(ω)), where vPi(cid:0)ω − (αi(ω)t−2 i + βi(ω)dti)(cid:1) > 0, i = 1, . . . , g. Since D is non-special, L is an injective map and hence an isomorphism. The differentials ωi and θi are obtained by choosing the only non-zero component of L to be αi = 1 and βi = 1 respectively. (cid:3) Remark 3. The choice of a non-special effective divisor D = P1 + · · · + Pg on X with distinct points Pi and uniformizers ti may be regarded as an algebraic analogue of the choice of a-cycles on a compact Riemann surface of genus g ≥ 1. Correspondingly, the differentials θi are analogues of differentials of the first kind with normalized a-periods, and differentials ωi are analogues of differentials of the second kind with second order poles, zero a-periods and normalized b-periods. The symplectic property of the basis {θi, ωi}g i=1 is an analogue of the reciprocity law for differentials of the first kind and the second kind (see [7, Ch. 5, §1] and [11, Ch. VI, §3]). Remark 4. It is not necessary to require all the points of the non-special effective divisor D of degree g to be distinct. Theorem 4 and all other results in this paper can be easily modified to include divisors with multiple points. A differential ω of the second kind is said to have zero a-periods if (ω, ωi)X = 0, i = 1, . . . , g. It follows from Theorem 4 that differential of the first kind with zero a- periods is zero. The vector space Ω(2nd) of differentials of the second kind with zero a-periods has the following properties. 0 Proposition 1. (i) The k-vector space Ω(2nd) 0 Ω(2nd) and is an isotropic complement of Ω(1st) in Ω(2nd) 0 = k · ω1 ⊕ · · · ⊕ k · ωg ⊕ dF. (ii) For every P ∈ X the k-vector space Ω0(∗ P ) of differentials of the second kind with zero a-periods and the only pole at P has a natural filtration {0} = Ω0(P ) ⊂ Ω0(2P ) · · · ⊂ Ω0(nP ) ⊂ . . . , 14 LEON A. TAKHTAJAN dimk Ω0(nP ) = n − 1. (iii) There is a direct sum decomposition Ω(2nd) 0 Ω0(∗ P ). = MP ∈X gXi=1 ω = df + ciωi, (iv) Every differential ω ∈ Ω0(nP ) can be written uniquely as where f ∈ H 0(X, OX (D + (n − 1)P )). Proof. Part (i) follows from Theorem 4 because D is non-special. Since dimk Ω1 F/k(nP ) = n − 1 + g, part (ii) follows from the decomposition Ω1 F/k(nP ) = Ω0(nP ) ⊕ Ω(1st). Part (iii) follows from part (ii) because every differential ω ∈ Ω(2nd) can be uniquely written as the sum of its principal parts at the poles. Since the divisor D = P1 + · · · + Pg is non-special, we have h0(D + (n − 1)P ) = n, and part (iv) also follows from Theorem 4. (cid:3) 0 Definition 1. A space of additive multi-valued functions on X (additive functions for brevity) is a subspace A(X) ⊂ AX with the following proper- ties. AF1) F ⊆ A(X). AF2) If a ∈ A(X), then da = ω ∈ Ω1 AF3) If a ∈ A(X) and da = 0, then a = c ∈ k. F/k (and hence ω ∈ Ω(2nd)). Remark 5. For every differential ω ∈ Ω(2nd), the corresponding ad`ele a = {aP }P ∈X = d−1ω is determined uniquely up to the choice of additive con- stants for every P ∈ X. Condition AF3) guarantees that for all f ∈ F these constants are compatible with the equation f = d−1(df ) + c. Example 1. Given any non-special effective divisor D = P1 + · · · + Pg of degree g on X with distinct points Pi, and any choice of the local uniformiz- ers ti at Pi, we have the following space A(X; D) of additive functions with zero a-periods. Let ηi ∈ AX be solutions of the equations dηi = ωi, i = 1, . . . , g, with any fixed choice of the additive constants at all points P ∈ X. Since the divisor D is non-special, the subspaces k · η1 ⊕ · · · ⊕ k · ηg and F of the k-vector space AX have zero intersection. Their direct sum A(X; D) = k · η1 ⊕ · · · ⊕ k · ηg ⊕ F ⊂ AX (3.1) possesses properties AF1)-AF3) and the map d : A(X; D) → Ω(2nd) jective. Indeed, by Proposition 1 every differential ω ∈ Ω(2nd) is sur- can be written 0 0 QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 15 uniquely in the form whence a = d−1ω = f + ciηi + c ∈ A(X; D). Remark 6. The additive functions a = d−1ω ∈ A(X, D) are algebraic ana- logues of abelian integrals of the second kind with zero a-periods on a com- pact Riemann surface of genus g (see, e.g., [7, Ch. V, §2]). We can define ω = df + ciωi, gXi=1 gXi=1 Z Q P ω = a(Q) − a(P ), (3.2) (3.3) (3.4) where a(P ) = aP mod p ∈ k for every P ∈ X. It is quite remarkable that using the additive functions in Example 1, one can naturally define the uniformizers tP at all points P ∈ X. They are uniquely determined by the following data: a choice of a non-special divisor D = P1 + · · · + Pg with distinct points, uniformizers ti at Pi and additive functions η1, . . . , ηg. For every P ∈ X let ω(2) P ∈ Ω0(2P ) be the unique differential of the second kind with the only second order pole at P and zero a-periods such that gXi=1(cid:0)θi, ω(2) P (cid:1)X = 1. Pi = ωi for i = 1, . . . , g. Let ηP = d−1ω(2) In particular, ω(2) P ∈ A(X; D) be an additive function with the only simple pole at P ∈ X. By (3.3), ηP is uniquely determined up to an overall additive constant. We fix this constant by requiring that the sum of constant terms of ηP Pi ∈ k((ti)) over all i = 1, . . . , g be equal to zero. In particular, ηPi = ηi + ci for some ci ∈ k. For every P ∈ X we now define the uniformizer tP by the formula tP = − , 1 ηP(cid:12)(cid:12)(cid:12)(cid:12)P = t−2 P dtP , P ∈ X. and for ω(2) P = dηP we have ω(2) P (cid:12)(cid:12)(cid:12)P P Extending this construction, we now endow the subspace Ω0(∗ P ) for every P ∈ X with a basis {ω(n+1) }∞ n=1 consisting of differentials of the second kind with the only pole at P of order n + 1 and zero a-periods, where the differentials ω(2) ∈ A(X; D) be an additive function with the only pole at P ∈ X of order n and with the following choice of the overall additive constant in (3.3). We put η(1) ∈ k((tP )) to be equal P are already specified by (3.4). Let η(n) P = ηP and require the constant term of η(n) P = d−1ω(n+1) P P (cid:12)(cid:12)(cid:12)P 16 LEON A. TAKHTAJAN to zero for all η(n) of η(n) P , n ∈ N. We have a decomposition P with n > 1. For every P ∈ X let AP (X, D) be the k-span A(X, D) = MP ∈X AP (X, D)! ⊕ k. (3.5) = d A(X; D) One can restate he property of isotropy of the subspace Ω(2nd) and the condition AF3) in the following way. 0 Lemma 3.1. (i) For all P, Q ∈ X and m, n ∈ N we have Q ) = ResQ(η(n) ResP (η(m) P d η(n) Q d η(m) P ). (ii) Every rational function f ∈ F admits a unique 'partial fraction ex- pansion' f = lXi=1 niXj=1 cijη(j) Qi + c, where n1Q1 + . . . nlQl = (f )∞ is the polar divisor of f and c, cij ∈ k. P , ω(n+1) 0 = (ω(m+1) Proof. Since ResQ(da) = 0 for all a ∈ AX, we get, for P 6= Q, Q ) + ResQ(η(m) Q ) − ResQ(η(n) P dη(n) P dη(n) P dη(n) )X = ResP (η(m) )X = ResP (η(m) = ResP (η(m) For P = Q we have 0 = (ω(m+1) N. Part (ii) follows directly from AF3) since there are cij ∈ k such that P dη(n) Q ) Q dη(m) P ). , ω(n+1) Q P P P ) for all m, n ∈ df − lXi=1 niXj=1 cijω(j+1) Qi ∈ Ω(2nd) 0 ∩ Ω(1st) = {0}. (cid:3) Remark 7. Part (i) of Lemma 3.1 is an algebraic analogue of the classical reciprocity law for differentials of the second kind with zero a-periods on a compact Riemann surface (see, e.g., [7, Ch. V, §1] and [11, Ch. VI, §3]). Remark 8. In the genus zero case X = P1 k = k ∪ {∞} we have F = k(z) and ω(n+1) P = dz (z − P )n+1 for P ∈ k, ω(n+1) P = −zn−1dz for P = ∞. Correspondingly, η(n) P (z) = − 1 n(z − P )n for P ∈ k, η(n) P (z) = − zn n for P = ∞. Remark 9. Put OX = AX(0) =QP ∈X OX . By Lemma 3.1 we have AX = A(X, Dns) + OX, while Serre's adelic interpretation of cohomology yields AX/(F + OX) = H 1(X, OX ). QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 17 Remark 10. The condition that the constant field k is algebraically closed is not necessary: all results in this section remain valid for any field of constants of characteristic 0 if we replace the field k by the residue class field k(P ) = Op/p and use the trace map Trk(P )/k : k(P ) → k. For example, for the bilinear form ( , )X we have (ω1, ω2)X = XP ∈X Trk(P )/k ResP (d−1ω1 ω2). Remark 11. The multiplicative analogue of a k-vector space A(X) of addi- tive multi-valued functions is the group M(X) of multiplicative multi-valued functions on X. This subgroup of the group of invertible elements of the ad`ele ring AX is defined by the following properties. It contains F ∗ as a subgroup, we have d log m = m−1dm = ω ∈ Ω1 F/k for all m ∈ M(X) and if m ∈ M(X) satisfies d log m = 0, then m = c ∈ k∗. It also seems natural to assume (as was done in a preliminary version of this paper) that the following multi- plicative analogue of Lemma 3.1 holds. Every rational function f ∈ F ∗ can be written uniquely as a product of multiplicative multi-valued functions with one zero and one pole obeying the natural generalized Weil reciprocity law on X (see [16],[13]). However, the referee pointed out that this assertion contradicts the non-triviality of Poincar´e bi-extension over the square of the Jacobian of X [5]. 4. Local Theory Let K be a complete closed field, that is, a complete discrete valuation field with valuation ring OK , maximal ideal p and algebraically closed residue field k = OK /p. Every local uniformizer t determines an isomorphism K ≃ k((t)). Therefore K may be interpreted as a 'geometric loop algebra' over k. The main example of a complete closed field is K = FP , where P is a point on an algebraic curve X over k. Here we describe some infinite-dimensional Lie algebras naturally associ- ated with K and construct their irreducible highest-weight modules. When K = FP , these objects determine local quantum field theories at P ∈ X. Specifically, we consider the following local quantum field theories (QFT): 1. the 'QFT of additive bosons', which corresponds to the Heisenberg Lie algebra g (a one-dimensional central extension of the geometric loop algebra gl1(K) = K), 2. the 'QFT of lattice bosons', which corresponds to the lattice Lie algebra l associated with the Heisenberg Lie algebra g and the lattice Z. 4.1. The Heisenberg algebra. Let Ω1 K/k be the K-module of Kahler dif- ferentials. We put Ω1 K/k/Q, where Q = ∩n≥0 pnd(O) (see Section 2.2). The abelian Lie algebra gl1(K) = K over the field k is endowed with K/k = Ω1 18 LEON A. TAKHTAJAN a natural bilinear skew-symmetric form c : ∧2K → k by the formula c(f, g) = − Res(f dg), f, g ∈ K, where dg ∈ Ω1 K/k. The bilinear form c is continuous with respect to the p-adic topology on K and the discrete topology on k. Hence c ∈ H 2 c (K, k) ≃ Homc(∧2K, k), where Homc(∧2K, k) is the group of continuous 2-cocycles on K with values in k. Definition 2. The Heisenberg Lie algebra g is the one-dimensional central extension of K 0 → k · C → g → K → 0 with the 2-cocycle c. Writing [ , ] for the Lie bracket in g = K ⊕ k · C, we have [f + a C, g + b C] = c(f, g) C, f, g ∈ K, a, b ∈ k. The Lie subalgebra g+ = OK ⊕ k · C is a maximal abelian subalgebra of g. Remark 12. Let Aut O = {u ∈ O : v(u) = 1} be the group of continuous automorphisms of the valuation ring O = k[[t]] (see [4]). One can easily show that every continuous linear map l : k((t)) ⊗k k((t)) → k which satisfies l(f ◦ u, g ◦ u) = l(f, g) for all f, g ∈ k((t)) and u ∈ Aut O is a constant multiple of c. This explains the natural role of the 2-cocycle c of K. In particular, every Aut O-invariant bilinear form l is necessarily skew-symmetric. This may be regarded as a simple algebraic analogue of the spin-statistics theorem. Definition 3. A module of the Heisenberg algebra g is a k-vector V with the discrete topology and with a k-algebra homomorphism ρ : g → End V such that the g-action on V is continuous and ρ(C) = I is the identity endomorphism of V . Equivalently, for every v ∈ V there is an open subspace U of K which is commensurable with p and annihilates v: ρ(U ) v = 0. Putting f = ρ(f ) ∈ End V for all f ∈ K, we have [f , g] = c(f, g)I and thus obtain a projective representation of the abelian Lie algebra K. Remark 13. Any choice of the uniformizer t for K determines an isomor- phism K ≃ k((t)) and a basis basis {tn}n∈Z in K. Putting αn = ρ(tn) and using the formulae c(tm, tn) = mδm,−n, we get the commutation relations of the 'oscillator algebra' They characterize free bosons in the two-dimensional QFT. [αm, αn] = mδm,−nI. QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 19 Definition 4. An irreducible highest-weight module of the Heisenberg alge- bra g is an irreducible g-module with a vector 1 ∈ V which is annihilated by the abelian subalgebra OK ⊕ {0}. The following result is well-known (see, for example, [8, Lemma 9.13]). Theorem 5. Each irreducible highest-weight module of the Heisenberg Lie algebra g is either the trivial one-dimensional module k = k · 1 with the highest vector 1 = 1 ∈ k, or the Fock module induced from the one-dimensional g+-module k. F = indg g+ k Remark 14. Let U g be the universal enveloping algebra of the Lie algebra g. By definition, F = U g ⊗ U g+ k, where U g is regarded as a right U g+-module. Equivalently, F = W /D, (4.1) where W is the Weyl algebra of g, that is, the quotient of U g by the ideal generated by C − 1 (with now 1 standing for the identity in U g) and D is the left ideal in W generated by OK ⊕ {0}. Explicit realization of the Fock module F (the bosonic Fock space) de- pends on a decomposition of K into a direct sum of subspaces isotropic with respect to the bilinear form c: K = K+ ⊕ K−, where the subspace K+ = OK is defined canonically. In this case, F ≃ Sym• K− (4.2) (4.3) is the symmetric algebra of the k-vector space K−. The Fock space F is a Z-graded commutative algebra F = F (n) ∞Mn=0 where F (n) ≃ Symn K−, F (0) = k · 1, and F (n) = {0} for n < 0. For every f = f+ + f− ∈ K the operator f = ρ(f ) ∈ End F is defined by the formula f · v = f− ⊙ v + c(f, vi) vi = f− ⊙ v − Res (f+ dvi) vi, (4.4) where v = v1 ⊙ · · · ⊙ vk ∈ F (k) and vi = v1 ⊙ · · · ⊙ vi ⊙ · · · ⊙ vk ∈ F (k−1), i = 1, . . . , k. Here ⊙ stands for the multiplication in Sym• K−. In particular, kXi=1 kXi=1 f · 1 = f−. 20 LEON A. TAKHTAJAN The Fock module F is endowed with the linear topology given by the fil- tration associated with the Z-grading and independent of the decomposition (4.2). Remark 15. Any choice of the uniformizer t determines an isomorphism K ≃ k((t)), and one can take K− = t−1k[t−1]. The map F (n) ∋ v = t−m1 ⊙ · · · ⊙ t−mn 7→ xm1 . . . xmn ∈ k[x1, x2, . . . ] determines an isomorphism F ≃ k[x1, x2, . . . ] between the bosonic Fock space and the polynomial ring in infinitely many variables {xn}n∈N. Under this map we have αn 7→ n∂/∂xn, α−n 7→ xn, n > 0 (the operator of multiplication by xn), and α0 7→ 0. Remark 16. For an arbitrary complete closed field K there is no canonical choice of the isotropic subspace K− complementary to K+ = OK . However, any choice of an effective non-special divisor D = P1 + · · · + Pg of degree g on an algebraic curve X and uniformizers ti at Pi, determines such isotropic subspaces K− for all fields K = FP , P ∈ X. Namely, let A(X, D) be the k-vector space of additive functions defined in Example 1, and let AP (X, D) be the subspace of additive functions with the only pole at P . We put K− = AP (X, D)P ⊂ K. By part part (i) of Lemma 3.1, the subspace K− is isotropic with respect to c and we have the decomposition (4.2). The subspace K− is spanned by the elements v(n) , n ∈ N, and dK− = Ω0(∗P )P . P = η(n) The bilinear form c has the one-dimensional kernel k. Since OK /k = p, the form c determines a continuous non-degenerate pairing c : p ⊗ K− → k, whence p = K ∨ − = Hom(K−, k) is the topological dual to the k-vector space K−. The topological dual of the bosonic Fock space F is accordingly equal to the k-vector space F ∨ = Sym• p which is the completion of Sym• p with respect to the linear topology given by the filtration {F n Sym•, p}∞ n=0, P (cid:12)(cid:12)(cid:12)P F n Sym• p = ⊕n i=0 Symi p. The continuous pairing ( , ) : F ∨ ⊗ F → k is uniquely determined by the pairing between Sym• p and F = Sym• K− and is defined recursively by the formula (u, v) = δkl c(u1, vi)(u1, vi), (4.5) where u = u1 ⊙ · · · ⊙ uk = u1 ⊙ u1 ∈ Symk p and v = v1 ⊙ · · · ⊙ vl = vi ⊙ vi ∈ F (l). The dual bosonic Fock space F ∨ is a right g-module with lowest-weight vector 1∨ annihilated by the subspace K− ⊕ k. The representation ρ of the Heisenberg algebra g on F determines a contragradient representation ρ∨ of g the Heisenberg algebra on F ∨ by the formula (u · ρ∨(f ), v) = (u, ρ(f ) · v), ∀ u ∈ F ∨, v ∈ F . lXi=1 QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 21 More explicitly, put f = f+ + f− ∈ K, where now f+ ∈ p and f− ∈ K− ⊕ k. Then it follows from (4.4) and (4.5) that the operator f = ρ∨(f ) ∈ End F ∨ is given by u · f = f+ ⊙ u + kXi=1 c(ui, f )ui = f+ ⊙ u + kXi=1 Res ( f−dui)ui, (4.6) where u = u1 ⊙· · ·⊙uk ∈ Symk p and ui = u1 ⊙· · ·⊙ ui ⊙· · ·⊙uk ∈ Symk−1 p. 4.2. The lattice algebra. Let k[Z] be the group algebra of the additive group Z. As a k-vector space, k[Z] has a basis {en}n∈Z, emen = em+n. For every decomposition (4.2) we define the 'constant term' of any f ∈ K as f (0) = f+ mod p ∈ k. Hence we have f (0) = 0 for f ∈ K−. Remark 17. If K = FP and K− = AP (X, D)P , then f (0) is the constant term of the formal Laurent expansion of f ∈ k((tP )) with respect to the uniformizer tP for K, defined in Section 3.1. Definition 5. The lattice algebra l associated with the decomposition (4.2) is a semidirect sum of the Heisenberg Lie algebra g and the abelian Lie algebra k[Z] with the Lie bracket [f + aC + αem, g + bC + βen] = c(f, g)C + mαg(0)em − nβf (0)en, where f + aC, g + bC ∈ g. The corresponding irreducible highest-weight module B for the lattice algebra l is given by B = k[Z] ⊗ F , where k[Z] acts by multiplication and K acts by the formula f (en ⊗ v) = −nf (0)en ⊗ v + en ⊗ (f · v), f ∈ K, v ∈ F . The module B (the Fock space of 'charged bosons') is a Z-graded com- mutative algebra, B =Mn∈Z B(n), B(n) = k · en ⊗ F . The elements en, n ∈ Z, correspond to the shift operators en = en in B, where e(en ⊗ v) = en+1 ⊗ v, v ∈ F . Remark 18. Using the canonical isomorphism K ∗/O∗ K ≃ Z induced by the valuation v : K ∗ → Z, one can also define the Fock space B as the space of all functions with finite support. F : K ∗/O∗ K → F 22 LEON A. TAKHTAJAN Remark 19. For every choice of the uniformizer t for K, the map B(n) ∋ en ⊗ (t−m1 ⊙ · · · ⊙ t−ml) 7→ enx0xm1 . . . xml ∈ enx0k[x1, x2, . . . ] establishes the isomorphism B ≃ k[ex0 , e−x0, x1, x2, . . . ]. Under this map we have αn 7→ n∂/∂xn, α−n 7→ xn, n > 0, α0 7→ −∂/∂x0, and e 7→ ex0 (the operator of multiplication by ex0). The topological dual of B is the k-vector space B∨ =Mn∈Z k·qn ⊗ F ∨, where {qn}n∈Z is the basis in k[Z]∨ dual to the basis {en}n∈Z. The continuous pairing ( , ) : B∨ ⊗ B → k is given by (qm ⊗ u, en ⊗ v) = (u, v)δmn, u ∈ F ∨, v ∈ F . As for the Heisenberg algebra, the representation ρ of the lattice Lie alge- bra l on B determines a contragradient representation ρ∨ on B∨. The dual Fock space B∨ is a right l-module with lowest-weight vector 1∨ annihilated by K−. 5. Global Theory Given an algebraic curve X over an algebraically closed field k of charac- teristic 0, we shall define the global versions of the local QFT's introduced in the previous section. One can briefly characterize these global QFT's as follows. 1. The 'QFT of additive bosons on X' corresponds to the global Heisen- berg algebra gX (the restricted direct sum of local Heisenberg alge- bras gP over all points P ∈ X). The global Fock space FX is defined as the restricted tensor product of the local Fock spaces FP over all points P ∈ X. The global Fock space FX is a highest-weight gX- module. There is a linear functional h · i : FX → k (the expectation value functional) which is uniquely determined by the properties of normalization and invariance with respect to the space of additive functions. 2. The 'QFT of charged bosons on X' corresponds to the global lattice algebra lX . The global charged Fock space BX is a highest-weight lX -module and there is a unique expectation value functional h · i : BX → k with similar properties. 5.1. Additive bosons on X. The QFT of additive bosons consists of the following data. AB1) An effective non-special divisor Dns = P1 + · · · + Pg of degree g on X with distinct points, uniformizers ti at Pi and the k-vector space of additive functions A(X, Dns) (a subspace of AX containing F = k(X)) introduced in Example 1. QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 23 AB2) The local QFT's of additive bosons (the highest-weight gP -modules FP for all points P ∈ X). AB3) The global Heisenberg algebra gX (the one-dimensional central ex- tension of the abelian Lie algebra gl1(AX) = AX by the cocycle cX =PP ∈X cP ). AB4) A highest-weight gX-module FX (the global Fock space, which is the restricted tensor product of FP over all points P ∈ X). AB5) An expectation value functional, that is, a linear map h · i : FX → k with the following properties: (i) h1Xi = 1, where 1X ∈ FX is the highest-weight vector, (ii) ha · vi = 0 for all a ∈ A(X, Dns) and v ∈ FX . The data AB1) and AB2) have already been described in Sections 3.1 and 4.1. Here we introduce the global Heisenberg algebra gX, construct the corresponding global Fock space FX and prove that there is a unique expectation value functional h · i with properties (i) and (ii). Let cX : AX × AX → k be the global bilinear form cX (x, y) = XP ∈X cP (xP , yP ) = −XP ∈X ResP (xP dyP ), x, y ∈ AX. Definition 6. The global Heisenberg Lie algebra gX is the one-dimensional central extension of the abelian Lie algebra AX by the two-cocycle cX . 0 → k C → gX → AX → 0 The Lie subalgebra g+ X = OX ⊕ kC is the maximal abelian subalgebra of gX. Definition 7. The global Fock space FX is an irreducible gX -module with vector 1X annihilated by the abelian subalgebra OX ⊕ {0}. As in the local case, the global Fock module is induced from the one- dimensional g+ X -- module: FX = indgX g+ X k. By what was said in the previous section, we have a decomposition (4.2) for K = FP , P ∈ X, where F (+) P = AP (X, D)P . This yields the following decomposition of the k-vector space AX into a direct sum of subspaces isotropic with respect to cX : P = OP and F (−) AX = OX ⊕ F (−) X . (5.1) Here F (−) X = aP ∈X F (−) P FX ≃ Sym• F (−) X . is the restricted direct product over all P ∈ X with respect to the zero subspaces {0} ⊂ F (−) P . The decomposition (5.1) gives rise to an isomorphism 24 LEON A. TAKHTAJAN The global Fock space FX carries a linear topology given by the natural filtration associated with the Z-grading. Equivalently, FX may be defined as the tensor product FP , FX = b⊗ P ∈X which is restricted with respect to the vectors 1P ∈ FP and is endowed with the product topology. In other words, 1X = ⊗P ∈X 1P , and FX is spanned by the vectors v = ⊗ P ∈X vP , where vP = 1P for all but finitely many P ∈ X. For every P ∈ X we have v = vP ⊗ vP , where vP = ⊗Q∈X vQ, vQ = vQ for Q 6= P and vP = 1P . We denote the corresponding representation of gP on FP , P ∈ X, by ρP , and the representation of gX on FX by ρ. Putting x = ρ(x) ∈ End FX for x = {xP }P ∈X ∈ AX and taking any v = ⊗P ∈XvP , we have x · v = XP ∈X xP · vP ⊗ vP , where xP = ρP (xP ) ∈ End FP . Put PX = YP ∈X p. The topological dual of the global Fock space FX is the k-vector space X = Sym• PX , which is the completion of Sym• PX with respect to the F ∨ linear topology given by the natural filtration associated with the Z-grading. The dual global Fock space F ∨ X is a right gX-module with lowest-weight vector 1∨ X ⊕ {0}. Equivalently, X annihilated by the abelian subalgebra F (−) is the completion of the tensor product restricted with respect to the vectors P . The completion is taken with respect to the double filtration {F mn Sym• PX }, 1∨ F mn Sym• PX = In other words, the elements of F ∨ X are infinite sums mXi=0 XP1,...,Pi∈X F ∨ P F ∨ P ∈X X = b⊗  nMl1+···+li=0 ∞Xn=0 XP1,...,Pn∈X u = aP1 ...PnuP1 ...Pn, Syml1 p1 ⊗ · · · ⊗ Symli pi . where the uP1 ...Pn ∈ F ∨ uct P1...Pn belong to the completion of the tensor prod- F ∨ P1...Pn = F ∨ P1 ⊗ · · · ⊗ F ∨ Pn QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 25 with respect to the filtration F mF ∨ P1...Pn = mMl1+···+ln=0(cid:16)Syml1 p1 ⊗ · · · ⊗ Symln pn(cid:17) . P (cid:12)(cid:12)(cid:12)Pon∈N P }n∈N be the basis of p dual to the basisnv(n) of F (−) Let {u(n) with respect to the pairing given by cP (see Section 4.1). Then we see that F ∨ P ]] of formal Taylor series in infinitely many variables u(n) X is used to prove the following main result in the QFT of additive bosons. P , P ∈ X, n ∈ N. This realization of F ∨ X is the completion of the space k[[un P = η(n) P Theorem 6. There is a unique linear functional h · i : FX → k (the expec- tation value functional) with the following properties: EV1) h1X i = 1, EV2) ha · vi = 0 for all a ∈ A(X, Dns) and v ∈ FX . The functional h · i is given by hvi = (ΩX, v) , where ΩX = exp − 1 2 ∞Xm,n=1 XP,Q∈X P Q = − ResQ(η(m) c(mn) P dη(n) Q ). Q  P Q u(m) c(mn) P u(n) ∈ F ∨ X , Proof. It follows from decomposition (3.5) that a linear functional of the form hvi = (Ω, v) possesses properties EV1) and EV2) if and only if it is normalized, (Ω, 1X ) = 1, and Ω ∈ F ∨ X satisfies the system of equations (n) P = 0 Ω · η for all P ∈ X and n ∈ N, where η P Q}Q∈X and γ(n) where β(n) P = {β(n) P = ρ∨(η(n) (n) P ). Write η(n) P = β(n) P Q}Q∈X ∈ AX are given by P = {γ(n) β(n) P Q = 0 η(n) P (cid:12)(cid:12)(cid:12)Q if Q = P , if Q 6= P , γ(n) P (cid:12)(cid:12)(cid:12)P P Q =( η(n) 0 if Q = P , if Q 6= P . (n) P acts on F ∨ X as differentiation with respect to (5.2) P + γ(n) P , It follows from (4.6) that γ u(n) P . For Q 6= P we have β(n) P Q = a(n) P Q + a(nm) P Q u(m) Q , ∞Xm=1 where a(n) P Q ∈ k and P Q = c(β(n) a(nm) P Q, v(m) Q ) = − ResQ(η(n) P dη(m) Q ) = c(nm) P Q . 26 LEON A. TAKHTAJAN Since c(nm) P P = 0 (see Lemma 3.1), we conclude that β X as a Q . One can rewrite the equations (5.2) in acts on F ∨ P Q u(m) (n) P the form multiplication by PQ∈X c(nm) + XQ∈X ∂u(n) P  ∂ It follows from part (i) of Lemma 3.1 that P Q u(m) c(nm) Q  Ω = 0, P ∈ X, n ∈ N. (5.3) P Q = c(nm) c(mn) QP , whence the system of differential equations (5.3) is compatible and ΩX is its unique normalized solution. (cid:3) Remark 20. Let g be a semi-simple Lie algebra over k with the Cartan- Killing form h , i. Then the k-vector space AX with bilinear form cX may be replaced by the k-vector space VX = g ⊗k AX with bilinear form −PP ∈X ResP hxP , dyP i. Theorem 6 extends to this case. The additive Ward identities hold for g ⊗k A(X, Dns) and the corresponding QFT is associated with the current algebra on X in the sense of [18]. 5.2. Charged additive bosons on X. The QFT of charged additive bosons is determined by the following data. CB1) An effective non-special divisor Dns = P1 + · · · + Pg of degree g on X with distinct points, uniformizers ti at Pi and the k-vector space of additive functions A(X, Dns) (a subspace of AX containing F = k(X)) introduced in Example 1. CB2) The local QFT's of charged additive bosons (the highest-weight lP - modules BP for all points P ∈ X). CB3) The global lattice algebra lX (a semi-direct sum of the global Heisen- berg algebra gX and the abelian Lie algebra k[Div0(X)] with gen- erators eD, D ∈ Div0(X), where k[Div0(X)] is the group algebra of the additive group Div0(X) of degree 0 divisors on X). CB4) A highest-weight lX -module BX (the global Fock space with the highest-weight vector 1X ∈ BX). CB5) An expectation value functional, that is, a linear map h · i : BX → k with the following properties: (i) heD · 1Xi = 1 for all D ∈ Div0(X), (ii) ha · ui = 0 for all a ∈ A(X, Dns) and u ∈ BX. As a k-vector space, the group algebra k[Div0(X)] has a basis {eD}D∈Div0(X), eD1eD2 = eD1+D2. For every x = {xP } ∈ AX and D = PP ∈X nP P ∈ Div0(X), we put x(D) = XP ∈X nP xP (0) ∈ k, QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 27 where xP (0) = x+ P mod p ∈ k is the constant term of xP ∈ FP , (it is determined by decomposition (4.2) associated with the non-special divisor Dns; see Section 4.2). Definition 8. The global lattice algebra lX is the semi-direct sum of the global Heisenberg algebra gX and the abelian Lie algebra k[Div0(X)] with Lie bracket [x + αC + γeD1, y + βC + δeD2 ] = cX (x, y)C + y(D1)γeD1 − x(D2)δeD2 , where x + αC, y + βC ∈ gX and γ, δ ∈ k. The global Fock space BX is the tensor product of the group algebra k[Div0(X)] and the Fock space of additive bosons FX : where BX = k[Div0(X)] ⊗ FX = MD∈Div0(X) BD X, BD X = k · eD ⊗ FX . BX is an irreducible lX -module where k[Div0(X)] acts by multiplication: eD1(eD2 ⊗ v) = eD1+D2 ⊗ v, v ∈ FX , and AX acts by the formula (5.4) (5.5) x(eD ⊗ v) = −x(D)eD ⊗ v + eD ⊗ (x · v), x ∈ AX, v ∈ FX . For every D =PP ∈X nP P ∈ Div0(X) the subspace BD X is an irreducible gX-module with the following property. If x = {xP }P ∈X ∈ AX with xP ∈ k for all P ∈ X, then the restriction of the operator x to BD X is equal to −x(D)I, where I is the identity operator. In particular, when x = c is a constant, we have x(D) = c deg D = 0, and x acts by zero on BX. Remark 21. One can also define an extended global lattice algebra lX as a semidirect sum of the global Heisenberg algebra gX and the abelian Lie algebra k[Div(X)]. The corresponding irreducible lX -module is the extended Fock space BX = k[Div(X)] ⊗ FX = MD∈Div(X) BD X. The action of lX on BX is given by the same formulas (5.4) -- (5.5), where now the constant ad`ele x = c acts on BD X by (c deg D)I. The dual Fock space B∨ dual spaces to BD infinite sums. Explicitly, X is defined as a completion of the direct sum of X over D ∈ Div0(X). This completion is given by formal where B∨ X = MD∈Div0(X) B∨ X(D), B∨ X(D) = k · qD ⊗ F ∨ X , 28 LEON A. TAKHTAJAN qD ∈ k[Div0(X)]∨ are dual to eD, and F ∨ X was defined in Section 5.1. Theorem 7. There is a unique linear functional h · i : BX → k (the expec- tation value functional) with the following properties: EV1) heD · 1Xi = 1 for all D ∈ Div0(X), EV2) ha · vi = 0 for all a ∈ A(X, Dns) and v ∈ BX. The functional h · i is given by hvi = ( ΩX , v), where ΩX = XD∈Div0(X) qD ⊗ exp( ∞Xn=1XP ∈X and ΩX is defined in Theorem 6. P (D)u(n) η(n) P ) ΩX ∈ B∨ X, Proof. We put Ω = XD∈Div0(X) qD ⊗ ΩD, ΩD ∈ F ∨ X . The condition (Ω, eD ⊗ 1X ) = 1 for all D ∈ Div0(X) is equivalent to the normalization (ΩD, 1X ) = 1. Since the constants act by zero on BX, it suffices to verify that (qD ⊗ ΩD) · η (n) P = 0 (5.6) nQ η(n) (0) qD qD · η P = −η(n) (n) (note that η(n) that ΩD satisfies the following system of differential equations (0) = 0 by the definition in Section 4.2), we see from (5.6) P (D) qD = −XQ∈X for all D =PQ∈X nQ Q ∈ Div0(X) and P ∈ X. Since P (cid:12)(cid:12)(cid:12)Q Q  ΩD = 0. Q  P (cid:12)(cid:12)(cid:12)P  ∂ ΩD = exp Remark 22. Theorems 6 and 7 hold for an arbitrary field k of constants of characteristic 0 (see Remark 10). This system has a unique normalized solution given by ∞Xm,n=1 XP,Q∈X (0) + XQ∈X ∞Xn=1XP ∈X P Q u(m) c(mn) P u(n) . (cid:3) P (D)u(n) η(n) P − 1 2 c(nm) P Q u(m) ∂u(n) P − XQ∈X nQ η(n) P (cid:12)(cid:12)(cid:12)Q Remark 23. All results in this section hold trivially in the case when X has genus 0. Using Remark 8, one can easily obtain elementary explicit formulae for the expectation value functional h · i for quantum additive and charged bosons on P1 k. QUANTUM FIELD THEORIES ON ALGEBRAIC CURVES 29 5.3. Invariant formulation. Here we present an invariant formulation and a proof of a generalization of Theorem 6 for the current algebra. They were suggested by the referee. Let V be a k-vector space regarded as abelian Lie algebra over k, and let c be a skew-symmetric bilinear form on V . We write V for the one-dimensional central extension of V 0 → k · C → V → V → 0 with the 2-cocycle c, and W for the Weyl algebra of the Lie algebra V , as in Section 4.1. Let U and W be isotropic subspaces of V with respect to c such that U ∩ W and V /(U + W ) are finite-dimensional and U ∩ W lies in the kernel of c. Lemma 5.1. There is a canonical isomorphism of k-vector spaces W /W ·(U + W ) ≃ Sym• (V /(U + W )) . Proof. This is proved by direct calculation in a symplectic basis of V com- patible with the corresponding bases in U and W . (cid:3) In the notation of Remark 20 we put V = VX = g ⊗k AX, c(x, y) = −XP ∈X ResP hxP , dyP i, and U = g ⊗k F , W = g ⊗k OX, where F = k(X). Then W /W ·W ≃ FX is the Fock space of the current algebra on X. Using Serre's adelic interpre- tation of cohomology in the form V (U + W ) ≃ g ⊗k H 1(X, OX ), we obtain from Lemma 5.1 that FX/(g ⊗k F )·FX ≃ Sym•(g ⊗k H 1(X, OX )). This shows that the global symmetries g ⊗k F do not uniquely determine the expectation value functional h · i except in the case when X = P1 k. To extend the Lie algebra of global symmetries, we consider a Lagrangian dR(X) ≃ Ω(2nd)/dF such that the restriction to L of the subspace L ⊂ H 1 natural map H 1 dR(X) → H 1(X, OX ) is an isomorphism. For example, take L = k · ω1 ⊕ · · · ⊕ k · ωg (see Theorem 4). Let L be the inverse image of L under the map Ω(2nd) → Ω(2nd)/dF . We claim that there is a subspace U0 ⊂ AX such that F ⊂ U0, U0 ∩ OX = k and dU0 = L. For example, take U0 = A(X, Dns). Then U = g ⊗k U0 is an isotropic subspace of V and, by Remark 9, we have Therefore, V /(U + W ) = {0}. FX /U · FX ≃ k, 30 LEON A. TAKHTAJAN which is essentially Theorem 6 (without an explicit formula for the vector ΩX). By Remark 10, the condition that the field k is algebraically closed is not necessary. References [1] E. Arbarello, C. De Concini, and V. G. Kac, The infinite wedge representation and the reciprocity law for algebraic curves, Proc. Sympos. Pure Math., 49:1 (1989), 171 -- 190. [2] C. Chevalley, Introduction to the theory of algebraic functions of one variable, Amer- ican Mathematical Society, Providence, R.I., 1963. [3] M. Eichler, Introduction to the theory of algebraic numbers and functions, Academic Press, New York, 1966. [4] E. Frenkel and D. Ben-Zvi, Vertex algebras and algebraic curves, American Mathe- matical Society, Providence, RI, 2004. [5] S.O. Gorchinskiı, The Poincar´e bi-extension and id`eles on an algebraic curve, Mat. Sb., 197:(1) (2006), 25 -- 38. [6] P. Griffiths and J. Harris, Principles of algebraic geometry, Wiley-Interscience, New York, 1978. [7] K. Iwasawa, Algebraic functions, American Mathematical Society, Providence, RI, 1993. [8] V.G. Kac, Infinite-dimensional Lie algebras, Cambridge University Press, Cambridge, 1990. [9] N. Kawamoto, Y. Namikawa, A. Tsuchiya, and Y. Yamada, Geometric realization of conformal field theory on Riemann surfaces, Comm. Math. Phys., 116:2 (1988), 247 -- 308. [10] D. Kazhdan. Free fermions on an algebraic curve, Talk at Amer. Math. Soc. Summer Institute Theta functions, 1987. [11] I. Kra, Automorphic forms and Kleinian groups, W. A. Benjamin, Inc., Reading, Mass., 1972. [12] S. Lefschetz, Algebraic Topology, American Mathematical Society, New York, 1942. [13] J.-P. Serre, Algebraic groups and class fields, Springer-Verlag, New York, 1988. [14] L.A. Takhtajan, Quantum field theories on an algebraic curve, Lett. Math. Phys., 52:1 (2000), 79 -- 91. [15] J. Tate, Residues of differentials on curves, Ann. Sci. ´Ecole Norm. Sup. (4), 1 (1968), 149 -- 159. [16] A. Weil, Sur les fonctions alg´ebriques `a corps de constantes fini, C. R. Acad. Sci. Paris, 210 (1940), 592 -- 594. [17] E. Witten, Free fermions on an algebraic curve, Proc. Sympos. Pure Math., 48 (1988), 329 -- 344. [18] E. Witten, Quantum field theory, Grassmannians, and algebraic curves, Comm. Math. Phys., 113:4 (1988), 529 -- 600. Department of Mathematics, Stony Brook University, Stony Brook, NY 11794-3651, USA The Euler International Mathematical Institute, St. Petersburg Depart- ment of Steklov Mathematical Institute, Pesochnaya nab. 10, St.Petersburg 197022, Russia E-mail address: [email protected]
1703.00870
1
1703
2017-03-02T17:44:15
Tempered subanalytic topology on algebraic varieties
[ "math.AG", "math.CV" ]
On a smooth algebraic variety over $\mathbb{C}$, we build the tempered subanalytic and Stein tempered subanalytic sites. We construct the sheaf of holomorphic functions tempered at infinity over these sites and study their relations with the sheaf of regular functions, proving in particular that these sheaves are isomorphic on Zariski open subsets. We show that these data allow to define the functors of tempered and Stein tempered analytifications. We study the relations between these two functors and the usual analytification functor. We also obtain algebraization results in the non-proper case and flatness results.
math.AG
math
Tempered subanalytic topology on algebraic varieties Fran¸cois Petit ∗ March 3, 2017 Abstract On a smooth algebraic variety over C, we build the tempered subanalytic and Stein tempered subanalytic sites. We construct the sheaf of holomorphic func- tions tempered at infinity over these sites and study their relations with the sheaf of regular functions, proving in particular that these sheaves are isomorphic on Zariski open subsets. We show that these data allow to define the functors of tem- pered and Stein tempered analytifications. We study the relations between these two functors and the usual analytification functor. We also obtain algebraization results in the non-proper case and flatness results. 7 1 0 2 r a M 2 ] . G A h t a m [ 1 v 0 7 8 0 0 . 3 0 7 1 : v i X r a Contents 1 Sheaves on the subanalytic site 1.1 Subanalytic topology, tempered functions and distributions: a review . 1.2 Tempered functions and distributions: complements . . . . . . . . . . . 1.3 Tempered holomorphic functions . . . . . . . . . . . . . . . . . . . . . 2 Tempered analytification 2.1 T P-spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 The tempered analytification functor . . . . . . . . . . . . . . . . . . . 2.3 The sheaves O t . . . . . . . . . . . . . . . . . . . . . . . . X tan and OXzar 3 Stein tempered analytification 3.1 The subanalytic Stein topology . . . . . . . . . . . . . . . . . . . . . . 3.2 The Stein tempered analytification functor . . . . . . . . . . . . . . . . 3.3 Comparison between the different analytification functors . . . . . . . . 3.4 The flatness of O t . . . . . . . . . . . . . . . . . . . . . . 3.5 Towards a tempered GAGA theorem . . . . . . . . . . . . . . . . . . . X sta over OXzar 3 3 5 8 9 9 10 13 15 15 18 19 20 23 ∗The author has been fully supported in the frame of the OPEN scheme of the Fonds National de la Recherche (FNR) with the project QUANTMOD O13/570706 2010 Mathematics Subject Classification: 32C38, 14A10 1 CONTENTS Introduction The problem of comparing algebraic and complex analytic geometry is a very classical question. This comparaison is carried out by first functorially associating to an alge- braic variety an complex analytic space with the help of the analytification functor. Then, one can try to compare the properties of these two spaces. In the case of proper algebraic varieties, this problem has been settle by Serre in his seminal paper [Ser55] and his famous GAGA theorem. In the non-proper case, this theorem does not hold but some comparison results between analytic and algebraic objects have been obtained under some growth condition on the analytic functions. One usually requires the holo- morphic functions to be tempered which means that they have polynomial growth with respect to the inverse of the distance to the boundary. A very classical instance of such results is Liouville's theorem asserting that a tempered entire function is a polynomial. Several authors have studied tempered holomorphic functions on affine algebraic va- rieties (see for instance [Bjo74, Rud67, RW80]). Our aim, in this paper, is to sheafify the study of tempered holomorphic functions on algebraic variety and to compare the tempered geometry we obtain with algebraic geometry. For that purpose we construct two functors of tempered analytification, study their relations, properties and relate them to the usual analytification functor. To carry out this idea, we rely upon Kashiwara-Schapira's theory of subanalytic sheaves [KS01] which allows one to define sheaves the sections of which satisfy growth conditions ([KS01], [GS16]). With the help of this theory, we associate to an algebraic variety the tempered subanalytic site, the Stein tempered subanalytic site and the sheaves of rings of tempered holomorphic functions over them. Then, we study the relationship between these sheaves and the sheaf of regular functions. We prove that on a Zariski open subset of X the ring of tempered holomorphic functions is isomorphic to the ring of regular functions. This is a generalization of the aforementioned Liouville's Theorem since it implies that on a Zariski open set a tempered holomorphic function is a regular function. In the last part of this paper, we discuss the possibility of getting a tempered GAGA theorem in the non-proper case. We construct an exact functor from the cate- gory of coherent algebraic sheaves to the category of modules over the sheaf of tempered holomorphic functions on the Stein tempered subanalytic site associated with X. We prove the fully faithfulness of this functor under the assumption of the vanishing of the tempered Dolbeaut cohomology on relatively compact subanalytic pseudo-convex open subset of Cn. This vanishing result seems to be known to some specialists. Acknowledgment: We would like to thank Pierre Schapira for many scientific advice and for generously sharing is knowledge of the theory of subanalytic sheaves. We also thank Jean-Pierre Demailly for several pointed explanations concerning tempered Dolbeaut cohomolgy as well as Mauro Porta for useful discussions. We also thanks Tony Pantev for his invitation and hospitality to the University of Pennsylvania where this work was completed. 2 1 Sheaves on the subanalytic site The definitions and results of the subsections 1.1 and 1.3 are due to [KS96, KS01] and we follow closely their presentation. We also refer to [Pre08], where the operations for subanalytic sheaves are developed without relying on the theory of ind-sheaves. 1.1 Subanalytic topology, tempered functions and distribu- tions: a review 1.1.1 The subanalytic topology Definition 1.1.1. (a) Let M be a real analytic manifold. Let OpM be the category where the objects are the open subsets of M and the morphisms are the inclusions between open subsets. The category OpMsa is the full subcategory of OpM spanned by the relatively compact subanalytic open subsets of M. (b) The site Msa is the category OpMsa endowed with the Grothendieck topology where a family {Ui}i∈I of objects is a covering of U ∈ OpMsa if for every i ∈ I, Ui ⊂ U and there exists a finite subset J ⊂ I such that Sj∈J Uj = U. Let k be a field. The Grothendieck category of sheaves of k-modules on Msa is denoted by Mod(kMsa). The next result is useful to construct subanalytic sheaves. It is a special case of [KS01, Proposition 6.4.1]. Proposition 1.1.2. A presheaf F on Msa is a sheaf if and only if F (∅) = 0 and for any pair (U1, U2) in OpMsa, the sequence 0 −→ F (U1 ∪ U2) −→ F (U1) ⊕ F (U2) −→ F (U1 ∩ U2) is exact. The morphism of sites ρsa induces the following adjoint pair of functors (1.1) ρ−1 sa : Mod(kMsa) / Mod(kM ) : ρsa∗ . Moreover, ρ−1 rise to the adjoint pair of functors sa is exact and the functor ρsa∗ is fully faithful and left exact. This gives (1.2) sa : Db(kMsa) ρ−1 / Db(kM ) : Rρsa∗ . The functor Rρsa∗ is fully faithful. The functor ρ−1 sa has also a left adjoint, denoted ρsa! : Mod(kM ) −→ Mod(kMsa). We refer the reader to [KS01, §6.6] for more details and only recall its construction and main properties. If F ∈ Mod(kM ), ρsa!F is defined as the sheafification of the presheaf OpMsa ∋ U 7→ F (U). The functor ρsa! is exact, fully faithful and commutes with tensor products. 3 / o o / o o 1.1 Subanalytic topology, tempered functions and distributions: a review The situation is summarized by the following diagram. Mod(kM ) ρsa ∗ sa ρ−1 ρsa ! Mod(kMsa). where (ρ−1 sa , ρsa∗) and (ρsa!, ρ−1 sa ) form adjoint pairs of functors. Note that M is not an object of OpMsa unless M is compact. Thus, we define the global section functor as follow. For an object F of D(kMsa), one sets (1.3) RΓ(Msa; F ) = RHom (kMsa, F ). It follows from the isomorphism kMsa ≃ ρsa!kM , the adjunction (ρsa!, ρ−1 mula (1.3) that sa ) and for- (1.4) RΓ(Msa; F ) ≃ RΓ(M; ρ−1 sa F ). 1.1.2 Induced topology Let U ∈ OpMsa. We endow U with the topology induced by Msa, that is, we consider the site UMsa defined as follows (a) the objects of OpUMsa are the open subanalytic subsets of U, no more necessarily relatively compact, (b) a covering of V ∈ OpUMsa and there is a finite subset J ⊂ I with V = Sj∈J Vj. is a family {Vi}i∈I of objects of OpMsa such that Vi ⊂ V Note that the natural morphism of sites UMsa : Usa −→ UMsa is not an equivalence of sites in general. 1.1.3 Tempered functions and distributions All along this paper M is a real analytic manifold. We denote by C∞ M the sheaf of C-valued functions of class C∞ on M and by DbM the sheaf of Schwartz's distributions on M. Definition 1.1.3. Let U be an open subset of M and f ∈ C∞ M (U). The function f has polynomial growth at p ∈ M if f satisfies the following condition: for a local coordinate system (x1, . . . , xn) around p, there exist a sufficiently small compact neighbourhood K of p and a positive integer N such that (1.5) (cid:0)dist(x, K \ U)N f (x))(cid:1) < ∞ . sup x∈K∩U Here, dist(x, K \ U) := inf {y − x ; y ∈ K \ U}, and we understand that the left-hand side of (1.5) is 0 if K ∩ U = ∅ or K \ U = ∅. We say that f has polynomial growth if it has polynomial growth at any point of M. We say that f is tempered at p if all its derivatives have polynomial growth at p. We say that f is tempered if it is tempered at any point of M. 4 / / / / o o 1.2 Tempered functions and distributions: complements Remark 1.1.4. (i) In the above definition, f has polynomial growth at any point of the open subset U. (ii) A holomorphic function which has polynomial growth on a relatively compact open subset U of Cn is tempered on U. This follows from the Cauchy formula (see [Siu70, Lemma 3] for a detailed proof). We recall Lojaciewicz's inequality. Theorem 1.1.5 (Lojaciewicz's inequality [BM88]). Let M be a real analytic manifold and let K be a subanalytic subset of M. Let f, g : K −→ R be subanalytic functions with compact graphs. If f −1(0) ⊂ g−1(0), then there exists c, r > 0 such that, for all x ∈ K, f (x) ≥ c g(x)r. The following metric property of subanalytic subsets of Rn is a consequence of the Lojaciewicz's inequality (See the seminal papers by Lojasiewicz [Loj59] and Mal- grange [Mal66]). This result is used to construct the sheaf of tempered smooth func- tions. Lemma 1.1.6. Let U and V be two relatively compact subanalytic open subsets of Rn. Then, there exist a positive integer N and a constant C > 0 such that dist(cid:0)x, Rn \ (U ∪ V )(cid:1)N ≤ C(cid:0)dist(x, Rn \ U) + dist(x, Rn \ V )(cid:1). We define the presheaf C∞,t Msa by OpMsa ∋ U 7→ C∞,t M (U) where C∞,t Msa which associates to a subanalytic open subset U of M the image Db t M (U) is the sub- M (U) consisting of tempered C∞-functions. We also consider the presheaf M (U) of the M (U) is called the space space of C∞ Db t restriction morphism Γ(M; DbM ) −→ Γ(U; DbM ). The space Db t of tempered distributions on U. Proposition 1.1.7. The presheaves U 7→ C∞,t Msa. M (U) and U 7→ Db t M (U) are sheaves on Proof. This follows from Lemma 1.1.6 and Proposition 1.1.2. Q.E.D. We have the following natural morphisms (see [KS01, p.122]) (1.6) and isomorphisms (1.7) ρ−1 sa ρsa!C∞ M ρsa!C∞ M −→ C∞,t Msa −→ ρsa∗C∞ M ∼−→ ρ−1 sa C∞,t Msa ∼−→ ρ−1 sa ρsa∗C∞ M ∼−→ C∞ M , ρ−1 sa Db t Msa ≃ DbM . 1.2 Tempered functions and distributions: complements In this subsection, we establish some complementary results concerning tempered func- tions and distributions. All manifolds M1, M2 etc. are real analytic. Set for short Mij = Mi × Mj, M123 = M1 × M2 × M3 and denote by pi the i-th projection defined on Mij or on M123 and by pij the (i, j) projection. Recall that if K is a closed subanalytic subset of M1, then, a function f : K −→ M2 is subanalytic if its graph Γf is subanalytic in M12. 5 1.2 Tempered functions and distributions: complements Lemma 1.2.1. Let M1, M2 and M3 be real analytic manifolds, K be a closed subanalytic subset of M1, f : K −→ M2 and g : M2 −→ M3 be continuous subanalytic functions. Then g ◦ f is a subanalytic function. Proof. By assumption Γf is a closed subanalytic subset of M12 and Γg is a closed subanalytic subset of M23. The set p−1 12 (Γf ) is subanalytic and closed in M123. 23 (Γg) ∩ p−1 Since p13 is proper over p−1 Γg◦f = p13(p−1 23 (Γg) ∩ p−1 12 (Γf ), the set 23 (Γg) ∩ p−1 12 (Γf )). is subanalytic in M13. We recall the following fact. Q.E.D. Lemma 1.2.2 ([BM88, Remark 3.11]). Let A be a subanalytic subset of Rn. Then, the function Rn −→ R, x 7→ dist(x, A) is subanalytic. Here, dist is the Euclidian distance. The following Proposition is probably well-known but we have no reference for it. So, we provide it with a proof. Proposition 1.2.3. Let M and N be two real analytic manifolds, let f : M −→ N be a real analytic map, let V be a subanalytic open subset of N and set U = f −1(V ). Let ψ be a smooth function with polynomial growth on V . Then ψ ◦ f has polynomial growth on U. Proof. The preimage of a subanalytic set by a real analytic map is subanalytic, thus U is subanalytic in M. Let p ∈ M. In view of Definition 1.1.3 and Remark 1.1.4, we can assume that p ∈ ∂U. There is a coordinates system (y1, . . . , ym) around f (p) as well as a sufficiently small subanalytic compact neighbourhood K ′ of f (p) and a non-negative integer N ′ such that (cid:0)dist(y, K ′ \ V )N ′ ψ(y)(cid:1) < ∞. sup y∈K ′∩V There is a coordinates system (x1, . . . , xm) around p and a sufficiently small subanalytic compact neighbourhood K of p such that ψ(K) ⊂ K ′. Shrinking K and K ′ if necessary, we can further assume that dist(x, K \ U) ≤ 1 and dist(f (x), K ′ \ V ) ≤ 1 for x ∈ K. The function f : K −→ N is subanalytic in M × N. It follows from Lemmas 1.2.2 and 1.2.1 that the continuous functions on K, dist(x, K \ U) and dist(f (x), K ′ \ V ) are subanalytic. Moreover, dist(f (x), K ′ \ V ) = 0 implies f (x) /∈ V hence x /∈ U and thus dist(x, K \ + such U) = 0. We deduce from the Lojaciewicz's inequality that there exists C, α ∈ R∗ that for every x ∈ K dist(x, K \ U)α ≤ C dist(f (x), K ′ \ V ). There is a non-negative integer N ≥ max(αN ′, N ′) such that for every x ∈ K ∩ U dist(x, K \ U)N ψ ◦ f (x) ≤ dist(x, K \ U)αN ′ ψ ◦ f (x) ≤ C dist(f (x), K ′ \ V )N ′ ψ ◦ f (x) which proves that sup x∈K∩U (cid:0)dist(x, K \ U)N ψ ◦ f (x)(cid:1) < ∞. Q.E.D. 6 1.2 Tempered functions and distributions: complements Corollary 1.2.4. Let M and N be two real analytic manifolds, let f : M −→ N be a real analytic map, let V be a subanalytic open subset of N and set U = f −1(V ). Let ψ be tempered smooth function on V . Then ψ ◦ f is a tempered smooth function on U. Finally, there is the following important result concerning Db t M sa. This is essentially a corollary of Lemma 2.5.7 of [KS16] but it has never been stated explicitly and due to its importance we provide a detailed proof. Consider a morphism of real analytic manifold f : M −→ N. Recall the natural isomorphism [KS16, Th. 2.5.6] (1.8) Dbt∨ Msa L ⊗DMsa DMsa−→Nsa ∼−→ f ! Dbt∨ Nsa. Recall that Dbt∨ (differential form of higher degree tensorized with the orientation sheaf). Msa is the tensor product of Db t Msa with the sheaf of analytic densities Proposition 1.2.5. Consider a morphism of real analytic manifolds f : M −→ N and let U ∈ OpMsa and V ∈ OpNsa. Assume that f induces an isomorphism U ∼−→ V . Then (a) f U induces an isomorphism of sites UMsa ≃ VNsa, (b) isomorphism (1.8) induces an isomorphism (1.9) f U ∗Db t MsaUMsa ≃ Db t NsaVNsa . Proof. (a) is clear. (b) It is enough to prove that for U ′ ∈ OpUMsa and V ′ ∈ OpVNsa with f (U ′) = V ′, we have an isomorphism Db t Msa(U ′) ≃ Db t Nsa(V ′). Changing our notations, we will prove this formula for U and V . To prove this statement, we work with indsheaves. This does not cause any problem since subanalytic sheaves form a full subcategory of the category of indsheaves (see Proposition 2.4.3 of [KS16] for more details). We follow the notations of [KS16] for all matters regarding ind-sheaves. We denote by Db t N the subanalytic sheaves Db t M and Db t Nsa viewed as indsheaves. In particular, we have Msa(U) ≃ RHom D(IkM )(CU , Db t M ) Msa and Db t Db t and similarly with Db t Nsa(V ). From Lemma 2.5.7 of [KS16], we have RIhom (CU , Db t M ) ≃ f ! RIhom (CV , Db t N ). Then, RHom D(IkM )(CU , Db t M ) ≃ RHom D(IkM )(CU , RIhom (CU , Db t M )) ≃ RHom D(IkM )(CU , f ! RIhom (CV , Db t ≃ RHom D(IkM )(CV ⊗ Rf!!CU , Db t N ) ≃ RHom D(IkM )(CV ⊗ Rf!CU , Db t N ) ≃ RHom D(IkM )(CV , Db t N ). N )) 7 Q.E.D. 1.3 Tempered holomorphic functions Remark 1.2.6. It follows from Corollary 1.2.4 that there is a morphism of sheaves of rings (1.10) Nsa −→ f∗C∞,t C∞,t Msa, ϕ 7→ ϕ ◦ f. It follows from the commutativity of Diagram (1.11) below (1.11) f!Db∨ U Rf / Db∨ V ·⊗dλ ≀ ≀ ·⊗dλ f!DbU DbV R (·)J ac(f )dλ R (·)dλ fU ∗C∞ U f ∗ C∞ V that the isomorphism (1.9) is compatible with the morphism (1.10). This shows that (1.10) induces an isomorphism of sheaves of rings (1.12) f U ∗C∞,t MsaUMsa ≃ C∞,t Nsa VNsa . 1.3 Tempered holomorphic functions In this subsection, we construct the sheaf of tempered holomorphic functions. Let X be a complex manifold. We denote by X c the complex conjugate manifold of X and by XR the underlying real analytic manifold. We write Xsa for the manifold XR endowed with the subanalytic topology and set Following [KS01], we define the following sheaves on Xsa. DXsa := ρsa! DX . (1.13) (1.14) (1.15) O ω Xsa O t Xsa OXsa OX, := ρsa! := RH om D Xc sa := Rρsa∗ OX. sa, Db t (O ω X c XR), The objects O ω isomorphic to the Dolbeault complex with coefficients in Db t Xsa and OXsa belong to the category Db(DXsa). The object O t Xsa is XR on the subanalytic site: Xsa, O t (1.16) 0 −→ Db t Xsa ∂−→ Dbt (0,1) Xsa ∂−→ · · · ∂−→ Dbt (0,dX ) Xsa −→ 0. One calls O t Xsa the sheaf of tempered holomorphic functions. We have natural mor- phisms in Db(DXsa): By [KS96, Theorem 10.5] we have the isomorphism O ω Xsa −→ O t Xsa −→ OXsa. (1.17) O t Xsa ≃ RH om D Xc sa (ρsa! OX c, C∞,t Xsa ) in Db(DXsa). 8 / O O O O O O O O o o This implies, in particular, that O t Xsa is represented by the differential graded algebra (1.18) 0 −→ C∞,t Xsa ∂−→ C∞,t (0,1) Xsa ∂−→ · · · ∂−→ C∞,t (0,dX ) Xsa −→ 0. Consider a morphism of complex manifolds f : X −→ Y First, recall the morphism of [KS01, Lem. 7.4.9] or [KS16, Cor. 3.1.4] (1.19) DXsa−→Ysa L ⊗f −1DYsa f −1O t Ysa −→ O t Ysa Note that this morphism is constructed from morphism (1.8). Proposition 1.3.1. Consider a morphism of complex manifolds f : X −→ Y and let U ∈ OpXsa and V ∈ OpYsa. Assume that f induces an isomorphism of complex analytic manifolds U ∼−→ V . Then isomorphism (1.8) induces an isomorphism (1.20) Rf U ∗ O t XsaUXsa ≃ O t YsaVYsa . Proof. We shall use Proposition 1.2.5. Recall first that f U induces an isomorphism of sites UXsa ≃ VYsa and that the morphism of DXsa-module DXsa −→ DXsa−→Ysa restricted to UXsa is an isomorphism. We have the sequence of isomorphisms f U ∗ O t XsaUXsa ≃ f U ∗RH om D sa, Db t (O ω X c Xc sa sa, Db t (O ω Ysa)VYsa Y c Xsa)UXsa ≃ RH om D ≃ O t YsaVYsa . Y c sa Remark 1.3.2. The sheaf H0(O t Xsa) is a sheaf of rings whose multipicative law is given by the multiplication of functions. Isomorphism (1.20) induces an isomorphism of sheaves of rings: Q.E.D. (1.21) f U ∗H0(O t Xsa)UXsa ≃ H0(O t Ysa)VYsa . This follows from the isomorphism (1.12). The following result seems known to some specialists but there is no reference for it in the literature so we do not label it as a theorem. Conjecture 1.3.3. Assume that X = Cn and that U is a relatively compact subana- Xsa) ∈ Db(C) is concentrated in lytic pseudo-convex open subset of X. Then RΓ(U; O t degree 0. 2 Tempered analytification 2.1 T P-spaces In this subsection, we review the notion of T P-space. This a notion which is interme- diate between the notion of topological space and the concept of site and will be better suited to our purpose than the notion of ringed sites. It was introduced in [KS01] and further studied in [EP16]. 9 2.2 The tempered analytification functor Definition 2.1.1. (i) A T P-space (X, T ) is the data of a set X together with a collection of subsets T of X such that (i) ∅ ∈ T , (ii) T is stable by finite unions and finite intersections. (ii) A morphism of T P-spaces f : (X, TX ) −→ (Y, TY ) is a map f : X −→ Y such that for every U ∈ TY , f −1(U) ∈ TX . We have thus defined the category of T P-spaces, that we denote by TS. The forgetful functor f or : Top −→ TS form the category of topological spaces to the category of T P-spaces admits a left adjoint (·)Top : TS −→ Top which associate to a T -space (X, T ) a topological space (X, T Top) where T Top := {SV ∈B V B ⊂ T }. When there is no risk of confusion we will write X Top instead of (X, T Top). A T P-space defines a presite, the morphisms U −→ V being the inclusions. We endow it with the following Grothendieck topology: A family {Ui}i∈I of objects of T is a covering of U if for every i ∈ I, Ui ⊂ U and there exists a finite subset J ⊂ I such that Sj∈J Uj = U. We denote this site by XT . A morphism of T P-spaces f : (X, TX) −→ (Y, TY ) induces a morphism of sites f : XTX −→ YTY , TY ∋ V 7→ f −1(V ) ∈ TX . Notation 2.1.2. If there is no risk of confusion, we will not make the distinction between a T P-space (X, T ) and the associated site XT . Given a T P-space X and U ∈ T , U is naturally endowed with a structure of T P- space (U, TU ) by setting TU = {V ⊂ U; V ∈ T }. We denote by UXT the site induced by XT on the presite (U, TU ). Hence, the coverings of UXT are those induced by the coverings in X. Definition 2.1.3. A site XT (associated with a T P-space) endowed with a sheaf of rings OX is called a ringed T P-space. One defines as usual a morphism of ringed T P-spaces . 2.2 The tempered analytification functor The aim of this subsection is to construct the tempered analytification functor. Notation 2.2.1. We denote by • Var (resp. Varsm) the category of complex algebraic varieties (resp. smooth com- plex algebraic varieties), • AnC the category of complex analytic spaces, • (·)an : Var −→ AnC the analytification functor, • TRgS the category of ringed T P-spaces. • TFS the category whose objects are the pair ((X, T ), F ) where (X, T ) is a T P- space and F ∈ Db(kXT ). A morphism in TFS is a pair (f, f ♯) : ((X, TX ), F ) −→ ((Y, TY ), G) where f : X −→ Y is a morphism of T P-space and f ♯ : f −1G −→ F is a morphism in Db(kXT ). 10 2.2 The tempered analytification functor In general, if there is no risk of confusion, we shall not write the functor (·)an. For example, if X is a smooth complex algebraic varieties, then Xsa is the subanalytic site associated with X an. We refer the reader to [SGA1, Expos´e XII] for a detailed study of the properties of the analytification functor. We recall a few classical facts concerning smooth compactifications of algebraic varieties. (i) It follows from the Nagata's compactification theorem and Hironaka's desingular- ization theorem that any smooth algebraic variety has a smooth algebraic com- pactification. (ii) Compactification is not functorial but let f : X0 −→ X1 be a morphism of algebraic varieties and j1 : X1 −→ Y1 be a compactification of X1. Then, there exists a morphism f : Y0 −→ Y1 where j0 : X0 −→ Y0 is a smooth compactification of X0 such that the following diagram commutes (2.1) Y0 j0 X0 f f / Y1 j1 / X1. This is a consequence of the following classical construction. zar Let f : X0 −→ X1 be a morphism of algebraic varieties. Let Y2 and Y1 be respec- tively smooth compactification of X0 and X1. Consider Γf the graph of f as a subset of Y2 × Y1 and consider its closure, in the Zariski topology, Γf . Note . By Hironaka's theorem there exists a smooth algebraic that Γf is open in Γf , such that g : g−1(Γf ) −→ Γf variety Y0 and a proper morphism g : Y0 −→ Γf is an isomorphism. This implies that there is an open embedding j0 : X0 −→ Y0. Thus j0 is a smooth compactification of X. Let pi : Y2 × Y1 −→ Yi (i = 1, 2) be the projection on Yi and we still denote by pi . We set bf = p1 ◦ g and notice that bf X0 = f . their respective restrcition to Γf Then, we have the diagram zar zar zar (2.2) p2 Y2 Γf g zar p1 Y1 <②②②②②②②②② bf j1 f / X1, j2 X0 Y0 j0 X0 (iii) Smooth compactifications are not unique but there is the following consequence of the construction in (ii). Consider two smooth compactifications of X, j1 : X ֒→ Y1 11 /  ? O O /  ? O O / / o o O O <  ? O O  ? O O /  ? O O 2.2 The tempered analytification functor and j2 : X ֒→ Y2. Applying (ii) with X0 = X1 = X and f = id, we get the following commutative diagram (2.3) Y0 q1 ~⑦⑦⑦⑦⑦⑦⑦ j0 Y1 j1 X q2 ❅❅❅❅❅❅❅ j2 Y2 where Y0 is a smooth compactification of X, q1 = p1 ◦g, q2 = p2 ◦g. The morphism q1 (resp. q2) induces an isomorphism q1 : j0(X) ∼−→ j1(X) (resp. q2 : j0(X) ∼−→ j2(X)). Lemma 2.2.2. Let X be a smooth algebraic variety and Y a smooth algebraic com- pactification of X. (a) The site XYsa does not depend on the choice of a smooth algebraic compactification Y of X, (b) the object O t YsaXYsa ∈ Db(kXYsa ) does not depend on the choice of an algebraic compactification Y of X. Proof. (a) is a consequence of Proposition 1.3.1 (a) and Diagram (2.3). (b) is a consequence of Proposition 1.3.1 (b) and Diagram (2.3). Q.E.D. As a consequence of Lemma 2.2.2 we can state Definition 2.2.3. Let X be a smooth algebraic variety. We define the tempered an- alytification (t-analytification for short) of X to be the object of TFS, (X tan, O t X tan) where X tan := (X(C), XYsa) and O t X tan := O t YsaXYsa . We will associate to a morphism of smooth algebraic varieties f : X0 −→ X1 is tem- pered f tan analytification. In order to do so, we need the following Lemma. Lemma 2.2.4. Let f : X0 −→ X1 be a morphism of smooth algebraic varieties. Let U ∈ OpX tan . Then (f an)−1(U) ∈ OpX tan . 0 1 Proof. There exist smooth algebraic compactifications Y0 and Y1 of X0 and X1 such that f extends to a regular morphism bf : Y0 −→ Y1 such that Diagram (2.1) commutes. Let U ∈ OpX tan, then f −1(U) = j−1 0 bf −1(j1(U)) = bf −1(U) ∩ X an 0 . It follows that f −1(U) is subanalytic in Y0. Q.E.D. 12 ~ / / o o O O 2.3 The sheaves O t X tan and OXzar Let f : X0 −→ X1 be a morphism of smooth algebraic variety. Using Lemma 2.2.4, we associate to f a morphism of site (2.4) f tan : X tan 0 −→ X tan 1 , OpX tan 1 ∋ U 7→ (f an)−1(U) ∈ OpX tan 0 . Lemma 2.2.5. Let f : X0 −→ X1 be a morphism of smooth algebraic varieties. Then f induces a canonical morphism in Db(kX tan 0 ) f tan ♯ : (f tan)−1O t X tan 1 −→ O t X tan 0 Proof. There exist smooth algebraic compactifications Y0 and Y1 of X0 and X1 such that f extends to a regular morphism bf : Y0 −→ Y1 such that Diagram (2.1) commutes. Using the presentation of O t Y1sa) by the complexe (1.18) and Corollary 1.2.4, the pullback of differential forms by f provides a morphism Y0 sa (resp. O t (C∞,t (0,•) Y1sa , ∂) −→ f∗(C∞,t (0,•) Y0sa , ∂)); ω 7→ f ∗ω Restricting to X tan and using the adjunction ( f −1, f∗), we get the desired morphism f tan, ♯ : (f tan)−1(C∞,t (0,•) X tan 1 , ∂) −→ (C∞,t (0,•) X tan 0 , ∂)). Moreover, this morphism does not depend of the choice of a compactification. Q.E.D. The datum of (f tan, f tan ♯) defines a morphism in TFS. If there is no risk of confusion, we write f tan instead of (f tan, f tan ♯). Finally, if g : Y0 −→ Y1 is an other regular morphism, on checks that (g ◦ f )tan = gtan ◦ f tan and that idtan = id. In view of the preceding construction we can state the following Theorem. Theorem 2.2.6. The functor (·)tan : Varsm −→ TFS, (X, OX) 7→ (X tan, O t X tan), f 7→ f tan is well defined and is called the functor of tempered analytification (t-analytification for short). We denote by TAnC its image in TFS. Remark 2.2.7. In a next paper, we will refine the construction of the functor (·)tan to get a functor valued in the ∞-category of E∞ ringed spaces through which the usual analytification functor will factor. We prove such a result for the Stein tempered analytification functor in the next section. 2.3 The sheaves O t X tan and OXzar Let (X, OX) be a smooth algebraic variety. If we want to emphasize that X is endowed with the Zariski topology we write Xzar instead of X. We first note that there is a natural morphism of site (2.5) ρtz : X tan −→ Xzar. We shall study its properties. 13 2.3 The sheaves O t X tan and OXzar Let Y be a smooth algberaic compactification of X. There are natural morphisms of sites that fit into the following commutative diagram. ρsa zar Ysa Yzar. jXYsa jXzar X tan / Xzar ρtz Lemma 2.3.1. Let X be a smooth algebraic variety and U a Zariski open subset of X. Let f ∈ OX(U). Then, the analytification of f is a tempered holomorphic function. Proof. Let p ∈ X and V an open affine subset of X containing p, we can further assume that U is affine and since X is an algebraic variety, the open set U ∩ V is again affine. In view of Proposition 1.2.3, we can assume that V is a smooth affine subvariety of An C. Then f is the restriction to U ∩ V of a rational function P (x)/Q(x) on An C. It follows immediately from Lojaciewicz's inequality that P (x)/Q(x) is tempered in p when p is a zero of Q. Q.E.D. Using the commutative differential graded algebra (1.18) to represent O t Ysa, the above observations implies that there is a canonical morphism of sheaves of differential graded rings Restrincting to Xzar and using the adjunction (ρ−1 tz , ρtz∗) this induces a morphism OYzar −→ ρsa zar∗ O t Ysa , ϕ 7→ ϕan. (2.6) Ltan : ρ−1 tz OXzar −→ O t X tan. By adjunction, we get the morphism (2.7) OXzar −→ Rρtz∗ O t X tan. Let (X ′, OX ′) be a complex manifold and Z ′ be an anlytic subset of X ′. We denote by IZ ′ the sheaf of holomorphic functions vanishing on Z ′ and set Γ[Z ′](OX ′) := lim−→ k Hom O X ′ (OX ′/I k Z ′, OX ′). Lemma 2.3.2. Let X be a smooth algebraic variety and Z be a closed subset of X. Then, (2.8) RΓ[Z an](OX an) ≃ (RΓZ(OXzar))an. Proof. Writing IZ an for the defining ideal of Z an in X an, we have that Γ[Z an](OX an) := lim−→ k Hom OX an (OX an/I k Z an, OX an) ≃ (lim−→ k Hom OX (OX/I k Z , OX))an. Moreover by [Har67, Theorem 2.8], on an algebraic variety, Hi Z(OX) ≃ lim−→ k E xti OX (OX/I k Z , OX). 14 Q.E.D. / /     / Theorem 2.3.3. The object Rρtz∗ (2.7) is an isomorphism. O t X tan is concentrated in degree zero and the morphism Proof. It is enough to check the isomorphism RΓ(U; O t Ysa) ≃ RΓ(U; OYzar) for any U ∈ OpXzar. Let Z := Y \ U. There is the following commutative diagram RΓZ(Y ; OYzar) RΓ(Y ; OYzar) RΓ(U; OYzar) [+1] Ltan Ltan Ltan RΓZ(Y ; O t Ysa) / RΓ(Y ; O t Ysa) / RΓ(U; O t Ysa) [+1] / . where the rows are distinguished triangle. As Y is proper RΓ(Y ; O t Then, RΓ(Y ; O t reduced to prove the local isomorphism Ysa) ≃ RΓ(Y an; OY an). Ysa) ≃ RΓ(Y ; OYzar) by the GAGA theorem [SGA1, Expos´e XII]. We are ρ−1 sa RΓZ(O t Ysa) ≃ RΓ[Z an](OY an). This result is Theorem 5.12 in [KS96]. This implies in particular that the morphism OYzar −→ Rρsa Q.E.D. Ysa is an isomorphism. O t zar∗ Corollary 2.3.4. A tempered holomorphic function on a Zariski open set is regular. Corollary 2.3.5. Let X and Y be two smooth algebraic varieties and f : X an −→ Y an an holomorphic map. Assume that Y is affine and that f induces a morphism of ringed T P-spaces f : X tan −→ Y tan, OpY tan ∋ U 7→ f −1(U) ∈ OpX tan H0(f tan ♯) : f −1H0(O t X tan), ϕ 7→ ϕ ◦ f. Y tan) −→ H0(O t Then ϕ is the analytification of a regular morphism. Proof. Since Y is affine, there is an algebraic closed immersion jY : Y −→ An composition, we get a tempered morphism jY f : X −→ An to the coordinate functions xi : An O t C. By C and applying this morphism C −→ C with 1 ≤ i ≤ n, we get that xi(jY f ) ∈ Q.E.D. Y tan(Y ) = OYzar(Y ). Thus, f is algebraic. 3 Stein tempered analytification 3.1 The subanalytic Stein topology In view of Conjecture 1.3.3 it is natural to introduce the sas-topology defined below. In all of Section 3, we are not assuming Conjecture 1.3.3 unless explicitely stated. 15 / /   / /     / / / / / 3.1 The subanalytic Stein topology Definition 3.1.1. Let X be a complex manifold. We denote by Xsas the presite for which the open sets are the finite unions of U ∈ OpXsa with U Stein. We endow Xsas with the topology induced by Xsa. We have the natural morphisms of sites ρsa X / Xsa ρsasa ρsas / Xsas where ρsasa := ρsas ◦ ρsa. The morphism of sites ρsas induces the following pairs of adjoint functors (3.1) and (3.2) ρ−1 sas : Mod(kXsas) / Mod(kXsa) : ρsas∗ ρ−1 sas : D+(kXsas) / D+(kXsa) : Rρsas∗. Note that the functors ρsas∗ and ρsa∗ commute with Hom and preserve injectives. The last observation implies that Rρsasa∗ ≃ Rρsas∗ ◦ Rρsa∗. Proposition 3.1.2. (a) The functor ρsasa∗ and Rρsasa∗ are fully faithful. (b) The functor ρsas∗ and Rρsas∗ are fully faithful. Proof. (a) This follows from the fact that any point x ∈ X has a fundamental sys- tem of neighbourhood composed of Stein subanalytic open sets. For details see [KS01, Proposition 6.6.1]. This, together with the exactness of ρ−1 sasa implies that ρ−1 sasa Rρsasa∗ −→ id is an isomorphism. (b) By definition of ρsasa, ρsasa∗ = ρsas∗ρsa∗. Since ρsasa∗ and ρsa∗ are fully faithful, the functor ρsas∗ is fully faithful. The proof for Rρsas∗ is similar. Q.E.D. We will also need for technical matters a variant of the sas topology. These two topologies have equivalent categories of sheaves. Definition 3.1.3. We denote by Xsat the presite for which the open sets are the U ∈ OpXsa with U a Stein open subset. We endow Xsat with the topology induced by Xsa. We have the natural morphisms of sites Xsas ρsat / Xsat The functor ρsat∗ : Mod(kXsas) ∼−→ Mod(kXsat) is an equivalence of categories the inverse of which ρ‡ sat : Mod(kXsat) ∼−→ Mod(kXsas) is given by F (Ui). ρ‡ satF (U) := lim←− Ui−→U Ui∈OpXsat More generally, the topos associated to Xsas and Xsat are equivalent. We set (3.3) O t Xsas := H0(Rρsas∗ O t Xsa) 16 5 5 / / / o o / o o / 3.1 The subanalytic Stein topology Xsa ∈ Remark 3.1.4. Assuming the Conjecture 1.3.3, we deduce that the object Rρsas∗ Db(CXsas) is concentrated in degree 0. To check it, it is enough to prove that any U ∈ OpXsas admits a finite covering U = Si Ui with RΓ(Ui; Rρsas∗ Xsa) concentrated in degree 0. We may assume U is Stein and we cover U with a finite union of open sets Vi, i ∈ I, such that Vi ∈ OpXsa, Vi is Stein and Vi is contained and relatively compact in a chart φi : Wi i = φi(Ui). Then Xsa) ≃ RΓ(U ′ RΓ(Ui; Rρsas∗ Cn) and this last complex is concentrated in degree 0. sa. Set Ui = U ∩ Vi and U ′ ∼−→ W ′ i ; O t i ⊂ Cn O t O t O t This justifies the notation (3.3). Let U ∈ OpXsas. We endow U with the topology induced by Xsas and we denote this site by UXsa. There is a natural morphsim of sites Usas −→ UXsas. In general, this morphism is not an equivalence of site. Proposition 3.1.5. Consider a morphism of complex manifolds f : X −→ Y and let U ∈ OpXsas and V ∈ OpYsas. Assume that f induces an isomorphism of complex analytic manifolds fU : U ∼−→ V . Then, fU induces (a) an isomorphism of sites fU : UXsas −→ VYsas , given by the functor f t U : OpUXsas −→ OpVYsas , W 7→ fU −1(W ) = f −1(W ) ∩ U, (b) an ismorphism of sheaves of rings (3.4) fU ∗ Proof. (a) clear. O t XsasUXsas ≃ O t YsasVYsas . (b) I follows from Proposition 1.3.1 and Remark 1.3.2 that we have an isomorphism of rings (3.5) f U ∗H0(O t Xsa)UXsa ≃ H0(O t Xsa)VYsa . Applying the functor ρVYsas ∗ to the isomorphism (3.5) and using the below commu- tative diagram of morphism of sites provides the isomorphism (3.4). Xsa jUXsa VYsa fU <②②②②②②②② ρUXsas UXsa Ysa f <①①①①①①①① jVXsa ρsas ρsas / Xsas ;✇✇✇✇✇✇✇✇✇ Ysas jVYsas ρVYsas jUXsas / VYsas fU ;✇✇✇✇✇✇✇✇ / UXsas 17 Q.E.D. / /     <   / ;   / < / ; 3.2 The Stein tempered analytification functor 3.2 The Stein tempered analytification functor We keep the notation of Section 2. The aim of this section is to construct the Stein tempered analytification functor. For that purpose, we need the following lemma. Lemma 3.2.1. Let X be a smooth algebraic variety and Y a smooth algebraic com- pactification of X. (a) The site XYsas does not depend on the choice of a smooth algebraic compactification Y of X, (b) The sheaf of rings O t YsasXYsas ∈ Mod(kXYsas ) does not depend on the choice of an algebraic compactification Y of X. Proof. (a) is a consequence of Proposition 3.1.5 (a) and Diagram (2.3). (b) is a consequence of Proposition 3.1.5 (b) and Diagram (2.3). Q.E.D. As a consequence of Lemma 3.2.1 we can state Definition 3.2.2. Let X be a smooth algebraic variety. The Stein tempered analyti- fication (st-analytification for short) of X is the ringed T P-space (X sta, O t X sta) where X sta := (X(C), XYsas) and O t X sta := O t YsasXYsas . There is a natural morphism of sites ρtan sta : X tan −→ X sta and O t X sta ≃ ρtan sta ∗H0(O t X tan). We associate to a morphsim of smooth algebriac varieties f : X0 −→ X1 its Stein tempered analytification f sta. We first establish the following lemma. Lemma 3.2.3. Let f : X0 −→ X1 be a morphism of smooth algebraic varieties. Let U ∈ OpX sta . Then (f an)−1(U) ∈ OpX sta . 1 0 Proof. There exist smooth algebraic compactifications Y0 and Y1 of X0 and X1 such that f extends to a regular morphism bf : Y0 −→ Y1 such that Diagram (2.1) commutes. Assume that U ⊂ X1 is a subanalytic Stein open subset of Y1. By Lemma 2.2.4, f −1(U) is subanalytic in Y0. The Zariski open subset X0 of Y0 has a finite covering (Vi)1≤i≤p by Stein subanalytic open subsets of Y0 (For instance, take the analytification of a finite open affine covering of X0). Let fi be the restriction of f to Vi. Then f −1 (U) = f −1(U) ∩ Vi is a subanalytic Stein open subset of Y0 and f −1(U) = S1≤i≤n f −1 (U). Thus f −1(U) is a finite union of subanalytic Stein open subsets of Y0. Finally, if U is a finite union of Stein open subsets (Ui)1≤i≤m which are subanalytic in Y1, the above argument applied to each Ui proves that f −1(U) is again a finite union of Stein open subsets of Y0. Q.E.D. i i Let f : X0 −→ X1 be a morphism of smooth algebraic varieties. By Lemma 3.2.3, the morphism of sites f tan : X tan 0 −→ X tan 1 induces a morphism f sta : X sta 0 −→ X sta 1 , OpX sta 1 ∋ U 7→ (f an)−1(U) ∈ OpX sta 0 . 18 3.3 Comparison between the different analytification functors Lemma 3.2.4. Let f : X0 −→ X1 be a morphism of smooth algebraic varieties. Then f induces a morphism of sheaves f sta ♯ : (f sta)−1O t X sta 1 −→ O t X sta 0 , ϕ 7→ ϕ ◦ f sta. Proof. This follows directly from Lemma 2.2.5. Q.E.D. The datum of (f an, f sta ♯) define a morphism of ringed T P-space. If there is no risk of confusion, we write f sta instead of (f sta, f sta ♯). Finally, if g : Y0 −→ Y1 is an other regular morphism, on checks that (g ◦ f )sta = gsta ◦ f sta and that idsta = id. The above constructions give rises to the Stein tempered analytification functor. Definition 3.2.5. The functor (·)sta : Varsm −→ TRgS, (X, OX) 7→ (X sta, O t X sta), f 7→ f sta is well defined and is called the functor of Stein tempered analytification (st- analytification for short). We denote by STAnC its image in TRgS Remark 3.2.6. The interest of the st-analytification functor is that it associates to a smooth algebraic variety, a ringed T P-space. The structure sheaf of this space has, under the assumption of Conjecture 1.3.3, the same cohomology that the sheaf of regular functions of the algebraic variety under consideration. This is not the case for the functor associating to a smooth algebraic variety the ringed T P-space (X tan, H0(O t X tan)) even under the assumption of Conjecture 1.3.3. Moreover, the st-analytification functor produces a sheaf of rings whereas an enhanced version of the t-analytification functor would produce a sheaf of E∞-ring. 3.3 Comparison between the different analytification functors A careful examination of the construction of the functor (·)sta shows that the following functor is well defined. Stn : TAnC −→ STAnC, (X tan, O t X tan) 7→ (X sta, O t X sta), f tan 7→ f sta. This means that (·)sta = Stn ◦ (·)tan. We now study the relation between the usual analytification functor, t-analytification and st-analytification functors. Let (X, OX) be smooth algebraic variety. There is a natural morphism of sites (3.6) ρsta : X an −→ X sta. We associate to the st-analytification (X sta, O t the ringed space ((X sta)Top, ρ−1 sta O t X sta). It is clear that the assignment X sta) of a smooth algebraic variety (X, OX), u : STAnC −→ AnC, (X sta, O t X sta) 7→ ((X sta)Top, ρ−1 sta O t X sta) is a well defined functor. 19 3.4 The flatness of O t X sta over OXzar We remark that that the topological space (X sta)Top is the underlying topological space of the complex manifold (X an, OX an) since X sta contains a basis of the topology of X an and X sta ⊂ X an. Moreover, there is a morphism of sheaves O t X sta −→ ρsta∗ OX an which induces by adjunction a morphism (3.7) ρ−1 sta O t X sta −→ OX an. It follows from the fact that O t of sheaves of rings. This implies the following proposition. X sta x ≃ OXx that the morphism (3.7) is an isomorphism Theorem 3.3.1. The below diagram is quasi-commutative. TAnC Stn / / STAnC (·)sta :ttttttttt u (·)tan Varsm (·)an / / AnC 3.4 The flatness of O t X sta over OXzar Let X be a smooth algebraic variety and Y be a smooth algebraic compactification of X. There is a canonical morphism of sites (3.8) ρstz : : X sta −→ Xzar. We study the properties of this morphism. By the results of the subsection 2.3, we have the morphism (2.6). Using that Rρtz∗ = Rρstz∗Rρtan sta ∗ and taking the zero degree cohomology, we obtain the morphism (3.9) Lsta : ρ−1 stz OXzar −→ O t X sta , ϕ 7→ ϕan. It is clear that Lsta does not depend of the choice of a compactification Y of X. By adjunction, we get the morphism (3.10) OXzar −→ ρzar∗ O t X sta , ϕ 7→ ϕan. Corollary 3.4.1. Let X be a smooth algebraic variety, the morphism (3.10) is an isomorphism. Proof. This is a direct consequence of Theorem 2.3.3 Q.E.D. Remark 3.4.2. Under the assumption of Conjecture 1.3.3, one obtain the following stronger version of Corollary 3.4.1. The morphism is an isomorphism. OXzar −→ Rρstz∗ O t X sta. 20   O O : 3.4 The flatness of O t X sta over OXzar 3.4.1 A flatness result In this subsection, we prove the flatness of O t X sta over ρ−1 stz OXzar. Lemma 3.4.3. Let X be a smooth affine algebraic variety and let W be an open subset of Xzar. Then OXzar(W ) is Noetherian. Proof. Let Z be the complement of W in X. The set Z is an algebraic subset of X and Z = Z≥2∪Z=1 where Z≥2 is the union of the irreducible components of Z of codimension at least two and Z=1 the union of the irreducible components of codimension exactly one. Let fW = X \ Z=1. By construction U ⊂ W ⊂ fW and since fW is the complement of a closed subset purely of codimension one in a smooth affine variety fW , it is affine. Since X is smooth it follows from Riemann second extension theorem that OXzar(W ) ≃ OXzar(fW ). This implies that OXzar(W ) is Noetherian. Q.E.D. We will need the following result. Lemma 3.4.4 ([PY14, Lemma 8.13]). Let X be a Stein manifold and U be a relatively compact Stein open subset of X. Then OX(U) is flat over OX(X). Lemma 3.4.5. Let Xzar be a smooth affine algebraic variety and U ∈ OpX sta be rela- tively compact Stein open subset of X an. Then OX an(U) is flat over OXzar(Xzar). Proof. It follows from Lemma 3.4.4 that OX an(U) is flat over OX an(X). Thus it is sufficient to prove that OX an(X) is flat over OXzar(X). Since OXzar(X) is Noetherian, we just need to show that for every exact sequence (3.11) OXzar(X)L A−→ OXzar(X)M B−→ OXzar(X)N the exact sequence obtained by applying OX an(X) ⊗OXzar (X) (·) to the sequence (3.11) is exact. As X is an affine variety the sequence (3.11) is equivalent to an exact sequence of sheaves (3.12) O L Xzar A−→ O M Xzar B−→ O N Xzar. This provides us with the following commutative diagram OX an(X) ⊗OX (X) OX(X)L A / OX an(X) ⊗OX (X) OX(X)M B / / OX an(X) ⊗OX (X) OX(X)N ≀ ≀ ≀ Γ(X; OX an ⊗ρ−1 stz OX ρ−1 stz O L X) A / / Γ(X; OX an ⊗ρ−1 stz OX ρ−1 stz O M X ) B / / Γ(X; OX an ⊗ρ−1 stz OX ρ−1 stz O N X ). The bottom line of the diagram is exact since OX an is flat over OXzar and Γ(X; ·) is exact since X is Stein. Q.E.D. Lemma 3.4.6. Let X be a smooth affine algebraic variety, let U ∈ OpX sta be a relatively compact Stein open subset of X an and W be a Zariski open subset of X containing U. Then O t X sta(U) is flat over OXzar(W ). 21 /       3.4 The flatness of O t X sta over OXzar Proof. Consider an exact sequence OXzar(W )L A−→ OXzar(W )M B−→ OXzar(W )N where A and B are two matrices with coefficients in OXzar(W ). Tensoring by O t we obtain the exact sequence X sta(U), (3.13) O t X sta(U)L A−→ O t X sta(U)M B−→ O t X sta(U)N . X sta(U)M such that Bw = 0. Then, by Lemma 3.4.3, there exists v ∈ O L Let w ∈ O t X an(U) such that Av = w. Since X is affine there exists f1, . . . , fn ∈ OXzar(X) such that W = D(f1) ∪ . . . ∪ D(fn) where D(fi) is the distinguished open set associated to fi. Thus, there exists m1, . . . mn ∈ N such that f m1 n A is an M × L matrix with coefficients in OXzar(X). Moreover f m1 . . . f mn are regular functions and f m1 n w. It follows from [Siu70, Theorem 2] that there exists u ∈ O t . . . f mn n w. This implies that Au = v on U ∩ D(f1) ∩ . . . ∩ D(fn) = U \ Z(f ) with f = f1 . . . fn. Since f is defined It follows that the sequence (3.13) is exact. As OXzar(W ) is on X, Au = v on U. Noetherian, this implies that O t Q.E.D. . . . f mn X sta(U)L such that f m1 X sta(U)M since the f mi X sta(U) is flat over OXzar(W ). n Av = f m1 n Au = f m1 n w ∈ O t 1 . . . f mn . . . f mn . . . f mn 1 1 1 1 1 i For a sheaf F on Xzar, that is, F ∈ Mod(CXzar), denote by ρ† of F in the category of presheaves on X sta. Then for U ∈ OpX sta, stzF the inverse image where W ranges over the family of objects of OpXzar such that U ⊂ W . ρ† stzF (U) = lim−→ W F (W ) Theorem 3.4.7. The sheaf of rings O t X sta is flat over ρ−1 stz OXzar. Ysas is flat OYzar and since Ysat is a dense subsite of Ysas this is equivalent to prove that Proof. Let Y be a smooth compactification of X. It sufficient to prove that O t over ρ−1 stz Ysas is flat over ρsat∗ρ−1 O t ρsat∗ stz Let (Yi)i∈I be an open affine covering of Y and (Vij)j∈J be a covering of Yi by Stein open sets relatively compact in Yi and subanalytic in Y . The family (Vij)i∈I,j∈J is a Stein subanalytic open covering of Y and since Y is compact, we can extract from it a finite covering of Y , say {V0, . . . , Vα} from (Vij)j∈J,i∈I. By construction, {V0, . . . , Vα} is a covering of Ysat. OYzar. ρ−1 stz Flatness being a local property, it sufficient to show that O t OYzarVkYsas , 0 ≤ k ≤ α. To prove this, O t it is sufficient by [TS16, Tag 03ET], to show that the presheaf OYzarVkYsas (Since Ysat is a dense subsite of Ysas, sheafi- ρsat∗ fication commutes with ρsat∗, see [TS16, Tag 03A0] for more details). This means, we have to show that for every V ′ ∈ OpYsat such that V ′ ⊂ Vk, O t Ysas(V ′) is flat over ρ† stz YsasVkYsas is flat over ρsat∗ρ† OYzar(V ′). But is flat over YsasVkYsas stz ρ† stz OYzar(V ′) = lim−→ W OYzar(W ) 22 3.5 Towards a tempered GAGA theorem where W ranges over the family of objects of OpYzar such that V ′ ⊂ W . Thus by [TS16, Tag 05UU], it is sufficient to prove that for such a W the ring Ysas(V ′) is flat over OYzar(W ). We can further assume that there is an i0 ∈ I such that O t all the W are contained in Yi0 and Vk is relatively compact subset of Yi0. The result follows then by Lemma 3.4.6. Q.E.D. 3.5 Towards a tempered GAGA theorem In this subsection, we present some results in the direction of a tempered version of the GAGA theorem of Serre. We assume all along this subsection that Conjecture 1.3.3 holds. One defines the functor ρ∗ stz : Mod(OXzar) −→ Mod(O t X sta) by (3.14) ρ∗ stz( • ) := O t X sta ⊗ρ−1 stz OXzar ρ−1 stz( • ). It follows from Theorem 3.4.7 that this functor is exact. We have the pairs of adjoint functors Mod(O t X sta) ρstz∗ / ρ∗ stz / Mod(OXzar). Assume Conjecture 1.3.3 and let X be a smooth algebraic variety over C. The stz is exact and fully faithful when restricted to Modcoh(OXzar). Its essential X sta) spanned by the X sta) the full subcategory of Mod(O t functor ρ∗ image is contained in Modlfp(O t O t X sta-modules locally of finite presentation. The only things that remains to prove in the above claim is that ρ∗ stz is fully faith- ful when restricted to Modcoh(OXzar). Under the assumption of Conjecture 1.3.3, this follows immediately from remark 3.4.2 and from the fact that any coherent sheaf on a smooth variety admits locally a finite free resolution. One shall notice that we do not ask X to be proper. Hence, this statement may be considered as a kind of weak Serre GAGA theorem in the non proper case. We would like to conclude this paper by two questions. Question 1: Is the sheaf O t X sta coherent? Question 2: Is the functor ρ∗ stz essentially surjective? We make the following straightforward observation concerning Question 2. If a X sta-module is locally of finite presentation on a cover formed of Zariski open subsets O t then it is in the essential image of ρ∗ stz. References [BM88] Edward Bierstone and Pierre D. Milman, Semianalytic and subanalytic sets, Inst. Hautes ´Etudes Sci. Publ. Math. 67 (1988), 5–42. 23 o o REFERENCES [Bjo74] Jan-Erik Bjork, On extensions of holomorphic functions satisfying a polynomial growth condi- tion on algebraic varieties in Cn, Ann. Inst. Fourier (Grenoble) 24 (1974), no. 4, vi, 157–165 (1975) (English, with French summary). [EP16] M´ario J. Emundo and Luca Prelli, Sheaves on T -topologies, J. Math. Soc. Japan 68 (2016), no. 1, 347–381. [Fri67] Jacques Frisch, Points de platitude d'un morphisme d'espaces analytiques complexes, Inven- tiones Math. 4 (1967), 118-138. [GS16] St´ephane Guillermou and Pierre Schapira, Construction of sheaves on the subanalytic site, Astrique, vol. 384, SMF, 2016. [Har67] Robin Hartshorne, Local cohomology, A seminar given by A. Grothendieck, Harvard Univer- sity, Fall, vol. 1961, Springer-Verlag, Berlin-New York, 1967. [Kas03] Masaki Kashiwara, D-modules and microlocal calculus, Translations of Mathematical Mono- graphs, vol. 217, American Mathematical Society, Providence, RI, 2003. [KS90] Masaki Kashiwara and Pierre Schapira, Sheaves on manifolds, Grundlehren der Mathematis- chen Wissenschaften, vol. 292, Springer-Verlag, Berlin, 1990. [KS96] , Moderate and formal cohomology associated with constructible sheaves, M´emoires, vol. 64, Soc. Math. France, 1996. [KS01] [KS16] 2016. , Ind-Sheaves, Ast´erisque, vol. 271, Soc. Math. France, 2001. , Regular and Irregular Holonomic D-Modules, Vol. 433, Cambridge University Press, [Loj59] Stanislaw Lojaciewicz, Sur le probl`eme de la division, Studia Math 8 (1959), 87-136. [Mal66] Bernard Malgrange, Ideals od differentiable functions, Tata Institute of Fundamental Re- search, Oxford University Press, 1966. [PY14] Mauro Porta and Tony Yue Yu, Higher analytic stacks and GAGA theorems, ArXiv e-prints (2014), available at https://arxiv.org/abs/1412.5166. [Pre08] Luca Prelli, Sheaves on subanalytic sites, Rend. Semin. Mat. Univ. Padova 120 (2008), 167– 216. [Rud67] Walter Rudin, A geometric criterion for algebraic varieties, J. Math. Mech. 17 (1967/1968), 671–683. [RW80] Kamil Rusek and Tadeusz Winiarski, Criteria for regularity of holomorphic mappings, Bull. Acad. Polon. Sci. S´er. Sci. Math. 28 (1980), no. 9-10, 471–475 (1981) (English, with Russian summary). [Ser55] Jean-Pierre Serre, G´eom´etrie alg´ebrique et g´eom´etrie analytique, Ann. Institut Fourier de Grenoble 6 (1955/1956), 1-42. [SGA1] Revetements ´etales et groupe fondamental (SGA 1), Documents Math´ematiques (Paris), 3, Soci´et´e Math´ematique de France, Paris, 2003. S´eminaire de g´eom´etrie alg´ebrique du Bois Marie 1960–61., Directed by A. Grothendieck, With two papers by M. Raynaud, Updated and annotated reprint of the 1971 original [Lecture Notes in Math., 224, Springer, Berlin. [Siu70] Yum-Tong Siu, Holomorphic functions of polynomial growth on bounded domains, Duke Math. Journal 37 (1970), 77-84. [Tay02] Joseph Taylor L, Several complex variables with connections to algebraic geometry and Lie groups, Graduates studies in Mathematics, vol. 46, Americam Mathematical Society, 2002. [TS16] The Stacks Project Authors, Stacks Project, 2016. Fran¸cois Petit, Mathematics Research Unit, University of Luxembourg email: [email protected] 24
1710.02471
5
1710
2019-01-03T16:21:45
Equivariant models of spherical varieties
[ "math.AG", "math.GR" ]
Let $G$ be a connected semisimple group over an algebraically closed field $k$ of characteristic 0. Let $Y=G/H$ be a spherical homogeneous space of $G$, and let $Y'$ be a spherical embedding of $Y$. Let $k_0$ be a subfield of $k$. Let $G_0$ be a $k_0$ -model ($k_0$-form) of $G$. We show that if $G_0$ is an inner form of a split group and if the subgroup $H$ of $G$ is spherically closed, then $Y$ admits a $G_0$-equivariant $k_0$-model. If we replace the assumption that $H$ is spherically closed by the stronger assumption that $H$ coincides with its normalizer in $G$, then $Y$ and $Y'$ admit compatible $G_0$-equivariant $k_0$-models, and these models are unique.
math.AG
math
EQUIVARIANT MODELS OF SPHERICAL VARIETIES MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI Abstract. Let G be a connected semisimple group over an algebraically closed field k of characteristic 0. Let Y = G/H be a spherical homogeneous space of G, and let Y ′ be a spherical embedding of Y . Let k0 be a subfield of k. Let G0 be a k0-model (k0-form) of G. We show that if G0 is an inner form of a split group and if the subgroup H of G is spherically closed, then Y admits a G0-equivariant k0-model. If we replace the assumption that H is spherically closed by the stronger assumption that H coincides with its normalizer in G, then Y and Y ′ admit compatible G0-equivariant k0-models, and these models are unique. Contents 0. Introduction 1. Semi-morphisms of k-schemes 2. Semi-morphisms of G-varieties 3. Quotients 4. Semi-morphisms of homogeneous spaces 5. Equivariant models of G-varieties 6. Spherical homogeneous spaces and their combinatorial invariants 7. Action of an automorphism of the base field 2 5 11 14 15 17 20 on the combinatorial invariants of a spherical homogeneous space 24 8. Equivariant models of automorphism-free spherical homogeneous spaces 9. Equivariant models of spherically closed spherical homogeneous spaces 10. Equivariant models of spherical embeddings of automorphism-free spherical homogeneous spaces Appendix A. Algebraically closed descent for spherical homogeneous spaces Appendix B. The action of the automorphism group on the colors of a spherical homogeneous space References 27 28 35 36 38 40 2010 Mathematics Subject Classification. 14M27, 14M17, 14G27, 20G15. Key words and phrases. Spherical variety, spherical homogeneous space, spherical embedding, color, model, form, semi-automorphism. This research was partially supported by the Hermann Minkowski Center for Geometry and by the Israel Science Foundation (grant No. 870/16). 1 2 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI 0. Introduction Let k be an algebraically closed field of characteristic 0, and let k0 ⊂ k be a subfield. Let X be a k-variety, that is, an algebraic variety over k. By a k0-model (k0-form) of X we mean a k0-variety X0 together with an isomorphism of k-varieties κX : X0 ×k0 k ∼ → X. Let G be a connected semisimple group over k. Let Y be a G-variety, that is, an algebraic variety over k together with a morphism θ : G ×k Y → Y defining an action of G on Y . We say that (Y, θ) is a G-k-variety or just that Y is a G-k-variety. Let G0 be a k0-model (k0-form) of G, that is, an algebraic group over k0 together with an isomorphism of algebraic k-groups κG : G0 ×k0 k ∼ → G. By a G0-equivariant k0-model of the G-k-variety (Y, θ) we mean a G0-k0-variety (Y0, θ0) ∼ → Y , that is, an isomorphism together with an isomorphism of G-k-varieties κY : Y0 ×k0 k of k-varieties κY such that the following diagram commutes: (1) G0,k ×k Y0,k θ0,k κG×κY Y0,k κY G ×k Y θ / Y where G0,k := G0 ×k0 k and Y0,k := Y0 ×k0 k. For a given k0-model G0 of G we ask whether there exists a G0-equivariant k0-model Y0 of Y . From now on till the end of the Introduction we assume that Y is a spherical homo- geneous space of G. This means that Y = G/H (with the natural action of G) for some algebraic subgroup H ⊂ G and that a Borel subgroup B of G has an open orbit in Y . Then the set of orbits of B in Y is finite; see, for example, Timashev [35, Section 25.1]. Let Y ֒→ Y ′ be a spherical embedding of Y = G/H. This means that Y ′ is a G-k- variety, that Y ′ is a normal variety, and that Y ′ contains Y as an open dense G-orbit. Then B has an open dense orbit in Y ′. Moreover the set of orbits of B (and hence, of G) in Y ′ is finite; see, for example, Timashev [35, Section 25.1]. Inspired by the works of Akhiezer and Cupit-Foutou [2], [1], [15], for a given k0-model G0 of G we ask whether there exist a G0-equivariant k0-model Y0 of Y and a G0-equivariant k0-model Y ′ 0 of Y ′. Since char k = 0, by a result of Alexeev and Brion [3, Theorem 3.1], see Knop's Math- Overflow answer [23] and Appendix A below, the spherical subgroup H of G is conjugate to some (spherical) subgroup defined over the algebraic closure of k0 in k. Therefore, from now on we assume that k is an algebraic closure of k0. We set Γ = Gal(k/k0) (the Galois group of k over k0). Let T be a maximal torus of G contained in a Borel subgroup B. We consider the Dynkin diagram Dyn(G) = Dyn(G, T, B). The k0-model G0 of G defines the so-called ∗- action of Γ = Gal(k/k0) on the Dynkin diagram Dyn(G); see Tits [36, Section 2.3, p. 39]. In other words, we obtain a homomorphism The k0-group G0 is called an inner form (of a split group) if the ∗-action is trivial, that is, if εγ = id for all γ ∈ Γ. For example, if G is a simple group of any of the types ε : Γ → Aut Dyn(G). / /     / EQUIVARIANT MODELS OF SPHERICAL VARIETIES 3 A1, Bn, Cn, E7, E8, F4, G2, then any k0-model G0 of G is an inner form, because in these cases Dyn(G) has no nontrivial automorphisms. If G0 is a split k0-group, then of course G0 is an inner form. Let D(Y ) denote the set of colors of Y = G/H, that is, the (finite) set of the closures of B-orbits of codimension one in Y . A spherical subgroup H ⊂ G is called spherically closed if the automorphism group AutG(Y ) = NG(H)/H acts on D = D(Y ) faithfully, that is, if the homomorphism AutG(Y ) → Aut(D) is injective. Here NG(H) denotes the normalizer of H in G, and Aut(D) denote the group of permutations of the finite set D. Example 0.1. Let k = C, G = PGL2,C, H = T (a maximal torus), Y = G/T . Then NG(T )/T = 2, and the spherical homogeneous space Y of G has exactly two colors, which are swapped by the non-unit element of NG(T )/T . We see that the subgroup H = T of G is spherically closed. Theorem 0.2. Let G be a connected semisimple group over an algebraically closed field k of characteristic 0. Let Y = G/H be a spherical homogeneous space of G. Let k0 be a subfield of k such that k is an algebraic closure of k0. Let G0 be a k0-model of G. Assume that: (i) G0 is an inner form, and (ii) H is spherically closed. Then Y admits a G0-equivariant k0-model Y0. Theorem 0.2 (which was inspired by Theorem 1.1 of Akhiezer [1] and by Corollary 1 of Cupit-Foutou [15, Section 2.5]), is a special case of the more general Theorem 9.2 below, where instead of assuming that G0 is an inner form, we assume only that for all γ ∈ Γ the automorphism εγ of Dyn(G) preserves the combinatorial invariants (Luna-Losev invariants) of the spherical homogeneous space Y . This assumption is necessary for the existence of a G0-equivariant k0-model of Y ; see Proposition 8.2 below. Remark 0.3. Necessary and sufficient conditions for the existence of G0-equivariant k0- model of Y = G/H were given by Moser-Jauslin and Terpereau [29, Theorem 3.18] in the case when k0 = R and H is a horospherical subgroup of G. Note that a horospherical subgroup H ⊂ G is not spherically closed unless it is parabolic, in which case NG(H) = H. The general case, when k0 is an arbitrary field of characteristic 0 and H is an arbitrary spherical subgroup of G, will be treated in the forthcoming article [11] of the author and G. Gagliardi. We have to assume that char k = 0 when dealing with spherical varieties, because we use Losev's uniqueness theorem [25, Theorem 1], which has been proved only in characteristic 0. Note that in general a G0-equivariant k0-model Y0 in Theorem 0.2 is not unique. The following theorem is a special case of the more general theorem 9.17 below. Theorem 0.4. In Theorem 0.2 the set of isomorphism classes of G0-equivariant k0-models of Y = G/H is naturally a principal homogeneous space of the finite abelian group H 1(Γ, AutG(Y )) ≃ Map(Ω(2), Hom(Γ, S2)) Here S2 is the symmetric group on two symbols, Ω(2) = Ω(2)(Y ) is the finite set defined in Section 6 below (before Definition 6.5), and Map(Ω(2), Hom(Γ, S2)) denotes the group of maps from the set Ω(2) to the abelian group Hom(Γ, S2). In particular, for k0 = R we have Hom(Γ, S2) ∼= S2, and therefore, the number of these isomorphism classes is 2s, where s = Ω(2). For G and Y as in Example 0.1 we have s = 1, 4 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI hence for each of the two R-models of G there are exactly two non-isomorphic equivariant R-models of Y ; see Example 9.19 below. Corollary 0.5 (Akhiezer's theorem). In Theorem 0.2, instead of (ii) assume that (ii′) H is self-normalizing, that is, NG(H) = H. Then Y = G/H admits a G0-equivariant k0-model Y0, and this model is unique up to a unique isomorphism. Indeed, since H is self-normalizing, we have AutG(Y ) = NG(H)/H = {1}, and hence, H is spherically closed. By Theorem 0.2, Y admits a G0-equivariant k0-model. The uniqueness assertion is obvious because AutG(Y ) = {1}. Corollary 0.5 generalizes Theorem 1.1 of Akhiezer [1], where the case k0 = R was considered. Theorem 0.6. Under the assumptions of Corollary 0.5, any spherical embedding Y ′ of Y = G/H admits a G0-equivariant k0-model Y ′ 0 is compatible with the unique G0-equivariant k0-model Y0 of Y from Corollary 0.5, and hence is unique up to a unique isomorphism. 0. This k0-model Y ′ Theorem 0.6 generalizes Theorem 1.2 of Akhiezer [1], who proved in the case k0 = R that the wonderful embedding of Y admits a unique G0-equivariant R-model. Our proof of Theorem 0.6 uses results of Huruguen [18]. Note that in Theorem 0.6 we do not assume that Y ′ is quasi-projective. Theorems 0.2, 0.4, and 0.6 seem to be new even in the case k0 = R. The plan of the rest of the article is as follows. In Sections 1 -- 5 we consider mod- els and semilinear morphisms for general G-varieties and homogeneous spaces of G, not necessarily spherical. In Sections 6 -- 7 we consider combinatorial invariants of spherical homogeneous spaces. Following ideas of Akhiezer [1, Theorem 1.1] and Cupit-Foutou [15, Theorem 3(1), Section 2.2], for γ ∈ Γ = Gal(k/k0) we give a criterion of isomorphism of a spherical homogeneous space Y = G/H and the "conjugated" variety γ∗Y = G/γ(H) in terms of the action of γ on the combinatorial invariants of G/H. In Sections 8 -- 10 we prove Corollary 0.5, Theorem 0.2, Theorem 0.4, and Theorem 0.6. In Appendix A, for a connected reductive group G0 defined over an algebraically closed field k0 of characteristic 0 and for an algebraically closed extension k ⊃ k0, it is proved that any spherical subgroup H of the base change G = G0 ×k0 k is conjugate to a (spherical) subgroup defined over k0. In Appendix B, following Friedrich Knop's MathOverflow answer [22] to the author's question, Giuliano Gagliardi gives a proof of an unpublished theorem of Ivan Losev de- scribing the image of AutG(G/H) = NG(H)/H in the group of permutations of D(G/H). Our proofs of Theorems 0.2, 0.4, and 0.6 use this result of Losev. Acknowledgements. The author is very grateful to Friedrich Knop for answering the author's numerous MathOverflow questions, especially for the answer [22], to Giuliano Gagliardi for writing Appendix B, and to Roman Avdeev for suggesting Example 9.12 and proving Proposition 9.13. It is a pleasure to thank Michel Brion for very helpful e-mail correspondence. The author thanks Dmitri Akhiezer, St´ephanie Cupit-Foutou, Cristian D. Gonz´alez-Avil´es, David Harari, Boris Kunyavskiı, and Stephan Snegirov for helpful discussions. The author thanks the referees for careful reading the article and very useful comments, which helped to improve the exposition. This article was written during the author's visits to the University of La Serena (Chile) and to the Paris-Sud University, and he is grateful to the departments of mathematics of these universities for support and excellent working conditions. Notation and assumptions. EQUIVARIANT MODELS OF SPHERICAL VARIETIES 5 k is a field. In Section 2 and everywhere starting Section 4, k is algebraically closed. Starting Section 6 we assume that char k = 0. k0 is a subfield of the algebraically closed field k such that k is a Galois extension of k0 (except for Appendix A), hence k0 is perfect. A k-variety is a geometrically reduced separated scheme of finite type over k, not nec- essarily irreducible. An algebraic k-group is a smooth k-group scheme of finite type over k, not necessarily connected. All algebraic k-subgroups are assumed to be smooth. Starting Section 4, all algebraic groups are assumed to be linear (affine). 1. Semi-morphisms of k-schemes 1.1. Let k be a field and let Spec k denote the spectrum of k. By a k-scheme we mean a pair (Y, pY ), where Y is a scheme and pY : Y → Spec k is a morphism of schemes. Let (Y, pY ) and (Z, pZ ) be two k-schemes. By a k-morphism, or a morphism of k-schemes, we mean a morphism of schemes λ : Y → Z such that the following diagram commutes: λ : (Y, pY ) → (Z, pZ ) Y pY Spec k λ id Z pZ / Spec k Let γ : k → k be an automorphism of k (we write γ ∈ Aut(k)). Let γ∗ := Spec γ : Spec k → Spec k denote the induced automorphism of Spec k; then (γγ′)∗ = (γ′)∗ ◦ γ∗. Let (Y, pY ) be a k-scheme. By abuse of notation we write just that Y is a k-scheme. We define the γ-conjugated k-scheme γ∗(Y, pY ) = (γ∗Y, γ∗pY ) to be the base change of (Y, pY ) from Spec k to Spec k via γ∗. By abuse of notation we write just γ∗Y for γ∗(Y, pY ). Lemma 1.2. Let (Y, pY ) be a k-scheme, and let γ ∈ Aut(k). Then the γ-conjugated k-scheme γ∗(Y, pY ) is canonically isomorphic to (Y, (γ∗)−1 ◦ pY ) Proof. Write (X, pX ) = γ∗(Y, pY ); then X comes with a canonical morphism λ : X → Y such that the following diagram commutes: X pX Spec k λ γ ∗ Y pY / Spec k Since (γ−1)∗(γ∗(Y, pY )) is canonically isomorphic to (Y, pY ), one can easily see that λ is an isomorphism of schemes. From the above diagram we obtain a commutative diagram X pX Spec k λ id Y pY Spec k (γ ∗)−1 / Spec k which gives a canonical isomorphism of k-schemes (X, pX ) ∼ → (Y, (γ∗)−1 ◦ pY ). (cid:3) / /     / / /     / / /       / 6 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI 1.3. We define an action of γ : k → k on k-points. Let y be a k-point of Y , that is, a morphism y : Spec k → Y such that pY ◦ y = idSpec k . We denote γ!(y) = y ◦ γ∗ : Spec k → Spec k → Y ; (2) then an easy calculation shows that γ!(y) is a k-point of γ∗Y , where we identify γ∗(Y, pY ) with (Y, (γ∗)−1 ◦ pY ). Thus we obtain a bijection (3) γ! : Y (k) → (γ∗Y )(k), y 7→ γ!(y). 1.4. Let G be a k-group scheme. Following Flicker, Scheiderer, and Sujatha [16, (1.2)], we define the k-group scheme γ∗G to be the base change of G from Spec k to Spec k via γ∗. Then the map (3) γ! : G(k) → (γ∗G)(k) is an isomorphism of abstract groups (because for any field extension λ : k ֒→ k′, the corresponding map on rational points λ! : G(k) → (G ×k k′)(k′) is a homomorphism). If H ⊂ G is a k-group subscheme, then γ∗H is naturally a k-group subscheme of γ∗G (because a base change of a group subscheme is a group subscheme). From the commutative diagram γ! γ! H(k) G(k) (γ∗H)(k) / (γ∗G)(k) we see that (γ∗H)(k) = γ!(H(k)) ⊂ (γ∗G)(k). Let (Y, θ) be a G-k-scheme (a G-scheme over k), where θ : G ×k Y → Y, is an action of G on Y . By abuse of notation we write just that Y is a G-k-scheme. We define the γ∗G-k-scheme γ∗(Y, θ) = (γ∗Y, γ∗θ) to be the base change of (Y, θ) from Spec k to Spec k via γ∗. Definition 1.5. Let (Y, pY ) and (Z, pZ ) be two k-schemes. A semilinear morphism (γ, ν) : (Y, pY ) → (Z, pZ ) is a pair (γ, ν), where γ : k → k is an automorphism of k and ν : Y → Z is a morphism of schemes, such that the following diagram commutes: Y pY ν Z pZ Spec k (γ ∗)−1 / Spec k We shorten "semilinear morphism" to "semi-morphism". We write "ν : (Y, pY ) → (Z, pZ ) is a γ-semi-morphism" if (γ, ν) : (Y, pY ) → (Z, pZ ) is a semi-morphism. Then by abuse of notation we write just that ν : Y → Z is a γ-semi-morphism. Note that if we take γ = idk, then a idk-semi-morphism (Y, pY ) → (Z, pZ ) is just a morphism of k-schemes. Lemma 1.6. If (γ, ν) : (Y, pY ) → (Z, pZ ) is a semi-morphism of nonempty k-schemes, then the morphism of schemes ν : Y → Z uniquely determines γ. / / _    _    / / /     / EQUIVARIANT MODELS OF SPHERICAL VARIETIES 7 Proof. We may and shall assume that Y and Z are affine, Y = Spec RY , Z = Spec RZ . Then we have a commutative diagram (4) RY ν ∗ RZ k γ−1 k Since k is a field, the vertical arrows are monomorphisms, and therefore, the homomor- phism of rings ν ∗ uniquely determines the automorphism γ−1. (cid:3) 1.7. We define an action of a semi-morphism (γ, ν) : (Y, pY ) → (Z, pZ ) on k-points. If y : Spec k → Y is a k-point of (Y, pY ), we set (5) (γ, ν)(y) = ν ◦ y ◦ γ∗ : Spec k → Z, which is a k-point of (Z, pZ ). This formula is compatible with the usual formula for the action of a k-morphism on k-points. By abuse of notation we write ν(y) instead of (γ, ν)(y). If (β, µ) : (Z, pZ ) → (W, pW ) is a semi-morphism of k-schemes, we set (β, µ) ◦ (γ, ν) = (βγ, µ ◦ ν). Then clearly (β, µ) ◦ (γ, ν) is a semi-morphism, and for every k-point y ∈ Y (k) we have (6) (µ ◦ ν)(y) = µ(ν(y)). Definition 1.8. By a γ-semi-isomorphism ν : (Y, pY ) → (Z, pZ ), where (Y, pY ) and (Z, pZ ) are two k-schemes, we mean a γ-semi-morphism ν : (Y, pY ) → (Z, pZ ) for which the morphisms of schemes ν : Y → Z is an isomorphism. By a γ-semi-automorphism of a k-scheme (Y, pY ) we mean a γ-semi-isomorphism µ : (Y, pY ) → (Y, pY ). 1.9. Let us fix γ ∈ Aut(k). Assume we have two k-schemes (Y, pY ) and (Z, pZ ). Let ν : Y → Z be a morphism of schemes. The diagram with commutative left-hand triangle (7) shows that Y pY ▲ ▲ ▲ ▲ Z pZ ν ▲ ▲ γ∗pY ▲ ▲ ▲ ▲ ▲ ▲ (γ ∗)−1 Spec k &▲ / Spec k (γ, ν) : (Y, pY ) → (Z, pZ ) is a semi-morphism, that is, the rectangle commutes, if and only if the right-hand triangle commutes, that is, if and only if (8) (idk, ν) : γ∗(Y, pY ) → (Z, pZ ) is a semi-morphism. In other words, ν : (Y, pY ) → (Z, pZ ) is a γ-semi-morphism if and only if ν : γ∗(Y, pY ) → (Z, pZ ) is a k-morphism. For brevity we write (9) ν♮ : γ∗Y → Z for the k-morphism (8); then the k-morphism ν♮ acts on k-points as follows: (10) (y′ : Spec k → γ∗Y ) 7−→ (ν ◦ y′ : Spec k → Z). o o O O O O o o / /   &   / 8 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI Example 1.10. Let (Y, pY ) be a k-scheme, and let γ ∈ Aut(k). Recall that γ∗(Y, pY ) = (Y, (γ∗)−1 ◦ pY ). The commutative diagram idY Y pY Y pY Spec k (γ ∗)−1 Spec k (γ ∗)−1 / Spec k shows that (γ, idY ) : Y → γ∗Y is a γ-semi-isomorphism. We denote this γ-semi-isomor- phism by γ! : Y → γ∗Y. Comparing formulas (2) and (5), we see that the γ-semi-isomorphism γ! : Y → γ∗Y acts on k-points as the bijective map γ! : Y (k) → (γ∗Y )(k) defined by formula (2). 1.11. Let ν : (Y, pY ) → (Z, pZ ) be a γ-semi-morphism. Then the diagram (7) commutes, and hence, ν = ν♮ ◦ γ! = (idk, ν) ◦ (γ, idY ) : Y γ!−−→ γ∗Y ν♮−−→ Z, where γ! is a γ-semi-isomorphism and ν♮ is a k-morphism (an idk-semi-morphism). follows that It (11) ν(y) = ν♮(γ!(y)) for y ∈ Y (k) (this can be also seen by comparing formulas (2), (5), and (10)). Example 1.12. Let Y0 be a k0-scheme, where k0 is a subfield of k. For simplicity, we assume that Y0 is affine, that is, Y0 = Spec R0 , where R0 is a k0-algebra. We set Then Y = Y0 ×k0 k := Y0 ×Spec k0 Spec k. Y = Spec R, where R = R0 ⊗k0 k. Let i : R0 ֒→ R denote the canonical embedding. Consider the sets of k-points Y0(k) = Homk0(R0, k) and Y (k) = Homk(R, k). We have a canonical morphism of schemes i∗ : Y → Y0 inducing a canonical map Y (k) → Y0(k), (y : Spec k → Y ) 7−→ (y0 = i∗ ◦ y : Spec k → Y → Y0), which in the language of rings can be written as (12) Y (k) → Y0(k) : (y : R → k) 7−→ (y0 = yR0 : R0 → k). The map (12) is bijective; the inverse map is given by Y0(k) → Y (k) : (y0 : R0 → k) 7−→ (r0 ⊗ λ 7→ y0(r0) · λ ∈ k) for r0 ∈ R0, λ ∈ k. Let γ ∈ Aut(k/k0), that is, γ is an automorphism of k that fixes all elements of k0. Consider µγ = idY0 × (γ∗)−1 : Y → Y. / /       / EQUIVARIANT MODELS OF SPHERICAL VARIETIES 9 It follows from the construction of µγ that the following diagram commutes: (13) Y0 i∗ Y pY ◆ ◆ ◆ ◆ ◆ ◆ ◆ γ∗pY ◆ ◆ y′ ◆ ◆ ◆ Spec k (γ ∗)−1 id µγ / Y0 i∗ Y y pY '◆ / Spec k We see that µγ is a γ-semi-automorphism of Y , and it induces an isomorphism of k-schemes (µγ)♮ : γ∗Y ∼ → Y ; see formula (9). By formula (10), (µγ)♮ takes a k-point y′ of γ∗Y to the k-point y = µγ ◦ y′ of Y . We see from the diagram (13) that i∗ ◦ y = i∗ ◦ µγ ◦ y′ = i∗ ◦ y′. This means that the isomorphism of k-schemes (µγ)♮ is compatible with the bijections Y (k) → Y0(k) and (γ∗Y )(k) → Y0(k). We identify the k-scheme γ∗Y with Y via (µγ)♮ . Then µγ = γ! : Y → γ∗Y = Y. Similar assertions are true when the k0-scheme Y0 is not assumed to be affine. If β, γ ∈ Aut(k/k0), then clearly (14) µβγ = µβ ◦ µγ . By (6) we obtain that for every y ∈ Y (k) we have (15) µβγ(y) = µβ(µγ(y)). Thus the group Aut(k/k0) acts on the set Y (k). Let Y be an affine k-variety, Y = Spec RY ; then RY is the ring of regular functions on Y . If f ∈ RY , then for any y ∈ Y (k) the value f (y) ∈ k is defined. Lemma 1.13. Let ν : (Y, pY ) → (Z, pZ ) be a γ-semi-isomorphism of affine k-varieties, where γ : k → k is an automorphism of k. Let Y = Spec RY , Z = Spec RZ , and let ν ∗ : RZ → RY denote the morphism of rings corresponding to ν. Let fZ ∈ RZ. Then (16) fZ(ν(y)) = γ((ν ∗fZ)(y)) for all y ∈ Y (k). Proof. The assumption that ν : Y → Z is a γ-semi-morphism means that the diagram (4) commutes. A k-point y ∈ Y (k) corresponds to a homomorphism of k-algebras ϕy : RY → k, and the following diagram commutes: RY ϕy ν ∗ RZ ϕν(y) k γ−1 k hence ϕν(y) = γ ◦ ϕy ◦ ν ∗. We set fY = ν ∗fZ ∈ RY ; then fY (y) = ϕy(fY ), and (16) means that (γ ◦ ϕy ◦ ν ∗)(fZ ) = γ(ϕy(ν ∗fZ)), which is obvious. (cid:3) / / /   O O '   O O / ` ` H H     o o o o 10 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI Now let ν : (Y, pY ) → (Z, pZ ) be a γ-semi-isomorphism of irreducible k-varieties, where γ : k → k is an automorphism of k. Then the isomorphism of schemes ν : Y → Z induces an isomorphism of the fields of rational functions ν∗ : K(Y ) → K(Z), f 7→ ν∗f. For any f ∈ K(Y ) and y ∈ Y (k), the value f (y) ∈ k ∪ {∞} of f at y is defined, where we write f (y) = ∞ if f is not regular at y. Corollary 1.14. Let ν : (Y, pY ) → (Z, pZ ) be a γ-semi-isomorphism of irreducible k- varieties, where γ : k → k is an automorphism of k. With the above notation we have (ν∗fY )(z) = γ(fY (ν −1(z))) for all fY ∈ K(Y ), z ∈ Z(k). Proof. We consider the isomorphism ν ∗ = ν −1 ∗ : K(Z) → K(Y ), fZ 7→ ν ∗fZ, where fZ ∈ K(Z). Set fZ = ν∗fY ; then fY = ν ∗fZ. We must prove that (16) holds. We may and shall assume that Y and Z are affine varieties, Y = Spec RY , Z = Spec RZ , fZ ∈ RZ , and that the morphism ν corresponds to a homomorphism of rings ν ∗ : RZ → RY . Now the corollary follows from Lemma 1.13. (cid:3) 1.15. (Classical language) In this subsection we describe the variety γ∗Y and the map γ! : Y (k) → (γ∗Y )(k) in the language of classical algebraic geometry. First, consider the k (k) = kn. Let k0 be the prime subfield of k, that k ; then An n-dimensional affine space An k = An is, the subfield generated by 1; then An ×k0 k. Let γ ∈ Aut(k) = Aut(k/k0); then k0 γ induces a γ-semi-automorphism µγ : An k ; see Example 1.12. As in Example 1.12, we identify γ∗An k using the k-isomorphism k with An k → An (µγ)♮ : γ∗An k ∼ → An k ; then µγ = γ! : An k (k). For i = 1, . . . , n, let fi denote the i-th coordinate k , which is a regular function. Since fi comes from a regular function on k → γ∗An k = An k , and µγ(x) = γ!(x) for x ∈ An function on An An k0 , we have (µγ)∗fi = fi, and by Lemma 1.13 we have for x ∈ An fi(µγ(x)) = γ(fi(x)) If we write x = (xi)n i=1 ∈ kn = An k (k) = kn. k (k), where xi = fi(x) ∈ k, then µγ(x) = γ(xi)n i=1 . Similarly, let Pn k denote the n-dimensional projective space over k; then Pn We denote by x = (x0 : x1 : · · · : xn) ∈ Pn x0, x1, . . . , xn. Then for γ ∈ Aut(k/k0), the γ-semi-automorphism µγ of Pn γ!(x) = (γ(x0) : γ(x1) : · · · : γ(xn)). ×k0 k. k (k) the k-point with homogeneous coordinates k takes x to k = Pn k0 Now let Y ⊂ An k be an affine variety (a closed subvariety of An k ). Let ι : Y ֒→ An k denote the inclusion morphism; then γ induces a k-morphism k = An k . γ∗ι : γ∗Y → γ∗An From the commutative diagram γ! Y (k) (γ∗Y )(k) ι γ∗ι An k (k) γ! / An k (k) we see that (γ∗ι)(y′) = γ!(ι(γ−1 ! (y′))) for y′ ∈ (γ∗Y )(k), / /     / EQUIVARIANT MODELS OF SPHERICAL VARIETIES 11 Now we assume that k is algebraically closed. As usual in classical algebraic geometry, we identify an affine variety Y ⊂ An k with the algebraic set Y (k) ⊂ An k (k) = kn. Furthermore, we identify γ∗Y with the algebraic set (γ∗ι)(γ∗Y (k)) = γ!(Y (k)) ⊂ kn . We see that and that the map γ∗Y = {γ(yi)n i=1 (yi)n i=1 ∈ Y }, i=1. If Y ⊂ kn sends a point y with coordinates (yi)n is defined by a family of polynomials (Pα)α∈A, then γ∗Y ⊂ kn is defined by the family (γ(Pα))α∈A , where γ(Pα) is the polynomial obtained from Pα by acting by γ on the coefficients. γ! : Y → γ∗Y i=1 to the point with coordinates γ(yi)n One can describe similarly γ∗Y when Y ⊂ Pn k is a projective or quasi-projective variety. Namely, γ∗Y = {γ!(y) y ∈ Y } ⊂ Pn k . 2. Semi-morphisms of G-varieties 2.1. In this section k is an algebraically closed field, and Y is a k-variety, that is, a reduced separated scheme of finite type over k. Let G be an algebraic group over k (we write also "an algebraic k-group"), that is, a smooth group scheme of finite type over k. Let (Y, θ) be a G-k-variety, that is, a k-variety Y together with an action θ : G ×k Y → Y of G on Y . If g ∈ G(k) and y ∈ Y (k), we write just g · Y y or g · y for θ(g, y) ∈ Y (k). Definition 2.2 (cf. [16, (1.2)]). Let γ ∈ Aut(k). A γ-semi-automorphism of an algebraic k-group G is a γ-semi-automorphism of k-schemes τ : G → G such that the corresponding isomorphism of k-varieties τ♮ : γ∗G → G, see (9), is an isomorphism of algebraic k-groups. This condition is the same as to require that certain diagrams containing τ commute; see [10, Section 1.2]. Let H ⊂ G be an algebraic k-subgroup. By Definition 2.2 we have τ (H(k)) = τ♮((γ∗H)(k)), where τ♮ : γ∗G → G is an isomorphism of algebraic k-groups. It follows that τ (H(k)) is the set of k-points of a k-subgroup of G, which we denote by τ (H). By definition, τ (H)(k) = τ (H(k)). 2.3. Let k0 ⊂ k be a subfield. We shall always assume that we are given a k0-model (G0, κG) of G; see the Introduction. Let γ ∈ Aut(k/k0), that is, γ be an automorphism of k fixing all elements of k0. Then we have γ-semi-automorphisms (17) compare Example 1.12, where G0,k = G0 ×k0 k and we obtain σγ from σ◦ isomorphism κG : G0,k → G. If β, γ ∈ Aut(k/k0), then by (14) we have γ = idG0 × (γ∗)−1 : G0,k → G0,k σ◦ and σγ : G → G, γ via the k- ∼ and by (15) we have (18) σβγ(g) = σβ(σγ(g)) for every g ∈ G(k). σβγ = σβ ◦ σγ , 12 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI If α is any k-automorphism of G, then τ := α ◦ σγ : G → G is a γ-semi-automorphism of G, and all γ-semi-automorphisms of G (for given γ) can be obtained in this way. If we assume that G = G0 ×k0 k, where G0 is an affine algebraic group contained in GLn,k0 and defined by polynomials (Pα) with coefficients in k0, then by Subsection 1.15 we have σγ(g) = γ(g) for every g ∈ G0(k), and clearly (19) σγ(g1 g2) = σγ(g1) · σγ(g2) for every g1, g2 ∈ G0(k). In general, for any k0-model G0 of G, since the multiplication law in G is "defined over k0", one can easily show that again (19) holds. By (18) the group Aut(k/k0) acts on the set G0(k) = G(k), and by (19) this action preserves the group structure in G(k). 2.4. Let G0 be as in 2.3 and let σγ be as in (17). If we have a G-k-variety (Y, θ) and a G0- equivariant k0-model (Y0, θ0) of (Y, θ) (see the Introduction), then for any γ ∈ Aut(k/k0) we obtain a γ-semi-automorphism µγ : Y → Y , compare Example 1.12 and Subsection 2.3. Since θ is "defined over k0", it is easy to see that µγ(g · y) = σγ(g) · µγ(y) for all g ∈ G(k), y ∈ Y (k). Definition 2.5. Let G be an algebraic k-group, and let (Y, θY ) and (Z, θZ ) be two G- k-varieties. Let γ ∈ Aut(k), and let τ : G → G be a γ-semi-automorphism of G. A τ -equivariant γ-semi-morphism is a γ-semi-morphism ν : Y → Z such that the following diagram commutes: ν : (Y, θY ) → (Z, θZ ) (20) G ×k Y θY τ ×ν Y ν G ×k Z θZ / / Z where we write τ × ν for the product of τ and ν over the automorphism (γ∗)−1 of Spec k. Since k is algebraically closed, G is smooth (reduced), Y and Z are reduced, we see that the diagram (20) commutes if and only if ν(g · y) = τ (g) · ν(y) for all g ∈ G(k), y ∈ Y (k). Construction 2.6. Let G be an algebraic k-group, and let (Y, θY ) be a G-k-variety. The group γ∗G naturally acts on γ∗Y : the action θ : G ×k Y → Y induces an action (21) γ∗θ : γ∗G ×k γ∗Y → γ∗Y. By definition, a γ-semi-automorphism τ of G defines an isomorphism of algebraic k-groups τ♮ : γ∗G → G. We identify G and γ∗G via τ♮ and obtain from (21) an action τ ∗γ∗θ : G ×k γ∗Y → γ∗Y, (g, y′) 7→ (γ∗θ)(τ −1 ♮ (g), y′) for g ∈ G(k), y′ ∈ (γ∗Y )(k). By abuse of notation, we write γ∗Y for the G-k-variety (γ∗Y, τ ∗γ∗θ). We write y′ g · τ for (τ ∗γ∗θ)(g, y′) = (γ∗θ)(τ −1 ♮ (g), y′), where g ∈ G(k), y′ ∈ (γ∗Y )(k). By formula (11) we have τ (g) = τ♮(γ!(g)), and hence, (22) g · τ y′ = (γ∗θ)(τ −1 ♮ (g), y′) = γ!(θ(γ−1 ! (τ −1 ♮ (g)), γ−1 ! (y′))) = γ!(θ(τ −1(g), γ−1 ! (y′))) = γ!(τ −1(g) · Y γ−1 ! (y′)). / /     EQUIVARIANT MODELS OF SPHERICAL VARIETIES 13 Lemma 2.7. Let G be an algebraic k-group, and let (Y, θ) be a G-k-variety. Let γ ∈ Aut(k), and let τ : G → G be a γ-semi-automorphism of G. Let y(0) ∈ Y (k) be a k-point, and write H = StabG(y(0)). Consider the action τ ∗γ∗θ : G ×k γ∗Y → γ∗Y. Then the stabilizer in G(k) of the point γ!(y(0)) ∈ (γ∗Y )(k) under the action τ ∗γ∗θ is τ (H(k)) = τ (H)(k). Proof. By formula (22) we have γ!(y(0)) = γ!(τ −1(g) · y(0)). g · τ Since the stabilizer in G(k) of y(0) ∈ Y (k) is H(k), the lemma follows. (cid:3) Note that the γ-semi-morphism ν in Definition 2.5 defines a k-morphism ν♮ : γ∗Y → Z; see (9). Lemma 2.8. Let γ ∈ Aut(k) and let τ : G → G be a γ-semi-automorphism of G. Let (Y, θY ) and (Z, θZ ) be two G-k-varieties. A morphism of schemes ν : Y → Z is a τ - equivariant γ-semi-morphism if and only if ν♮ : γ∗Y → Z is a G-equivariant morphism of k-varieties (where we write γ∗Y for (γ∗Y, τ ∗γ∗θY )). Proof. By (7) the morphism of schemes ν is a γ-semi-morphism Y → Z if and only if it is a k-morphism γ∗Y → Z. Let g ∈ G(k), y′ ∈ (γ∗Y )(k). Using formula (22) we obtain (23) We have also (24) ν♮(g · τ y′) = ν(γ−1 ! (g · τ y′)) = ν(τ −1(g) · Y γ−1 ! (y′)). g · Z ν(γ−1 ! (y′)) = g · Z ν♮(y′). If ν is τ -equivariant, then ν(τ −1(g) · Y γ−1 ! (y′)) = g · Z ν(γ−1 ! (y′)), and we obtain using (24) that (25) ν(τ −1(g) · Y γ−1 ! (y′)) = g · Z ν♮(y′). From (23) and (25) we obtain that ν♮(g · τ y′) = ν(τ −1(g) · Y γ−1 ! (y′)) = g · Z ν♮(y′) for all g ∈ G(k), y′ ∈ (γ∗Y )(k), hence ν♮ is G-equivariant. Conversely, if ν♮ is G-equivariant, then ν♮(g · τ y′) = g · Z ν♮(y′), and we obtain using (23) and (24) that (26) ν(τ −1(g) · Y γ−1 ! (y′)) = ν♮(g · τ y′) = g · Z ν♮(y′) = g · Z ν(γ−1 ! (y′)). Set g′ = τ −1(g) ∈ G(k), y = γ−1 ! (y′) ∈ Y (k). Then we obtain from (26) that ν(g′ · Y Thus ν is τ -equivariant. y) = τ (g′) · Z ν(y) for all g′ ∈ G(k), y ∈ Y (k). (cid:3) 14 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI Corollary 2.9. Let γ be an automorphism of k, and let τ : G → G be a γ-semi-automor- phism of G. Let (Y, θ) be a G-k-variety. There exists a τ -equivariant γ-semi-automorphism µ : Y → Y if and only if the G-k-variety (γ∗Y, τ ∗γ∗θ) is isomorphic to (Y, θ). Proof. We take Z = Y in Lemma 2.8. (cid:3) 3. Quotients 3.1. Let k be a field (not necessarily algebraically closed). By an algebraic scheme over k we mean a scheme of finite type over k. By an algebraic group scheme over k we mean a group scheme over k whose underlying scheme is of finite type over k. Let H be an algebraic group subscheme of an algebraic group k-scheme G. A quotient of G by H is an algebraic scheme Y over k equipped with an action θ : G ×k Y → Y and a point y(0) ∈ Y (k) fixed by H satisfying certain conditions (a) and (b); see Milne [28, Definition 5.20]. By [28, Theorem 5.28] there does exist a quotient of G by H. By [28, Proposition 5.22] this quotient (Y, θ, y(0)) has the following universal property: (U). Let Z be a k-scheme on which G acts, and let z(0) ∈ Z(k) be a point fixed by H. Then there exists a unique G-equivariant map Y → Z making the following diagram commute: g7→g·y(0) G ◆ ◆ ◆ ◆ ◆ ◆ ◆ ◆ g7→g·z(0) ◆ ◆ ◆ ◆ &◆ Y Z Clearly the universal property (U) uniquely determines the quotient up to a unique isomorphism, so we may take (U) as a definition of the quotient. 3.2. We return to our settings: k is an algebraically closed field. Let G be a linear algebraic k-group (a smooth affine group k-scheme of finite type over k) and H be a smooth algebraic k-subgroup of G. Under these assumptions, a classical construction of Chevalley (see, for instance, [28, Theorem 4.27]) shows that the quotient Y exists as a quasi-projective variety, see [28, Theorem 7.18]. Thus Y is a k-variety, and therefore, in the universal property (U) defining Y we may assume that Z is a k-variety. Since k is algebraically closed and H is smooth, the condition "fixed by H" is equivalent to "fixed by H(k)". Thus we arrive to the following definition of Springer: Definition 3.3 (cf. Springer [34, Section 5.5]). Let k be an algebraically closed field, and let G be a linear algebraic k-group. Let H ⊂ G be a (smooth) k-subgroup. A quotient of G by H is a pointed G-k-variety (Y, θ : G ×k Y → Y, y(0) ∈ Y (k) ) such that H(k) fixes y(0), with the following universal property: (U′). For any pointed G-k-variety (Z, θZ , z(0)) with the k-point z(0) ∈ Z(k) fixed by H(k), there exists a unique morphism of pointed G-k-varieties (Y, θ, y(0)) → (Z, θZ , z(0)). 3.4. For G and H as in Definition 3.3, let (Y, θ, y(0)) be a quotient of G by H. The action of G on Y induces a G-k-morphism (a morphism of G-k-varieties) (27) G → Y, g 7→ g · y(0) for g ∈ G, where G acts on itself by left translations. As usual, we write G/H for Y and g · H or gH for g · y(0), where g ∈ G(k). In particular, we write 1 · H for y(0). The G-equivariant morphism G/H = Y → Z of (U′) sends 1 · H ∈ (G/H)(k) to z(0), hence for every g ∈ G(k) it sends the k-point gH ∈ (G/H)(k) to g · z(0) ∈ Z(k). Thus the quotient G/H has the following universal property: / / &   EQUIVARIANT MODELS OF SPHERICAL VARIETIES 15 (U′′). For any pointed G-k-variety (Z, θZ , z(0)) such that the k-point z(0) is fixed by H(k), there exists a unique G-k-morphism G/H → Z sending gH to g · z(0) for every g ∈ G(k). By [28, Definition 5.20(a)] the morphism (27) induces an injective map G(k)/H(k) → (G/H)(k), g · H(k) 7→ gH. By [28, Proposition 5.25] the morphism (27) is faithfully flat, and therefore, since k is algebraically closed, we see that the induced map G(k)/H(k) → (G/H)(k) is surjective. We conclude that this map is bijective. Thus any k-point of G/H is of the form gH, where g ∈ G(k). 4. Semi-morphisms of homogeneous spaces Let k be an algebraically closed field. All algebraic k-groups are assumed to be linear and smooth, and all k-subgroups are assumed to be smooth. Lemma 4.1 (well-known). Let G be a linear algebraic k-group over an algebraically closed field k, and let H1, H2 be two k-subgroups. Then Y1 = G/H1 and Y2 = G/H2 are isomor- phic as G-k-varieties if and only if the subgroups H1 and H2 are conjugate. To be more precise, for a ∈ G(k) the following two assertions are equivalent: (i) There exists an isomorphism of G-k-varieties φa : G/H1 → G/H2 taking g · H1 to ga−1 · H2 for g ∈ G(k); (ii) H1 = a−1H2 a. Proof. (i)⇒(ii). Clearly StabG(k)(1 · H1) = H1(k) and StabG(k)(a−1 · H2) = a−1 · H2(k) · a. Since φa(1 · H1) = a−1 · H2, these stabilizers coincide, whence (ii). (ii)⇒(i). Set Y2 = G/H2, y(0) 2 ∈ Y2(k); then StabG(k)(y′) = a−1H2(k)a = H1(k), so by the property (U′′) of the quotient G/H1 there exists a unique morphism of G-varieties φa : G/H1 → G/H2 such that 2 = 1 · H2 ∈ Y2(k), y′ = a−1 · y(0) φa(g · H1) = g · a−1 · H2 for g ∈ G(k). Similarly, since the stabilizer in G(k) of a · H1 ∈ (G/H1)(k) unique morphism of G-varieties ψa : G/H2 → G/H1 such that is H2(k), there exists a ψa(g · H2) = g · a · H1 for g ∈ G(k). Clearly these two morphisms are mutually inverse, hence both φa and ψa are isomorphisms. (cid:3) Definition 4.2. Let G be a linear algebraic group over an algebraically closed field k. Let Y be a G-k-variety. We denote by AutG(Y ) the group of G-equivariant k-automorphisms of Y , that is, of k-automorphisms ψ : Y → Y such that ψ(g · y) = g · ψ(y) for g ∈ G, y ∈ Y. Corollary 4.3 (well-known). Let G be a linear algebraic group over an algebraically closed field k. Let Y be a homogeneous G-k-variety, that is, Y = G/H, where H is a k-subgroup of G. Set N = NG(H), the normalizer of H in G. For n ∈ N (k) we define a map on k-points n∗ : G/H → G/H, gH 7−→ gHn−1 = gn−1H for g ∈ G(k). Then (i) The map n∗ is induced by some automorphism φn ∈ AutG(G/H); 16 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI (ii) The map (28) φ : N (k) → AutG(G/H), n 7→ φn is a homomorphism inducing an isomorphism N (k)/H(k) ∼ → AutG(G/H). Proof. By assumption n−1Hn = H, and by Lemma 4.1 there exists an isomorphism φn : G/H → G/H such that φn(g · H) = gn−1 · H, which proves (i). Clearly the map φ of (28) is a homomorphism with kernel H(k). To prove (ii) it remains (cid:3) to show that φ is surjective, which is straightforward. Corollary 4.4. If NG(H) = H, then AutG(G/H) = {1}. (cid:3) 4.5. Let k be an algebraically closed field. Let G be a linear algebraic group over k. Let γ ∈ Aut(k). Let τ : G → G be a γ-semi-automorphism of G. Let H ⊂ G be a (smooth) k-subgroup. Set Y = G/H; then we have a morphism θ : G ×k Y → Y defining the action of G on Y . Furthermore, the variety Y has a k-point y(0) = 1 · H such that StabG(k)(y(0)) = H(k), and the group of k-points G(k) acts on Y (k) transitively. Consider the variety γ∗Y , the action γ∗θ : γ∗G ×k γ∗Y → γ∗Y of γ∗G on γ∗Y , and the k-point γ!(y(0)) ∈ (γ∗Y )(k). As in Construction 2.6 we obtain an action τ ∗γ∗θ : G ×k γ∗Y → γ∗Y. Lemma 4.6. Let k be an algebraically closed field, and let G be a linear algebraic group over k. Let γ ∈ Aut(k), and let τ : G → G be a γ-semi-automorphism of G. Let H ⊂ G be a smooth k-subgroup. Set Y = G/H and let θ : G ×k Y → Y denote the canonical action. Consider the map on k-points (29) (G/H)(k) → (G/τ (H))(k), g · H 7→ τ (g) · τ (H) for g ∈ G(k). Then the following assertions hold: (i) The pointed G-k-variety (γ∗Y, τ ∗γ∗θ, γ!(y(0))) is isomorphic to G/τ (H); (ii) the map (29) is induced by some γ-semi-isomorphism ν : G/H → G/τ (H). Proof. Consider the pointed G-k-variety (γ∗Y, τ ∗γ∗θ, γ!(y(0))). By Lemma 2.7, the sub- group τ (H(k)) = τ (H)(k) of G(k) fixes γ!(y(0)). Now let (Z, θZ , z(0)) be a pointed G-k-variety such that τ (H(k)) fixes z(0). Consider the pointed G-k-variety where the action ((γ−1)∗Z, (τ −1)∗(γ−1)∗θZ , (γ−1)! (z(0))), (τ −1)∗(γ−1)∗θZ : G ×k (γ−1)∗Z → (γ−1)∗Z (30) is defined as in Construction 2.6, but for the pair (γ−1, τ −1) instead of (γ, τ ). By Lemma 2.7 applied to (γ−1, τ −1), the group H(k) fixes (γ−1)! (z(0)) ∈ (γ−1)∗Z. For any morphism of pointed G-k-varieties (31) κ : (Y, θ, y(0)) → ((γ−1)∗Z, (τ −1)∗(γ−1)∗θZ, (γ−1)! (z(0))) we obtain a morphism of pointed G-k-varieties (32) γ∗κ : (γ∗Y, τ ∗γ∗θ, γ!(y(0))) → (Z, θZ , z(0)). We see that the map κ 7→ γ∗κ is a bijection between the set of morphisms as in (31) and the set of morphisms as in (32). Since Y = G/H and H(k) fixes (γ−1)! (z(0)) under the action (30), we conclude by the universal property (U′) for the quotient Y = G/H, that the EQUIVARIANT MODELS OF SPHERICAL VARIETIES 17 former set contains exactly one element. It follows that the latter set contains exactly one element, that is, the pointed G-k-variety (γ∗Y, τ ∗γ∗θ, γ!(y(0))) has the universal property (U′). This means that, by Definition 3.3, the pointed G-k-variety (γ∗Y, τ ∗γ∗θ, γ!(y(0))) is a quotient of G by τ (H), which proves (i). It follows that there exists an isomorphism of G-k-varieties (33) We set (34) λ : γ∗Y → G/τ (H) such that γ!(y(0)) 7−→ g · τ (H). g · τ ν = λ ◦ γ! : G/H = Y γ!−−→ γ∗Y λ−−→ G/τ (H), where γ! : Y → γ∗Y is the γ-semi-morphism of Example 1.10. Then we have ν♮ = λ. Since ν♮ is an isomorphism of G-k-varieties, by Lemma 2.8 ν is a τ -equivariant γ-semi- isomorphism. Since by (33) and (34) we have ν(1 · H) = ν(y(0)) = λ(γ!(y(0))) = λ(1 · τ γ!(y(0))) = 1 · τ (H) and since ν is τ -equivariant, we obtain that ν(g · H) = ν(g · (1 · H)) = τ (g) · ν(1 · H) = τ (g) · τ (H), which proves (ii). (cid:3) Corollary 4.7. Let G be a linear algebraic k-group and H ⊂ G be an algebraic k-subgroup. Set Y = G/H. Let γ ∈ Aut(k) and let τ : G → G be a γ-semi-automorphism of G. The following three conditions are equivalent: (i) There exists a τ -equivariant γ-semi-automorphism µ : Y → Y ; (ii) The G-k-variety G/τ (H) is isomorphic to G/H; (iii) The algebraic subgroup τ (H) ⊂ G is conjugate to H. Proof. By Corollary 2.9 there exists µ : Y → Y as in (i) if and only if the G-k-variety (γ∗Y, τ ∗γ∗θ) is isomorphic to (Y, θ). By construction (Y, θ) = G/H, and by Lemma 4.6, (γ∗Y, τ ∗γ∗θ) ∼= G/τ (H). Thus (i)⇔(ii). By Lemma 4.1 (ii)⇔(iii). (cid:3) Remark 4.8. Lemma 4.6 and Corollary 4.7 are the main results of Sections 1 -- 4. 5. Equivariant models of G-varieties 5.1. Let k be an algebraically closed field, and k0 ⊂ k be a subfield such that k is a Galois extension of k0, that is, k0 is a perfect field and k is an algebraic closure of k0. We write Γ = Gal(k/k0) := Aut(k/k0). Let Y be a k-variety. Let γ, γ′ ∈ Γ. If µ is a γ-semi-automorphism of Y , and µ′ is a γ′-semi-automorphism of Y , then µ ◦ µ′ is a γ ◦ γ′-semi-automorphism of Y and µ−1 is a γ−1-semi-automorphism of Y . We denote by SAutk/k0(Y ) or just by SAut(Y ) the group of all γ-semi-automorphisms µ of Y where γ runs over Γ = Gal(k/k0). A k0-model of Y is a k0-variety Y0 together with an isomorphism of k-varieties κY : Y0 ×k0 k ∼ → Y. Note that γ ∈ Γ defines a γ-semi-automorphism of Y0 ×k0 k idY0 × (γ∗)−1 : Y0 ×Spec k0 Spec k −→ Y0 ×Spec k0 Spec k and thus, via κY , a γ-semi-automorphism µγ of Y . We obtain a homomorphism Γ → SAut(Y ), γ 7→ µγ . Conversely: 18 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI Lemma 5.2 (Borel and Serre [8, Lemma 2.12]). Let k, k0, Γ, Y be as above. Assume that for every γ ∈ Γ we have a γ-semi-automorphism µγ of Y such that the following conditions are satified: (i) the map Γ → SAutk/k0(Y ), γ 7→ µγ, is a homomorphism, (ii) the restriction of this map to Gal(k/k1) for some finite Galois extension k1/k0 in k comes from a k1-model Y1 of Y , (iii) Y is quasi-projective. Then there exists a k0-model Y0 of Y that defines this homomorphism γ 7→ µγ. (cid:3) 5.3. Let k and k0 be as in 5.1, and let G be a linear algebraic group over k. We denote by SAutk/k0(G) or just by SAut(G) the group of all γ-semi-automorphisms σ of G where γ runs over Γ = Gal(k/k0). We assume that we are given a k0-model of G, that is, a linear algebraic group G0 ∼ over k0 together with an isomorphism of algebraic k-groups κG : G0 ×k0 k → G. For γ ∈ Γ, the automorphism (γ∗)−1 of Spec k induces a γ-semi-automorphism idG0 × (γ∗)−1 of G0 ×Spec k0 Spec k. We identify G with G0 ×Spec k0 Spec k via κG; then for any γ ∈ Γ we obtain a γ-semi-automorphism σγ : G → G. The map Γ → SAut(G), γ 7→ σγ is a homomorphism. We identify γ∗G with G using (σγ)♮ : γ∗G → G. For a G-k-variety (Y, θ) and for a given k0-model G0 of G, we ask whether there exists a G0-equivariant k0-model (Y0, θ0) of (Y, θ). Let (Y, θ) be a G-k-variety. We write g · y for θ(g, y). Recall (Definition 2.5) that a γ-semi-automorphism µ of Y is σγ-equivariant if the following diagram commutes: G ×k Y σγ ×µ G ×k Y θ θ Y µ / Y This is the same as to require that µ(g · y) = σγ(g) · µ(y) for all g ∈ G(k), y ∈ Y (k). A G0-equivariant k0-model (Y0, θ0) of (Y, θ) defines a homomorphism Γ → SAutk/k0(Y ), γ 7→ µγ , where for every γ ∈ Γ, the γ-semi-automorphism µγ of Y is σγ-equivariant. Conversely: Lemma 5.4. Let k, k0, Γ = Gal(k/k0), G, (Y, θ) be as in 5.3 and let G0 be a k0-model of G. Assume that for every γ ∈ Γ we have a γ-semi-automorphism µγ of Y such that the following conditions are satisfied: (i) the map Γ → SAutk/k0(Y ), γ 7→ µγ is a homomorphism, (ii) the restriction of this map to Gal(k/k1) for some finite Galois extension k1/k0 in k comes from a G1-equivariant k1-model Y1 of Y , where G1 = G0 ×k0 k1, (iii) Y is quasi-projective, (iv) for every γ ∈ Γ, the γ-semi-automorphism µγ is σγ-equivariant. Then there exists a G0-equivariant k0-model (Y0, θ0) of (Y, θ) that defines this homomor- phism γ 7→ µγ. Proof. By Lemma 5.2 the homomorphism Γ → SAut(Y ), γ 7→ µγ / /     / EQUIVARIANT MODELS OF SPHERICAL VARIETIES 19 defines a k0-model Y0 of Y . Using Galois descent for morphisms (see, for example, Jahnel [19, Proposition 2.8]) we obtain from condition (iv) that θ comes from some morphism θ0 : G0 ×k0 Y0 → Y0, and the k0-model (Y0, θ0) of (Y, θ) is G0-equivariant. (cid:3) Remark 5.5. If in Lemma 5.4 we do not assume that Y is quasi-projective, then we obtain a k0-model Y0 in the category of algebraic k0-spaces (see Wedhorn [37, Proposition 8.1]), but not necessarily in the category of k0-schemes (even when k = C and k0 = R, see Huruguen [18, Theorem 2.35]). We need a proposition. Proposition 5.6 (well-known; see, for example, EGA IV3 [17, Theorem 8.8.2]). Let k be an algebraically closed field. (i) For any k-variety X and any subfield k0 of k, there exists a k1-model X1 of X for some finitely generated extension k1 of k0 in k; (ii) If k1 is a subfield of algebraically closed field k, f : X → Y a morphism of k- varieties, and X1, Y1 are k1-models of X, Y , respectively, then there exists a finitely generated field extension k2 of k1 in k such that, if we set X2 = X1 ×k1 k2 and Y2 = Y1 ×k1 k2, then there exists a k2-morphism f2 : X2 → Y2 such that the triple (X2, Y2, f2) ×k2 k is isomorphic to (X, Y, f ). Proof (communicated by an anonymous MathOverflow user). (i) A variety X is a finite union of affine open subvarieties Xi. Since X is separated, the intersections Xi ∩ Xj are affine varieties; see, for example, Liu [24, Ch. 3, Proposition 3.6 on p. 100]. Now X can be reconstructed from the affine varieties Xi, Xi ∩ Xj and the morphisms of affine varieties Xi ∩ Xj → Xi. Obviously, this system is defined over a subfield of k finitely generated over k0. (ii) The graph of f is a closed subvariety of X × Y , and so is defined by an ideal in the structure sheaf of X × Y , which is obviously defined over a subfield of k finitely generated over k1. (cid:3) Corollary 5.7. Let G be a linear algebraic group over an algebraically closed field k, and let θ : G ×k Y → Y be an action of G on a k-variety Y . Let k0 be a subfield of k, and let G0 be a k0-model of G. Then there exists a finitely generated extension k2 of k0 in k, a k2-variety Y2, and a k2-action θ2 : G2 ×k2 Y2 → Y2 such that (Y2, θ2) is a k2-model of (Y, θ). Here G2 = G0 ×k0 k2. Proof. By Proposition 5.6(i) there exists a k1-model Y1 of the variety Y for some finitely generated extension k1 of k0 in k. We obtain a k1-model G1 ×k1 Y1 of G ×k Y , where G1 = G0 ×k0 k1. By Proposition 5.6(ii) the action θ can be defined over a finitely generated extension k2 of k1 in k. (cid:3) Lemma 5.8. Let k, k0, Γ = Gal(k/k0), G, (Y, θ) be as in 5.3, and let G0 be a k0-model of G. Assume that AutG(Y ) = {1}. Assume that for every γ ∈ Γ there exists a γ-semi- automorphism µγ of Y satisfying condition (iv) of Lemma 5.4. Then such µγ is unique, and the map γ 7→ µγ satisfies conditions (i) and (ii) of Lemma 5.4. γ another such γ-semi-automorphism, then µ−1 Proof. If µ′ γ = µγ. If β, γ ∈ Γ, then µ−1 µ′ γ 7→ µγ is a homomorphism, that is, condition (i) is satisfied. γ ∈ AutG(Y ) = {1}, hence βγ µβ µγ ∈ AutG(Y ) = {1}, hence µβγ = µβ µγ, hence the map γ µ′ By Corollary 5.7, there exists a finite field extension k1/k0 in k and a G1-equivariant k1-model (Y1, θ1) of (Y, θ), where G1 = G0 ×k0 k1. This k1-model defines a homomorphism γ 7→ µ′ γ : Gal(k/k1) → SAut(Y ) 20 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI such that µ′ automorphism is unique, we see that for all γ ∈ Gal(k/k1) we have µγ = µ′ the restriction of the map γ is σγ-equivariant for all γ ∈ Gal(k/k1). Since a σγ-equivariant γ-semi- γ, and hence, γ 7→ µγ : Γ → SAut(Y ) to Gal(k/k1) comes from the k1-model (Y1, θ1) of (Y, θ), that is, condition (ii) of Lemma 5.4 is satisfied. (cid:3) 6. Spherical homogeneous spaces and their combinatorial invariants Starting this section, k is an algebraically closed field of characteristic 0. Let G be a connected reductive k-group. We describe combinatorial invariants (invariants of Luna and Losev) of a spherical homogeneous space Y = G/H of G. 6.1. We start with combinatorial invariants of G. We fix T ⊂ B ⊂ G, where B is a Borel subgroup and T is a maximal torus. Let BRD(G) = BRD(G, T, B) denote the based root datum of G. We have BRD(G, T, B) = (X, X ∨, R, R∨, S, S ∨) where X = X∗(T ) := Hom(T, Gm,k) is the character group of T ; X ∨ = X∗(T ) := Hom(Gm,k, T ) is the cocharacter group of T ; R = R(G, T ) ⊂ X is the root system; R∨ ⊂ X ∨ is the coroot system; S = S(G, T, B) ⊂ R is the system of simple roots (the basis of R) defined by B; S ∨ ⊂ R∨ is the system of simple coroots. There is a canonical pairing X × X ∨ → Z, (χ, x) 7→ hχ, xi, and a canonical bijection α 7→ α∨ : R → R∨ such that S ∨ = {α∨ α ∈ S}. See Springer [33, Sections 1 and 2] for details. We consider also the Dynkin diagram Dyn(G) = Dyn(G, T, B), which is a graph with the set of vertices S. The edge between two simple roots α, β ∈ S is described in terms of the integers hα, β∨i and hβ, α∨i. We call a pair (T, B) as above a Borel pair. If (T ′, B ′) is another Borel pair, then by Theorem 11.1 and Theorem 10.6(4) in Borel's book [7], there exists g ∈ G(k) such that (35) g · T · g−1 = T ′, g · B · g−1 = B ′. This element g induces an isomorphism g∗ : BRD(G, T ′, B ′) ∼ → BRD(G, T, B). If g′ ∈ G(k) another element as in (35), then g = gt for some t ∈ T (k), and therefore, the isomorphism (g′)∗ : BRD(G, T ′, B ′) ∼ → BRD(G, T, B) coincides with g∗. Thus we may identify the based root datum BRD(G, T ′, B ′) with BRD(G, T, B) and write BRD(G) for BRD(G, T, B). We say that BRD(G) is the canonical based root datum of G. We see that the based root datum BRD(G) is an invariant of G. In particular, the character lattice X = X∗(T ) with the subset S ⊂ X is an invariant, and the Dynkin diagram Dyn(G) is an invariant. 6.2. We describe the combinatorial invariants of a homogeneous spherical G-variety Y = G/H. Let K(Y ) denote the field of rational functions of Y . The group G(k) acts on K(Y ) by (g · f )(y) = f (g−1 · y) for f ∈ K(Y ), g ∈ G(k), and y ∈ Y (k). EQUIVARIANT MODELS OF SPHERICAL VARIETIES 21 For χ ∈ X∗(B), let K(Y )(B) k-space of rational functions f ∈ K(X) such that χ denote the space of χ-eigenfunctions in K(Y ), that is, the b · f = χ(b) · f for all b ∈ B(k). Since B has an open dense orbit in Y , the k-dimension of K(Y )(B) X = X (Y ) ⊂ X∗(B) denote the set of characters χ of B such that K(Y )(B) X is a subgroup of X∗(B) called the weight lattice of Y . We set χ χ is at most 1. Let 6= {0}. Then V = V (Y ) = HomZ(X , Q). Let Val(K(Y )) denote the set of Q-valued valuations of the field K(Y ) that are trivial on k. The group G(k) naturally acts on K(Y ) and on Val(K(Y )). We shall consider the set ValB(K(Y )) of B(k)-invariant valuations and the set ValG(K(Y )) of G(k)-invariant valuations. We have a canonical map ρ : Val(K(Y )) → V, v 7→ (χ 7→ v(fχ)), where v ∈ Val(K(Y )), χ ∈ X , fχ ∈ K(Y )(B) 1.8]) that the restriction of ρ to ValG(K(Y )) is injective. We denote by χ , fχ 6= 0. It is known (see Knop [20, Corollary V = V(Y ) := ρ(ValG(K(Y ))) ⊂ V the image of ValG(K(Y )) in V . It is a cone in V called the valuation cone of Y ; see Perrin [30, Corollary 3.3.6]. Let D = D(Y ) denote the set of colors of Y , that is, the set of B-invariant prime divisors in Y . Each color D ∈ D defines a B-invariant valuation of K(Y ) that we denote by val(D). Thus we obtain a map By abuse of notation we denote ρ(val(D)) ∈ V by ρ(D). Thus we obtain a map val : D → Val(K(Y )). ρ : D → V, which in general is not injective (for example, it is not injective for G and Y as in Example 0.1). For D ∈ D, let StabG(D) denote the stabilizer of D ⊂ Y in G. Clearly StabG(D) ⊃ B, hence StabG(D) is a parabolic subgroup of G. For α ∈ S, let Pα ⊃ B denote the corresponding minimal parabolic subgroup of G containing B. Let ς(D) denote the set of α ∈ S for which Pα is not contained in StabG(D). We obtain a map ς : D → P(S), where P(S) denotes the set of all subsets of S. Lemma 6.3. Any fiber of the map ς has ≤ 2 elements. Proof. Let D ∈ D. Since Y is a homogeneous G-variety, clearly StabG(D) 6= G, hence ς(D) 6= ∅. We see that there exists α ∈ ς(D). Consider the set D(α) consisting of those D ∈ D for which α ∈ ς(D). By Proposition B.3 in Appendix B below we have D(α) ≤ 2, and the lemma follows. (cid:3) Consider the map ρ × ς : D → V × P(S). Corollary 6.4. Any fiber of the map ρ × ς has ≤ 2 elements. (cid:3) 22 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI Consider the subset Ω := im(ρ × ς) ⊂ V × P(S). Let Ω(2) (resp. Ω(1)) denote the subset of Ω consisting of the elements with two preimages (resp. with one preimage) in D. We obtain two subsets Ω(1), Ω(2) ⊂ V × P(S), and by Corollary 6.4 we have Ω = Ω(1) ⊔ Ω(2) (disjoint union). Definition 6.5. Let G be a connected reductive group over an algebraically closed field k of characteristic 0. Let Y = G/H be a spherical homogeneous space of G. By the combinatorial invariants of Y we mean X ⊂ X∗(B), V ⊂ V := HomZ(X , Q), and Ω(1), Ω(2) ⊂ V × P(S). 6.6. Let G be a connected reductive k-group. Let H1 ⊂ G be a spherical subgroup; then we set Y1 = G/H1. We consider the set of colors D(Y1) and the canonical maps ρ1 : D(Y1) → V (Y1), ς1 : D(Y1) → P(S). If H2 ⊂ G is another spherical subgroup, then we set Y2 = G/H2 and consider the set of colors D(Y2) and the canonical maps ρ2 : D(Y2) → V (Y2), ς2 : D(Y2) → P(S). Now assume that there exists a ∈ G(k) such that H2 = aH1a−1. Then we have an isomorphism of G-varieties of Lemma 4.1 φa : Y1 → Y2, g · H1 7→ ga−1 · H2 . It follows that X (Y1) = X (Y2), V (Y1) = V (Y2), V(Y1) = V(Y2). Moreover, the G- equivariant map φa : Y1 → Y2 induces a bijection satisfying (36) It follows that Conversely: (φa)∗ : D(Y1) → D(Y2) ρ2 ◦ (φa)∗ = ρ1, ς2 ◦ (φa)∗ = ς1 . Ω(1)(Y1) = Ω(1)(Y2) and Ω(2)(Y1) = Ω(2)(Y2). Proposition 6.7 (Losev's Uniqueness Theorem [25, Theorem 1]). Let G be a connected reductive group over an algebraically closed field k of characteristic 0. Let H1, H2 ⊂ G be two spherical subgroups, and let Y1 = G/H1 and Y2 = G/H2 be the corresponding spherical homogeneous spaces. If X (Y1) = X (Y2), V(Y1) = V(Y2), Ω(1)(Y1) = Ω(1)(Y2), and Ω(2)(Y1) = Ω(2)(Y2), then there exists a ∈ G(k) such that H2 = aH1a−1. 6.8. Consider the group AutG(Y ) = NG(H)/H (cf. Corollary 4.3). This group acts on D. We consider the surjective map (37) ζ = ρ × ς : D → Ω. By Corollary 6.4 each of the fibers of ζ has either one or two elements. We denote by AutΩ(D) the group of permutations π : D → D such that ζ ◦ π = ζ. From (36) we see that the group AutG(Y ), when acting on D, acts on the fibers of the map ζ, and so we obtain a homomorphism AutG(Y ) → AutΩ(D). Theorem 6.9 (Losev, unpublished). Let G be a connected reductive group over an alge- braically closed field k of characteristic 0. Let Y = G/H be a spherical homogeneous space of G. Then, with the above notation, the homomorphism AutG(Y ) → AutΩ(D). (38) is surjective. This theorem will be proved in Appendix B; see Theorem B.5. EQUIVARIANT MODELS OF SPHERICAL VARIETIES 23 Corollary 6.10 (Strong version of Losev's Uniqueness Theorem). Let G, H1, H2, Y1 = G/H1, Y2 = G/H2 be as in Proposition 6.7, in particular, X (Y1) = X (Y2), V(Y1) = V(Y2), Ω(1)(Y1) = Ω(1)(Y2), and Ω(2)(Y1) = Ω(2)(Y2). Let ϕ : D(Y1) → D(Y2) be any bijection satisfying ρ2 ◦ ϕ = ρ1, ς2 ◦ ϕ = ς1 (such a bijection exists because Ω(1)(Y1) = Ω(1)(Y2) and Ω(2)(Y1) = Ω(2)(Y2)). Then there exists a′ ∈ G(k) such that H2 = a′H1(a′)−1 and (φa′)∗ = ϕ : D(Y1) → D(Y2). Proof. By Proposition 6.7 there exists a ∈ G(k) such that H2 = aH1a−1. This element a defines a map satisfying (36). Set then ψ satisfies (φa)∗ : D(Y1) → D(Y2) ψ = (φa)−1 ∗ ◦ ϕ : D(Y1) → D(Y1); ρ1 ◦ ψ = ρ1, ς1 ◦ ψ = ς1 , hence ψ ∈ AutΩ D(Y1). By Theorem 6.9 there exists n ∈ NG(H1) such that We set a′ = an; then a′H1(a′)−1 = H2, φa′ = φa ◦ φn, and (φn)∗ = ψ : D(Y1) → D(Y1). (φa′ )∗ = (φa)∗ ◦ (φn)∗ = (φa)∗ ◦ ψ = ϕ. (cid:3) Corollary 6.11. In Theorem 6.9, if H is spherically closed, then the homomorphism (38) is an isomorphism. Proof. Indeed, by the definition of a spherically closed spherical subgroup, the homo- morphism (38) is injective, and by Theorem 6.9 it is surjective, hence it is bijective, as required. (cid:3) Note that AutΩ(D) = Yω∈Ω Aut(ζ −1(ω)), where Aut(ζ −1(ω)) is the group of permutations of the set ζ −1(ω). It is clear that for every ω ∈ Ω, the restriction homomorphism (39) is surjective. AutΩ(D) → Aut(ζ −1(ω)) Corollary 6.12. In Theorem 6.9, if NG(H) = H, then the surjective map ζ of (37) is bijective, hence D injects into V × P(S). Proof. It follows from Theorem 6.9 and the surjectivity of the homomorphism (39), that the group AutG(Y ) = NG(H)/H acts transitively on the fiber ζ −1(ω) for every ω ∈ Ω. Since by assumption NG(H)/H = {1}, we conclude that each fiber of ζ has exactly one element, hence ζ is bijective, as required. (cid:3) 24 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI 7. Action of an automorphism of the base field on the combinatorial invariants of a spherical homogeneous space 7.1. Let k be an algebraically closed field of characteristic 0, G be a connected reductive group over k, H1 ⊂ G be a spherical subgroup, and Y1 = G/H1 be the corresponding spherical homogeneous space. Let k0 ⊂ k be a subfield and let γ ∈ Aut(k/k0). We assume that "G is defined over k0", that is, we are given a k0-model G0 of G. Then we have a γ-semi-automorphism σγ of G; see Subsection 5.3. Set H2 = σγ(H1) ⊂ G and denote by Y2 := G/H2 the corresponding spherical homogeneous space. We wish to know whether the spherical homogeneous spaces Y1 and Y2 are isomorphic as G-varieties. For this end we compare their combinatorial invariants. We fix a Borel pair (T, B); then T ⊂ B ⊂ G. Consider σγ(T ) ⊂ σγ(B) ⊂ G. Then (σγ(T ), σγ(B)) is again a Borel pair and hence, there exists gγ ∈ G(k) such that gγ · σγ(T ) · g−1 γ = T, gγ · σγ(B) · g−1 γ = B. Set τ = inn(gγ) ◦ σγ : G → G; then τ is a γ-semi-automorphism of G, and (40) τ (B) = B, τ (T ) = T. Set H ′ and Y ′ 2 = τ (H1) ⊂ G and Y ′ 2 are isomorphic, and we wish to know whether Y1 and Y ′ 2. We have H ′ 2 = G/H ′ 2 = gγ · H2 · g−1 γ , so by Lemma 4.1 Y2 2 are isomorphic. By (40), τ naturally acts on the characters of T and B; we denote the corresponding automorphism by εγ. By definition (41) εγ(χ)(b) = γ(χ(τ −1(b))) for χ ∈ X∗(B), b ∈ B(k), and the same holds for the characters of T (recall that X∗(B) = X∗(T )). Since τ (B) = B, we see that εγ, when acting on X∗(T ), preserves the set of simple roots S = S(G, T, B) ⊂ X∗(T ). It is well known (see for example [12, Section 3.2 and Proposition 3.1(a)]) that the automorphism εγ does not depend on the choice of gγ and that the map ε : Aut(k/k0) → Aut(X∗(T ), S), γ 7→ εγ is a homomorphism. Since εγ acts on X∗(B) and on S, one can define εγ(X (Y1)), εγ(V(Y1)), εγ(Ω(1)(Y1)), εγ(Ω(2)(Y1)). Following Akhiezer [1], we compute the combinatorial invariants of the spherical homo- geneous space Y ′ 2. We define a map on k-points (42) Y1(k) → Y ′ 2(k), g · H1 7→ τ (g) · H ′ 2, where g ∈ G(k). Recall that H ′ that we have 2 = τ (H1). It is easy to see that the map ν is well-defined and bijective and (43) ν −1(g′ · y′ 2) = τ −1(g′) · ν −1(y′ 2) for g′ ∈ G(k), y′ 2 ∈ Y ′ 2(k). By Lemma 4.6 the map (42) is induced by some τ -equivariant γ-semi-isomorphism and so we obtain an isomorphism of the fields of rational functions ν∗ : K(Y1) → K(Y ′ 2). ν : Y1 → Y ′ 2 , Lemma 7.2. Let χ1 ∈ X∗(B) and assume that f1 ∈ K(Y1)(B) where χ2 = εγ(χ1). χ1 . Then ν∗f1 ∈ K(Y ′ 2)(B) χ2 , EQUIVARIANT MODELS OF SPHERICAL VARIETIES 25 χ1 , we have Proof. Since f1 ∈ K(Y1)(B) (44) We write f ′ 1.14 we have 2 = ν∗f1 ∈ K(Y ′ f1(b−1y1) = χ1(b) · f1(y1) for all y1 ∈ Y1(k), b ∈ B(k). 2). Since ν : Y1 → Y ′ 2 is a γ-semi-isomorphism, by Corollary f ′ 2(y′ (45) Note that τ −1 : G → G is a γ−1-semi-automorphism of G, and ν −1 : Y ′ 2 → Y1 is a τ −1- equivariant γ−1-semi-isomorphism. Moreover, τ −1(T ) = T and τ −1(B) = B. Using (45), (43), (44), and (41), we compute: 2(k). 2))) 2) = γ(f1(ν −1(y′ 2 ∈ Y ′ for y′ 2(b−1 · y′ f ′ 2) = γ(f1(ν −1(b−1 · y′ = γ(χ1(τ −1(b))) · γ(f1(ν −1(y′ 2))) = γ(f1((τ −1(b))−1 · ν −1(y′ 2))) = εγ(χ1)(b) · f ′ 2))) 2(y′ 2). Thus f ′ 2 ∈ K(Y ′ 2)(B) χ2 where χ2 = εγ(χ1), as required. (cid:3) Corollary 7.3. X (Y ′ 2) = εγ(X (Y1)). Proof. By Lemma 7.2 we have εγ(X (Y1)) ⊂ X (Y ′ (γ−1, τ −1, ν −1), we obtain that εγ−1(X (Y ′ X (Y ′ 2) = εγ(X (Y1)), as required. 2)) ⊂ X (Y1), hence X (Y ′ 2). Applying Lemma 7.2 to the triple 2) ⊂ εγ(X (Y1)). Thus (cid:3) Let v1 ∈ ValB(K(Y1)). We define ν∗v1 ∈ ValB(K(Y ′ for f ′ (ν∗v1)(f ′ 2) = v1(ν −1 ∗ (f ′ 2)) by 2 ∈ K(Y ′ 2)) 2). We consider the maps ρ1 : ValB(K(Y1)) → V (Y1) and ρ′ Lemma 7.4. For any v1 ∈ ValB(K(Y1)) we have 2 : ValB(K(Y ′ 2)) → V (Y ′ 2). ρ′ 2(ν∗v1) = εγ(ρ1(v1)). Proof. See Huruguen [18, Proposition 2.18]. Corollary 7.5. V(Y ′ 2) = εγ(V(Y1)). (cid:3) (cid:3) Let D1 ∈ D(Y1) be a color, that is, D1 is the closure of a B-orbit of codimension one in Y1. We set D′ 2 := ν(D1) ⊂ Y ′ 2 ∈ D(Y ′ 2). We also write ν∗D1 for ν(D1). 2; then D′ 2 ∈ D(Y ′ Let D1 ∈ D(Y1) and D′ 2(D′ 2) ∈ ValB(K(Y ′ val′ 2)) the corresponding B-invariant valuations. 2); then we denote by val1(D1) ∈ ValB(K(Y1)) and Lemma 7.6. For any D1 ∈ D(Y1) we have val′ 2(ν∗D1) = ν∗(val1(D1)). Proof. See Huruguen [18, Proposition 2.19]. (cid:3) Remark 7.7. Propositions 2.18 and 2.19 of Huruguen [18, Section 2.2] were proved in his article under certain additional assumptions. Namely, Huruguen assumes that k/k0 is a Galois extension, that the triple (G, Y, θ) has a k0-model (G0, Y0, θ0), and that Y0 has a k0-point y(0). Those assumptions are not used in his proof. By abuse of notation, if D1 ∈ D(Y1) and D′ 2). 2)) ∈ V (Y ′ 2) for ρ′ V (Y1) and ρ′ 2(val′ 2(D′ 2(D′ 2 ∈ D(Y ′ 2), we write ρ1(D1) for ρ1(val1(D1)) ∈ Corollary 7.8 (from Lemma 7.4 and Lemma 7.6). For any D1 ∈ D(Y1) we have ρ′ 2(ν∗D1) = εγ(ρ1(D1)). (cid:3) 26 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI Lemma 7.9. For any D1 ∈ D(Y1) we have: (i) StabG(ν∗D1) = τ (StabG(D1)); (ii) ς(ν∗D1) = εγ(ς(D1)). Proof. (i) follows from the fact that the map ν : Y1(k) → Y ′ follows from (i). 2(k) is τ -equivariant, and (ii) (cid:3) Corollary 7.10 (from Corollary 7.8 and Lemma 7.9). Ω(1)(Y ′ 2) = εγ(Ω(1)(Y1)) and Ω(2)(Y ′ 2) = εγ(Ω(2)(Y1)). (cid:3) Proposition 7.11. X (Y2) = εγ(X (Y1)), V(Y2) = εγ(V(Y1)), Ω(1)(Y2) = εγ(Ω(1)(Y1)), Ω(2)(Y2) = εγ(Ω(2)(Y1)). Proof. Since H ′ 2 and Y2 are isomorphic, hence they have the same combinatorial invariants, and the proposition follows from Corollaries 7.3, 7.5, and 7.10. (cid:3) 2 and H2 are conjugate, by Lemma 4.1 the G-varieties Y ′ Note that Proposition 7.11 generalizes Propositions 5.2, 5.3, and 5.4 of Akhiezer [1]. Namely, in the case when γ2 = 1, our Proposition 7.11 is equivalent to those results of Akhiezer. Our proofs are similar to his. Proposition 7.12. With the notation and assumptions of Subsection 7.1, the subgroups H1 and H2 = σγ(H1) are conjugate if and only if εγ preserves the combinatorial invariants of Y1, that is εγ(X (Y1)) = X (Y1), (46) εγ(V(Y1)) = V(Y1), εγ(Ω(1)(Y1)) = Ω(1)(Y1), εγ(Ω(2)(Y1)) = Ω(2)(Y1). Proposition 7.12 generalizes Theorem 3(1) of Cupit-Foutou [15], where the case k0 = R was considered. Proof. If H1 and H2 are conjugate, then by Lemma 4.1 the homogeneous spaces Y1 = G/H1 and Y2 = G/H2 are isomorphic as G-varieties and hence, they have the same combinatorial invariants, that is, (47) X (Y2) = X (Y1), V(Y2) = V(Y1), Ω(1)(Y2) = Ω(1)(Y1), Ω(2)(Y2) = Ω(2)(Y1), and (46) follows from Proposition 7.11. Conversely, if equalities (46) hold, then by Propo- sition 7.11 the equalities (47) hold, and by Proposition 6.7 (Losev's Uniqueness Theorem) the subgroups H1 and H2 are conjugate. (cid:3) Corollary 7.13. With the notation and assumptions of Subsection 7.1, there exists a σγ- equivariant γ-semi-automorphism µ : Y1 → Y1, if and only if εγ preserves the combinatorial invariants of Y1, that is, equalities (46) hold. Proof. By Corollary 4.7 there exists a σγ-equivariant γ-semi-automorphism µ of Y1 if and only if the subgroup σγ(H1) of G is conjugate to H1. Now the corollary follows from Proposition 7.12. (cid:3) EQUIVARIANT MODELS OF SPHERICAL VARIETIES 27 8. Equivariant models of automorphism-free spherical homogeneous spaces 8.1. Let k0 be a field of characteristic 0 and let k be a fixed algebraic closure of k0 with Galois group Γ = Gal(k/k0). Let G be a connected reductive group over k. Let T ⊂ B ⊂ G be as in Section 6. We consider the based root datum BRD(G) = BRD(G, T, B). Let G0 be a k0-model of G. For any γ ∈ Γ, this model defines a γ-semi-automorphism σγ : G → G, which induces an automorphism εγ ∈ Aut BRD(G); see Subsection 7.1. We obtain a homomorphism ε : Γ → Aut BRD(G), γ 7→ εγ . Let Y = G/H be a spherical homogeneous space of G. We consider the combinatorial invariants of Y : X = X (Y ) ⊂ X∗(T ), V = V(Y ) ⊂ HomZ(X , Q), Ω(1) = Ω(1)(Y ), Ω(2) = Ω(2)(Y ) ⊂ HomZ(X , Q) × P(S); see Section 6. Since εγ acts on BRD(G), we can define εγ(X ), εγ(V), εγ(Ω(1)), εγ(Ω(2)). Recall that Y = G/H. By Lemma 4.6(i) we have γ∗Y ∼= G/σγ(H). Proposition 8.2. If Y = G/H admits a G0-equivariant k0-model Y0, then for all γ ∈ Γ, the automorphism εγ preserves the combinatorial invariants of Y , that is (48) εγ(X ) = X , εγ(V) = V, εγ(Ω(1)) = Ω(1), εγ(Ω(2)) = Ω(2). Proposition 8.2 follows from formulas of Huruguen [18, Section 2.2]. For the reader's convenience we prove it here. Proof. A G0-equivariant k0-model Y0 of Y defines, for any γ ∈ Γ, a σγ-equivariant γ-semi- automorphism µγ of Y , hence an isomorphism of G-k-varieties (µγ)♮ : G/σγ (H) ∼= γ∗Y ∼ −−→ Y. We see that the G-varieties G/H and G/σγ(H) are isomorphic, hence they have the same combinatorial invariants. By Proposition 7.11 the combinatorial invariants of the spherical homogeneous space G/σγ(H) are εγ(X ), εγ(V), εγ(Ω(1)), εγ(Ω(2)), and (48) follows. (cid:3) The next theorem is a partial converse of Proposition 8.2. Theorem 8.3. Let k, k0, Γ, G, H, G0 be as in 8.1. Assume that: (i) For all γ ∈ Γ, εγ preserves the combinatorial invariants of Y = G/H, that is, equalities (48) hold, and (ii) NG(H) = H. Then Y admits a G0-equivariant k0-model Y0. This k0-model is unique up to a unique isomorphism. 28 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI Proof. Let γ ∈ Γ. Since εγ preserves the combinatorial invariants of Y , by Corollary 7.13 there exists a σγ-equivariant γ-semi-automorphism µγ : Y → Y. Thus condition (iv) of Lemma 5.4 is satisfied. Since NG(H) = H, by Corollary 4.4 we have AutG(Y ) = {1}, and by Lemma 5.8 conditions (i) and (ii) of Lemma 5.4 are satisfied. The variety Y = G/H is quasi-projective (see Springer [33, Corollary 5.5.6]) and hence, condition (iii) of Lemma 5.4 is satisfied. By Lemma 5.4 there exists a G0-equivariant k0-model Y0 of Y . Since AutG(Y ) = {1}, for any given γ ∈ Γ the γ-semi-automorphism µγ is unique, hence the model Y0 is unique up to a unique isomorphism. (cid:3) Recall that a k0-model G0 of a connected reductive k-group G is called an inner form (of a split group) if εγ = 1 for all γ ∈ Γ = Gal(k/k0). Lemma 8.4. Let k, k0, Γ, G, H, G0 be as in 8.1. Then each of the conditions below imply condition (i) of Theorem 8.3. (i) G0 is an inner form; (ii) G0 is absolutely simple (that is, G is simple) of one of the types A1, Bn, Cn, E7, E8, F4, G2; (iii) G0 is split. Proof. (i) If G0 is an inner form, then εγ = 1 for every γ ∈ Γ, hence condition (i) of Theorem 8.3 is clearly satisfied. (ii) In these cases the Dynkin diagram Dyn(G) has no nontrivial automorphisms, hence Γ acts trivially on Dyn(G) and on BRD(G). We see that (ii) implies (i). (iii) In this case clearly G0 is an inner form. (cid:3) Corollary 8.5. If NG(H) = H and at least one of the conditions (i -- iii) of Lemma 8.4 is satisfied, then Y admits a G0-equivariant k0-model, and this k0-model is unique. (cid:3) Remark 8.6. Assume that k = R and NG(H) = H. The assertion that if condition (iii) of Lemma 8.4 is satisfied, then Y has a unique G0-equivariant R-model Y0, is Theorem 4.12 of Akhiezer and Cupit-Foutou [2]. The similar assertion when only condition (i) of Lemma 8.4 is satisfied, is Theorem 1.1 of Akhiezer [1]. Our article is inspired by this result of Akhiezer, and our proof of Theorem 8.3 is similar to his proof. 9. Equivariant models of spherically closed spherical homogeneous spaces In this section we do not assume that NG(H) = H. 9.1. Let k, G, H, Y = G/H, T ⊂ B ⊂ G be as in Section 6, in particular, k is an algebraically closed field of characteristic 0. The group AutG(Y ) = NG(H)/H acts on Y and on the set D of colors of Y . Recall that a spherical homogeneous space Y = G/H is called spherically closed if NG(H)/H acts on D faithfully, that is, if the homomorphism AutG(Y ) → Aut(D) is injective. (Here Aut(D) denotes the group of permutations of the finite set D.) Let k0 ⊂ k be a subfield such that k is an algebraic closure of k0. Let G0 be a k0-model of G, and for γ ∈ Γ := Gal(k/k0) let σγ : G → G be the γ-semi-automorphism defined by G0. Let εγ : X∗(T ) → X∗(T ) be as in (41). EQUIVARIANT MODELS OF SPHERICAL VARIETIES 29 Theorem 9.2. Let k be an algebraically closed field of characteristic 0. Let G, H, Y = G/H be as in Section 6. Let G0 be a k0-model of G, where k0 ⊂ k is a subfield such that k is an algebraic closure of k0. Assume that (i) εγ preserves the combinatorial invariants of Y for all γ ∈ Γ, and (ii) Y is spherically closed. Then Y admits a G0-equivariant k0-model. Theorem 9.2 generalizes the existence assertion of Theorem 8.3. It was inspired by Corollary 1 of Cupit-Foutou [15, Section 2.5], where the case k0 = R was considered. In order to prove the theorem we need a few lemmas. Lemma 9.3. Let ζ : D → Ω be a mapping of nonempty finite sets. Let Γ be a group acting on Ω by a homomorphism s : Γ → Aut(Ω), γ 7→ sγ. Assume that for every γ ∈ Γ there exists a permutation mγ : D → D covering sγ, that is, such that the following diagram commutes: D ζ Ω mγ sγ D ζ / Ω Then there exists a homomorphism m′ : Γ → Aut(D) such that: (i) for every γ ∈ Γ the permutation m′ (ii) for every γ ∈ Γ we have m′ (iii) m′ γ = idD for all γ ∈ ker s. γ covers sγ; γ = aγ ◦ mγ, where aγ ∈ AutΩ(D); Proof. We may and shall assume that Γ acts transitively on Ω. Let ω, ω′ ∈ Ω; then there exists γ ∈ Γ such that sγ(ω) = ω′. By hypotheses there exists mγ ∈ Aut(D) covering sγ . Then mγ induces a bijection ζ −1(ω) → ζ −1(ω′), hence the cardinalities of ζ −1(ω) and ζ −1(ω′) are equal. We see that ω 7→ ζ −1(ω) is a constant function on Ω; we denote its value by n. For each ω ∈ Ω we fix some bijection between ζ −1(ω) and the set {1, . . . , n}; we denote the element of ζ −1(ω) ⊂ D corresponding to i ∈ {1, . . . , n} by d(i) ω . Then we construct m′ γ ∈ Aut(D) by setting m′ γ(d(i) ω ) = d(i) sγ (ω). Since s : γ 7→ sγ is a homomorphism, we see that m′ : γ 7→ m′ clearly m′ assertion (iii) holds by construction. γ covers sγ, which proves (i). Set aγ = m′ γ is a homomorphism, and γ ; then clearly (ii) holds, and the (cid:3) γ ◦ m−1 9.4. Write then the map ζ = ρ × ς : D → V × P(S), Ω = im ζ, sγ = εγΩ : Ω → Ω; is a homomorphism. Assume that for all γ ∈ Γ there exists a σγ-equivariant γ-semi- automorphism µ : Y → Y , that is, a γ-semi-automorphism satisfying Γ → Aut(Ω), γ 7→ sγ (49) µ(g · y) = σγ(g) · µ(y) for all g ∈ G(k), y ∈ Y (k). The following lemma is obvious: / /     / 30 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI Lemma 9.5. If γ, δ ∈ Aut(k/k0), µ is a σγ-equivariant γ-semi-automorphism, and ν is a σδ-equivariant δ-semi-automorphism, then µν is a σγδ-equivariant γδ-semi-automorphism and µ−1 is a σγ−1-equivariant γ−1-semi-automorphism. (cid:3) 9.6. Consider σγ(T ) ⊂ σγ(B) ⊂ G. There exists gγ ∈ G(k) such that if we set σ′ inn(gσ) ◦ σγ, then γ = (50) σ′ γ(T ) = T and σ′ γ(B) = B. Let µ : Y → Y be as in (49). We define a γ-semi-automorphism µ′ = gγ ◦ µ : Y → Y, Then for g ∈ G(k), y ∈ Y (k), we have y 7→ gγ · µ(y) for y ∈ Y (k). (51) µ′(g · y) = gγ · µ(g · y) = gγ · σγ(g) · µ(y) = (gγ · σγ(g) · g−1 γ ) · (gγ · µ(y)) = σ′ γ(g) · µ′(y). Let D ∈ D = D(Y ) be a color, this means that D is the closure of a codimension one γ(B) = B, it follows from (51) that the divisor µ′(D) in Y is the B-orbit in Y . Since σ′ closure of a codimension one B-orbit, that is, a color. We obtain a permutation (52) mµ : D → D, D 7→ µ′(D), covering sγ. Since gγ for which (50) holds is defined uniquely up to multiplication on the left by an element t ∈ T (k) ⊂ B(k), we see that mµ depends only on µ and does not depend on the choice of gγ. Lemma 9.7. The map µ 7→ mµ is a homomorphism: for γ, µ, δ, ν as in Lemma 9.5, we have mµ◦ν = mµ ◦ mν. Proof. A straightforward calculation. (cid:3) 9.8. Proof of Theorem 9.2. Let γ ∈ Γ. Since εγ preserves the combinatorial invariants of Y = G/H, by Corollary 7.13 there exists a σγ-equivariant γ-semi-automorphism µγ : Y → Y . Set mγ = mµγ ∈ Aut(D), see Subsection 9.6; then mγ covers sγ, where sγ ∈ Aut(Ω) is the restriction of εγ to Ω. By Lemma 9.3 there exists a homomorphism m′ : Γ → Aut(D), γ 7→ m′ γ such that for every γ ∈ Γ the permutation m′ have m′ γ ∈ Aut(D) covers sγ (property (i)) and we γ = aγ ◦ mγ, where aγ ∈ AutΩ(D) (property (ii)). By Theorem 6.9 there exists an automorphismeaγ ∈ AutG(Y ) inducing aγ on D. We set µ′ Then µ′ 5.4 is satisfied. By Lemma 9.7, µ′ γ is a σγ-equivariant γ-semi-automorphism of Y and hence, condition (iv) of Lemma γ =eaγ ◦ µγ . γ acts on D by aγ ◦ mγ = m′ γ. δ (m′ Let γ, δ ∈ Γ; then µ′ γ m′ m′ hence µ′ of Lemma 5.4 is satisfied. γ ◦ µ′ γδ)−1 = idD. Since Y is spherically closed, we conclude that µ′ γδ = µ′ γδ)−1 ∈ AutG(Y ) and by Lemma 9.7 it acts on D by γδ)−1 = idY , γ is a homomorphism and hence, condition (i) δ. Thus the map γ 7→ µ′ δ (µ′ δ (µ′ γ µ′ γ µ′ By Corollary 5.7, the G-variety Y admits a Gk2-equivariant k2-model Y2 over some finite Galois extension k2/k0 in k. Let Γ2 = Gal(k/k2), and for γ ∈ Γ2 let µ′′ γ denote the γ-semi-automorphism of Y defined by the k2-model Y2. After passing to a finite extension, we may assume that for γ ∈ Γ2 we have sγ = idΩ, and by property (iii) of Lemma 9.3 we γ = idD . Moreover, we may assume that for γ ∈ Γ2 the semi-automorphism µ′′ have m′ γ EQUIVARIANT MODELS OF SPHERICAL VARIETIES 31 γ)−1µ′ acts trivially on D. It follows that (µ′′ γ ∈ AutG(Y ). Since Y is spherically closed, we conclude that (µ′′ γ)−1µ′ γ = idY , and hence, µ′ γ for γ ∈ Γ2. We see that the homomorphism γ 7→ µ′ γ = µ′′ γ satisfies condition (ii) of Lemma 5.4. Note that Y = G/H is quasi-projective, that is, condition (iii) of Lemma 5.4 is satisfied as well. By Lemma 5.4 the homomorphism γ 7→ µ′ γ defines a G0-equivariant k0-model Y0 of Y , which completes the proof of the theorem. (cid:3) γ acts trivially on D, and clearly (µ′′ γ)−1µ′ Remark 9.9. Let k = C, k0 = R, Γ = Gal(C/R) = {1, γ}. In this case Theorem 9.2 means that if εγ corresponding to a real structure σγ : G → G preserves the combinatorial invariants of Y = G/H, and Y is spherically closed, then there exists an anti-holomorphic σγ-equivariant semi-automorphism µγ : Y → Y such that µ2 γ = 1. Note that in general it is not true that then for every anti-holomorphic σγ-equivariant semi-automorphism µγ : Y → Y we have µ2 γ = 1; see Example 9.10 below. Example 9.10. Let G1 = G2 = SO3,C , H1 = H2 = SO2,C , Y1 = G1/H1 , Y2 = G2/H2 , G = G1 ×C G2 , Y = Y1 ×C Y2 . Then Y is a spherically closed spherical homogeneous space of G. Let D1 = {D+ 1 , D− 1 } denote the set of colors of Y1, and let D2 = {D+ 2 } denote the set of colors of Y2; then the set of colors D of Y can be identified with D1 ⊔ D2 (disjoint union). Let G0 = RC/RSO3,C (the Weil restriction of scalars from C to R), which is a real model of G, and let H0 = RC/RSO2,C ⊂ G0. We set Y0 = G0/H0 = RC/RY . Let σγ : G → G and µγ : Y → Y correspond to the real models G0 and Y0, respectively. Then µγ acts on D by the permutation of order 2 2 , D− where (D+ 1 , D+ 2 ) and (D− 2 ) are transpositions. Consider the transposition 1 , D+ 2 ) · (D− 1 , D− 2 ) ∈ Aut(D), mµ = (D+ 1 , D− then the permutation a is induced by the nontrivial element a = (D+ 1 , D− 1 ) ∈ Aut(D); Set ea ∈ O2,C/SO2,C = AutG1(Y1) ⊂ AutG(Y ). µ′ γ =ea ◦ µγ : Y → Y, which is an anti-holomorphic σγ-equivariant semi-automorphism of Y . Since µ′ D as a ◦ mγ, which is a permutation of order 4, we conclude that (µ′ γ)2 6= 1. γ acts on 9.11. In Example 0.1 we considered a spherically closed spherical variety Y = G/H, where G = SL2,k and H = T , a maximal torus in G. In this case it is obvious that for any k0-model G0 of G there exists a G0-equivariant k0-model Y0 of Y = G/H. Indeed, there exists a maximal torus T0 ⊂ G0 defined over k0, and it is clear that Y0 := G0/T0 is a G0-equivariant k0-model of Y = G/T . In the following example we consider a spherically closed spherical subgroup that is not conjugate to a subgroup defined over k0. Example 9.12. Let k = C, k0 = R. Following a suggestion of Roman Avdeev, we take G = SO2n+1,C, where n ≥ 2, and we take for H a Borel subgroup of SO2n,C, where SO2n,C ⊂ SO2n+1,C = G. By Proposition 9.13 below, H is a spherically closed spherical subgroup of G and NG(H) 6= H. Take G0 = SO2n+1,R; then G0 is an anisotropic (compact) R-model of G. Since the Dynkin diagram Bn of G has no nontrivial automorphisms, G0 is an inner form. We wish to show that Y = G/H admits a G0-equivariant R-model. The subgroup H is not conjugate to any subgroup H0 of G0 defined over R because H is not reductive; see Lemma 9.14 below. We see that we cannot argue as in Subsection 9.11. Since NG(H) 6= H, we cannot apply Theorem 8.3 either. However, since H is spherically 32 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI closed, by Theorem 9.2 the homogeneous variety Y = G/H does admit a G0-equivariant R-model Y0. Proposition 9.13 (Roman Avdeev, private communication). Let G = SO2n+1,C, where n ≥ 2. Let H be a Borel subgroup of SO2n,C, where SO2n,C ⊂ SO2n+1,C = G. Then H is a spherically closed spherical subgroup of G, and NG(H) 6= H. Proof. Write g = Lie(G). Choose a Borel subgroup B ⊂ G and a maximal torus T ⊂ B. Let X = X∗(T ) denote the character lattice of T and let R = R(G, T ) ⊂ X be the root system. The Borel subgroup B defines a set of positive roots R+ ⊂ R and the corresponding set of simple roots S ⊂ R+ ⊂ R. Let U denote the unipotent radical of B and put u = Lie(U ). We have g = Lie(T ) ⊕Mβ∈R gβ , u = Mβ∈R+ gβ , where gβ is the root subspace corresponding to a root β. Let Rl ⊂ R denote the root subsystem consisting of the long roots. Observe that R is l = R+ ∩ Rl. We set a root system of type Bn, and Rl is a root system of type Dn. Set R+ gl = Lie(T ) ⊕ Mβ∈Rl gβ , ul = Mβ∈R+ l gβ , where the direct sums are taken over long roots. Let Gl (resp., Ul) be the connected algebraic subgroup of G with Lie algebra gl (resp., ul). Set H = T Ul. Then Gl ≃ SO2n,C and H is a Borel subgroup of Gl. It is well known that H is a spherical subgroup of G. For example, this fact follows from Theorem 1 of Avdeev [4] (to apply this theorem one needs to check that the short positive roots in R are linearly independent). By Avdeev [6, Proposition 5.25] H is spherically closed. We consider the Weyl group W = W (G, T ) = W (R). Let r ∈ W = NG(T )/T denote the reflection with respect to the short simple root, and let ρ be a representative of r in NG(T ). Since r preserves R+ l , we see that ρ ∈ NG(H). Since ρ /∈ T and NG(T )∩B = T , we see that ρ /∈ B. By construction H ⊂ B, and we conclude that ρ /∈ H, hence NG(H) 6= H. In fact, NG(H) = H ∪ ρH by Avdeev [5, Theorem 3]. (cid:3) Lemma 9.14 (well-known). Let k0 be a field of characteristic 0, and let G0 be a connected, reductive, anisotropic k0-group. Then any connected k0-subgroup H0 ⊂ G0 is reductive. Proof. For the sake of contradiction, assume that H0 is not reductive. Let U0 denote the unipotent radical of H0, which is a nontrivial unipotent k0-subgroup. Let u ∈ U0(k0) ⊂ H0(k0) ⊂ G0(k0) be a non-unit element of U0(k0). We see that G0(k0) contains a non-unit nilpotent element. On the other hand, by Borel -- Tits [9, Corollary 8.5], all k0-points of a connected, reductive, anisotropic k0-group are semisimple elements. Contradiction. (cid:3) The following example shows that G/H might have no G0-equivariant k0-model when H is not spherically closed. Example 9.15. Let k = C, k0 = R. Let n ≥ 1, G = Sp2n,C ×C Sp2n,C, Y = Sp2n,C, the group G acts on Y by (g1, g2) ∗ y = g1 y g−1 2 , g1, g2, y ∈ Sp2n(C). Let H denote the stabilizer in G of 1 ∈ Sp2n(C) = Y (C); then H = Sp2n,C embedded diagonally in G. We have Y = G/H, and Y is a spherical homogeneous space of G. We have NG(H) = Z(G)·H, where Z(G) denotes the center of G. It follows that NG(H)/H ≃ {±1} 6= {1}. Clearly NG(H)/H acts trivially on D(G/H), so H is not spherically closed. EQUIVARIANT MODELS OF SPHERICAL VARIETIES 33 Consider the following real model of G: G0 = Sp2n,R ×R Sp(n), where Sp(n) is the compact real form of Sp2n. We show that Y cannot have a G0- equivariant real model, although G0 is an inner form of a split group. Indeed, assume for the sake of contradiction that such a real model Y0 of Y exists. We have Y = Sp2n,C, and Y0 is simultaneously a principal homogeneous space of Sp2n,R and of Sp(n). Since H 1(R, Sp2n,R) = 1 (see Serre [32, III.1.2, Proposition 3]), we see that Y0(R) is not empty. It follows that the topological space Y0(R) is simultaneously a principal homogeneous space of Sp(2n, R) and of Sp(n). Thus Y0(R) is simultaneously homeomorphic to the noncompact Lie group Sp(2n, R) and to the compact Lie group Sp(n), which is clearly impossible. Thus, there is no G0-equivariant real model Y0 of Y . 9.16. Let k, k0, Γ, G, H, G0 be as in Subsection 8.1, in particular, k is an algebraically closed field of characteristic 0 and Γ = Gal(k/k0). We assume that H is spherically closed and that Y = G/H admits a G0-equivariant k0-model Y0. Then by Corollary 7.13, εγ preserves the combinatorial invariants of Y for all γ ∈ Γ, in particular, Γ acts on the finite set Ω(2) = Ω(2)(Y ). Let U1, U2, . . . , Ur be the orbits of Γ in Ω(2). For each i = 1, 2, . . . , r, let us choose a point ui ∈ Ui. Set Γi = StabΓ(ui). Theorem 9.17. With the notation and assumptions of 9.16 we have: (i) The set of isomorphism classes of G0-equivariant k0-models of Y is canonically a principal homogeneous space of the abelian group H 1(Γ, AutΩ(D)); i=1 Hom(Γi, S2), where S2 is the symmetric group on two (ii) H 1(Γ, AutΩ(D)) ≃ Qr symbols. Corollary 9.18. In Theorem 9.17 assume that Γ = 2. Then the number of isomorphism classes of G0-equivariant k0-models of Y = G/H is 2s, where s is the number of fixed points of Γ in Ω(2). Proof. Let Ui be an orbit of Γ in Ω(2). If Ui = 2, then Γi = {1}, hence Hom(Γi, S2) = 1. If Ui = 1, then Γi = Γ, and hence, Hom(Γi, S2) = 2. Now the corollary follows from the theorem. (cid:3) Proof of Theorem 9.17. Since Y is quasi-projective, there is a canonical bijection between the set of the isomorphism classes in the theorem and the pointed set H 1(Γ, AutG(Y )); see Serre [32], Proposition 5 in Section III.1.3. By Theorem 2 of Losev [25] (see also Subsection B.1 below), the group AutG(Y ) is abelian, hence H 1(Γ, AutG(Y )) is an abelian group. It follows that the set of isomorphism classes in the theorem is canonically a principal homogeneous space of this abelian group; see Serre [32], Proposition 35 bis and Remark (1) thereafter in Section I.5.3. By Corollary 6.11 there is a canonical isomorphism of abelian groups AutG(Y ) → AutΩ(D), and (i) follows. ∼ We compute H 1(Γ, AutΩ(D)). Recall that we have a surjective map ζ : D → Ω. Set D(2) = ζ −1(Ω(2)); then clearly AutΩ(D) = AutΩ(2)(D(2)) = Yω∈Ω(2) hence H 1(Γ, AutΩ(D)) = rYi=1 S2 , S2 = Yω∈Ui rYi=1 H 1Γ, Yω∈Ui S2 . 34 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI Since Γ acts on Ui transitively, the Γ-module Qω∈Ui S2 is induced by the Γi-module S2 (see Serre [32, Section I.2.5]) and hence, by the lemma of Faddeev and Shapiro, see Serre [32, I.2.5, Proposition 10], we have H 1Γ, Yω∈Ui S2 ≃ H 1(Γi, S2) = Hom(Γi, S2). Thus which proves (ii). H 1(Γ, AutΩ(D)) ≃ Hom(Γi, S2), rYi=1 (cid:3) Example 9.19. Let G = SO3,C ≃ PGL2,C. Consider the maximal torus H = SO2,C ⊂ SO3,C = G. Consider the affine quadric Y in A3 2 + x2 C given by the equation x2 1 + x2 3 = 1. The group G = SO3,C ⊂ GL(3, C) naturally acts in A3 C and preserves Y . The stabilizer in G of the point (0, 0, 1) ∈ Y (C) is H = SO2,C, and therefore, we may identify Y with G/H. The homogeneous space Y = G/H is a spherically closed spherical homogeneous space of G; see Example 0.1. We have AutG(G/H) = NG(H)/H = O2,C/SO2,C ≃ {±1}. Let G0 be an R-form of G; then there exists a maximal torus H0 in G0 (defined over R), and clearly G0/H0 is a G0-equivariant R-model of Y = G/H (so we do not have to refer to Theorem 9.2 in order to see that Y admits a G0-equivariant real model). Since H 1(Γ, AutG(G/H)) = Hom(Γ, {±1}) ≃ {±1}, the variety Y has exactly two G0-equivariant R-models. We describe these models for each R-model of G = SO3,C. There exist two non-isomorphic R-models of SO3,C : the split model SO2,1 and the compact (anisotropic) model SO3,R. Consider the indefinite real quadratic form in three variables F2,1(x1, x2, x3) = x2 1 + x2 2 − x2 3, xi ∈ R. Set G0 = SO(F2,1) = SO2,1. We consider the affine quadric Y + equation F2,1(x) = +1, and the affine quadric Y − −1. The group G0 = SO2,1 naturally acts on Y + the points (0, 0, i) ∈ Y + Y + 2,1 and Y − is a hyperboloid of one sheet, hence it is connected, while Y − sheets, hence it is not connected. non-isomorphic SO2,1-equivariant R-models of Y = G/H. R given by the R given by the equation F2,1(x) = 2,1, and the stabilizers in G0,C of 2,1(C) are both equal to SO2,C. We see that 2,1(R) 2,1(R) is a hyperboloid of two 2,1 are two 2,1 are SO2,1-equivariant R-models of Y = G/H. It is well known that Y + It follows that the R-varieties Y + 2,1(C) and (0, 0, 1) ∈ Y − 2,1 ⊂ A3 2,1 and Y − 2,1 and Y − 2,1 ⊂ A3 Now consider the positive definite real quadratic form in three variables F3(x1, x2, x3) = x2 1 + x2 2 + x2 3, xi ∈ R. Set G0 = SO(F3) = SO3,R. We consider the affine quadric Y + F3(x) = +1, and the affine quadric Y − group G0 = SO3,R naturally acts on Y + (0, 0, 1) ∈ Y + are SO3,R-equivariant R-models of Y = G/H. Clearly, Y + hence it is nonempty, while Y − are two non-isomorphic SO3,R-equivariant R-models of Y = G/H. R given by the equation R given by the equation F3(x) = −1. The 3 , and the stabilizers in G0,C of the points 3 and Y − 3 3 (R) is the unit sphere in R3, 3 and Y − 3 3 (R) is empty. It follows that the R-varieties Y + 3 (C) are both equal to SO2,C. We see that Y + 3 (C) and (0, 0, i) ∈ Y − 3 ⊂ A3 3 and Y − 3 ⊂ A3 EQUIVARIANT MODELS OF SPHERICAL VARIETIES 35 Remark 9.20. Let k = C, k0 = R, Γ = Gal(C/R) = {1, γ}. Let Y = G/H be a spherical homogeneous space of a connected reductive group G over C, and let G0 be a real model of G with the corresponding homomorphism σ : Γ → SAut(G), that is, with an anti-holomorphic involution σγ : G → G. Assume that there exists an anti-holomorphic σγ- equivariant semi-automorphism µγ : Y → Y (such µγ exists if and only if εγ corresponding to σγ preserves the combinatorial invariants of Y ). Corollary 1 in Section 2.5 of Cupit- Foutou [15] (which inspired this section of the present article) claims that if, moreover, G is semisimple and Y is spherically closed, then (a) any such µγ is involutive, that is, µ2 γ = 1, and (b) such involutive µγ is unique. Unfortunately, this corollary is erroneous. Example 9.10 disproves (a) (see Remark 9.9), and Example 9.19 disproves (b) (see also Theorem 9.17). However, it is true that there exists an involutive anti-holomorphic σγ-equivariant semi-automorphism µγ ; see Theorem 9.2. 10. Equivariant models of spherical embeddings of automorphism-free spherical homogeneous spaces In this section we assume that NG(H) = H. Theorem 10.1. Let k, G, H, Y = G/H, k0, Γ, G0 be as in Subsection 8.1, in particular, char k = 0. We assume that (i) G0 is an inner form of a split group, and (ii) NG(H) = H. Let Y ֒→ Y ′ be an arbitrary spherical embedding of Y . Then Y ′ admits a G0-equivariant k0-model Y ′ 0. This model is compatible with the k0-model Y0 of Y from Theorem 8.3 and is unique up to a canonical isomorphism. This theorem generalizes Theorem 1.2 of Akhiezer [1], who considered the case k0 = R. Note that Akhiezer considered only the wonderful embedding of Y , while we consider an arbitrary spherical embedding, so our result is new even in the case k0 = R. Proof. We show that a k0-model of Y ′, if exists, is unique. 0 be such a γ : Y ′ → Y ′ be the corresponding γ-semi- k0-model. For γ ∈ Γ := Gal(k/k0), automorphism of Y ′. Since Y is the only open G-orbit in Y ′, it is stable under µ′ γ for all γ ∈ Γ. Since k0 is a field of characteristic 0, hence perfect, this defines a G0-equivariant k0-model Y0 of Y , which is unique because NG(H) = H and hence, AutG(Y ) = {1}. Since Y is Zariski-dense in Y ′, we conclude that the model Y ′ Indeed, let Y ′ let µ′ 0 of Y ′ is unique. We prove the existence. By Theorem 8.3, Y admits a unique G0-equivariant k0-model Y0. The model Y0 defines an action of Γ on the finite set D; see, for example, Huruguen [18, 2.2.5]. Namely, for every γ ∈ Γ we have a σγ-equivariant γ-semi-automorphism µγ, which induces an automorphism mγ : D → D covering sγ : Ω → Ω; see (52). We show that this action of Γ on D is trivial. Indeed, since NG(H) = H, by Corollary 6.12 the surjective map ζ : D → Ω is bijective. Since by assumption G0 is an inner form, for all γ ∈ Γ we have εγ = 1, hence sγ = 1. Thus Γ acts trivially on Ω and on D. Let CF(Y ′) denote the colored fan of Y ′ (see Knop [20] or Perrin [30, Definition 3.1.9]) which is a set of colored cones (C, F) ∈ CF(Y ′), where C ⊂ V and F ⊂ D. We know that Γ acts trivially on V = HomZ(X , Q) and on D. It follows that for every γ ∈ Γ and for every colored cone (C, F) ∈ CF(Y ′), we have (53) γ∗(C) = C, γ∗(F) = F. 36 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI It follows that the colored fan CF(Y ′) is Γ-stable. Moreover, it follows from (53) that the hypothesis of Theorem 2.26 of Huruguen [18] is satisfied, that is, Y ′ has a covering by G-stable and Γ-stable open quasi-projective subvarieties. By that theorem Y ′ admits a G0-equivariant k0-model compatible with Y0. (cid:3) Remark 10.2. Huruguen [18] assumes that Y0 has a k0-point, but he does not use that assumption. Remark 10.3. In Theorem 10.1 we do not assume that Y ′ is quasi-projective. Appendix A. Algebraically closed descent for spherical homogeneous spaces The proofs in this appendix were communicated to the author by experts. Theorem A.1. Let G0 be a connected reductive group defined over an algebraically closed field k0 of characteristic 0. Let k ⊃ k0 be a larger algebraically closed field. We set G = G0 ×k0 k, the base change of G0 from k0 to k. Let H ⊂ G be a spherical subgroup of G (defined over k). Then H is conjugate to a (spherical) subgroup defined over k0. The proof is based on the following result of Alexeev and Brion [3]: Proposition A.2 ([3, Theorem 3.1]). Let G be a connected reductive group over an alge- braically closed field k of characteristic 0. For any G-scheme X of finite type, only finitely many conjugacy classes of spherical subgroups of G occur as isotropy groups of points of X. Proof of Theorem A.1. The theorem will be proved in five steps. 1) Let X0 be a variety equipped with an action of G0. Then X0 is the disjoint union of locally closed G0-stable subvarieties X m 0 consisting of all orbits of a fixed dimension m. Note that the dimension d(x) of the G0-orbit of x ∈ X0 is a lower semi-continuous function on X0 (see, for instance, Popov and Vinberg [31, Section 1.4]). This means that for every number ξ ∈ R, the subset {x ∈ X0 d(x) > ξ} is open in X0. It follows that the union of the G0-orbits in X0 of maximal dimension is an open subvariety. 2) Take for X0 the variety of Lie subalgebras of g0 = Lie G0 of a fixed codimension, say r, and let X be the k-variety obtained from X0 by scalar extension. Then X is the variety of Lie subalgebras of codimension r in g = Lie G. We write xh ∈ X for the point corresponding to a Lie subalgebra h ⊂ g, and we write hx ⊂ g for the Lie subalgebra corresponding to a point x ∈ X. The group G acts on X via the adjoint representation in g, and the stabilizer of a k-point xh in X is the normalizer N (h) of h in G. So the dimension of the orbit of xh is dim(G) − dim NG(h) = dim(G) − dim(h) − (dim NG(h) − dim(h)) = r − dim ng(h)/h, where ng(h) denotes the normalizer of h in g. Thus, if there exists h such that h = ng(h), then the Lie subalgebras h satisfying this property correspond to the k-points of the open subset consisting of the orbits of maximal dimension. Note that ng(h) is the Lie algebra of NG(h). So if h = ng(h), then h is an algebraic Lie algebra. 3) Let H be a spherical subgroup of G with Lie algebra h such that ng(h) = h. We claim that the orbit G · xh in X is open. First, note that the homogeneous G-variety G · xh is spherical because the stabilizer of xh in G is the subgroup NG(h) with Lie algebra ng(h) = h = Lie H. Hence for a suitable Borel subgroup B ⊂ G we have dim(B · xh) = dim(G · xh). Consider the open subset U = {x′ ∈ X : dim(B · x′) ≥ dim(B · xh)} ⊂ X. EQUIVARIANT MODELS OF SPHERICAL VARIETIES 37 Since by Step 2 the orbit G·xh has maximal dimension among the G-orbits in X, for every x′ ∈ U we have dim(B · x′) ≥ dim(B · xh) = dim(G · xh) ≥ dim(G · x′), hence dim(B · x′) = dim(G · x′) = dim(G · xh), that is, if we write h′ = hx′ , then NG(h′) ⊂ G is spherical and h′ = ng(h′). By Proposition A.2 (due to Alexeev and Brion), the set of conjugacy classes of spherical subgroups of the form NG(hx′) for x′ ∈ X is finite. Hence, since hx′ = ng(hx′) is the Lie algebra of NG(hx′) for every x′ ∈ U ⊂ X, the set G · U contains only finitely many G-orbits, which are all of the same (maximal) dimension. It follows that all these orbits are open; in particular, the orbit G · xh in X is open. 4) By Step 3, the Lie algebras h of spherical subgroups H of G such that ng(h) = h form finitely many G-orbits, and the closures of these orbits are irreducible components of the variety X, which is defined over k0. Since k0 is algebraically closed, it follows that every such orbit is defined over k0 and, moreover, every such G-orbit has a k0-point, which proves the theorem for spherical subgroups such that ng(h) = h. Also NG(h) = NG(H 0) = NG(H), where the latter equality follows from Corollary A.4 below. Thus if NG(H)/H is finite, then NG(h)/H 0 is finite, and hence, ng(h) = h. 5) To handle the case of an arbitrary spherical k-subgroup H of G, consider the spherical closure of H, that is, the algebraic subgroup H of NG(H) containing H such that H/H = ker [NG(H)/H → Aut D(G/H)] . By Corollary A.6 below, the spherical closure H is spherically closed, that is, NG(H)/H acts faithfully on the finite set of colors of G/H, hence the group NG(H)/H is finite, and therefore, ng(Lie H) = Lie H. By Step 4 we may assume that H is defined over k0. Now H is an intersection of kernels of characters of H (since the quotient H/H is diagonalizable) and every such character is defined over k0. Thus H is defined over k0, as required. An alternative proof, also based on Proposition A.2 due to Alexeev and Brion, sketched in Knop's MathOverflow answer [23]. is (cid:3) Lemma A.3. Let G be an abstract group and H ⊂ G a subgroup. Let H0 ⊂ H be a characteristic subgroup of H (this means that all automorphisms of H preserve H0). If the group NG(H0)/H0 is abelian, then NG(H) = NG(H0). Proof. Since H0 is characteristic, we have NG(H) ⊂ NG(H0). In particular, H0 is normal in H, hence H ⊂ NG(H0). Consider the inclusions of groups H0 ⊂ H ⊂ NG(H) ⊂ NG(H0) and the inclusions of the corresponding quotient groups H/H0 ⊂ NG(H)/H0 ⊂ NG(H0)/H0 . Since NG(H0)/H0 is abelian, the subgroup H/H0 is normal in NG(H0)/H0 , and hence, H is normal in NG(H0). We see that NG(H0) ⊂ NG(H) and thus NG(H0) = NG(H). (cid:3) Corollary A.4 (Brion and Pauer [14, Corollary 5.2]). Let G be a linear algebraic group over an algebraically closed field k of characteristic 0. Let H ⊂ G be a spherical subgroup, and let H 0 denote the identity component of H. Then NG(H) = NG(H 0). Proof. Clearly, H 0 is a characteristic subgroup of H. Since H is spherical, the subgroup H 0 is spherical as well, and therefore, NG(H 0)/H 0 is diagonalizable, hence abelian; see subsection B.1 below. Now the corollary follows from Lemma A.3. (cid:3) 38 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI From now on till the end of this appendix we follow Avdeev [6]. Let G be a connected reductive group over an algebraically closed field k of characteristic 0. Fix a finite covering group eG → G such that eG is a direct product of a torus with a simply connected semisimple group. For every simple eG-module V (a finite dimensional irreducible representation V of eG), the corresponding projective space P(V ) has a natural structure of a G-variety. Every G-variety arising in this way is said to be a simple projective G-space. Proposition A.5 (Bravi and Luna [13, Lemma 2.4.2]). (See also Avdeev [6, Corollary 3.24].) For any spherical subgroup H of a connected reductive group G over an algebraically closed field k of characteristic 0, the spherical closure H of H is the common stabilizer in G of all H-fixed points in all simple projective G-spaces. Corollary A.6 (well-known). Let H be a spherical subgroup of a connected reductive group G defined over an algebraically closed field k of characteristic 0. Let H ′ denote the spherical closure of H, and let H ′′ denote the spherical closure of H ′. Then H ′′ = H ′, that is, H ′ is spherically closed. This result was stated without proof in Section 6.1 of Luna [27] (see also Avdeev [6, Corollary 3.25]). Deduction of Corollary A.6 from Proposition A.5. Let P(V ) be a simple projective G-space. Let P(V )H denote the set of fixed points of H in P(V ). Since H ⊂ H ′, we have P(V )H ′ ⊃ P(V )H. Thus P(V )H ′ ⊂ P(V )H . By Proposition A.5 applied to H, we have P(V )H ′ = P(V )H . By Proposition A.5 applied to H ′, the group H ′′(k) is the set of g ∈ G(k) that fix P(V )H ′ for all simple projective G-spaces P(V ). By Proposition A.5 applied to H, the group H ′(k) is the set of g ∈ G(k) that fix P(V )H for all simple projective G-spaces P(V ). Since P(V )H ′ (cid:3) = P(V )H, we have H ′′ = H ′, as required. Appendix B. The action of the automorphism group on the colors of a spherical homogeneous space By Giuliano Gagliardi In this appendix we prove Theorem 6.9, which we restate below as Theorem B.5. Our proof is based on Friedrich Knop's MathOverflow answer [22] to Borovoi's question. Knop writes that Theorem B.5 was communicated to him by Ivan Losev. Let G be a connected reductive group over an algebraically closed field k of character- istic 0. Let Y = G/H be a spherical homogeneous space. B.1. We use the notation of Section 6. Let φ ∈ AutG(Y ) be a G-equivariant automorphism of Y . For every χ ∈ X , the automorphism φ preserves the one-dimensional subspace K(Y )(B) and thus acts on this space by multiplication by a scalar aφ,χ ∈ k×. It is easy to see that we obtain a homomorphism χ κ : AutG(Y ) → Hom(X , k×), φ 7→ (χ 7→ aφ,χ). The group Hom(X , k×) is naturally identified with the group of k-points of the k-torus with character group X . According to Knop [21, Theorem 5.5], the homomorphism κ is injective and its image is closed. B.2. We present results of Knop [21] and Losev [25] describing AutG(Y ). EQUIVARIANT MODELS OF SPHERICAL VARIETIES 39 The uniquely determined set Σ ⊂ X of linearly independent primitive elements γ of the lattice X such that V = \γ∈Σ {v ∈ V : hv, γi ≤ 0}. is called the set of spherical roots of Y . Since the image κ(AutG(Y )) ⊂ Hom(X , k×) is closed, this image corresponds to a sublattice Λ ⊂ X such that im κ = {φ ∈ Hom(X , k×) φ(χ) = 1 for all χ ∈ Λ ⊂ X }. According to Losev [25, Theorem 2], there exist integers (cγ )γ∈Σ equal to 1 or 2 such that each γ′ := cγ · γ ∈ X is a primitive element of the lattice Λ. The set ΣN = {cγ · γ}γ∈Σ ⊂ Λ generates the lattice Λ; see Knop [21, Corollary 6.5]. It follows that we have AutG(Y ) ∼= {ψ ∈ Hom(X , k×) : ψ(ΣN ) = {1}}. Losev has shown how the coefficients cγ can be computed from the combinatorial invariants of Y , but we shall only need the property recalled in Proposition B.4 below. For further details, we refer to Losev [25]. For α ∈ S, let D(α) denote the set of colors D ∈ D such that the parabolic subgroup Pα moves D, that is, α ∈ ς(D). We need the following results of Luna [26], [27]: Proposition B.3. Let α ∈ S. (1) We have D(α) ≤ 2. Moreover, D(α) = 2 if and only if α ∈ Σ ∩ S. (2) Assume D(α) = 2 and write D(α) = {D+ α ) = ρ(D− α ), χi = 1 α , D− 2 hα∨, χi for all χ ∈ X , where α∨ ∈ X∗(T ) (i) we have hρ(D+ α ), χi = hρ(D− α }. If ρ(D+ α ), then: is the corresponding simple coroot; (ii) we have ς(D+ α ) = ς(D− α ) = {α}. Proof. For (1), see Luna [26, Sections 2.6 and 2.7] or Timashev [35, Section 30.10]. For (2), we use that [27, Theorem 2] or [35, Theorem 30.22] implies that the invariants of a spherical homogeneous space satisfy the axioms of a homogeneous spherical datum. These axioms are stated in [27, Sections 2.1 and 2.2] and [35, Definition 30.21]. In particular, we have ρ(D+ α ), βi = 1. With the assumption ρ(D+ (cid:3) α ) = α∨X and for every β ∈ ς(D± α ) we have β ∈ X and hρ(D± α ), we obtain (i) and then (ii). α ) = ρ(D− α ) + ρ(D− We need the following result of Losev: Proposition B.4. If α ∈ Σ ∩ S and hρ(D+ then 2α ∈ ΣN (hence α /∈ ΣN ). α ), χi = hρ(D− α ), χi = 1 2 hα∨, χi for all χ ∈ X , Proof. See Losev [25, Theorem 2 and Definition 4.1.1(1)]. (cid:3) The following theorem is the main result of this appendix: Theorem B.5 (Losev, unpublished). The homomorphism AutG(Y ) → AutΩ(D) is surjective. α ). By Proposition B.3, for every α ∈ A we have ς(D+ Proof. Let A denote the set of simple roots α ∈ S such that D(α) = 2 and ρ(D+ ρ(D− map α 7→ {D+ such that (ρ × ς)(D+ α } is a bijection between A and the set of unordered pairs {D+ α ) = α ) = {α}, hence the α , D− α } α ). Note that there is a canonical bijection α ) = (ρ × ς)(D− α ) = ς(D− α , D− A → Ω(2), α 7→ (ρ × ς)(D+ α ). 40 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI By Proposition B.3(1), for every α ∈ A we have α ∈ S ∩ Σ ⊂ X (because Σ ⊂ X ), and α with fα 6= 0. Moreover, from Propositions B.3 and B.4 hence, there exists fα ∈ K(Y )(B) we obtain that 2α ∈ ΣN (and α /∈ ΣN ). D+ We want to show that for every α ∈ A there exists φα ∈ AutG(Y ) such that φα swaps α and D− β for β ∈ A, β 6= α. We have a natural homomorphism of algebraic k-tori α , but fixes all D+ β and D− Hom(X , k×) → Map(Σ, k×), ψ 7→ ψΣ , where Map(Σ, k×) denotes the group of maps Σ → k×. Since the set Σ ⊂ X is linearly independent, this homomorphism is surjective. It follows easily that any element of finite order in the group Map(Σ, k×) can be lifted to an element of finite order in the group Hom(X , k×). Now let α ∈ A. By the previous paragraph, there exists a homomorphism ψα : X → k× with ψα(α) = −1 and ψα(γ) = 1 for every γ ∈ Σ r {α}, and such that ψα is of finite order in the group Hom(X , k×). Then we have ψα(ΣN ) = {1}. By B.2 there exists an automorphism of finite order φα ∈ AutG(Y ) with (54) φα(fβ) =(−fβ fβ for β = α, for β ∈ A r {α}, where fβ ∈ K(Y )(B) β . Let eH ⊂ NG(H) denote the subgroup containing H such that eH/H = hφαi ⊂ NG(H)/H = AutG(Y ), same notation for the combinatorial objects associated to the spherical homogeneous space where hφαi denotes the finite subgroup generated by φα. We set eY = G/eH. We use the eY as for Y , but with a tilde above the respective symbol. The morphism of G-varieties Y → eY induces an embedding K(eY ) ֒→ K(Y ), and K(eY ) is the fixed subfield of φα. Since K(eY ) is a G-invariant subfield of K(Y ), we have eX ⊂ X . By (54) we have φα(fα) = −fα 6= fα . We see that fα /∈ K(eY ), hence α ∈ X r eX ; in particular α /∈eΣ. By Proposition B.3(1) we have eD(α) ≤ 1, hence the two colors in D(α) are mapped to one color by the map Y → eY , that is, φα swaps D+ fβ ∈ K(eY ) and β ∈ eX . Since β is a primitive element of X , it is a primitive element of eX ⊂ X . The natural map V → eV induced by Y 7→ eY is bijective and identifies V and eV (see Knop [20, Section 4]). Since β ∈ Σ is dual to a wall of −V, it is dual to a wall of −eV = −V. It follows that β ∈ S ∩eΣ; hence eD(β) = 2, and the two colors in D(β) are mapped to distinct colors under Y → eY , that is, φα fixes D+ On the other hand, for every β ∈ A r {α}, by (54) we have φα(fβ) = fβ, hence β and D− β . α and D− α . (cid:3) References [1] D. Akhiezer, Satake diagrams and real structures on spherical varieties, Internat. J. Math. 26 (2015), no. 12, Art. ID 1550103, 13 pp. [2] D. Akhiezer, S. Cupit-Foutou, On the canonical real structure on wonderful varieties, J. reine angew. Math. 693 (2014), 231 -- 244. [3] V. Alexeev, M. Brion, Moduli of affine schemes with reductive group action, J. Algebraic Geom. 14 (2005), 83 -- 117. [4] R. Avdeev, On solvable spherical subgroups of semisimple algebraic groups, Trans. Moscow Math. Soc. 2011, 1 -- 44. [5] R. Avdeev, Normalizers of solvable spherical subgroups, Math. Notes 94 (2013), 20 -- 31. [6] R. Avdeev, Strongly solvable spherical subgroups and their combinatorial invariants, Selecta Math. (N.S.) 21 (2015), 931 -- 993. [7] A. Borel, Linear Algebraic Groups, Second edition, Graduate Texts in Mathematics, Vol. 126, Springer- Verlag, New York, 1991. EQUIVARIANT MODELS OF SPHERICAL VARIETIES 41 [8] A. Borel, J.-P. Serre, Th´eor`emes de finitude en cohomologie galoisienne, Comm. Math. Helv. 39 (1964), 111 -- 164. [9] A. Borel, J. Tits, Groupes r´eductifs, Publ. Math. Inst. Hautes ´Etudes Sci. 27 (1965), 55 -- 150. [10] M. Borovoi, Abelianization of the second nonabelian Galois cohomology, Duke Math. J. 72 (1993), 217 -- 239. [11] M. Borovoi, G. Gagliardi, Existence of equivariant models of spherical homogeneous spaces and other G-varieties, arXiv:1810.08960[math.AG]. [12] M. Borovoi, B. Kunyavskiı, N. Lemire, Z. Reichstein, Stably Cayley groups in characteristic zero, Int. Math. Res. Not. IMRN 2014, No. 19, 5340 -- 5397. [13] P. Bravi, D. Luna, An introduction to wonderful varieties with many examples of type F4, J. Algebra 329 (2011), 4 -- 51. [14] M. Brion, F. Pauer, Valuations des espaces homog`enes sph´eriques, Comment. Math. Helv. 62 (1987), 265 -- 285. [15] S. Cupit-Foutou, Anti-holomorphic involutions and spherical subgroups of reductive groups, Transform. Groups 20 (2015), no. 4, 969 -- 984. [16] Y. Z. Flicker, C. Scheiderer, R. Sujatha, Grothendieck's theorem on non-abelian H 2 and local-global principles, J. Amer. Math. Soc. 11 (1998), 731 -- 750. [17] A. Grothendieck, ´El´ements de g´eom´etrie alg´ebrique. IV: ´Etude locale des sch´emas et des morphismes de sch´emas. (Troisi`eme partie), Publ. Math. Inst. Hautes ´Etudes Sci. 28 (1966), 1 -- 255. [18] M. Huruguen, Toric varieties and spherical embeddings over an arbitrary field, J. Algebra 342 (2011), 212 -- 234. [19] J. Jahnel, The Brauer-Severi variety associated with a central simple algebra: a survey, https://www.math.uni-bielefeld.de/LAG/man/052.pdf. [20] F. Knop, The Luna-Vust theory of spherical embeddings, in: Proceedings of the Hyderabad Conference on Algebraic Groups (Hyderabad, 1989), Manoj Prakashan, Madras, 1991, pp. 225 -- 249. [21] F. Knop, Automorphisms, root systems, and compactifications of homogeneous varieties, J. Amer. Math. Soc. 9 (1996), 153 -- 174. [22] F. Knop on the https://mathoverflow.net/q/271795. spherical colors (https://mathoverflow.net/users/89948/friedrich-knop), Action space G/H, URL (version: homogeneous of a of N (H)/H 2017-06-09): [23] F. Knop (https://mathoverflow.net/users/89948/friedrich-knop), Is any spherical subgroup con- jugate to a subgroup defined over a smaller algebraically closed field?, URL (version: 2017-08-01): https://mathoverflow.net/q/277708. [24] Q. Liu, Algebraic Geometry and Arithmetic Curves, Oxford Graduate Texts in Mathematics, Vol. 6, Oxford Science Publications, Oxford University Press, Oxford, 2002. [25] I. V. Losev, Uniqueness property for spherical homogeneous spaces, Duke Math. J. 147 (2009), 315 -- 343. [26] D. Luna, Grosses cellules pour les vari´et´es sph´eriques, in: Algebraic Groups and Lie Groups, Austral. Math. Soc. Lect. Ser., Vol. 9, Cambridge Univ. Press, Cambridge, 1997, pp. 267 -- 280. [27] D. Luna, Vari´et´es sph´eriques de type A, Publ. Math. Inst. Hautes ´Etudes Sci. 94 (2001), 161 -- 226. [28] J. S. Milne, Algebraic Groups. The Theory of Group Schemes of Finite Type over a Field, Cambridge Studies in Advanced Mathematics, Vol. 170, Cambridge University Press, Cambridge, 2017. [29] L. Moser-Jauslin, R. Terpereau, with an appendix by M. Borovoi, Real structures on horospherical varieties, arXiv:1808.10793[math.AG]. [30] N. Perrin, On the geometry of spherical varieties, Transform. Groups 19 (2014), no. 1, 171 -- 223. [31] V. L. Popov, E. B. Vinberg, Invariant theory, in: A. N. Parshin and I. R. Shafarevich (Eds.), Algebraic Geometry IV, Encyclopaedia of Mathematical Sciences, Vol. 55, Springer-Verlag, Berlin, 1994, pp. 123 -- 278. [32] J.-P. Serre, Galois Cohomology, Springer-Verlag, Berlin, 1997. [33] T. A. Springer, Reductive groups, in: Automorphic Forms, Representations and L-Functions (Proc. Sympos. Pure Math., Oregon State Univ., Corvallis, Ore., 1977), Proc. Sympos. Pure Math., Vol. 33, Amer. Math. Soc., Providence, RI, 1979, Part 1, pp. 3 -- 27. [34] T. A. Springer, Linear Algebraic Groups, Second ed., Progress in Math., Vol. 9, Birkhauser, Boston, MA, 1998. [35] D. A. Timashev, Homogeneous Spaces and Equivariant Embeddings, Encyclopaedia of Mathematical Sciences, Vol. 138, Berlin, Springer-Verlag, 2011. [36] J. Tits, Classification of algebraic semisimple groups, in: Algebraic Groups and Discontinuous Sub- groups (Proc. Sympos. Pure Math., Vol. 9, Boulder, Colo., 1965), Amer. Math. Soc., Providence, R.I., 1966, pp. 33 -- 62. [37] T. Wedhorn, Spherical spaces, Ann. Inst. Fourier (Grenoble) 68 (2018), 229 -- 256. 42 MIKHAIL BOROVOI WITH AN APPENDIX BY GIULIANO GAGLIARDI Borovoi: Raymond and Beverly Sackler School of Mathematical Sciences, Tel Aviv Uni- versity, 6997801 Tel Aviv, Israel E-mail address: [email protected] Gagliardi: Raymond and Beverly Sackler School of Mathematical Sciences, Tel Aviv University, 6997801 Tel Aviv, Israel E-mail address: [email protected]
1506.00872
1
1506
2015-06-02T13:13:20
On equivariant quantum Schubert calculus for G/P
[ "math.AG", "math.AT", "math.CO" ]
We show a Z^2-filtered algebraic structure and a "quantum to classical" principle on the torus-equivariant quantum cohomology of a complete flag variety of general Lie type, generalizing earlier works of Leung and the second author. We also provide various applications on equivariant quantum Schubert calculus, including an equivariant quantum Pieri rule for any partial flag variety of Lie type A.
math.AG
math
ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P YONGDONG HUANG AND CHANGZHENG LI Abstract. We show a Z2-filtered algebraic structure and a "quantum to clas- sical" principle on the torus-equivariant quantum cohomology of a complete flag variety of general Lie type, generalizing earlier works of Leung and the second author. We also provide various applications on equivariant quantum Schubert calculus, including an equivariant quantum Pieri rule for partial flag variety F ℓn1,··· ,nk ;n+1 of Lie type A. . G A h t a m [ 1 v 2 7 8 0 0 . 6 0 5 1 : v i X r a 1. Introduction The complex Grassmannian Gr(m, n + 1) parameterizes m-dimensional complex vector subspaces of Cn+1. The integral cohomology ring H ∗(Gr(m, n+1), Z) has an additive basis of Schubert classes σν , indexed by partitions ν = (ν1, · · · , νm) inside an m × (n + 1 − m) rectangle: n + 1 − m ≥ ν1 ≥ · · · ≥ νm ≥ 0. The initial classical Schubert calculus, in modern language, refers to the study of the ring structure of H ∗(Gr(m, n + 1), Z). The content includes (1) a Pieri rule, giving a combinatorial formula for the cup product by a set of generators of the cohomology ring, for instance by the special Schubert classes σ1p , 1 ≤ p ≤ m, where 1p = (1, · · · , 1, 0, · · · , 0) has precisely p copies of 1; (2) more generally, a Littlewood-Richardson rule, giving a (manifestly positive) µ,ν in the cup product combinatorial formula of the structure constants N η σµ ∪ σν =Pη N η µ,νση; (3) a ring presentation of H ∗(Gr(m, n + 1), Z); (4) a Giambelli formula, expressing every σν as a polynomial in special Schu- bert classes. The complex Grassmannian Gr(m, n + 1) is a special case of homogeneous varieties G/P , where G denotes the adjoint group of a complex simple Lie algebra of rank n, and P denotes a parabolic subgroup of G. The classical Schubert calculus, in gen- eral, refers to the study of the classical cohomology ring H ∗(G/P ) = H ∗(G/P, Z). There are various extensions of the classical Schubert calculus by replacing "classical" with "equivariant", "quantum", or "equivariant quantum". The equi- variant quantum Schubert calculus for G/P refers to the study of the (integral) torus-equivariant quantum cohomology ring QH ∗ T (G/P ), which is a deformation of the ring structure of the torus-equivariant cohomology H ∗ T (G/P ) by incorpo- rating genus zero, three-point equivariant Gromov-Witten invariants. In analogy with H ∗(Gr(m, n + 1)), the ring QH ∗ T (G/P ) has a basis of Schubert classes σu over H ∗ T (pt)[q1, · · · , qk] where k := dim H2(G/P ), indexed by elements in a subset W P of the Weyl group W of G. The structure coefficients N w,d u,v of the equivariant 1 2 YONGDONG HUANG AND CHANGZHENG LI quantum product σu ⋆ σv = Xw∈W P ,d∈H2(G/P,Z) N w,d u,v σwqd are homogeneous polynomials in H ∗ T (pt) = Z[α1, · · · , αn] with variables αi being simple roots of G. They contain all the information in the former kinds of Schubert calculus. For instance, the non-equivariant limit of N w,d u,v , given by evaluating (α1, · · · , αn) = 0, recovers an ordinary Gromov-Witten invariant, which counts the number of degree d rational curves in G/P passing through three Schubert subvarieties associated to u, v, w. When G = P SL(n + 1, C), the Weyl group W is a permutation group Sn+1 gen- erated by transpositions si = (i, i + 1). Every homogenous variety P SL(n+ 1, C)/P is of the form F ℓn1,··· ,nk;n+1 := {Vn1 6 · · · 6 Vnk 6 Cn+1 dimC Vnj = nj, ∀1 ≤ j ≤ k}, parameterizing partial flags in Cn+1. As an algebra over H ∗ T (pt)[q1, · · · , qk], the equivariant quantum cohomology ring QH ∗ T (F ℓn1,··· ,nk;n+1) is generated (see e.g. [1, 31]) by special Schubert classes σc[ni,p], where c[ni, p] := sni−p+1 · · · sni−1sni. One of the main results of our present paper is the following equivariant quantum Pieri rule. The non-equivariant limit of it recovers the quantum Pieri rule, which was first given by Ciocan-Fontanine [14], and was reproved by Buch [5]. The clas- sical limit (by evaluating q = 0) is a slight improvement of Robinson's equivariant Pieri rule [44]. By evaluating all equivariant parameters αi and all quantum param- eters qj at 0, we have the classical Pieri rule due to Lascoux and Schutzenberger [32] and Sottile [46]. Theorem 3.10. For any 1 ≤ i ≤ k, 1 ≤ p ≤ ni and any u ∈ W P , we have σc[ni,p] ⋆ σu = Xd∈Piei,p(u) p−di Xj=0 Xw ξni−di−j,p−di−j(µw·φd,u·τd,ni−di)σwqd with the last summation over those w ∈ Per(d) satisfying w · φd ∈ Sni−di,j(u · τd). Here Piei,p(u), etc., are combinatorial sets to be described in section 3.1; the element µ = µw·φd,u·τd,ni−di is a partition inside an (ni−di−j)×(n+1−ni+di+j) rectangle. Each structure coefficient ξni−di−j,p−di−j(µ) coincides with the coefficient of σµ in the equivariant product σc[ni,p] ◦ σµ in H ∗ T (Gr(ni − di − j, n + 1)). Geometrically, it is the restriction of the Grassmannian Schubert class σ(1p−di−j ) of H ∗ T (Gr(ni − di − j, n+1)) (labeled by the special partition of (p−di −j) copies of 1) to a T -fixed point labeled by the partition µ. These restriction type structure coefficients can be easily computed in many ways [3, 11, 16, 23 -- 25, 27, 47]. For completeness, we will include a known formula in Definition 3.4. We remark that our formula above is different from the one in [30] by Lam and Shimozono, which concerns the multiplications by σspsp−1···s2s1sθ in the ring QH ∗ T (F ℓ1,2,··· ,n;n+1). Here θ denotes the highest root, and a ring isomorphism [29,34,43] between QH ∗ T (F ℓ1,2,··· ,n;n+1) and the equivariant homology H T ∗ (ΩSU (n + 1)) of based loop groups after localization is involved. In the special case of complex Grassmannians, H2(Gr(m, n + 1), Z) = Z, so that there is only one quantum variable q. Let Pm,n+1 denote the set of partitions inside an m × (n + 1 − m) rectangle. For ν = (ν1, · · · , νm) and η = (η1, · · · , ηm) ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 3 in Pm,n+1 satisfying ηi − νi ∈ {0, 1} for all i, we introduce an associated partition ην ∈ Pm−r,n+1, ην := (νj1 − j1 + r + 1, νj2 − j2 + r + 2, · · · , νjm−r − jm−r + m), where j1 < j2 < · · · < jm−r denote all entries with ηji = νji . In other words, the Young diagram of η is obtained by adding a vertical strip to the Young diagram of ν; the associated partition ην can also described with this flavor by using a simple join-and-cut operation (see Definition 3.15 and the figures therein for more details). A further simplification of the above theorem leads to the following equivariant quantum Pieri rule for complex Grassmannians. Theorem 3.17. Let 1 ≤ p ≤ m and ν = (ν1, · · · , νm) ∈ Pm,n+1. In QH ∗ 1)), T (Gr(m, n+ σ1p ⋆ σν = p Xr=0Xη ξm−r,p−r(ην)ση + p−1 Xr=0Xκ ξm−1−r,p−1−r(κ′ ν′ )σκq, where the second sum is over partitions η = (η1, · · · , ηm) ∈ Pm,n+1 satisfying η = ν + r and ηi − νi ∈ {0, 1} for all i; the q-terms occur only if ν1 = n + 1 − m, and when this holds, the last sum is over partitions κ = (κ1, · · · , κm−1, 0) such that κ′ := (κ1 + 1, · · · , κm−1 + 1) and ν′ := (ν2, · · · , νm) satisfy κ′ = ν′ + r and κi + 1 − νi+1 ∈ {0, 1} for all 1 ≤ i ≤ m − 1. For instance in QH ∗ T (Gr(3, 7)), we have σ(1,1,1) ⋆ σ(4,0,0) = (α1 + · · · + α6)σ(4,1,1) + q. The special case when p = 1 has been given by Mihalcea [40]. The full Pieri- type formula could have been obtained by combining the study of the equivariant quantum K-theory [10] with an equivariant Pieri rule [16, 27] using Grassmannian algebras. It might also be deduced by the results in [29, 30]. However, an explicit statement, which is a very important component of the equivariant quantum Schu- bert calculus, has not been given anywhere else yet. Actually, the equivariant Pieri rule, which is read off from the classical part of Theorem 3.17, is different from the aforementioned known formulations. It has even inspired the second author and Ravikumar to find an equivariant Pieri rule for Grassmannians of all classi- cal Lie types with a geometric approach in their recent work [38]. On the other hand, Buch has recently shown a (manifestly positive) equivariant puzzle rule for two-step flag varieties [6], generalizing the rules in [7, 23]. Therefore he obtains a Littlewood-Richardson rule for QH ∗ T (Gr(m, n + 1)), due to the equivariant "quan- tum to classical" principle [10]. It will be interesting to study how we simplify Buch's general rule in the special case of Pieri rule to obtain a more compact form as above. In analogy with the contents (1),(2),(3),(4) of the classical Schubert calculus, there are the corresponding equivariant quantum extensions, say (1)′, (2)′, (3)′ and (4)′, in the equivariant quantum Schubert calculus. The problem (2)′ of finding a manifestly positive formula of the structure constants for the equiavariant quantum cohomology remains open except for very few cases including complex Grassman- nians [6] (which is widely open even for the classical Schubert calculus). For (3)′ and (4)′, there have been a few developments (see [21] and [1, 20, 31] respectively). For QH ∗ T (F ℓn1,··· ,nk;n+1), we expect that our Theorem 3.10 leads to an alternative 4 YONGDONG HUANG AND CHANGZHENG LI approach to the earlier studies on (3)′ and (4)′. Indeed, we will illustrate this for the special case of complex Grassmannians [42], by using Theorem 3.17. It is another one of the main results of the present paper that the "quantum to classical" principle holds among various equivariant Gromov-Witten invariants of G/P . Applying it for G = P SL(n + 1, C), we achieve the above theorems. In [33, 37], Leung and the second author discovered a functorial relationship be- tween the quantum cohomologies of complete and partial flag varieties, in terms of filtered algebraic structures on QH ∗(G/B). A "quantum to classical" principle among various Gromov-Witten invariants of G/B was therefore obtained [35] in a combinatorial way. Such a principle was also shown for some other cases [8, 12, 13] in a geometric way. It has led to nice applications in finding Pieri-type formulas, such as [9, 36]. In the present paper, we generalize the work [33] to the equivariant quantum cohomology QH ∗ T (G/B) in the following case. Theorem 2.5. Let P ⊃ B be a parabolic subgroup of G such that P/B is iso- morphic to the complex projective line P1. With respect to the natural projection G/B → G/P , there is a Z2-filtration F = {Fa} on QH ∗ T (G/B) respecting the alge- bra structure: Fa ⋆ Fb ⊂ Fa+b for any a, b ∈ Z2. Moreover, the associated graded algebra GrF (QH ∗ T (G/P ) as Z2-graded Z-algebras, after localization. T (G/B)) is isomorphic to QH ∗(P/B)⊗Z QH ∗ Explicit construction of the Z2-filtration will be given in section 2.2. As a con- sequence, we obtain Theorem 2.7, giving identities among various equivariant Gromov-Witten invariants of G/B. This is an extension of Theorem 1.1 [35] in exactly the same form. Together with an equivariant extension of Peterson- Woodward comparison formula (see Proposition 2.10), we expect that Theorem 2.7 leads to nice applications in the equivariant quantum Schubert calculus for various G/P . Indeed, for G = P SL(n + 1, C), we can reduce all the relevant equivariant Gromov-Witten invariants in Theorem 3.10 to Pieri-type structure coefficients for H ∗ T (F ℓ1,2,··· ,n;n+1), by applying Theorem 2.7 repeatedly. Combining such reduc- tions with Robinson's equivariant Pieri rule [44], we obtain Theorem 3.10. It will be very interesting to explore a simpler and conceptually much cleaner proof by a kind of inductive argument based on Mihalcea's characterization of the struc- ture coefficients via the equivariant quantum Chevalley rule [41]. We can also find nice applications on the equivariant quantum Pieri rules for orthogonal isotropic Grassmannians, in a joint work in progress by the second author and Ravikumar. This paper is organized as follows. In section 2, we introduce basic notations, and prove the main technical results for homogeneous varieties G/P of general Lie type. In section 3, we show an equivariant quantum Pieri rule for F ℓn1,··· ,nk;n+1, and give a simplification in the special case of complex Grassmannians. Finally in the appendix, we give alternative proofs of a ring presentation of QH ∗ T (Gr(m, n+1)) and the equivariant quantum Giambelli formula for Gr(m, n + 1). Acknowledgements. The authors thank Leonardo Constantin Mihalcea for his generous helps. The authors also thank Anders Skovsted Buch, Ionut Ciocan- Fontanine, Thomas Lam, and Naichung Conan Leung for helpful conversations. The authors would like to thank the anonymous referees for their valuable comments on an earlier version of the manuscript. The second author is supported by IBS- R003-D1. ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 5 2. A Z2-filtration on QH ∗ T (G/B) and its consequences In this section, we show a Z2-filtered algebraic structure on QH ∗ T (G/B), with respect to a choice of a simple root. We also obtain a number of identities among various equivariant Gromov-Witten invariants of G/B. 2.1. Preliminaries. We fix the notions here, following [15, 18, 19]. G be the (connected) adjoint group of g, and B ⊂ G be the Borel subgroup with Let g be a complex simple Lie algebra of rank n, h be a Cartan subalgebra of g, and ∆ = {α1, · · · , αn} ⊂ h∗ be a basis of simple roots. Let R denote the root i=1 Z≥0αi called the system of (g, h). We have R = R+ ⊔ (−R+) with R+ = R ∩Ln set of positive roots, and have the Cartan decomposition g = hL(cid:0)Lγ∈R gγ(cid:1). Let b := Lie(B) = hL(Lγ∈R+ gγ(cid:1). Each subset ∆′ of ∆ gives a root subsystem R∆′ = ∆′ = R+T(cid:0)Lα∈∆′ Zα(cid:1), and defines a parabolic subalgebra by p(∆′) := bL(cid:0)Lγ∈−R+ gγ(cid:1). This gives rise to a one-to-one correspondence between the subsets ∆P of ∆ and the parabolic subgroups P ⊂ G that contain B. In particular, we denote by Pβ the parabolic subgroup corresponding to a subset {β} ⊂ ∆. We notice that Pβ is a minimal subgroup among those parabolic subgroups P ) B, and Pβ/B is isomorphic to the complex projective line P1. ∆′) where R+ ∆ ⊔ (−R+ R+ ∆′ Let {α∨ 1 , · · · , χ∨ 1 , · · · , α∨ n } ⊂ h be the simple coroots, {χ∨ tal coweights, {χ1, · · · , χn} ⊂ h∗ be the fundamental weights, and ρ := Pn n} ⊂ h be the fundamen- i=1 χi. Let h·, ·i : h∗ × h → C denote the natural pairing. Every simple root αi labels a simple reflection si := sαi, which maps λ ∈ h and γ ∈ h∗ to si(λ) = λ − hαi, λiα∨ i and si(γ) = γ − hγ, α∨ i iαi respectively. Let W denote the Weyl group generated by all the simple reflections, and WP denote the subgroup of W generated by {sα α ∈ ∆P }. Let ℓ : W → Z≥0 denote the standard length function, ω (resp. ωP ) denote the longest element in W (resp. WP ), and W P denote the subset of W that consists of minimal length representatives of the cosets in W/WP . Denote i . Every γ ∈ R is given by γ = w(αi) i ) ∈ Q∨ and the reflection for some (w, αi) ∈ W × ∆, then the coroot γ∨ := w(α∨ sγ := wsiw−1 ∈ W are both independent of the expressions of γ. P := Lαi∈∆P Q∨ := Ln i and Q∨ i=1 Zα∨ Zα∨ Let T be the maximal complex torus of G with h = Lie(T ), and N (T ) denote the ∼=→ N (T )/T by w 7→ wT . normalizer of T in G. There is a canonical isomorphism W We then have a Bruhat decomposition of the homogeneous variety G/P , given by G/P = Fw∈W P B− wP/P , where B− denotes the opposite Borel subgroup, and each cell B− wP/P is isomorphic to CdimC G/P −ℓ(w). As a consequence, the integral (co)homology of the homogeneous variety G/P has an additive Z-basis of Schubert (co)homology classes σw (resp. σw) of (co)homology degree 2ℓ(w), indexed by w ∈ W P . Here σw = P.D.([X w]) is the Poincar´e dual of the fundamental class of the Schubert subvariety X w := B− wP/P ⊂ G/P , and σw is the fundamental class of the Schubert subvariety Xw := B wP/P ⊂ G/P . We consider the integral T -equivariant cohomology H ∗ T (G/P ) with respect to the natural (left) T -action on G/P . Every Schubert subvariety X w is T -invariant and of complex codimension ℓ(w), and hence determines an equivariant cohomology class in H 2ℓ(w) (X), which we still denote as σw by abuse of notations. The equivariant T (pt)-module with an H ∗ cohomology H ∗ T (G/P ) is an H ∗ T (pt)-basis of the equivariant Schubert classes σw. Here H ∗ T (pt), denoting the T -equivariant cohomology of a T 6 YONGDONG HUANG AND CHANGZHENG LI point equipped with the trivial T -action, is isomorphic to the symmetric algebra of the character group of T . Since G is adjoint, we have S := H ∗ T (pt) = Z[α1, · · · , αn]. The second integral homology H2(G/P, Z) has a basis of Schubert curve classes 7→ {σsα}α∈∆\∆P . Therefore, it can be identified with Q∨/Q∨ ajσsαj P , by Pαj ∈∆\∆P qaj α∨ j +Q∨ P is a monomial in ajα∨ j + Q∨ λP = Pαj ∈∆\∆P tegers, i.e., if the associated function qλP := Qαj ∈∆\∆P P . We call λP effective, if all aj are nonnegative in- P j +Q∨ the polynomial ring Z[q] of indeterminate variables qα∨ . The integral (small ) quantum cohomology ring QH ∗(G/P ) = (H ∗(G/P ) ⊗ Z[q], •P ) of G/P is a defor- mation of the ring structure of H ∗(G/P ). The quantum multiplication is defined by incorporating genus zero, three-point Gromov-Witten invariants, i.e., intersection numbers on the moduli spaces of stable maps M0,3(G/P, d), with respect to three classes pull-back from H ∗(G/P ) via the natural evaluation maps. The moduli space M0,3(G/P, d) admits a natural T -action induced from the one on the target space G/P , and the evaluation maps are all T -equivariant. The so-called T -equivariant Gromov-Witten invariants are polynomials in S, defined by pulling back classes in H ∗ with the equivariant Gysin push forward map [21]. The Schubert classes σu form an S[q]-basis of the commutative T -equivariant quantum cohomology ring QH ∗ T (G/P ). The structure coefficients N w,λP T(cid:0)M0,3(G/P, d)(cid:1) and taking integration over the moduli space in the equivariant quantum product, T (G/P ) to H ∗ u,v σu ⋆P σv = Xw∈W P ,λP ∈Q∨/Q∨ P N w,λP u,v σwqλP , u,v u,v coincides with the coef- T (G/P ). The non-equivariant are homogenous polynomials in S. The classical limit N w,0 ficient of σw in the equivariant product σu ◦ σv in H ∗ limit N w,λP cient of σwqλP in the quantum product σu •P σv in QH ∗(G/P ). (cid:12)(cid:12)α1=···=αn=0 is a Gromov-Witten invariant, coinciding with the coeffi- There is an equivariant quantum Chevalley formula stated by Peterson [43] and proved by Mihalcea [41], which concerns the multiplication by Schubert divisor classes in QH ∗ T (G/P ). We review the special case of it when P = B as follows. In this case, we notice that Q∨ B = 0, WB = {1} and W B = W . Hence we will simply denote λ := λB and qj := qα∨ , whenever there is no confusion. j Proposition 2.1 (Equivariant quantum Chevalley formula for G/B). For any simple reflection si and any u in W , in QH ∗ T (G/B), we have σsi ⋆ σu = (χi − u(χi))σu +Xhχi, γ∨iσusγ +Xhχi, γ∨iqγ∨σusγ , the first summation over those γ ∈ R+ satisfying ℓ(usγ) = ℓ(u) + 1, and the second summation over those γ ∈ R+ satisfying ℓ(usγ) = ℓ(u) + 1 − h2ρ, γ∨i. Despite of the lack of geometric meaning, the structure coefficients N w,λP for T (G/P ) enjoy a positivity property [39]. Here is a precise statement for P = B. u,v QH ∗ Proposition 2.2 (Positivity). Let u, v, w ∈ W , λ ∈ Q∨, and d := ℓ(u) + ℓ(v) − ℓ(w) − h2ρ, λi. The structure coefficient N w,λ T (G/B) is a homogeneous polynomial of degree d in Z≥0[α1, · · · , αn], provided that λ is effective and d ≥ 0, and zero otherwise. u,v for QH ∗ ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 7 We remark that the structure coefficients for equivariant (quantum) product of the equivariant (quantum) Schubert classes determined by the T -invariant Schu- bert varieties Xω0w enjoy the Graham-positivity [17, 39], i.e., they take values in Z≥0[−α1, · · · , −αn] instead. 2.2. Main results. Let β ∈ ∆. The natural projection G/B → G/Pβ is a bundle with fiber Pβ/B ∼= P1. As in [35], we define a map sgnβ : W → {0, 1} by sgnβ(w) = 1 if w(β) ∈ −R+, or 0 otherwise. In other words, we have sgnβ(w) =(1, 0, if ℓ(w) − ℓ(wsβ) > 0 if ℓ(w) − ℓ(wsβ) ≤ 0 . For I = (i1, · · · , in) ∈ Zn, we denote I := i1 + · · · + in and αI := αi1 1 · · · αin n Definition 2.3. With respect to β ∈ ∆, we define a map grβ : W ×Zn×Q∨ −→ Z2, grβ(w, I, λ) := (sgnβ(w) + hβ, λi, ℓ(w) + I + h2ρ, λi − sgnβ(w) − hβ, λi). The equivariant quantum cohomology ring QH ∗ T (G/B) admits a Z-basis σwαI qλ, with w ∈ W and αI qλ ∈ Z[α, q]. Naturally, we define the grading of σwαI qλ to be grβ(w, I, λ). Therefore, we obtain a family of Z-vector subspaces of QH ∗ T (G/B) by F := {Fa}a∈Z2 with Fa := Mgrβ (w,I,λ)≤a ZσwαI qλ ⊂ QH ∗ T (G/B). Here we are considering the lexicographical order on Z2. That is, (a1, a2) < (b1, b2) if and only if either a1 < b1 or (a1 = b1 and a2 < b2). The associated Z2-graded vector space with respect to F is then given by GrF (QH ∗ T (G/B)) = Ma∈Z2 GrF a where GrF a := Fa(cid:14) ∪b<a Fb. Lemma/Definition 2.4 (Lemma 1 of [48]). Let λP ∈ Q∨/Q∨ unique λB ∈ Q∨ such that λP = λB + Q∨ We call λB the Peterson-Woodward lifting of λP . P . Then there is a P and hγ, λBi ∈ {0, −1} for all γ ∈ R+ P . Thanks to the above lemma, we obtain an injective morphism of S-modules ψ∆,∆P : QH ∗ T (G/P ) −→ QH ∗ T (G/B) B P qλP 7→ σwωP ωP ′ defined by σw qλB . Here ωP ′ denotes the longest element in the Weyl subgroup generated by the simple reflections {sα α ∈ ∆P , hα, λBi = 0}, and the subscript "P " in the Schubert classes σw P for G/P is used in order to distinguish them from those Schubert classes for G/B. In the special case when P = Pβ, we simply denote ψβ := ψ∆,{β}. Our first main result is the next theorem, giving an equivariant generalization of the special case of Theorems 1.2 and 1.4 of [33] when P = Pβ. We take an isomorphism QH ∗(P1) ∼= Z[x,t] hx2−ti of algebras. Theorem 2.5. The filtration F gives a Z2-filtered algebraic structure on QH ∗ That is, we have Fa ⋆ Fb ⊂ Fa+b for any a, b ∈ Z2. GrF ver : QH ∗(P1) −→ GrF (i,0) ⊂ GrF (QH ∗ The map Ψβ T (G/B)), T (G/B). ver := Li∈Z 8 YONGDONG HUANG AND CHANGZHENG LI defined by x 7→ σsβ and t 7→ qβ∨, is an isomorphism of Z-algebras. The map Ψβ hor : QH ∗ T (G/Pβ) −→ GrF hor :=Mj∈Z GrF (0,j) ⊂ GrF (QH ∗ T (G/B)), ), is an isomorphism of S-algebras. 7→ ψβ(σwαI qλPβ defined by σwαI qλPβ Remark 2.6. There is a Z2-filtration F ′ on QH ∗ from F . The above Ψβ ver, Ψβ Ψβ ver ⊗ Ψβ hor : GrF ′(cid:0)QH ∗ hor induce an isomorphism of Z2-graded Z-algebras: T (G/B)[q−1 β∨ ](cid:1) ∼=−→ QH ∗(P1)[t−1] ⊗Z QH ∗ T (G/P ). Our second main result is the next generalization of [35, Theorem 1.1] to QH ∗ with the statements exactly of the same form. We simply denote sgni := sgnαi . Theorem 2.7. Let u, v, w ∈ W and λ ∈ Q∨. The coefficient N w,λ equivariant quantum product σu ⋆ σv in QH ∗ T (G/B) satisfies the following. u,v of σwqλ in the T (G/B), (1) N w,λ u,v = 0 unless sgni(w) + hαi, λi ≤ sgni(u) + sgni(v) for all 1 ≤ i ≤ n. (2) If sgnk(w) + hαk, λi = sgnk(u) + sgnk(v) = 2 for some 1 ≤ k ≤ n, then T (G/B)[q−1 β∨ ], naturally extended u,v = N w,λ−α∨ N w,λ k usk,vsk =  N wsk,λ−α∨ u,vsk k , if sgnk(w) = 0 N wsk,λ u,vsk , if sgnk(w) = 1 . Corollary 2.8. Let u, v, w ∈ W , α ∈ ∆ and λ ∈ Q∨. (1) If hα, λi = sgnα(u) = 0 and sgnα(w) = sgnα(v) = 1, then N w,λ (2) If hα, λi = sgnα(u) = 1and sgnα(w) = sgnα(v) = 0, then N w,λ u,v = N wsα,λ u,vsα . u,v = N wsα,λ−α∨ . Proof. For part (1), we note hα, λ+α∨i = 2, sgnα(wsα) = 0, sgnα(usα) = sgnα(v) = 1. Applying "(u, v, w, λ, αk)" in Theorem 2.7 (2) to (v, usα, wsα, λ + α∨, α), we have N wsα,λ+α∨ vsα,u = N w,λ v,u . Similarly, we conclude (2), by applying "(u, v, w, λ, αk)" to (u, vsα, wsα, λ, α). (cid:3) = N wsαsα,λ+α∨−α∨ = N wsα,λ+α∨−α∨ . That is, N wsα,λ vsα,usαsα v,usαsα usα,v v,usα Remark 2.9. When λ = 0, Corollary 2.8 (1) was also known as the "descent- cycling" condition for H ∗ T (G/B) in [22]. 2.3. Equivariant Peterson-Woodward comparison formula. There is a com- parison formula, originally stated by Peterson [43] and proved by Woodward [48]. It tells that every genus zero, three-point Gromov-Witten invariant of G/P coincides with a corresponding Gromov-Witten invariant of G/B. It has played an important role in the earlier works [33, 35, 37]. In order to prove our main results, we need the equivariant extension of the comparison formula as follows. Our readers may skip this subsection first, by assuming the following proposition. Proposition 2.10 (Equivariant Peterson-Woodward comparison formula). For any u, v, w ∈ W P and λP ∈ Q∨/Q∨ N w,λP P , we have u,v = N wωP ωP ′ ,λB u,v , where λB denotes the Peterson-Woodward lifting of λP , and ωP ′ denotes the longest element in the Weyl subgroup generated by {sα α ∈ ∆P , hα, λBi = 0}. P qλP in the equivariant quantum product σu P ⋆P σv That is, the coefficient of σw QH ∗ T (G/P ) coincides with the coefficient of σwωP ωP ′ P in T (G/B). B in QH ∗ qλB in σu B ⋆B σv B ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 9 Remark 2.11. The above statement is exactly of the same as Theorem 10.15 (2) of [29], which is an equivalent version of the non-equivariant Peterson-Woodward comparison formula in terms of Gromov-Witten invariants in [48]. The geometric method in [48] might also be valid in the equivariant setting, while a rigorous argument is missing. In this subsection, we will devote to a proof of Proposition 2.10, by a direct translation of Corollary 10.22 of [29] by Lam and Shimozono. We will have to introduce some notations on the affine Kac-Moody algebras, which, however, will be used in the rest of this subsection only. The affine Weyl group of G is the semi-direct product Waf := W ⋉ Q∨, in which the image of λ ∈ Q∨ in Waf is a translation, denoted as tλ. We have tw(λ) = wtλw−1 and tλtλ′ = tλ+λ′ for all w ∈ W and λ, λ′ ∈ Q∨. Let Q∨ denote the set of anti- dominant elements in Q∨, i.e., Q∨ = {λ ∈ Q∨ hα, λi ≤ 0 for all α ∈ ∆}. Denote af := {wtλ λ ∈ Q∨, and if α ∈ ∆ satisfies hα, λi = 0 then w(α) ∈ R+}, which W − consists of the minimal length representatives of the cosets Waf/W . The (level zero) action of Waf on the affine root system Raf = R+ af) is given by wtλ(γ+mδ) = w(γ) + (m − hγ, λi)δ, in which δ denotes the null root and R+ af := {γ + mδ m ∈ Z+ or (m = 0 and γ ∈ R+)}. Denote the subgroup (WP )af := WP ⋉ Q∨ P and the subset (W P )af := {x ∈ Waf x(γ + mδ) ∈ R+ af with γ ∈ RP }. Lemma 2.12 (See e.g. Lemma 10.6 and Proposition 10.10 of [29]). For every x ∈ Waf, there is a unique factorization x = x1x2 with x1 ∈ (W P )af and x2 ∈ (WP )af. This defines a map1 φP : Waf → (W P )af by x 7→ x1. Then for any w ∈ W P , we have φP (wx) = wφP (x). Proposition 2.13 (Corollaries 9.3 and 10.22 of [29]). Let u, v, w ∈ W P and λP ∈ af ∩(W P )af Q∨/Q∨ P . Write uφP (tµ) = u′tµ′ , vφP (tκ) = v′tκ′ and wφP (tη) = where λP = η − µ − κ + Q∨ w′tη′, where u′, v′, w′ ∈ W and µ′, κ′, η′ ∈ Q∨. Then we have P . Pick µ, κ, η ∈ Q∨ such that uφP (tµ), vφP (tκ) and wφP (tη) ∈ W − afF(−R+ af for all γ + mδ ∈ R+ N w,λP u,v = N w′,η′−µ′−κ′ u′,v′ , in which the left-hand (resp. right-hand) side is a structure coefficient of the equi- variant quantum product σu T (G/B)). T (G/P ) (resp. σu′ B ∈ QH ∗ B ⋆B σv′ P ∈ QH ∗ P ⋆P σv χ∨ α and η = 2µ + λB, where M := max{hα, λBi + 1 α ∈ ∆}. Then µ, η are both in Q∨ (since the Proof of Proposition 2.10. Let µ = −12(n + 1)MPα∈∆\∆P determinant of the Cartan matrix(cid:0)hαi, α∨ Clearly, for α ∈ ∆ \ ∆P , we have hα, µi < 0 and hα, ηi < 0. For γ ∈ R+ P (in particular for γ ∈ ∆P ), we have hγ, µi = 0 and hγ, ηi = hγ, λBi ∈ {0, −1}. Hence, µ, η are both in Q∨, and utµ, wωP ωP ′ tη are both in W − af by noting u ∈ W P and wωP ωP ′ ∈ W P ′ j i(cid:1) is equal to 1, 2, 3, 4 or n + 1). Let γ + mδ ∈ R+ af with γ ∈ RP . Since tµ(γ + mδ) = γ + (m − hγ, µi)δ = γ + mδ ∈ R+ af, tµ is in (W P )af. Note ωP ωP ′tη(γ + mδ) = ωP ωP ′(γ)+ (m− hγ, λBi)δ. Clearly, hγ, λBi is in {0, 1, −1}, and it vanishes if and only if γ ∈ RP ′ . Note ωP ωP ′(R+ . P ′ ) ⊂ R+ and ωP ωP ′ (R+ P \ R+ i) if m > 2 or (m = 1 and γ ∈ R+ ii) if m = 1 and γ ∈ −R+ P ′ ) ⊂ −R+. As a consequence, we have P ), then m − hγ, λBi > 0; P ′, then m − hγ, λBi = 1 > 0; 1The map is denoted as πP in [29]. 10 YONGDONG HUANG AND CHANGZHENG LI iii) if m = 1 and γ ∈ (−R+ P ) \ (−R+ P ′ ), then m − hγ, λBi = 0, and we note ωP ωP ′ (γ) ∈ R+ in this case; iv) if m = 0, then γ ∈ R+ P . Further, if γ ∈ R+ P \ R+ P ′ , then m − hγ, λBi = 1 > 0; if γ ∈ R+ P ′, then m − hγ, λBi = 0, and we note ωP ωP ′ (γ) ∈ R+. Hence, ωP ωP ′ tη is also in (W P )af. Since ωP ωP ′ ∈ WP , η − ωP ωP ′(η) ∈ Q∨ we have the factorizations tµ = tµ ·id and tη = (ωP ωP ′ tη)·(cid:0)(ωP ωP ′ )−1tη−ωP ωP ′ (η)(cid:1). Hence, we have φP (tµ) = tµ and φP (tη) = ωP ωP ′ tη due to the uniqueness of the factorization. Hence, utµ, wωP ωP ′ tη are in (W P )af, by noting u, w ∈ W P and using Lemma 2.12. P . Thus Now we set κ := µ and use the same notation as in Proposition 2.13. It follows immediately from the above arguments that µ, κ, η satisfy all the hypotheses of Proposition 2.13, for which we have u′ = u, v′ = v, w′ = wωP ωP ′ , µ′ = µ, κ′ = κ, η′ = η and η′ − µ′ − κ′ = λB. Therefore the statement follows. (cid:3) 2.4. Proof of theorems. The proofs of the theorems in section 2.2 are similar to the corresponding ones in the non-equivariant case in [33, 35]. 2.4.1. Preliminary propositions. We will need the next combinatorial fact. Lemma 2.14 (Lemmas 3.8 and 3.9 of [33]). Let u ∈ W and γ ∈ R+ satisfy ℓ(usγ) = ℓ(u) + 1 − h2ρ, γ∨i. If hα, γ∨i > 0 for some α ∈ ∆, then ℓ(u) − ℓ(usα) = ℓ(usγsα) − ℓ(usγ) = 1. Furthermore if γ 6= α, then hα, γ∨i = 1. The next proposition is the generalization of (a special case of) the Key Lemma and Proposition 3.23 of [33] to the equivariant quantum cohomology QH ∗ T (G/B). We would like to remind our readers that a simple root β has been fixed in prior. Proposition 2.15. For any 1 ≤ i ≤ n and u ∈ W , we have Fa ⋆ Fb ⊂ Fa+b, where a := grβ(si, 0, 0) and b := grβ(u, 0, 0). Furthermore, we have σsi ⋆ σu = σusi in GrF (QH ∗ (⋄). T (G/B)), if the following hypotheses (⋄) hold: si = sβ and u ∈ W Pβ Proof. Write σsi ⋆ σu =P cw,I,λσwαI qλ, and denote d := grβ(w, I, λ). The state- ment to prove is equivalent to the following: i) d ≤ a + b whenever the coefficient cw,I,λ does not vanish; ii) under the additional hypotheses (⋄), d = a+b if and only if σwαI qλ = σusi . Note cw,I,λ 6= 0 only if d = ℓ(w) + I + h2ρ, λi = 1 + ℓ(u) = a + b. Thus for nonzero cw,I,λ, d is less than (resp. equal to) a + b if and only if d1 is less than (reps. equal to) a1 + b1, where a = (a1, a2), b = (b1, b2) and d = (d1, d2). Note a1 + b1 = sgnβ(si) + sgnβ(u) and d1 = sgnβ(w) + hβ, λi. Due to Proposition 2.1, if cw,I,λ 6= 0, then one of the following cases must hold. (1) σwαI qλ = σuαj for some 1 ≤ j ≤ n, which must come from (χi − u(χi))σu. Clearly, d1 = sgnβ(u) = b1 ≤ a1 + b1, and '<" holds if we assume (⋄). (2) σwαI qλ = σusγ with ℓ(usγ) = ℓ(u) + 1. If either of a1, b1 is nonzero, then d1 ≤ 1 ≤ a1 + b1. Otherwise, we have a1 = b1 = 0, i.e., si, u ∈ W Pβ . Due to the canonical injective morphism H ∗(G/Pβ ) ֒→ H ∗(G/B), σusγ occurs T (G/B) only if usγ lies in W Pβ as well, i.e., in σu ∪ σv ∈ H ∗(G/B) ⊂ QH ∗ d1 = 0. Thus d1 ≤ a1 + b1. Furthermore we assume (⋄), then d1 = a1 + b1 only if usγ = vsi with ℓ(v) = ℓ(u); cvsi,0,0 6= 0 implies that u ≤ vsi with respect to the Bruhat order, i.e., u is obtained by deleting a simple reflection from a reduced expression of vsi, which implies u = v. Thus if both (⋄) and d1 = a1 + b1 hold, then usγ = usi and cusi,0,0 = hχi, α∨ i i = 1. ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 11 (3) σwαI qλ = σusγ qγ∨ with ℓ(usγ) = ℓ(u) + 1 − h2ρ, γ∨i. Furthermore, we have a1 = 1, assuming (⋄). (a) If hβ, γ∨i < 0, then d1 ≤ 0 ≤ a1 + b1, and "<" holds if we assume (⋄). (b) If hβ, γ∨i = 0, then usγ(β) = u(β), which implies d1 = sgnβ(usγ) = sgnβ(u) = b1 ≤ a1 + b1, and "<" holds if we assume (⋄). (c) If hβ, γ∨i > 0, then sgnβ(u) = 1 and sgnβ(usγ) = 0 by Lemma 2.14. If γ 6= β, then d1 = hβ, γ∨i = 1 = b1 ≤ a1 + b1, and "<" holds if we assume (⋄). If γ = β, then hχi, γ∨i 6= 0 implies that αi = β, and consequently d1 = 2 = 1 + 1 = a1 + b1. Since sgnβ(u) = 1, u 6∈ W Pβ . Hence, the hypotheses (⋄) cannot hold in this case. Hence, the statement follows. (cid:3) Remark 2.16. The main body of [33] is devoted to a complicated proof of the Key Lemma therein with respect to a general P . The above proposition can be obtained as an easy consequence. Nevertheless, we provide a detailed proof for P = Pβ for both the sake of completeness and the purpose of exposition of the Key Lemma. it suffices to show σw ⋆ 2.4.2. Proof of Theorem 2.5. For the first statement, σuαI qλ ∈ Fa+b, for any σw, σuαI qλ ∈ QH ∗ T (G/B) with a = grβ(w, 0, 0) and b = grβ(u, I, λ). Clearly, it holds when ℓ(w) = 0, for which σw = σid is the unit in QH ∗ If ℓ(w) = 1, then w = si and it is done by Proposition 2.15. Assume ℓ(w) > 1 now. Take v ∈ W and i ∈ {1, · · · , n}, such that grβ(w, 0, 0) = grβ(v, 0, 0) + grβ(si, 0, 0) and that the coefficient of σw T (G/B). We use induction on ℓ(w). If sgnβ(w) = 0, then we write w = sjv ∈ W Pβ with si = sβ and v = wsβ. ℓ(v) = ℓ(w) − 1, and simply take αi ∈ ∆ \ {β} such that hχi, γ∨i > 0, which exists in the cup product σv ∪ σsi is nonzero. (cid:0)Precisely, if sgnβ(w) = 1, then we take by noting γ := v−1(αj) 6= β.(cid:1) By the induction hypothesis, we have σsi ⋆ (σv ⋆ σuαI qλ) ∈ σsi ⋆ Fgrβ (v,0,0)+b ⊂ Fa+b. On the other hand, we have (σsi ⋆ σv) ⋆ σuαI qλ = (hχi, γ∨iσw +X csi,v w′,I ′,λ′σw′ αI ′ qλ′ ) ⋆ σuαI qλ with hχi, γ∨i > 0 and all the coefficients csi,v w′,I ′,λ′ ≥ 0. There will be no cancelation, when we expand the product, due to the positivity (see Proposition 2.2). Hence, we conclude σw ⋆ σuαI qλ ∈ Fa+b. It follows directly from Definition 2.3 that there is a unique term σuαI qλ of β∨ if m is even, or grading (m, 0) for every nonnegative integer m. It is given by q m 2 m−1 σsβ q 2 β∨ otherwise. By Proposition 2.1, σsβ ⋆ σsβ = qβ∨ + βσsβ +Xhχβ, sβ(α∨)iσsαsβ , the summation over those simple roots α adjacent to β in the Dynkin diagram of ∆. Thus we have σsβ ⋆ σsβ = qβ∨ in GrF (QH ∗ ver is an isomorphism of (graded) algebras (with respect to the given gradings on both sides). T (G/B)). That is Ψβ As remarked earlier, Lemma 2.4 implies that Ψβ hor is an injective morphism of S-modules. For any σw′ T (G/B) of grading (0, ∗), we have sgnβ(w′) + hβ, λi = 0. This implies that w := w′ ∈ W Pβ if hβ, λi = 0, or w := w′sβ ∈ W Pβ and hβ, λi = −1 otherwise. Denote λP := λ + Zβ∨ ∈ Q∨/Q∨ . Then Pβ σw′ hor is a bijection. Let u, v ∈ W Pβ . Consequently, αI qλ = ψβ(σwαI qλP ). Thus Ψβ αI qλ in QH ∗ 12 YONGDONG HUANG AND CHANGZHENG LI hor(σu ⋆Pβ σv) = Ψβ we have Ψβ µP ∈ Q∨/Q∨ Pβ Ψβ Ψβ hor(σu) ⋆ Ψβ hor(σv), by Proposition 2.10. For any hor(qµP ) equals qµB if hβ, µBi = 0, or σsβ qµB otherwise. Thus hor(qµP ⋆Pβ σv) = hor(qµP ). By Proposition 2.15, Ψβ hor is an isomorphism of S-algebras. hor(qλP ⋆Pβ qµP ) = Ψβ hor(qµP ) ⋆ Ψβ hor(qλP )⋆Ψβ hor(σv). Hence, Ψβ , Ψβ 2.4.3. Proof of Theorem 2.7. The first half is a direct consequence of Theorem 2.5. Now we assume the hypothesis in the second half of the statement, and consider the Z2-filtration F on QH ∗ T (G/B) with respect to β := αk. Write σu ⋆ σv =Xw,λ N w,λ u,v σwqλ = Xw,I,λ cw,I,λσwαI qλ where N w,λ Thus in GrF (QH ∗ u,v =PI cw,I,λαI is nonzero only if I = ℓ(u) + ℓ(v) − ℓ(w) − h2ρ, λi ≥ 0. T (G/B)), we have σu ⋆ σv =X cw,I,λσwαI qλ with sgnβ(w) + hβ, λi = sgnβ(u) + sgnβ(v) = 2. Since sgnβ(W ) = {0, 1}, sgnβ(u) = sgnβ(v) = 1. Thus u′ := usβ, v′ := vsβ are both in W Pβ . By Proposition 2.15, σu′ ⋆ σsβ = σu ∈ GrF grβ (u,0,0) Since the graded algebra GrF (QH ∗ and σv′ ⋆ σsβ = σv ∈ GrF grβ (v,0,0). T (G/B)) is associative and commutative, following from Theorem 2.5. In QH ∗ σu′ σu ⋆ σv =(cid:0)σu′ ⋆ σv′(cid:1) ⋆(cid:0)σsβ ⋆ σsβ(cid:1) = Ψβ = Xw′,I ′,λβ hor(σu′ T (G/Pβ), we write w′,λPβ u′,v′ σw′ ⋆Pβ σv′ qλPβ N = Xw′,λPβ σu ⋆ σv =X cw′,I ′,λPβ ψβ(σw′ qλPβ )αI ′ qβ∨ Then we have ⋆Pβ σv′ ) ⋆ qβ∨, cw′,I ′,λPβ σw′ αI ′ qλPβ . Hence, the second half of the statement follows, by comparing coefficients of both expressions of σu ⋆ σv. Indeed, we note hβ, λ − β∨i = −sgnβ(w) ∈ {0, −1}. It follows that the Peterson- is given by λB = λ − β∨. Set w′ := w if )qβ∨ = σwqλ, and conse- Woodward lifting of λPβ := λ + Q∨ Pβ sgnβ(w) = 0, or wsβ if sgnβ(w) = 1. Then ψβ(σw′ quently cw,I,λ = cw′,I,λPβ for all I. Hence, by Proposition 2.10, qλPβ N w,λ u,v =XI cw,I,λαI =XI cw′,I,λPβ αI = N w′,λPβ u′,v′ = N w,λ−β∨ u′,v′ . To show the remaining identities, we consider the expansion σu ⋆ σv = (σu ⋆ σv′ ) ⋆ σsβ =X N w,λ u,v′ qλσ w ⋆ σsβ =X c w,I,λσ wsβ αI qλ +X c w,I,λσ wsβ αI qλ+β∨ , where sgnβ( w) + hβ, λi = 1 and sgnβ( w) = 0 (resp. 1) hold in the former (resp. latter) summation. Hence, if sgnβ(w) = 0, then for every I we have cw,I,λσwαI qλ = c w,I,λσ wsβ αI qλ+β∨ for a unique term in the latter summation, i.e., for ( w, I, λ) = (wsβ, I, λ − β∨). Thus N w,λ . u,v = PI cw,I,λαI = PI cwsβ ,I,λ−β∨αI = N wsβ ,λ−β∨ u,vsβ ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 13 If sgnβ(w) = 1, then for every I we have cw,I,λσwαI qλ = c w,I,λσ wsβ αI qλ for a unique term in the former summation, i.e., for ( w, I, λ) = (wsβ , I, λ). In this case, N w,λ u,v =PI cw,I,λαI =PI cwsβ ,I,λαI = N wsβ ,λ u,vsβ . 3. Application: an equivariant quantum Pieri rule for F ℓn1,··· ,nk;n+1 Throughout the rest of the present paper, we let G = P SL(n + 1, C), which is the quotient group of G = SL(n + 1, C) by its center Z( G). We make the Dynkin diagram of ∆ in the standard way: ◦−−◦ · · · ◦−−◦ αn . The standard Borel subgroup α1 α2 B of G is the quotient of the subgroup of upper triangular matrices in G by Z( G). Each proper parabolic subgroup P ⊃ B is in one-to-one correspondence with a proper subset ∆P = ∆\{αn1, · · · , αnk }, where n0 := 0 < n1 < n2 < · · · < nk < n+1 =: nk+1. Then F ℓn1,··· ,nk;n+1 := P SL(n + 1, C)/P parameterizes partial flags in Cn+1: F ℓn1,··· ,nk;n+1 = {Vn1 6 · · · 6 Vnk 6 Cn+1 dimC Vni = ni, i = 1, · · · , k}. For each i, we denote by πi : F ℓn1,··· ,nk;n+1 → Gr(ni, n + 1) the natural projection. The equivariant quantum cohomology ring QH ∗ T (F ℓn1,··· ,nk;n+1) is generated (see e.g. [1, 31]) by special Schubert classes σc[ni,p] where c[ni, p] := sni−p+1 · · · sni−1sni . In this section, we will show an equivariant quantum Pieri rule for F ℓn1,··· ,nk;n+1, giving the equivariant quantum multiplication by σc[ni,p]. 3.1. Equivariant quantum Pieri rule. In order to state the formula, we need to introduce some notions, mainly following [5, 14, 44]. The Weyl group W for P SL(n + 1, C) is isomorphic to the permutation group Sn+1, by mapping each simple reflection si to the transposition (i(i + 1)). In particular, each reflection sγ from a positive root γ = αi + αi+1 + · · · + αj is sent to the transposition (i(j + 1)), where 1 ≤ i ≤ j ≤ n. Furthermore, Schubert classes σw in QH ∗ T (F ℓn1,··· ,nk;n+1) are indexed by w ∈ W P with W P = {w ∈ Sn+1 w(ni−1 + 1) < w(ni−1 + 2) < · · · < w(ni), i = 1, · · · , k + 1}. For Gr(m, n + 1) = F ℓm;n+1, we have a bijection ϕm : W P ≃−→ Pm,n+1 to the partitions Pm,n+1 := {(a1, · · · , am) ∈ Zm n + 1 − m ≥ a1 ≥ a2 ≥ · · · ≥ am ≥ 0}, (3.1) w 7→ ϕm(w) = (w(m) − m, · · · , w(2) − 2, w(1) − 1). We simply call such w an m-th Grassmannian permutation, whenever n + 1 is well understood. Set P0,n+1 := {(0)}. Review that the length of u ∈ W = Sn+1 is given by ℓ(u) = {(i, j) 1 ≤ i < j ≤ n + 1 and u(i) > u(j)}. Definition 3.1. Let ζ = (rip · · · i2i1) be a (p + 1)-cycle in W . For any u ∈ W , we say that uζ is special j-superior to u of degree p if all the following hold: (1) i1, · · · , ip ≤ j < r, (2) u(r) > u(i1) > · · · > u(ip), (3) ℓ(uζ) = ℓ(u) + p. 14 YONGDONG HUANG AND CHANGZHENG LI More generally, if ζ1, · · · , ζd are pairwise disjoint cycles such that each uζs is special s=1 ps = p = ℓ(uζ1 · · · ζd) − ℓ(u), then we say that uζ1 · · · ζd is special j-superior to u of degree p, and denote j-superior to u of degree ps and Pd Sj,p(u) := {w ∈ W w is special j-superior to u of degree p}. Furthermore for w = uζ1 · · · ζd ∈ Sj,p(u) above, we sort the values {u(1), · · · , u(j)} \ {u(i) i occurs in some ξs} to get a decreasing sequence [µ1 + j − p, · · · , µj−p−1 + 2, µj−p + 1], and then obtain an associated partition µw,u,j := (µ1, µ2, · · · , µj−p) ∈ Pj−p,n+1. We denote the set of such associated partitions as PSj,p(u) := {µw,u,j w ∈ Sj,p(u)} ⊂ Pj−p,n+1. Remark 3.2. If p = j, then µw,u,j = (0) is the zero partition. Example 3.3. For P SL(7, C)/P = F l2,4;7, we take the same u = [3715246] ∈ W P in one-line notation as in Example 2 of [5]. Since w := [4725136] = u(35)(16) is in S4,2(u), 1, 3, 5, 6 are the indices occurring in u−1w. Sorting {u(1), · · · , u(4)} \ {u(1), u(3), u(5), u(6)}, we obtain a decreasing sequence [7, 5]. Hence, the associated partition is given by µw,u,4 = (7 − 2, 5 − 1) = (5, 4) ∈ P2,7, which corresponds to the 2nd Grassmannian permutation [5712346] for Gr(2, 7). u,v of an equivariant product σu ◦ σv of H ∗ So far there have been no manifestly positive formulas for general structure coefficients N w,0 T (G/P ) except for the case of complex Grasssmannians and two-step flag varieties. However, there does be one for the special case N w,0 w,v (i.e., when u = w) in terms of a linear combination of products of positive roots [3,24] (for G of general Lie type). Geometrically, G/P has finitely many T -fixed points parameterized by the minimal length representatives in W P . We let ιw : pt → G/P denote the natural inclusion of the T -fixed point labeled by w into G/P . Then we have N w,0 T (pt), localizing the equivariant Schubert class σv at such T -fixed point. We will reduce all the relevant coefficients in our equivariant quantum Pieri rule to such kind of coefficients for G/P = Gr(m, n + 1) with v = 1p being a special partition, for which there are much more manifestly positive formulas. With the partitions in Pm,n+1, we give a precise description of ξm,p(a) := N a,0 w(σv) ∈ H ∗ w,v = ι∗ a,1p following [3, 24]. Proposition/Definition 3.4. Let 0 ≤ m ≤ n and a = (a1, · · · , am) ∈ Pm,n+1. Denote by aT = (aT n+1−m) ∈ Pn+1−m,n+1 the transpose of the partition a. Then si1 si2 · · · sia gives a reduced expression of ϕ−1 s=1 as and m (a) ∈ W , where a =Pm 1 , · · · , aT [i1, · · · , ia] : = [n − aT n+1−m + 1, n − aT n+1−m + 2, · · · , n; n−m + 1, (n − 1) − aT n−m + 2, · · · , n − 1; (n − 1) − aT · · · ; m − aT 1 + 1, m − aT 1 + 2, · · · , m]. Consequently, γb := si1 · · · sib−1 (αib ) is a positive root for any 1 ≤ b ≤ a. Further- more, we have ξm,0(a) = 1; for 1 ≤ p ≤ m, the structure coefficient ξm,p(a) = N a,0 a,1p ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 15 is a homogeneous polynomial of degree p in Z≥0[α1, · · · , αn] given by where the sum is over all subsequences 1 ≤ j1 < · · · < jp ≤ a that satisfy [ij1, · · · , ijp ] = [m − p + 1, m − p + 2, · · · , m]. ξm,p(a) =X γj1 · · · γjp , Remark 3.5. If p > aT satisfying the constraint. 1 , then ξm,p(a) = 0, for which there does not exist [j1, · · · , jp] Example 3.6. Let n = 6. For a := (5, 4) ∈ P2,7, we have aT = (2, 2, 2, 2, 1) ∈ P5,7 and [i1, · · · , ia] = [6, 4, 5, 3, 4, 2, 3, 1, 2]. Hence, ξ2,1(a) = s6s4s5s3s4(α2) + s6s4s5s3s4s2s3s1(α2) = α1 + 2α2 + 2α3 + 2α4 + α5 + α6, ξ2,2(a) = s6s4s5s3s4s2s3(α1) · s6s4s5s3s4s2s3s1(α2) = (α1 + · · · + α4)(α1 + · · · + α6). Definition 3.7. Let u ∈ W P , 1 ≤ i ≤ k and 1 ≤ p ≤ ni. We denote by Piei,p(u) the set of elements d = (d1, · · · , dk) with [d0, d1, · · · , dk, dk+1] being of the form [0, · · · , 0 , 1, · · · , 1 , 2, · · · , 2 , · · · , m, · · · m , · · · , 2, · · · , 2 , 1, · · · , 1 , 0, · · · , 0 ] that satisfy both di = m ≤ p and the next property {z } (∗) : {z } {z } {z } {z } {z } {z } u(nhj ) > max{u(r) nhj + 1 ≤ r ≤ nlj +1} for all 1 ≤ j ≤ m. Here 1 ≤ h1 < · · · < hm ≤ lm < · · · < l1 ≤ k denote all the jumps, namely dhj = dlj = j and dhj −1 = dlj +1 = j − 1 for all 1 ≤ j ≤ m. (Note d0 = dk+1 = 0.) Given the above d, we denote by τd ∈ Sn+1 the unique permutation defined by τd(nlj +1 − j + 1) = nhj , j = 1, · · · , m, together with the property that the restriction of τd on the remaining elements {1, · · · , n + 1} \ {nlj +1 − j + 1 1 ≤ j ≤ m} preserves the usual order. Similarly, we denote by φd ∈ Sn+1 the unique permutation given by φd(nlj − j + 1) = nhj −1 + 1, j = 1, · · · , m, together with the property that φd(cid:12)(cid:12){1,··· ,n+1}\{nlj −j+1 1≤j≤m} preserves the usual order. In addition, we denote Per(d) := {w ∈ W P w(nhj −1 + 1) < min{w(r) nhj −1 + 2 ≤ r ≤ nlj + 1}, ∀ j}. The permutations τd and φd 2 can be expressed in terms of products of simple reflections ([5, 14]): τd = τ (m) · · · τ (1) and φd = φ(m) · · · φ(1), where τ (j) = snhj φ(j) = snhj −1+1 · · · snlj −2snlj −1, for each 1 ≤ j ≤ m. Moreover, the above expressions are reduced, implying · · · snlj +1−2snlj +1−1, ℓ(τd) = m Xj=1 (nlj +1 − nhj ) and ℓ(φd) = −m + (nlj − nhj−1). m Xj=1 Remark 3.8. For d ∈ Piei,p(u), di = m < k d = 0 ∈ Piei,p(u), τ0 = φ0 = id ∈ Sn+1 and Per(0) = W P . 2 + 1. When m = 0, we have 2τd coincides with the permutation γd in [5]. With notations in [14], τd = γhl and φd = δ−1 hl . 16 YONGDONG HUANG AND CHANGZHENG LI Example 3.9. Let P SL(7, C)/P = F l2,4;7. Take u = [3715246] ∈ W P , i = 2, p = 3. Then d ∈ Pie2,3(u) only if d = (0, 0), (0, 1) or (1, 1). For (0, 1), the jumps are given by h = l = 2. Since max{u(5), u(6), u(7)} = 6 > 5 = u(n2), (∗) is not satisfied. For (1, 1), we have d2 = m = max{d1, d2} = 1 < 3 = p and 1 = h < l = 2. Clearly, u(n1) = u(2) = 7 > max{u(3), · · · , u(7)}. Thus, Pie2,3(u) = {(0, 0), (1, 1)}. For F ℓn1,··· ,nk;n+1, there are k quantum variables qα∨ ni simply denote as ¯qi respectively. +Q∨ P , 1 ≤ i ≤ k, which we Theorem 3.10 (Equivariant quantum Pieri rule for F ℓn1,··· ,nk;n+1). For any 1 ≤ i ≤ k, 1 ≤ p ≤ ni and any u ∈ W P , we have σc[ni,p]⋆σu = X(d1,··· ,dk)∈Piei,p(u) p−di Xj=0 Xw ξni−di−j,p−di−j(µw·φd,u·τd,ni−di)σw ¯qd1 1 · · · ¯qdk k , where the last sum is over all w ∈ Per(d) satisfying w · φd ∈ Sni−di,j(u · τd). Let us say a few words on the constraints in the theorem. Given d = (d1, · · · , dk) of the form in Definition 3.7 with di = max{d1, · · · , dk} ≤ p, we have (see [5, 14]). d ∈ Piei,p(u) ⇐⇒ ℓ(u · τd) = ℓ(u) − ℓ(τd); w ∈ Per(d) ⇐⇒ ℓ(w · φd) = ℓ(w) + ℓ(φd). Remark 3.11. The above formula is different from the one given in [30] by Lam and Shimozono, who worked on the side of equivariant homology of affine Grass- mannians. In [30], special Schubert classes are of the form σspsp−1···s2s1sθ where θ denotes the highest root, and they generate QH ∗ T (F ℓ1,··· ,n;n+1) as well. These special classes, in general, are not pullback from H ∗(F ℓn1,··· ,nk;n+1), and therefore do not induce equivariant quantum Pieri rules for F ℓn1,··· ,nk;n+1 immediately. Example 3.9 (Continued). Note τ(1,1) = (234567), φ(1,1) = (1234), u · τ(0,0) = u = [3715246] and u · τ(1,1) = [3152467]. Write w · φd = (u · τd) · ζ for w ∈ Per(d)T(cid:0)S4−d2,j(u · τd)(cid:1) · (φd)−1. Denote a := µw·φd,u·τd,4−d2. Precisely, we have w · φd w d a 2 (24) (1, 1) [1325467] [1425367] [1326457] By Definition 3.4, we can write down ξ4−di−j,3−di−j(a) immediately. By abuse of notation, we simply denote each Schubert class σv as v. In conclusion, we have (15)(24) (24)(36) [2, 1] c[4, 3] ⋆ [3715246] = α2(α2 + α3 + α4)(α2 + · · · + α6)[3715246] ζ id (16) (35) (47) (16)(35) (16)(47) (35)(47) (16)(35)(47) (0, 0) 1 2 j 0 3 1 [3715246] [4715236] [3725146] [3716245] [4725136] [4716235] [3726145] [4726135] [3251467] [4251367] [3261457] (3, 2, 1, 0) (4, 3, 0) (4, 3, 2) (4, 1, 0) (5, 4) (5, 0) (5, 2) (6) (3, 2) (4) (2) [i1, · · · , ia] [6, 4, 5, 2, 3, 4] [6, 4, 5, 3, 4, 2, 3] [6, 4, 5, 2, 3, 4, 1, 2, 3] [6, 5, 4, 2, 3] [6, 4, 5, 3, 4, 2, 3, 1, 2] [6, 5, 4, 3, 2] Coincide with w · φd [6, 5, 4, 2, 3, 1, 2] [6, 5, 4, 3, 2, 1] [4, 2, 3, 1, 2] [4, 3, 2, 1] ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 17 + α2(α2 + · · · + α6)[3716245] + (α2 + α3 + α4)(α2 + · · · + α6)[4715236] +(cid:0)α2(α2 + α3 + α4) + α2(α1 + · · · + α6) + (α1 + · · · + α4)(α1 + · · · + α6)(cid:1)[3725146] + (α1 + 2α2 + α3 + α4 + α5 + α6)[3726145] + (α2 + · · · + α6)[4716235] + (α1 + 2α2 + 2α3 + 2α4 + α5 + α6)[4725136] + [4726135] + ¯q1 ¯q2[1326457] + ¯q1 ¯q2[1425367] + (α1 + 2α2 + α3 + α4)¯q1 ¯q2[1325467] 3.2. Proof of the equivariant quantum Pieri rule for F ℓn1,··· ,nk;n+1. This subsection is devoted to a proof of Theorem 3.10. We will show it by reducing all the relevant structure coefficients to certain structure coefficients of degree zero using Theorem 2.7, so that we can apply Robinson's equivariant Pieri rule [44]. 3.2.1. Robinson's equivariant Pieri rule. For u ∈ W and w ∈ Sr,j(u) where 0 ≤ j ≤ r ≤ n, we denote by {i1 < i2 < · · · < ij} the set of indices is such that is < r and is occurs in a cycle decomposition of u−1w. Here we mean the empty set if j = 0. In [44], Robinson introduced an associated element v[w,u,r] = [v(1) · · · v(n + 1)] that is obtained from u by moving the entries u(i1), · · · , u(ij) to positions r − j + 1, r − j + 2, · · · , r, respectively, and preserving the relative positions of all other entries. That is, v[w,u,r] is of the form [∗ · · · ∗ u(i1)u(i2) · · · u(ij)u(r + 1) · · · u(n + 1)]. The next equivariant Pieri rule for F ℓn+1 := F ℓ1,2,··· ,n;n+1 is due to Robinson. Proposition 3.12 (Theorem A of [44]). Let u ∈ W and 1 ≤ p ≤ r ≤ n. We have σc[r,p] ◦ σu = N v[w,u,r],0 c[r−j,p−j],v[w,u,r] σw in H ∗ T (F ℓn+1). p Xj=0 Xw∈Sr,j(u) In the following, we further reduce the structure coefficients in the above equi- variant Pieri rule to a more special type for complex Grassmannians. Corollary 3.13. Let 1 ≤ i ≤ k, 1 ≤ p ≤ ni and u ∈ W P . In H ∗ T (F ℓn1,··· ,nk;n+1), σc[ni,p] ◦ σu = p Xj=0 Xw∈Sni,j (u) ξni−j,p−j(µw,u,ni)σw. Furthermore, a coefficient ξni−j,p−j(µw,u,ni) vanishes if and only if p − j is larger than the first entry µT 1 of the transposed partition µT 1 , · · · ) ∈ Pn+1−ni+j,n+1. w,u,ni = (µT Proof. We let v ∈ Sn+1 be the (ni − j)-th Grassmannian permutation (which has at most a descent at the (ni − j)-th position) determined by the property that [v(1) · · · v(ni − j)] is an increasing sequence obtained from u by sorting the values {u(1), · · · , u(ni)} \ {u(d) d ≤ ni, d occurs in a cycle decomposition of u−1w}. Then by definition, µw,u,ni = ϕni−j(v) is a partition in Pni−j,n+1. Moreover, x := v−1v[w,u,ni] is in the Weyl subgroup generated by {sα α 6= αni−j, α ∈ ∆}, and ℓ(v[w,u,ni]) = ℓ(v)+ℓ(x). We notice that sgnα(c[ni −j, p−j]) for any α 6= αni−j. It follows immediately from Corollary 2.8 (1) that N v[w,u,ni],0 c[ni−j,p−j],v[w,u,ni] = N vx,0 c[ni−j,p−j],vx = N v,0 2 , · · · , µT 1 , µT Write the transpose of µw,u,ni as (µT Proposition 3.4 that ξni−j,p−j(µw,u,ni) 6= 0 if and only if p − j ≤ µT 1 . n+1−ni+j). It follows directly from (cid:3) c[ni−j,p−j],v = ξni−j,p−j(µw,u,ni). 18 YONGDONG HUANG AND CHANGZHENG LI 3.2.2. Proof of Theorem 3.10 for F ℓn+1. In this subsection, we will prove the the- orem for the case k = n: F ℓn1,··· ,nk;n+1 = F ℓ1,2,··· ,n;n+1 = F ℓn+1. This will be done by a combination of a number of claims. Our readers can first focus on the statements themselves without referring to the technical arguments, in order to get an outline of the proof of our Pieri rule. For F ℓn+1, there are n quantum variables q1, · · · , qn, and ni = i for i = 1, · · · , n. The statement to prove concerns the equivariant quantum multiplication by Schu- bert classes σc[ni,p] with c[ni, p] = sni−p+1 · · · sni−1sni. We notice that sgnr(c[ni, p]) is equal to 1 if r = ni, or 0 otherwise. On the other hand, for any nonzero effective coroot λ ∈ Q∨, there always exists a simple root In many cases, we can find α 6= αni with hα, λi > 1, α such that hα, λi > 0. implying that qλ never occurs in the corresponding multiplication by Theorem 2.7 (1). Therefore, qλ occurs in σc[ni,p] ⋆ σu only if λ is of particular type. Precisely, Claim A: Assume N w,λ n. Then we have c[ni,p],u 6= 0, where λ = d1α∨ 1 + · · · + dnα∨ (1) di ≤ p; (2) 0 ≤ d1 ≤ · · · ≤ di; (3) di ≥ · · · ≥ dn ≥ 0. Proof. Clearly, σc[ni,p] occurs in (σsni )p. Since N w,λ c[ni,p],u 6= 0, qλ occurs in the product (σsni )p ⋆ σu by the positivity property. By Proposition 2.1, there is 0 ≤ j for each ) where kj ≤ ni ≤ k′ kj +1 + · · · + α∨ k′ j j=1(α∨ kj + α∨ j. Thus di = m = max{d1, · · · , dn} ≤ p. This proves (1). m ≤ p such that λ =Pm If (2) did not hold, then {j j < i, dj > dj+1} is non-empty, so that we can take the minimum b of it. Then 1 ≤ b < i − 1 and db−1 ≤ db. If db−1 < db, then we have the following inequalities by noting sgnb(c[ni, p]) = 0: sgnb(w) + hαb, λi = sgnb(w) + (2db − db−1 − db+1) ≥ 2 > sgnb(c[ni, p]) + sgnb(u). This would imply N w,λ c[ni,p],u = 0 by Theorem 2.7 (1), which makes a contradiction. If db−1 = db, then for a := min{j dj = db} ≤ b − 1, we have da−1 < da. If c[ni,p],u = 0 again by using sgna. If da −da−1 = 1, then c[ni,p],u 6= 0, b+1i ≥ = 0, which makes a contradiction again. (cid:3) da −da−1 ≥ 2, we conclude N w,0 by using Corollary 2.8 (2) repeatedly, we have N wsasa+1···sb−1,λ′ where λ′ = λ−α∨ 2. It would follow that N wsasa+1···sb−1,λ′ b−1. Note hαb, λ′i = hαb, (db−1)α∨ Hence, 0 ≤ d1 ≤ · · · ≤ di. Similarly, we can show di ≥ · · · ≥ dn ≥ 0. Thanks to the above claim, we have m := max{d1, · · · , dn} = di if N w,λ = N w,λ b +db+1α∨ a −· · ·−α∨ c[ni,p],usasa+1···sb−1 c[ni,p],u 6= 0. c[ni,p],usasa+1···sb−1 b−1+dbα∨ Moreover, we can denote by 1 ≤ hr1 < · · · < hm−1 < hm ≤ lm < lm−1 < · · · < lr2 ≤ n all the jumps among dr's, namely dhj −1 < dhj for r1 ≤ j ≤ m, and dlj > dlj +1 for m ≥ j ≥ r2. As we will show in Claim C, there are exactly m increasing (decreasing) jumps (i.e., r1 = r2 = 1), together with lots of constraints on u and w. To make this conclusion, we denote j := shj · · · si−2si−1 · slj · · · si+2si+1, ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 19 τ (j) := shj shj +1 · · · slj , and φ(j) := shj shj +1 · · · slj −1 for any max{r1, r2} ≤ j ≤ m. Define λ(j−1) inductively by λ(m) := λ = d1α∨ 1 + · · · + dnα∨ n and λ(j−1) := λ(j) − αr ∨. lj Xr=hj We will prove the conclusion by induction on λ(j). Here is the first step of the induction, which we prove by applying Corollary 2.8 repeatedly. Claim B: N w,λ(m) c[ni,p],u 6= 0 if and only if all the following hold: (1) dhm−1 = dlm+1 = m − 1; (2) ℓ(uτ (m)) = ℓ(u) − ℓ(τ (m)) and ℓ(w φ(m)) = ℓ(w) + ℓ( φ(m)); (3) N w φ(m),λ(m−1) (4) N w φ(m),λ(m−1) c[ni−1,p−1],uτ (m) 6= 0; c[ni−1,p−1],uτ (m) = N w,λ(m) c[ni,p],u; (5) ℓ(umsi) = ℓ(u) − ℓ(msi). Proof. We first assume N w,λ(m) c[ni,p],u 6= 0 and discuss all the possibilities as follows. i) Assume hm = lm = i (i.e., both an increasing jump and a decreasing jump happen at the (i = ni)-th position). If (1) did not hold, then hαi, λi = (m − di−1) + (m − di+1) > 2. This would imply N w,λ c[ni,p],u = 0, making a contradiction. Hence, (1) holds, and hαi, λi = 2. In this case, τ (m) = φ(m) = si and m = id. Hence, all (2), (3), (4), (5) follow immediately from Theorem 2.7. ii) Assume hm < i and lm = i. Since N w,λ c[ni,p],u 6= 0, it follows that 1 ≥ sgnhm(u) = sgnhm(c[ni, p])+sgnhm(u) ≥ sgnhm(w)+hαhm , λi ≥ m−dhm−1 ≥ 1. Hence, all the inequalities are in fact equalities. Thus we have dhm−1 = m − = 1, ℓ(ushm) = ℓ(u) − 1, ℓ(wshm ) = ℓ(w) + 1, and consequently N N w,λ c[ni,p],u 6= 0 by Corollary 2.8 (2). For hm < a < ni = lm, we note sgna(c[ni, p]) = 0 and hαa, λ − α∨ a−1i = 1. Using Corollary 2.8 (2) repeatedly, hm we conclude ℓ(ushmshm+1 · · · si−1) = ℓ(u) − (i − hm), ℓ(wshm shm+1 · · · si−1) = wshm shm +1···si−1,λ−α∨ ℓ(w) + (i − hm), and N c[ni,p],u 6= 0. Since c[ni,p],ushm shm +1···si−1 the reduced structure coefficient is nonzero, wshm ,λ−α∨ c[ni,p],ushm − · · · − α∨ a−2 − α∨ = N w,λ i−2−α∨ −···−α∨ i−1 hm hm 2 ≥ hαi, λ − α∨ hm − · · · − α∨ i−2 − α∨ i−1i = 1 + m − di+1 ≥ 2. Hence, dlm+1 = di+1 = m − 1, and consequently and wshm shm +1···si−1,λ−α∨ c[ni,p],ushm shm +1···si−1 c[ni−1,p−1],uτ (m) = N N w φ(m),λ(m−1) hm −···−α∨ i−2−α∨ i−1 6= 0 by Theorem 2.7 (2). That is, the statements (1), (3), (4) hold. It is easy to see that (2) and (5) hold as well. iii) Assume hm = i and lm > i. The claim holds by similar arguments to ii). iv) Assume hm < i and lm > i. Again by similar arguments to ii), we conclude dhm−1 = dlm+1 = m − 1, ℓ(umsi) = ℓ(u) − ℓ(msi), ℓ(wm) = ℓ(w) + ℓ(m), and 0 6= N w,λ c[ni,p],u = N wm,λ(m−1)+α∨ c[ni,p],um i = N wm,λ(m−1) c[ni−1,p−1],umsi . For every i + 1 ≤ a ≤ lm, we have sgna(c[ni − 1, p − 1]) = 0, hαa, λ(m−1)i = 0 and ℓ(wshm · · · si−2si−1 · slm · · · sa+1sa) = ℓ(wshm · · · si−2si−1 · slm · · · sa+2sa+1) + 1. 20 YONGDONG HUANG AND CHANGZHENG LI By Corollary 2.8 (1), we conclude N wm,λ(m−1) c[ni−1,p−1],umsi = N wshm ···si−2si−1,λ(m−1) c[ni−1,p−1],umsisi+1···slm and ℓ(umsisi+1 · · · slm) = ℓ(umsi)−lm+i. Note msisi+1 · · · slm = τ (m)slm−1 · · · si+1si. It follows that ℓ(uτ (m)slm−1 · · · si+1si) = ℓ(u) − (lm − hm + 1) − lm + i = ℓ(u) − ℓ(τ (m)) − lm + i. As a consequence, we have ℓ(uτ (m)) = ℓ(u)−ℓ(τ (m)), and ℓ(uτ (m)slm−1 · · · sb+1sb) = ℓ(uτ (m)slm−1 · · · sb) + 1 for all i ≤ b ≤ lm − 1. Hence, by Corollary 2.8 (1), we have N wshm ···si−2si−1,λ(m−1) c[ni−1,p−1],uτ (m)slm−1···si+1si = N w φ(m),λ(m−1) c[ni−1,p−1],uτ (m) and ℓ(w φ(m)) = ℓ(wshm · · · si−2si−1) + lm − i = ℓ(w) + ℓ( φ(m)). In a summary, all (1) -- (5) hold. The other direction is obvious. (In fact, (4) is a consequence of the hypotheses (cid:3) (1), (2) and (3).) By using claims A and B, the next claim follows immediately by induction on λ(j). Claim C: N w,λ c[ni,p],u 6= 0 only if all the following hold: (a) r1 = r2 = 1, namely there are exactly 2m jumps among [0, d1, · · · , dn, 0]. (b) ℓ(u · τ (m) · · · τ (1)) = ℓ(u) − ℓ(τ (m) · · · τ (1)) and ℓ(w · φ(m) · · · φ(1)) = ℓ(w) + ℓ( φ(m) · · · φ(1)). (c) ℓ(umsim−1si−1 · · · 1si−m+1) = ℓ(u) −Pm Whenever both (a) and (b) hold, we have j=1 ℓ(jsi−m+j). N w· φ(m)··· φ(1),0 c[ni−m,p−m],u·τ (m)···τ (1) = N w,λ c[ni,p],u. Since ni = i in the case of F ℓn+1, we have τ (j) = τ (j) and φ(j) = φ(j) for all j. Therefore we have τ (m) · · · τ (1) = τd and φ(m) · · · φ(1) = φd. Hence, we finish the proof of Theorem 3.10 for F ℓn+1, by using Corollary 3.13 together with the fact that the hypotheses (a), (b) in Claim C are equivalent to the hypotheses d ∈ Piei,p(u) and w ∈ Per(d). In order to show the general case in next subsection, we make one more claim. Claim D: Let 1 ≤ p ≤ r ≤ n and u ∈ W . Suppose qd1 σc[r,p] ⋆ σu in QH ∗ then we have ℓ(usj) < ℓ(u). n occurs in the product T (F ℓn+1). If j coincides with some jump hb or lb with 1 ≤ b < m, 1 · · · qdn 1 + · · · + dnα∨ Proof. By the hypothesis, there exists w ∈ W such that N w,λ d1α∨ c[r,p],u 6= 0, where λ = n . By Claim C (b), (c), we have ℓ(u · τ (m) · · · τ (1)) = ℓ(u) − j=1 ℓ(jsr−m+j). Thus we have ℓ(usa) < ℓ(u), whenever a reduced expression of τ (m) · · · τ (1) or msr · · · 1sr−m+1 starts with sa. Hence, we are done, due to the following: ℓ(τ (m) · · · τ (1)) and ℓ(umsrm−1sr−1 · · · 1sr−m+1) = ℓ(u) −Pm (τ (m) · · · τ (1))−1(αj) =(−(αlb+1−b+1 + · · · αlb−b + αlb−b+1) if hb = hb+1 − 1 if hb < hb+1 − 1 −(αhb−b+1 + · · · αlb−b + αlb−b+1) . ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 21 Hence, τ (m) · · · τ (1) admits a reduced expression starting with shb . Similarly, we conclude (msr · · · 1sr−m+1)−1(αlb ) 6∈ R+ by direct calculations. Consequently, msr · · · 1sr−m+1 admits a reduced expression starting with slb. (cid:3) 3.2.3. Proof of Theorem 3.10 for F ℓn1,··· ,nk;n+1. In this subsection, we prove the theorem for general F ℓn1,··· ,nk;n+1 by reducing all the relevant structure coefficients to the case of F ℓn+1, thanks to the equivariant Peterson-Woodward comparison formula. We will use ¯ to distinguish from the notations for F ℓn+1. For instance, we denote by ¯qj = qα∨ T (F ℓn1,··· ,nk;n+1). For nj nj + Q∨ P and u, w ∈ W P , by Proposition 2.10 we have the quantum variables in QH ∗ ¯djα∨ +Q∨ P λP =Pk j=1 N w,λP c[ni,p],u = N wωP ωP ′ ,λB c[ni,p],u for a unique λB = d1α∨ We investigate all the nonzero N w,λP 1 + · · · + dnα∨ n. c[ni,p],u. By Claim C (a), we can denote all the jumps of the sequence [0, d1, · · · , dn, 0] as 1 ≤ h1 < · · · < hm ≤ lm < · · · < l1 ≤ n. Since λP = λB + Q∨ P , we have dnj = ¯dj for all 1 ≤ j ≤ k. Claim E: Assume N w,λP sequence [0, ¯d1, · · · , ¯dk, 0]: c[ni,p],u 6= 0. Then there are 2m jumps in total among the 1 ≤ ¯h1 < · · · < ¯hm ≤ ¯lm < · · · ¯l1 ≤ k, which are given by the jumps for λB. Precisely, for all 1 ≤ j ≤ m, we have hj = n¯hj , lj = n¯lj , and ¯d¯hj = dhj = dlj = ¯d¯lj = j. Proof. It follows from Claim A and Claim B (1) that dhm = m ≥ dhm+1 ≥ m − 1 and dhm−1 = m − 1. Hence, hαhm , λBi ∈ {1, 2}. Since hα, λBi ∈ {0, −1} for all α ∈ ∆P , we have αhm 6∈ ∆P . Thus hm ∈ {n1, · · · , nk}, i.e., hm = n¯hm for some 1 ≤ ¯hm ≤ k. Similarly, we conclude dlm = m, lm ∈ {n1, · · · , nk}, and hence lm = n¯lm. It follows that ¯di = dni = m = max{d1, · · · , dn} = max{ ¯d1, · · · , ¯dk}, and the first two jumps around ¯di occur exactly on ¯hm ≤ ¯lm. By Claim D, we have ℓ(usa) < ℓ(u) for all a ∈ {h1, · · · , hm−1, l1, · · · , lm−1}. This implies αa 6∈ ∆P , since u ∈ W P . Thus {h1, · · · , hm−1, l1, · · · , lm−1} ⊂ {n1, · · · , nk}. The statement becomes a direct consequence of Claim C (a). (cid:3) By Claim C, we have N wωP ωP ′ ,λB c[ni,p],u = N wωP ωP ′ φ(m)··· φ(1),0 c[ni−m,p−m],uτ (m)···τ (1) ; ℓ(uτ (m) · · · τ (1)) = ℓ(u) − ℓ(τ (m) · · · τ (1)) and ℓ(wωP ωP ′ φ(m) · · · φ(1)) = ℓ(wωP ωP ′ ) + ℓ( φ(m) · · · φ(1)), with φ(j) = shj · · · slj −1 = sn¯hj sn¯hj +1 · · · sn¯lj −1, τ (j) = shj · · · slj = sn¯hj · · · sn¯lj . Note ∆P ′ = {α ∈ ∆P hα, λBi = 0} = ∆P \ {αnhj −1, αnlj +1 j = 1, · · · , m} where ∆P = ∆ \ {αn1 , · · · , αnk }. It follows that ωP ωP ′ = u1 · · · umvm · · · v1, with uj := sn¯hj −1+1sn¯hj −1+2 · · · sn¯hj −1 and vj := sn¯lj +1−1sn¯lj +1−2 · · · sn¯lj +1. 22 YONGDONG HUANG AND CHANGZHENG LI Clearly, u1, · · · , um, v1, · · · , vm are pairwise commutative. Denote v[j−1] j := sn¯lj +1−jsn¯lj +1−j−1 · · · sn¯lj −j+2, which does not contain sni−m. It follows that ωP ωP ′ φ(m) · · · φ(1) = um φ(m) · · · u1 φ(1)v[0] 1 v[1] = φ(m) · · · φ(1)v[0] For ¯d = ( ¯d1, · · · , ¯dk), we recall φ¯d = φ(m) · · · φ(1). Since w ∈ W P , we have 2 · · · v[m−1] m 1 v[1] 2 · · · v[m−1] m . ℓ(wωP ωP ′ φ(m) · · · φ(1)) = ℓ(w) + ℓ(φ¯d) + ℓ(v[0] 1 v[1] 2 · · · v[m−1] m ). Hence, by Corollary 2.8(1), we have N wωP ωP ′ φ(m)··· φ(1),0 c[ni−m,p−m],uτ (m)···τ (1) = N wφ¯dv = N φ¯d,0 [0] 1 v [1] 2 ···v [m−1] m ,0 c[ni−m,p−m],uτ (m)···τ (1) c[ni−m,p−m],uτ (m)···τ (1)·(v [0] 1 v [1] 2 ···v [m−1] m )−1 = N w·φ¯d,0 c[ni−m,p−m],u·τ¯d and ℓ(u · τ¯d) = ℓ(u) − ℓ(τ (m) · · · τ (1)) − ℓ((v[0] Hence, 1 v[1] 2 · · · v[m−1] m )−1) = ℓ(u) − ℓ(τ¯d). N w,λP c[ni,p],u = N wωP ωP ′ ,λB c[ni,p],u = N w·φ¯d,0 c[ni−m,p−m],u·τ¯d . Then we are done by Corollary 3.13. 3.3. Specialization to complex Grassmannians. In this subsection, we will further simplify our equivariant quantum Pieri rule for the special case Gr(m, n + 1) = F ℓm;n+1. The bijection map ϕm : W P ≃→ Pm,n+1 sends c[m, p] = sm−p+1 · · · sm−1sm to 1p := (1, · · · , 1, 0, · · · , 0) ∈ Pm,n+1 Therefore we will also denote the special Schubert classes σc[m,p] as σ1p that σ1p of the dual of the tautological subbundle (see e.g. [42, §5.1]). is related with (but different from) the p-th equivariant Chern class cT . We remark p (S∗) (p copies of 1). The equivariant quantum multiplication by σ1 was given by Mihalcea [40]. Here we will give a neat formula of the multiplication by all σ1p by simplifying Theorem 3.10. We remark that the classical part of our formula, i.e., the equivariant Pieri rule, is different from those known rules in [16, 27]. It is obtained by simplifying Robinson's Pieri rule in a purely combinatorial way. Nevertheless, our formulation has inspired the second author and Ravikumar to find an equivariant Pieri rule for Grassmannians of all classical Lie types [38] in a geometric way. Definition 3.14. Let ν = (ν1, · · · , νm) and η = (η1, · · · , ηm) be partitions in Pm,n+1 with ηi − νi ∈ {0, 1} for all 1 ≤ i ≤ m. Denote by j1 < j2 < · · · < jm−r all those ηji = νji . We define a partition ην in Pm−r,n+1 associated to (η, ν) by ην := (νj1 − j1 + r + 1, νj2 − j2 + r + 2, · · · , νjm−r − jm−r + m). The above definition can be alternatively described by the language of Young diagrams as follows. We also provide an example illustrated by Figures 1 and 2. Definition/Example 3.15. Let ν, η be partitions in Pm,n+1 such that the Young diagram of η is obtained by adding a vertical strip to the Young diagram of ν. Denote by r the number of boxes in the strip η/ν. We define an associated partition ην in Pm−r,n+1 by a simple join-and-cut operation as follows. ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 23 Step 1: Whenever a row of the Young diagram of η inside the m × (n + 1 − m) rectangle does not contain a box in the strip η/ν, we add A boxes, where A counts the remaining rows of the rectangle below the given one. We then move them to an (m − r) × (n + 1 − m + r) rectangle preserving the relative positions, which could be beyond the boundary of the rectangle on the right. Step 2: For each row in the (m − r) × (n + 1 − m + r) rectangle, we remove B boxes, where B counts the remaining rows of the rectangle below the given one. As a result, we obtain a partition in Pm−r,n+1, denoted as ην. η = ν, then r = 0 and ην = ν. In particular if Figure 1 illustrates the case of ν = (6, 3, 2, 2, 0, 0) and η = (6, 3, 3, 2, 1, 1) in P6,13, for which we have r = 3. Then the associated partition in P3,13 is given by ην = (9, 6, 4) as illustrated by Figure 2. Figure 1. Young diagrams of partitions η, ν ν = (6, 3, 2, 2, 0, 0) η = (6, 3, 3, 2, 1, 1) η/ν : vertical strip Figure 2. Associated parition ην by a join-and-cut operation Step 1: join Step 2: cut (11, 7, 4) /∈ P3,13 ην = (9, 6, 4) ∈ P3,13 Lemma 3.16. Let 1 ≤ p ≤ m and v, w ∈ W P . Denote ν := ϕm(v) and η := ϕm(w). Then w ∈ Sm,p(v) if and only if both of the following hold: (i) η = ν + p; (ii) ηj − νj ∈ {0, 1} for all 1 ≤ j ≤ m. Furthermore when this holds, we have µw,v,m = ην. Proof. We assume w ∈ Sm,p(v) first. Write w = vζ1 · · · ζd where ζ1, · · · , ζd are pair- wise disjoint cycles. Since v, w ∈ W P , it follows that vζ1 · · · ζk ∈ W P for all 1 ≤ k ≤ d. Denote pk := ℓ(vζ1 · · · ζk) − ℓ(vζ1 · · · ζk−1). Then vζ1 · · · ζk ∈ Sm,pk (vζ1 · · · ζs−1) follows from the definition. In particular, write ζ1 = (rip1 · · · i2i1), then we have (1) i1, · · · , ip1 ≤ m < r; (2) v(r) > v(i1) > · · · > v(ip1 ). Claim F: Denote i0 := r. We have (a)[ip1 , · · · , i2, i1] = [i1−p+1, · · · , i1−1, i1]; Assuming the above claim first, we write ϕm(vζ1) = (η(1) any 1 ≤ s ≤ m distinct from those m + 1 − ij, we have (b) v(ij) = v(ij+1)+1, 0 ≤ j ≤ p1−1. 1 , · · · , η(1) m ) =: η(1). For η(1) s = vζ1(m + 1 − s) − (m + 1 − s) = v(m + 1 − s) − (m + 1 − s) = ηs. 24 YONGDONG HUANG AND CHANGZHENG LI By Claim F (b), we have η(1) m+1−ij = vζ1(ij) − ij = v(ij−1) − ij = v(ij) + 1 − ij = ηm+1−ij + 1. Together with Claim F (a), we obtain (∗∗) : η(1) = (ν1, · · · , νm−i1 , νm−i1+1 + 1, · · · , νm−i1+p1 + 1, νm−i1+p1+2, · · · , νm), where p1 ≤ i1 ≤ m. Thus if d = 1, then we are done. Now we assume d ≥ 2. Write ζ2 = (r′i′ p2 · · · i′ 2i′ s = η(1) 1, · · · , m+1−i′ 1). Since ζ1, ζ2 are disjoint cycles, 1 , · · · , η(2) m ). s whenever s ∈ {m + 1 − i1, · · · , m + 1 − ip1}. Using the same s ∈ {0, 1} for all s ∈ {1, · · · , m} \ {m + s − νs ∈ {0, 1} for all 1 ≤ s ≤ m. Hence, m+1−ij 6∈ {m+1−i′ Then η(2) arguments as above, we conclude η(2) s − η(1) 1 − i1, · · · , m + 1 − ip1}. Thus we have η(2) both (i) and (ii) hold by induction on k. p2}. Write ϕm(vζ1ζ2) = η(2) = (η(2) As a direct consequence of the above arguments, we observe that 1 ≤ a ≤ m occurs in some cycle ζs if and only if ηm+1−a − νm+1−a = 1. Hence, [v(m + 1 − j1), v(m + 1 − j2), · · · , , v(m + 1 − jm−p)] is the decreasing sequence obtained by sorting {v(1), · · · , v(m)} \ {v(a) a occurs in ζs for some 1 ≤ s ≤ d}. Hence, the partition µw,v,m = (µ1, · · · , µm−p) in Pm−p,n+1 coincides with ην , by noting µi = v(m+1−ji)−(m−p+1−i) = v(m+1−ji)−(m+1−ji)+p+i−ji = νji +p+i−ji. On the other hand, we assume the hypotheses (i) and (ii) both hold now. If ϕm(w) = η is given by (∗∗), we define ij := i1 − j + 1 for every 1 ≤ j ≤ p1 and define r to be the element satisfying v(r) = v(i1) + 1. It is easy to check r > m. Consequently, we have w = v(rip1 · · · i2i1) ∈ Sm,p1 (v) with p1 = ℓ(w) − ℓ(v) = p. In general, there are d nests of consecutive 1, for which we can construct pairwise disjoint cycles ζ1, · · · , ζd by induction, such that w = vζ1 · · · ζd ∈ Sm,p(v). It remains to show Claim F. It follows from v ∈ W P and properties (1), (2) that m ≥ i1 > i2 > · · · > ip1. If (a) did not hold, then ij > ij+1 + 1 for some 1 ≤ j ≤ p1 − 1, and we would deduce a contradiction: v(ij) = vζ1(ij+1) < vζ1(ij+1 + 1) = v(ij+1 + 1) < v(ij). Hence, (a) holds. If (b) did not hold, then v(ij) > v(ij+1)+1 for some 0 ≤ j ≤ p1−1. If j > 0, then ij+1 + 1 = ij by Claim F (a), and consequently v(r) > v(ij+1) + 1 = v(a) for some m + 1 < a < r. In this case, we deduce a contradiction: v(a) = vζ1(a) < vζ1(r) = v(ip1 ) ≤ v(ij+1) = v(a) − 1. If j = 0, then v(r) > v(i1) + 1 = v(b) for some b < r. If b > m, then we would have v(b) = vζ1(b) < vζ1(r) = v(ip1 ) ≤ v(i1) = v(b) − 1. If b ≤ m, then b > i1 and consequently we have v(r) = vζ1(i1) < vζ1(b) = v(b) < v(r). Either cases deduces a contradiction again. (cid:3) Using the above lemma, we can simplify Theorem 3.10 for the special case of complex Grassmannians, and therefore obtain the following. The proof is essen- tially the same as Corollary 3.3 of [14]. There is only one quantum variable in QH ∗ T (Gr(m, n + 1)), which we simply denote as q. ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 25 Theorem 3.17 (Equivariant quantum Pieri rule for complex Grassmannians). Let 1 ≤ p ≤ m and ν = (ν1, · · · , νm) ∈ Pm,n+1. In QH ∗ T (Gr(m, n + 1)), we have ξm−r,p−r(ην)ση + ξm−1−r,p−1−r(κ′ ν′ )σκq, σ1p ⋆ σν = p Xr=0Xη p−1 Xr=0Xκ where the second sum is over those η = (η1, · · · , ηm) ∈ Pm,n+1 satisfying η = ν+r and ηi −νi ∈ {0, 1} for all 1 ≤ i ≤ m; the q-terms occur only if ν1 = n+1−m, and when this holds, the last sum is over those κ = (κ1, · · · , κm−1, 0) ∈ Pm,n+1 such that κ′ := (κ1 + 1, · · · , κm−1 + 1) and ν′ := (ν2, · · · , νm) satisfy κ′ = ν′ + r and κi + 1 − νi+1 ∈ {0, 1} for all 1 ≤ i ≤ m − 1. Proof. Denote v = ϕ−1 k = 1, and hence Pie1,p ⊂ {(0), (1)}. By Theorem 3.10, we have m (ν). Using the same notations as in Theorem 3.10, we have σ1p ⋆σν = p Xj=0 Xw∈Sm,j ξm−j,p−j(µw,v,m)σw+ǫ p−1 Xj=0Xw ξm−1−j,p−1−j(µw·φ(1),v·τ(1),m−1)σwq, where ǫ = 1 if ℓ(v · τ(1)) = ℓ(v) − ℓ(τ(1)), or 0 otherwise; the last sum is over those w ∈ Pie((1)) satisfying w · φ(1) ∈ Sm−1,j(v · τ(1)). The classical part of the formula to prove is referred to as the equivariant T (Gr(m, n + 1)) ֒→ T (F ℓn+1) that ξm−j,p−j(µw,v,m) 6= 0 only if w ∈ W P . Hence, the equivariant Pieri rule. It follows from the canonical injective morphism H ∗ H ∗ Pieri rule follows directly from Lemma 3.16. When d = (1), we have τd = smsm+1 · · · sn and φd = s1s2 · · · sm−1. Note v(m) = max{v(1), · · · , v(m)} and v(n + 1) = max{v(m + 1), · · · , v(n + 1)}. As a consequence, the following are all equivalent: i) ℓ(v·τ(1)) = ℓ(v)−ℓ(τ(1)); ii) vτd(αn) ∈ R+; iii) v(m) > v(n+1); iv) v(m) = n+1. Hence, we have ǫ = 1 if and only if ν1 = v(m) − m = n + 1 − m. Furthermore when this holds, v · τ(1) is a Grassmannian permutation for Gr(m − 1, n + 1), which corresponds to the partition ϕm−1(v · τ(1)) = (ν2, · · · , νm−1) =: ν′ in Pm−1,n+1 (by noting vτd(j) = v(j) for 1 ≤ j ≤ m − 1). Write ϕm(w) = (κ1, · · · , κm) =: κ. Then the following are equivalent: (1) w ∈ Pie((1)); (2) ℓ(w · φ(1)) = ℓ(w) + ℓ(φ(1)); (3)κm = 0. It follows that w · φ(1) is a Grassmannian permutation for Gr(m − 1, n + 1), which corresponds to the partition ϕm−1(w · φ(1)) = (κ1 + 1, · · · , κm−1 + 1) =: κ′. Hence, the q-part also becomes a direct consequence of Lemma 3.16. (cid:3) Remark 3.18. The non-equivariant quantum Pieri rule [2] can be obtained by using Proposition 11.10 of [29] and the Pieri-type formula of H∗(ΩSU (n + 1)) in [28]. It will be very interesting to generalize this approach to the equivariant quantum cohomology of complex (or more generally, cominuscule) Grassmannians. Example 3.19. Among the product σ(1,1,0) ⋆ σ(4,2,1) in H ∗ q0ση and q1σκ can be read off from the following figure: T (Gr(3, 7)), two terms η/ν ην = (5, 3) κ = (1, 0, 0) κ′ = ν ′ = κ′ ν ′ = (2, 1) ξ3−1,2−1(ην) = α1 + 2α2 + 2α3 + α4 + α5 + α6 ξ3−1−0,2−1−0(κ′ ν ′ ) = α1 + α2 + α3 26 YONGDONG HUANG AND CHANGZHENG LI By calculating the remaining terms in the product by Theorem 3.17, we have σ(1,1,0) ⋆ σ(4,2,1) =(α1 + α2 + α3)(α1 + · · · + α6)σ(4,2,1) + (α1 + 2α2 + 2α3 + α4 + α5 + α6)σ(4,2,2) + (α1 + · · · + α6)σ(4,3,1) + σ(4,3,2) + qσ(1,1,0) + qσ(2,0,0) + (α1 + α2 + α3)qσ(1,0,0) Corollary 3.20. In QH ∗ T (Gr(m, n + 1)), we have σ1p σ1m ⋆ σ(n+1−m,0,··· ,0) = σ1p ⋆ σ(n+1−m,0,··· ,0) = (α1 + · · · + αn)σ(n+1−m,1,··· ,1) + q. ◦ σ(n+1−m,0,··· ,0), for 1 ≤ p < m; Proof. Let 1 ≤ p ≤ m and ν = (n+1−m, 0, · · · , 0). It follows directly from Theorem 3.17 that all possible partitions are given by η(r) := (n + 1 − m, 1, · · · , 1, 0, · · · , 0) where η(r) = n + 1 − m + r, 0 ≤ r ≤ m − 1; and the q-terms occur only if there exists κ satisfying p − 1 ≥ r = κ′ = m − 1 + κ ≥ m − 1. Hence, if p < m, then σ1p ⋆ σ(n+1−m,0,··· ,0) involves no q-terms. If p = m, then κ = 0, namely (0, · · · , 0) is the only partition satisfying the required properties. Hence, σ1m ⋆ σν = ξm−r,m−r(η(r) ν )ση + ξm−1−(m−1),m−1−(m−1)(cid:0)(1, · · · , 1)(0,··· ,0)(cid:1)σidq, m−1 Xr=0 in which we note ξ0,0(cid:0)(1, · · · , 1)(0,··· ,0)(cid:1) = 1. By definition, we have η(r) 1 − m + r, 0, · · · , 0) ∈ Pm−r,n+1. Hence, ξm−r,m−r(η(r) Furthermore when r = m − 1, we have ν = (n + ν ) = 0 unless r = m − 1. ξm−r,m−r(η(r) ν ) = ξ1,1((n, 0, · · · , 0)) = snsn−1 · · · s2(α1) = α1 + · · · + αn. Hence, the statement follows. (cid:3) Appendix: Equivariant quantum Giambelli formula for complex Grassmannians We expect out equivariant quantum Pieri rule to have further applications in the equivariant quantum Schubert calculus. To illustrate our expectation, we will reprove [42, Theorem 3.22]. That is, we will study QH ∗ T (Gr(m, n + 1)), giving alternative proofs of the ring presentation and the equivariant quantum Giambelli formula. In our approach, we use the equivariant quantum Pieri rule as in Theorem 3.17, together with the equivariant Giambelli formula [26, 42]. This is completely similar to the one given by Buch [4] for the non-equivariant quantum cohomology QH ∗(Gr(m, n + 1)). We follow [42] for the next facts on equivariant cohomology H ∗ Treat S = Z[α1, · · · , αn] as a subring of Z[t] = Q[t1, · · · , tn+1] via T (Gr(m, n + 1)). αi 7→ tn+2−i − tn+1−i, i = 1, · · · , n. By convention, we denote ti = 0 if i ≤ 0 or i ≥ n + 2. Let ei = ei(x1, · · · , xm; t), i = 1, · · · , m (resp. hj = hj(x1, · · · , xm; t), j = 1, · · · , · · · , n + 1 − m) denote the ele- mentary (resp. complete) homogeneous factorial Schur functions. By convention, we denote e0 = h0 = 1, ei = 0 if i < 0 or i > m, and hj = 0 if j < 0 or j > n + 1 − m. Define τ s inductively by τ 0ep := ep and τ sep := τ s−1ep + (ts − ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 27 tm−p+s+1)τ s−1ep−1. Denote by λT := (λT 1 , · · · , λT n+1−m) ∈ Pn+1−m,n+1 the trans- pose of a given partition λ ∈ Pm,n+1. Let Hk := det(cid:0)τ j−1e1+j−i(cid:1)1≤i,j≤k. We will need the next lemma, which follows directly from equation (2.10) of [42] Lemma 3.21. For any M ∈ Z+, in H ∗ T (Gr(m, n + 1)), we have (−1)pσ1p ◦ τ 1−M HM−p = 0 m Xp=0 with τ 0Hj := Hj and τ −sHj := τ 1−sHj + (tj+m−s − t1−s)τ 1−sHj−1. Theorem 3.22 (Equivariant quantum Giambelli formula; Theorem 4.2 of [42]). There is a canonical isomorphism of S[q]-algebras, S[q][e1, · · · , em]/hHn−m+2, · · · , Hn, Hn+1 + (−1)mqi −→ QH ∗ T (Gr(m, n + 1)), . Under this isomorphism, σr = Hr for r ≤ n + 1 − m, and defined by ep 7→ σ1p i +j−i(cid:1)1≤i,j≤n+1−m. Proof. It is sufficient (1) to calculate Hk and det(cid:0)τ j−1eλT σλ = det(cid:0)τ j−1eλT equivariant quantum product and (2) to subtract the quantum corrections, by an equivariant quantum extension of [15, Proposition 11] (or [45, Proposition 2.2]). The known ring presentation of H ∗ T (Gr(m, n + 1)) is read off from the first half of the statement by evaluating q = 0, and the known equivariant Giambelli formula is exactly of the same form as in the second half. Thus for any λ ∈ Pm,n+1, we have i +j−i(cid:1) with respect to the det(cid:0)τ j−1eλT i +j−i(cid:1)1≤i,j≤n+1−m = σλ + q · gλ in QH ∗ T (Gr(m, n + 1)), for some element gλ ∈ QH ∗ QH ∗ T (Gr(m, n + 1)). The determinant det(cid:0)τ j−1eλT i +j−i(cid:1) in T (Gr(m, n + 1)) ⊗S Z[t] is a summation of the form f (t)σ1i1 ⋆ σ1i2 ⋆ · · · ⋆ σ1in+1−m . By Theorem 3.17 and induction, the expansion of σ1i1 ⋆ σ1i2 ⋆ · · · ⋆ σ1ij involves no q-terms, and all Schubert classes in the expansion are of the form σµ, µ = (µ1, · · · , µm) with µ1 ≤ j. Hence, gλ = 0. That is, the second part of the statement holds, by noting that the determinant lies in QH ∗ T (Gr(m, n + 1)). In particular, we have Hr = σr in QH ∗ T (Gr(m, n + 1)), for r = 0, 1, . . . , n − m + 1. Clearly, τ j−1e1+j−i is zero if 1 + j < i, or of degree 1 + j − i otherwise. Hence, deg Hr = r. Since deg q = hc1(TGr(m,n+1)), σsk i = n + 1, it follows that no q-term is involved in the expansion of Hr in QH ∗ T (Gr(m, n + 1)), whenever r < n + 1. T (Gr(m, n + 1)) it follows from Lemma 3.21 with respect to M = n + 1 that In H ∗ Hn+1 = (−1)m−1emHn+1−m+ fi(t)Hi+ gm,j(t)emHj + gp,j(t)epHr n Xi=0 n−m Xj=0 m−1 Xp=1 n−p Xr=0 for some fi, gm,j, gp,r ∈ Z[t]. Now we compute the q-terms in the expansion of the right-hand side as multiplications in QH ∗ T (Gr(m, n + 1)) ⊗S Z[t]. With respect to the equivariant quantum multiplications, we have shown Hr = 0 if n − m + 2 ≤ r ≤ n, and Hr = σr if 0 ≤ r ≤ n − m + 1. Hence, it follows from Theorem 3.17 Hr) involves no q-terms. Hence, the only q-term in the expansion of Hn+1 comes from (resp. Corollary 3.20) thatPn−m ⋆σp (resp. Pm−1 p=1 Pn−p j=0 gm,j(t)σ1m r=0 gp,j(t)σ1p (−1)m−1σ1m σn−m+1 = (−1)m−1((α1 + · · · + αn)σ(n+1−m,1,··· ,1) + q), 28 YONGDONG HUANG AND CHANGZHENG LI by Corollary 3.20 again. Hence, Hn+1 = (−1)m−1q in QH ∗ T (Gr(m, n + 1)). (cid:3) References [1] D. Anderson and L. Chen, Equivariant quantum Schubert polynomials, Adv. Math. 254 (2014), 300 -- 330. [2] A. Bertram, Quantum Schubert calculus, Adv. Math. 128 (1997), no. 2, 289 -- 305. [3] S. C. Billey, Kostant polynomials and the cohomology ring for G/B, Duke Math. J. 96 (1999), no. 1, 205 -- 224. [4] A. S. Buch, Quantum cohomology of Grassmannians, Compositio Math. 137 (2003), no. 2, 227 -- 235. [5] [6] , Quantum cohomology of partial flag manifolds, Trans. Amer. Math. Soc. 357 (2005), no. 2, 443 -- 458 (electronic). , Mutations of puzzles and equivariant cohomology of two-step flag varieties, to appear in Ann. of Math. (2). Preprint at arxiv: math.CO/1401.3065. [7] A. S. Buch, A. Kresch, K. Purbhoo, and H. Tamvakis, The puzzle conjecture for the coho- mology of two-step flag manifolds, preprint at arXiv: math.CO/1401.1725. [8] A. S. Buch, A. Kresch, and H. Tamvakis, Gromov-Witten invariants on Grassmannians, J. Amer. Math. Soc. 16 (2003), no. 4, 901 -- 915 (electronic). [9] , Quantum Pieri rules for isotropic Grassmannians, Invent. Math. 178 (2009), no. 2, 345 -- 405. [10] A. S. Buch and L. C. Mihalcea, Quantum K-theory of Grassmannians, Duke Math. J. 156 (2011), no. 3, 501 -- 538. [11] A. S. Buch and R. Rim´anyi, Specializations of Grothendieck polynomials, C. R. Math. Acad. Sci. Paris 339 (2004), no. 1, 1 -- 4. [12] P. E. Chaput, L. Manivel, and N. Perrin, Quantum cohomology of minuscule homogeneous spaces, Transform. Groups 13 (2008), no. 1, 47 -- 89. [13] P. E. Chaput and N. Perrin, On the quantum cohomology of adjoint varieties, Proc. Lond. Math. Soc. (3) 103 (2011), no. 2, 294 -- 330. [14] I. Ciocan-Fontanine, On quantum cohomology rings of partial flag varieties, Duke Math. J. 98 (1999), no. 3, 485 -- 524. [15] W. Fulton and R. Pandharipande, Notes on stable maps and quantum cohomology, Alge- braic geometry -- Santa Cruz 1995, Proc. Sympos. Pure Math., vol. 62, Amer. Math. Soc., Providence, RI, 1997, pp. 45 -- 96. [16] L. Gatto and T. Santiago, Equivariant Schubert calculus, Ark. Mat. 48 (2010), no. 1, 41 -- 55. [17] W. Graham, Positivity in equivariant Schubert calculus, Duke Math. J. 109 (2001), no. 3, 599 -- 614. [18] J. E. Humphreys, Introduction to Lie algebras and representation theory, Graduate Texts in Mathematics, vol. 9, Springer-Verlag, New York, 1978. [19] , Linear algebraic groups, Springer-Verlag, New York, 1975. Graduate Texts in Math- ematics, No. 21. [20] T. Ikeda, L.C. Mihalcea, and H. Naruse, Factorial P - and Q-Schur functions represent equi- variant quantum Schubert classes, preprint at arxiv: math.CO/1402.0892. [21] B. Kim, On equivariant quantum cohomology, Internat. Math. Res. Notices 17 (1996), 841 -- 851. [22] A. Knutson, A Schubert Calculus recurrence from the non-complex W -action on G/B, preprint at arxiv: math.CO/0306304. [23] A. Knutson and T. Tao, Puzzles and (equivariant) cohomology of Grassmannians, Duke Math. J. 119 (2003), no. 2, 221 -- 260. [24] B. Kostant and S. Kumar, The nil Hecke ring and cohomology of G/P for a Kac-Moody group G, Adv. in Math. 62 (1986), no. 3, 187 -- 237. [25] V. Kreiman, Equivariant Littlewood-Richardson skew tableaux, Trans. Amer. Math. Soc. 362 (2010), no. 5, 2589 -- 2617. [26] V. Lakshmibai, K. N. Raghavan, and P. Sankaran, Equivariant Giambelli and determinantal restriction formulas for the Grassmannian, Pure Appl. Math. Q. 2 (2006), no. 3, Special Issue: In honor of Robert D. MacPherson., 699 -- 717. [27] D. Laksov, Schubert calculus and equivariant cohomology of Grassmannians, Adv. Math. 217 (2008), no. 4, 1869 -- 1888. ON EQUIVARIANT QUANTUM SCHUBERT CALCULUS FOR G/P 29 [28] T. Lam, L. Lapointe, J. Morse, and M. Shimozono, Affine insertion and Pieri rules for the affine Grassmannian, Mem. Amer. Math. Soc. 208 (2010), no. 977. [29] T. Lam and M. Shimozono, Quantum cohomology of G/P and homology of affine Grassman- nian, Acta Math. 204 (2010), no. 1, 49 -- 90. [30] [31] , Equivariant Pieri rule for the homology of the affine Grassmannian, J. Algebraic Combin. 36 (2012), no. 4, 623 -- 648. , Quantum double Schubert polynomials represent Schubert classes, Proc. Amer. Math. Soc. 142 (2014), no. 3, 835 -- 850. [32] A. Lascoux and M.-P. Schutzenberger, Polynomes de Schubert, C. R. Acad. Sci. Paris S´er. I Math. 294 (1982), no. 13, 447 -- 450 (French, with English summary). [33] N. C. Leung and C. Li, Functorial relationships between QH ∗(G/B) and QH ∗(G/P ), J. Differential Geom. 86 (2010), no. 2, 303 -- 354. [34] [35] , Gromov-Witten invariants for G/B and Pontryagin product for ΩK, Trans. Amer. Math. Soc. 364 (2012), no. 5, 2567 -- 2599. , Classical aspects of quantum cohomology of generalized flag varieties, Int. Math. Res. Not. IMRN 16 (2012), 3706 -- 3722. [36] , Quantum Pieri rules for tautological subbundles, Adv. Math. 248 (2013), 279 -- 307. [37] C. Li, Functorial relationships between QH ∗(G/B) and QH ∗(G/P ), (II), Asian J. Math. 19, no. 2, 203-234. [38] C. Li and V. Ravikumar, Equivariant Pieri rules for isotropic Grassmannians, preprint at arxiv: math.AG/1406.4680. [39] L. C. Mihalcea, Positivity in equivariant quantum Schubert calculus, Amer. J. Math. 128 (2006), no. 3, 787 -- 803. [40] [41] [42] , Equivariant quantum Schubert calculus, Adv. Math. 203 (2006), no. 1, 1 -- 33. , On equivariant quantum cohomology of homogeneous spaces: Chevalley formulae and algorithms, Duke Math. J. 140 (2007), no. 2, 321 -- 350. , Giambelli formulae for the equivariant quantum cohomology of the Grassmannian, Trans. Amer. Math. Soc. 360 (2008), no. 5, 2285 -- 2301. [43] D. Peterson, Quantum cohomology of G/P , Lecture notes at MIT, 1997. Notes by J. Lu and K. Rietsch. [44] S. Robinson, A Pieri-type formula for H ∗ T (SLn(C)/B), J. Algebra 249 (2002), no. 1, 38 -- 58. [45] B. Siebert and G. Tian, On quantum cohomology rings of Fano manifolds and a formula of Vafa and Intriligator, Asian J. Math. 1 (1997), no. 4, 679 -- 695. [46] F. Sottile, Pieri's formula for flag manifolds and Schubert polynomials, Ann. Inst. Fourier (Grenoble) 46 (1996), no. 1, 89 -- 110. [47] H. Thomas and A. Yong, Equivariant Schubert calculus and jeu de taquin, to appear in Annales de l'Institut Fourier. Preprint at arXiv: math.CO/1207.3209. [48] C. T. Woodward, On D. Peterson's comparison formula for Gromov-Witten invariants of G/P , Proc. Amer. Math. Soc. 133 (2005), no. 6, 1601 -- 1609 (electronic). Department of Mathematics, Jinan University, Guangzhou, Guangdong, China E-mail address: [email protected] Center for Geometry and Physics, Institute for Basic Science (IBS), Pohang 790-784, Republic of Korea E-mail address: [email protected]
1103.3936
3
1103
2012-05-17T08:12:38
Relative singularity category of a non-commutative resolution of singularities
[ "math.AG", "math.RT" ]
In this article we study the triangulated category of singularities associated with a non-commutative resolution of singularities. In particular, we give a complete description of this category in the case of a curve with nodal singularities, classifying its indecomposable objects and computing its Auslander-Reiten quiver and K-group.
math.AG
math
RELATIVE SINGULARITY CATEGORY OF A NON-COMMUTATIVE RESOLUTION OF SINGULARITIES IGOR BURBAN AND MARTIN KALCK Abstract. In this article, we study a triangulated category associated with a non- commutative resolution of singularities. In particular, we give a complete description of this category in the case of a curve with nodal singularities, classifying its indecomposable objects and computing its Auslander -- Reiten quiver and K -- group. 1. Introduction This article grew up from an attempt to generalize the following statement, which is a consequence of a theorem of Buchweitz [5, Theorem 4.4.1] and results on idempotent completions of triangulated categories [17, 22]. Let X be an algebraic variety with isolated Gorenstein singularities, Z = Sing(X) =(cid:8)x1, . . . , xp(cid:9) and bOi := bOX,xi for all 1 ≤ i ≤ p. Then we have an equivalence of triangulated categories Perf(X) !ω Db(cid:0)Coh(X)(cid:1) ∼−→ p_i=1 MCM(cid:16)bOi(cid:17) . (1) (2) The left-hand side of (1) stands for the idempotent completion of the Verdier quotient Db(cid:0)Coh(X)(cid:1)/Perf(X) (known to be triangulated by [2]), whereas on the right-hand side MCM(bOi) denotes the stable category of maximal Cohen-Macaulay modules over bOi. We want to generalize this construction as follows. Let F ′ ∈ Coh(X), F := O ⊕ F ′ and A := EndX(F). Consider the ringed space X := (X, A). It is well-known that the functor F L ⊗X − : Perf(X) −→ Db(cid:0)Coh(X)(cid:1) viewed as a non-commutative (or categorical) resolution of singularities of X, in the spirit of works of Van den Bergh [27], Kuznetsov [18] and Lunts [19]. To measure the difference is fully faithful, see for instance [7, Theorem 2]. If gl.dim(cid:0)Coh(X)(cid:1) < ∞ then X can be between Perf(X) and Db(cid:0)Coh(X)(cid:1), we suggest to study the triangulated category Perf(X) !ω ∆X(X) := Db(cid:0)Coh(X)(cid:1) , which we shall call relative singularity category. Assuming F to be locally free on U := X \ Z, we prove an analogue of the "localization equivalence" (1) for the category ∆X(X). Using the negative K-theory of derived categories of Schlichting [23], we also describe the Grothendieck group of ∆X(X). 2000 Mathematics Subject Classification. Primary 14F05, Secondary 14A22, 18E30. 1 2 IGOR BURBAN AND MARTIN KALCK The main result of our article is a complete description of ∆Y (Y) in case Y is an arbitrary curve with nodal singularities and F ′ := IZ is the ideal sheaf of the singular locus of Y . We show that ∆Y (Y) splits into a union of p blocks: ∆Y (Y) i=1 ∆i, where p is the number of singular points of Y . Moreover, each block ∆i turns out to be equivalent to the category ∆nd defined as follows: ∼−→ Wp where And is the completed path algebra of the following quiver with relations: ∆nd := Hotb(cid:0)pro(And)(cid:1) Hotb(cid:0)add(P∗)(cid:1) , − α β * ∗ γ δ 3 + δα = 0, βγ = 0 and P∗ is the indecomposable projective And -- module corresponding to the vertex ∗. We prove that the category ∆nd is idempotent complete and Hom -- finite. Moreover, we give a complete classification of indecomposable objects of ∆nd. Finally, we show that ∆nd has the following interesting description: ∆nd ∼−→ Db(cid:0)Λ − mod(cid:1) Band(Λ) !ω , where Λ is the path algebra of the following quiver with relations ◦ a c 4 ◦ b d 4 ◦ ba = 0, dc = 0 and Band(Λ) is the category of the band objects in Db(cid:0)Λ − mod(cid:1), i.e. those objects which are invariant under the Auslander -- Reiten translation in Db(cid:0)Λ − mod(cid:1). Using this result, we describe the Auslander -- Reiten quiver of ∆nd. Acknowledgement. We would like to thank Nicolas Haupt, Jens Hornbostel, Bernhard Keller, Henning Krause, Helmut Lenzing, Michel Van den Bergh and Dong Yang for helpful discussions of parts of this article. We are also grateful to the anonymous referee for his/her comments and suggestions. This work was supported by the DFG grant Bu -- 1866/2 -- 1. 2. Generalities on the non-commutative category of singularities Let k be an algebraically closed field, X be a separated excellent Noetherian scheme over k such that any coherent sheaf on X is a quotient of a locally free sheaf, and Z be the singular locus of X. Let F ′ be a coherent sheaf on X, F = O ⊕ F ′ and A := EndX (F). Consider the non-commutative ringed space X = (X, A). Note that F is a locally projective coherent left A -- module. The following result is well-known, see e.g. [7, Theorem 2]. Proposition 2.1. The functor F := F L ⊗X − : Perf(X) → Db(cid:0)Coh(X)(cid:1) is fully faithful. * k k 3 t t * * 4 * * 4 SINGULARITY CATEGORY OF A NON-COMMUTATIVE RESOLUTION 3 Let P(X) be the essential image of Perf(X) under F. This category can be characterized in the following intrinsic way, see for instance [7, Proposition 2]1 Ob(cid:0)P(X)(cid:1) =nH• ∈ Ob(cid:16)Db(cid:0)Coh(X)(cid:1)(cid:17)(cid:12)(cid:12)(cid:12) H• x ∈ Im(cid:16)Hotb(cid:0)add(Fx)(cid:1) −→ Db(cid:0)Ax − mod(cid:1)(cid:17)o. idempotent completion of the Verdier quotient Db(cid:0)Coh(X)(cid:1)/P(X). Recall that according Definition 2.2. In the above notations, the relative singularity category ∆X(X) is the to Balmer and Schlichting [2], ∆X(X) has a natural structure of a triangulated category. Remark 2.3. In case X is an affine scheme, triangulated categories of the form ∆X (X) were also considered by Chen [10], Thanhoffer de Volcsey and Van den Bergh [24]. The following result seems to be well-known to experts. However, we were not able to find a reference in the literature and therefore give a proof here. Lemma 2.4. Let A be a ring and O be its center. Assume that O is Noetherian of Krull dimension d and A is finitely generated as a left O -- module. For any 0 ≤ e ≤ d let A−mode be the full subcategory of left Noetherian A -- modules A − mod, whose support over O is at most e -- dimensional and Db e(A − mod) be the full subcategory of Db(A − mod) consisting of complexes whose cohomology belongs to A − mode. Then the canonical functor is an equivalence of triangulated categories.2 Db (A − mode) −→ Db e(A − mod) Proof. By [15, Proposition 1.7.11], it is sufficient to show the following Statement. Let M be an arbitrary object of A − mode, N an arbitrary object of A − mod and φ : M → N an arbitrary injective A -- linear map. Then there exists an object K of A − mode and a morphism ψ : N → K such that ψφ is injective. Indeed, let E = E(M ) be an injective envelope of M and θ : M → E the corresponding embedding. Then there exists a morphism α : N → E such that αφ = θ. Note that K := Im(α) is a left Noetherian A -- module. As in [4, Lemma 3.2.5] one can show that for ∼= E(Mp). Hence, Kp = 0 for all p /∈ Supp(M ). This implies any p ∈ Spec(O) we have: Ep (cid:3) that kr. dim(cid:0)Supp(K)(cid:1) ≤ e. Remark 2.5. Lemma 2.4 is no longer true if A is assumed to be just left Noetherian. For example, let g be a finite dimensional simple Lie algebra over C and U = U (g) its universal enveloping algebra. Then U is left Noetherian, see for instance [20, Section I.7]. By Weyl's complete reducibility theorem, the category U − mod0 of finite dimensional left U -- modules is semi-simple. However, higher extensions between finite dimensional modules do not necessarily vanish in U − mod, see for instance [14]. In particular, the triangulated categories Db(U − mod0) and Db 0(U − mod) are not equivalent. Globalizing the proof of Lemma 2.4, we get the following result. 1Although the quoted results were stated in [7] in a weaker form, their proofs can be generalized literally to our case. 2We would like to thank B. Keller and M. Van den Bergh for an enlightening discussion on this subject. 4 IGOR BURBAN AND MARTIN KALCK Lemma 2.6. Let CohZ(X) be the category of coherent left A -- modules, whose support be- longs to Z and Db whose cohomology is supported at Z. Then the canonical functor Z(cid:0)Coh(X)(cid:1) be the full subcategory of Db(cid:0)Coh(X)(cid:1) consisting of complexes is an equivalence of triangulated categories. Db(cid:0)CohZ(X)(cid:1) −→ Db Z(cid:0)Coh(X)(cid:1) Our next goal is to prove that the category ∆X(X) depends only on an open neighborhood of the singular locus Z. Proposition 2.7. Let Db complexes whose cohomology is supported in Z and PZ (X) = P(X) ∩ Db the canonical functor Z(cid:0)Coh(X)(cid:1) be the full subcategory of Db(cid:0)Coh(X)(cid:1) consisting of Z(cid:0)Coh(X)(cid:1). Then H : is fully faithful. Db Z(cid:0)Coh(X)(cid:1) PZ (X) −→ Db(cid:0)Coh(X)(cid:1) P(X) P • C• Proof. Our approach is inspired by a recent paper of Orlov [22]. By [15, Proposition 1.6.10], it is sufficient to show that for any P • ∈ Ob(cid:0)P(X)(cid:1), C• ∈ Ob(cid:0)Db ϕ : P • → C• there exists Q• ∈ Ob(cid:0)PZ (X)(cid:1) and a factorization ϕ Z(cid:0)Coh(X)(cid:1)(cid:1) and ϕ′ !❈❈❈❈❈❈❈❈ =⑤⑤⑤⑤⑤⑤⑤⑤ Z(cid:0)Coh(X)(cid:1) → Db(cid:0)CohZ (X)(cid:1) is an equivalence t ≥ 1 such that I t annihilates every term of C•. Consider the ringed space Y =(cid:0)Z, A/I t(cid:1). By Lemma 2.6, we know that the functor Db of categories. Hence, we may without loss of generality assume that C• is a bounded complex of objects of CohZ (X). Let I = IZ be the ideal sheaf of Z. Then there exists Then we have a morphism of ringed spaces η : Y → X and an adjoint pair Q• ϕ′′ ( η∗ = forgetf : η∗ = A/I t L D−(cid:0)Coh(Y)(cid:1) → D−(cid:0)Coh(X)(cid:1) ⊗A − : D−(cid:0)Coh(X)(cid:1) → D−(cid:0)Coh(Y)(cid:1). Next, there exists E • ∈ Ob(cid:0)Db(cid:0)Coh(Y)(cid:1)(cid:1) such that C• = η∗(E •). Moreover, we have an iso- morphism γ : HomY(cid:0)η∗P •, E •(cid:1) −→ HomX(cid:0)P •, η∗(E •)(cid:1) such that for ψ ∈ HomY(cid:0)η∗P •, E •(cid:1) the corresponding morphism ϕ = γ(ψ) fits into the commutative diagram ξP • η∗η∗P • P • "❉❉❉❉❉❉❉❉❉ ϕ z✉✉✉✉✉✉✉✉✉ η∗(ψ) η∗E • where ξ : 1D−(X) → η∗η∗ is the unit of adjunction. Thus, it is sufficient to find a factor- ization of the morphism ξP • through an object of PZ(X). By definition of P(X), there exists a bounded complex of locally free OX -- modules R• such that the complexes P • and F ⊗X R• are isomorphic in Db(cid:0)Coh(X)(cid:1). Note that we / / ! = / / " z SINGULARITY CATEGORY OF A NON-COMMUTATIVE RESOLUTION 5 have the following commutative diagram in the category Comb(X) of bounded complexes of coherent left A -- modules: F ⊗X R• (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ ζR• 1⊗θR• F ⊗X(cid:0)O/I t ⊗X R•(cid:1) t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ∼= A/I t ⊗A(cid:0)F ⊗X R•(cid:1) where ζR• = ξP • in D−(cid:0)Coh(X)(cid:1) and θR• : R• → O/I t ⊗X R• is the canonical map. Since any coherent sheaf on X is a quotient of a locally free sheaf, there exists a bounded complex K• of locally free OX -- modules (Koszul complex of I t) K• =(cid:0)0 −→ Km −→ . . . −→ K1 −→ K0 −→ 0(cid:1) such that • K0 = O and H0(K•) ∼= O/I t, • H−i(K•) are supported at Z for all 1 ≤ i ≤ m. Hence, we have a factorization of the canonical morphism O → O/I t in the category of complexes Comb(cid:0)Coh(X)(cid:1): O[0] → K• → O/I t[0], which induces a factorization R• −→ K• ⊗X R• −→ O/I t ⊗X R• of the canonical map θR•. Note that the complex K• ⊗X R• is perfect and its cohomology is supported at Z. Hence, we get a factorization of the (derived) adjunction unit ξP • P • ∼= F L ⊗X R• −→ Q• := F L ⊗X (cid:0)K• L ⊗X R•(cid:1) −→ A/I t L ⊗A(cid:0)F L ⊗X R•(cid:1) ∼= A/I t L ⊗A P • (cid:3) we are looking for. This concludes the proof. Theorem 2.8. In the notations of Proposition 2.7, the induced functor Hω : Db PZ (X) !ω Z(cid:0)Coh(X)(cid:1) −→ Db(cid:0)Coh(X)(cid:1) P(X) !ω is an equivalence of triangulated categories. Proof. Proposition 2.7 implies that the functor Hω is fully faithful. Hence, we have to show it is essentially surjective. It suffices to prove the following Statement. For any M• ∈ Ob(cid:0)Db(cid:0)Coh(X)(cid:1)/P(X)(cid:1) there exist fM• ∈ Ob(cid:0)Db(cid:0)Coh(X)(cid:1)/P(X)(cid:1) and N • ∈ Ob(cid:0)Db Z(cid:0)Coh(X)(cid:1)/PZ (X)(cid:1) such that M• ⊕ fM• ∼= H(N •). / / ( t 6 IGOR BURBAN AND MARTIN KALCK Note that we have the following diagram of categories and functors Db Db(cid:0)Coh(X)(cid:1) Z(cid:0)Coh(X)(cid:1) Db(cid:0)Coh(X)(cid:1) P ⊗X −  F A / Db(cid:18) Coh(X) CohZ (X)(cid:19) ı∗ Perf(X) / Db(cid:0)Coh(U)(cid:1), where both compositions Perf(X) → Db(cid:0)Coh(U)(cid:1) are isomorphic. / Perf(U ) ⊗U − ∗  • The functor P is the canonical projection on the Verdier quotient. • The functor A is the canonical equivalence of triangulated categories constructed F(cid:12)(cid:12)U by Miyachi [21, Theorem 3.2].  ⊗U − : F(cid:12)(cid:12)U is an equivalence of triangulated categories induced by a Morita-type equivalence Perf(U ) = Db(cid:0)Coh(U )(cid:1) −→ Db(cid:0)Coh(U)(cid:1) canonical restrictions on an open subset. The functor ı∗ is an equivalence of triangulated categories induced by a canonical equivalence of abelian categories Coh(X)/ CohZ (X) → Coh(U). • For U = X \ Z let U be the ringed space (cid:0)U, A(cid:12)(cid:12)U ). The functor ∗ is the • Since the coherent sheaf F(cid:12)(cid:12)U is locally free, the functor F(cid:12)(cid:12)U ⊗U − : Coh(U ) → Coh(U). By a result of Thomason and Trobaugh [26, Lemma 5.5.1], for any S • ∈ Ob(cid:0)Perf(U )(cid:1) there exist eS • ∈ Ob(cid:0)Perf(U )(cid:1) and R• ∈ Ob(cid:0)Perf(X)(cid:1) such that ∗R• ∼= S • ⊕ eS •. Us- ing the fact that A, ı∗ and F(cid:12)(cid:12)U that for any M• ∈ Ob(cid:0)Db(cid:0)Coh(X)(cid:1)(cid:1) there exist fM• ∈ Ob(cid:0)Db(cid:0)Coh(X)(cid:1)(cid:1) and R• ∈ Ob(cid:0)Perf(X)(cid:1) such that P • := F ⊗X R• is isomorphic to M• ⊕fM• in the Verdier quotient Db(cid:0)Coh(X)(cid:1)/Db Z(cid:0)Coh(X)(cid:1). The last statement is equivalent to the fact that there exists T • ∈ Ob(cid:0)Db(cid:0)Coh(X)(cid:1)(cid:1) and a pair of distinguished triangles in Db(cid:0)Coh(X)(cid:1) such that C• are isomorphic in the Verdier quotient Db(cid:0)Coh(X)(cid:1)/P(X), we get a distinguished triangle in Db(cid:0)Coh(X)(cid:1)/P(X). The functor H : Db Z(cid:0)Coh(X)(cid:1)/PZ (X) → Db(cid:0)Coh(X)(cid:1)/P(X) is fully faithful, see Proposition 2.7. Hence, M• ⊕ fM• belongs to the essential image of H. Thus, θ −→ M• ⊕ fM• −→ C• −→ M• ⊕ fM• −→ C• θ −→ T • θ belong to the category Db ξ and C• and C• the functor Hω is essentially surjective, what concludes the proof. Z(cid:0)Coh(X)(cid:1). Since C• ⊗U − are equivalences of categories, this implies θ−→ P • −→ C• ξ −→ T • C• θ and T • α−→ C• ξ [1] θ [1]   ξ ξ [1] C• ξ (cid:3) /   O O O O / / SINGULARITY CATEGORY OF A NON-COMMUTATIVE RESOLUTION 7 t t Lemma 2.9. The canonical functor ∨p dimensional left Ai -- modules. In this case, Lemma 2.6 yields the following statement. iFi. Note that bAi ∼= EndbOi objects admitting a bounded resolution by objects of add(Fi). Then this functor restricts to an equivalence ∨p iOi to be the mi -- adic completion of Oi, bAi := lim←− Ai/m For any 1 ≤ i ≤ p we denote Oi := Oxi, mi the maximal ideal in Oi, Ai = Axi and Fi = Fxi. t iAi and From now on, we assume X has only isolated singularities and Z = Sing(X) =(cid:8)x1, . . . , xp(cid:9). Next, we set bOi = lim←− Oi/m (bFi). Let Ai − fdmod denote the category of finite bFi := lim←− Fi/m i=1Db(cid:0)Ai−fdmod(cid:1) → Db Z(cid:0)Coh(X)(cid:1) is an equivalence of triangulated categories. Let Pi be the full subcategory of Db(cid:0)Ai − fdmod(cid:1) consisting of Our next aim is to show that the Verdier quotient Db(cid:0)Ai − fdmod(cid:1)/Pi does not change Lemma 2.10. Let Perf fd(Oi) (respectively Perf fd(bOi)) be the full subcategory of Db(cid:0)Oi − fdmod) (respectively Db(cid:0)bOi − fdmod)) consisting of those complexes which are quasi- isomorphic to a bounded complex of finite rank free Oi -- (respectively bOi) -- modules. Let Perf fd(Oi) → Perf fd(bOi) and Db(cid:0)Ai − fdmod(cid:1) → Db(cid:0)bAi − fdmod(cid:1) be the exact functors in- duced by taking the completion. Then they are both equivalences of categories. Moreover, we have a diagram of categories and functors under passing to the completion. i=1Pi → PZ (X). L − L Fi bFi ⊗ bOi ⊗Oi − Perf fd(Oi) Db(cid:0)Ai − fdmod(cid:1) is fully faithful. In order to show it is essentially surjective, it is sufficient to prove that a non-perfect complex can not become perfect after applying the completion functor. Perf fd(bOi) / Db(cid:0)bAi − fdmod(cid:1), where both compositions Perf fd(Oi) → Db(cid:0)bAi − fdmod(cid:1) are isomorphic. Proof. Since the functors Oi − fdmod → bOi − fdmod and Ai − fdmod → bAi − fdmod are equivalences of categories, they induce equivalences Db(cid:0)Oi − fdmod(cid:1) → Db(cid:0)bOi − fdmod(cid:1) and Db(cid:0)Ai −fdmod(cid:1) → Db(cid:0)bAi −fdmod(cid:1). In particular, the functor Perf fd(Oi) → Perf fd(bOi) Indeed, X • ∈ Ob(cid:0)Db(cid:0)Oi − mod(cid:1)(cid:1) is perfect if and only if there exists n0 ∈ N such that for all n ≥ n0 we have: Hom(cid:0)X •, Oi/mi[n](cid:1) = 0. But this property is obviously preserved Corollary 2.11. Let ePi (respectively bPi) be the essential image of the triangle functor − : Perf fd(bOi) → Db(cid:0)bAi − bFi ⊗bOi ⊗bOi fdmod(cid:1)). Then we have an equivalence of triangulated categories Db(cid:0)bAi − fdmod(cid:1) − : Perf(bOi) → Db(cid:0)bAi − mod(cid:1) (respectively bFi bPi Db(cid:0)Ai − fdmod(cid:1) under the passing to the completion. −→ Pi (cid:3) L L . / /     / 8 IGOR BURBAN AND MARTIN KALCK Going along the same lines as in Theorem 2.8, one can show that the canonical functor is an equivalence. Summing up, we have equivalences of triangulated categories −→ Db(cid:0)bAi − mod(cid:1) !ω !ω Db(cid:0)bAi − fdmod(cid:1) !ω bPi p_i=1 Db(cid:0)Ai − fdmod(cid:1) ∼←− Pi !ω ePi ∼−→ Db(cid:0)Coh(X)(cid:1) P(X) !ω =: ∆X (X). p_i=1 Db(cid:0)bAi − mod(cid:1) ePi Now, let Y be a nodal algebraic curve, Z = {x1, . . . , xp} the singular locus of Y , I = IZ the ideal sheaf of Z and F = O ⊕ I. Then A = EndY (F) is the Auslander sheaf of orders introduced in [7]. Let Y = (Y, A) be the corresponding non-commutative curve. According to [7, Theorem 2], we have: gl.dim(cid:0)Coh(Y)(cid:1) = 2. Thus, Y is a non-commutative resolution of Y and Corollary 2.11 specializes to the following statement. Corollary 2.12. The triangulated category ∆Y (Y) splits into a union of p blocks ∆nd, where ∆nd is the "local" contribution of a singular point of Y (see also Section 4 for an explicit description of ∆nd). The goal of the subsequent part of this article is to answer the following questions. • Is the category ∆Y (Y) Hom -- finite? What are its indecomposable objects? • What is the Grothendieck group of ∆Y (Y)? • Assume E is a plane nodal cubic curve. What is the relation of ∆E(E) with the "quiver description" of Db(cid:0)Coh(E)(cid:1) from [7, Section 7]? 3. On the K-theory of the relative singularity category ∆X(X) Let O be a complete Gorenstein local ring and F = O ⊕ F1 ⊕ · · · ⊕ Fr ∈ O − mod, where F1, . . . , Fr are indecomposable and pairwise non-isomorphic and such that O does not belong to add(F1 ⊕· · ·⊕Fr). Let A = EndO(F ), P(O) be the essential image of Perf(O) under the exact embedding F ⊗O − : Perf(O) → Db(A − mod) and L ∆O(A) :=(cid:18) Db(A − mod) P(O) (cid:19) . The following result is well-known to specialists. Lemma 3.1. The triangulated category Db(O − mod)/ Perf(O) is idempotent complete. Proof. By a result of Buchweitz [5], we have an equivalence of triangulated categories Db(O − mod) Perf(O) ∼−→ MCM(O), where MCM(O) is the stable category of maximal Cohen -- Macaulay modules over O. Hence, it suffices to show that MCM(O) is idempotent complete. Since the ring O is complete, the endomorphism algebra of an indecomposable Noether- ian O -- module is local, see [11, Proposition 6.10]. Let M be any object of MCM(O). Then it admits a decomposition M ∼= M1 ⊕· · ·⊕Mp, such that the ring EndO(Mi) is local for any 1 ≤ i ≤ p (in other words, MCM(O) is a local category). Hence, MCM(O) is idempotent complete, see for example [8, Corollary 13.9]. (cid:3) SINGULARITY CATEGORY OF A NON-COMMUTATIVE RESOLUTION 9 The main result of this section is the following. Theorem 3.2. The category ∆O(A) is idempotent complete. Moreover, if gl. dim(A) < ∞ then K0(cid:0)∆O(A)(cid:1) ∼= Zr. Proof. First note that we have the following long exact sequences of abelian groups K0(cid:0)Perf(O)(cid:1) can−→ K0(cid:0)Db(O − mod)(cid:1) −→ K0(cid:0)MCM(O)(cid:1) −→ 0 Since by Lemma 3.1 the stable category MCM(O) is idempotent complete, we also obtain K -- groups of stable Frobenius categories of Schlichting [23, Section 4], associated with the and K0(cid:0)Perf(O)(cid:1) can−→ K0(cid:0)Db(O − mod)(cid:1) → K0(cid:0)(MCM(O))ω(cid:1) → K−1(cid:0)Perf(O)(cid:1) → K−1(cid:0)Db(O − mod)(cid:1), where K−1(cid:0)Perf(O)(cid:1) and K−1(cid:0)Db(O − mod)(cid:1) denote the negative Frobenius pairs(cid:16)Comb(cid:0)add(O)(cid:1), Comb ac(cid:0)add(O)(cid:1)(cid:17) and(cid:16)Com−, b(cid:0)add(O)(cid:1), Com− ac(cid:0)add(O)(cid:1)(cid:17) respectively, see [23, Theorem 1]. By [23, Theorem 7] we have K−1(cid:0)Db(O − mod)(cid:1) = 0. the vanishing K−1(cid:0)Perf(O)(cid:1) = 0. In a similar way, we have long exact sequences and K0(cid:0)P(O)(cid:1) can−→ K0(cid:0)Db(A − mod)(cid:1) → K0(cid:0)∆O(A)ω(cid:1) → K−1(cid:0)P(O)(cid:1) → 0. Since the F ⊗O − :(cid:16)Comb(cid:0)add(O)(cid:1), Comb ac(cid:0)add(F )(cid:1)(cid:17) 7] implies that K−1(cid:0)P(O)(cid:1) = K−1(cid:0)Perf(O)(cid:1) = 0. Hence, the canonical homomorphism of abelian groups K0(cid:0)∆O(A)(cid:1) → K0(cid:0)∆O(A)ω(cid:1) is an isomorphism. By a result of Thomason K0(cid:0)P(O)(cid:1) can−→ K0(cid:0)Db(A − mod)(cid:1) −→ K0(cid:0)∆O(A)(cid:1) −→ 0 ac(cid:0)add(O)(cid:1)(cid:17) −→(cid:16)Comb(cid:0)add(F )(cid:1), Comb [25, Theorem 2.1], the canonical functor ∆O(A) → ∆O(A)ω is an equivalence of triangu- lated categories, i.e. the triangulated category ∆O(A) is idempotent complete. induces an equivalence of the corresponding stable Frobenius categories, [23, Proposition morphism of Frobenius pairs Since O is a complete ring, A is semi-perfect with r + 1 pairwise non-isomorphic in- decomposable projective modules. If gl. dim(A) < ∞ then [11, Proposition 16.7] implies that K0(cid:0)Db(A − mod)(cid:1) ∼= K0(cid:0)A − mod(cid:1) ∼= Zr+1. Moreover, the image of the canonical homomorphism can : K0(cid:0)P(O)(cid:1) → K0(cid:0)Db(A − mod)(cid:1) is the free abelian group generated by the class of the projective module F . Hence, K0(cid:0)∆O(A)(cid:1) ∼= coker(can) ∼= Zr. (cid:3) 4. Description of the category ∆nd In this section And denotes the arrow ideal completion of the path algebra of the following quiver with relations ~Qnd (3) − α β * ∗ γ δ 3 + δα = 0, βγ = 0. Remark 4.1. Note that And = EndOnd(cid:0)Ond ⊕ kJuK ⊕ kJvK(cid:1) is the Auslander algebra of the nodal curve singularity Ond = kJu, vK/uv. In particular gl. dim(And) = 2, see [1] or [7, Remark 1]. * k k 3 t t 10 IGOR BURBAN AND MARTIN KALCK Our goal is to study the triangulated Verdier quotient category ∆nd := Db(And − mod) Hotb(cid:0)add(P∗)(cid:1) ∼= Hotb(cid:0)pro(And)(cid:1) Hotb(cid:0)add(P∗)(cid:1) ∼= ∆Ond(And), where P∗ is the indecomposable projective And -- module corresponding to the vertex ∗. By Theorem 3.2 we know that ∆nd is idempotent complete and K0(∆nd) ∼=(cid:10)[P−], [P+](cid:11) ∼= Z2. Definition 4.2. Let σ, τ ∈ {−, +} and l ∈ N. A minimal string Sτ (l) is a complex of indecomposable projective And -- modules · · · / 0 / Pσ / P∗ / · · · / P∗ / Pτ / 0 / · · · of length l + 2 with differentials given by non-trivial paths of minimal possible length and Pτ located in degree 0. Note, that σ is uniquely determined by τ and l: (σ = τ σ 6= τ if l is even, if l is odd. Example 4.3. The two complexes depicted below are minimal strings: • S+(1) = · · · • S+(2) = · · · / 0 / 0 / 0 / P− / P+ ·δ / P∗ ·β ·αβ / P∗ / P∗ ·γ ·γ / P+ / P+ / 0 / 0 / · · · / · · · It is interesting to note that the images of minimal strings remain to be indecomposable in ∆nd. In order to prove this, we need the following result of Verdier [28, Proposition II.2.3.3], playing a key role in the sequel. Lemma 4.4. Let T be a triangulated category and let U ⊆ T be a full triangulated sub- category. Let Y be an object in ⊥U =(cid:8)T ∈ Ob(T )(cid:12)(cid:12) HomT (T, U ) = 0(cid:9) and let P : HomT (Y, X) −→ HomT /U (Y, X) be the map induced by the localization functor. Then P is bijective for all X in T . Of course, there is a dual result for Y in U ⊥. Lemma 4.5. Let τ ∈ {+, −} and l ∈ N. Then any minimal string S = Sτ (l) belongs to Proof. First note that EndHotb(pro(And))(S) ∼= k. Hotb(cid:0)add(P∗)(cid:1)(cid:1) ∩ Ob(cid:0)Hotb(cid:0)add(P∗)(cid:1)⊥(cid:1). Moreover, S is indecomposable in ∆nd. Ob(cid:0)⊥ Hotb(cid:0)pro(And)(cid:1). Next, it is easy to check that for all m ∈ Z HomHotb(pro(And))(cid:0)P∗[m], S(cid:1) = 0 = HomHotb(pro(And))(cid:0)S, P∗[m](cid:1) holds. Now, Verdier's Lemma 4.4 implies indecomposability of S in ∆nd. In particular, S is indecomposable in (cid:3) Definition 4.6. Let T be an idempotent complete triangulated category and X1, · · · , Xn ∈ Ob(T ) an arbitrary collection of objects. Then Tria(X1, · · · , Xn) ⊆ T is the smallest full triangulated subcategory of T containing all Xi and closed under taking direct summands. / / / / / / / / / / / / / / / / / / / / / / SINGULARITY CATEGORY OF A NON-COMMUTATIVE RESOLUTION 11 Remark 4.7. The projective resolutions of the simple And -- modules S+ and S− are ·α−→ P− → S− → 0. Thus S± ∼= S±(1) are minimal strings. Let ρ, σ, τ ∈ {−, +} and l ∈ N. The cone of ·γ −→ P+ → S+ → 0 and 0 → P+ ·β −→ P∗ ·δ−→ P∗ 0 → P− Sτ (l) 0 / Pσ ·d3 P∗ ·d4 / · · · ·dl+2 / P∗ ·dl+3 / Pτ / 0 Sσ(1)[l + 1] 0 / Pρ ·d1 / P∗ ·d2 / Pσ / 0 is isomorphic to the following minimal string id Sτ (l + 1)[1] 0 / Pρ ·d1 / P∗ ·d2d3 / · · · ·dl+2 / P∗ ·dl+3 / Pτ / 0. Hence, the minimal strings are generated by S+ and S−. contained in Tria(S+, S−) ⊆ Db(And − mod), for all τ ∈ {−, +}, l ∈ N and n ∈ Z. In other words, Sτ (l)[n] is Theorem 4.8. We use the notations from above. (a) Let X be an indecomposable complex in Hotb(cid:0)pro(And)(cid:1). Then the image of X in ∆nd is either zero or isomorphic to one of the following objects where m, n ∈ Z, l ∈ N and σ, τ ∈ {+, −}. Pσ[n] ⊕ Pτ [m], Pτ [n] or Sτ (l)[n], (b) Let σ, τ ∈ {+, −} and n ∈ Z. We have the following formula: Hom∆nd(cid:0)Pσ, Pτ [n](cid:1) ∼= if n ∈ 2Z≤0 and σ = τ, if n ∈ 2Z≤0 − 1 and σ 6= τ, otherwise. k k 0 • String objects. In particular, Pσ[n] is indecomposable in ∆nd for any σ ∈ {+, −} and n ∈ Z. morphic in ∆nd if and only if their discrete parameters coincide. Proof. Since And is a nodal algebra, by the work of Burban and Drozd [6] the indecom- (c) Two objects from the set (cid:8)Pσ[n], Sτ (l)[m](cid:12)(cid:12) σ, τ ∈ {+, −}, n, m ∈ Z, l ∈ N(cid:9) are iso- posable objects in Hotb(cid:0)pro(And)(cid:1) are explicitly known. They are • Band objects. These are contained in Hotb(cid:0)add(P∗)(cid:1) and thus are zero in ∆nd. The string objects in Hotb(cid:0)pro(And)(cid:1) can be described in the following way. Let Z ~A∞ ∞ be the oriented graph obtained by orienting the edges in a Z2 -- grid as indicated in Example 4.9 below. Let ~θ ⊆ Z ~A∞ ∞ be a finite oriented subgraph of type An for a certain n ∈ N. Let Σ and T be the terminal vertices of ~θ and σ, τ ∈ {−, ∗, +}. We insert the projective modules Pσ and Pτ at the vertices Σ and T respectively. Next, we plug in P∗ at all intermediate vertices of ~θ. Finally, we put maps (given by multiplication with non-trivial paths in ~Qnd) on the arrows between the corresponding indecomposable projective modules. This has to be done in such a way that the composition of two subsequent arrows is always zero. Additionally, at the vertices where ~θ changes orientation, the inserted paths have to be "alternating", i.e. if one adjacent path involves α or β then the second should involve γ or / / /   / / / / / / / / / / / / / / 12 IGOR BURBAN AND MARTIN KALCK δ. Taking a direct sum of modules and maps in every column of the constructed diagram, we get a complex of projective And -- modules S, which we shall simply call string. Example 4.9. ◦ ◦ ◦ ?⑧⑧⑧⑧⑧⑧⑧ ❄❄❄❄❄❄❄ ?⑧⑧⑧⑧⑧⑧⑧ ❄❄❄❄❄❄❄ ◦ ◦ ◦ ❄❄❄❄❄❄❄ ?⑧⑧⑧⑧⑧⑧⑧ ❄❄❄❄❄❄❄ ?⑧⑧⑧⑧⑧⑧⑧ ?⑧⑧⑧⑧⑧⑧⑧ ❄❄❄❄❄❄❄ ?⑧⑧⑧⑧⑧⑧ ❄❄❄❄❄❄ ·(αβ)l P∗ ·(γδ)k P∗ P− ·(γδ)m P∗ ·β(αβ)n P∗ ❄❄❄❄❄❄ ?⑧⑧⑧⑧⑧⑧ ❄❄❄❄❄❄❄ ?⑧⑧⑧⑧⑧⑧⑧ ◦ ?⑧⑧⑧⑧⑧⑧⑧ ❄❄❄❄❄❄❄ ?⑧⑧⑧⑧⑧⑧⑧ ❄❄❄❄❄❄❄ ◦ ◦ ◦ S = · · · / 0 d1 / P∗ P− ⊕ P ⊕2 ∗ d2 / P∗ / 0 / · · · ·(γδ)k(cid:1)tr ·(γδ)m 0(cid:1) . and d2 =(cid:0)·β(αβ)n 1. If Pτ = P∗ and ~θ = Σ → · · · hold, then there exists a distinguished triangle Note, that the strings with Pσ = Pτ = P∗ vanish in ∆nd. Therefore, in what follows we may and shall assume that σ or τ ∈ {−, +}. where d1 =(cid:0)0 ·(αβ)l Proof of (a). Let S ∈ Ob(cid:0)Hotb(cid:0)pro(And)(cid:1)(cid:1) be an indecomposable string as defined above. with cone(f ) ∈ Ob(cid:0)Hotb(cid:0)add(P∗)(cid:1)(cid:1), yielding an isomorphism S ∼= Pσ[n] in ∆nd. Similarly, with cone(f ) ∈ Ob(cid:0)Hotb(cid:0)add(P∗)(cid:1)(cid:1) and hence an isomorphism S ∼= Pσ[n] in ∆nd. if ~θ = Σ ← · · · holds, then we obtain a triangle Pσ[n] 2. We may assume that σ, τ ∈ {−, +}. If the graph ~θ defining S is not linearly oriented (i.e. contains a subgraph (⋆) ◦ ◦ ◦ ), then there exists a distinguished triangle of the following form or (⋆⋆) ◦ / ◦ ◦ −→ S −→ cone(f ) −→ Pσ[n + 1] S −→ Pσ[n] −→ cone(f ) −→ S[1] f f (⋆) P∗[s] / S / S ′ ⊕ S ′′ / P∗[s + 1] (⋆⋆) P∗[s − 1] / S ′ ⊕ S ′′ / S / P∗[s] and therefore S ∼= S ′ ⊕ S ′′ ∼= Pσ[n] ⊕ Pτ [m] is decomposable in ∆nd. 3. Hence, without loss of generality, we may assume σ, τ ∈ {−, +} and ~θ to be linearly oriented. If S has a "non-minimal" differential d = ·p (i.e. the path p in ~Qnd contains δγ     ?  ?  ?  ?  ?  ?  ?  ?  ? ? / / / / / / / / o o / / o o / / / / / / SINGULARITY CATEGORY OF A NON-COMMUTATIVE RESOLUTION 13 or βα as a subpath), then we consider the following morphism of complexes S ′ f S ′′ Pσ / P∗ / · · · / P∗ d P∗ / P∗ / · · · / P∗ / Pτ which can be completed to a distinguished triangle in Hotb(cid:0)pro(And)(cid:1) S ′ f −→ S ′′ −→ S −→ S ′[1]. By our assumption on d, the morphism f factors through P∗[s] for some s ∈ Z and therefore vanishes in ∆nd. Hence, we have a decomposition S ∼= S ′[1] ⊕ S ′′ ∼= Pσ[n] ⊕ Pτ [m] in ∆nd. 4. If σ, τ ∈ {−, +}, ~θ is linearly oriented and S has only minimal differentials, then S is a minimal string. This concludes the proof of part (a) of Theorem 4.8. Proof of (b). Every morphism Pσ → Pτ [n] in ∆nd is given by a roof Pσ f, g are morphisms in Hotb(cid:0)pro(And)(cid:1) and cone(f ) ∈ Ob(cid:0)Hotb(cid:0)add(P∗)(cid:1)(cid:1). By a common abuse of terminology, we call f a quasi-isomorphism. Our aim is to find a convenient representative in each equivalence class of roofs. It turns out that σ ∈ {+, −} and n ∈ Z determine Q and f of our representative and g is either 0 or determined by τ up to scalar. g −→ Pτ [n], where f ←− Q 1. Without loss of generality, we may assume that Q has no direct summands from Hotb(cid:0)add(P∗)(cid:1). Indeed, if Q ∼= Q′ ⊕ Q′′ with Q′′ ∈ Ob(cid:0)Hotb(cid:0)add(P∗)(cid:1)(cid:1), then the diagram f ′ w♥♥♥♥♥♥♥♥♥♥♥♥♥♥♥♥ f =( f ′ f ′′ ) Q′ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ 0 (cid:17) (cid:16) id Q′ ⊕ Q′′ g=( g′ g′′ ) g′ Pσ / Pτ [n] yields an equivalence of roofs Pσ f ←− Q g −→ Pτ [n] and Pσ ←− Q′ g′ f ′ −→ Pτ [n]. 2. Using our assumptions on f and Q in conjunction with the description of indecompos- able strings in Hotb(cid:0)pro(And)(cid:1), it is not difficult to see that Q can (without restriction) taken to be an indecomposable string with τ = ∗ and f to be of the following form: · · · / P∗ / P∗ / P∗ Q f Pσ Pσ id Pσ / P∗ / · · · / P∗ ❇❇❇❇❇❇❇ / P∗ 3. Without loss of generality, we may assume that Q is constructed from a linearly oriented graph ~θ. Indeed, otherwise we may consider the truncated complex Q≤ defined   / / /   / / / / ( w   o o / / / /     / / / / 14 IGOR BURBAN AND MARTIN KALCK in the diagram below and replace our roof by an equivalent one. · · · / P∗ / P∗ / P∗ Pσ id Pσ d1 d1 / P∗ id / P∗ d2 d2 dn−1 / · · · dn−1 · · · / P∗ id / P∗ Q q Q≤ ❇❇❇❇❇❇❇ dn / P∗ dn id P∗ f q xqqqqqqqqqqqqq f Pσ Q≤ q Q gq '❖❖❖❖❖❖❖❖❖❖❖❖❖ g / Pτ [n] In particular, n > 0 implies that Hom∆nd(cid:0)Pσ, Pτ [n](cid:1) = 0 holds. 4. By the above reductions, g has the following form: Q g Pτ [n] Pσ / P∗ / · · · / P∗ P∗ / · · · / P∗ g Pτ We may truncate again so that Q ends at degree −n: Q≤−n w♦♦♦♦♦♦♦♦♦♦♦♦♦ Pσ Q 'PPPPPPPPPPPPP Pτ [n] 5. Next, we may assume that Q has minimal differentials (see Definition 4.2) and thus is uniquely determined by σ and n. Indeed, otherwise there exists a quasi-isomorphism: Q′ q Q Pσ P∗ · · · / P∗ id id id Pσ / P∗ / · · · / P∗ d′ d / 0 P∗ d′′ / P∗ / P∗ / · · · 6. Summing up, our initial roof can be replaced by an equivalent one of the following form Pσ id Pσ Pσ f Q g Pτ [n] / P∗ / · · · / P∗ g Pτ / / / / / / / O O O O / / / O O / / / O O O O ' x   o o /   / / / / / /   / ' w o o / / O O   / /   / /   / / /     /   / / / / / / O O   O O / / /   SINGULARITY CATEGORY OF A NON-COMMUTATIVE RESOLUTION 15 If g is not minimal (i.e. not given by multiplication with a single arrow), then it factors over P∗[n] and therefore vanishes in ∆nd. Thus the morphism space Hom∆nd(Pσ, Pτ [n]) is at most one dimensional. Moreover, g can be non-zero only if n has the right parity. g f ←− Q g −→ Pτ [n] as in the previous step and assume that g is non-zero 7. Consider a roof Pσ and minimal. We want to show that the roof defines a non-zero homomorphism in ∆nd. We have a triangle Q Pσ → Pτ [n] → Sτ (−n)[n] → Pσ[1] in ∆nd. Since Sτ (−n)[n] is indecomposable, the map is non-zero. The claim follows. → Pτ [n] → Sτ (−n)[n] → Q[1] in Hotb(cid:0)And − mod(cid:1) yielding a triangle Proof of (c). Note that for X ∈ Ob(cid:0)Tria(S+, S−)(cid:1) we have [X] = n ·(cid:0)[P+] + [P−](cid:1) ∈ indecomposable strings in Hotb(cid:0)pro(And)(cid:1), it remains to show that Pσ[n] ∼= Pτ [m] implies K0(∆nd) for a certain n ∈ Z. Thus, the images of indecomposable projective And -- modules P+ and P− are not contained in Tria(S+, S−). By Lemma 4.4 and the classification of σ = τ and n = m. Assume that n > m holds. Then using Lemma 4.4 again, we obtain Hom∆nd(cid:0)Pσ[n], Sσ(1)[n](cid:1) ∼= k 6= 0 = Hom∆nd(cid:0)Pτ [m], Sσ(1)[n](cid:1). This is a contradiction. Similarly, the assumption σ 6= τ leads to a contradiction. (cid:3) Remark 4.10. Theorem 4.8 and Lemma 4.4 reduce the computation of morphism spaces in ∆nd to a computation in Hotb(cid:0)pro(And)(cid:1). Moreover, every minimal string may be presented as a cone of a morphism Pσ[n] → Pτ [m] in ∆nd (see step 7 in the proof of Theorem 4.8 (b)). Using this fact and the long exact Hom-sequence, one can show that dimk Hom∆nd(X, Y ) ≤ 1 holds for all indecomposable objects X and Y in ∆nd. Corollary 4.11. The indecomposable objects of the triangulated subcategory Tria(S−, S+) ⊂ Db(cid:0)And − mod(cid:1) are precisely the shifts of the minimal strings Sτ (l). Proof. By Remark 4.7, we know that all minimal strings belong to Tria(S−, S+). Hence, we just have to prove that there are no other indecomposable objects. According to Lemma 4.4 and Lemma 4.5, the functor Tria(S−, S+) → ∆nd is fully faithful. Therefore, the indecomposable objects of the category Tria(S−, S+) and its essential image in ∆nd are the same. By Theorem 4.8, all indecomposable objects of ∆nd are known and the shifts of the objects P+ and P− are not contained in Tria(S−, S+). Hence, the minimal strings are the only indecomposable objects of Tria(S−, S+). (cid:3) 5. Connection with the category(cid:0)Db(Λ − mod)/ Band(Λ)(cid:1)ω Let Λ be the path algebra of the following quiver with relations (4) 1 a c 4 2 b d 4 3 ba = 0, dc = 0 and Band(Λ) be the full subcategory of Db(Λ − mod) consisting of those objects, which are invariant under the Auslander -- Reiten translation in Db(Λ − mod). By [7, Corollary 6], the subcategory Band(Λ) is triangulated. Hence, we can define the triangulated category e∆nd :=(cid:0)Db(Λ − mod)/ Band(Λ)(cid:1)ω, * * 4 * * 4 16 IGOR BURBAN AND MARTIN KALCK i.e. the idempotent completion of the Verdier quotient Db(Λ − mod)/ Band(Λ) (see [2]). The main goal of this section is to show that e∆nd and ∆nd are triangle equivalent. Lemma 5.1. The indecomposable projective Λ -- modules are pairwise isomorphic in e∆nd. Proof. Complete the following exact sequences of Λ -- modules 0 −→ P2 −→ P1 −→ k 0 −→ P3 −→ P2 −→ 0 1 1 6 k 6 k 1 1 6 0! −→ 0 6 k! −→ 0 to triangles in Db(Λ−mod) and note that the modules on the right-hand side are bands. (cid:3) Let P ∈ Ob(cid:16)e∆nd(cid:17) be the common image of the indecomposable projective Λ -- modules. Lemma 5.2. The endomorphisms of P , which are given by the roofs (5) e+ = P1 ·(a+c) ←−−−− P2 ·a−→ P1 and e− = P1 ·(a+c) ←−−−− P2 ·c−→ P1 satisfy e−e+ = 0 = e+e− and e− + e+ = idP and thus are idempotent. In particular, we have a direct sum decomposition P ∼= P + ⊕ P −, where P + = (P, e+) and P − = (P, e−). Proof. It is clear that e− + e+ = idP . The equality e+e− = 0 follows from the diagram ·(b+d) ~⑥⑥⑥⑥⑥⑥⑥⑥ ❆❆❆❆❆❆❆❆ ·a e+e− = P2 ·(a+c) ~⑥⑥⑥⑥⑥⑥⑥⑥ P1 P3 ·d ❆❆❆❆❆❆❆❆ ~⑥⑥⑥⑥⑥⑥⑥⑥ ·(a+c) P1 = P1 ·(da+bc) ←−−−−− P3 0−→ P1. P2 ·c ❆❆❆❆❆❆❆❆ P1 The second equality e−e+ = 0 follows from a similar calculation. Hence, e2 ± = e±. (cid:3) Next, note the following easy but useful result. Lemma 5.3. Let A be an abelian category and let S : Db(A) → Db(A) be a triangle equivalence. If X1, X2 ∈ Ob(cid:0)Db(A)(cid:1) and n1, n2, m1, m2 ∈ Z satisfy Sm1X1 ∼= X1[n1], Sm2X2 ∼= X2[n2] and d = m1n2 − m2n1 6= 0, then HomDb(A)(X1, X2) = 0 = HomDb(A)(X2, X1). Proof. By the symmetry of the claim, it suffices to show that HomDb(A)(X1, X2) vanishes. Since S is an equivalence, we have a chain of isomorphisms HomDb(A)(X1, X2) ∼= HomDb(A)(cid:0)S±m1m2X1, S±m1m2X2(cid:1) ∼= HomDb(A)(cid:0)X1[±m2n1], X2[±m1n2](cid:1) ∼= HomDb(A)(cid:0)X1, X2[±d](cid:1) ∼= HomDb(A)(cid:0)X1, X2[±kd](cid:1) ( ( 6 ( ( 6 ( ( 6 ( ( 6 ~ ~ ~ SINGULARITY CATEGORY OF A NON-COMMUTATIVE RESOLUTION 17 for all k ∈ N. Hence, the claim follows from the boundedness of X1 and X2 together with A (A1, A2) ∼= HomDb(A)(A1, A2[−n]), the fact that there are no non-trivial Ext -- groups Ext−n where A1, A2 ∈ Ob(A) and n is a positive integer. A direct calculation in Db(Λ − mod) yields the following result. (cid:3) Lemma 5.4. Let S : Db(Λ − mod) → Db(Λ − mod) be the Serre functor, X+ = k 1 0 6 k 0 1 6 k and X− = k 0 1 6 k 1 0 6 k . Then S(X±) ∼= X∓[2]. In particular, X± are 4 2 -fractionally Calabi -- Yau objects. Corollary 5.5. The following composition of the inclusion and projection functors Tria(X+, X−) ֒→ Db(Λ − mod) −→ Db(Λ − mod) Band(Λ) is fully faithful. Proof. Lemma 5.4 and Lemma 5.3 applied to the Serre functor S in Db(Λ − mod) imply Band(Λ)(cid:1)∩Ob(cid:0)Band(Λ)⊥(cid:1). Hence, the claim follows from Lemma 4.4. (cid:3) Theorem 5.6. There exists an equivalence of triangulated categories that X± ∈ Ob(cid:0)⊥ Proof. Let E = V (zy2 − x3 − x2z) ⊂ P2 be a nodal cubic curve and F ′ = I be the ideal sheaf of the singular point of E. Let F = O ⊕ I, A = EndE(F) and E = (E, A). By a result of Burban and Drozd [7, Section 7], there exists a triangle equivalence G : . Db(And − mod) Band(Λ) !ω Hotb(cid:0)add(P∗)(cid:1) −→ Db(Λ − mod) T : Db(cid:0)Coh(E)(cid:1) −→ Db(Λ − mod) identifying the image of the category Perf(E) with the category Band(Λ). Moreover, by [7, Proposition 12], the functor T restricts to an equivalence Tria(S+, S−) → Tria(X+, X−). This can be summarized by the following commutative diagram of categories and functors ∼ Db(And − mod) Hotb(cid:0)add(P∗)(cid:1) (cid:18)Db(And − fdmod) fd(cid:0)add(P∗)(cid:1) (cid:19)ω Hotb can Db(And − fdmod) (6) G ∼ P(E) (cid:18) Db(cid:0)Coh(E)(cid:1) (cid:19)ω / Db(cid:0)Coh(E)(cid:1) can Tria(S+, S−) ∼ T ∼ Band(Λ) (cid:19)ω /(cid:18) Db(Λ − mod) can / Db(Λ − mod) / Tria(X+, X−) ( ( 6 ( ( 6 ( ( 6 ( ( 6 ) ) / / O O / O O / O O / O O ?  O O / ?  O O 18 IGOR BURBAN AND MARTIN KALCK where G : ∆nd → e∆nd is the induced equivalence of triangulated categories. Lemma 5.7. The indecomposable objects of the triangulated category e∆nd are • P ±[n] ∼= G(cid:0)P±[n](cid:1), n ∈ Z. • The indecomposables of the full subcategory Tria(X+, X−) ∼= G(cid:0)Tria(S+, S−)(cid:1). Proof. Consider the projective resolution of the simple And-module S∗ (cid:3) 0 / P− ⊕ P+ ( ·β ·δ ) / P∗ / S∗ / 0 . Completing it to a distinguished triangle yields an isomorphism S∗[−1] ∼= P+ ⊕ P− in ∆nd. In the notations of the diagrams (6) and (4), we have T(S∗) ∼= P3[1] and therefore G(P+ ⊕ P−) ∼= G(S∗[−1]) ∼= P ∼= (P + ⊕ P −). ·d−→ P2 ·c−→ P1(cid:1), where P1 is located in degree 0 and P ± := Recall that X+ ∼= (cid:0)P3 (P, e±) ∈ Ob(cid:0)e∆nd(cid:1), with e± as defined in (5). A direct calculation shows that the obvious morphism from P1 to X+ induces a non-zero morphism P + = (P, e+) → X+ in e∆nd, (P +, X−) = 0. Moreover, it was shown in [7] that T(S±) ∼= X±. This whereas Home∆nd implies G(S±) ∼= X± and thus G(P±) ∼= P ±. Theorem 4.8 and Corollary 4.11 yield the stated classification of indecomposables in e∆nd. (cid:3) 6. Concluding remarks on ∆nd Proposition 6.1. The category Tria(S+, S−) ⊂ ∆nd has Auslander -- Reiten triangles. Proof. As mentioned above, we have an exact equivalence of triangulated categories Tria(S+, S−) ∼= Tria(X+, X−) ⊂ Db(Λ − mod). The category Db(Λ − mod) has a Serre functor S and therefore has Auslander -- Reiten triangles, see [13]. Let τ = S ◦ [−1] be the Auslander -- Reiten translation. Using that τ is an equivalence and Lemma 5.4, we obtain τ(cid:0)Tria(X+, X−)(cid:1) ∼= Tria(cid:0)τ (X+), τ (X−)(cid:1) ∼= Tria(X+, X−). Now, the restriction of τ to Tria(X+, X−) is the Auslander -- Reiten transla- tion of this subcategory. (cid:3) Remark 6.2. One can show that the Auslander -- Reiten quiver of Tria(S+, S−) consists of two ZA∞ -- components. We draw one of them below, indicating the action of the Auslander -- Reiten translation by o o❴ ❴ ❴ . The other component is obtained from this one by changing the roles of + and −. S−(3)[1] o❴ ❴ ❴ S+(3) o❴ ❴ ❴ S−(3)[−1] o❴ ❴ S+(3)[−2] o❴ ❴ S−(3)[−3] S−(2)[1] o❴ ❴ ❴ S+(2) o❴ ❴ ❴ S−(2)[−1] o❴ ❴ S+(2)[−2] <③③③③ "❉❉❉❉ "❉❉❉❉ <③③③③ <③③③③ "❉❉❉❉ "❉❉❉❉ <③③③③ <③③③③ "❉❉❉❉ "❉❉❉❉ <③③③③ <③③③③ "❉❉❉❉ "❉❉❉❉ <③③③③ S+(1)[2] o❴ ❴ ❴ S−(1)[1] o❴ ❴ ❴ S+(1) o❴ ❴ ❴ S−(1)[−1] o❴ ❴ S+(1)[−2] / / / / < " " < o < o " o " < o < o " o " < o < o " o " < o o SINGULARITY CATEGORY OF A NON-COMMUTATIVE RESOLUTION 19 The category ∆nd does not have Auslander -- Reiten triangles, but we may still consider the quiver of irreducible morphisms in ∆nd, which has two additional A∞ ∞ -- components. / P±[2] / P∓[1] / P± / P∓[−1] / P±[−2] Proposition 6.3. The respective triangulated categories Tria(X+, X−) and ∆nd are not triangle equivalent to the bounded derived category of a finite dimensional algebra. Proof. Assume that there exists a triangle equivalence to the derived category of a finite dimensional algebra A. Then Db(A − mod) is of discrete representation type. Hence, A is a gentle algebra occuring in Vossieck's classification [29]. In particular, A is a Gorenstein algebra [12]. Therefore, the Nakayama functor defines a Serre functor on Hotb(cid:0)pro(A)(cid:1) [13], whose action on objects is described in [3, Theorem B]. On the other hand, S2(X) ∼= X[4] holds for all objects X in Tria(X+, X−), by Lemma 5.4 and Proposition 6.1. This yields a contradiction. (cid:3) The following proposition generalizes Theorem 5.6. Proposition 6.4. Let n ≥ 1 and Λn be the path algebra of the following quiver identify ◦ _❅❅❅❅❅❅❅ w− 1 ?⑦⑦⑦⑦⑦⑦⑦ w+ 1 ◦ ◦ _❅❅❅❅❅❅❅ w− 2 ◦ ?⑦⑦⑦⑦⑦⑦⑦ w+ 2 ◦ a❇❇❇❇❇❇❇❇ =⑤⑤⑤⑤⑤⑤⑤⑤ · · · ◦ _❅❅❅❅❅❅❅ w− n ◦ ?⑦⑦⑦⑦⑦⑦⑦ w+ n ◦ u1 v1 u2 v2 ◦ ◦ · · · un vn ◦ subject to the relations w− i ui = 0 and w+ i vi = 0 for all 1 ≤ i ≤ n. Then ∆n :=(cid:18) Db(Λn − mod) Band(Λn) (cid:19)ω ∼= ∆nd. n_i=1 In particular, the category ∆n is representation discrete, Hom-finite and K0(∆n) ∼= (Z2)⊕n. where IZ is the ideal sheaf of the singular locus Z. By [7, Proposition 10], there exists an Proof. Let E = En be a Kodaira cycle of n projective lines and E =(cid:0)E, EndE(O ⊕ IZ)(cid:1), equivalence of triangulated categories Db(cid:0)Coh(E)(cid:1) ∼−→ Db(Λn −mod) identifying Perf(E) ∼= P(E) ⊂ Db(cid:0)Coh(E)(cid:1) with Band(Λn) ⊂ Db(Λn − mod), see [7, Corollary 6]. Thus, Corollary 2.11 yields the proof. (cid:3) 6 ◦ and Band(A) Remark 6.5. Let A be the path algebra of the Kronecker quiver ◦ be the full subcategory of Db(A − mod) consisting of those objects, which are invariant under the Auslander -- Reiten translation. Note, that the objects of Band(A) are direct sums of indecomposable objects lying in tubes. Moreover, Band(A) is closed under taking cones and direct summands. In particular, it is a triangulated subcategory. It is interesting to note, that the Verdier quotient category Db(A − mod)/ Band(A) is not Hom-finite. / / / / / / / _ ? _ ? a = ? _ K K S S K K S S K K S S ( ( 6 20 IGOR BURBAN AND MARTIN KALCK Indeed, the well-known tilting equivalence Db(A − mod) → Db(cid:0)Coh(P1)(cid:1) identifies Band(A) with the category Db(cid:0)Tor(P1)(cid:1), where Tor(P1) is the category of torsion coherent sheaves on P1. Hence, by Miyachi's theorem [21] we have: Db(A − mod) Band(A) ∼= Db(cid:0)Coh(P1)(cid:1) Db(cid:0)Tor(P1)(cid:1) ∼= Db(cid:18) Coh(P1) Tor(P1)(cid:19) ∼= Db(cid:0)k(t) − mod(cid:1), where k(t) is the field of rational functions. Therefore, the category Db(A−mod)/ Band(A) is not Hom -- finite. We conclude this paper by giving a relation between our non-commutative singularity cate- gory ∆nd and the (classical) singularity category MCM(Ond) for the ring Ond = kJu, vK/uv. Proposition 6.6. There is an equivalence of triangulated categories ∆nd Tria(S+, S−) ∼−→ MCM(Ond). Proof. The functor HomAnd(P∗, −) : And − mod → Ond − mod is exact and induces an ∼−→ Ond − mod [7, Theorem equivalence of abelian categories And − mod/ add(S+ ⊕ S−) 4.8]. Using Miyachi's compatibility of Serre and Verdier quotients [21, Theorem 3.2], we see that P = HomAnd(P∗, −) : Db(cid:0)And − mod(cid:1) → Db(cid:0)Ond − mod(cid:1) is a quotient functor in ∼−→ Db(cid:0)Ond − mod(cid:1). A direct calculation shows the sense of [9], i.e. Db(cid:0)And − mod(cid:1)/ker(P) that HomAnd(P∗, P∗) ∼= Ond. Hence, P induces a functor I : ∆nd → MCM(Ond) and the following diagram commutes. Hotb(cid:0)add(P∗)(cid:1)  P Db(And − mod) P Perf(Ond)  / Db(Ond − mod) can can ∆nd I / MCM(Ond) By [9, Lemma 2.1], I is again a quotient functor. Using the classification of indecomposable objects in ∆nd and the fact that HomAnd(P∗, P+ ⊕ P−) ∼= kJuK ⊕ kJvK, we obtain ker(I) = Tria(S+, S−). This concludes the proof. (cid:3) Remark 6.7. After submitting this article, we learned that Thanhoffer de Volcsey and Van den Bergh proved a general Theorem, which contains Proposition 6.6 as a special case. Namely, in the notations of Section 3 assume that O is an isolated singularity and A has finite global dimension. Let e ∈ A be the idempotent corresponding to the identity endomorphism of O. We denote the simple A -- modules S0, . . . , Sr in such a way that S0 has projective cover Ae. Then the exact functor eA ⊗A − : A − mod → O − mod induces an equivalence of triangulated categories ∆O(A) (7) Tria(S1, . . . , Sr) ∼−→ MCM(O), see [24, Theorem 5.1.1 and Lemma 5.1.3]. Moreover, they show that ∆O(A) is Hom -- finite in this generality, see [24, Proposition 5.1.4]. Using quite different methods, slight generalizations of both results were subsequently obtained in recent work of Kalck and Yang [16].  / /   / /      / / SINGULARITY CATEGORY OF A NON-COMMUTATIVE RESOLUTION 21 7. Summary In this section, we collect the major results obtained in this article. Let Y be a nodal algebraic curve, Z its singular locus, I = IZ and Y = (Y, A) for A = EndY (O ⊕ I). Similarly, let O = kJu, vK/(uv), m = (u, v) and A = EndO(O ⊕ m). Then the following results are true. • The category ∆Y (Y) splits into a union of p blocks ∆nd, where p is the number of singular points of Y and ∆nd = ∆O(A), see Corollary 2.12. • The category ∆nd is Hom -- finite and representation discrete. In particular, its inde- composable objects and the morphism spaces between them are explicitly known, see Theorem 4.8. Moreover, one can compute its Auslander -- Reiten quiver, see Remark 6.2. • We have: K0(cid:0)∆nd(cid:1) ∼= Z2, see Theorem 3.2. Moreover, the category ∆nd admits an alternative "quiver description" in terms of repre- sentations of a certain gentle algebra Λ, see Section 5. References [1] M. Auslander, K. Roggenkamp, A characterization of orders of finite lattice type, Invent. Math. 17 (1972), 79 -- 84. [2] P. Balmer, M. Schlichting, Idempotent completion of triangulated categories, J. Algebra 236 (2001), no. 2, 819 -- 834. [3] G. Bobi´nski, C. Geiss, A. Skowro´nski, Classification of discrete derived categories, Central European J. Math. 2 (2004), 19 -- 49. [4] W. Bruns, J. Herzog, Cohen-Macaulay rings, Cambridge Studies in Advanced Mathematics 39, Cam- bridge University Press (1993). [5] R.-O. Buchweitz, Maximal Cohen-Macaulay modules and Tate-Cohomology over Gorenstein rings, Preprint 1987, available at http://hdl.handle.net/1807/16682. [6] I. Burban, Yu. Drozd, Derived categories of nodal algebras, J. Algebra 272 (2004), no. 1, 46 -- 94. [7] I. Burban, Yu. Drozd, Tilting on non-commutative rational projective curves, Mathematische Annalen, vol. 351, no. 3, 665 -- 709 (2011). [8] I. Burban, Yu. Drozd, Maximal Cohen-Macaulay modules over non-isolated surface singularities, arXiv:1002.3042. [9] X.-W. Chen, Unifying two results of Orlov on singularity categories, Abh. Math. Semin. Univ. Ham- burg 80 (2010), no. 2, 207 -- 212. [10] X.-W. Chen, Relative singularity categories and Gorenstein-projective modules, Math. Nachr. 284 (2011), no. 2 -- 3, 199 -- 212. [11] C. Curtis, I. Reiner, Methods of representation theory. Vol. I. With applications to finite groups and orders, Pure and Applied Mathematics. A Wiley-Interscience Publication (1981). [12] C. Geiss, I. Reiten, Gentle algebras are Gorenstein, Representations of algebras and related topics, Fields Inst. Comm. 45 (2005), 129 -- 133. [13] D. Happel, Triangulated categories in the representation theory of finite-dimensional algebras, LMS Lecture Note Series 119, Cambridge University Press (1988). [14] J. Humphreys, Representations of semisimple Lie algebras in the BGG category O, Graduate Studies in Mathematics 94, American Mathematical Society, Providence, RI (2008). [15] M. Kashiwara, P. Schapira, Sheaves on Manifolds 292, Springer (1990). [16] M. Kalck, D. Yang, DG-models for algebraic triangulated categories I: Knorrer's periodicity revisited, arXiv:1205.1008. [17] B. Keller, D. Murfet, M. Van den Bergh, On two examples by Iyama and Yoshino, Compos. Math. 147 (2011), no. 2, 591 -- 612. 22 IGOR BURBAN AND MARTIN KALCK [18] A. Kuznetsov, Lefschetz decompositions and categorical resolutions of singularities, Selecta Math. 13 (2008), no. 4, 661 -- 696. [19] V. Lunts, Categorical resolution of singularities, J. Algebra 323 (2010), no. 10, 2977 -- 3003. [20] J. McConnell, J. Robson, Noncommutative Noetherian rings. With the cooperation of L. W. Small, Graduate Studies in Mathematics 30, American Mathematical Society, Providence, RI (2001). [21] J. Miyachi, Localization of triangulated categories and derived categories, J. Algebra 141 (1991), 463 -- 483. [22] D. Orlov, Formal completions and idempotent completions of triangulated categories of singularities, Adv. Math. 226 (2011), no. 1, 206 -- 217. [23] M. Schlichting, Negative K-theory of derived categories, Math. Z. 253 (2006), no. 1, 97 -- 134. [24] L. Thanhoffer de Volcsey, M. Van den Bergh, Explicit models for some stable categories of maximal Cohen -- Macaulay modules, arXiv:1006.2021. [25] R. Thomason, The classification of triangulated subcategories, Compositio Math. 105 (1997), no. 1, 1 -- 27. [26] R. Thomason, T. Trobaugh, Higher algebraic K-theory of schemes and of derived categories, The Grothendieck Festschrift, Vol. III, 247 -- 435, Progr. Math. 88, Birkhauser (1990). [27] M. Van den Bergh, Three-dimensional flops and noncommutative rings, Duke Math. J. 122 (2004), no. 3, 423 -- 455. [28] J. L. Verdier, Des Cat´egories D´eriv´ees des Cat´egories Ab´eliennes, Ast´erisque (Soci´et´e Math´ematique de France, Marseilles) 239, 1996. [29] D. Vossieck, The algebras with discrete derived category, J. Algebra 243 (2001), no. 1, 168 -- 176. Universitat zu Koln, Mathematisches Institut, Weyertal 86-90, D-50931 Koln, Germany E-mail address: [email protected] Mathematisches Institut, Universitat Bonn, Endenicher Allee 60, D-53115 Bonn, Germany E-mail address: [email protected]
1305.6417
1
1305
2013-05-28T09:08:57
The Gromov-Winkelmann theorem for flexible varieties
[ "math.AG" ]
An affine variety $X$ of dimension $\ge 2$ is called {\em flexible} if its special automorphism group SAut$(X)$ acts transitively on the smooth locus $X_{reg}$ \cite{AKZ}. Recall that the special automorphism group SAut$(X)$ is the subgroup of the automorphism group Aut$(X)$ generated by all one-parameter unipotent subgroups \cite{AKZ}. Given a normal, flexible, affine variety $X$ and a closed subvariety $Y$ in $X$ of codimension at least 2, we show that the pointwise stabilizer subgroup of $Y$ in the group SAut$(X)$ acts infinitely transitively on the complement $X\backslash Y$, that is, $m$-transitively for any $m\ge 1$. More generally we show such a result for any quasi-affine variety $X$ and codimension $\ge 2$ subset $Y$ of $X$. In the particular case of $X=\AA^n$, $n\ge 2$, this yields a Theorem of Gromov and Winkelmann \cite{Gr1}, \cite{Wi}.
math.AG
math
THE GROMOV-WINKELMANN THEOREM FOR FLEXIBLE VARIETIES H. FLENNER, S. KALIMAN, M. ZAIDENBERG Abstract. An affine variety X of dimension ≥ 2 is called flexible if its special au- tomorphism group SAut(X) acts transitively on the smooth locus Xreg [1]. Recall that SAut(X) is the subgroup of the automorphism group Aut(X) generated by all one-parameter unipotent subgroups [1]. Given a normal, flexible, affine variety X and a closed subvariety Y in X of codimension at least 2, we show that the point- wise stabilizer subgroup of Y in the group SAut(X) acts infinitely transitively on the complement X\Y , that is, m-transitively for any m ≥ 1. More generally we show such a result for any quasi-affine variety X and codimension ≥ 2 subset Y of X. In the particular case of X = An, n ≥ 2, this yields a Gromov -- Winkelmann Theorem [5], [13]. Contents Introduction 1. Main theorem 1.1. Basic notions and the main result 1.2. Transitivity versus flexibility on quasi-affine varieties 1.3. Generation of subgroups by LND's. 2. m-blowups, tangency, and m-contractions 2.1. m-blowups and tangency 2.2. m-contractions 3. Replicas as m-contractions 4. Proof of the main theorem 4.1. Algebraic families of automorphisms 4.2. Proof of the main theorem References 1 3 3 6 7 9 9 11 15 20 20 21 25 Introduction Throughout the paper X will be an algebraic variety of dimension ≥ 2 over an alge- braically closed field k of characteristic 0. The special automorphism group SAut(X) Date: July 28, 2018. The research of the second author was supported by an NSA grant No. H98230-10-1-0185. This work was done during a stay of the authors at the Max Planck Institut fur Mathematik at Bonn, and a stay of the first and the second authors at the Institut Fourier, Grenoble. The authors thank these institutions for hospitality. 2010 Mathematics Subject Classification: 14R20, 32M17. Key words: affine varieties, group actions, one-parameter subgroups, transitivity. 1 2 H. FLENNER, S. KALIMAN, M. ZAIDENBERG of such a variety X is the subgroup of the full automorphism group Aut(X) generated by all one-parameter unipotent subgroups of Aut(X).1 Let U(X) denote the set of all these subgroups. A quasi-affine variety X is called flexible, if the tangent space TxX in any smooth point x ∈ Xreg is spanned by the tangent vectors at x to the orbits U.x, where U runs over U(X). If X is affine then this amounts to the notion of flexibility as introduced in [1, 2]. For such varieties the flexibility is equivalent to the transitivity, and even to infinite transitivity of the group SAut(X) acting on the smooth locus Xreg of X (see [2, The- orem 0.1]). (We say that a group action is infinitely transitive if it is m-transitive for any m ≥ 1.) These characterizations of flexibility can be extended to any quasi-affine variety (see Remarks 1.7 and Theorem 1.11 in Sect. 1). It is worthwhile mentioning that the class of flexible varieties is rather wide. It includes in particular - homogeneous spaces of semi-simple groups (and even homogeneous spaces of exten- sions of semi-simple groups by unipotent radicals); - non-degenerate toric varieties (i.e. toric varieties without nonconstant invertible reg- ular functions); - cones over flag varieties and anti-canonical cones over Del Pezzo surfaces of degree at least 4; - normal hypersurfaces of the form uv = p(¯x) in Cn+2 u,v,¯x; - homogeneous Gizatullin surfaces; see [1], [2], [9]. If on a quasi-affine variety X the group SAut(X) has an open orbit, then this open orbit is a flexible quasi-affine variety. A normal quasi-affine variety X is flexible if and only if so is Xreg. In its simplest form the main result of this paper is the following theorem; see Sect. 1 for generalizations and refinements. Theorem 0.1. Let X be a smooth quasi-affine variety of dimension ≥ 2 and Y ⊆ X a closed subscheme of codimension ≥ 2. If X is flexible then so is X\Y . That is, if SAut(X) acts transitively on X then SAut(X\Y ) acts transitively on X\Y . We note that in the setup of the Theorem any action of a unipotent group on X\Y extends to an action on X preserving Y ; see Proposition 1.8 for a more general statement. Moreover, our main result (see Theorem 1.6) yields that the pointwise stabilizer SAutY (X) acts transitively on X\Y . This answers in affirmative a question posed in [2, 4.22(2)]. Partial results in this direction were obtained in Theorem 2.5 and Proposition 4.19 in [2], see also Proposition 1.11 below. Let us note that Theorem 0.1 does not hold for subsets Y of X of codimension 1, in general; see [2, Proposition 4.13]. In this sense the result above is optimal. For an affine space X = An, n ≥ 2, the flexibility of X\Y was first observed by M. Gromov in [5, §2.1.5, p. 72, Exercise (b′)], cf. also 4.6(b) and 5.3(c) in [6]. The transitivity of SAutY (X) in X\Y was proven in this particular case by J. Winkelmann [13, §2, Proposition 1]. 1I.e. by subgroups isomorphic to Ga. By abuse of language we do not distinguish between one- parameter unipotent subgroups of the group Aut(X) and effective Ga-actions on X. GROMOV-WINKELMANN THEOREM 3 The paper is organized as follows. In Section 1 we recall some useful facts from [2] and formulate, after introducing necessary definitions, a stronger version of Theorem 0.1, see Theorem 1.6. As an important ingredient of the proof we show that for any flexible variety X one can find a subgroup of SAut(X) acting with an open orbit on X, which is generated by two locally nilpotent derivations δ0, δ1 along with their replicas f0δ0, f1δ1, where f0 ∈ ker δ0 and f1 ∈ ker δ1; see Proposition 1.14. In Sections 2 and 3 we prepare the setup for the proof of Theorem 1.6. The proof is then contained in Section 4. It should be possible, after reading Section 1, to go directly to Section 4 addressing results in Sections 2 and 3 when necessary. Let us sketch the scheme of the proof of Theorem 0.1. By a result in [2] the pointwise stabilizer SAutY (X) of Y in SAut(X) has an open orbit, say, O in X. We consider a completion ¯X of X compatible with partial quotients by the two Ga-subgroups U0 = exp(kδ0) and U1 = exp(kδ1), where δ0 and δ1 are as in Proposition 1.14. These quotients define on ¯X two P1-fibrations ¯0, ¯1 with privileged sections D0, D1, which lie on the boundary of X in ¯X. Acting with a suitable replica of U0 one can move the part of the boundary ∂Y ∩ D1 to a fixed proper subset of D1, and symmetrically for U1 and ∂Y ∩ D0, see Proposition 3.11. Up to a controllable (and so negligible) proper subset of D0 ∪ D1, this property is preserved when we iterate subsequently actions by suitable replicas of U0 and U1, see Proposition 4.11. Using the transitivity property of the subgroup H ⊆ SAut(X) generated by U0, U1 and their replicas, we can move a given codimension ≥ 2 subset Y as in Theorem 0.1 and, simultaneously, a given point x ∈ X\Y to a generic fiber, say, F of the P1-fibration ¯0 so that F does not meet ∂Y ∩ D0. Using the Transversality Theorem from [2] we can achieve that F does not meet Y hence in total F and ¯Y are disjoint. This enables us to find a U0-invariant function f ∈ OX (X), which vanishes on Y and not in x. The corresponding replica U ′ 0 of U0 fixes Y and moves x along F . Since the fiber F is generic it meets the open orbit O of SAutY (X), hence so does U ′ 0.x. Thus x belongs to O, and so O = X\Y , as stated. In order to prove Propositions 3.11 and 4.11 we develop in Sections 2 and 3 a ma- chinery, which allows to reduce the proof to the model case of a standard birational transformation of a ruled surface induced by a Ga-action. This reduction is the most lengthly part of the proof. We thank M. Gizatullin for his interest in our work and in particular for his suggestion to treat in Theorem 1.6 also non-reduced subschemes Y of X. 1. Main theorem 1.1. Basic notions and the main result. We let An = An k and Ga = Ga(k). In the sequel X denotes a quasi-affine variety over k. Thus X can be embedded into an affine variety X ′ = Spec B as an open subset. We let A = OX(X) so that B is a finitely generated k-subalgebra of A. The embedding X ֒→ X ′ factors as X → Spec A → Spec B. Furthermore X ֒→ Spec A is an open embedding. We note that A is in general not a finitely generated algebra over k. Lemma 1.1. With the notation as above the following hold. 4 H. FLENNER, S. KALIMAN, M. ZAIDENBERG (a) Every action of an algebraic group on X extends in a canonical way to Spec A. (b) Every subgroup U ∈ U(X) with infinitesimal generator δ yields a locally nilpotent k-derivation on A. Proof. (a) is standard, and (b) is a consequence of (a). (cid:3) Let us recall some notions and useful facts from [2]. Given a subgroup U ∈ U(X) we let δ denote an infinitesimal generator of U; the latter is uniquely determined up to a nonzero constant factor. Thus δ is a locally nilpotent derivation of the algebra A = OX (X) such that U = exp(kδ). Geometrically δ can be viewed as a complete vector field on X with phase flow ut = exp(tδ), t ∈ k. The tangent vector at the point x ∈ X given by this vector field is denoted δx. Lemma 1.2. Let Y be a closed (not necessarily reduced) subscheme of the quasi-affine variety X with ideal sheaf I ⊆ OX , and consider the ideal of global sections I = I(X) ⊆ A = OX (X). Given U ∈ U(X) with an infinitesimal generator δ the following hold. (a) δ(A) ⊆ I if and only if uY = idY for any u ∈ U. (b) δ(I) ⊆ I if and only if u.Y ⊆ Y for any u ∈ U.2 Let us fix the following notation. Notation 1.3. (a) Let as before X be a quasi-affine variety and A = OX (X) be its ring of regular functions. If a ⊆ A is the ideal of the complement Spec(A)\X, then the set of nonzero locally nilpotent derivations δ of A with δ(a) ⊆ a is denoted by LND(X). In view of Lemmas 1.1 and 1.2(b) any element δ ∈ LND(X) gives rise to a one- parameter subgroup U = exp(kδ) in U(X) and vice versa. (b) In order to deal with quasi-affine varieties we choose a k-subalgebra Λ of A such that the induced map X → Spec Λ is an open embedding. Letting b be the ideal of the complement Spec(Λ)\X we let LNDΛ(X) denote the set of all locally nilpotent derivations δ on Λ with δ(b) ⊆ b. Every such derivation induces as before a one- parameter subgroup U ∈ U(X) and consequently extends to an element in LND(X). Thus LNDΛ(X) can be considered as a subset of LND(X). (c) Given a collection N ⊆ LNDΛ(X) of nonzero locally nilpotent derivations we let G = GN = hN i be the subgroup of the group SAut(X) generated by the corresponding one-parameter unipotent subgroups U = exp(kδ), δ ∈ N . Remarks 1.4. 1. We emphasize that the subring Λ of A is not supposed to be finitely generated over k so that the choice Λ = A is also possible. In other words, we consider X as an open subset of an affine k-scheme Spec Λ, which is not necessarily an algebraic variety, in contrast with [2]; see also Remark 1.7 below. 2. We observe as well that the G-action on X as in 1.3(c) extends to a G-action on the affine scheme Spec Λ. Let us recall some notation and standard facts. 2In the terminology of [4, p. 10] this means that I is an integral ideal. GROMOV-WINKELMANN THEOREM 5 1.5. (1) Given a group G = GN as before, the set of all one-parameter unipotent subgroups of G will be denoted by U(G), and the set of all nonzero locally nilpotent derivations on Λ generating one-parameter subgroups of G by LNDΛ(G) or simply LND(G). (2) A Λ-replica of a subgroup U = exp(kδ) ∈ U(G) is a subgroup Uf = exp(kf δ) ∈ LNDΛ(G), where f ∈ Λ is in the kernel of δ ([2]). (3) We say that N is Λ-saturated if N is closed under conjugation by elements in G and taking Λ-replicas i.e., f δ ∈ N ∀δ ∈ N and ∀f ∈ kerΛ δ . Hereafter Λ will be fixed, hence in most cases we omit the symbol Λ and say simply 'replica' or 'saturated'. (4) A point x ∈ X is called G-flexible if TxX = Span(N (x)), where N (x) denotes the set of tangent vectors δx with δ ∈ N . We say that X is G-flexible if Xreg consists of G-flexible points. (5) Given a (not necessarily reduced) closed subscheme Y in X we let GN ,Y denote the subgroup of G generated by all replicas f δ in N vanishing on Y in the ideal theoretic sense, see Lemma 1.2(a). Therefore GN ,Y ⊆ GY , where GY = {g ∈ G : gY = idY } stands for the 'pointwise' stabilizer of Y in G in the scheme theoretic sense. The following result is our main theorem. Theorem 1.6. Let X be a quasi-affine variety of dimension ≥ 2 and X ֒→ Spec Λ be an open embedding into an affine k-scheme, see 1.3(b). Let G = hN i be a subgroup of the group SAut(X) generated by a Λ-saturated set N of locally nilpotent derivations as in 1.5. Suppose that X is G-flexible. If Y is a closed (possibly non-reduced3) subscheme of X of codimension ≥ 2, then the complement X\Y is GN ,Y -flexible. In the case of a smooth variety X applying Theorem 1.6 to the group G = SAut(X) we get Theorem 0.1 from the Introduction. Remarks 1.7. 1. Since G ⊆ Aut(Spec Λ) the variety X satisfies the requirements of Theorem 1.6 whenever so does its (G-stable) regular locus Xreg. Therefore it suffices to prove Theorem 1.6 under the assumption that X is smooth. This explains the necessity to fix a subring Λ ⊆ A as in 1.3(b). Indeed, A can be properly contained in A′ = OXreg(Xreg). If instead of fixing Λ we consider always LND's and their replicas with respect to the ring A = OX (X), then an A′-replica is possibly not an A-replica and so the notion of saturated set of derivations could change when passing from X to Xreg. 2. The viewpoint of the paper [2] is slightly different as it deals with open subsets X of affine algebraic varieties Z = Spec B, and with subgroups G of SAut(Z) stabilizing X. It might happen in principle that although Aut(X) acts transitively on Xreg there is no subgroup G of Aut(Z) acting transitively on Xreg, whatever is the choice of an 3The authors are grateful to M. Gizatullin for the suggestion to take also into account non-reduced subschemas Y of X. 6 H. FLENNER, S. KALIMAN, M. ZAIDENBERG embedding of X into an affine variety Z; cf. Question 1.10 below. Thus a priori our viewpoint here is more general. 3. Working with quasi-affine varieties has yet another advantage: given a subgroup G ⊆ SAut(X), in the subsequent proofs we may at any step replace X by an open orbit of G. This considerably simplifies our notation. It is worthwhile to note that if X as in Theorem 1.6 is normal then the group SAut(X\Y ) is in a natural way a subgroup of SAut(X). This is a consequence of the following proposition. Proposition 1.8. Let X be a normal quasi-affine variety and Y ⊆ X a subset of codimension ≥ 2. Then every Ga-action on X\Y extends to a Ga-action on X that stabilizes Y . Proof. A Ga-action on X\Y corresponds to a locally nilpotent derivation on A = OX(X\Y ) such that the ideal, say, c of the complement Z\(X\Y ) is stabilized by δ, where Z = Spec A. Because of codimX Y ≥ 2 the k-algebras A and OX(X) coincide. Consider the ideal a ⊆ A of the complement X c = Z\X and the ideal b ⊆ A of the closure ¯Y so that c = a ∩ b is the ideal of the complement of X\Y in Z. We have to show that a is stabilized by δ. This is easy in the case that A is finitely generated, thus Z is an affine algebraic variety. Indeed, if U stabilizes X c ∪ ¯Y then it stabilizes all irreducible components of that set (see e.g. [4, Proposition 1.14(b)]), thus also X c and ¯Y and consequently their respective ideals. In the general case, by Lemma 1.9 below A is a direct limit of its ∂-stable finitely generated subalgebras Ai such that X embeds as an open subset into Spec Ai. Applying the first case to every Ai the result follows easily. (cid:3) The following fact is an easy consequence of the Lemma of Cartier [10, Chapt. I, §1]. Lemma 1.9. Given δ ∈ LNDΛ(X) and a finite dimensional k-subspace E ⊆ Λ there is a finitely generated δ-stable k-subalgebra Λ′ ⊆ Λ containing E such that X embeds as on open subset of the affine variety Spec Λ′. Since X is quasi-affine there is a finitely generated subalgebra C of B such that X embeds as an open subset in Spec C. We may suppose that E contains a finite set of generators of C. Since ∂ is locally nilpotent, the set E′ = Si≥1 ∂i(E) is finite. Since it is also ∂-stable, it generates a subalgebra Λ′ of C with the desired properties. We do not know whether this result remains true for any finite collection of locally nilpotent derivations. More precisely: Question 1.10. Suppose that N ⊆ LND(X) is a finite subset. Does there exist a finitely generated N -stable k-subalgebra Λ′ of A = OX (X) such that X embeds into Spec Λ′ as an open subset? 1.2. Transitivity versus flexibility on quasi-affine varieties. Let X = Spec A be an affine variety. By the main result in [2] the flexibility of X is equivalent to the GROMOV-WINKELMANN THEOREM 7 transitivity of SAut(X) on Xreg, which in turn is equivalent to infinite transitivity. In the sequel we need this and related facts in the more general setting of quasi-affine varieties. We will state the necessary results in the generality that we need below. The proofs in [2] can be carried over to our more general quasi-affine setup without any difficulty. Let us start with the main result of [2], see 1.11 and 2.2 in loc.cit. Theorem 1.11. Let X be a smooth, quasi-affine variety of dimension ≥ 2, and let G = hN i be a subgroup of SAut(X) generated by a Λ-saturated set N ⊆ LNDΛ(X) as in Notation 1.3 and 1.5. Then the following are equivalent. (i) X is GN -flexible. (ii) GN acts transitively on X. (iii) GN acts infinitely transitively on X. In the proof of Theorem 1.6 we use the following auxiliary results. They are estab- lished in 2.5, 4.19, and 4.2 in [2] in the case of affine schemes X and reduced subvarieties Y of X. The proofs given there carry immediately over to our more general situation. Proposition 1.12. Let X and GN be as in Theorem 1.11, and let Y be a closed subscheme of X. If X is GN -flexible4 then the following hold. (1) The group GN ,Y acts on X\Y with a dense open orbit, say, OY , which consists of all GN ,Y -flexible points of X\Y . Consequently, the GN ,Y -action on OY is infinitely transitive. (2) If Y is finite then OY = X\Y . (3) If x ∈ X then the image of the tangent representation GN ,x → GL(TxX) given by the differential coincides with the special linear group SL(TxX). Finally we need the following interpolation result, see [2, Theorem 4.14 and Remark 4.16]. Proposition 1.13. Let X and GN be as in Theorem 1.11. If G acts transitively on X then for any finite subset Z ⊆ X there exists an automorphism g ∈ G with g(x) = x for x ∈ Z and prescribed tangent map dxg ∈ SL(TxX) at the points x ∈ Z.5 1.3. Generation of subgroups by LND's. Let as before X be a quasi-affine al- gebraic variety of dimension n ≥ 2 equipped with an open embedding into an affine k-scheme Spec Λ, where Λ ⊆ OX (X). Given a set of locally nilpotent derivations N ⊆ LNDΛ(X) we enrich it by adding all the Λ-replicas of derivations in N . Letting N be this enlarged set we consider the subgroup hhN ii := h N i of the group Aut(X) generated by N . In this section we prove the following result. 4Equivalently, if GN acts transitively on X. 5In fact this proposition holds more generally for any finite collection of m-jets provided these jets fix the corresponding points and preserve local volume forms on X at these points; see [2, Remark 4.16]. 8 H. FLENNER, S. KALIMAN, M. ZAIDENBERG Proposition 1.14. Let G = hN i ⊆ SAut(X) be a subgroup generated by a Λ-saturated set N of locally nilpotent derivations. Suppose that G acts transitively on X. Then for any locally nilpotent derivation δ0 ∈ N one can find another one δ1 ∈ N such that the subgroup (1) H = hhδ0, δ1ii generated by δ0, δ1 and all their replicas acts with an open orbit on X. To deduce this result let us recall a few facts. Let U be a one-parameter unipotent subgroup with an infinitesimal generator δ ∈ LNDΛ(X) (see Notation 1.3). By assump- tion X is contained as an open subset in Spec Λ and by Lemma 1.9 even in Spec Λ′ for some δ-stable finitely generated subalgebra Λ′ of Λ. By the Rosenlicht Theorem (see [11, Theorem 2.3]) one can find a finite set of U-invariant functions f1, . . . , fm ∈ Λ′U , which separate general U-orbits. Let B be the integral closure of the finitely generated k-algebra k[f1, . . . , fm]. It is a standard result that B is again finitely generated, see e.g. [3, Theorem 4.14]. Definition 1.15. The normal affine variety QU = Spec B will be called a partial quotient of X by U. In general it depends on the choice of the functions f1, . . . , fm.6 The inclusion B ֒→ OX (X) defines a dominant morphism U : X → QU such that the general fibers of U are general orbits of U. Proof of Proposition 1.14. Let as before 0 : X → Q0 be a partial quotient of X by U 0, where dim Q0 = n − 1. Since n ≥ 2 there exists σ ∈ N such that ker σ 6= ker δ0 and so U 0 and U = exp(Cσ) have different general orbits. We can choose x ∈ X such that the tangent vector δ0,x of δ0 at x is nonzero, hence dim U 0.x = 1. Choosing x in an appropriate way there are points x1, . . . , xn−1 on the orbit U 0.x such that the vectors v′ i = σxi ∈ TxiX are all nonzero. Letting q = 0(x) ∈ Q0 we fix for each i = 1 . . . , n−1 a tangent vector vi ∈ TxiX in such a way that the vectors d0(vi) ∈ TqQ0, i = 1, . . . , n − 1, generate the tangent space TqQ0 to Q0 at q. For every i = 1, . . . , n − 1 we can choose a 1-jet of a local automorphism at the point xi that fixes xi and sends i to vi. This amounts to choosing αi ∈ SL(TxiX) such that αi(v′ v′ i) = vi. According to Proposition 1.13 one can interpolate these jets by an automorphism, say, α ∈ G such that α(xi) = xi and dα(v′ i) = vi for i = 1, . . . , n − 1. Replacing U by U 1 = α ◦ U ◦ α−1 = exp(Cδ1) ∈ U(G) . we obtain a one-parameter unipotent subgroup with tangent vector vi at xi, i = 1, . . . , n − 1. We claim that the locally nilpotent derivation δ1 satisfies our require- Indeed, δ1 ∈ N since N is saturated and so, in particular, is closed under ment. conjugation in G. Consider the conjugated one-parameter subgroups U 1 i = α−1 i ◦ U 1 ◦ αi = exp(Cσi) ∈ U(H), i = 1, . . . , n − 1 , where αi ∈ U 0 is an element which maps x to xi. Here H is as in (1) and σi is a conjugate of δ1 under the action of H for i = 1, . . . , n − 1. For any i in this range 6Alternatively, one could use the Winkelmann quotient [14]. This quasi-affine quotient is canonically defined, but has the disadvantage to be non-affine, in general. GROMOV-WINKELMANN THEOREM 9 the vector ui = dαi(vi) is tangent to the orbit U 1 i .x at the point x ∈ X. Furthermore, the vectors d0(ui) = d0(vi) ∈ TqQ0, i = 1, . . . , n − 1 still generate TqQ0. Hence the vectors u0 = δ0,x, u1 = σ1,x, . . . , un−1 = σn−1,x ∈ TxX span TxX as well. Consequently, x is an H-flexible point and so the H-orbit H.x is open and dense in X (see [2, Corollary 1.11(a)]). (cid:3) 2. m-blowups, tangency, and m-contractions This section is technical; we use its results and notions (see especially Definitions 2.5 and 2.8 and Proposition 2.15) in the proof of Proposition 3.11 in the next section. 2.1. In the sequel we deal with rational maps g : X .......✲ Y which fit into a diagram X h ✛ g g′ ✲ X ............................. ✲ Y where h is a sequence of blowups and g′ is a proper morphism. This somewhat restricted class of rational maps is suitable for our purposes. Given subsets A ⊆ X and B ⊆ Y we let g(A) = g′(h−1(A)) and g−1(B) = h(g′−1(B)) denote the total image and preimage, respectively.7 Since any two resolutions of the indeterminacy set are dominated by a third one, the total image and the total preimage are well defined. 2.1. m-blowups and tangency. In the next Definition we introduce a setup which is used repeatedly in this and the next section. Definition 2.2. Let X be an algebraic variety and C, D be divisors in X, which are Cartier near C ∩ D. The m-blowup σm : Xm → X of D along C is defined recursively as follows. With X0 = X we let X1 be the blowup of X along the subscheme C ∩ D. If Xm−1 is already defined for some m ≥ 2, then we let Xm → Xm−1 be the blowup along D(m−1) ∩ Em−1, where D(m−1) is the proper transform of D in Xm−1 and Em−1 the exceptional set of the previous blowup Xm−1 → Xm−2. In the following we call the proper transforms E′ 1, . . . , E′ m ⊆ X ′ = Xm of the exceptional sets Ei of Xi → Xi−1 the exceptional sets of the m-blowup of D along C. The proper transforms of C and D will always be denoted C ′, D′, respectively. 7These notions should be treated with caution, because they are not compatible with composition of rational maps. 10 H. FLENNER, S. KALIMAN, M. ZAIDENBERG Example 2.3. Suppose that S is a complete smooth surface and C ∩ D = {p}, where the intersection is transversal. Then the dual graph of C ′ ∪ E′ m ∪ D′ is a linear chain: 1 ∪ . . . ∪ E′ (2) C 2 − 1 ❝ C ′ −2 ❝ E′ 1 . . . −1 ❝ E′ m D2 − m ❝ D′ . Let us consider next the effect of an m-blowup as in Definition 2.2 on the boundary of a closed subset of X. Proposition 2.4. We keep the notation and assumptions as in Definition 2.2. Given a closed subset Y ⊆ X we let Y ′ denote its proper transform in X ′ and ∂Y ′ its boundary ∂Y ′ = Y ′ ∩ σ−1 m (C ∪ D). Then with P = Y ∩ D\C, for m ≫ 0 ∂Y ′ ⊆ E′ 1 ∪ . . . ∪ E′ m−1 ∪ σ−1 m (P ) . Proof. The assertion is local around points in C ∩ D\P . Thus we may assume that P = ∅, X = Spec A is affine, and that D = V (x), C = V (y) with functions x, y ∈ A. The subset m−1 U ′ = X ′\ E′ i [i=0 of X ′ is affine with coordinate ring A′ = A[u] , where u = x/ym, cf. Lemma 2.10 below for the special case of surfaces. Furthermore m = {y = 0} and U ′ ∩ D′ = {u = 0}. U ′ ∩ E′ (3) If I ⊆ A is the ideal of Y then B = A/I is the affine coordinate ring of Y . Since Y ∩ D\C = ∅ the set Y ∩ D is contained in C ∩ D and so the localization (B/xB)y is zero. Hence there exists a natural number m such that ym−1 ∈ xB. In other words, we can find a ∈ A such that ym−1 − a · x ∈ I . (4) In the blowup ring A′ the ideal I ′ of Y ′ is given by Since u = x/ym condition (4) can be rewritten in the form I ′ = {g ∈ A′ ∃k ∈ N : ykg ∈ IA′}. ym−1 · (1 − yau) ∈ IA′. Hence 1 − yau ∈ I ′. This shows that in the affine coordinate ring B′ = A′/I ′ of U ′ ∩ Y ′ the residue classes of y and u are units. In view of (3) this implies that U ′ ∩ Y ′ ∩ E′ m = ∅ and U ′ ∩ Y ′ ∩ D′ = ∅, which immediately yields the required result. (cid:3) Definition 2.5. We say that a closed subset Y of X is at most m-tangent to D along C, if the conclusion of Proposition 2.4 holds with this particular value of m. The subset N = C ∩ Y ∩ D\C of C ∩ D will be called the defect set. GROMOV-WINKELMANN THEOREM 11 We note that if Y is at most m-tangent to C along D then it is also at most m′- tangent to C along D for all m′ ≥ m. The following observation is important. Lemma 2.6. If codimX Y ≥ 1 and Y \D is dense in Y then the defect set N is nowhere dense in C ∩ D. Proof. If codimX Y ≥ 1 then the set Y ∩ D has codimension ≥ 1 in D. Hence its closure cannot contain any component of C ∩ D. (cid:3) Remark 2.7. In the setup of Proposition 2.4 suppose that (Ys)s∈S is a family of proper closed subsets of X. Then there is a natural m such that Ys is at most m-tangent to D along C for any s ∈ S. This follows easily from the fact that the construction of Proposition 2.4 can be done al least generically in the given family and that it is then compatible with restriction to the general fiber. More precisely, one can find an open dense subset U ⊆ S so that all fibers Ys are at most m-tangent to D along C with m independent of s ∈ U, and with a defect set Ns = C ∩Ys ∩ D\C. Restricting the family to S′ = S\U and applying induction on dim S, we may assume that Ys is at most m-tangent to D along C for any s ∈ S′. Hence the assertion follows. 2.2. m-contractions. Definition 2.8. Let C, D be divisors on an the algebraic variety X, which are Cartier near C ∩ D. Consider a birational map g : X .......✲ X and a resolution of the indeter- minacy set of g which factors through the m-blowup σm : X ′ = Xm → X of D along C, see Definition 2.2: X m h g′ h g is called an m-contraction for C along D if the following hold. ✛ X ′ = Xm σm ✲ ✲ ❄ g X .......... ✲ X (1) g is biregular in the points of X\C; (2) with gm = g ◦ σm, the total image8 gm(C ′ + E′ where E′ 1, . . . , E′ m are as in Definition 2.2. 1 + . . . + E′ m−1) is a subset of D, Clearly, an m-contraction for C along D is also an m′-contraction for C along D for any m′ ≤ m. The following example is important and serves as a model case. Notation 2.9. Let Γ = (Γ, o) be a germ of a smooth affine curve with a uniformizing parameter u such that u(o) = 0, and let d(u) denote a nowhere vanishing function on Γ. We consider homogeneous coordinates (ζ1 : ζ2) on P1 and an affine coordinate v = ζ1/ζ2 on A1 = P1\{(1 : 0)}. The product S := Γ × P1 is a P1-fibered surface over Γ. Its fiber, say, C over o ∈ Γ and the section D = Γ × {(0 : 1)} ⊆ Γ × A1 can be described in coordinates by C = {u = 0} and D = {v = 0}. 8See 2.1 12 H. FLENNER, S. KALIMAN, M. ZAIDENBERG Let us study the rational map gm : S 99K S, where m ∈ N, given in affine coordinates by (5) gm(u, v) = (cid:18)u, umv d(u)v + um(cid:19) . Its indeterminacy set consists of the intersection point C ∩ D = {u = v = 0}, which will be denoted by ¯0. Lemma 2.10. Let S′ gm g′ m ✲ ✲ S m σ ✛ S be the minimal resolution of indeterminacies of gm, where σm is a sequence of blowups and g′ m is a morphism. Then the total transform of C + D on S′ under σm has weighted dual graph (6) −1 ❝ C ′ −2 ❝ E′ 1 ❝D′ −m −2 ❝ E′ m . . . . . . −2 −1 ❝ ❝ , E′ 2m−1 E′ 2m where C ′ and D′ are the proper transforms of C and D, respectively. The map σm contracts the components E′ m contracts the curves C ′, E′ 2m) = C. 2m to the origin ¯0 ∈ S, while g′ m(E′ 2m−1 to ¯0 ∈ S. Furthermore g′ m(D′) = D and g′ 1, . . . , E′ 1, . . . , E′ Proof. Letting v0 = v we define a sequence of coordinates charts (u, vi) on S′, i = 0, . . . , 2m, so that the 2m blowing-downs over the origin with exceptional curves E′ 2m that constitute the map 1, . . . , E′ σ : (u, v2m) 7→ (u, v2m−1) 7→ . . . 7→ (u, v1) 7→ (u, v) can be described by the formulae (7) v1 = v/u, v2 = v1/u = v/u2, . . . , vm = vm−1/u = v/um , and (8) vm+1 = (1 + d(u)vm)/u, vm+2 = vm+1/u, . . . , v2m = v2m−1/u = (1 + d(u)vm)/um . The map gm can be written in these coordinate charts as (u, v) 7→ (cid:18)u, umv d(u)v + um(cid:19) = (cid:18)u, umv1 d(u)v1 + um−1(cid:19) = . . . (cid:19) = . . . = (cid:18)u, d(u)umv2m − 1 v2m (cid:19) . . . . = (cid:18)u, umvm 1 + d(u)vm(cid:19) = (cid:18)u, d(u)umvm+1 − um−1 vm+1 Hence the curve E′ i given in the chart (u, vi) by equation u = 0 is contracted under g′ m for every i = 0, . . . , 2m − 1, while the curve E′ 2m given by the same equation in the chart (u, v2m) maps birationally onto the curve C in S. Now the assertion follows. (cid:3) GROMOV-WINKELMANN THEOREM 13 An immediate consequence is the following corollary. Corollary 2.11. The birational map gm in (5) is an m-contraction of C along D. Let us note that gm is not an (m + 1)-contraction of C along D. This example can be generalized to higher dimensions as follows. Notation 2.12. Instead of a curve Γ in 2.9 we consider now a smooth affine algebraic variety Q and a smooth divisor T ⊆ Q given by the equation {u = 0}, where u ∈ OQ(Q). The product X = Q × P1 is P1-fibered over Q and contains the divisors C = T × P1 and D = Q × {(0 : 1)} ⊆ Q × A1, where we equip P1 with homogeneous coordinates (ζ1 : ζ2). As before v = ζ1/ζ2 stands for an affine coordinate on A1 = P1\{(1 : 0)}. Thus we have C = {u = 0} and D = {v = 0}. Lemma 2.13. Given a nowhere vanishing function d(q) on Q and m ∈ N the rational map (9) gm : X 99K X , where is an m-contraction of C along D. Proof. A resolution gm(q, v) = (cid:18)q, u(q)mv d(q)v + u(q)m(cid:19) , X ′ gm g′ m ✲ ✲ X m σ ✛ X of the indeterminacy points of gm can be obtained (with obvious changes) by the same sequence of blowups as in the proof of Lemma 2.10. Letting v0 = v we define a sequence of coordinates charts (q, vi) ∈ Ui = Q × A1 on X ′, i = 0, . . . , 2m, so that the 2m blowdowns over C ∩ D with exceptional divisors E′ 2m that constitute the map 1, . . . , E′ σ : (q, v2m) 7→ (q, v2m−1) 7→ . . . 7→ (q, v1) 7→ (q, v) can be described by the formulae in (7) and (8), where u is now the function u(q). With the same calculation as before the map gm can be written in these coordinate charts as (q, v) 7→ (cid:18)q, d(q)u(q)mv2m − 1 (cid:19) . As in the proof of 2.10 the exceptional set E′ u = 0, and it is contracted under g′ Finally, the exceptional set E′ isomorphically onto the divisor C in X. Since the divisors C ′, E′ contracted under g′ v2m i is given in the chart Ui by the equation m to the subset C ∩ D for every i = 0, . . . , 2m − 1. 2m given by {u = 0} in the chart U2m maps under g′ m−1 in X ′ are (cid:3) m to C ∩ D, the result follows. 1, . . . , E′ m Next we show that m-contractions are compatible with certain blowups. 14 H. FLENNER, S. KALIMAN, M. ZAIDENBERG Proposition 2.14. Let X be an algebraic variety and C, D be connected divisors on X, which are Cartier near C ∩ D. Let g : X .......✲ X be an m-contraction of C along D and p : Z → X be a modification, which is an isomorphism over D ∪ (X\C). Then the rational map f : Z .......✲ Z induced by g is an m-contraction of CZ = p−1(C) along DZ = p−1(D) ∼= D. Proof. Let Xm → X and Zm → Z be the m-blowups of X and Z, respectively. Since p is an isomorphism in the points near D, the exceptional sets E′ m of Xm → X can be identified in a natural way with the exceptional sets, say, E′ m,Z of Zm → Z. Consider the composed rational maps 1,Z, . . . , E′ 1, . . . , E′ Z ′ m and a diagram fm .................... ✲ Z and X ′ m gm ..................... ✲ Z . Z m h f′ m ✛ ............................ fm ✲ ✲ Z ′ m Z p ❄ p′ ❄ X ′ m gm ............................ ✲ X where Z is a resolution of the indeterminacy locus of fm and then also of gm. By our assumption the set (p′ ◦ hm)−1(C ′ ∪ E′ 1 ∪ . . . ∪ E′ m−1) = h−1 m (C ′ Z ∪ E′ 1,Z ∪ . . . ∪ E′ m−1,Z) is contracted under p◦ f ′ set is already contracted under f ′ m to a subset of D. Since p is an isomorphism near D the latter (cid:3) m to a subset of D. This proves the assertion. Let us now study the effect of an m-contraction of C along D on the boundary of a closed subset Y of X. Proposition 2.15. Let X be an algebraic variety and C, D divisors on X, which are Cartier near C ∩ D. Assume that g : X .......✲ X is an m-contraction of C along D and that Y ⊆ X is a closed subset, which is at most m-tangent to C along D with defect set N = C ∩ Y ∩ D\C. Then the proper image Y of Y under g satisfies ∂ Y ⊆ D ∪ g(N) , where g(N) is the total image of N and ∂ Y denotes the intersection of Y with D ∪ C. 1, . . . , E′ Proof. Let σm : X ′ = Xm → X be the m-blowup of C along D with exceptional m and consider the composition gm = g ◦ σm : X ′ .......✲ X. We can find a sets E′ resolution of the indeterminacy locus of gm X m h ✛ g′ m ✲ gm X ′ ........................... ✲ X . GROMOV-WINKELMANN THEOREM 15 Since Y is at most m-tangent to C along D, the boundary ∂Y ′ of the proper transform Y ′ of Y in X ′ satisfies where P = Y ∩ D\C, see Proposition 2.4. By condition (2) in Definition 2.8 ∂Y ′ ⊆ E′ 1 ∪ . . . ∪ E′ m−1 ∪ σ−1 m (P ) , h−1 m (C ′ ∪ E′ 1 ∪ . . . ∪ E′ m−1) is contracted under g′ m to a subset of D. Hence m (∂Y ′)) ⊆ D ∪ g′ m(h−1 m (σ−1 g′ m(h−1 m (P )))) = D ∪ g(P ). Since g′ m is proper the set on the right is easily seen to contain ∂ Y , as stated. (cid:3) 3. Replicas as m-contractions Notation 3.1. (a) Let X be a smooth quasi-affine algebraic variety and GN a group of automorphisms on X generated by a set of Λ-saturated locally nilpotent derivations N ⊆ LNDΛ(X), see Notation 1.3 and 1.5. Suppose that GN acts transitively on X. (b) We choose two locally nilpotent derivations δ, δ0 ∈ LNDΛ(X) such that ker δ 6= ker δ0. Let U, U 0 denote the associated one-parameter subgroups and choose partial quotients : X → Q and 0 : X → Q0 as introduced in 1.15. (c) We can embed Q and Q0 into normal projective varieties ¯Q and ¯Q0, respectively. Let ¯X be a smooth projective completion of X. After blowing up ¯X in the boundary ∂X = ¯X\X, if necessary, we may extend and 0 to morphisms ¯X ¯0 ✲ ¯Q0 ¯ ❄ ¯Q The general fiber of is an orbit of U isomorphic to A1. Clearly ¯−1(q) ∼= P1 for a general point q ∈ Q. Hence there is a unique divisor D ⊆ ¯X\X which maps birationally onto ¯Q. Similarly there is a unique divisor D0 in ¯X\X mapping birationally onto ¯Q0. Thus both D and D0 are contained in the boundary ∂X = ¯X\X. The following observations will be important. Lemma 3.2. (1) Let ϕ ∈ ker δ\ ker δ0 be a regular function on X. Then ϕ is a rational function on ¯X with poles at general points of D0. (2) We have ¯(D0) ⊆ ¯Q\Q and ¯0(D) ⊆ ¯Q0\Q0 . In particular, D 6= D0. 16 H. FLENNER, S. KALIMAN, M. ZAIDENBERG Proof. (1) Since D0 → ¯Q0 is dominant, an orbit closure H0.x of a general point x ∈ X meets D0 at a general point ¯x ∈ D0. Let us consider ϕ as a rational map ¯X ✲ P1. Since the indeterminacy set of ϕ on ¯X is of codimension at least 2, ϕ is regular on the orbit closure H0.x ∼= P1 for a general x ∈ X. Since ϕ 6∈ ker δ0 this map is not constant on general orbits of H0. In particular it restricts to a dominant morphism ϕ : H0.x → P1 such that ϕ(¯x) = ∞. (2) It is sufficient to prove the first part. If ¯(D0) ∩ Q 6= ∅ then a function ϕ ∈ (cid:3) O(Q)\ ker δ0 would be holomorphic in a general point of D0 contradicting (1). Lemma 3.3. After blowing up the boundaries ∂X = ¯X\X and ∂Q = ¯Q\Q suitably we can achieve that (a) T = ¯(D0) is a divisor in ¯Q, and (b) ¯X, D and D0 are smooth. Proof. (a) By Lemma 3.2(2) T sits in the boundary of ¯Q. According to a theorem of Zariski, see [15] and [8, Theorem 1.3], there is a blowup ¯Q′ → ¯Q with a center in ¯(D0) such that the proper transform of D0 in ¯X ¯Q′ maps onto a divisor in ¯Q′. Thus replacing ¯Q by ¯Q′ we can achieve that T is a divisor. Since X is smooth and does not meet D ∪ D0, by a suitable blowup of the boundary (cid:3) ¯X\X we can achieve that (b) holds. Lemma 3.4. There is a closed subset B0 of ¯Q with codim ¯Q B0 ≥ 2 such that the following hold. (a) Sing ¯Q ∪ Sing T ⊆ B0. (b) D → ¯Q is an isomorphism in the points D\¯−1(B0). (c) ¯X → ¯Q is flat in the points over ¯Q\B0. Proof. (a) can be satisfied as ¯Q is normal and T is reduced. Since D → ¯Q is a birational map, also (b) can be achieved. (c) By the theorem on generic flatness [3, Theorem 14.4] there is a proper closed subset E in ¯Q such that ¯ is flat in the points over ¯Q\E. Applying the theorem on generic flatness again gives that the restricted map ¯E : ¯−1(E) → E is flat over a subset E\B′ of E, where B′ is a nowhere dense closed subset of E. Using Corollary 6.9 in [3] it follows that f is flat over the set ¯Q\B′′, where B′′ = B′ ∪ {s ∈ E : E is not a Cartier divisor in ¯Q at x} Since ¯Q is normal this set has codimension ≥ 2 in ¯Q. Adding B′′ to B0, also (c) is satisfied. (cid:3) The following facts should be well known; in lack of a reference we provide a brief argument. Lemma 3.5. Let p : S → Γ be a P1-fibration of a smooth surface S over a smooth affine curve Γ admitting a smooth section D ⊆ S so that D ∼= Γ. Then for any point t ∈ Γ the fiber F = p−1(t) over t is a tree of rational curves. Furthermore the following hold. GROMOV-WINKELMANN THEOREM 17 (a) If {x} = F ∩ D then h0(F, OF (x)) = 2 and H i(F, OF (x)) = 0 for i ≥ 1. (b) The sheaf OF (x) is generated by its global sections. (c) If s0, s1 ∈ H 0(F, OF (x)) is a basis, then the map (s0 : s1) : F → P1 is an isomor- phism near x. Proof. Blowing down successively (−1)-curves in the fibers of p not meeting D we obtain a locally trivial P1-bundle V → Γ. The curve D can as well be considered as a section of V → Γ and so we have an isomorphism V ∼= ProjΓ(p∗(OV (D))). If S = V then the assertions (a)-(c) are trivial. Blowing up subsequently points in the fibers these assertions also follow for p : S → Γ. (cid:3) In what follows we may assume that the conditions (a), (b) in Lemma 3.3 are satis- fied. Lemma 3.6. Letting ¯Xq = ¯−1(q) and Dq = D ∩ ¯Xq there is a closed subset B of codimension ≥ 2 in ¯Q such that for q ∈ ¯Q\B the following assertions hold. (a)q h0( ¯Xq, O ¯Xq (Dq)) = 2 and H i( ¯Xq, O ¯Xq (Dq)) = 0 for i ≥ 1. (b)q The sheaf O ¯Xq (Dq)) is generated by its global sections. (c)q If s0, s1 ∈ H 0( ¯Xq, O ¯Xq (Dq)) is a basis, then the map (s0 : s1)) : ¯Xq → P1 is an isomorphism near Dq. (d)q The map ¯∗(O ¯X (D))q → H 0( ¯Xq, O ¯Xq (Dq)) is surjective, and ¯∗(O ¯X(D))q is free of rank 2. Proof. Let B0 ⊆ ¯Q be a set as in Lemma 3.4. We choose a proper closed subset P of ¯Q such that any fiber over ¯Q\P is isomorphic to P1. For any q ∈ ¯Q\P the assertions (a)q-(d)q follow easily. Let a curve Γ in ¯Q be an intersection of n − 1 general ample divisors in ¯Q. Since ¯Q is normal and codim B0 ≥ 2, Γ meets neither Sing ¯Q nor B. By Bertini's theorem both Γ and the surface S = ¯−1(Γ) are smooth. The restriction ¯S : S → P1 is a P1-fibration. This P1-fibration admits a section, namely D ∩ S. The intersection D ∩ S is smooth in view of Bertini's theorem and Lemma 3.3(b). The fiber of S → Γ over q ∈ Γ ⊆ ¯Q coincides with ¯Xq. By Lemma 3.5 such a fiber ¯Xq is a tree of rational curves satisfying (a)q-(c)q. Since Γ meets every component, say, Pi of P of codimension 1 and does not meet B0, for some qi ∈ Pi\B0 the conditions (a)qi-(c)qi are satisfied. By semicontinuity (see[7, III, 12.8]) we obtain the inequalities hj( ¯Xp, O ¯Xp(Dp)) ≤ hj( ¯Xq, O ¯Xq (Dq)) ≤ hj( ¯Xqi, O ¯Xqi (Dqi)), j ≥ 0, where q ∈ Pi is a point near qi and p ∈ ¯Q\P is a point near q. Since the outer terms are equal, condition (a)q holds for q in some open dense subset P o i of Pi. i , which are satisfied for some q ∈ P o By Grauert's criterion (see [7, III, 12.9]) now also (d)q is satisfied. Since (b)q and i , they are satisfied (cid:3) (c)q are open conditions on P o generically on Pi. Now the lemma follows. Corollary 3.7. There is a proper closed subset B ⊆ ¯Q containing Sing T and Sing ¯Q with codimT (T ∩ B) ≥ 1 such that, letting X o = ¯X\¯−1(B) , Qo = ¯Q\B , T o = T \B and C = ¯−1(T ) , 18 H. FLENNER, S. KALIMAN, M. ZAIDENBERG there is a birational morphism (10) ϕ : X o −→ X = Qo × P1 compatible with the projection to Qo, which restricts to a biregular morphism (11) X o\C −→ X \C = (Qo\T ) × P1 , where C = T o × P1. Furthermore ϕ is biregular in a neighborhood of Do = D ∩ X o. Proof. Let B ⊆ ¯Q be the subset constructed in Lemma 3.6. Enlarging it in a suitable way we may assume that it contains Sing T ∪ Sing ¯Q. According to Lemma 3.6(c) the sheaf E = ¯∗(OX o(D)) is locally free of rank 2 on Qo. Thus enlarging B we may suppose that ¯∗(OX o(D)) is free. Choose two sections s0, s1 which form a basis of this bundle. They provide a morphism ϕ = (¯, (s0 : s1)) : X o → Qo × P1. Restricting to a fiber over q ∈ Qo, in view of Lemma 3.6(c)q this yields an isomorphism near Dq. Hence ϕ is an isomorphism near Do. Enlarging B further we may also assume that all fibers in Qo\T are isomorphic to P1. This implies that the restricted morphism (11) is an isomorphism. (cid:3) Notation 3.8. Consider the restriction of the locally nilpotent vector field δ to X o∩X. The associated action of U = exp(kδ) has no fixed points in this set and extends to an action on X o\C, where as before C = ¯−1(T ). The fibers of X o\C → Qo\T are preserved under U. Under the isomorphism X o\C ≃ X = (Qo\T )×P1 the second factor can be equipped with a homogeneous coordinate system (ζ1 : ζ2) such that the image, say, D of Do = D ∩ X o in X o is defined by the equation ζ1 = 0. We treat v = ζ1/ζ2 as a coordinate in the neighborhood X \{ζ2 = 0} of D in X . We fix a function f ∈ k[Q] such that its pullback on X belongs to ker δ\ ker δ0. This pullback induces rational functions on X o and on X denoted by the same symbol f . By Lemma 3.2(1) f has poles along D0 ∩ X o. By our choice of B in Corollary 3.7 T o is a submanifold of Qo. Thus locally the ideal of T o is generated by some function, say, u on Qo. On Qo the function f is of form a/us. Here s ≥ 1 is the pole order of f along T o, so a is a rational function on Qo, which is nonzero in the general point of T o. Later on we will replace f by a sufficiently large power f k. By this we can achieve that the pole order s is arbitrary large. Recall that Uf stands for the replica of U associated with the locally nilpotent vector field f δ. We note that Uf is well defined on the set X o\C ∼= (Qo\T o) × P1, cf. Corollary 3.7. Its element at moment τ ∈ k will be denoted by hf,τ . Considered as an automorphism of (Qo\T ) × P1 it preserves the first factor but not the second one. The action of hf,τ on v is described by the following Lemma. GROMOV-WINKELMANN THEOREM 19 Lemma 3.9. There exist a regular function d = d(f ) on Qo, which does not vanish at general points of T , and an integer l such that the automorphism of (Qo\T ) × P1 defined by hf,τ is given in the coordinates (q, v) by the formula hf,τ : (q, v) 7→ (cid:18)q , u(q)mv u(q)m + τ d(q)v(cid:19) , where m = s − l. In particular D ∩ C = {u = v = 0} is the set of indeterminacy points of hf,τ . Proof. In homogeneous coordinates (ζ1 : ζ2) the action of U = exp(kδ) on (Qo\T ) × P1 is of form (ζ1 : ζ2) → (ζ1, ζ2 + τ cζ1) where c is non-vanishing function on Qo\T . That is, c = c0ul where c0 is a non-vanishing function on Qo and l ∈ Z. Hence hf,τ is of form (ζ1 : ζ2) 7→ (ζ1 : ζ2 + τ d us−l ζ1), where d does not vanish at general points of T o. Note that m > 0 since f δ has a pole along D0. Passing to the affine coordinate v = ζ1/ζ2 this yields the desired conclusion. (cid:3) Letting s be the pole order of f along T we consider the set (12) Pf = {q ∈ T : locally f = a/us with a(q) = 0 or a 6∈ O ¯Q,q} , where u is as before (i.e. u = 0 is a local equation of T near q) and a is a rational function. This set is a proper closed subset of T . The next proposition is the main result of this section. Proposition 3.10. Given m and a function f ∈ k[Q] ∩ ker δ\ ker δ0 there exists a positive integer k0 such that any transformation is an m-contractions of C along D over the points of Qo\Pf . h ∈ Uf k, h 6= id, k ≥ k0, Proof. Let s, l be as in Notation 3.8 and Lemma 3.9. If we chose k0 in such a way that m′ = k0s − l ≥ m then by Lemma 2.13 the map h = hf k,τ is indeed an m-contraction for any τ 6= 0. (cid:3) Let now Y ⊆ X be a closed subset. Consider the partial boundary ∂0Y = ¯Y ∩ D0 . For U ∈ U(X) we let U ∗ = U\{id}. With this notation the following result holds. Proposition 3.11. Let the notation and conventions be as in Notation 3.1 and assume that (a), (b) in Lemma 3.3 are satisfied. Let (Yα,β)(α,β)∈A×B be a flat family of proper closed subsets of X. Suppose that there is a flat family (Eα)α∈A of proper, closed subset of D such that ∂Yα,β ∩ D ⊆ Eα for all (α, β) ∈ A × B. Given an invariant function f ∈ ker δ\ ker δ0, there is a dense open subset Ao of A and a flat family (E′ α)α∈Ao of proper closed subset of D0 satisfying ∂0h.Y(α,β) ⊆ E′ α ∀ (α, β) ∈ Ao × B, ∀ h ∈ U ∗ f k , ∀ k ≥ k0 . 20 H. FLENNER, S. KALIMAN, M. ZAIDENBERG Proof. According to Proposition 2.4 and Remark 2.7 the closure ¯Yαβ of Yαβ in ¯X is at most m-tangent to D along C for m ≫ 0 and for all (α, β) ∈ A × B simultaneously. Let Nαβ = C ∩ D ∩ Yαβ\C denote the defect set. By Proposition 3.10 for k ≫ 0 any map h ∈ U ∗ f k is an m-contraction of C along D over the points of Qo\Pf . Applying Proposition 2.15 the image h.Yαβ satisfies (13) h.Yαβ ∩ (Do ∪ C o) ⊆ D ∪ h(Nαβ) ∪ ¯−1(Pf ) where h(Nαβ) stands for the total transform of Nαβ under h. By our assumption the defect set Nαβ is contained in Nα = C ∩ Eα\C. Since our birational transformation h is compatible with the fibration ¯, the total image h(Nαβ) is contained in ¯−1(¯(Nα)). Taking in (13) the intersection with D0 gives ∂0(h.Yαβ) ⊆ E′ α = (D ∪ ¯−1(B ∪ ¯(Nα) ∪ Pf )) ∩ D0, where B = ¯Q\ ¯Qo is as in Corollary 3.7. Using the theorem on generic flatness it is easily seen that over an open dense subset Ao of A the sets E′ α form a flat family of closed subsets of D0. This yields the assertion. (cid:3) 4. Proof of the main theorem 4.1. Algebraic families of automorphisms. Following Ramanujam [12] let us in- troduce the following notion. Definition 4.1. Given irreducible algebraic varieties X and A and a map ϕ : A → Aut(X) we say that (A, ϕ) is an algebraic family of automorphisms on X if the induced map A × X → X, (α, x) 7→ ϕ(α).x, is a morphism. By abuse of notation, we do not distinguish in the sequel A and its image ϕ(A), and we identify α ∈ A with its image ϕ(α) in Aut(X). As in the case of group action, given a point x ∈ X the set A.x will be called the A-orbit of x, and the set Ax = {α ∈ A α(x) = x} the stabilizer of x in A. The stabilizer admits a natural linear representation dx : Ax → GL(TxX), α 7→ dαTxX, called the tangent representation. The following result allows to work with finite dimensional algebraic families instead of dealing with infinite dimensional groups of automorphisms. Lemma 4.2. Let X be a smooth quasi-affine variety and G = GN a group of auto- morphisms generated by a saturated set of locally nilpotent derivations so that G acts transitively on X. Then there exists an algebraic family of automorphisms A ⊆ G such that for any x ∈ X we have (a) A.x = X and (b) dx(Ax) = SL(TxX). Proof. According to Proposition 1.5 in [2] there exist one-parameter unipotent sub- groups H1, . . . , Hs of G such that with H = H1 · . . . · Hs ⊆ G we have H.x = G.x for any x ∈ X. In particular, (a) holds with the algebraic family A = H. By Theorem 4.2 [2] and its proof, for a fixed point x ∈ X the group SL(TxX) is 1 · . . . · H ′ r, r are suitable one-parameter subgroups of GN ,x. Taking the product equal to the image in dx(H ′) ⊆ GL(TxX) for an algebraic family H ′ = H ′ where H ′ 1, . . . , H ′ GROMOV-WINKELMANN THEOREM 21 A = HH ′H −1, where H is as in (a) and H −1 = Hs · . . . · H1, we thus achieve that both (a) and (b) are satisfied at every point x ∈ X. (cid:3) Notation 4.3. (a) As before we let X be a smooth quasi-affine variety and G = GN a group of automorphisms generated by a saturated set of locally nilpotent derivations as in Notation 3.1(a). We suppose that G acts transitively on X. According to Theorem 1.14 there are derivations δ0, δ1 ∈ N such that the group H = hhδ0, δ1ii ⊆ G generated by δ0, δ1 and their replicas acts with an open orbit on X.9 These locally nilpotent vector fields generate one-parameter unipotent subgroups U 0, U 1 ∈ U(G). Any function f ∈ ker δ0\ ker δ1 yields a replica U 0 f , and similarly g ∈ ker δ1\ ker δ0 yields a replica U 1 g . (b) To any sequence of invariant functions (14) F = {f1, . . . , fs, g1, . . . , gs}, where fi ∈ ker δ1\ ker δ0 and gi ∈ ker δ0\ ker δ1 , we associate an algebraic family of automorphisms A2s → Aut(X) defined by the product (15) More generally, given a tuple κ = (ki, li)i=1,...,s ∈ N2s the product ⊆ H Uκ = U F g1 ⊆ H . (16) U F = U 1 fs · U 0 gs · . . . · U 1 f1 · U 0 κ = U 1 f ks s · U 0 gls s · . . . · U 1 f · U 0 l1 g 1 k1 1 is as well an algebraic family of automorphisms. Corollary 4.4. There is a finite collection of invariant functions F as in (14) such that for any sequence κ = (ki, li)i=1,...,s ∈ N2s the algebraic family of automorphisms Uκ as in (16) has a dense open orbit in X. This orbit O(Uκ) coincides with O(H) and so does not depend on the choice of κ ∈ N2s. Proof. According to Proposition 1.5 in [2] there is a sequence F as in (14) such that H.x = U F .x ∀x ∈ X . In particular, for x ∈ O(H) the orbit U F .x = O(H) is open in X. It is easily seen that for any κ ∈ N2s we have O(Uκ) = O(U F ) = O(H). Indeed, O(H) consists of all the U F -flexible points in X. Now the assertions follow. (cid:3) 4.2. Proof of the main theorem. Notation 4.5. We keep the notation and assumptions from 4.3(a). (a) Let 0 : X → Q0 and 1 : X → Q1 be partial quotients with respect to the unipotent subgroups U 0 and U 1, respectively. Let us choose open embeddings X ֒→ ¯X, Q0 ֒→ ¯Q0, and Q1 ֒→ ¯Q1 into normal projective varieties, see Notation 3.1. We can assume that the following conditions are satisfied. 9In contrast to Notation 3.1(a) in this section the role of δ0 and δ1 will be symmetric so that it is convenient to replace the former δ by δ1. 22 H. FLENNER, S. KALIMAN, M. ZAIDENBERG (i) 0 and 1 extend to morphisms ¯0 : ¯X → ¯Q0 and ¯1 : ¯X → ¯Q1. Let D0 and D1 as in 3.1 be the unique horizontal divisors that map birationally onto ¯Q0 and ¯Q1, respectively. (ii) ¯X, D0 and D1 are smooth, see Lemma 3.3(b). (iii) T0 = ¯(D0) and T1 = ¯(D1 are divisors in ¯Q0 and ¯Q1, respectively; see Lemma 3.3(a). (b) Given a closed subscheme Y ⊆ X of codimension ≥ 2 we call ∂0Y = ¯Y ∩ D0 and ∂1Y = ¯Y ∩ D1 the partial boundaries. Furthermore OY will denote the open orbit of GN ,Y in X\Y . 4.6. In the course of the proof of the main Theorem we move the given pair (Y, x) to another one (Yα, xα) by means of an automorphism α ∈ GN , where Yα = α.Y and xα = α.x. In this way we can adopt the position of our pair with respect to the P1-fibration ¯0 : ¯X → ¯Q0 so that the conditions (i)-(iii) below hold. (i) U 0.xα ∩ OYα 6= ∅ ; (ii) U 0.xα ∩ Yα = ∅ ; (iii) ∂0(U 0.xα) /∈ ∂0(Yα). The following lemma allows to deduce Theorem 1.6 provided that (i)-(iii) hold for any x ∈ X\Y with some α ∈ G depending on x. Lemma 4.7. If for a point x ∈ X\Y and for some α ∈ G conditions (i)-(iii) in 4.6 are fulfilled then x ∈ OY . If these conditions are fulfilled for any x ∈ X\Y with some α ∈ G depending on x, then the conclusion of Theorem 1.6 holds. Proof. Since OY α = α.OY we have x ∈ OY ⇐⇒ xα ∈ OYα . Replacing (Y, x) by (Yα, xα) we will assume that (i)-(iii) hold for the pair (Y, x) and α = id. We need to show that then x ∈ OY . Conditions (ii) and (iii) yield that 0(x) ∈ 0(OY )\0(Y ) . Therefore there exists a regular function h ∈ O(Q0) such that h(0(x)) = 1 and h vanishes on 0(Y ). Replacing h by a suitable power of h we may suppose that the δ0-invariant function f = h ◦ 0 on X vanishes on Y . Thus the replica U 0 f = exp(kf δ0) of U 0 fixes Y pointwise i.e. U 0 f such that u.x ∈ OY . Hence also x ∈ OY , as stated. (cid:3) f ∈ U(GN ,Y ). By (i) one can find u ∈ U 0 Thus to prove Theorem 1.6 it is enough to show that (i)-(iii) hold for every point x ∈ X\Y with a suitable α ∈ G depending on x. Lemma 4.8. Given a point x ∈ X\Y and an algebraic family of automorphisms ϕ : A → Aut(X) the following hold. (a) The set of all α ∈ A satisfying (i) is open in A. (b) The set of all α ∈ A satisfying (ii) is constructible in A. GROMOV-WINKELMANN THEOREM 23 Proof. (a) The subset B ⊆ A where (i) does not hold is the set of α ∈ A satisfying U 0.xα ⊆ Yα or, equivalently, α−1U 0α.x ⊆ Y. Thus B = Tu∈U 0 Bu, where Bu = {α ∈ A : α−1uα.x ∈ Y } is the preimage of Y under the morphism A → X, α 7→ α−1uα.x. Hence B is closed in A. This proves (a). (b) Similarly, the subset C ⊆ A where (ii) does not hold is the set of α ∈ A with α−1U 0α ∩ Y 6= ∅. Consider the set C ′ = {(α, u) ∈ A × U 0 : α−1uα.x ∈ Y } . This set is closed in A × U 0 since it is the preimage of Y under the morphism A × U 0 → X, (α, u) 7→ α−1uα.x. Since C is the image of C ′ under the projection to A, (b) follows. (cid:3) The next proposition allows to verify conditions (i) and (ii). Proposition 4.9. Let as before x ∈ X\Y . (a) If A is an algebraic family of automorphisms of X with dx(Ax) ⊇ SL(TxX), then the set of all α ∈ A satisfying (i) is a dense open subset of A. (b) There exists an algebraic family A∗ ⊆ Gx transitive in X ∗ = X\{x} such that for any subgroup U 0 ∈ U(X) condition (ii) holds for a general α ∈ A∗. (c) Given an algebraic family B ⊆ Aut(X) we let A = B · A∗ ⊆ Aut(X), where A∗ ⊆ Gx is as in (b). Then (ii) holds for a general α ∈ A. Proof. (a) By Lemma 4.8 it suffices to find α ∈ A satisfying (i), or, equivalently, such that α−1U 0α.x ∩ OY 6= ∅. By our assumptions in (a) for any nonzero vector v ∈ TxX there is an element α ∈ Ax such that v is tangent to the orbit through x of the one- parameter group α−1U 0α ⊆ Aut(X). These orbits form an algebraic family of smooth rational curves in X through the point x that dominates X and so meets the open orbit OY , as required. (b) By the Transversality Theorem [2, 1.16] there exists an algebraic family A∗ ⊆ Gx transitive in X ∗ such that for any two subvarieties Y, Z ⊆ X there is a dense open subset A0 ⊆ A∗ with the property that for any α ∈ A0 the varieties α.Y and Z are transversal. Applying this to Z = U 0.x the varieties U 0.x and α.Y are disjoint, because under our assumptions dim U 0.x + dim Y < dim X . Since xα = x, (b) follows. To deduce (c) we note that the set, say C of points α ∈ A, where (ii) fails is the set of α = (β, α) with α−1β−1U 0βα.x ∩ Y 6= ∅. Consider similarly as in the proof of Lemma 4.8(b) the closed subset of B × A∗ × U 0 C ′ = {(β, α, u) ∈ B × A∗ × U 0 : α−1β−1uβα.x ∈ Y } , where A∗ satisfies the conclusion of (b). According to (b) for any β ∈ B the set C ′ β = C ′ ∩ ({β} × A∗ × U 0) 24 H. FLENNER, S. KALIMAN, M. ZAIDENBERG maps under the projection to A∗ to a nowhere dense subset. Hence also the image C of C ′ under the projection to A = B × A∗ will be nowhere dense. Thus its complement contains an open dense subset proving (c). (cid:3) Notation 4.10. Given a one-parameter group U ∈ U(X) we let as before U ∗ = U\{id}. Given a collection F of invariant functions and Uκ = U 1 f ks s f1, . . . , fs ∈ ker δ1\ ker δ0 · U 0 gls s · U 0 l1 g 1 · . . . · U 1 f U ∗ k1 1 and g1, . . . , gs ∈ ker δ0\ ker δ1 as in (15), we let κ = U 1∗ f ks s · U 0∗ gls s · . . . · U 1∗ k1 f 1 · U 0∗ l1 g 1 . Using Proposition 3.11 we can deduce the following result. Proposition 4.11. Let (Yα)α∈A be a flat family of proper closed subsets of X. Assume that the partial boundaries ∂iYα (see Notation 4.5) are contained in Eα,i, where the (Eα,i)α∈A, i = 0, 1, form flat families of proper closed subsets of Di. Then one can find an open dense subset Ao of A, flat families of proper, closed subsets (Eo α,i)α∈Ao of Di (i = 0, 1), and a sequence κ = (k1, l1, . . . , ks, ls) ∈ N2s such that for any element h ∈ U ∗ κ we have ∂i(h.Yα) ⊆ Eo α,i , i = 0, 1 , ∀ α ∈ Ao . Proof. The proof proceeds by induction on s. For s = 0 the assertion clearly holds with Ao = A and Eα,i = ∂iYα, i = 0, 1. Assume that it holds at step s − 1, i.e. we can find κ′ = (kj, lj)j=1,...,s−1 ∈ N2s−2, a dense open subset A′ ⊆ A and flat families of proper closed subsets (Eα,i)α∈A′ of Di such that for α ∈ A′ ∂i(h.Yα) ⊆ Eα,i , i = 0, 1, ∀ h ∈ U ∗ κ′ . The varieties (h.Yα)(h,α)∈U ∗ can find an open dense subset A′′ ⊆ A′ and flat families (E′ closed subsets of Di such that κ′ ×A′ form a flat algebraic family. By Proposition 3.11 one α,i)α∈A′′, i = 0, 1, of proper ∂i(h′h.Yα) ⊆ E′ α,i (i = 0, 1) ∀ ls ≫ 0, ∀α ∈ A′′, ∀ (h′, h) ∈ U 0∗ gls s × U ∗ κ′ . Fixing a sufficiently large ls and applying the same argument again one can find an open dense subset Ao ⊆ A′′ and flat families (Eo α,i)α∈Ao, i = 0, 1, of proper closed subsets of Di such that ∂i(h′′h′h.Y ) ⊆ Eo α,i (i = 0, 1) ∀k1 ≫ 0, ∀α ∈ Ao , ∀ (h′′, h′, h) ∈ U 1∗ f ks s × U 0∗ gls s × U ∗ κ′ . This concludes the induction. (cid:3) Using Proposition 4.11 and Corollary 4.4 we can now deduce Theorem 1.6. Proof of Theorem 1.6. Let x ∈ X\Y be a fixed point. We show that for a suitable choice of an algebraic family A of automorphisms conditions (i)-(iii) are satisfied for the pair (Yα, xα), if α ∈ A is generic. Then our theorem follows by applying Lemma 4.6. Step 1. Consider an algebraic family A ⊆ G satisfying conditions (a) and (b) of Lemma 4.2. Applying Proposition 4.9(a) condition (i) holds when α varies in a dense GROMOV-WINKELMANN THEOREM 25 open subset of A. Replacing the original pair (Y, x) by a suitable new one (Yα, xα) = (α.Y, α.x) we may suppose that (Y, x) satisfies (i). Step 2. In the following we construct an algebraic family B of automorphisms such that for a generic choice of β ∈ B the translates (Yβ, xβ) satisfy (ii), (iii). Since by Proposition 4.9(a) condition (i) is open then the pair (Yβ, xβ) also satisfies (i). Let A∗ be a family of automorphisms as in Proposition 4.9(b). The translates Yα = α.Y , α ∈ A∗, form a flat family of proper closed subsets of X. Using the theorem of generic flatness it is easily seen that over an open dense subset A′ ⊆ A∗ also the partial boundaries Eα,i = ∂iYα, α ∈ A′, form flat families of proper closed subsets of Di, i = 0, 1. Let now F , Uκ, and U ∗ κ be as in Notation 4.10. By Proposition 4.11 we can find κ = (k1, l1, . . . , ks, ls) ∈ N2s, a dense open subset Ao ⊆ A′, and families (Eo α,i)α∈Ao, i = 0, 1, of proper closed subsets of Di such that ∂i(h.Yα) ⊆ Eo α,i for i = 0, 1, α ∈ Ao and all h ∈ U ∗ κ. We claim that for a generic choice of (h, α) ∈ B = U ∗ κ ×A∗ conditions (ii) and (iii) are satisfied for h.Yα. To check (ii) we note that h.Yα = hα.Y . Thus applying Proposition 4.9(c) to the family B = U ∗ κ × A∗ condition (ii) is indeed satisfied for a generic choice of (h, α). It remains to show that (iii) is satisfied for a generic choice of (h, α). Condition (iii) is equivalent to ¯0(h.xα) 6∈ ∂0(h.Yα). By construction ∂0(h.Yα) ⊆ Eα,0 ⊆ D0 for any h ∈ U ∗ κ, fill in a dense subset of X, and so their images ¯0(h.x) fill in a dense subset of Q0 ⊆ ¯Q0\¯0(D0). Thus (iii) holds for a generic choice of (h, α) ∈ U ∗ κ × Ao. This concludes the proof of Theorem 1.6. (cid:3) κ, while for a fixed α ∈ Ao the points h.xα, h ∈ U ∗ References [1] I.V. Arzhantsev, K. Kuyumzhiyan, M. Zaidenberg: Flag varieties, toric varieties, and sus- pensions: three instances of infinite transitivity. Sb. Math. 203:7-8 (2012), 923 -- 949. [2] I. V. Arzhantsev, H. Flenner, S. Kaliman, F. Kutzschebauch, M. Zaidenberg: Flexible varieties and automorphism groups. Duke Math. J. 162:4 (2013), 767 -- 823. [3] D. Eisenbud: Commutative algebra. With a view toward algebraic geometry. Graduate Texts in Mathematics, 150. Springer-Verlag, New York, 1995. xvi+785 pp. [4] G. Freudenburg: Algebraic theory of locally nilpotent derivations. Encyclopaedia of Mathe- matical Sciences, 136. Invariant Theory and Algebraic Transformation Groups, VII. Springer- Verlag, Berlin, 2006. [5] M. Gromov: Partial differential relations. Ergebnisse der Mathematik und ihrer Grenzgebiete (3) 9. Springer-Verlag, Berlin, 1986. [6] M. Gromov: Oka's principle for holomorphic sections of elliptic bundles. J. Amer. Math. Soc. 2 (1989), 851 -- 897. [7] R. Hartshorne: Algebraic Geometry. Springer-Verlag, New York-Heidelberg, 1977. [8] J. Koll´ar: Rational curves on algebraic varieties. Ergebnisse der Mathematik und ihrer Gren- zgebiete. 3. Folge, 32. Springer-Verlag, Berlin, 1996. [9] S. Kovalenko: Transitivity of automorphism groups of Gizatullin surfaces. arXiv:1304.7116, 32p. [10] D. Mumford, J. Fogarty, F. Kirwan: Geometric invariant theory. Ergebnisse der Mathematik und ihrer Grenzgebiete (2) 34. Springer-Verlag, Berlin, 1994. [11] V. L. Popov, E. B. Vinberg: Invariant Theory. In: Algebraic geometry IV, A. N. Parshin, I. R. Shafarevich (eds.), Berlin, Heidelberg, New York: Springer-Verlag, 1994. 26 H. FLENNER, S. KALIMAN, M. ZAIDENBERG [12] C. P. Ramanujam: A note on automorphism groups of algebraic varieties. Math. Ann. 156 (1964), 25 -- 33. [13] J. Winkelmann: On automorphisms of complements of analytic subsets in Cn. Math. Z. 204 (1990), 117 -- 127. [14] J. Winkelmann: Invariant rings and quasiaffine quotients. Math. Z. 244 (2003), no. 1, 163 -174. [15] O. Zariski: The reduction of the singularities of an algebraic surface. Ann. of Math. 40 (1939), 639-689. Fakultat fur Mathematik, Ruhr Universitat Bochum, Geb. NA 2/72, Universitats- str. 150, 44780 Bochum, Germany E-mail address: [email protected] Department of Mathematics, University of Miami, Coral Gables, FL 33124, USA E-mail address: [email protected] Universit´e Grenoble I, Institut Fourier, UMR 5582 CNRS-UJF, BP 74, 38402 Saint Martin d'H`eres c´edex, France E-mail address: [email protected]
1209.3725
4
1209
2013-11-19T12:03:13
Bernstein-Sato ideals and local systems
[ "math.AG" ]
The topology of smooth quasi-projective complex varieties is very restrictive. One aspect of this statement is the fact that natural strata of local systems, called cohomology support loci, have a rigid structure: they consist of torsion-translated subtori in a complex torus. We propose and partially confirm a relation between Bernstein-Sato ideals and local systems. This relation gives yet a different point of view on the nature of the structure of cohomology support loci of local systems. The main result is a partial generalization to the case of a collection of polynomials of the theorem of Malgrange and Kashiwara which states that the Bernstein-Sato polynomial of a hypersurface recovers the monodromy eigenvalues of the Milnor fibers of the hypersurface. We also address a multi-variable version of the Monodromy Conjecture, prove that it follows from the usual single-variable Monodromy Conjecture, and prove it in the case of hyperplane arrangements.
math.AG
math
BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS NERO BUDUR Abstract. The topology of smooth quasi-projective complex varieties is very restrictive. One aspect of this statement is the fact that natural strata of local systems, called cohomology support loci, have a rigid structure: they consist of torsion-translated subtori in a complex torus. We propose and partially confirm a relation between Bernstein-Sato ideals and local systems. This relation gives yet a different point of view on the nature of the structure of cohomology support loci of local systems. The main result is a partial generalization to the case of a collection of polynomials of the theorem of Malgrange and Kashiwara which states that the Bernstein-Sato polynomial of a hypersurface recovers the mon- odromy eigenvalues of the Milnor fibers of the hypersurface. We also address a multi-variable version of the Monodromy Conjecture, prove that it follows from the usual single-variable Monodromy Conjecture, and prove it in the case of hyperplane arrangements. Contents Introduction 1. 2. Cohomology support loci 3. Sabbah specialization and local systems 4. Bernstein-Sato ideals 5. D-modules 6. Hyperplane arrangements 7. Examples References 1 10 11 23 32 36 41 42 1. Introduction 1.1. Bernstein-Sato ideals and local systems. We first propose a conjectural picture relating Bernstein-Sato ideals with local systems. It is known that the topology of smooth quasi-projective complex varieties is very restrictive. One as- pect of this statement is the fact that natural strata of local systems, called coho- mology jump loci, have a rigid structure: they consist of torsion-translated subtori in a complex torus, see Budur-Wang [11]. The structure of cohomology jump loci in various setups is the main theme in previous works such as Green-Lazarsfeld 2010 Mathematics Subject Classification. 14F10, 32S40, 14B05, 32S05, 32S22. Key words and phrases. Bernstein-Sato ideal, Bernstein-Sato polynomial, b-function, D- local systems, cohomology jump loci, characteristic variety, Sabbah specialization, modules, Alexander module, Milnor fiber, Monodromy Conjecture, hyperplane arrangements. 1 2 NERO BUDUR [19, 20], Arapura [1, 2], Simpson [42], Budur [9], Libgober [27], Dimca-Papadima- Suciu [17], Popa-Schnell [37], Dimca-Papadima [16]. The union of the cohomology jump loci forms the cohomology support locus. The conjectural picture we pro- pose gives yet a different point of view on the nature of the structure of cohomology support loci of local systems. To be more precise, let F = (f1, . . . , fr) be a collection of non-zero polynomials fj in C[x1, . . . , xn]. The Bernstein-Sato ideal of F is the ideal BF generated by polynomials b ∈ C[s1, . . . , sr] such that b(s1, . . . , sr)f s1 1 · · · f sr r = P f s1+1 · · · f sr+1 r 1 for some algebraic differential operator P ∈ C(cid:20)x1, . . . , xn, ∂ ∂x1 , . . . , ∂ ∂xn , s1, . . . , sr(cid:21) . The existence of non-zero Bernstein-Sato ideals BF has been proved by Sabbah [39], see also Bahloul [3] and Gyoja [21]. In the one-variable case r = 1, the monic generator of the ideal BF is the classical Bernstein-Sato polynomial. In general, the ideal BF and its radical are not always principal, for such examples see Bahloul- Oaku [4, §4.1]. The ideal BF is generated by polynomials with coefficients in the subfield of C generated by the coefficients of F [4, §4]. Conjecture 1. The Bernstein-Sato ideal BF is generated by products of linear polynomials of the form with αj ∈ Q≥0, and α ∈ Q>0. α1s1 + . . . + αrsr + α This would imply the same for the radical ideal of BF . The conjecture would refine a result of Sabbah [39] and Gyoja [21] which states that BF contains at least one element of this type. In the one-variable case r = 1, Conjecture 1 is due to Kashiwara [22]. When n = 2, every element of BF is divisible by the linear polynomials defining (r −1)-dimensional faces of the jumping polytopes of the local mixed multiplier ideals of f1, . . . , fr, by Cassou-Nogu`es and Libgober [12, Theorem 4.1]. We have originally arrived to conjecture that Bernstein-Sato ideals have this particular shape from computing examples with the library dmod.lib in Singular [13, 24]. This paper evolved out of the effort to understand this behavior. The interpretation of this behavior in terms of cohomology support loci of local systems which we give in this paper is new to our knowledge. Apart from the positivity statement, Conjecture 1 can be seen as the conse- quence of the following situation, similar to ones occurring frequently in arithmetic geometry and model theory. Let Exp : Cr −→ (C∗)r be the map x 7→ exp(2πix). We conjecture that Exp of the zero locus V (BF ) of the Bernstein-Sato ideal BF satisfies the conditions of the following result of M. Laurent: BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 3 Theorem 1. [25] Let Z be a Zariski closed subset of (C∗)r defined over Q with a Zariski dense subset of torsion points. Then Z is a finite union of torsion translates of complex subtori. Since in all computed examples Exp (V (BF )) satisfies the assumptions of Theo- rem 1, the interesting question is then: where do the torsion translates of subtori come from? Next, we will formulate a conjecture answering this question. We will then prove this conjecture in one direction and almost prove it in the other direction as well. 1 . . . f mr The idea is that one can produce lots of torsion points by restricting to one- parameter subgroups using the classical result of Malgrange and Kashiwara for the hypersurfaces f m1 . Then the torsion-translated subtori are obtained by interpolating over all one-parameter restrictions the Milnor monodromies, or equivalently the nearby cycles complexes, of these hypersurfaces. The interpola- tion is achieved by Sabbah's specialization complex attached to F = (f1, . . . , fr). We show that the support of the Sabbah's specialization complex is related with cohomology support loci of rank one local systems. r Let us give more details now. It is important for the rest of the paper to work locally at a point x in X := Cn. In this case, we replace in all the above BF by the local Bernstein-Sato ideal BF,x of the germ of F at x and we also propose the local version of Conjecture 1. It is known that see [4, Corollary 6]. Thus, letting V (I) denote the zero locus of an ideal I, BF = \x∈X V (BF ) = [x∈X BF,x, V (BF,x), so the local version implies the global version of Conjecture 1 for the radical ideals. Moreover, this is a finite union since there is a constructible stratification of X such that for x running over a given stratum the Bernstein-Sato ideal at x is constant [7]. The relation with local systems is in two steps. First, we propose a generalization of the well-known result of Kashiwara [23] and Malgrange [30] which states that the roots of the classical Bernstein-Sato polynomial of a polynomial germ f give the monodromy eigenvalues on the Milnor fiber. In this case, the cohomology of the Milnor fiber is packaged into Deligne's nearby cycles complex ψf CX . When r ≥ 1, a generalization of Deligne's nearby cycles functor is the Sabbah specialization functor ψF : Db c(X, C) → Db c(D, A), where D := r[j=1 V (fj) 4 NERO BUDUR is the union of the zero loci of the fj, A := C[t1, t−1 1 , . . . , tr, t−1 r ], and Db c(., R) is the bounded derived category of constructible sheaves in the analytic topology over a ring R. This functor has been introduced in [40]. The action of A on ψF CX generalizes the monodromy of the Milnor fiber from the case r = 1. For a point x in D, denote by Supp x(ψF CX) the support of ψF CX at x as an A-module, see Definition 3.6. The ambient space of the support is the torus (C∗)r with affine coordinate ring A. For our purposes, we have to take into account the possibility that some fj do not vanish at x, and thus we are lead to define the uniform support Supp unif x (ψF CX) ⊂ (C∗)r, see Definition 3.20. See [32, 3.8.1] for the next result in the l-adic setting: Theorem 2. Supp unif (C∗)r. x (ψF CX) is a finite union of torsion translated subtori of The following would generalize the classical result of Kashiwara and Malgrange. Conjecture 2. Exp (V (BF,x)) = [y∈D near x Supp unif y (ψF CX ). The union is taken over points y ∈ D in a small ball around x. However, one can take only the general points y of a fine enough stratification of the singular locus of D. Conjecture 2 almost implies Conjecture 1 for codimension part of the radical of the Bernstein-Sato ideal: Proposition 1. Assume Conjecture 2. Let Z be an irreducible component of V (BF,x). If Z has codimension 1, then Z is the zero locus of a linear polynomial of the form with αj ∈ Q≥0 and α ∈ Q>0. α1s1 + . . . + αrsr + α The case of hyperplane arrangements, where the support of the Sabbah spe- cialization complex is a combinatorial invariant, provides a checking ground and evidence for Conjecture 2, see Corollary 2 and the Remark thereafter. We show the following partial confirmation of Conjecture 2. Theorem 3. Exp (V (BF,x)) ⊃ [y∈D near x Supp unif y (ψF CX ). We also make a significant step toward proving the converse of Theorem 3. Let DX be the sheaf of holomorphic differential operators on X. BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 5 Proposition 2. The converse of Theorem 3, and so the Conjecture 2, holds for any F if the following holds for any F such that the fj with fj(x) = 0 define mutually distinct reduced and irreducible hypersurface germs at x: locally at x, for all α ∈ V (BF,x), (sj − αj)DX [s1, . . . , sr]f s1 1 . . . f sr r 6≡ DX [s1, . . . , sr]f s1 1 . . . f sr r rXj=1 modulo DX[s1, . . . , sr]f s1+1 1 . . . f sr+1 r . The relation of Bernstein-Sato ideals with local systems is achieved by relating the latter with the Sabbah specialization complex. For a connected finite CW- complex M, let L(M) denote the space of complex local systems of rank one on M. Then L(M) = Hom(H1(M, Z), C∗). Define the cohomology support locus (also called the characteristic variety) of M to be the subset V(M) of L(M) consisting of local systems with non-trivial coho- mology, (1) V(M) := {L ∈ L(M) H k(M, L) 6= 0 for some k}. There are more refined cohomology jump loci of M which can be defined, but we will not be concerned with them in this article. It is known that V(M) is a Zariski closed subset of L(M) defined over Q. For a point x in X, let UF,x be the complement of D in a small open ball centered at x, UF,x := Ballx r (Ballx ∩ D). There is a natural embedding of L(UF,x) into the torus (C∗)r induced by F . Theorem 4. If the polynomials fj with fj(x) = 0 define mutually distinct reduced and irreducible hypersurface germs at x, then Supp x(ψF CX) = V(UF,x). There is absolutely no difficulty to understand the relation between Supp x(ψF CX ) and cohomology support loci if the assumptions are dropped, cf. 3.35. We can define the uniform cohomology support locus with respect to F at x, (ψF CX) via which we denote by V unif (UF,x), such that it agrees with Supp unif Theorem 4, see Definition 3.21. Hence: x Theorem 5. If the polynomials fj with fj(x) = 0 define mutually distinct reduced and irreducible hypersurface germs at x, then Exp (V (BF,x)) ⊃ [y∈D near x V unif (UF,y). Assuming Conjecture 2, equality holds. 6 NERO BUDUR Again, there is absolutely no difficulty to understand what happens if the assump- tions are dropped. Let us mention the connection with local Alexander modules. The cohomologies of the stalks of ψF CX are the multi-variable local homology Alexander modules, as shown by Sabbah [40], see Proposition 3.12. In the special case when all the polynomials fj are homogeneous, the cohomologies of the stalk at the origin of ψF CX are the multi-variable universal homology Alexander modules introduced by Dimca-Maxim [15], see Proposition 3.24. 1.2. The geometry of Bernstein-Sato ideals. Next, information about uniform supports and cohomology support loci leads to better understanding of the question of what do zero loci of Bernstein-Sato ideals look like. In the case when all fj are homogeneous polynomials, we give a formula which reduces the computation of uniforms supports to a lower-dimensional, but possibly non-homogeneous case, see Proposition 3.27. Hence, conjecturally, the same holds for Exp (V (BF )). In particular, we obtain: Corollary 1. Let F = (f1, . . . , fr) with 0 6= fj ∈ C[x1, . . . , xn] irreducible and ho- mogeneous of degree dj defining mutually distinct hypersurfaces with gcd(d1, . . . , dr) = 1. Let V be the complement in Pn−1 of the union of the zero loci of fj. If χ(V ) 6= 0, then {d1s1 + . . . + drsr + k = 0} ⊂ V (BF ) for some k ∈ Z. It is tempting to conjecture that k = n in Corollary 1. We do so below for hyperplane arrangements. In the case of hyperplane arrangements, the homogenous reduction formula can be applied repeatedly to obtain precise combinatorial formulas for the uniform supports of the Sabbah specialization complex, see Proposition 6.9. Let F = (f1, . . . , fr) be such that fj are non-zero linear forms defining mutually distinct hyperplanes. The following terminology is defined in Section 6. For an edge W j=1 fj, let FW be the restriction in the sense of of the hyperplane arrangement Qr hyperplane arrangements of F to W . Let be a total splitting of FW . If we set F (i) FW = F (i) W lWYi=1 W = (f (i) 1,W , . . . , f (i) d(i) j,W := deg f (i) j,W . r,W ), let Corollary 2. Let F = (f1, . . . , fr) with fj ∈ C[x1, . . . , xn] non-zero linear forms defining mutually distinct hyperplanes. Then d(i) 1,W V (ht 1 d(i) r,W r − 1 i = 1, . . . , lW i) ⊂ Exp (V (BF )), . . . t where the union is over the edges W of the hyperplane arrangementQr suming Conjecture 2, equality holds. j=1 fj. As- [W BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 7 Remark. This corollary provides support for Conjecture 2 in the sense that the conjectured equality in Corollary 2 can be checked for many particular examples, see Section 7. Note that the left-hand side is completely combinatorial. The conditions in the Corollary can be relaxed, see Remark 6.10, however we opted to keep only an esthetically cleaner statement. By specializing F = (f1, . . . , fr) toQr and fW =QlW j=1 fr in the above Corollary we obtain the following. Let f be a hyperplane arrangement, fW the restriction to the edge W , W . Denote by bf the classical one-variable Bernstein-Sato polynomial of f , and by Mf,x the Milnor fiber of f at x. With this notation: W a total splitting of fW . Let d(i) W = deg f (i) i=1 f (i) Corollary 3. Let f ∈ C[x1, . . . , xn] be a hyperplane arrangement. Then Exp (V (bf )), which equals the set of all eigenvalues of the monodromy on H •(Mf,x, C) for x rang- ing over f −1(0), is a combinatorial invariant. If f is reduced, this is the set V (htd(i) W − 1 i = 1, . . . , lW i), [W where the union is over the edges W of f . In contrast, U. Walther has announced that the Bernstein-Sato polynomial bf of a hyperplane arrangement is not a combinatorial invariant. A different proof of Corollary 3 involving [26, Theorem 3.1] was noticed and communicated to us by A. Libgober. The following is a multi-variable generalization of [10, Conjecture 1.2]. This statement has implications for the Multi-Variable Strong Monodromy Conjecture, see Theorem 8 below. Conjecture 3. Let F = (f1, . . . , fr), where fj are central hyperplane arrangements j=1 fj is a central essential in Cn, not necessarily reduced, of degree dj, and Qr indecomposable hyperplane arrangement. Then {d1s1 + . . . + drsr + n = 0} ⊂ V (BF ). 1.3. Multi-Variable Monodromy Conjecture. We discuss the relation between multi-variable topological zeta functions on one hand, and Sabbah specialization complexes and Bernstein-Sato ideals, on other hand. Let F = (f1, . . . , fr) with 0 6= fj ∈ C[x1, . . . , xn]. We keep the notation from 1.1. Let µ : Y → X be a log resolution ofQj fj. Let Ei for i ∈ S be the collection of irreducible components of the zero locus of (Qj fj) ◦ µ. Let ai,j be the order of vanishing of fj along Ei, and let ki be the order of vanishing of the determinant of the Jacobian of µ along Ei. For I ⊂ S, let E◦ I = ∩i∈IEi r ∪i∈SrIEi. With this notation, the topological zeta function of F = (f1, . . . , fr) is Z top F (s1, . . . , sr) :=XI⊆S χ(E◦ I ) ·Yi∈I 1 ai,1s1 + . . . ai,rsr + ki + 1 . This rational function is independent of the choice of log resolution. Define P L(Z top F (s1, . . . , sr)) 8 NERO BUDUR to be the polar locus in Cr. The following is the Topological Multi-Variable Monodromy Conjecture, slightly different than phrased by Loeser, see [32, 29]: Conjecture 4. Exp (P L(Z top Supp unif x (ψF CX). F )) ⊂ [x∈D When r = 1, this is the Topological Monodromy Conjecture of Igusa-Denef- Loeser saying that poles of the topological zeta function give eigenvalues of the Milnor monodromy. In fact, in response to a question of V. Shende, the general case follows from the r = 1 case: Theorem 6. Let C be a class of non-zero polynomials stable under multiplication. If the Monodromy Conjecture holds for polynomials in C, then the Multi-Variable Monodromy Conjecture holds for maps F = (f1, . . . , fr) with fj in C. Examples of classes of polynomials stable under multiplication for which the Monodromy Conjecture is known include plane curves [28] and hyperplane arrange- ments [10]. Thus Theorem 6 reproves the Multi-Variable Monodromy Conjecture for plane curves due to Nicaise [32], and proves it for hyperplane arrangements. One can ask how natural is to specialize the Monodromy Conjecture. We de- fine later what it means to specialize F to another collection G of possibly fewer polynomials, see Definition 3.30. For example, F = (f1, . . . , fr) specializes to j=1 fj. In the first example, the spe- cialization loses in some sense fr, where as in the second example none of the fj are lost. We call the second type a non-degenerate specialization, see Definition 3.30. We show the following naturality with respect to non-degenerate specializations of the Monodromy Conjecture: G = (f1, . . . , fr−1), and it also specializes toQr Theorem 7. Assume that Conjecture 4 holds for a given F . degenerate specialization of F , then Conjecture 4 also holds for G. If G is a non- Up to now, there has been no multi-variable version of the Strong Monodromy Conjecture since it was not clear which ideal of Bernstein-Sato type was the right candidate, see 4.1. Since a strong version should imply the weaker version, the search for the right candidate is related with the search for the multi-variable generalization of the Malgrange-Kashiwara result. Thanks to V. Levandovskyy, we were able access and experiment with implemented algorithms for computing various types of Bernstein-Sato ideals. Based on these computations and based on the other supporting evidence for Conjecture 2 put forth in this paper, we make the following Topological Multi-Variable Strong Monodromy Conjecture: Conjecture 5. P L(Z top F ) ⊂ V (BF ). Conjecture 5 implies Conjecture 4 if we believe Conjecture 2, hence the adjective "strong". At the moment we cannot conclude that the r = 1 case for the Strong Monodromy Conjecture implies the r ≥ 1 case, nor that the Strong Monodromy BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 9 Conjecture is compatible with non-degenerate specializations, but see Remark 4.31. For hyperplane arrangements we reduce Conjecture 5 to Conjecture 3, a result which was proved for r = 1 in [10]: Theorem 8. If each fj defines a (possibly-nonreduced) hyperplane arrangement in Cn and if Conjecture 3 holds for the restriction FW = (fj,W fj(W ) = 0, j ∈ {1, . . . , r}) of the hyperplane arrangements to any dense edge W of Qr 5 holds for F . j=1 fj, then Conjecture On a different note, there has been recent interest in zeta functions attached to differential forms and possible connections with monodromy-type invariants, see N´emethi-Veys [31]. Let dx = dx1 ∧ . . . ∧ dxn and let ω be an n-form on X. Define Z top, ω F (s1, . . . , sr) in a similar fashion as Z top F (s1, . . . , sr), but with ki replaced by ordEiω. Note that Z top, dx F (s1, . . . , sr) = Z top F (s1, . . . , sr). One can ask what would a Monodromy Conjecture predict for Z top, ω ? See [31, 1.2] for a discussion. We propose an answer which is very natural and which says that the Monodromy Conjecture for Forms is a special case of the Multi-Variable Monodromy Conjecture. Clearly F Z top, frdx (f1,...,fr−1)(s1, . . . , sr−1) = Z top F (s1, . . . , sr−1, 1). Hence, the Topological Multi-Variable Monodromy Conjecture for Forms should be: and the Topological Multi-Variable Strong Monodromy Conjecture for Forms should be: P L(Z top, frdx (f1,...,fr−1)) ⊂ V (BF ) ∩ V (sr − 1). The Topological Multi-Variable Monodromy Conjecture for Forms is thus equiva- lent with the usual single-variable Topological Monodromy Conjecture, by Theorem 6. It is a standard procedure to adjust statements involving topological zeta func- tions to obtain statements involving: local topological zeta functions, (local) p-adic zeta functions, and, more generally, (local) motivic zeta functions. For brevity, we shall skip this discussion. 1.4. Applications. One of the main applications of the theory of D-modules is that it leads to algorithms which can be implemented to compute topological in- variants. For example, the classical result of Malgrange and Kashiwara led to al- gorithms for computing Milnor monodromy eigenvalues via the classical Bernstein- Sato polynomial, the first such algorithm being due to Oaku [33]. Similarly, Con- jecture 2 would provide already-implemented algorithms to compute cohomology Exp (P L(Z top, frdx Supp unif x (ψF CX ) ∩ V (tr − 1), (f1,...,fr−1))) ⊂ [x∈D 10 NERO BUDUR support loci of hypersurface germs complements. There are no other known algo- rithms for cohomology support loci applicable in general. Note that Bernstein-Sato ideals are essential for the current algorithms computing cohomology of local sys- tems on complements of projective hypersurfaces, see Oaku-Takayama [34]. 1.5. Acknowledgement. We would like to thank V. Levandovskyy for help with computing examples and for corrections. For computations of the Bernstein-Sato ideals in this paper we used the library dmod.lib in Singular [13, 24]. We would also like to thank F. J. Castro-Jim´enez, A. Dimca, A. Libgober, L. Maxim, M. Schulze, V. Shende, W. Veys, U. Walther, and Y. Yoon for helpful discussions, and the University of Nice for hospitality during writing part of the article. Special thanks are due to B. Wang who helped correct many mistakes in the original version. This work was partially supported by the National Security Agency grant H98230-11-1-0169 and by the Simons Foundation grant 245850. 1.6. Notation. All algebraic varieties are assumed to be over the complex number field. A variety is not assumed to be irreducible. The notation (f1, . . . , fr) stands for a tuple, while hf1, . . . , fri will mean the ideal generated by elements fj of some ring. Loops and monodromy around divisors are meant counterclockwise, i.e. going once around {x = 0} in a small loop sends xα to e2πiαxα for any α ∈ C. 2. Cohomology support loci 2.1. Local systems and cohomology support loci. Let M be a connected finite CW-complex of dimension n. Let L(M) be the group of rank one complex local systems on M. We can identify L(M) = Hom(π1(M), C∗) = Hom(H1(M, Z), C∗) = H 1(M, C∗). Consider the ring B := C[H1(M, Z)]. Then L(M) is an affine variety with affine coordinate ring equal to B. Example 2.2. If U denotes the complement in a small open ball centered at a point x in Cn of r mutually distinct analytically irreducible hypersurface germs, then H1(U, Z) = Zr is generated by the classes of small loops around the branches, and L(U) = (C∗)r, see [14, (4.1.5)]. By Libgober [27], the cohomology support locus V(U) is a finite union of torsion translated subtori of L(U). Subtori are automatically defined over Q. Example 2.3. If V denotes the complement in Pn−1 of r mutually distinct reduced and irreducible hypersurfaces of degrees d1, . . . , dr, then H1(V, Z) =" rMj=1 Z · γj# / Z(d1γ1 + . . . + drγr), where γj is the class of a small loop centered at a general point on the j-th hy- persurface, see [14, (4.1.3)]. By Budur-Wang [11], the cohomology support locus eV(M) :=[k Supp (Hk(M ab, C)) BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 11 V(V ) of any smooth complex quasi-projective variety V is a finite union of torsion translated subtori of L(V ). Let M ab be the universal abelian cover of M. In other words, M ab is the cover of M given by the kernel of the natural abelianization map ab : π1(M) → H1(M, Z). Definition 2.4. The homological Alexander support of M is the subset of L(M), where Supp (Hk(M ab, C)) is the support of the B-module Hk(M ab, C) . The homological Alexander support is almost the same as the cohomology sup- port locus (1) from the Introduction, see [36, Theorem 3.6]: implies: homology. Since H k(M, L−1) = Hk(M, L)∨ for a rank one local system L, the last result Theorem 2.5. eV(M) is the set of local systems of rank one on M with non-trivial Corollary 2.6. V(M) = {L L−1 ∈ eV(M)}. Example 2.7. V((C∗)r) is just a point in L((C∗)r) = (C∗)r, corresponding to the trivial local system. 3. Sabbah specialization and local systems In this section we write down some properties of the Sabbah specialization com- plexes. We also prove Theorems 2, 4, 5, as well as the homogeneous reduction formula mentioned in 1.2 and Corollary 1. For a ring R and a variety X, let Db c(X, R) denote the bounded derived category of R-constructible sheaves on the underlying analytic variety of X. 3.1. Sabbah specialization. Let F = (f1, . . . , fr) be a collection of non-zero polynomials fj ∈ C[x1, . . . , xn]. Let X = Cn, Dj = V (fj), D = ∪r j=1Dj, U = X r D. the affine coordinate ring of S∗ by Let S = Cr, S∗ = (C∗)r, and denote byfS∗ the universal cover of S∗. We denote A = C[t1, t−1 1 , . . . , tr, t−1 r ]. Consider the following diagram of fibered squares of natural maps: D  iD / X F S U? j FU S∗ jS F eU eUp fS∗ pS  /   _ o o   o o   ? _ o o o o 12 NERO BUDUR Definition 3.2. The Sabbah specialization functor of F is c(X, C) → Db DRj∗Rp!(j ◦ p)∗ : Db ψF = i∗ c(D, A). We call ψF CX the Sabbah specialization complex. The constructibility ψF over A follows from the part (a) of the Lemma 3.4 below. Remark 3.3. (1) This definition is slightly different than [40, 2.2.7] where it is called the nearby Alexander complex. To obtain the definition in loc. cit., one has to restrict further to ∩jDj. (2) This definition is also slightly different than the one in [32], where Rp! is replaced by Rp∗. (3) When r = 1, ψf F as defined here equals ψf F [−1] as defined by Deligne, see [8, p.13]. A different expression for Sabbah specialization is as follows. Let L = R(pS)!CfS∗ = (pS)!CfS∗. This is the rank-one local system of free A-modules on S∗ corresponding to the isomorphism Define π1(S∗) → Zr = π1(S∗). LF := F ∗ U L. This is a local system of A-modules on U. The following is essentially a particular case of [40, 2.2.8] in light of Remark 3.3 (1): Lemma 3.4. (a) ψF F = i∗ DRj∗(j∗F ⊗CU LF ). (b) In particular, ψF CX = i∗ DRj∗LF . Proof. By the projection formula [40, 2.1.3], Rj∗Rp!p∗j∗F = Rj∗(j∗F ⊗CU Rp!C eU ). On the other hand, we have Rp!C eU = Rp!F ∗ (cid:3) eU 3.5. Multi-variable monodromy zeta function. We recall Sabbah's multi- variable generalization of A'Campo's formula for the monodromy zeta function. U Rp!CfS∗ = LF . CfS∗ = F ∗ Definition 3.6. For an A-module G, we denote by Supp (G) the support of G in S∗ = Spec A. For an A-constructible sheaf G on X and a point x in X, the support of G at x is the support of the stalk: The support at the point x of a complex G ∈ Db Supp x(G) := Supp (Gx) ⊂ S∗ c(D, A) is Definition 3.7. The multi-variable monodromy zeta function of G ∈ Db the point x is defined to be the cycle c(D, A) at Supp x(G) :=[k Supp x(H k(G)) ⊂ S∗. ζx(G) :=XP χx(GP ) · V (P ), BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 13 where the sum is over prime ideals P of A of height one among those such that their zero locus V (P ) ⊂ Supp x(G), GP is the localization of G at the prime ideal P , and χx is the stalk Euler characteristic. Codimension-one cycles on S∗ can be viewed as rational functions in t1, . . . , tr up to multiplication by a monomial. We will use the rational function notation for the multi-variable monodromy zeta function of Sabbah specialization complex: ζx(ψF CX)(t1, . . . , tr) := ζx(ψF CX). Let µ : Y → X be a log resolution ofQj fj. Let Ei for i ∈ S be the collection of irreducible components of the zeros locus of (Qj fj) ◦ µ. Let ai,j be the order of vanishing of fj along Ei, and let ki be the order of vanishing of the determinant of the Jacobian of µ along Ei. For I ⊂ S, let E◦ I = ∩i∈IEi r ∪i∈S−IEi. The following is Sabbah's generalization of A'Campo's formula, see [40, 2.6.2]: Theorem 3.8. If fj(x) = 0 for all j, then ζx(ψF CX)(t1, . . . , tr) = Yi∈I with µ(Ei)=x (tai,1 1 . . . tai,r r − 1)−χ(E◦ i ). Corollary 3.9. If fj are non-zero homogeneous polynomials of degree dj for all j, and χ(V ) 6= 0, where V = Pn−1 r P(D), then V (td1 1 . . . tdr r − 1) ⊂ Supp 0(ψF CX). Proof. In this case, one can take a log resolution µ : Y → X which factors through the blow-up at 0, such that the strict transform E of the blow-up exceptional divisor E′ is the only exceptional divisor of Y mapping to 0, and such that µ is an isomorphism outside D. In this case, E′ = Pn−1, E◦ = V . By construction, the log resolution µ is the blow-up Y ′ of X at 0, followed by the extension to Y ′ of any fixed log resolution of P(D) in E′, see [38, 1.4]. This is possible because Y ′ is a line bundle over E′. Then, by Theorem 3.8, ζ0(ψF CX)(t1, . . . , tr) = (td1 1 . . . tdr r − 1)−χ(V ), and the conclusion follows. (cid:3) For r = 1 one recovers a well-known formula for the monodromy zeta function of a homogeneous polynomial, see [14, p.108]. 3.10. Local Alexander modules. Let x be a point in D and ix : {x} → D the natural inclusion. Let Ballx be a small open ball centered at x in X, and let Let U ab F,x be the universal abelian cover of UF,x, and let UF,x = Ballx r D. Let B = C[H1(UF,x, Z)]. gUF,x := UF,x ×S∗fS∗. 14 NERO BUDUR Consider the commutative diagram of fibered squares Ballx p UF,x F S S∗ gUF,x fS∗. Definition 3.11. The k-th local homology Alexander module of F at x is the A- module The k-th local homology Alexander module of UF,x is the B-module Hk(gUF,x, C). Hk(U ab F,x, C). The following is essentially a particular case of [40, 2.2.5]: Proposition 3.12. The k-th cohomology of the stalk of the Sabbah specialization complex is isomorphic as an A-module, up to the switch between of the action of tj with that of t−1 , to the k-th local homology Alexander module of F at x: j Proof. By Lemma 3.4, H k(i∗ at x equals the stalk of the presheaf V 7→ H k(U ∩ V, LF ). Hence xψF CX) = (Rkj∗LF )x. The stalk of the sheaf Rkj∗LF H k(i∗ xψF CX) ∼=A Hk(gUF,x, C). (2) We have H k(i∗ xψF CX) = H k(UF,x, LF ). Hk(gUF,x, C) = H 2n−k = H 2n−k c c (gUF,x, C) = H 2n−k(Ra!p!C gUF,x (UF,x, LF ) = H 2n−k(Ra!LF ), ) where a is the map to a point. On the other hand, letting (LF )∨ be the A-dual local system of LF , and DA be the Verdier duality functor, we have H 2n−k(Ra!LF ) = H n−k(Ra!DA((LF )∨[n])) = H n−k(DARa∗((LF )∨)[n]) = DAH k(Ra∗(LF )∨) = DAH k(UF,x, (LF )∨). The last A-module is non-canonically isomorphic, after the change of tj with t−1 , with H k(UF,x, LF ). (cid:3) j Lemma 3.13. If the polynomials fj with fj(x) = 0 define mutually distinct reduced and irreducible hypersurface germs at x, then and B = A/htj − 1 fj(x) 6= 0i. F,x gUF,x = U ab Proof. The second assertion follows from Example 2.2. For the first assertion, it is enough to show that gUF,x corresponds to the kernel of the abelianization map π1(UF,x) → H1(UF,x, Z).   ? _ o o   o o   ? _ o o o o BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 15 By definition, gUF,x is given by the kernel of the composition π1(UF,x) F∗−→ π1(S∗) = Zr id−→ H1(S∗, Z) = Zr. Since the codomain is abelian, it is enough to show that the natural direct image F∗ = ((f1)∗, . . . , (fr)∗) : H1(UF,x, Z) −→ H1(S∗, Z) is injective. By Example 2.2, H1(UF,x, Z) is free abelian generated by the classes of loops γj centered a general point of Ballx ∩ Dj for those j such that fj(x) = 0. Let δj be a generator for the first homology of the j-th copy of C∗ in S∗. Both assertions of the Lemma follow then from the fact that fj(γj′) ∼(cid:26) 0 δj if fj(x) 6= 0, or j 6= j′, if fj(x) = 0 and j = j′. Indeed, if fj(x) 6= 0, then γj′ can be chosen such that fj(γj′) is a loop homologically equivalent to 0. If j 6= j′, then γj′ can be chosen such that fj(γj′) is a point. If fj(x) = 0 and j = j′, then γj can be chosen to intersect at most once every fiber of fj, hence fj(γj) is homologically equivalent to δj. (cid:3) From Proposition 3.12 and Lemma 3.13 we obtain: Corollary 3.14. If the polynomials fj with fj(x) = 0 define mutually distinct reduced and irreducible hypersurface germs at x, then the k-th cohomology of the stalk of ψF CX at x is the k-th local homology Alexander module of UF,x. More precisely, the action of A on H k(i∗ xψF CX ) factors through the action of B, and H k(i∗ xψF CX) ∼= Hk(U ab F,x, C) as B-modules after replacing on the right-hand side the tj-action with the t−1 j - action. Remark 3.15. In the case when one the polynomials fr is nonsingular outside ∪r−1 j=1Dj, or, when fr is a generic linear polynomial through x, more information is available about the local Alexander modules from [40, 2.6.3 and 2.6.4]. 3.16. Proof of Theorem 4. In this case, L(UF,x) = Spec B = \j:fj(x)6=0 V (tj − 1) ⊂ S∗ taking reciprocals coordinate-wise due to the change in action of tj by t−1 by Example 2.2. Corollary 3.14 implies that Supp x(ψF CX ) equals eV(UF,x) via Corollary 2.6,eV(UF,x) equals V(UF,x) via taking reciprocals coordinate-wise. Hence, Supp x(ψF CX) = V(UF,x). (cid:3) 3.17. Uniform support. Even if the polynomials fj vanishing at x do not define mutually distinct reduced and irreducible hypersurface germs at x, the proof of Lemma 3.13 together with Proposition 3.12 show: . By j Lemma 3.18. The action of A on H k(i∗ xψF CX) factors through the action of A/(htj − 1 fj(x) 6= 0i). 16 NERO BUDUR As a consequence, for a point x ∈ D, the support Supp x(ψF CX) lies in a sub- torus of S∗ of codimension exactly the number of polynomials fj with fj(x) 6= 0. More precisely, let Tx := V (htj − 1 fj(x) 6= 0i) ⊂ S∗. Then Supp x(ψF CX ) ⊂ Tx. Let rx be the codimension of Tx in S∗, in other words the number of j's with fj(x) 6= 0. Definition 3.19. The F -natural splitting of S∗ at x is the splitting S∗ = Tx × (C∗)rx compatible with the splitting {j fj(x) = 0} ∪ {j fj(x) 6= 0}. Definition 3.20. The uniform support at x of ψF CX is Supp unif x (ψF CX ) := (Supp x(ψF CX)) × (C∗)rx ⊂ S∗, the last inclusion being induced by the F -natural splitting of S∗. By definition, the uniform support coincides with the usual support when Tx is empty, or in other words, when all fj vanish at x. In other words, the uniform support is defined by the same equations as the usual support except we discard the equations tj = 1 for those j such that the hypersurface fj does not pass through x. Similarly, consider the cohomology support locus V(UF,x). This is a subvariety of the space of rank one local systems L(UF,x), and L(UF,x) ⊂ Tx, with Tx as above. We have an equality L(UF,x) = Tx if the germs fj which vanish at x define mutually distinct reduced and irreducible hypersurface germs. Definition 3.21. The uniform cohomology support locus of F at x is V unif (UF,x) := V(UF,x) × (C∗)rx ⊂ S∗, the last inclusion being induced by the F -natural splitting of S∗. By definition, the uniform cohomology support locus coincides with the usual cohomology support locus when Tx is empty, or in other words, when all fj vanish at x. 3.22. Proof of Theorem 5. It follows from Theorem 4 and Theorem 3 which we prove latter. (cid:3) BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 17 3.23. Homogeneous polynomials. Assume now that fj are non-zero homoge- neous polynomials for all j. We show first that i∗ 0ψF CX recovers the universal Alexander modules of Dimca-Maxim [15, §5]. Let V = Pn−1 r P(Dj). r[j=1 Let dj be the degree of fj. Let V ab be the universal abelian cover of V . Then Hk(V ab, C) admits an action of If fj are mutually distinct irreducible homogeneous polynomials, which is the sit- uation considered in [15], B := C[H1(V, Z)]. B = A/htd1 1 · · · tdr r − 1i, and see Example 2.3. L(V ) = V (td1 1 · · · tdr r − 1) ⊂ S∗, Proposition 3.24. If fj are irreducible homogeneous polynomials of degree dj defining mutually distinct hypersurfaces with gcd(d1, . . . , dr) = 1, then the action of A on i∗ 0ψF CX factorizes through B, and H k(i∗ 0ψF CX) ∼= Hk(V ab, C) as B-modules after replacing on the right-hand side the tj-action with the t−1 j - action. Proof. Consider UF,0, the complement in a small open ball at the origin of D. The natural projectivization map UF,0 → V has fibers diffeomorphic with C∗ and is a deformation retract of the restriction of the tautological line bundle of Pn−1 to V . Since gcd(d1, . . . , dr) = 1, the Picard group of V is trivial. Hence, topologically, UF,0 ≈ C∗ × V. Fix a section σ : V → UF,0. First, we show that via this section V ab ≈ V ×UF,0 U ab F,0. The cover on the right-hand side is given by the kernel of the composition π1(V ) σ∗−→ π1(UF,0) ab−→ H1(UF,0, Z). The cover on the left-hand side is given by the kernel of π1(V ) ab−→ H1(V, Z). Hence, it is enough to show that the map σ∗ : H1(V, Z) −→ H1(UF,0, Z) is injective. By assumption, both groups are free abelian of rank r − 1 and, respec- tively, r. Hence σ∗ is compatible with the Kunneth decomposition H1(UF,0, Z) = H1(V, Z) ⊕ Z, 18 NERO BUDUR by the naturality of the Kunneth decomposition via cross products, and the injec- tivity follows. This also shows that A acts on Hk(V ab, C) via the surjection A = C[H1(UF,0, Z)] −→ B = C[H1(V, Z)] = A/htd1 1 . . . tdr r − 1i induced by σ∗. Now the Proposition follows from Corollary 3.14 and the fact that V ab is a deformation retract of U ab F,0 ≈ (C∗)ab × V ab ≈ C × V ab. (cid:3) Proposition 3.25. If fj are irreducible homogeneous polynomials of degree dj defining mutually distinct hypersurfaces with gcd(d1, . . . , dr) = 1, then, in S∗: Supp 0(ψF CX ) = V(V ) ⊂ L(V ) = V (td1 1 · · · tdr r − 1). The four sets are equal if, in addition, χ(V ) 6= 0. Proof. The first equality is new. It follows from Proposition 3.24 and Corollary 2.6. If χ(V ) 6= 0, the equality follows from Corollary 3.9. (cid:3) Remark 3.26. In the case when one of the homogeneous polynomials is a generic linear form, more information is available about V(V ), see [15, Theorem 3.6] and see also Remark 3.15. For a point y ∈ X = Cn different than the origin, let [y] denote the point in Pn−1 with homogeneous coordinates given by y. We denote by UF,[y] ⊂ Pn−1 the complement in a small ball around [y] of P(D). Note that if we consider an affine space neighborhood of [y], then [y] ∈ An−1 ⊂ Pn−1, UF,[y] = UFAn−1 ,[y]. There is a homotopy equivalence UF,y ≈ht UF,[y]. Hence with Ty as in 3.17. Moreover, V(UF,y) = V(UF,[y]) = V(UFAn−1 ,[y]) ⊂ Ty, V unif (UF,y) = V unif (UFAn−1 ,[y]) ⊂ S∗ because the F -natural splitting S∗ = Ty × (C∗)ry is the same as the FAn−1-natural splitting. We define: V unif (UF,[y]) := V unif (UF,y) = V unif (UFAn−1 ,[y]). From this discussion together with Proposition 3.25, we obtain the following com- putation reduction to a lower-dimensional, possibly non-homogeneous, case: BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 19 Proposition 3.27. If the polynomials fj are homogeneous of degree dj and define mutually distinct reduced and irreducible hypersurfaces, and if gcd(d1, . . . , dj) = 1, then [y∈D Supp unif y ψF CX = [y∈D V unif (UF,y) = V(V ) ∪ [[y]∈Pn−1 V unif (UF,[y]). If, in addition, χ(V ) 6= 0, then V(V ) = L(V ) = V (td1 1 · · · tdr r − 1). 3.28. Proof of Corollary 1. It follows from Theorem 3, which we will prove later, and Proposition 3.25. (cid:3) 3.29. Specialization of polynomial maps. Next, we address the question of what happens with the support of ψF CX under specialization of the map F in the following sense: Definition 3.30. (a) We say that F = (f1, . . . , fr) specializes to G = (g1, . . . , gp), if fj and gk are non-zero polynomials in C[x1, . . . , xn], r ≥ p ≥ 1, and G is the composition F ′ ◦ F where F ′ = (f ′ 1, . . . , f ′ p) : Cr −→ Cp is such that f ′ surjective. We will also write k are monomial maps and the induced map on tori (C∗)r → (C∗)p is where the matrix M = (mkj) in Np×r is obtained from writing k : (τ1, . . . , τr) 7→ τ mk1 f ′ 1 . . . τ mkr r . G = F M (b) We say that F M is a non-degenerate specialization of F if, in addition, k=1 mkj 6= 0 for all j such that fj is non-constant. The condition in (b) guarantees that no non-constant fj is lost during the spe- Pp cialization. We will use the notation: S = Cr, SM = Cp, S∗ = (C∗)r, S∗ M = (C∗)p. There is a natural identification of S∗ with L(S∗), a tuple of non-zero complex num- bers describing the monodromy of a rank one local system around the coordinate axes. The pull-back of local systems defines a map given by (λ1, . . . , λp) 7→ (λm11 1 ). Denote by φM : S∗ p M ) −→ S∗ = L(S∗) M = L(S∗ · · · λmp1 A = C[t1, t−1 AM = C[u1, u−1 , . . . , λm1r · · · λmpr p 1 , . . . , tr, t−1 r ], 1 , . . . , up, u−1 p ] 1 the coordinate rings of S∗ and S∗ morphism M , respectively. Then φM corresponds to the ring mapping tj toQk umkj k . The following is a consequence of [40, 2.3.8]: φ# M : A −→ AM 20 NERO BUDUR Proposition 3.31. If G = F M is a non-degenerate specialization of the polynomial map F , then for all x M (Supp unif φ−1 (ψF CX)) = Supp unif (ψGCX ). x x Proof. Let DM be the union of the zero loci of gk. The non-degeneracy assumption is equivalent to saying that DM = D. We can assume x is in D. x Indeed, φ−1 Supp x(ψF CX ⊗L The statement holds for the usual support. A AM ), and by [40, 2.3.8], Supp x(ψF CX ⊗L M (Supp x(ψF CX)) = A AM ) = Supp x(ψGCX ). Recall that the uniform support at x is obtained from the F -natural splitting S∗ = Tx × (C∗)rx, where Tx is the zero locus V (htj − 1 fj(x) 6= 0i) in S∗ and contains Supp x(ψF CX ), by setting Supp unif (ψF CX ) = Supp x(ψF CX ) × (C∗)rx. Similarly, S∗ M , and Supp unif M = TM,x×(C∗)px, where TM,x is the zero locus V (huk − 1 gk(x) 6= 0i) in S∗ (ψGCX ) = Supp x(ψGCX) × (C∗)px, where the coordinates of the last term (C∗)px correspond to uk such that gk(x) 6= 0. Hence it is enough to show that φM , or equivalently φ# M , is compatible with the splittings. The splitting on S∗ is by definition compatible with the splitting x The splitting on S∗ {1, . . . , r} = {j fj(x) = 0} ∪ {j fj(x) 6= 0}. M is compatible with the splitting {1, . . . , p} = {k gk(x) = 0} ∪ {k gk(x) 6= 0}. Hence it is enough to show that these two splittings are compatible under φ# M . More precisely, let A′ = C[t± fj(x) 6= 0], so that A = j A′ ⊗C A′′. Similarly define A′ M = C[uk gk(x) 6= 0], so that AM = A′ M . We need to show that the middle horizontal map in the diagram fj(x) = 0] and A′′ = C[t± j M = C[uk gk(x) 6= 0] and A′′ M ⊗ A′′ A′ /❴❴❴❴❴❴❴❴❴ A′ M A = A′ ⊗ A′′ φ# M / AM = A′ M ⊗ A′′ M A′ /❴❴❴❴❴❴❴❴❴ A′ M induces the other two horizontal maps making the diagram commute, where the top vertical maps are id ⊗ 1 and the bottom vertical maps are the natural quotient maps given by the F and FM natural splittings. In other words, we need to show that the set {k ∈ {1, . . . , p} uk does not appear in φ# M (tj) for any j with fj(x) = 0} is the same as the set This is true since both sets equal the set {k ∈ {1, . . . , p} gk(x) 6= 0}. {k ∈ {1, . . . , p} mkj = 0 for all j with fj(x) = 0}. (cid:3) /     /         / BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 21 Example 3.32. F = (f1, . . . , fr) specializes to G =Qr j=1 fj via the 1 × r matrix M = (1 . . . 1). Hence this specialization is non-degenerate. Here S∗ M → S∗ is the diagonal inclusion and φ : A → AM = C[u, u−1] is given by tj 7→ u. In this case, the uniform support of ψGCX at x is the same as the usual support: Supp unif x (ψGCX) = Supp x(ψGCX), and it consists of the eigenvalues of the monodromy on the Milnor fiber of G at x by Remark 3.3. Example 3.33. F = (f1, . . . , fr) specializes to G = (f1, . . . , fr−1). To see this, in Definition 3.30 let F ′ : S → SM = Cr−1 be defined by (τ1, . . . , τr) 7→ (τ1, . . . , τr−1), that is M is the square identity matrix with the last row deleted. Hence this specialization is degenerate if fr is non-constant. Here the inclusion S∗ M → S∗ is given by (τ1, . . . , τr−1) 7→ (τ1, . . . , τr−1, 1). The map A → AM is the natural one induced by the quotient AM = A/htr − 1i. Lemma 3.34. Let x be a point in X. For any collection G = (g1, . . . , gp) of hyper- surface germs at x in X, there exists a collection F = (f1, . . . , fr) of hypersurface germs at x such that the set of fj with fj(x) = 0 define mutually distinct reduced and irreducible germs, and such that G is a non-degenerate specialization of F . Proof. Suppose we find F and a matrix M = (mkj) such that G = F M . Then we need to ensure the surjectivity of the map F ′ : S∗ → S∗ M defined by (τ1, . . . , τr) 7→ (. . . , τ mkj k , . . .). rYj=1 This is achieved if the rank of the matrix M is p, since the linear transformation associated to M is the differential of the map F ′ on the associated Lie algebras of S∗ and S∗ Let r by the number of analytically irreducible components of the germQp Let fj for 1 ≤ j ≤ r be those components. Write k=1 gk. M . gk = f mkj j rYj=1 and let M be the p × r matrix (mkj). We can assume by permuting the germs gk that the first rank(M) rows of M are linearly independent. Let F = (f1, . . . , fr, 1, . . . , 1), where the number of 1's added after fr is p − rank(M). Then F satisfies the conditions of the lemma and G is the specialization of F via the rank p matrix of size p × (p + r − rank(M)) (cid:20)M O I (cid:21) , where O is the zero matrix, and I is the identity square matrix of size p − rank(M). (cid:3) 22 NERO BUDUR 3.35. Proof of Theorem 2. By construction of the uniform support, it is enough to prove the statement for Supp x(ψF CX ). By Lemma 3.34, the germ of F at x in X is the non-degenerate specialization GM of a map germ G = (g1, . . . , gp) via some matrix M, where the set of gk with gk(x) = 0 define mutually distinct reduced and irreducible germs. Thus Supp x(ψF CX) equals φ−1 M (Supp x(ψGCX)) by specialization of the support. By Theorem 4, Supp x(ψGCX) is the cohomology support locus of UG,x, and thus it is a finite union of torsion translated subtori of L(UG,x), see Example 2.2. Since φM is a torus homomorphism, the same is true for Supp x(ψF CX). (cid:3) 3.36. Thom-Sebastiani. Next, we state a multiplicative Thom-Sebastiani type of result for the support of the Sabbah specialization complex. First, we have: Lemma 3.37. For i = 1, 2, let Xi = Cni, Fi = (fi1, . . . , firi) with fij 6= 0, Di = ∪jf −1 ij (0), xi ∈ Di. Then, in (C∗)r1 × (C∗)r2, we have Supp unif (x1,x2)(ψF1×F2CX1×X2) = Supp unif x1 (ψF1CX1) × Supp unif x2 (ψF2CX2). Proof. Using the notation of Lemma 3.4 adapted to our situation, there is an equality of local systems of A1 ⊗ A2-modules LF1×F2 = LF1 ⊠ LF2, where ⊠ denotes the external direct product on U1 × U2, with Ui = Xi r Di, and Ai is the coordinate ring of S∗ i = (C∗)ri. Using then Lemma 3.4 and standard arguments, one can show that ψF1×F2CX1×X2 = ψF1CX1 L ⊠ ψF2CX2, Hence the claim holds for the usual supports. See also [36, Proposition 3.1] for the same statement for the cohomology support loci. The claim for the uniform support follows easily from Definition 3.20. (cid:3) The following is the multiplicative Thom-Sebastiani property for the support of the Sabbah specialization complex: Proposition 3.38. With notation as in Lemma 3.37, let r = r1 = r2. Let G be the map G = F1 · F2 : X1 × X2 −→ (C∗)r defined by (x1, x2) 7→ (f11(x1)f21(x2), . . . , f1r(x1)f2r(x2)) for xi ∈ Xi, i = 1, 2. Then Supp unif (x1,x2)(ψGCX1×X2) = \i=1,2 Supp unif xi (ψFiCXi). Proof. Let S = Cr and let F ′ : S × S → S be defined by multiplication coordinate- wise. Then G = F ′ ◦ (F1 × F2) and thus F1 × F2 specializes to G, cf. Defini- tion 3.30. Hence, by Proposition 3.31, Supp unif (x1,x2)(ψF1×F2CX1×X2) specializes to Supp unif (x1,x2)(ψGCX1×X2) via intersection with the diagonal in S∗ × S∗. The claim then follows from Lemma 3.37. (cid:3) BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 23 4. Bernstein-Sato ideals In this section we develop some properties of Bernstein-Sato ideals. We use them to prove geometrically a weaker version of Theorem 3. With a similar proof, we then prove Theorems 6 and 7. 4.1. Ideals of Bernstein-Sato type. There are ways to define ideals of Bernstein- Sato type different than presented in the Introduction. These other ideals are useful for understanding the Bernstein-Sato ideal BF . So we start with a more general definition. Let X = Cn. Let DX denote the Weyl algebra of algebraic differential operators on X, DX = C(cid:20)x1, . . . , xn, ∂ ∂x1 , . . . , ∂ ∂xr(cid:21) with the usual commutation relations. Let F = (f1, . . . , fr) with 0 6= fj ∈ C[x1, . . . , xn]. Let M = {mk ∈ Nr k = 1, . . . , p} be a collection of vectors, which we also view as an p × r matrix M = (mkj) with mkj = (mk)j, with r, p ≥ 1. Definition 4.2. The Bernstein-Sato ideal associated to F and M is the ideal F = Bm1,...,mp of all polynomials b(s1, . . . , sr) such that B M F ⊂ C[s1, . . . , sr] b(s1, . . . , sr) f sj j = rYj=1 Pk pXk=1 rYj=1 f sj+mkj j for some algebraic differential operators Pk in DX [s1, . . . , sr]. Remark 4.3. (a) BF , as defined before, is B 1 F , where 1 = (1, . . . , 1). (b) For a point x in X, the local Bernstein-Sato ideal B M F,x is similarly defined by replacing DX with the ring DX,x of germs of holomorphic differential operators at x. Then B M F = \x∈X B M F,x, see [4, Corollary 6]. (c) The ideals B M F,x are non-zero by Sabbah [39], see also [3], [21]. Example 4.4. If fj are monomials, write fj = Qn Pr j=1 ai,jsj. Let ai =Pr j=1 ai,j. Then BF = h (li(s) + 1) · · · (li(s) + ai)i. i=1 xai,j i . Let li(s1, . . . , sr) = nYi=1 24 NERO BUDUR Lemma 4.5. [21, Lemma 1.5] By the correspondence sj ↔ −∂tj tj and fj ↔ δ(tj − fj), where δ(u) denotes the Dirac delta function, i.e. the standard gen- erator of DA1/DA1u with u the affine coordinate on A1, there is an isomorphism of DX[s1, . . . , sr]-modules DX[s1, . . . , sr] rYj=1 f sr j ∼= DX[−∂t1t1, . . . , −∂tr tr] δ(tj − fj). rYj=1 The action of tj on the right-hand side corresponds to replacing sj by sj + 1 on left-hand side. Let Y = X × Cr with affine coordinates x1, . . . , xn, t1, . . . , tr. Define for m ∈ Nr, V mDY := DX ⊗C Xβ,γ∈Nr β−γ≥m Ctβ1 1 . . . tβr r ∂γ1 t1 . . . ∂γr tr ⊂ DY . The following is the D-module theoretic interpretation of Bernstein-Sato ideals and it is a consequence of Lemma 4.5: Proposition 4.6. Let m ∈ Nr. The Bernstein-Sato ideal Bm polynomials b(s1, . . . , sr) such that F consists of the b(−∂t1 t1, . . . , −∂tr tr) · V 0DY · δ(tj − fj) ⊂ V mDY · rYj=1 δ(tj − fj). rYj=1 The next result unveils somewhat the structure of the Bernstein-Sato ideals. This can be used in practice to compute BF in cases where the current implementations do not work. It also explains to some extent the nature of the mysterious shifts which appear in Bernstein-Sato ideals. Theorem 4.7. Let m ∈ Nr. For j = 1, . . . , r, let tj be the ring isomorphism of C[s1, . . . , sr] defined by tj(si) = si + δij. Then there are inclusions of ideals in C[s1, . . . , sr] Y1≤j≤r mj >0 mj−1Yk=0 tm1 1 . . . tmj−1 j−1 tk j B ej F ⊂ Bm F ⊂ \1≤j≤r mj >0 mj −1\k=0 tm1 1 . . . tmj−1 j−1 tk j B ej F . Here δij = 0 if i 6= j, and δii = 1. Also, we denote by ej the r-tuple with the 1 . . . tar r r I is the image of the ideal I k-th entry δjk. By convention, t0 means the obvious composition of maps, and ta1 under this product map. j is the identity map, the product map ta1 1 . . . tar The first inclusion in Theorem 4.7 can be strict, see Examples 4.13 and 4.20. We do not know examples for which the second inclusion is strict, raising the obvious question if equality holds in general. The radical of product of ideals equals the radical of the intersection of the ideal. Thus, letting V (I) denote the zero locus of an ideal I ⊂ C[s1, . . . , sr] in Cn, and writing tj also for the corresponding linear map of Cn, we obtain: BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 25 Proposition 4.8. With the notation as above, V (Bm F ) = [1≤j≤r mj >0 mj −1[k=0 tm1 1 . . . tmj−1 j−1 tk j V (B ej F ). Theorem 4.7 is a consequence of the following result, in which we will use the notation tm =Qr j=1 tmj j . Lemma 4.9. Let m, n ∈ Nr. Then there are inclusions of ideals Bm F · (tm B n F ) ⊂ Bm+n F ⊂ Bm F ∩ (tm B n F ). Proof. We will use the notation f s =Qr F . Write b1f s = P1f s+m and b2f s = P2f s+n for some P1 and P2 in DX [s1, . . . , sr]. Apply tm to both sides of b2f s = P2f s+n. We obtain then that (tmb2)f s+m = (tmP2)f s+m+n. Applying P1 on the left on both sides of the equality, we have j . Let b1 ∈ Bm F and b2 ∈ B n j=1 f sj P1(tmP2)f s+m+n = P1(tmb2)f s+m = (tmb2)P1f s+m = (tmb2)b1f s. Thus b1(tmb2) is in Bm+n , which implies the first inclusion. F Take now b ∈ Bm+n . Write bf s = P f s+m+n for some P in DX [s1, . . . , sr]. Then bf s = P f nf s+m, so b ∈ Bm F . Now, multiply by f m on the left on both sides of the last equality. We obtain that bf s+m = f mP f s+m+n. This shows that b ∈ tmB n F . Hence b ∈ Bm (cid:3) F ), which proves the second inclusion. F ∩ (tmB n F Proof of Theorem 4.7. Write m = (m1, 0, . . . , 0)+(0, m2, . . . , mr) and apply Lemma 4.9 to obtain that Bm F is squeezed between the product and the intersection of the ideals Bm1 e1 and so on. It remains to squeeze the ideals Bmj ej , for which there is the obvious procedure: write mjej = (mj − 1)ej + ej, use Lemma 4.9, and repeat. (cid:3) . Repeat the procedure with B(0,m2,...,mr) and tm1 e1B(0,m2,...,mr) F F F F Remark 4.10. Note that one can choose different decompositions of m than the ones used in the proof of Theorem 4.7. In particular, if π is any permutation of {1, . . . , r}, there are inclusions PF,m,π := Y1≤j≤r mπ(j)>0 mπ(j)−1Yk=0 mπ(1) t π(1) and . . . t mπ(j−1) π(j−1) tk e π(j) B π(j) F ⊂ Bm F Bm F ⊂ \1≤j≤r mπ(j)>0 mπ(j)−1\k=0 mπ(1) t π(1) . . . t mπ(j−1) π(j−1) tk e π(j) B π(j) F =: IF,m,π. Again, we do not know an example where Bm F 6= IF,m,π. The ideals PF,m,π could be different for different permutations π, see Example 4.20. Hence we the record the following as a strengthening of Theorem 4.7 and Proposition 4.8. 26 NERO BUDUR Proposition 4.11. With the notation as above, there is an inclusion of ideals Xπ PF,m,π ⊂ Bm IF,m,π F ⊂\π where π ranges over all permutations of {1, . . . , r}. Hence, there is an equality of zero loci V (Bm F ) = [1≤j≤r mπ(j)>0 mπ(j)−1[k=0 mπ(1) t π(1) . . . t mπ(j−1) π(j−1) tk π(j) V (B F e π(j) ). Example 4.12. Let F = (f ) for a polynomial f . This is the case r = 1. Let m = (m) with m ∈ N. Then Bm F = hbf m(s/m)i, where bf m(s) is the classical Bernstein- j=0 bf (s + j) is divisible by bf m(s/m) which, in turn, is divisible by the lowest common multiple lcm{bf (s + j) j = 0, . . . , m − 1}. Sato polynomial of f m. Thus Theorem 4.7 states in this case thatQm−1 Example 4.13. In the previous example, let f = x4 − y2z2. Then bf (s) = (s + 1)3(4s + 3)2(4s + 5)2(2s + 3)(4s + 7). Thus, lcm{bf (s), bf (s + 1)} = bf (s)bf (s + 1) 4s + 7 . One also computes that the right-hand side equals bf 2(s/2). Hence the first inclu- sion in Theorem 4.7 is strict in this case. We write next a few immediate consequences of Theorem 4.7 and Proposition 4.8. Lemma 4.14. Let m ∈ Nr. Then there is an equality Exp (V (Bm F )) = In particular, there is a decomposition Exp (V (Bmj ej )). F Exp (V (BF )) = Exp (V (B ej F )). r[j=1 r[j=1 Example 4.15. The last lemma does not necessarily hold without exponentiating. For example, let F = (xy, (1 − x)y). Then one can compute with dmod.lib [24] that BF = h(s1 + 1)(s2 + 1)(s1 + s2 + 1)(s1 + s2 + 2)i, F = h(s1 + 1)(s1 + s2 + 1)i and B e2 but B e1 component V (s1 + s2 + 2) of V (BF ) does not show up in either V (B F = h(s2 + 1)(s1 + s2 + 1)i. Hence the ej F ). Lemma 4.16. If m ∈ N is nonzero, then Exp (V (Bmej F )) = Exp (V (B ej F )) for 1 ≤ j ≤ r. BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 27 Example 4.17. The last lemma does not necessarily hold without exponenti- ating. Let r = 1, F = (x), and m = 2. Then B e1 F = BF = hs + 1i, but B2e1 F = h(s + 1)(s + 2)i. Lemma 4.18. If m = (mj), m′ = (m′ m′ j = 0 for all 1 ≤ j ≤ r, then j) ∈ Nr such that mj = 0 if and only if Exp (V (Bm F )) = Exp (V (Bm′ Exp (V (B ej F )). F )) = [j:mj6=0 The following will not be used, but we state it for clarification purposes and it follows from the definition. Lemma 4.19. For a matrix M ∈ Np×r with row vectors mk for 1 ≤ k ≤ p, B M F ⊃ Bmk F , V (B M F ) ⊂ V (Bmk F ). pXk=1 p\k=1 All statements in this subsection are true for local Bernstein-Sato ideals as well. Example 4.20. Let F = (z, x4 + y4 + 2zx2y2). This appeared in [4, Example 3]. One computes with dmod.lib [24]: BF = h(s1 + 1)(s2 + 1)2(2s2 + 1)(4s2 + 3)(4s2 + 5)(2s2 + 3)i, BF,0 = h(s1 + 1)(s2 + 1)2(2s2 + 1)(4s2 + 3)(4s2 + 5)(s1 + 2), (s1 + 1)(s2 + 1)2(2s2 + 1)(4s2 + 3)(4s2 + 5)(2s2 + 3)i, B e1 F = hs1 + 1i ∩ h2s2 + 1, s1 + 2i, B e2 F = h(s2 + 1) (4s2 + k)i, 6Yk=2 B e1,e2 F = hs1 + 1, (s2 + 1)2i ∩ hs1 + 2, 2s2 + 1i ∩ \k=2,3,5 hs1 + 1, 4s2 + ki. Theorem 4.7 and also its strengthening, Proposition 4.11, imply that B e1 (t2B e1 F · (t1B e2 F ) · B e2 F ) ⊂ BF ⊂ B e1 F ⊂ BF ⊂ (t2B e1 F ∩ (t1B e2 F ), F ) ∩ B e2 F , which can be checked easily from the above formulas. In this example the three ideals B e1 F , and BF are mutually distinct, and the two inclu- sions in Proposition 4.11 are equalities. F ), (t2B e1 F · (t1B e2 F ) · B e2 4.21. Bernstein-Sato ideals and specialization. With F as above, let F M be a specialization of F via a p × r matrix M = (mkj) as in 3.29, possibly without the surjectivity assumption of that definition. With the notation as in 3.29, the pull-back of local systems and the exponential maps from the tangent spaces at the trivial local systems define a commutative diagram 28 (3) NERO BUDUR fS∗ M := T1L(S∗ M ) ≈ Cp Exp S∗ M = L(S∗ M ) eφM φM /fS∗ = T1L(S∗) ≈ Cr Exp / S∗ = L(S∗) The top right-hand side space was denotedfS∗ in 3.1 and we keep the notation. How- ever, we keep in mind that T1L(S∗) is the natural ambient space of the Bernstein- Sato ideal BF . The bottom horizontal map is φM : (λ1, . . . , λp) 7→ (λm11 1 · · · λmp1 p , . . . , λm1r 1 · · · λmpr p ). The top map is the linear map given by multiplication on the left by the matrix M: eφM : (α1, . . . , αp) 7→ (m11α1 + · · · + mp1αp, . . . , m1rα1 + · · · + mprαp). Let C[s1, . . . , sr] and C[v1, . . . , vp] be the coordinate rings on fS∗ and fS∗ tively. The induced map on coordinate rings is M , respec- The following is straight-forward from the definition: Lemma 4.22. For all non-zero M ∈ Np×r and all vectors m ∈ Np, M : sj 7→ mkjvk. pXk=1 M (Bm·M F ) ⊂ Bm F M . M (V (Bm·M F )) ⊃ V (Bm F M ). eφ# eφ# eφ−1 F Hence eφ−1 Example 4.23. The converse does not necessarily hold. Let F = (y2 − x3, x5), m = 1, M = [1 0]. Then Bm·M = B e1 F = h(s1 + 1)Qk=5,7,9(6s1 + 10s2 + k)i. The ) = h(v + 1)Qk=5,7,9(6v + k)i. F M = BF M = h(v + 1)(6v + 5)(6v + 7)i. M (Bm·M F map eφ# However, F M = (y2 − x3) and Bm M sends s1 7→ v and s2 7→ 0. Hence eφ# Proposition 4.24. For all non-zero M ∈ Np×r , φ−1 M (Exp (V (BF ))) ⊃ Exp (V (BF M )). Proof. We apply the last lemma with m the unit vector in Np to obtain that M (V (B 1·M ))) ⊃ Exp (V (BF M )). Now we apply Lemma 4.18 to show that )) ⊃ V (BF M ). Hence φ−1 M (Exp (V (B 1·M F F φ−1 M (Exp (V (B 1·M F ))) ⊂ φ−1 M (Exp (V (BF ))). (cid:3)   /   / and (6s1 + 6s2 + k)i BF M = h BF = h Yk=5,6,7,11,12,13 Y Exp (V (BF )) = V ( Yk=1,5,6 k=5,6,7,11,12,13,17,18,19,23,24,25 (24s + k)i. (t1t2 − e−2πi k 6 )). The last ideal is generated by the classical Bernstein-Sato polynomial of f 4. Thus BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 29 Conjecture 4.25. For all non-zero M ∈ Np×r and all vectors m ∈ Np, φ−1 M (Exp (V (Bm·M F ))) = Exp (V (Bm F M )). In particular, for all M with nonzero columns, φ−1 M (Exp (V (BF ))) = Exp (V (BF M )). Remark 4.26. For this conjecture to hold it suffices that F M )) F ))) ⊂ Exp (V (B ek φ−1 M (Exp (V (B ej for all 1 ≤ j ≤ r, 1 ≤ k ≤ p such that mkj 6= 0, according to Lemma 4.18. Example 4.27. Consider F = (f, f ), with f = x2 + y3, and M = (2 2), so that F M = (f 4). Then, one computes using [24] that Since the map φM is λ 7→ (λ2, λ2), we see that M (Exp (V (BF ))) = {e−2πi k φ−1 24 k = 1, 5, 6, 7, 11, 12, 13, 17, 18, 19, 23, 24}, which is the same as Exp (V (BF M )). Note that eφ−1 inclusion ⊃ does not hold. M (V (BF )) 6= V (BF M ) since the All statements in this subsection are true for local Bernstein-Sato ideals as well. 4.28. Geometric proof of a weaker version of Theorem 3. We shall give now a proof of the statement that one gets by replacing in Theorem 3 the set Exp (V (BF,x)) with its analytic Zariski closure, namely that (Exp (V (BF,x)))cl ⊃ [y∈D near x Supp unif y (ψF CX ). The method is to use specialization of polynomial maps to reduce the statement to the case r = 1 for which equality is known by Malgrange and Kashiwara. We define a subset of S∗ by So = [m∈Nr Im (φm), where φm : S∗ m → S∗ is the map λ 7→ (λm1, . . . , λmr ) from (3). Note that a torsion point in S∗ must lie in So. The Zariski closure, analytic or algebraic, of So is S∗. y ψF CX are torsion trans- lated subtori of S∗. By [6, Proposition 3.3.6], for each component the torsion By Theorem 2, the irreducible components of Supp unif 30 NERO BUDUR points are Zariski dense. Hence the algebraic and analytic Zariski closure of So ∩ Supp unif y ψF CX equal Supp unif y ψF CX. On the other hand, So ∩ Supp unif y Im (φm) ∩ Supp unif y (ψF CX) (ψF CX ) = [m∈Nr Consider the specialization F m of F via the vector m. Then F m =Qr Proposition 3.31, Im (φm) ∩ Supp unif (ψF CX) = φm(Supp unif (ψF mCX)) y y j=1 f mj j . By if the specialization is non-degenerate, that is, if no coordinate of m is zero. There- fore, for such m, by the classical result of Malgrange and Kashiwara we have φm(Supp unif y (ψF mCX)) = φm(Exp (V (BF m,x))), where BF m is generated by the classical Bernstein-Sato polynomial of F m. By Proposition 4.24, this set is included in Im (φm) ∩ Exp (V (BF )). Hence So ∩ Supp unif (ψF CX) ⊂ So ∩ Exp (V (BF,x)) y [y∈D near x [y∈D near x away from the zero locus V (Qr we obtain the claim since V (Qr Sy∈D near x Supp unif Dj near x. (cid:3) y j=1(tj − 1)). Passing to the analytic Zariski closures, j=1(tj − 1)) lies anyway in both Exp (V (BF,x) and (ψF CX), being contributed by the smooth points y of the germs F 4.29. Proof of Theorem 6. We follow the same strategy as in 4.28. We view the multi-variable topological zeta function Z top as a rational function on the same Similarly for a matrix M ∈ Np×r, Z top F M is a rational function on T1L(S∗ M ≈ M on coordinate rings partially extends to one on the M , and which is defined for rational functions ambient space as for the Bernstein-Sato ideal BF , namely on T1L(S∗) =fS∗ ≈ Cr. M ) = fS∗ Cp. The induced map eφ# function fields, which we also denoteeφ# with polar locus not containing the image vector subspace Im(eφM ). In particular, F M = eφ# if F M is a non-degenerate specialization of F , see Definition 3.30. In this case, for polar loci we have M (Z top F ) Z top M (P L(Z top F )) ⊂ P L(Z top F M ) ∪ WM , where WM is the union of the linear codimension-one subvarieties of eS∗ in both eφ−1 M which lie F )), where ZL stands for the zero locus of a rational function. Hence M (P L(Z top M (ZL(Z top M (P L(Z top F ))) ⊂ φ−1 M (Exp (P L(Z top F ))), Exp (P L(Z top (4) with the last inclusion an equality if the map φM is injective. Also in this case, P L(Z top F M ) ⊂ eφ−1 F )) and eφ−1 F M )) ⊂ Exp (eφ−1 M (Exp (P L(Z top φ−1 F ))) ⊂ Exp (P L(Z top F M )) ∪ Exp (WM ). By assumption, the Monodromy Conjecture holds for F m. Thus there is an inclu- sion φm(Exp ((P L(Z top φm(Supp unif x (ψF mCX)), [m which is in turn included in F m)))) ⊂[m [x [x So ∩ Supp unif x (ψF CX ) BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 31 The map φM is injective if for example if M consists of only one row m with the greatest common divisor of the entries equal to 1. Let us cover S∗ discretely by such images. More precisely, let So be as in 4.28. Note it is enough to consider the union of Im(φm) over m ∈ Nr with the greatest common divisor of the entries equal to 1. Since we want to consider only non-degenerate specializations via m, we shall furthermore restrict in the definition of So to vectors m with non-zero entries. Note that the Zariski algebraic closure of So is still S∗ after all these restrictions. F )) is algebraically Zariski closed, it is the algebraic Zariski Since Exp (P L(Z top closure of So ∩ Exp (P L(Z top F )). Also, So ∩ Exp (P L(Z top φm(Exp ((P L(Z top φm(Exp (Wm)). F )) ⊂[m F m)))) ∪[m x by specialization of supports. Taking closure, we obtain that Exp (P L(Z top F )) (ψF CX) with the algebraic Zariski clo- is included in the union of Sx Supp unif sure ofSm φm(Exp (Wm)). It remains to show that we can ignore this last term Sm φm(Exp (Wm)). Note that Wm has dimension zero and that Sm φm(Exp (Wm)) equals So ∩ Exp (P L(Z top F )∩ZL(Z top F ). Write T ∩ ZL(Z top F ) = Zlin ∪ Znonlin, where Zlin is the union of linear irreducible com- ponents and Znonlin is the union of all the other components. Since Zlin ( T and Exp (Zlin) is algebraically Zariski closed, running the previous argument with So replaced by So = ∪mIm(φm) with Im(φm) ∩ Exp (Zlin) = ∅, we obtain that Exp (T ) (ψF CX) with the algebraic Zariski closure F )). Let T be an irreducible component of P L(Z top is included in the union ofSx Supp unif ofSm φm(Exp (Wm) where the union is over those m such that φm(Exp (Wm)) ∈ Exp (Znonlin). So we can assume that Zlin is empty and it remains to deal with the non-linear term Znonlin. x The problem with Znonlin is that Exp (Znonlin) might have the same algebraic Zariski closure as Exp (T ). However, if Znonlin 6= ∅, we change slightly the ar- gument: instead of filling the component Exp (T ) discretely with points of type Im(φm), we fill it discretely with restrictions of higher dimensional subtori Exp (T )∩ Im(φM ). More precisely, we let now So be the union of Im(φM ) over matrices of natural numbers M of size p × r with 1 < p < r of rank p and such that φM is injective and such that M has no non-zero columns. This ensures in particular that M gives a non-degenerate specialization. Note that the non-emptiness of Znonlin implies that r > 2 and T ∩ WM = ∅. Running the previous argument with this So, namely using the inductive assumption that the Monodromy Conjecture holds for 32 NERO BUDUR F M , using the specialization of supports, and taking algebraic Zariski closure, we obtain that Exp (T ) is included inSx Supp unif x 4.30. Proof of Theorem 7. Let F specialize to G via the matrix M, that is G = F M . The assumption the Topological Monodromy Conjecture holds for F implies that (ψF CX ). (cid:3) M (Exp (P L(Z top φ−1 M (Supp unif x (ψF CX)). By Proposition 3.31, the right-hand side equals eφ−1 F ))) ⊂[x [x Supp unif x (ψGCX). Hence, by (4), the Topological Monodromy Conjecture holds for G as well. Note that for (4) we need to assume that M gives a non-degenerate specialization. (cid:3) Remark 4.31. The analogs of Theorems 6 and 7 for the Strong Monodromy Con- jecture, while expected to hold, do not follow with a similar proof. The reason M (V (BF )) can fail, see Example 4.27. Note F ), namely that compatibility with specializations would place bounds on P L(Z top it would have to be included in is that the inclusion V (BF M ) ⊂ eφ−1 When r = 1, this means that P L(Z top ) should be a subset of \M eφM (V (BF M )) ∩ V (BF ). \m>0 (mV (bf m)) ∩ V (bf ). f In this section we prove Theorem 3 and Propositions 1 and 2. 5. D-modules 5.1. Explicit Riemann-Hilbert correspondence. Let DX be the sheaf of holo- morphic differential operators on the complex manifold X = Cn. Let F = (f1, . . . , fr) with 0 6= fj ∈ C[x1, . . . , xr], let f = fj, rYj=1 D = f −1(0), and let j : U = X r D → X be the natural inclusion. For α in Cr, consider the left DX-modules f αj j , rYj=1 Mα = OX[f −1] Pα = DX f αj j . rYj=1 On U, Mα and Pα define the same rank one locally free OU -module, hence an integrable connection. Let Lα be the associated rank one local system on U. Let BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 33 IC(Lα[n]) be the intersection complex on X of the perverse sheaf Lα[n], and let DR denote the De Rham functor. The following is known to experts: Theorem 5.2. (a) Mα is a regular holonomic DX-module, and DR(Mα) = DR(Pα−m·1) = Rj∗Lα[n] for integers m ≫ 0. (b) For integers m ≫ 0, Pα+m·1 is a regular holonomic simple DX-module, and DR(Pα+m·1) = IC(Lα[n]). j=1 f αj Proof. That Mα = Dxf −mQr j = Pα−m·1, for integers m ≫ 0, it is proved In loc. cit., Mα is called a Deligne module and it is in [34, Proposition 3.5]. identified with the sheaf of sections of moderate growth of j∗(OU ⊗CU Lα), For this identification, see in [5]: 6.1.10 and remark after 6.3.13. For the fact that the Deligne module Mα is a regular holonomic DX-module, see [5, 5.3.8]. The unique object of Db r.h.(DX) such that its image under DR is Rj∗Lα[n] is called B+(U, Lα) in [5, 5.5.5]. Note that in loc. cit. the definition of perverse sheaves is shifted by [−n] compared with the commonly accepted definition of today. The identification of B+(U, Lα) with Mα is in two steps. Firstly, B+(U, Lα) is the D-module direct image, under a log resolution Y of (X, D) keeping U intact, of the Deligne module attached to U and Lα on Y , see [5, 5.5.28]. Secondly, this direct image is the Deligne module attached to U and Lα on X, see the proof of [5, 4.5.2]. This shows (a). The minimal Deligne extension Mα,⊗ of the Deligne module Mα is defined in [5, 4.4.8] and also in [18, 4.4.3]. The fact that it is regular holonomic is proved in [5, Sublemma 2, p. 211] and also in [18, 4.4.5.6]. The fact that Mα,⊗ = Pα+m·1 for integers m ≫ 0 is proved in [18, 4.4.7]. The fact that DR(Mα,⊗) = IC(Lα[n]) follows from the functoriality in the Riemann-Hilbert correspondence, see [5, 5.5.11], or see the proof of the main properties characterizing the intersection complex in [18, 4.4.4]. This shows (b). (cid:3) For a point x in X, let Ux denote a small ball around x in X, and UF,x denote, as before, the complement of D in Ux. 5.3. Proof of Theorem 3. By Proposition 3.31, Lemma 3.34, and Proposition 4.24, it is enough to restrict to the case when the fj with fj(x) = 0 define mutually distinct reduced and irreducible hypersurface germs at x. By Theorem 4, it is then enough to prove that Since It is enough to show that Exp (V (BF,x)) ⊃ [y∈D near x V (BF,x) = [y∈D near x V unif (UF,y). V (BF,y), Exp (V (BF,x)) ⊃ V unif (UF,x). 34 NERO BUDUR When the fj vanishing at x are mutually distinct reduced and irreducible hyper- surface germs at x, the restriction of the local system defined above, Lα, to UF,x is the local system with monodromy Exp (αj) around fj. Let α ∈ Cr such that Exp (α) is not in Exp (V (BF,x)). It is then enough to show that Lα is not in V(UF,x). Since H k(UF,x, Lα)∨ = H n−k(UF,x, Lα), it is enough to show Lα has trivial cohomology on UF,x. Since α + m · 1 is not in V (BF,x) for any integer m, Pα = Pα+m·1 on Ux for all integers m, see [34, Proposition 3.3]. Thus, by Theorem 5.2, on Ux we have Rj∗Lα = IC(Lα). In particular Rj∗Lα = j∗Lα, and so Rkj∗Lα = 0 for k > 0. Since Ux is very small, this means that H k(UF,x, Lα) = 0 for k > 0. When k = 0, H 0(UF,x, Lα) 6= 0 if and only if Lα is the trivial local system on UF,x. However, we have excluded this case since V (BF,x) contains the hyperplanes V (sj + 1) for all 1 ≤ j ≤ r with fj(x) = 0. Indeed, hsj + 1i = BF,y for y near x such that y is a nonsingular point of D with fj(y) = 0. (cid:3) 5.4. Proof of Proposition 1. If Conjecture 2 is true, then, by Theorem 2, Exp (V (BF,x)) is a finite union of torsion translated subtori of S∗. Since a subtorus of S∗ must be the Exp image of a linear subvariety defined over Q of S, it follows that V (BF,x) is a union of linear subvarieties defined over Q. Suppose V (BF,x) ⊃ V (α1s1 + . . . + αrsr + α) for some αj, α in Q. By [21], we know that V (BF,x) is included in a finite union V (βi1s1 + . . . βirsr + βi) [i [i with βij ∈ Q≥0 and βi ∈ Q>0. obtain that If αj 6= 0, by restricting to V (hsk k 6= ji), we {−βi/βij βij 6= 0} ⊃ {−α/αj}. In particular, α 6= 0 and has the same sign as αj. Thus we can assume αj ∈ Q≥0 for all j and α ∈ Q>0. Indeed, the only other case that can occur is when all nonzero αj and α are negative, in which case we can replace them with their absolute values. This proves the claim. (cid:3) 5.5. The converse of Theorem 3. While the converse of Theorem 3 can be phrased in terms of D-modules using Sabbah specialization for D-modules, see [40, §4 and §5], we opt for a different strategy which has the advantage of being easier to state. Let N := DX[s1, . . . , sr]f s1 1 . . . f sr r as a DX-submodule of OX [f −1, s1, . . . , sr]f s1 1 . . . f sr r . There is an injective map ∇ : N −→ N BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 35 which sends (s1, . . . , sr) to (s1 + 1, . . . , sr + 1). Restating Proposition 4.6 locally, for a point x ∈ D, one has the following D-module description on Bernstein-Sato ideals: Lemma 5.6. BF,x = AnnC[s1,...,sr](Nx/∇Nx). Lemma 5.7. The following are equivalent: (1) For all α ∈ V (BF,x), (sj − αj)Nx/∇Nx ( Nx/∇Nx; rXj=1 (2) For all maximal ideals I of C[s1, . . . , sr] containing AnnC[s1,...,sr](Nx/∇Nx), I(Nx/∇Nx) 6= Nx/∇Nx. Remark 5.8. When r = 1, the equivalent statements in the previous lemma hold. In this case, N /∇N is a holonomic DX -module, hence artinian. Thus the map s − α on N /∇N is surjective if and only if it is an isomorphism, or, in other words, if and only if α is not a root of the classical one-variable Bernstein-Sato polynomial of f . But here is another proof of this case, without using any D-module theory. Consider the following statement: (*) Let M be a module over the localization R of C[s1, . . . , sr] at the origin. Let I be the maximal ideal of R. Assume that 0 6= AnnRM 6= R. Then IM 6= M. When M is finitely generated, (*) is true due to Nakayama's Lemma. When r = 1, (*) is again true. Indeed, in this case AnnRM = I a, for some a > 0. If IM = M, then M = IM = I 2M = . . . = I aM = 0, contradicting that AnnRM 6= R. Now, after a linear change of the coordinates sj and after localization, (*) is equivalent with the statement (2) of the above lemma. Hence, we have given another, simpler, proof that the statements in the previous lemma hold in the case r = 1. Remark 5.9. One way to see where lies the difficulty with proving the equivalent statements in the previous lemma for the case r > 1 is to see why the statement (*) fails in general for non-finitely generated modules M. With the notation as in (*), let R be the localization at the origin of the affine coordinate ring of a generic line through the origin in Ar. One can reduce to the case r = 1 if the assumptions of (*) imply AnnR′(M ⊗R R′) = AnnR(M) ⊗R R′. The left-hand side always includes the right-hand side, but the other inclusion is not always true. 5.10. Proof of Proposition 2. By Proposition 3.31, Lemma 3.34, and Propo- sition 4.24, it is enough to restrict to prove the converse of Theorem 3 for the case when the fj with fj(x) = 0 define mutually distinct reduced and irreducible hypersurface germs at x. By Theorem 4, it is then enough to prove that Exp (V (BF,x)) ⊂ [y∈D near x V unif (UF,y). 36 NERO BUDUR We follow the strategy as in the case r = 1 from [5, 6.3.5]. Let α ∈ V (BF,x). Suppose that Exp (α) 6∈ V unif (UF,y) for all y ∈ D near x. Then H k(UF,y, Lα) = 0 for all k and all y ∈ D near x, where Lα is as in 5.1. In particular, Rkj∗Lα has no sections over a small ball around such y, Uy, included in a small ball around x, Ux. Hence (Rj∗Lα)Ux∩D = 0, and so Rj∗Lα = IC(Lα) on Ux. Hence Mα = Pα. In particular, Pα = Pα+m·1 for any integer m. We will show that this contradicts the assumption. Let Then ∇ induces a map Nα = N / rXj=1 (sj − αj)N . ρα : Nα+1 −→ Nα. The assumption is equivalent to the statements of Lemma 5.7. Hence locally at x (sj − αj)N /∇N ( N /∇N , rXj=1 and thus the map ρα is not surjective on the stalks at x. There is a natural commutative diagram of DX-modules ρα Nα+1 Nα Pα+1 Pα where the vertical maps are surjective. Replace, if necessary, α by α − m · 1 for some positive integer m to obtain that α ∈ V (BF,x), but α − m · 1 6∈ V (BF,x), for all m ∈ Z>0. It is possible to do so by [34, Proposition 3.2]. Note that Mα is unchanged. Then, by the local version of [34, Proposition 3.6], the right-most map gives an isomorphism locally at x . . . f αr r . Since ρα is not surjective locally at x, it follows that ∼= DX f α1 Nα 1 DXf α1+1 1 . . . f αr+1 r ( DXf α1 1 . . . f αr r , which is what was claimed. (cid:3) 6. Hyperplane arrangements In this section we give a combinatorial formula for the support of the Sabbah specialization complex of a collection of hyperplanes, prove Corollaries 2 and 3, give a different proof of the fact the Multi-Variable Monodromy Conjecture holds for hyperplane arrangements, and prove Theorem 8. / /         BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 37 6.1. Terminology. Let D = ∪r Cn. Let fj be a linear polynomial defining Dj. Let f =Qj f mj j=1Dj be a finite collection of hyperplanes in X = j with mj ≥ 1 be a possibly non-reduced polynomial such that the zero locus V (f ) is D. We call both D and f hyperplane arrangements. The hyperplane arrangement is called central if each Dj is a linear subspace of codimension one of X, that is, if f is homogeneous. A central hyperplane arrangement f is indecomposable if there is no linear change of coordinates on X such that f can be written as the product of two non-constant polynomials in disjoint sets of variables. Note that indecomposability is a property of the underlying reduced zero locus D of f . An edge is an intersection of hyperplanes Dj. An arrangement is essential if {0} is an edge. For a linear subset W ⊂ X, the restriction of D to W is the hyperplane arrangement DW given by the image of ∪Dj⊃W Dj in the vector space quotient X/W , which is defined as soon as we make a linear change of coordinates such that 0 ∈ W . Similarly, one defines the restriction fW of f to W as a polynomial map on X/W , by keeping track of the multiplicities along the hyperplanes Dj which contain W . An edge W of D is called dense if the restriction arrangement DW is indecomposable. For example, Dj is a dense edge for every j. The canonical log resolution of D is the map µ : Y → Cn obtained by composition of, in increasing order for i = 0, 1, . . . , n − 2, the blowups along the (proper transform of) the union of the dense edges of dimension i. Theorem 6.2. ( [41, Theorem 3.1]) The canonical log resolution µ : Y → X is a log resolution of f . Proposition 6.3. ([41, Proposition 2.6]) If f is a central hyperplane arrangement and V = Pn−1 − P(D), then f is indecomposable if and only if χ(V ) 6= 0. Remark 6.4. The Euler characteristic of the complement of a hyperplane arrange- ment can be determined only from the lattice of intersections of the hyperplanes in the arrangement, see [35]. Hence the previous Proposition also implies that indecomposability and density are combinatorial conditions. 6.5. Sabbah specialization complex for arrangements. From now on we use the same setup as in 3.1. Assume that fj are central hyperplane arrangements in X = Cn, not necessarily reduced, of degree dj. The following two lemmas are immediate consequences of Corollary 3.9, Propo- sition 3.25, and Proposition 6.3: Lemma 6.6. If f =Qr 1 · · · tdr j=1 fj is an indecomposable central hyperplane arrangement, r − 1) ⊂ Supp 0(ψF CX). then V (td1 Lemma 6.7. There is an equality in Lemma 6.6, if, in addition, f is reduced. Definition 6.8. We say that polynomial map F = (f1, . . . , fr) with fj ∈ C[x1, . . . , xn] splits into G · H, and that G · H is a splitting of F , if, up to a different choice of co- ordinates, there exists m with 1 ≤ m ≤ n and there are polynomials gj(x1, . . . , xm) and hj(xm+1, . . . , xn) for j = 1, . . . , r, such that not all gj are constant, not all hj are constant, and fj = gjhj. If so, we set G = (g1, . . . , gr) and H = (h1, . . . , hr). Otherwise, we say that F is does not split. We say that a splitting F = F (1) · . . . · F (l) 38 NERO BUDUR is total if each F (i) does not split. Let F = (f1, . . . , fr) be such that fj are linear forms defining mutually distinct hyperplanes. Up to multiplication by constants, a total splitting of F is unique. For an edge W of the hyperplane arrangement f =Qr FW = (f1,W , . . . , fr,W ) : X/W −→ S∗, j=1 fj, let where fj,W is the restriction of the hyperplane arrangement fj to W as defined in 6.1. More precisely, fj,W = fj X/W if fj(W ) = 0, and fj,W = 1 otherwise. Note that W is a dense edge if and only if FW does not split. For every edge, let be a total splitting of FW . If we set F (i) 1,W , . . . , f (i) r,W ), let FW = F (i) W lWYi=1 W = (f (i) d(i) j,W := deg f (i) j,W . Note that d(i) j,W is either 0 or 1. Proposition 6.9. (a) If fj are linear forms defining mutually distinct hyperplanes, then (5) Supp unif x d(i) 1,W V (ht 1 d(i) r,W r − 1 i = 1, . . . , lW i), . . . t [x∈D (ψF CX) =[W where the union is over the edges W of the hyperplane arrangement Qr particular, the codimension-one part is the zero locus in S∗ of j=1 fj. In YW  Yj : fj (W )=0 tj − 1 , where the first product is over dense edges W of f =Qj fj. Proof. First, we assume that f = Qr Then (b) Assuming Conjecture 2, formula (5) also holds for Exp (V (BF )). Supp unif x (ψF CX) = Supp unif 0 Supp unif y (ψF CX ). j=1 fj is a central hyperplane arrangement. Let us focus on the first term of the right-hand side. Let F0 = F = F (1) 0 · . . . · F (l0) 0 be a total splitting of F0, with F (i) 3.38, we have 0 defined on Xi, and X = ×l0 i=1Xi. By Proposition Supp unif 0 (ψF CX) = Supp unif 0 (ψF (i) 0 CXi). l0\i=1 [x∈D (ψF CX) ∪ [06=y∈D BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 39 Since F (i) 0 able. Hence, by Proposition 6.3, for each F (i) is does not split, the hyperplane arrangement Qr j,0 is indecompos- 0 we are in the case Lemma 6.7. Thus, j=1 f (i) Supp 0(ψF (i) 0 d(1) 1,0 CXi) = V (t 1 (l0 ) d r,0 r − 1) . . . t inside the torus ]). By the definition of the uniform support, Supp 0(ψF (i) but inside the possibly-bigger torus 0 ±d(1) 1,0 Spec (C[t 1 ±d , . . . , t r (l0) r,0 CXi) has the same equations, Spec (C[t±1 1 , . . . , t±1 r ]) = S∗. Thus, Supp unif 0 d(1) 1,0 (ψF CX ) = V (ht 1 (l0) d r,0 r − 1 i = 1, . . . , l0i). . . . t The rest of the claim follows by replacing W = 0 in the above argument with other edges W of the hyperplane arrangementQr equations of the support of ψF CX at x inside the torus j=1 fj. If f is not central, fix x ∈ D. The above argument for the central case gives the Tx = V (htj − 1 fj(x) 6= 0i) ⊂ S∗. By the definition of the uniform support, Supp unif in S∗, and the claim follows. x (ψF CX ) has the same equations (cid:3) Remark 6.10. If F = (f1, . . . , fr) is such thatQr j=1 fj is a possibly non-reduced hyperplane arrangement, then F is the specialization of a polynomial map satisfying the conditions of Proposition 6.9 up to the harmless appearance of additional con- stant polynomials, as explicited in Lemma 3.34. Since we know how the supports behave under specialization, there is no mystery then what happens in Proposition 6.9 after dropping the conditions. 6.11. Proof of Corollary 2. This is an immediate consequence of Proposition 6.9 and Theorem 3. (cid:3) 6.12. Proof of Corollary 3. We can assume that f is reduced. The general case will follow from this case in light of Remark 6.10. First, we need to clarify the notation used in the statement. With the notation as in Proposition 6.9, for an edge W of D let fW be the restriction of the hyperplane arrangement f to W as in 6.1. If then f (i) W := fW = f (i) j,W , f (i) W rYj=1 lWYi=1 40 NERO BUDUR is a total splitting of fW . Let d(i) W := deg f (i) W = d(i) j,W . rXj=1 Supp unif x (ψF CX) [x∈D Let fj for j = 1, . . . , r, be the irreducible factors of f . Specialize F = (f1, . . . , fr) j=1 fj as in Example 3.32. As in the proof of Theorem 7, to f =Qr specializes via the restriction to the diagonal t1 = . . . = tr to the set {eigenvalues of monodromy on H k(Mf,x, C)}, ⋆ = [x∈D[k where Mf,x is the Milnor fiber of f at x, see Remark 3.3. Since we have {t1 = . . . = tr} ∩ [x∈D Supp unif x d(i) W = d(i) j,W , rXj=1 (ψF CX) =[W V (htd(i) W − 1 i = 1, . . . , lW i), by Proposition 6.9. The conclusion follows from the fact that Exp (V (bf )) = ⋆, by the theorem of Malgrange and Kashiwara. (cid:3) Theorem 6.13. If each fj define a (possibly nonreduced) hyperplane arrangement in Cn, then Conjecture 4 holds. Proof. We give another proof of this result, different than the one given by Theorem 6. We deal first with the central case. For j = 1, . . . , r, let fj be central hyperplane arrangements in X = Cn. Using the canonical log resolution, we see that the polar locus P L(Z top F ) is a hyperplane sub-arrangement of ∪W PW , with PW = {aW,1s1 + . . . + aW,rsr + kW + 1 = 0}, where W varies over the dense edges of the hyperplane arrangement D, aW,j = ordW (fj), and kW = codim (W ) − 1. Fix a dense edge W , and the corresponding hyperplane PW which candidates for a component of the polar locus. Let DW , fW , fj,W be the restrictions of the hyperplane arrangements D, f , fj, respectively, to W as defined in 6.1. We have fW =Qj fj,W , where the product is over those j with fj(W ) = 0. We can assume {j fj(W ) = 0} = {1, . . . , p} for some integer p. Now, take a point x ∈ W not lying on any hyperplane in D which does not contain W . After choosing a splitting of W ⊂ Cn, we have locally around x, D = DW × W ⊂ Cn = Cn/W × W and f = fW · u, where u is a (locally) invertible function. Hence, by Lemma 6.6, V ( pYj=1 taW,j j − 1) ⊂ Supp x(ψfW,1,...,fW,pCCn/W ) ⊂ (C∗)p. BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 41 On the other hand, by the definition of the uniform support, we have V ( pYj=1 taW,j j − 1) ⊂ Supp unif x (ψF CX) ⊂ (C∗)p × (C∗)r−p = S∗. Let λ ∈ PW . Then rYj=1(cid:0)e2πiλj(cid:1)aW,j − 1 = e−2πi·codim(W ) − 1 = 0. This shows that Exp (PW ) ⊂ V (taW,1 1 which was the claim. · · · taW,r r − 1) ⊂ Supp unif x (ψf1,...,frCX), The non-central case follows as in the proof of Proposition 6.9. (cid:3) 6.14. Proof of Theorem 8. We prove the claim for the case when f =Qr j=1 fj is central, since this implies the non-central case as well. As in the proof of Theorem 6.13, the polar locus P L(Z top F ) is included in the hyperplane arrangement ∪W PW , with PW = {aW,1s1 + . . . + aW,psp + kW + 1 = 0}, where W varies over the dense edges of the hyperplane arrangement D, aW,j = ordW (fj), and kW = codim (W ) − 1. Hence, PW = {deg(f1,W )s1 + . . . deg(fp,W )sp + dim(Cn/W ) = 0}. Note that fW = Qp j=1 fW,j is indecomposable, and automatically central and es- sential. By assumption, PW is in the zero locus of the ideal BFW , where FW = (f1,W , . . . , fp,W ) as before. Now, as in the proof of Theorem 6.13, take a point x ∈ W not lying on any hyperplane in D which does not contain W . We have BFW = BF,x, and the zero locus of BF,x is included in the zero locus of BF . The claim follows. (cid:3) 7. Examples Example 7.1. Let F = (x, y, x + y, z, x + y + z). Then the product of all entries of F forms a central essential indecomposable hyperplane arrangement in C3. The Bernstein-Sato ideal BF of F is currently intractable via computer. However, Conjecture 2 predicts via Corollary 2 that, in (C∗)5, (6) Exp (V (BF )) = V (h(t1t2t3 − 1)(t3t4t5 − 1)(t1 . . . t5 − 1) (tj − 1)i). 5Yj=1 42 NERO BUDUR We can actually check this as follows. One can compute with dmod.lib [24]: B e1 B e2 B e3 B e4 B e5 F = h(s1 + 1)(s1 + s2 + s3 + 2)(s1 + s2 + s3 + s4 + s5 + 3)i, F = h(s2 + 1)(s1 + s2 + s3 + 2)(s1 + s2 + s3 + s4 + s5 + 3)i, F = h(s3 + 1)(s1 + s2 + s3 + 2)(s3 + s4 + s5 + 2)(s1 + s2 + s3 + s4 + s5 + 3)i, F = h(s4 + 1)(s3 + s4 + s5 + 2)(s1 + s2 + s3 + s4 + s5 + 3)i, F = h(s5 + 1)(s3 + s4 + s5 + 2)(s1 + s2 + s3 + s4 + s5 + 3)i. Then (6) follows from Lemma 4.14 which says that Exp (V (BF )) =S5 Using a different computation, U. Walther has also checked that (6) holds. The local topological zeta function Z top F,0 of F at the origin has a degree-7 irre- j=1 Exp (V (B ej F )). ducible numerator, and the denominator is equal to 5Yj=1 (s1 + s2 + s3 + 2)(s3 + s4 + s5 + 2)(s1 + . . . + s5 + 3) (sj + 1). This illustrates the (local version of the) Multi-Variable Monodromy Conjecture which we proved for hyperplane arrangements. This also shows that Conjecture 3 and the (local version of the) Multi-Variable Strong Monodromy Conjecture hold for F . Example 7.2. Conjecture 3, which we proved to imply the Multi-Variable Strong Monodromy Conjecture for hyperplane arrangements, can fail for decomposable arrangements. Let F = (x, y, x + y, z). Then the product of all entries of F forms a central essential, but decomposable, hyperplane arrangement. The hyperplane s1 + . . . + s4 + 3 = 0 does not lie in the zero locus of the Bernstein-Sato ideal BF = h(s1 + 1)(s2 + 1)(s3 + 1)(s4 + 1) (s1 + s2 + s3 + k)i. 4Yk=2 Example 7.3. If F is as in Example 4.20, the multi-variable local topological zeta function at the origin is Z top F,0 (s) = 2s1s2 + s2 + 1 (s1 + 1)(s2 + 1)(2s2 + 1) . This illustrates the (local version of the) Multi-Variable Strong Monodromy Con- jecture. References [1] D. Arapura, Higgs line bundles, Green-Lazarsfeld sets, and maps of Kahler manifolds to curves. Bull. Amer. Math. Soc. 26 (1992), no. 2, 310 -- 314. 2 [2] D. Arapura, Geometry of cohomology support loci for local systems. I. J. Algebraic Geom. 6 (1997), no. 3, 563 -- 597. 2 [3] R. Bahloul, D´emonstration constructive de l'existence de polynomes de Bernstein-Sato pour plusieurs fonctions analytiques. Compos. Math. 141 (2005), no. 1, 175 -- 191. 2, 23 [4] R. Bahloul and T. Oaku, Local Bernstein-Sato ideals: algorithm and examples. J. Symbolic Comput. 45 (2010), no. 1, 46 -- 59. 2, 3, 23, 27 BERNSTEIN-SATO IDEALS AND LOCAL SYSTEMS 43 [5] J.-E. Bjork, Analytic D-modules and applications. Mathematics and its Applications, 247. Kluwer Academic Publishers Group, Dordrecht, 1993. xiv+581 pp. 33, 36 [6] E. Bombieri and W. Gubler, Heights in Diophantine geometry. New Mathematical Mono- graphs, 4. Cambridge University Press, Cambridge, 2006. xvi+652 pp. 29 [7] J. Brian¸con, P. Maisonobe, and M. Merle, Constructibilit´e de l'id´eal de Bernstein. Singularities -- Sapporo 1998, 79 -- 95, Adv. Stud. Pure Math., 29, Kinokuniya, Tokyo, 2000. 3 [8] J.-L. Brylinski, Transformations canoniques, dualit´e projective, th´eorie de Lefschetz, trans- formations de Fourier et sommes trigonom´etriques. G´eom´etrie et analyse microlocales. Ast´erisque No. 140 -- 141 (1986), 3 -- 134, 251. 12 [9] N. Budur, Unitary local systems, multiplier ideals, and polynomial periodicity of Hodge numbers. Adv. Math. 221 (2009), 217 -- 250. 2 [10] N. Budur, M. Mustat¸a, and Z. Teitler, The monodromy conjecture for hyperplane arrange- ments. Geom. Dedicata 153 (2011), no. 1, 131 -- 137. 7, 8, 9 [11] N. Budur and B. Wang, Cohomology jump loci of quasi-projective varieties. arXiv:1211.3766. 1, 10 [12] P. Cassou-Nogu`es and A. Libgober, Multivariable Hodge theoretical invariants of germs of plane curves. J. Knot Theory Ramifications 20 (2011), no. 6, 787 -- 805. 2 [13] W. Decker, G.-M. Greuel, G. Pfister, and H. Schonemann, Singular 3-1-3 -- A computer algebra system for polynomial computations. http://www.singular.uni-kl.de (Feb 2010). 2, 10 [14] A. Dimca, Singularities and topology of hypersurfaces. Universitext. Springer-Verlag, New York, 1992. xvi+263 pp. 10, 13 [15] A. Dimca and L. Maxim, Multivariable Alexander invariants of hypersurface complements. Trans. Amer. Math. Soc. 359 (2007), no. 7, 3505 -- 3528. 6, 17, 18 [16] A. Dimca and S¸. Papadima, Nonabelian cohomology jump loci from an analytic viewpoint. arXiv:1206.3773. 2 [17] A. Dimca, S¸. Papadima, and A. Suciu, Topology and geometry of cohomology jump loci. Duke Math. J. 148 (2009), no. 3, 405 -- 457. 2 [18] V. Ginzburg, Lectures on D-modules. With collaboration of V. Baranovsky and S. Evens, 1998. 33 [19] M. Green and R. Lazarsfeld, Deformation theory, generic vanishing theorems, and some conjectures of Enriques, Catanese and Beauville. Invent. Math. 90 (1987), no. 2, 389 -- 407. 2 [20] M. Green and R. Lazarsfeld, Higher obstructions to deforming cohomology groups of line bundles. J. Amer. Math. Soc. 4 (1991), no. 1, 87 -- 103. 2 [21] A. Gyoja, Bernstein-Sato's polynomial for several analytic functions, J. Math. Kyoto Univ. 33 (1993), 399 -- 411. 2, 23, 24, 34 [22] M. Kashiwara, B-functions and holonomic systems. Rationality of roots of B-functions. Invent. Math. 38 (1976/77), no. 1, 33 -- 53. 2 [23] M. Kashiwara, Vanishing cycle sheaves and holonomic systems of differential equations. Algebraic geometry (Tokyo/Kyoto, 1982), 134 -- 142, Lecture Notes in Math., 1016, Springer, Berlin, 1983. 3 [24] V. Levandovskyy and J. Mart´ın Morales, dmod.lib. A Singular 3-1-3 library for algebraic D-modules (2011). 2, 10, 26, 27, 29, 42 [25] M. Laurent, ´Equations diophantiennes exponentielles. Invent. Math. 78 (1984), no. 2, 299 -- 327. 3 [26] A. Libgober, Eigenvalues for the monodromy of the Milnor fibers of arrangements. Trends in singularities, 141 -- 150, Birkhauser, Basel, 2002. 7 [27] A. Libgober, Non-vanishing loci of Hodge numbers of local systems. Manuscripta Math. 128 (2009), no. 1, 1 -- 31. 2, 10 [28] F. Loeser, Fonctions d'Igusa p-adiques et polynomes de Bernstein, Amer. J. Math. 110 (1988), no. 1, 1 -- 21. 8 44 NERO BUDUR [29] F. Loeser, Fonctions zeta locales d'Igusa `a plusieurs variables, int´egration dans les fibres, et discriminants. Ann. Sci. ´Ecole Norm. Sup. (4) 22 (1989), no. 3, 435 -- 471. 8 [30] B. Malgrange, Polynomes de Bernstein -- Sato et cohomologie ´evanescente. Analysis and topol- ogy on singular spaces, II, III (Luminy, 1981), 243 -- 267, Ast`erisque, 101-102, Soc. Math. France, Paris, 1983. 3 [31] A. N´emethi and W. Veys, Generalized Monodromy Conjecture in dimension two. Geometry & Topology 16 (2012), 155 -- 217. 9 [32] J. Nicaise, Zeta functions and Alexander modules. math.AG/0404212. 4, 8, 12 [33] T. Oaku, An algorithm of computing b-functions. Duke Math. J. 87 (1997), 115 -- 32. 9 [34] T. Oaku and N. Takayama, An algorithm for de Rham cohomology groups of the comple- ment of an affine variety via D-module computation. Effective methods in algebraic geometry (Saint-Malo, 1998). J. Pure Appl. Algebra 139 (1999), no. 1-3, 201 -- 233. 10, 33, 34, 36 [35] P. Orlik and H. Terao, Arrangements of hyperplanes. Springer-Verlag, Berlin, 1992. xviii+325 pp. 37 [36] S¸. Papadima and A. Suciu, Bieri-Neumann-Strebel-Renz invariants and homology jumping loci. Proc. Lond. Math. Soc. (3) 100 (2010), no. 3, 795 -- 834. 11, 22 [37] M. Popa and C. Schnell, Generic vanishing theory via mixed Hodge modules. arXiv:1112.3058. 2 [38] B. Rodrigues and W. Veys, Holomorphy of Igusa's and topological zeta functions for homo- geneous polynomials. Pacific J. Math. 201 (2001), no. 2, 429 -- 440. 13 [39] C. Sabbah, Proximit´e ´evanescente. I. La structure polaire d'un D-module. Compositio Math. 62 (1987), no. 3, 283 -- 328. Proximit´e ´evanescente. II. ´Equations fonctionnelles pour plusieurs fonctions analytiques. ibid. 64 (1987), no. 2, 213 -- 241. 2, 23 [40] C. Sabbah, Modules d'Alexander et D-modules. Duke Math. J. 60 (1990), no. 3, 729 -- 814. 4, 6, 12, 13, 14, 15, 19, 20, 34 [41] V. Schechtman, H. Terao and A. Varchenko, Local systems over complements of hyperplanes and the Kac-Kazhdan conditions for singular vectors. J. Pure Appl. Algebra 100 (1995), 93 -- 102. 37 [42] C. Simpson, Subspaces of moduli spaces of rank one local systems. Ann. Sci. ´Ecole Norm. Sup. (4) 26 (1993), no. 3, 361 -- 401. 2 KU Leuven, Department of Mathematics, Celestijnenlaan 200B, B-3001 Leuven, Belgium E-mail address: [email protected] University of Notre Dame, Department of Mathematics, 255 Hurley Hall, IN 46556, USA E-mail address: [email protected]
0902.0601
3
0902
2010-05-11T19:35:01
Symplectic automorphisms and the Picard group of a K3 surface
[ "math.AG" ]
We consider the symplectic action of a finite group G on a K3 surface. The Picard group of the K3 surface has a primitive sublattice determined by G. We show how to compute the rank and discriminant of this sublattice. We then describe moduli spaces of K3 surfaces with symplectic G-action, extending results of Nikulin in the abelian case. We use our moduli spaces to develop techniques for classifying all possible symplectic actions of a group G.
math.AG
math
Symplectic Automorphisms and the Picard Group of a K3 Surface Ursula Whitcher ∗ 1 Introduction Let X be a K3 surface, and let G be a finite group acting on X by automorphisms. The action of G on X induces an action on the cohomology of X. We assume G acts symplectically: that is, G acts as the identity on H 2,0(X). In this case, the minimum resolution Y of the quotient X/G is itself a K3 surface. In [N80a], Nikulin classified the finite abelian groups which act symplectically on K3 surfaces by analyzing the relationship between X and Y . Nikulin also described moduli spaces of K3 surfaces with G actions for the case that G is an abelian group; these topological spaces are subspaces of the moduli space of marked K3 surfaces. Mukai showed in [M88] that any finite group G with a symplectic action on a K3 surface is a subgroup of a member of a list of eleven groups, and gave an example of a symplectic action of each of these maximal groups. Xiao gave an alternate proof of the classification in [X96] by listing the possible types of singularities, and Kond¯o showed in [K98] that the action of G on the K3 lattice extends to an action on a Niemeier lattice. The Picard group of X has a primitive sublattice SG determined by the action of G. The rank of SG varies from 8 to 19, depending on G. Thus, K3 surfaces which admit symplectic group actions provide a rich source of examples of families of K3 surfaces with high-rank Picard groups. The monodromy and mirror symmetry properties of algebraic K3 surfaces which admit a sublattice SG of rank 18, and therefore have a Picard group of rank 19, have been extensively studied. (cf. [N01, S07, DK08]) Conversely, if the structure of Pic(X) is known, one may examine its sublattices to detect symplectic group actions on X. Morrison used the structure of SG for G = Z/2Z to study K3 surfaces which admit Shioda-Inose structures in [M84]. Recently, Garbagnati and Sarti have computed SG for all possible abelian groups with symplectic action, correcting an earlier computation of Nikulin's; Garbagnati has also studied SG for dihedral groups, and Hashimoto calculated the invariants of SG for the permutation group G = S5 (see [GS08, G08a, G08b, G09, H09]). In Section 2, we discuss the relationship between the lattice SG and the singularities of X/G for any symplectic G-action, and show how to compute the rank and discriminant of SG. In Section 3, we show that the maps between X, Y , and X/G can be generalized to the realm of moduli spaces, and describe moduli spaces of K3 surfaces with symplectic G-action. The key observation is that we may work backwards from a K3 surface Y endowed with a set of exceptional curves to the K3 surface X. We use our moduli spaces to develop techniques for classifying all possible symplectic actions of a group G. 2 A Sublattice of the Picard Group The cup product induces a bilinear form h , i on H 2(X, Z) ∼= H ⊕ H ⊕ H ⊕ E8 ⊕ E8. (We take E8 to be negative definite.) Using this form, we define SG = (H 2(X, Z)G)⊥. The Picard group of X, Pic(X), consists of H 1,1(X) ∩ H 2(X, Z); the group T (X) ⊆ H 2(X, Z) of transcendental cycles is defined as (Pic(X))⊥. Nikulin showed that the groups Pic(X) and SG are related: Proposition 2.1. [N80a, Lemma 4.2] SG ⊆ Pic(X) and T (X) ⊆ H 2(X, Z)G. The lattice SG is nondegenerate and negative definite. ∗I thank the referee for comments which improved the exposition of the paper, Alice Garbagnati, Paul Hacking, and Kenji Hashimoto for useful discussion, Charles Doran for his generous guidance, and the organizers of the 2007 GAeL conference in Istanbul, where I presented an early version of this work, for their support. Further, I gratefully acknowledge partial support in the preparation of this work by the National Science Foundation under the Grant DMS-083996. 1 In this section, we show how to compute the rank and discriminant of SG, and relate SG to the singularity structure of X/G. The number of fixed points of an element g of a group G acting symplectically on a K3 surface X depends only on the order of g, by the results of [N80a, §5]. Proposition 2.2. [M88, §3] [O03, Proposition 4.5] Let m(n) be the number of elements in G of order n, and let f (n) be the number of fixed points of an element of order n. Then, rank H ∗(X, Z)G = 1 G (24 + 8 X n=2 m(n)f (n)). Since G acts as the identity on H 0(X, Z) and H 4(X, Z) as well as H 2,0(X) and H 0,2(X), we also know that rank H ∗(X, Z)G ≥ 4. Because G acts symplectically on X, X/G has a minimal resolution Y which is also a K3 surface. Let {pi} be the singular points of X/G. The inverse image in Y of pi is a configuration Ψi of (−2)-curves of type Al, Dm or En; let ci be the number of curves in this configuration. The configurations Ψi generate a lattice K in Pic(Y ) ⊂ H 2(Y, Z) of rank Pi ci. Let M be the minimal primitive sublattice of H 2(Y, Z) containing K. Then M also has rank Pi ci, and H 2(Y, Z)/M is a free abelian group. Xiao showed in [X96, Lemma 6] that M is uniquely determined by the Ψi. Remark 2.1. [X96, Theorem 3] If G is isomorphic to Q8, the group of unit quaternions, or T24, the binary tetrahedral group of order 24, then K may be one of two different lattices, depending on the action of G. In all other cases, K (and thus M ) is uniquely determined by G. Let {qij} be the inverse images in X of pi, and let Gi be the stabilizer group of any qij ; set Ni = Gi. Proposition 2.3. [X96, Lemma 1] X ci = i 24(G − 1) G − k X i=1 Ni − 1 Ni . Proposition 2.4. rank SG = Pi ci. Nikulin discusses this proposition for the case that G is abelian in [N80a, §10]. We use Propositions 2.2 and 2.3 to give a brief proof for any G. Proof. We calculate: rank H ∗(X, Z)G + X ci = 24 − rank SG + X ci i i = 1 G (24 + 8 X n=2 m(n)f (n)) + 24(G − 1) G − k X i=1 Ni − 1 Ni = 24 + 1 G 8 X n=2 m(n)f (n) − k X i=1 Ni − 1 Ni . Thus, it suffices to show that 8 X n=2 m(n)f (n) = k X i=1 G Ni (Ni − 1). P8 n=2 m(n)f (n) counts each non-identity element g of G once for each point of X which g fixes. Ni − 1 preimages qij in X; (Ni − 1) counts the non-identity elements of the stabilizer group Gi. The point pi has precisely G by definition, the elements of Gi fix the qij. Summing over all singular points pi, we see that Pk also counts every element of G other than the identity once for each point of X which that element fixes. Ni G Ni i=1 2 Though the lattices SG and M are primitive sublattices of the K3 lattice H ⊕ H ⊕ H ⊕ E8 ⊕ E8 and have the same rank, they are not isomorphic: by [N80a, Lemma 4.2], SG contains no elements with square −2. Instead, the relationship between SG and M is given by the fact that SG = (H 2(X, Z)G)⊥ and the following exact sequence. Theorem 2.1. There exists an exact sequence 0 −−−−→ M/K −−−−→ H 2(Y, Z)/K θ−−−−→ (H 2(X, Z))G −−−−→ H 3(G, Z) −−−−→ 0 where hθ(m), θ(n)i = G hm, ni. Proof. Let X ′ = X − (∪i,jqij ) and let Y ′ = Y − (∪iΨi). Since X is a simply connected complex surface, X ′ is also simply connected; since Y ′ = X ′/G, X ′ is the universal covering space of Y ′. By [EC56, Application XVI.1], there exists an exact sequence 0 −−−−→ H 2(G, Z) −−−−→ H 2(Y ′, Z) θ−−−−→ (H 2(X ′, Z))G −−−−→ H 3(G, Z) ζ −−−−→ H 3(Y ′, Z). Since θ is induced by the quotient map X ′ → Y ′, hθ(m), θ(n)i = G hm, ni. Xiao showed in[X96, Lemma 2] that H 2(G, Z) = M/K and H 2(Y ′, Z) = H 2(Y, Z)/K; because X is a complex surface (and therefore has four real dimensions), H 2(X, Z) = H 2(X ′, Z). Since G is a finite group, H 3(G, Z) is a finite abelian group. We shall show that H 3(Y ′, Z) is a free abelian group, so ζ must be the zero map. Let Ni be a tubular neighborhood of the configuration of exceptional curves Ψi in Y , and let Li be the boundary of Ni. Consider the Mayer-Vietoris sequence . . . −−−−→ H 3(Y, Z) −−−−→ H 3(Y ′, Z) ⊕ Li H 3(Ni, Z) −−−−→ Li H 3(Li, Z) −−−−→ H 4(Y, Z) −−−−→ . . . Since Y is a K3 surface, H 3(Y ) = 0 and H 4(Y ) = Z. Because Ni is a tubular neighborhood of an ADE configuration of curves, Ni is homotopy equivalent to a bouquet of ci 2-spheres, so H 3(Ni) = 0. Since Li is a smooth real 3-manifold, H 3(Li) = Z. Furthermore, the map Li H 3(Li) → H 4(Y ) is given by f : Zk → Z, where f ((x1, . . . , xci )) = x1 + .. + xk. Thus, H 3(Y ′) is isomorphic to the kernel of f , a free abelian group of rank k − 1. Remark 2.2. In [G08a, Proposition 2.4], Garbagnati proved a variant of Theorem 2.1 in the case that G is an abelian group, correcting Nikulin's claim that θ is surjective. Lemma 2.1. [N80a, Lemma 10.2] Let J = Im(θ). Then the lattice discriminants d(J) and d(M ) are related by d(J) = − G22−rank(M) d(M ) . Example 2.1. Let X be a K3 surface which admits a symplectic action by the permutation group G = S4. Then Pic(X) admits a primitive sublattice SG which has rank 17 and discriminant d(SG) = −26 · 32. Proof. According to [X96, Table 2], when G = S4, K is the rank 17 lattice given by (A3)2 ⊕ (A2)3 ⊕ (A1)5, and M/K ∼= Z/(2Z). Next we use the fact that if lattices L and L′ have the same rank, and L ⊂ L′, then the discriminants d(L) and d(L′) are related by d(L)/d(L′) = [L′ : L]2, where [L′ : L] is the index of L in L′ as an abelian group. Since d(K) = −29 · 33, we see that d(M ) = −27 · 33. By Lemma 2.1, the discriminant d(J) = 28 · 32. The cohomology group H 3(S4, Z) is isomorphic to Z/(2Z), so [(H 2(X, Z))G : J] = 2 and d((H 2(X, Z))G) = 26 · 32. Since SG is the perpendicular complement of (H 2(X, Z))G in the unimodular K3 lattice H ⊕ H ⊕ H ⊕ E8 ⊕ E8, we conclude that d(SG) = −d((H 2(X, Z))G) = −26 · 32. Example 2.2. Let X be a K3 surface which admits a symplectic action by the Chevalley group G = L2(7) ∼= P SL(2, F7). Then (H 2(X, Z))G has rank 3 and discriminant 196. 3 Proof. Consulting [X96, Table 2], we find that K is the rank 19 lattice given by A6 ⊕ (A3)2 ⊕ (A2)3 ⊕ A1, and M ∼= K. Thus, d(M ) = −7 · 42 · 33 · 2. The order of L2(7) is 23 · 3 · 7, so by Lemma 2.1, the discriminant d(J) = 24 · 7. We may use the computer algebra system [SAGE] to show that H 3(G, Z) ∼= Z/2Z. Thus, [(H 2(X, Z))G : J] = 2, so d(H 2(X, Z)) = (24 · 72)/22 = 196. Remark 2.3. The result of Example 2.2 is the "Key Lemma" of [OZ02]; that paper gives a longer and more involved proof by constructing an embedding of SG in a Niemeier lattice. 3 Classifying Symplectic Group Actions In [N80a], Nikulin showed that, when G is abelian, symplectic actions of G are unique up to overall isomor- phisms. In this section, we develop techniques for classifying the symplectic actions of any group, and show that certain non-abelian groups admit multiple distinct symplectic actions. To do so, we construct moduli spaces of K3 surfaces which can be realized as resolutions of quotients by symplectic group actions. Our discussion extends and refines the constructions of [N80a] in the non-abelian case. We begin by reviewing the standard constructions of moduli spaces of K3 surfaces. We follow the exposition and notation of [BHPV04, §VIII]. We call a choice of isomorphism α : H 2(X, Z) → L a marking of X, and refer to the pair (X; α) as a marked K3 surface. Let us write h , i for the bilinear form on L; we set LR = L ⊗ R and LC = L ⊗ C, and extend the bilinear form appropriately. For any nonzero element ω of LC, let [ω] be the corresponding element of the projective space P(LC). Let Ω = {[ω] ∈ P(LC) hω, ωi = 0, hω, ¯ωi > 0}. Let (X; α) be a marked K3 surface, and let ωX be a nowhere- vanishing holomorphic two-form on X. (The form ωX is unique up to a scalar multiple.) The image of ωX under αC determines a point [αC(ωX )] in P(LC). Since ωX ∧ ωX = 0 and ωX ∧ ¯ωX > 0, [αC(ωX )] is an element of Ω, which we refer to as the period point. There exists a universal marked family of K3 surfaces. The base space N is a non-Hausdorff "smooth analytic space" of dimension 20. The period points of marked K3 surfaces yield a period map τN : N → Ω. We now consider marked K3 surfaces with specified Kahler class. We wish to specify the Kahler class in a manner consistent with our marking. For any [ω] ∈ Ω, let E(ω) be the oriented 2-plane in LR spanned by {Re ω, Im ω}. Let KΩ be the set {(κ, [ω]) ∈ LR × Ω hκ, λi = 0 ∀ λ ∈ E(ω) and hκ, κi > 0}. Then KΩ is a fiber bundle over Ω. For any [ω] ∈ Ω, let Cω be the cone {x ∈ LR hx, ωi = 0, hx, xi > 0}. We may choose a connected component C+ ω varies continuously with our choice of [ω]. If (X; α) is a marked K3 surface and κ ∈ H 1,1(X) a Kahler class, we say that (X, κ) is a marked pair if αC(κ) ∈ C+ ω , where [ω] is the period point of X. There exists a universal object M for marked pairs and a natural forgetful map M → N . The space M is a 60-dimensional real-analytic manifold. ω of Cω in such a way that C+ Let (KΩ)0 be the subset of KΩ consisting of those points (κ, [ω]) such that hκ, di 6= 0 for every d ∈ L such that hd, di = −2 and hω, di = 0. The subset (KΩ)0 is open in KΩ. We may define a real-analytic map τM : M → (KΩ)0 called the refined period map as follows: if m ∈ M corresponds to the marked pair (X, κ), we set τM(m) = (αC(κ), [ω]). The period map τN and the refined period map τM together with the forgetful maps fit into a commutative diagram: M τM−−−−→ (KΩ)0 y y τN−−−−→ Ω N Theorem 3.1. (See [BHPV04, Theorem VIII.12.3] for a modern proof.) The refined period map τM is injective. We also have a surjectivity result due to Todorov (see [T80]): Theorem 3.2. (cf. [BHPV04, Theorem VIII.14.1]) The refined period map τM is surjective. Remark 3.1. In [N80a], Nikulin uses a moduli space of Kahler K3 surfaces which has two components; by working with marked pairs, we have essentially fixed our choice of component. 4 We now describe moduli spaces which will parametrize possible resolutions Y . Definition 3.1. [N80a, Definition 2.1] A condition T is a primitive sublattice M of L and a finite subset {ci} of M such that c2 i = −2 for each i. We work with conditions T where M is negative definite. Definition 3.2. [N80a, Definition 2.2] A marked K3 surface with condition T is a K3 surface Y together with an isometry α : H 2(X, Z) → L such that α−1(M ) ⊂ H 1,1(Y ) and α−1(ci) is represented by a nonsingular rational curve on Y for each i. Remark 3.2. A nonsingular rational curve with self-intersection −2 in a K3 surface is uniquely determined by its homology class. (cf. [BHPV04, Proposition VIII.3.7]) We will often identify the cohomology classes α−1(ci) with the corresponding curves. Definition 3.3. A marked pair with condition T is a marked pair (Y, κ) such that Y is a marked K3 surface with condition T . Any marked pair with condition T must satisfy hκ, α−1(ci)i > 0 for each i, because α−1(ci) is represented by a nonsingular rational curve. Definition 3.4. Let MT be the subspace of M corresponding to the marked pairs with condition T = {cj} ⊂ M ⊂ L. Let NT be the image of MT under the forgetful map. Remark 3.3. Note that by taking M to be isomorphic to a primitive sublattice of H 2(Y, Z), we have fixed the primitive embedding of M in L up to automorphisms of L. Let (KΩ)0 M be the subset of (KΩ)0 given by the refined period points (κ, [ω]) such that M is contained in the perpendicular complement of ω and ¯ω. Suppose m ∈ M corresponds to a marked K3 surface (Y ; α) with Kahler class κ, and suppose τM(m) = (κ, [ω]). Let ∆m be the set given by {δ ∈ αC(H 1,1(X)) hδ, δi = −2}, M such that ci ∈ ∆+ and let ∆+ and ci is an irreducible element of ∆+ Proposition 3.1. [N80a, Proposition 2.8] The point m ∈ MT if and only if τM(m) ∈ (KΩ)0 T . m be the subset of ∆m given by hκ, δi > 0. Let (KΩ)0 m for each i. The following proposition follows immediately. T be the subset of (KΩ)0 m Proposition 3.2. [N80a, Proposition 2.9] Let M be a negative definite lattice with rank M ≤ 19. Then (KΩ)0 M − (KΩ)0 M which is the union of at most countably many closed complex subspaces of (KΩ)0 M is a closed smooth complex subspace of (KΩ)0. Furthermore, (KΩ)0 T is a closed subset of (KΩ)0 M is connected, and (KΩ)0 M . Theorem 3.3. MT is path-connected. Remark 3.4. Nikulin proved a variant of Theorem 3.3 under the assumption that rank M ≤ 18 by construct- ing paths between elements (see [N80a, Theorem 2.10]). We give a quick argument using the bijectivity of the refined period map τM. Proof. Proposition 3.2 implies that (KΩ)0 τM is injective and surjective, Proposition 3.1 implies that MT is also path-connected. T is connected and path-connected. Since the refined period map Corollary 3.1. NT is connected and path-connected. Let G be a group which admits a symplectic action on some marked K3 surface (Xν ; βν). Let G be the group which has the same elements of G, but where the group operation is written in the reverse order. Then G acts on H 2(X, Z), and we may use β0 to define an embedding φ : G ֒→ O(L). In the following discussion, we treat the group G and the embedding φ as fixed. Definition 3.5. [N80a, Definition 4.9] A marked K3 surface with symplectic automorphism group G and action φ on the integral cohomology is a triple (X, i; β) such that (X; β) is a marked K3 surface, i : G ֒→ Aut(X) is an embedding where G acts symplectically on X, and β · i(g)∗ · β−1 = φ(g) for any g ∈ G. We say that two such triples (X, i; β) and (X ′, i′; β′) are isomorphic if there exists an isomorphism t : X → X ′ such that β′ = β · t∗ and i′(g) = t · i(g) · t−1 for any g ∈ G. 5 Definition 3.6. A marked pair with symplectic automorphism group G and action φ is a triple (X, i; β) which is a marked K3 surface with symplectic automorphism group G and action φ together with a Kahler class κ such that (X, κ) is a marked pair. We say that two such pairs (X, κ) and (X ′, κ′) are isomorphic if there exists an isomorphism t : X → X ′ such that t∗(κ′) = κ and the underlying triples (X, i; β) and (X ′, i′; β′) are isomorphic. Definition 3.7. Let MSG be the subspace of M corresponding to the marked pairs with the condition R given by {} ⊂ SG ⊂ L. Let NSG be the image of MSG under the forgetful map. Given a marked pair with symplectic automorphism group G and action φ, we may obtain a marked pair with condition R. The global Torelli theorem for K3 surfaces implies that two marked pairs with symplectic automorphism group G and action φ correspond to the same marked pair with condition R if and only if they are isomorphic. Let MG,φ be the subspace of MSG corresponding to marked pairs with symplectic automorphism group G and action φ; by [N80a, Theorem 4.10], MG,φ is open in MSG. Let NG,φ be the image of MG,φ under the forgetful map, and let XG,φ be the subset of the universal family X of marked K3 surfaces lying over MG,φ. Consider the minimal resolution Yν of the quotient Xν/G. Let ∆ = {Ci} be the exceptional curves of Yν. Definition 3.8. [SZ01, §4] We say that a simple normal crossing divisor ∆ on a K3 surface Y is an ADE configuration of smooth rational curves, or, more briefly, an ADE configuration, if each irreducible component of ∆ is a smooth rational curve and the intersection matrix of the irreducible components of ∆ is a direct sum of the Cartan matrices of type Al, Dm or En. (We take these matrixes to be negative definite.) Fix a marking αν of Yν. We may use the image αν(∆) of the ADE configuration ∆ to define a condition T ; the marked K3 surface (Yν ; αν) corresponds to a point in NT . We wish to show that all points of NT correspond to resolutions of symplectic quotients of K3 surfaces. We will need the following classification of the covering spaces of the complements of ADE configurations of rational curves on K3 surfaces: Theorem 3.4. [SZ01, Proposition 4.1] [C04] Let ∆ be an ADE configuration of smooth rational curves on a K3 surface Y . Let Y ′ = Y − ∆, and let X ′ be the universal covering space of Y ′. Then X ′ and π1(X ′) satisfy one of the following conditions: 1. X ′ ∼= Y ′ and π1(Y ′) is trivial. 2. X ′ is isomorphic to the complement of a discrete set of points A in C2, and π1(Y ′) is infinite. Further- more, there exists a map f from C2 − A to a two-dimensional complex torus T and a map g from T to Y ′ such that g is the quotient of T by a finite group of automorphisms Γ and g ◦ f is the covering map. 3. X ′ is isomorphic to a K3 surface with a finite set of points removed, and the group of covering transfor- mations (which is naturally isomorphic to π1(Y ′)) acts symplectically on this surface. Theorem 3.5. Suppose there exists ν ∈ NT , corresponding to a marked K3 surface (Yν , αν ), such that Yν is the resolution of the quotient of a K3 surface Xν by a symplectic G-action. Let q ∈ NT , and let (Yq, αq) be the corresponding marked K3 surface. Then there exists a K3 surface Xq and a symplectic action of G on Xq such that Yq is a resolution of Xq/G. Proof. For any n ∈ NT , we may choose a neighborhood Un of n such that for all n′ in Un, there exists a diffeomorphism α : Yn → Yn′ such that α∗(α−1 n′ (M )) = α−1 n (M ) and (since rational curves in K3 surfaces are uniquely determined by their homology classes) α∗(α−1 n′ ({cj})) = α−1 m ({cj}) and Yn′ − α−1 n′ ({cj}) are isomorphic, and π1(Yn − α−1 n ({cj})) = π1(Yn′ − α−1 n ({cj}). Thus Yn − α−1 By Theorem 3.3, there exists a path in NT from q to ν. Covering this path by a finite number of the q ({cj})) = q ({cj}) is isomorphic to a K3 surface Xq with a finite number neighborhoods Un, we see that π1(Yq −α−1 G. By Theorem 3.4, the covering space of Yq −α−1 of points removed, and G acts symplectically on Xq. Thus, Yq is the resolution of Xq/G, as desired. q ({cj})) is isomorphic to π1(Yµ −α−1 µ ({cj})), so π1(Yq −α−1 n′ ({cj})). Starting with Yq, we obtained a pair (Xq, iq : G ֒→ Aut X). 6 Definition 3.9. We say that two points n, n′ ∈ NT determine the same action of G on the two-dimensional integral cohomology of K3 surfaces if there exist corresponding pairs (Xn, in : G ֒→ Aut X), (Xn′ , in′ : G ֒→ Aut X) and an isomorphism φ : H 2(Xn, Z) → H 2(Xn′ , Z) which preserves the cup product and satisfies the relation in′ (g)∗ = φ · in(g) · φ−1 for any g ∈ G. The condition that points determine the same action of G defines an equivalence relation on NT . Theorem 3.6. Let n0 ∈ NT , and suppose n0 corresponds to the pair (Xn, in : G ֒→ Aut G). The set of points in NT which determine the same action of G as n0 is open. Proof. We construct an open neighborhood of n0 in which the action coincides with the action determined by n0 and a corresponding neighborhood in NG,φ. Fix a marking βn0 : H 2(X, Z) → L. The triple (Xn0, in0, βn0 ) defines a point ν0 in the moduli space NG,φ. The usual map u : X → N restricts to a map uG,φ : XG,φ → NG,φ. Following [N80a, §8.5], we obtain a neighborhood V of ν0 in NG,φ, a corresponding neighborhood X V G,φ in XG,φ, and a resolution Y V G,φ/G such that the following diagram commutes: G,φ of X V X V G,φ π ............ ............................................................................................................................................................................... ............................................................................................................................................................................................................................................... ............ uG,φ ........................................................................................ ............ V X V G,φ/G Y V G,φ σ ............................................................................................................................................................................................ ............ ............................................................................................................................................................................................................................................... v ............ Each curve Ej in Yn0 extends uniquely to an effective divisor Ej on Y V G,φ. For each ν ∈ V , Ej · Yν = Eν j j } is the set of components of the curves obtained from the G,φ = XG,φ − {fixed points of G} and Y ′ G,φ = YG,φ − ∪Ej, is a nonsingular rational curve on Yn, where {Eν resolution of singularities of Xν/G. We set X ′ obtaining a new commutative diagram: ............π′ ........................................................................................ X V G,φ ............ ........................................................................... (X V G,φ)′ ............................................................................................................................................................................................................................................... ............ ........................................................................................................................................ ............ uG,φ u′ G,φ (Y V G,φ)′ ........................................................................................ ............ Y V G,φ v′ ............................................................................................................................................................................................................................................... ........................................................................................................................................ ............ ............ v These maps induce corresponding maps on G-sheaves: V R2uG,φ∗Z ∗ i −→ R2uG,φ ′ ∗Z ′∗ π ←− R2vG,φ ′ ∗Z j∗ ←− R2vG,φ∗Z [N80a] showed that there exists a map θ = (i∗)−1 ◦ π′∗ ◦ ¯j∗ : R2vG,φ∗Z/ ⊕ ZEj → (R2vG,φ∗Z)G which satisfies θ(x) · θ(y) = G(x · y) for x, y ∈ (⊕ZEj)⊥ and fits into an exact sequence 0 → ker θ −→ R2v∗Z/ ⊕ ZEj θ−→ (R2uG,φ∗Z)G. [N80a] also showed that ker θ is the torsion subsheaf of R2v∗Z/ ⊕ ZEj . Over µ0, we may use the markings αn0 and βn0 to obtain the exact sequence 0 −−−−→ M/ ⊕ Zcj −−−−→ L/ ⊕ Zcj βn0 ◦θ◦α−1 n0 −−−−−−−→ Lφ(G). Note that θ restricts to an injective map θ : M ⊥ ֒→ Lφ(G). Proposition 2.4 implies that M ⊥ and Lφ(G) have the same rank, so we may extend θ to an isomorphism from M ⊥ ⊗ C to Lφ(G) ⊗ C. The theorem now follows from the argument in the abelian case (see [N80a, §8.5]). 7 Remark 3.5. [N80a] claimed that θ is a surjective map when G is abelian. As Garbagnati and Sarti ob- served, this is not the case (see [G08a, Proposition 2.4] and [GS08]). In general, the discrepancy is given by Theorem 2.1. Corollary 3.2. All points of NT determine the same action. Together, Theorem 3.5 and Corollary 3.2 show that we may classify symplectic actions on K3 surfaces by classifying the conditions T which are obtained from symplectic actions. [X96, Table 2] lists the ADE configurations corresponding to finite groups which can act symplectically; we shall refer to these ADE config- urations as symplectic ADE configurations. In most cases, a group G corresponds to a single symplectic ADE configuration; the exceptions are Q8, the group of unit quaternions, and T24, the binary tetrahedral group of order 24, each of which corresponds to two different symplectic ADE configurations. Nikulin showed by direct computation that when G is abelian, the primitive lattice M generated by the singular curves has a unique embedding in the K3 lattice, so T is uniquely determined by G (see [N80a, Theorem 7.2]). The condition T (and thus the action of G) is not uniquely determined by G for every non-abelian group G. For instance, Hashimoto showed in [H09, Proposition 2.12] that there are two distinct symplectic actions of the symmetric group G = S5. Question 3.1. Does every embedding of a symplectic ADE configuration of rational curves in the K3 lattice yield a symplectic group action? Theorem 3.4 tells us that we may approach Question 3.1 by analyzing the possible fundamental groups of the complement of a given configuration. Let T 2 be a two-dimensional complex torus, and let Γ be a finite group of automorphisms of T 2. Fujiki classified the possible finite groups Γ in [F88], and Bertin, Onsiper, and Sertoz classified the resulting singularities of T 2/Γ (see [B88] and [OS99, Proposition 3]): Group C2 C3 C4 C6 Q8 Q12 T24 T24 Singularities of T 2/Γ 16A1 9A2 4A3 + 6A1 A5 + 4A2 + 5A1 4D4 + 3A1 D5 + 3A3 + 2A2 + A1 A5 + 2A3 + 4A2 E6 + D4 + 4A2 + A1 (Here Ck is the cyclic group of order k, Q8 and Q12 are binary dihedral groups, and T24 is the binary tetrahedral group.) The list of K3 singularities obtained from group actions in [X96, Table 2] is disjoint from the list above. We next consider whether there exists an ADE configuration ∆ which can be obtained in two ways: from a singular K3 surface whose smooth part has trivial fundamental group, and as the ADE singularity of another K3 surface whose smooth part has non-trivial fundamental group. Most of the cases can be eliminated using the following lemma, as stated in [SZ01, Lemma 4.6]: Lemma 3.1. [X96, Lemma 2] Let ∆ be an ADE configuration of rational curves on a K3 surface, let Z[∆] be the sublattice of the K3 lattice L generated by the curves in ∆, and let M∆ be the smallest primitive sublattice of L containing Z[∆]. Then the dual of the abelianisation of π1(X − ∆) is canonically isomorphic to M∆/Z[∆]. In particular, if π1(X − ∆) is trivial, then Z[∆] embeds in L as a primitive sublattice. [X96, Table 2] lists M∆/Z[∆] for each ADE configuration which can occur as the exceptional divisor of a resolution of the quotient of a K3 surface by a group of symplectic automorphisms. Using Lemma 3.1, we conclude that none of the configurations in [X96, Table 2] can yield a trivial fundamental group, save possibly the following list of symplectic ADE configurations obtained from perfect groups: Group A5 L2(7) A6 M20 Symplectic ADE Configuration 2A4 + 3A2 + 4A1 A6 + 2A3 + 3A2 + A1 2A4 + 2A3 + 2A2 + A1 D4 + 2A4 + 3A2 + A1 8 (Here A5 and A6 are alternating groups, L2(7) is the Chevalley group P SL(2, F7), and M20 is a subgroup of the Mathieu group M24 which is isomorphic to the semidirect product (Z/2Z)4 ⋉ A5.) Symplectic actions of these groups have been extensively studied using Niemeier lattices. Mukai studied the lattice invariants of SG when G = M20 in an appendix to [K98]; Oguiso and Zhang investigated finite non-symplectic extensions of an L2(7) action in [OZ02]; Keum, Oguiso, and Zhang studied extensions of A6 actions in [KOZ05] and [KOZ07]; and Hashimoto considered actions induced by A5 ֒→ S5 in [H09]. When G = A5, the lattice M = M∆ has rank 18 and discriminant group M ∗/M ∼= (Z/5Z)2 ⊕ (Z/3Z)3 ⊕ (Z/2Z)4. Therefore, the primitive embedding of M in the K3 lattice L is unique up to isometries of L by the results of [N80b], and A5 corresponds to a single condition T and moduli space MT . For each of the groups A6, L2(7), and M20, the lattice M has rank 19; thus, its orthogonal complement M ⊥ in L will be a positive definite lattice of rank 3. Isomorphism classes of positive definite lattices are not always uniquely determined by their invariants. Using the computer algebra system Magma, one may check that when M20 acts symplectically, the lattice M ⊥ is uniquely determined up to isomorphism (see [BCP97]). However, in the cases of A6 and L2(7) a similar analysis in Magma shows that there are two distinct candidates for each M ⊥, and therefore two possible embeddings of each lattice M in L (up to overall isometry). Determining whether these embeddings can be constructed using symplectic actions of A6 and L2(7) is an interesting question for further research. References [BHPV04] Barth, W.P., Hulek, K., Peters, C.A.M., and Van de Ven, A. Compact Complex Surfaces. Berlin, Springer: 2004. [BC94] Batyrev, V. and Cox, D. On the Hodge structure of projective hypersurfaces in toric varieties. Duke Mathematical Journal 75, 1994. [B88] Bertin, J. R´eseaux de Kummer et surfaces K3. Inventiones Mathematicae 93, 1988. [BCP97] Bosma, W., Cannon, J. and Playoust, C. The Magma algebra system. I. The user language. Journal of Symbolic Computation, 24(3-4):235-265, 1997. [C04] Campana, F. Orbifoldes `a Premi`ere Classe de Chern Nulle. arXiv:math.AG/0402243 v2, 2004. [EC56] Cartan, H. and Eilenberg, S. Homological Algebra. Princeton: Princeton University Press, 1956, 1999. [C96] Cox, D. Toric residues. Arkiv for matematik 34, 1996. [CK99] Cox, D. and Katz, S. Mirror Symmetry and Algebraic Geometry. Providence: American Mathe- matical Society, 1999. [DK08] Doran, C. and Kerr, M. Algebraic K-theory of toric hypersurfaces. arXiv:0809.4669v1, 2008. [F88] Fujiki, A. Finite automorphism groups of complex tori of dimension two. Publications of the Re- search Institute for Mathematical Sciences 24 no. 1, 1988. [G08a] Garbagnati, A. Symplectic Automorphisms on Kummer Surfaces. arXiv:0802.0369v1, 2008. [G08b] [G09] [GS08] Garbagnati, A. The Dihedral Group D5 as Group of Symplectic Automorphisms on K3 Surfaces. arXiv:0812.4518v1, 2008. Garbagnati, A. Elliptic K3 surfaces with abelian and dihedral groups of symplectic automorphisms. arXiv.org:0904.1519, 2009. Garbagnati, A. and Sarti, A. Elliptic fibrations and symplectic automorphisms on K3 surfaces. Journal of Algebra 318:1, 2007. [H09] Hashimoto, K. Period map of a certain K3 family with an S5-action. arXiv.org:0904.0072, 2009. 9 [HLOY04] Hosono, S., Lian, B.H., Oguiso, K., and Yau, S.-T. Autoequivalences of derived category of a K3 surface and monodromy transformations. Journal of Algebraic Geometry 13, no. 3, 2004. [KOZ05] Keum, J., Oguiso, K., and Zhang, D.-Q. The alternating group of degree 6 in the geometry of the Leech lattice and K3 surfaces. Proc. London Math. Soc. (3), 90, no. 2, 2005. [KOZ07] Keum, J., Oguiso, K., and Zhang, D.-Q. Extensions of the alternating group of degree 6 in the geometry of K3 surfaces. European Journal of Combinatorics 28, no. 2, 2007. [K98] Kond¯o, S. Niemeier lattices, Mathieu groups, and finite groups of symplectic automorphisms of K3 surfaces. With an appendix by Shigeru Mukai. Duke Mathematical Journal 92, no. 3, 1998. [M84] Morrison, D.R. On K3 surfaces with large Picard number. Inventiones Mathematicae 75, 1984. [M88] [N01] [N80a] [N80b] [O03] [OS99] [OZ02] Mukai, S. Finite groups of automorphisms and the Mathieu group. Inventiones Mathematicae 94, 1988. Narumiya, N. and Shiga, H. The mirror map for a family of K3 surfaces induced from the sim- plest 3-dimensional reflexive polytope. Proceedings on Moonshine and related topics, American Mathematical Society, Providence, RI, 2001. Nikulin, V. Finite automorphism groups of Kahler K3 surfaces. Transactions of the Moscow Math- ematical Society 38, 1980. Nikulin, V. Integral symmetric bilinear forms and some of their geometric applications. Math USSR-Izv. 14, no. 1, 1980. Oguiso, K. A characterization of the Fermat quartic K3 surface by means of finite symmetries. arXiv:math.AG/0308062 v1, 2003. Onsiper, H. and Sertoz, S. Generalized Shioda-Inose Structures on K3 Surfaces. Manuscripta Math- ematica 98, 1999. Oguiso, K. and Zhang, D.-Q. The simple group of order 168 and K3 surfaces. Complex geometry (Gottingen, 2000). Berlin, Springer: 2002. [SAGE] SAGE Mathematics Software, Version 3.4, http://www.sagemath.org/ [SZ01] [S07] [T80] Shimada, I. and Zhang, D.-Q. Classification of extremal elliptic K3 surfaces and fundamental groups of open K3 surfaces. Nagoya Mathematical Journal 161, 2001. Smith, J.P. Picard-Fuchs Differential Equations for Families of K3 Surfaces. University of Warwick, 2006; arXiv:0705.3658v1, 2007. Todorov, A.N. Applications of the Kahler-Einstein Calabi-Yau metric to moduli of K3 surfaces. Inventiones Mathematicae 61, 1980. [X96] Xiao, G. Galois covers between K3 surfaces. Annales de l'Institut Fourier 46, no. 1, 1996. 10
1810.00824
3
1810
2018-12-26T18:01:24
Three plots about the Cremona groups
[ "math.AG", "math.AT", "math.GR" ]
The first group of results of this paper concerns the compressibility of finite subgroups of the Cremona groups. The second concerns the embeddability of other groups in the Cremona groups and, conversely, the Cremona groups in other groups. The third concerns the connectedness of the Cremona groups.
math.AG
math
THREE PLOTS ABOUT THE CREMONA GROUPS VLADIMIR L. POPOV To the memory of V. A. Iskovskikh Abstract. The first group of results of this paper concerns the compressibility of finite subgroups of the Cremona groups. The second concerns the embeddability of other groups in the Cremona groups and, conversely, the Cremona groups in other groups. The third concerns the connectedness of the Cremona groups. 1. Introduction 1.1. The Cremona group Crn(k) of rank n over the field k is the group of k- automorphisms of the field k(x1, . . . , xn) of rational functions over k in the variables x1, . . . , xn. It admits a geometric interpretation: if the field k(x1, . . . , xn) is iden- tified by means of a k-isomorphism with the field k(X) of an irreducible algebraic variety X defined and rational over k, then each element σ of the group Birk(X) of all k-birational self-maps X 99K X determines the element σ∗ ∈ Crn(k), σ∗(f ) := f ◦ σ, f ∈ k(X), (1.1) and the mapping Birk(X) → Crn(k), σ 7→ (σ−1)∗, is an isomorphism of groups. For this reason, the group Birk(X) is called the Cremona group as well and denoted by Crn(k). Which interpretation of Crn(k) is meant -- algebraic or geometric -- is usually clear from context. The naturally defined concept of a morphism of an al- gebraic variety into the Cremona group (or the concept of an algebraic family of elements of the Cremona group) allows one to endow it with the Zariski topology [Se10, 1.6]. Besides this property, there is a number of others which permit to speak about the far-reaching analogies between the Cremona groups and affine algebraic groups, see [Po131], [Po132], [Po142], [Po17]. The Cremona groups are classical objects of research, intensity of which in re- cent years increased significantly and led to essential advances in understanding the structure of these groups. Among the most impressive is tour de force [DI09] by I. V. Dolgachev and V. A. Iskovskikh on the classification of finite subgroups of Cr2(C). 1.2. In this paper, three aspects of the structure of the Cremona groups are ex- plored. The topic of the (longest) Section 2 is the comparison of different finite subgroups of the Cremona group Crn(k), where k is an algebraically closed field of characteristic zero. So far, in the studies of these subgroups, including that in [DI09], they were all considered on an equal footing. However, in reality it is necessary to consider some of them as "not basic", since they are obtained from others by a standard "base change" construction [Po141, 3.4]. This leads to the problem, formulated in [Po141, 3.4], [Po16, Quest. 1], of finding those subgroups in the classification lists that are 1 2 VLADIMIR L. POPOV obtained by such a nontrivial change, or, in another terminology, are nontrivialy "compressible" (see definitions in Subsection 2.1). Developing this topic, in Section 2 we prove a series of statements about such subgroups. Some of them are of a general nature, while some concern cases n = 1 and 2. For example, we obtain the following result (Theorem 2.1), which immedi- ately implies the nontrivial self-compressibility of any finite subgroup G in Cr1: for the corresponding binary group eG of linear transformations of the affine plane, we find an infinite increasing sequence of integers d > 0 such that eG admits a ho- mogeneous polynomial self-compression of degree d, which descends to a nontrivial self-compression of the group G. The proof allows us in principle to specify these self-compressions by explicit formulas. For n = 2, we prove, for example, that if G is a non-Abelian finite subgroup of GL2(k) ⊂ Cr2(k) that is not isomorphic to a dihedral group, then every finite subgroup in Cr2(k), isomorphic to G as an abstract group, is obtained from G by a nontrivial base change (Theorem 2.19). Other state- ments on this subject, proved in Section 2, see below in Theorems 2.8 -- 2.19 and their Corollaries. 1.3. The subject of Section 3 is the embeddability of other groups in the Cremona groups and, conversely, the embeddability of the Cremona groups in other groups. This theme originates from the question of J.-P. Serre [Se09, §6, 6.0] on the existence of finite groups that are nonembeddable in Cr3(C). By now (September 2018) signi- ficant information is accumulated on it (including the affirmative answer to this question). The most essential contribution to its obtaining is related to the Jordan property (see Definition 3.1 below) of the Cremona groups Crn(k), whose proof for any n has been completed recently1. Although the statements about the group embeddings proved in Section 3 are also related to the Jordan property, which is in the focus of attention already for a long time, in the published literature they did not occur to the author. The fact that, for char k = 0, every finite p-subgroup of Crn(k) is Abelian for sufficiently big p, immediately follows from the Jordan property of the Cremona groups (this was noted already in [Se09, §6, 6.1]). Therefore, every non-Abelian finite p-group (such exist for any p) is nonembeddable in Crn(k) for sufficiently big p. We prove (Corollary 3.7), for any Cremona group Crn(k) with char k = 0, the existence of an integer bn,k > 0 such that every product of groups G1 × · · · × Gs, each of which contains a non-Abelian finite subgroup, is nonembeddable in the group Crn(k) if s > bn,k. In particular, for any (and not only for sufficiently big) prime integer p, there exists a non-Abelian finite p-group that is nonembeddable in Crn(k). Considering p-subgroups delivers invariants, which allow us to prove in some cases that one group is nonembeddable in another. Some applications are obtained on this way. 1In [PS16, Thm. 1.8], it was given the conditional (modulo the so-called BAB conjecture) proof of the Jordan property of the group Birk(X) for any rationally connected algebraic k-variety X in the case of char k = 0 (and therefore, the conditional proof of the Jordan property of any Cremona group Crn(k)). The BAB conjecture was then proved in [Bi17, Thm. 3.7]. This completed the proof of the Jordan property of the groups Birk(X). THREE PLOTS ABOUT THE CREMONA GROUPS 3 For example (special case of Corollary 3.14), we prove that if k is an algebraically closed field of characteristic zero, and with each integer d > 0 any abstract group Hd from the following list is associated: (a) Crd(k), (b) Aut(Ad (c) a connected affine algebraic group over k with maximal tori of dimension d, (d) a connected real Lie group with maximal tori of dimension d, k), then the group Hn is nonembeddable in Hm if n > m. In particular, the groups Hn and Hm for n 6= m are not isomorphic. For instance, Crn(k) is embeddable in Crm(k) if and only if n 6 m; in particular, Crn and Crm are isomorphic if and only if n = m (this was previously proven in [Ca14, Thm. B], [PS16, Rem. 1.11]). Another example (Theorem 3.20): we prove that if M is a compact connected n-dimensional topological manifold, and BM is the sum of its Betti numbers with respect to homology with coefficients in Z, then for d > pn2 + 4n(n + 1)BM + n 2 + log2BM , the Cremona group Crd(k) is nonembeddable in the homeomorphism group of the manifold M. Concerning other statements on nonembeddable groups proved in Section 3, see below Lemma 3.2, Theorems 3.12, 3.13, 3.21 and their Corollaries. 1.4. In Section 4, we return back to the question of J.-P. Serre on the connectedness of the Cremona group Crn(k) in the Zariski topology [Se10, 1.6]. It was answered in the affirmative in [BZ18], where the linear connectedness (and therefore the con- nectedness) of the group Crn(k) is proved in the case of an infinite field k (for an algebraically closed field k, this was proved earlier in [Bl10]). We give a short new proof for the case of an infinite field k. It is based on an argument, ideologically close to that of Alexander, which he used in [Al23] in proving the connectedness of the homeomorphism group of the ball, and which was then adapted in [Sh82, Lem. 4], [Po142, Thm. 6], and [Po17] to the proofs of connectedness of the groups Aut(An) and their affine-triangular subgroups, respectively. The author is grateful to J.-P. Serre, Ch. Urech, and the referee for the comments. 1.5. Notations and conventions. k is a fixed algebraically closed field containing k. Crn := Crn(k), Bir(X) := Birk(X), Aut(X) := Autk(X). o = (0, . . . , 0) ∈ An. hSi is a linear span of a subset S of a linear space over k. Grass(n, V ) is the Grassmannian of all n-dimensional linear subspaces of a finite- dimensional linear space V over k. P(V ) := Grass(1, V ). We put P({0}) = ∅ and dim(∅) = −1. L⊕m is the direct sum of m copies of a linear space V over k (for m = 0, it is considered to be zero). Gs is the direct product of s copies of a group G. 4 VLADIMIR L. POPOV "Variety" means "algebraic variety over k". Its irreducibilitty means geometric irreducibility, and points mean k-points. The set of k-points of a variety X is denoted by X(k). Dom(ϕ) is the domain of definition of a rational map ϕ. Ta,X is the tangent space of a variety X at a point a. daϕ : Ta,X → Tϕ(a),X is the differential of a rational map ϕ : X 99K X at a point a ∈ Dom(ϕ). k[x1, . . . , xn]d is the space of all forms of degree d in variables x1, . . . , xn and with coefficients in k. Ld := L ∩ k[x1, . . . , xn]d for any k-linear subspace L in k[x1, . . . , xn]. F H = {f h f ∈ F, h ∈ H} for any nonempty sets F, H ⊆ k[x1, . . . , xn]. The variables x1, . . . , xn in the definition of the Cremona group are assumed to be the standard coordinate functions on An: a := (a1, . . . , an) ∈ An. For any rational map σ : An 99K An, we use the notation xi(a) := ai, σ = (σ1, . . . , σn) : An 99K An, where σi := σ∗(xi). (1.2) We call σ polynomial homogeneous map of degree d, if σ1, . . . , σn ∈ k[x1, . . . , xn]d. In these notation, if for a rational map τ : An 99K An the composition ν := σ ◦ τ is defined, then it is described by the formula νi = τ ∗(σi) for all i, (1.3) i.e., νi is obtained from the rational function σi in x1, . . . , xn by means of plugging in τj in place of xj for every j. The map (1.2) is called affine (respectively, linear), if all nonzero functions σi are polynomial in x1, . . . , xn of degree 6 1 (respectively, are forms in x1, . . . , xn of degree 1). The set of all invertible affine (respectively, linear) maps An → An is the subgroup Aff n (respectively, GLn) of Crn. 2. Compressing finite subgroups of the Cremona groups In this section, k = k and char k = 0. 2.1. Terminology. First, we fix the terminology. Here, unless a special reservation is made, a rational action of a finite group G on an irreducible variety X is understood as a faithful (that is, with trivial kernel) action by birational self-maps of this variety. Specifying such an action is equivalent to specifying a group embedding : G ֒→ Bir(X); therefore, hereinafter the very homomorphism is called a rational action. The integer dim(X) is called the di- mension of the action . We say that (G) is the subgroup of Bir(X) defined by the action . If (G) ⊆ Aut(X), then the action is called regular. Any regular action ρ of the group G on an irreducible smooth complete variety Y such that there is a G-equivariant birational isomorphism X 99K Y is called a regularization of the action ; combining the results of [Ro56, Thm. 1] and [BM97] shows that a regularization always exists. If there is a regularization ρ such that Y G 6= ∅, then we say that has a fixed point. Consider two rational actions i : G ֒→ Bir(Xi), i = 1, 2. Let πi : Xi 99K Xi---G, i = 1, 2, be the corresponding rational quotients, see [PV94, 2.4]. Assume that there THREE PLOTS ABOUT THE CREMONA GROUPS 5 is a G-equivariant dominant rational map ϕ : X1 99K X2. Let ϕG : X1---G 99K X2---G be the dominant rational map induced by ϕ. Then the following properties hold (see, e.g., [Re001, 2.6]): First, the commutative diagram X1 ϕ /❴❴❴❴❴ X2 π1 π2 X1---G ϕG /❴❴❴ X2---G (2.1) is cartesian, i.e., π1 is obtained from π2 by the base change ϕG. In particular, X1 is birationally G-equivariantly isomorphic to the variety X2 × X2--- G (X1---G) := {(x, y) ∈ Dom(π2) × Dom(ϕG) π2(x) = ϕG(y)} (2.2) (the bar in (2.2) means the closure in X2 × X1---G), on which G acts through X2. Second, for every irreducible variety Z and every dominant rational map β : Z 99K Z is irreducible, the latter inherits through X2---G such that the variety X2 × X2 a rational action of the group G such that commutative diagram (2.1) with X1 = X2 × Z, ϕG = β, and ϕ = pr1 holds. X2--- G X2--- G It is said [Re04] that ϕ is a compression of the action 1 into the action 2 (or that 2 is obtained by the compression ϕ from 1), and also [Po141, 3.4] that 1 is obtained by the base change ϕG from 2. A compression that is not (or, respectively, is) a birational isomorphism is called nontrivial (respectively, trivial ); in this case, we say that 1 is obtained by a nontrivial (respectively, trivial ) base change from 2. If for 1 there is 2, which is obtained from 1 by a nontrivial compression, then we say that 1 is nontrivially compressible, and otherwise, that it is incompressible. Similar terminology applies to groups: if Gi ⊆ Bir(Xi), i = 1, 2, are finite subgroups isomorphic to G, then we say that G1 is compressible into G2, if there are rational actions i : G ֒→ Bir(Xi), i = 1, 2, such that i(G) = Gi, i = 1, 2, and 2 is obtained by a compression ϕ from 1. If 1, 2, ϕ can be chosen so that ϕ is nontrivial, then G1 is called nontrivialy compressible into G2. If G1 does not admit any nontrivial compression into any subgroup in Bir(X2), then G1 is called incompressible. In the case when X1 = X2 and 1 = 2, we are talking about self-compressions of the action and the group. In particular, if in this case there exists a nontrivial compression, then we say that 1(G) is a nontrivially self-compressible subgroup of Bir(X). 2.2. Self-compressibility of finite subgroups in Cr1: reformulation. First we consider the problem of nontrivial self-compressibility of finite subgroups in the Cremona group Cr1 of rank 1. It can be reformulated as follows. We assume that A1 = {(a0 : a1) ∈ P1 a0 6= 0} and denote the standard coordinate function x1 ∈ k[A1] by z. The elements of every finite subgroup G of the Cremona group Cr1 are fractional-linear functions from the field k(z) (considered as rational maps A1 99K A1). The restriction to A1 defines a bijection between the set of self-compressions P1 → P1 of the group G and the set of rational functions /   ✤ ✤ ✤   ✤ ✤ ✤ / 6 VLADIMIR L. POPOV f = f (z) ∈ k(z), which are solutions of the following system of functional equations: f(cid:16)az + b cz + d(cid:17) = af + b cf + d for all az + b cz + d ∈ G. (2.3) In this setting, the nontriviality of the self-compression defined by the rational function f is equivalent to the condition deg(f ) > 2. Note that in (2.3) instead of all functions from the group G it suffices to consider only the generators of this group. Thus, the question of nontrivial compressibility of the group G is equivalent to the question of the existence of a rational function f of degree > 2 among the solutions of system (2.3). 2.3. Compressibility of binary polyhedral groups: formulation of the re- sult. The comprehensive answer to the above question can be obtained for any finite subgroup of the Cremona group Cr1: all of them are nontrivially compress- ible. This asnwer is an immediate corollary of a more subtle result, which we obtain here. Namely, we prove that there exists infinitely many homogeneous polynomial self-compressions A2 → A2 of any binary polyhedral group, descending to the non- trivial self-compressions P1 → P1 of the corresponding polyhedral group. The proof is effective and gives a way to explicitly specify these self-compressions by formulas (see Remark (c) in Subsection 2.8). We now give the precise formulation of this result. Let G be a nontrivial finite subgroup of the group PSL2 = Aut (P1) = Cr1. We consider the canonical homomorphism whose kernel is the center Z := (−id). The group ν : SL2 → PSL2 eG := ν −1(G) ⊂ SL2 (2.4) is either a binary rotation group of one of the regular polyhedra (dihedron, tetra- hedron, octahedron, or icosahedron), or a cyclic group of even order > 4. The subset A2 ∗ := A2 \ o is open in A2 and stable with respect to the actions on A2 of the groups eG and T := {(tx1, tx2) t ∈ k×}. Let π : A2 ∗ := A2 \ o → P1 be the natural projection. The pair (π, P1) is a geometric quotient for the action of the torus T on A2 ∗. The morphism π is eG-equivariant if we assume that the action of eG on P1 is the restriction on eG of the homomorphism ν (this action is not faithful, its kernel is Z). If a self-compression eϕ = (eϕ1, eϕ2) : A2 99K A2 of the group eG is polynomial homogeneous of degree d, then the morphism π ◦ eϕ ∗ and, therefore, factors through π, i.e., there is a is constant on the T -orbits in A2 morphism ϕ : P1 → P1, (2.6) such that ϕ ◦ π = eϕ ◦ π. It is dominant (and therefore surjective) in view of the dominance of the morphism eϕ. From the eG-equivariance of the morphisms π and eϕ it follows the eG-equivariance -- and therefore the G-equivariance -- of the morphism (2.5) THREE PLOTS ABOUT THE CREMONA GROUPS 7 ϕ. Consequently, ϕ is the self-compression of the natural action of the group G on P1. We say that the self-compression ϕ is a descent of the self-compression eϕ. Theorem 2.1. Let G be a nontrivial finite subgroup of the Cremona group Cr1 = PSL2 = Aut (P1). Associate with it the formal power series SG(t) = X n>0 sntn ∈ Z[[t]] (2.7) of the following form: (a) If G is a rotation group of tetrahedron, octahedron, or icosahedron, then SG(t) = t2a−1(1 + t4a−6)X t2naX t(4a−4)n + t4a−5X n>0 n>0 n>0 t(4a−4)n, (2.8) where, respectively, a = 3, 4, or 6. (b) If G is either a dihedral group of order 2ℓ > 4 or a cyclic group of order ℓ > 2, then SG(t) = X n>0 t2ℓ(n+1)−1. (2.9) Suppose that the coefficient sd of the series (2.7) is different from zero. Then there exists a polynomial homogeneous self-compression (2.5) of the binary group eG (see (2.4)), whose degree is d, and descent (2.6) is a nontrivial self-compression of the group G. The proof of Theorem 2.1 will be given in Subsection 2.7, after proving several necessary auxiliary statements in Subsection 2.6. 2.4. Application: self-compressibility of finite subgroups of Cr1. Theorem 2.1 immediately implies statement (i) of the following theorem. Theorem 2.2. (i) Every finite subgroup of Cr1 is nontrivially self-compressible. (ii) Every compression of a finite subgroup of Cr1 is a compression P1 → P1 into a conjugate subgroup. Proof of (ii). Since every variety, to which P1 maps dominantly, is rational, (ii) follows from the definition of compression and the well-known fact that two finite subgroups of Cr1 are isomorphic if and only if they are conjugate. (cid:3) Remark 2.3. Another proof of statement (i) of Theorem 2.2 is given in [GA16, Cor. 1.3]. This proof consists of presenting explicit formulas, in relation to which the reader is supposed to verify by direct computations that they define G-equivariant maps. In [GA16] there are no comments about the origin of these formulas. For example (see [GA16, Lemma 9.7]), if ω5 ∈ k is a primitive fifth root of 1 and G is the lying in Cr1 rotation group of the icosahedron generated by the fractional-linear transformations ω5z and (ω5 + ω−1 5 )z + 1 z − (ω5 + ω−1 5 ) , then such a formula has the the appearance P1 → P1, (x : y) 7→ (x11 + 66x6y5 − 11xy10 : −11x10y − 66x5y6 + y11). 8 VLADIMIR L. POPOV Below (see Remark (c) in Subsection 2.8) we explain how in principle the explicit formulas can be found that define any self-compression specified in Theorem 2.1. 2.5. Notations. To prove Theorem 2.1 we need several notations. dule. A := k[A2] = k[x1, x2] We denote by eG∨ the set of characters of all simple keG-modules. The action of the group eG on the affine plane A2 defines on the algebra the structure of a keG-module. The latter is graded: each space An is its keG-submo- We denote by χ the character of the submodule A1. If the group eG is not cyclic, For any simple keG-module M with character γ ∈ eG∨, we denote by A(γ) the isotypic component of type M in the keG-module A; it is its graded submodule. In particular, A(1) is the subalgebra of eG-invariants in A. We will also need the following set of characters: this submodule is simple. (2.10) ֒→ L}. (2.11) It is not empty, because 1 ∈ [χ]. [χ] := {γ ∈ eG∨ dim(γ) = 1, γχ = χ}. For any finite-dimensional keG-module L, we put multχ(L) := max{d there exists an embedding of keG-modules A⊕d 1 2.6. Auxiliary statements. We now prove several auxiliary statements that are used in the proof of Theorem 2.1. Lemma 2.4. Let H be a subgroup of GLn and let L be a finite-dimensional k-linear subspace in k[x1, . . . , xn]. (a) The following conditions are equivalent: (a1) L is a submodule of the kH-module k[x1, . . . , xn] that is isomorphic to the kH-module k[x1, . . . , xn]1. (a2) There exists a basis σ1, . . . , σn of the linear space L such that the mor- phism σ := (σ1, . . . , σn) : An → An is H-equiavriant. (b) Suppose that the equivalent conditions (a1) and (a2) hold. (b1) The morphism σ from (a2) is dominant if and only if σ1, . . . , σn are algebraically independent over k. (b2) If n = 2 and σ1, σ2 ∈ k[x1, x2]d for some d, then σ1, σ2 are algebraically independent over k. Proof. (a1) ⇒ (a2): Let k[x1, . . . , xn]1 → L be an isomorphism of kH-modules and let σi be the image of xi with respect to this isomorphism. Then the H-equivariance of σ follows directly from the definitions and formulas (1.1), (1.2). (a2)⇒(a1): It follows from (1.1), (1.2) that the restriction of σ∗ to k[x1, . . . , xn]1 is an isomorphism of linear spaces k[x1, . . . , xn]1 → L. From this and the H-stability of k[x1, . . . , xn]1 it follows the H-stability of L, so this restriction is an isomorphism of kH-modules. (b1): The dominance of σ is equivalent to the triviality of the kernel of the homo- morphism σ∗ of the algebra k[x1, . . . , xn], which, in view of (1.2), is equivalent to the algebraic independence of σ1, . . . , σn over k. THREE PLOTS ABOUT THE CREMONA GROUPS 9 (b2): Suppose that σ1, σ2 are algebraically dependent over k, i.e., there exists a nonzero polynomial F = F (t1, t2) ∈ k[t1, t2], where t1, t2 are variables, such that F (σ1, σ2) = 0. (2.12) Since σ1 and σ2 are forms in x1, x2 of the same degree, we can (and shall) assume that F is a form in t1, t2, say, of degree s: F (t1, t2) = sX i=0 αits−i 1 ti 2, α0, . . . , αs ∈ k. (2.13) In view of (a2), the polynomial σ2 is nonzero, so that we can consider the rational function σ1/σ2 ∈ k(x1, x2). From (2.12), (2.13) we get for it the relation 0 = sX i=0 αi(cid:16)σ1 σ2(cid:17)s−i . (2.14) It follows from the linear independence of the polynomials σ1, σ2 over k that the rational function σ1/σ2 is not an element of k, and therefore takes on A2 infinitely In view of (2.14), each of these values is the root of the many different values. i=0 αits−i ∈ k[t], t = t1/t2. This contradiction proves the algebraic independence of σ1, σ2 over k. (cid:3) nonzero polynomial Ps Lemma 2.5. Let (2.5) be a polynomial homogeneous self-compression of the group eG, whose degree is d. Let a form a ∈ A be the greatest common divisor of the forms eϕ1 and eϕ2 that define (2.5). The following properties are equivalent: (a) the descent (2.6) of the self-compression (2.5) is trivial; (b) deg(a) = d − 1; (c) there exists a character γ ∈ [χ] and an element s ∈ A(γ)d−1 such that Proof. (a)⇔(b): If we consider x1 and x2 as homogeneous coordinates on P1, then the self-compression (2.6) is given by the formula eϕ∗(A1) = sA1. ϕ = (cid:16)eϕ1 a (cid:17) : eϕ2 a (2.15) (2.16) (see [Sh13, Chap. III, §1, 4]). Since every k-automorphism of the field of rational functions in one variable over k is a fractional linear transformation [Wa67, §73], it follows from (2.16) and the inclusion eϕ1, eϕ2 ∈ Ad that the self-compression ϕ is trivial if and only if the forms eϕ1/a eϕ2/a are linear, i.e., (b) is satisfied. (b)⇔(c): Suppose that (b) holds. Then the equality and the definition of the form a imply the equality (2.15) for s = a. eϕ∗(A1) = heϕ1, eϕ2i (2.17) In view of the eG-invariance of the subspaces eϕ∗(A1) and A1, for every g ∈ eG, we obtain from (2.15) the following equalities: aA1 = g · (aA1) = (g · a)(g · A1) = (g · a)A1. (2.18) We take some linear form l ∈ A1, whose zero in P1 does not coincide with any of zeros of the form a. Since, in view of (2.18), the form (g · a)l is divisible by a, and deg(g · a) = deg(a), this means that the divisors of the forms a and g · a on P1 10 VLADIMIR L. POPOV coincide, hence hai = hg · ai. Therefore, a is a semi-invariant of the group eG. Let γ ∈ eG∨ be the character of the one-dimensional keG-module hai. Then a ∈ A(γ)d−1, and γχ is the character of the keG-module aA1. But since the keG-modules A1 and eϕ∗(A1) are isomorphic, it follows from (2.15) that the character of the keG-module greatest common divisor of the forms eϕ1 and eϕ2, and therefore hsi = hai, and hence Conversely, if (c) is satisfied, then it follows from (2.15) and (2.17) that s is the aA1 is χ. Therefore, γ ∈ [χ]. This proves (b)⇒(c). (b) is satisfied. This proves (c)⇒(b). (cid:3) Lemma 2.6. Let H be a group,let L be a nonzero kH-module of dimension s < ∞, and let m be a positive integer. The Grassmanninan Grass(s, L⊕m) contains a closed irreducible (m − 1)-dimensional subset such that all the s-dimensional linear subspaces of the kH-module L⊕m corresponding to its points are the submodules isomorphic to L. Proof. We assign to any nonzero vector (λ1, . . . , λm) ∈ km the following embedding of the kH-modules: ι(λ1,...,λm) : L ֒→ L⊕m, v 7→ (λ1v, . . . , λmv). The images of the embeddings ι(λ1,...,λm) and ι(µ1,...,µm) coincide if and only if the vectors (λ1, . . . , λm) and (µ1, . . . , µm) are proportional, i.e., the corresponding points (λ1 : . . . : λm) and (µ1 : . . . : µm) of the projective space P(km) coincide. Conse- quently, the mapping P(km) → Gr(s, L⊕m), which assigns to every point (λ1 : . . . . . . : λm) ∈ P(km) the image of the embedding ι(λ1,...,λm), is injective. It is not difficult to see that this mapping is a morphism. Therefore, its image is an irreducible closed subset of dimension dim(P(km)) = m − 1. It is this image that should be taken as the subset specified in the formulation of Lemma 2.6. (cid:3) Lemma 2.7. If, for a positive integer d, the inequality multχ(Ad) > max{dimk(A(γ)d−1) γ ∈ [χ]}, (2.19) holds, then the group eG admits a polynomial homogeneous self-compression (2.5) of degree d, whose descent (2.6) is nontrivial. Proof. Assume that the inequality (2.19) holds. For the sake of brevity, put m := multχ(Ad). (2.20) It follows from (2.19) that m > 0. According to (2.11), in the keG-module Ad there exists a submodule M isomorphic In view of dim(A1) = 2 and Lemma2.6, in Grass(2, M) there exists an to A⊕m irreducible closed subset X such that . 1 dim(X) = m − 1 (2.21) and all the 2-dimensional linear subspaces of M corresponding to its points are the submodules isomorphic to A1. Since M is a linear subspace in Ad, the variety Grass(2, M), and hence X as well, is a closed subset of Grass(2, Ad). It follows from the definition of the set [χ] (see (2.10)) that for every character γ ∈ [χ] and every nonzero element s ∈ A(γ)d−1, the linear subspace sA1 is a submodule of the keG-module Ad isomorphic to A1. This submodule does not change when s THREE PLOTS ABOUT THE CREMONA GROUPS 11 is multilied by nonzero elements of k, therefore, assigning the submodule sA1 to the element s defines a mapping P(A(γ)d−1) → Grass(2, Ad). It is not difficult to see that it is a morphism. Hence its image Y (γ) is an irreducible closed subset in Grass(2, Ad), and dim(Y (γ)) 6 dim(P(A(γ)d−1)) = dimk(A(γ)d−1) − 1. (2.22) In view of the finiteness of the set [χ], it follows from (2.19), (2.20), (2.21), (2.22) that X \Sγ∈[χ] Y (γ) (2.23) is a nonempty subset of the Grassmannian Grass(2, Ad). We consider a point of the set (2.23) and the two-dimensional linear subspace L in Ad corresponding to it. Then it follows from the above properties of the sets X and Y (γ) that (ii) there are no γ ∈ [χ] and s ∈ A(γ)d−1 such that L = sA1. (i) L is a submodule of the keG-module Ad isomorphic to A1; In view of (i) and Lemma 2.4, there is a basis eϕ1, eϕ2 of L such that (2.5) is a polynomial homogeneous self-compression of the group eG of degree d, for which eϕ∗(A1) = L. It follows from (ii) and Lemma 2.5 that the descent (2.6) of this (cid:3) self-compression is nontrivial. 2.7. Proof of Theorem 2.1. The plan of the proof of Theorem 2.1 is as follows. For each noncyclic finite subgroup G of PSL2 = Aut (P1) = Cr1, we exlicitly describe for eG the set [χ] and the Poincar´e series P (χ, t) := X (multχ(An))tn, P (γ, t) := X n>1 n>0 (dimk(A(γ)n))tn, where γ ∈ [χ]. (2.24) Comparing the coefficients of these series, we show that if a coefficient sd of the series (2.7) is nonzero, then the inequality (2.19) holds, from which, according to Lemma 2.7, the statement of Theorem 2.1 for G follows. The case of a cyclic finite subgroup G is reduced to that of the corresponding dihedral one. Proof of Theorem 2.1. We consider separately three possible types of the group eG. (a) eG is a primitive subgroup of the group SL2, i.e., a binary tetrahedral, octahed- ral, or icasahedral group. In view of [Sp87, 3.2(a)] and the definition of the set [χ] (see (2.10)), in this case we have From [Sp87, 4.2] we obtain: [χ] = {1}. P (χ, t) = t + t2a−1 + t4a−5 + t6a−7 (1 − t2a)(1 − t4a−4) , P (1, t) = 1 + t6a−6 (1 − t2a)(1 − t4a−4) , (2.25) (2.26) 12 VLADIMIR L. POPOV where a = 3, 4, and 6 respectively for binary tetrahedral, octahedral, and icosahedral group. From (2.24), (2.26) we then deduce the following: P (χ, t) − tP (1, t) = X n>1 (multχ(An) − dimk(A(1)n−1))tn = t2a−1 1 + t2a−4 + t4a−6 − t4a−4 (1 − t2a)(1 − t4a−4) = t2a−1 1 + t4a−6 + t4a−5 (1 − t2a)(1 − t4a−4) 1 − t4a−4 = t2a−1(1 + t4a−6)X t2naX t(4a−4)n + t4a−5X t(4a−4)n n>0 n>0 n>0 1 (2.27) (2.8) == SG(t). From (2.27) and (2.7) we obtain that sd = multχ(Ad) − dimk(A(1)d−1) for every d > 0. In view of (2.25), this gives sd = multχ(Ad) − max{dimk(A(γ)d−1) γ ∈ [χ]} for every d. (2.28) As (2.8) shows, if sd 6= 0, then sd > 0. In view of (2.28) and Lemma 2.7, this implies the statement of Theorem 2.1 in case (a). dral group of order 4ℓ > 8. (b) eG is an irreducible imprimitive subgroup of the group SL2, i.e., a binary dihe- In this case, the McKay correspondence [Sp87, Sect. 2] juxtaposes to the group eG the extended Dynkin diagram of type D(1) ℓ+2 with ℓ + 3 vertices. According to [Sp87, 2.3(a)], the vertex juxtaposed to the character χ is a branch point of this diagram. In eG∨ there are exactly four one-dimensional characters 1, θ, θ′, θ′′ (see [Sp87, 4.3]); the vertices of the diagram juxtaposed to them is the set of all its endpoints. In view of [Sp87, 2.2] and (2.10), an endpoint corresponds to a character from [χ] if and only if it is connected by an edge to the vertex juxtaposed to the character χ. Therefore, apart from 1, there is at least one more character in [χ] (we denote it by θ), and two possibilities occur: -- If ℓ > 3, then the vertex juxtaposed to the character χ is connected by edges to only two endpoints of the diagram, which are juxtaposed to the one-dimensional characters 1 and θ: ◦ ■■■■ χ ◦ ✉✉✉✉ ◦ 1 θ · · · α◦ ◦ ✉✉✉✉ ■■■■ ◦ θ′ θ′′ , χ 6= α. (2.29) Thus, we obtain (2.30) -- If ℓ = 2, then the vertex juxtaposed to the character χ is connected by edges to all four endpoints, which are juxtaposed to the one-dimensional characters 1, θ, θ′ θ′′: [χ] = {1, θ} for ℓ > 3. 1 θ ◦ ◦ ❆❆❆ ⑦⑦⑦ ◦ χ ❆❆❆ ⑦⑦⑦ ◦ ◦ θ′ θ′′ . Therefore, we obtain [χ] = {1, θ, θ′, θ′′} for ℓ = 2. (2.31) THREE PLOTS ABOUT THE CREMONA GROUPS 13 It follows from [Sp87, (9)] that, for every ℓ > 2, θ is the character denoted in [Sp87, p. 103] by c1. From this and [Sp87, 4.4] we obtain: P (χ, t) := t + t3 + t2ℓ−1 + t2ℓ+1 (1 − t4)(1 − t2ℓ) P (1, t) := P (θ, t) = 1 + t2ℓ+2 (1 − t4)(1 − t2ℓ) , t2 + t2ℓ (1 − t4)(1 − t2ℓ) . In addition, according to [Sp87, 4.4], ,   for every ℓ > 2. (2.32) P (θ, t) = P (θ′, t) = P (θ′′, t) for ℓ = 2. (2.33) From (2.24), (2.32) we obtain: P (χ, t) − tP (1, t) − tP (θ, t) = X n>1(cid:0)multχ(An) − dimk(A(1)n−1) − dimk(A(θ)n−1)(cid:1)tn = t2ℓ−1 1 − t4 (1 − t4)(1 − t2ℓ) = X n>0 t2ℓ(n+1)−1 (2.34) (2.9) == SG(t). It follows from (2.34) and (2.7) that sd = multχ(Ad) − dimk(A(1)d−1) − dimk(A(θ)d−1) for every d > 0. In view of (2.30), (2.31),and (2.33), this gives sd 6 multχ(Ad) − max{dimk(A(γ)d−1) γ ∈ [χ]} for every d. (2.35) As (2.9) shows, if sd 6= 0, then sd > 0. In view of (2.35) and Lemma 2.7, this implies the statement of Theorem 2.1 in case (b). (c) eG is a cyclic subgroup of order 2ℓ > 4 in the group SL2. Since eG is a subgroup of a binary dihedral group of order 4ℓ, every polynomial eG. Therefore, the statement of Theorem 2.1 for eG follows from its statement, already homogeneous self-compression of the latter group is a self-compression of the group proved, for binary dihedral groups. (cid:3) 2.8. Remarks about Theorems 2.1 and 2.2. Concluding the discussion of self- compressibility of finite subgroups of the Cremona group of rank 1, we will make a few remarks about Theorems 2.1 and 2.2. (a) For every nontrivial finite subgroup G of the Cremona group Cr1, Theorem 2.1 yields an infinite set of natural numbers d, for which there exists a polynomial homogeneous self-compression (2.5) of the corresponding binary group eG, whose degree is d and descent (2.6) is a nontrivial self-compression of the group G. From formulas (2.8), (2.9) we obtain the minimal of these d: it is equal to 5, 7, and 11 respectively for the rotation group of a regular tetrahedron, octahedron, and icosahedron, and to 2ℓ − 1 for the dihedral group of order 2ℓ > 4 and cyclic group of order ℓ > 2. 14 VLADIMIR L. POPOV (b) Let K be a field of algebraic functions in one variable over k. By Theorem 2.2(i), if the genus of the field K is equal to 0, then every finite subgroup of Autk(K) is nontrivially self-compressible. For the fields K of genus > 2, this, in general, is not so, see [Re04, Ex. 6], [GA16]. are specified in Theorem 2.1. (gℓ)(a) := g(ℓ(g−1(a))), g ∈ eG, ℓ ∈ L (A1, Ad), a ∈ A1, (2.36) Let d be such a degree. In view of what was said in part (c) of the proof of the character χ is irreducible. Consider the linear space L (A1, Ad) of all linear maps (c) Let us briefly explain how one can find explicitly the forms eϕ1, eϕ2, which define the homogeneous polynomial self-compressions (2.5) of the group eG, whose degrees Theorem 2.1, we can (and shall) assume that the group eG is not cyclic, and hence A1 → Ad. The group eG acts linearly on it by the rule and the eG-equivariant maps are precisely the fixed points of this action. They form image of the Reynolds operator eG−1Pg∈ hSm that L := (Pm in L (A1, Ad) a linear subspace L (A1, Ad)G, see [PV94, 3.12]. The latter is the eG g for the action (2.36), see [PV94, 3.4], and therefore, is described effectively. If ℓ1, . . . , ℓm is a basis of L (A1, Ad)G, then i=1 ℓi(A1)i = A(χ)d. Similarly, effectively are described the isotypic components According to the proof of Theorem 2.1, the set of all (α1, . . . , αm) ∈ km such i=1 αiℓi)(A1) does not lie in hA(γ)d−1A1i for all γ ∈ [χ], is nonempty. Effective finding of such a (α1, . . . , αm) reduces to finding for some explicitly de- scribed nonzero polynomial in m variables with coefficients in k any values of these variables that do not make this polynomial zero. A(γ)d−1 for all γ ∈ [χ]. space A1 (it suffices to ensure this only for the system of generators of the group The linear space L is a keG-submodule of Ad isomorphic to A1. As a couple of forms eϕ1, eϕ2 one can now take a basis of this subspace such that the matrices of the elements of the group eG in this basis are the same as in the basis x1, x2 of the eG, containing two elements for dihedral group and three for the others, see [Sp87]). Effective finding of such a basis reduces to finding a solution of a system of linear equations for the coefficients of the transition matrix, which satisfies an inequality equivalent to the nondegeneracy of this matrix. (d) Theorem 2.2 naturally leads to the question of whether its statements (i) and (ii) will remain true if Cr1 isreplaced by Cr2 in them. As is shown below (see Theorem 2.14), concerning statement (ii) the answer is negative. Concerning the statement (i), at the time of this writing (September 2018) the author does not know the answer, and the following question seems to him to be of a principal importance: Qustion ([Po16, Quest. 1]). Is there an incompressible rational action of a finite group on A2? In the case of a positive answer to this question, the problem of finding all incom- pressible actions in the list found in [DI09] naturally arises. 2.9. Self-compressing linear actions. The remaining results of Section 3 are divided into two groups: one relates to the general case, the other to the case of Cr2. The following theorem applies to the general case. THREE PLOTS ABOUT THE CREMONA GROUPS 15 Theorem 2.8. Let G be a finite subgroup of GLn, n > 1. (a) If k[x1, . . . , xn]G d 6= 0, then G admits a polynomial homogeneous self-compres- sion An → An of degree d + 1. For d 6= 0, its is nontrivial. (b) If G divides d, then k[x1, . . . , xn]G d 6= 0. Proof. (a) We take a nonzero polynomial f ∈ k[x1, . . . , xn]G phism d and consider the mor- (2.37) In view of the G-invariance of f and the linearity of the action of G on An, for any g ∈ G and a ∈ An we have a 7→ f (a)a, ϕ : An → An, ϕ(g · a) (2.37) == f (g · a)(g · a) = f (a)(g · a) = g ·(cid:0)f (a)a(cid:1) (2.37) == g ·(cid:0)ϕ(a)(cid:1), i.e., ϕ is a G-equivariant morphism. From (2.37) and f ∈ k[x1, . . . , xn]d we obtain ϕ(ta) = td+1f (a)a = td+1ϕ(a) for any a ∈ An, t ∈ k, (2.38) so ϕ is a polynomial homogeneous map of degree d + 1. From (2.38) it follows that if a line L in An contains 0 and a point a ∈ U := {c ∈ An f (c) 6= 0} different from 0, then (i) ϕ(L) = L; (ii) the degree of the morphism ϕL : L → L is equal to d + 1. In view of (i), the image of ϕ contains a set U open in An, therefore, ϕ is dominant, hence is a self-compression of the group G. Suppose that ϕ is a birational isomorphism. Then the restriction of ϕ to some nonempty open subset U ′ of An is injective. Since An is irreducible, U ∩ U ′ 6= ∅. Let a ∈ U ∩ U ′. Then, in the previous notation, the degree of the morphism ϕL is equal to 1 in view of its injectivity on the subset L ∩ U ′, which is open in L. From (ii) we then obtain d = 0, which completes the proof of (a). (b) The kernel of the natural action of G on k[x1, . . . , xn]1 is trivial. Therefore there is a nonzero linear form ℓ ∈ k[x1, . . . , xn]1 such that its G-orbit contains exactly G is a nonzero G-invariant form of degree sG for (cid:3) elements. Therefore, (cid:0)Qg∈G g · ℓ(cid:1)s any integer s > 0, which proves (b). Corollary 2.9. Every finite subgroup of Crn, which is conjugate to a subgroup of the group GLn, is nontrivially self-compressible. 2.10. Compressing actions with a fixed point. The following result is an ap- plication of Theorem 2.8. Theorem 2.10. Every (faithful ) rational action of a finite group G on an n- dimensional irreducible variety, which has a fixed point, is obtained by a nontrivial base change from a (faithful ) linear action of the group G on an n-dimensional linear space. Proof. Let Y be an irreducible smooth complete variety and let G ֒→ Aut(Y ) be a regularization of the action , such that Y G 6= ∅. Let y ∈ Y G. We consider a nonempty open affine subset U of Y , containing y. Since Tg∈G g · U is a G-stable open affine subset, which contains y, replacing U by it, we can (and shall) assume that U is G-stable. Since U is dense in Y , the action of G on U is faithful. 16 VLADIMIR L. POPOV Let Ty,U be the tangent space to U at the point y. The tangent action τ : G → GL(Ty,U ) ⊂ Bir(Ty,U ) of the group G on the space Ty,U is faithful [Po141, Lem. 4]. According to [LPR06, Lem. 10.3], there is a G-equivariant dominant morphism α : U → Ty,U . In view of Theorem 2.8, the linearity of the action τ implies the existence of a nontrivial self- compression β : Ty,U → Ty,U of the group G (so that deg(β) > 1). Then β ◦ α : Y 99K Ty,U is a nontrivial (because deg(β ◦ α) = deg(β) deg(α) > 1) self-compression of the action . (cid:3) Corollary 2.11. Every incompressible rational action of a finite group on an irre- ducible variety has no fixed points. Remark 2.12. In [DD16] all rational actions of finite groups on A2, having a fixed point, are found. There are quite a few of them. By Theorem 2.10 they all are ob- tained by nontrivial base changes from the linear actions on A2 (the classification of which has long been known, see, e.g., [NPT08]). Recall that every finite Abelian group G decomposes into a direct sum of cyclic subgroups of orders m1, . . . , mr, where mi divides mi+1 for i = 1, . . . , r − 1, and m1 > 1 for G > 1. The sequence m1, . . . , mr is uniquely determined by G and called the sequence of invariant factors of the group G. The integer r is called its rank; the latter is equal to the minimal number of generators of the group G. For every integer n > r, we distinguish in GLn ⊂ Crn the following subgroup isomorphic to G: Tn(m1, . . . , mr) := {(t1x1, . . . , trxr, xr+1, . . . , xn) ti ∈ k, tmi i = 1, 1 6 i 6 r}. (2.39) Theorem 2.13. Let G be a finite Abelian group with the sequence of invariant fac- tors m1, . . . , mr. If a (faithful ) rational action of the group G on an n-dimensional irreducible variety has a fixed point, then n > r and is obtained by a nontrivial base change from a linear action λ : G ֒→ GLn(k) on An, such that λ(G) = Tn(m1, . . . , mr). Proof. We use the notation from the proof of Theorem 2.10. Fixing an isomorphism of the space Ty,U with An, we identify the group Bir(Ty,U ) with Crn. Since the groups G and τ (G) are isomorphic, they have the same invariant factors. According to [Po132, Thm. 1], every finite Abelian subgroup of Aff n, whose invariant factors are m1, . . . , mr, is transformed to the group Tn(m1, . . . , mr) by means of conjugation in the group Crn. Hence this is true for the subgroup τ (G). From here, arguing as in the proof of Theorem 2.10, we get the statement to be proved. (cid:3) 2.11. Compressing actions of cyclic groups. According to [Bl06, Thm. A], the sets of conjugacy classes of cyclic subgroups of Cr2 of some fixed orders d < ∞ are infininite and even there are parameter-dependent families of such classes (this is the case if d is even, d/2 is odd). The following theorem (a special case of which is proved in [Re04, Ex. 5]) implies that all these subgroups are obtained by the base changes from a single such subgroup. Theorem 2.14. Let n, m, d be any positive integers, and n > m. Every (faithful ) rational action of a finite cyclic group G of order d on an n-dimensional irreducible variety X is obtained by a nontrivial base change from a linear action λ : G ֒→ GLm(k) on Am such that λ(G) = Tm(d). THREE PLOTS ABOUT THE CREMONA GROUPS 17 Proof. Let G∨ be the group of all homomorphisms G → k×, and put K := k(X). If χ ∈ G∨, then (2.40) is a linear subspace of the linear space K over k. Since dim(hG · f i) < ∞ and G is Kχ := {f ∈ K g · f = χ(g)f for all g ∈ G} Abelian, we have K = Lχ∈G∨ Kχ, and (2.40) implies that {χ ∈ G∨ Kχ 6= 0} is a subgroup of G∨. Since G∨ is a cyclic group of order d and the action of G on K is faithful, this subgroup coincides with G∨. Let χ0 be a generator of G∨ and let f ∈ Kχ0 , f 6= 0. We can (and shall) assume that f /∈ k (and hence f is transcendental over k): if d > 1, this holds automatically; if d = 1, this follows from K 6= k. Since K/K G is a finite field extension, trdegkK G = trdegkK = n. Let h1, . . . , hn ∈ K G be a transcendence basis of the field K G over k. Then trdegkk(f, h1, . . . , hn) = n. Since f is transcendental over k, this and [La65, Chap. X, §1, Thm. 1] imply that after possibly renumbering elements h1, . . . , hn the subfield L := k(f, h1, . . . , hn−1) of K is a purely transcendental extension of the filed k of degree n. The construction implies the G-invariance of L and the faithfulness of the action of G on L. It follows from the definition of χ0 that the linear action µ : G ֒→ GLn(k) on An, given by the formula g · (a1, . . . , an) := (χ0(g−1)a1, a2, . . . , an), is faithful, µ(G) = Tn(d), and the map ϕ : X 99K An is a contraction of the action into the action µ. Since, in turn, µ is compressed into a linear action of the group G on Am by means of the projectionAn → Am, (a1, . . . , an) 7→ (a1, . . . , am), the assertion being proved now follows from Corollary 2.9. (cid:3) 2.12. Auxiliary statement: embeddings of G-modules into coordinate al- gebras. In what follows we shall need the following general statement: Lemma 2.15. If a finite group G acts regularly (and faithfully ) on an irreducible affine variety X, then every finite-dimensional kG-module M is isomorphic to a submodule of the kG-module k[X]. Proof. We can (and shall) assume that dim(X) > 0. Since tr degk(k(X)G) = dim(X) − dim(G) = dim(X) (see [PV94, Sect. 2.3, Cor.]), and k(X)G is the field of fractions of the algebra k[X]G (see [PV94, Lemma 3.2]), the latter is an infinite-dimensional linear space over k: dimk(k[X]G) = ∞. (2.41) From char k = 0 and the finiteness of the group G it follows that the kG-modules M and k[X] are completely reducible. Therefore, to prove the lemma, it suffices to establish that, for every nonzero simple kG-module S, the multiplicity of its occurrence in the S-isotypic component of the kG-module k[X] is infinite, which is equivalent to the infinite-dimensionality of this isotypic component as a linear space over k. In turn, for this is sufficient to establish that this S-isotypic component is nonzero. Indeed, the multiplication of functions defines on it a structure of a k[X]G- module. Therefore, if this S-isotypic component contains a nonzero function, its infinite-dimensionality follows from (2.41) and the absence of zero-divisors in k[X]. Having in view this reduction, we shall now prove that the S-isotypic component of the kG-module k[X] is indeed nonzero. The set of fixed points of every element of G is closed in X. Since G is fnite, X is irreducible, and the action of G on X is faithful, this implies that there exists a point 18 VLADIMIR L. POPOV x of X, whose G-stabilizer Gx is trivial. Its G-orbit G · x is a G-stable and (in view of the finiteness) closed subset of X. Its closedness implies that the homomorphism of kG-modules k[X] → k[G · x], f 7→ f G·x, is surjective. Therefore, to prove the nontriviality of the S-isotypic component of the kG-module k[X], it suffices to prove the nontriviality of the S-isotypic component of the kG-module k[G·x]. But it follows from the Frobenius duality that the multiplicity of the occurence of the kG-module S in the kG-module k[G · x] is equal to the dimension of the space of Gx-fixed points in the dual kG-module S ∗ (see [PV94, Thm. 3.12]). Since the group Gx is trivial, this shows that the specified multiplicity is equal to dim(S) > 0. Therefore, the S-isotypic component of the kG-module k[G · x] is indeed nonzero. This completes the proof of Lemma 2.15. (cid:3) 2.13. rdimk(G) and the existence of compressions. For any finite group G and any field ℓ we put rdimℓ(G) := min{m ∈ Z, m > 0 there is a group embedding G ֒→ GLm(ℓ)}. (2.42) In other words, rdimℓ(G) is the minimum of dimensions of faithful linear representa- tions of the group G over the field ℓ. Thus G has a faithful n-dimensional linear representation over ℓ if and only if n > rdimℓ(G). Note that if the group G is Abelian, then rdimk(G) is equal to its rank. Theorem 2.16. Let be a (faithful ) rational action of a finite group G on an n-dimensional irreducible variety. (i) If has a fixed point, then n > rdimk(G). (ii) If n > rdimk(G), then is compressible into a (faithful) rational action of a smaller dimension. (iii) If there is an n-dimensional faithful linear representation over k λ : G → GLn ⊂ Aut(An), (2.43) then either is compressible into a (faithful ) rational action of a smaller dimension or is obtained by a nontrivial base change from a linear action λ of the group G on An. Proof. Consider a regular (faithful) action of the group G on an n-dimensional smooth variety X, which is a regularization of the action . (i) If has a fixed point, we choose X so that X G 6= ∅. Let x ∈ X G. Since the tangent action G → GL(Tx,X) (2.44) of the group G on Tx,X is faithful [Po141, Lem. 4], the homomorphism (2.44) is injective. From this and (2.42) it follows that n = dim(X) = dim(Tx,X) > rdimk(G). (ii) In view of the inequality edk(G) 6 rdimk(G), which follows from (2.42) and the definition of edk(G) (see [BR97, Thm. 3.1(b)]), the statement (ii) follows from the inequality edk(X) 6 edk(G) proved in [BR97, Thm. 3.1(c)]. Other proof of the statement (ii), not using [BR97, Thm. 3.1(b, c)], is obtained in the course of the proof of (iii) below, see Remark 2.17. (iii) As in the proof of Theorem 2.10, replacing X by an appropriate invariant open subset, in the sequel we can (and shall) assume that X is affine. THREE PLOTS ABOUT THE CREMONA GROUPS 19 Since the representation λ is faithful, the dual representation λ∗ : G → GLr(k) is faithful as well. From Lemma 2.15 it follows that there a linear subspace L in k[X] with the following properties: (a) L is G-stable; (b) dim(L) = n; (c) the action of G on L is the representation λ∗. Consider in k[X] the k-subalgebra A generated by the subspace L. Since dim(L) < ∞, it is finitely generated and therefore isomorphic to the algebra of regular functions of an affine variety Y . It follows from (b) that dim(Y ) 6 n. (2.45) The identity embedding A ֒→ k[X] determines a dominant morphism ϕ : X → Y. From (a) the G-invarince of A follows. The action of G on A determines a regular action ϑ of the group G on the variety Y . The morphism ϕ is G-equivariant with respect to ϑ. In view of (c) and the faithfulness of the representation ∗, the action ϑ is faithful. Therefore, ϕ is a compression of into ϑ. Suppose that this compression does not reduce the dimension of the action , i.e., dim(Y ) = dim(X) = n. (2.46) Hence, in this case any basis of the linear space L over k consists of the elements of the algebra A, which are algebraically independent over k, because, by construction, this algebra is generated by them; its transcendental degree over k is then equal to the number of these elements: tr degk(A) = dim(Y ) (2.46) === n (b) = dim(L). This proves that there is a G-equivariant isomorphism α : Y → L∗, where L∗ is a kG-module dual to the kG-module L. From (c) it follows that the action of G on L∗ is the representtion (λ∗)∗ = λ. By Theorem 2.8 there is a G-equivariant dominant morphism ε : L∗ → L∗, which is not a birational isomorphism. Therefore, the composition ε ◦ α ◦ ϕ : X → L∗ is a nontrivial compression of the action into the action λ. (cid:3) Remark 2.17. In view of (2.42), as in the proof of statement (iii) we establish (under the assumption of affinity of X) the existence of -- an rdimk(G)-dimensional kG-submodule M in k[X], on which the action of the group G is faithful; -- an affine G-variety Z and a dominant G-equivariant morphism ψ : X → Z such that ψ∗(k[Z]) is the k-subalgebra of k[X] generated by the subspace M. Since the action of G on Z is faithful, ψ is a compression of the action . If n > rdimk(G), then ψ reduces the dimension of , because dim(Z) 6 dimk(M) = rdimk(G). This gives another proof of the statement (ii) of Theorem 2.16. 2.14. Compressing Abelian subgroups of rank 2 of the group Cr2. Theorem 2.14 answers the question about constructing finite Abelian subgroups of rank 1 of the Cremona group Crn by means of base changes. For n = 2, the next theorem answers the analogous question about the Abelian subgroups of rank 2, i.e., the noncyclic subgroups isomorphic to Z/aZ ⊕ Z/bZ, a > 2, b > 2. 20 VLADIMIR L. POPOV Theorem 2.18. Let be a (faithful ) rational action of a finite Abelian group G of rank 2 on A2 and let m1, m2 be the sequence of invariant factors of the group G. (i) In every of the following cases (a) G 6= 4; (b) G = 4 and has a fixed point the rational action is obtained by a nontrivial base change from a linear action λ : G ֒→ GL2(k) on A2 such that λ(G) = T2(m1, m2) (see (2.39)). In these cases the rational action is incompressible into a rational action of a smaller dimension. (ii) If G = 4 and does not have a fixed point, then G is a dihedral group (that is a Klein's Vierergruppe), and is obtained by a base change from the action γ : G → Cr1 on P1, for which the group γ(G) is generated by the elements σ, τ ∈ Aut(P1) given by the formulas σ · (a0 : a1) = (a0 : −a1), τ · (a0 : a1) = (a1 : a0) for all (a0 : a1) ∈ P1. (2.47) Proof. Since G is an Abelian group of rank 2, there exists a faithful linear represen- tation (2.43) with n = 2 and λ(G) = T2(m1, m2). If is compressible into a rational action ϑ of a smaller dimension, then ϑ is a faithful rational action of the group G on an irreducible algebraic curve C. In view of the existence of a dominant rational map A2 99K C (a compression of into ϑ), the curve C is rational. Therefore, we can (and shall) assume that C = P1 and hence G is isomorphic to a subgroup of Cr1 = Bir(P1) = Aut(P1) = PSL2. It follows from the well-known description of finite subgroups in PSL2 that noncyclic Abelian among them are only the subgroups conjugate to the dihedral subgroup of order 4, which is generated by the elements σ and τ given by formulas (2.47). They do not have fixed points on P1. In view of the "going down" property for fixed points (see [RY002, Prop. A.2]), it follows from (P1)G = ∅ that does not have a fixed point. If is not compressible into a rational action of smaller dimension, then by Theo- rem 2.16(iii), is obtained by a nontrivial base change from a linear action λ of the group G on A2. From this and the "going up" property for fixed points (see [RY002, Prop. A.4]) it follows that if in the considered case both invariant factors m1 and m2 are equal to the same prime numer, then has a fixed point. In particular, this is the case if G = 4. This completes the proof of Theorem 2.18. (cid:3) 2.15. Compressing other subgroups. The classification of finite Abelian sub- groups in Cr2 up to conjugacy is given in [Bl06]. In view of Theorems 2.14 and 2.18, it follows from it that among these subgroups only the subgroups isomorphic to Z/2dZ ⊕ (Z/2Z)2, d > 1; (Z/4Z)2 ⊕ Z/2Z; (Z/3Z)3, and (Z/2Z)4 remain unexplored for nontrivial compressibility. Their ranks are 3, 3, 3, and 4, respectively. By Theorem 2.16(i), all of these subgroups do not have fixed points. Theorem 2.19. Let G be a non-Abelian finite group different from dihedral group and admitting a faithful linear representation λ : G ֒→ GL2(k). Then every (faithful ) rational action of the group G on A2 is obtained by means of a nontrivial base change from its linear action λ on A2. THREE PLOTS ABOUT THE CREMONA GROUPS 21 Proof. The statement will follow from Theorem 2.16(iii) if we prove that from this theorem cannot be compressed into a faithful rational action of a smaller dimension. Arguing on the contrary, assume that such a compression exists. Then, as in the proof of Theorem 2.18 we obtain that G is isomorphic to a subgroup of the group Aut(P1) = PSL2. Since noncyclic and nondihedral finite subgroups in PSL2 are only the rotation groups of a regular tetrahedron, octahedron, and icosahedron, G is isomorphic to one of them. However, this is impossible because the rotation group of an icosahedron does not have nontrivial two-dimensional representations, and even though the rotation groups of a regular tetrahedron and octahedron have them, the kernels of these representations are nontrivial (their orders are 4), see, e.g., [Vi85]. Contradiction. (cid:3) 3. Group embeddings and the Cremona groups In this section, the characteristic k is zero. 3.1. Properties of abstract Jordan groups. We recall the concepts introduced in [Po11, Def. 2.1], [Po141, Def. 1]. For any finite group H, put mH := min S [H : S], where S runs over all normal Abelian subgroups of the group H. Definition 3.1. Let G be a group and let JG := sup mF F (3.1) (3.2) where F runs over all finite subgroups of G. If JG < ∞, then G is called a Jordan group (one also says that G has the Jordan property), and JG its Jordan constant. Lemma 3.2. For any groups G1, . . . , Gs, the inequality JG1×···×Gs > JG1 · · · JGs (3.3) holds (if JGi = ∞, then by definition, (3.3) means that JG1×···×Gs = ∞). Proof. Let Fi be a finite subgroup of Gi and let N be a normal Abelian subgroup of the finite subgroup F1 ×· · ·×Fs of the group G1 ×· · ·×Gs. Let πi : F1 ×· · ·×Fs → Fi be the projection to the ith factor. Then πi(N) is a normal Abelian subgroup of the group Fi, therefore, (3.1) implies the inequality Fi mFi πi(N) 6 . (3.4) From the inclusion N ⊆ π1(N) × · · · × πs(N) and the inequality (3.4), we get: N 6 π1(N) × · · · × πs(N) = sY i=1 πi(N) 6 sY i=1 Fi mFi = F1 × · · · × Fs mF1 · · · mFs . (3.5) It follows from (3.5) that [(F1 × · · · × Fs) : N] > mF1 · · · mFs, whence, in view of (3.1), we obtain the inequality mF1×···×Fs Now (3.3) follows from (3.6) and (3.2). > mF1 · · · mFs. (3.6) (cid:3) 22 VLADIMIR L. POPOV Remark 3.3. We set jG := sup F min A [F : A], where F runs over all finite subgroups of G, and A over all Abelian (not necessarily normal) subgroups of F . Clearly jG 6 JG. The conditions JG < ∞ and jG < ∞ turn out to be equivalent, see [Po11, Rem. 2.2]. Omitting the assumption of the normality of the subgroup N in the proof of Lemma 3.2, we obtain for any groups G1, . . . , Gs the proof of the inequality jG1×···×Gs > jG1 · · · jGs. Theorem 3.4. Let P be a Jordan group and let Q1, . . . , Qs be the groups, each of which contains a non-Abelian finite subgroup. Then the group Q1 × · · · × Qs is nonembeddable in the group P if s > log2(JP ). Proof. It follows from Definition 3.1 that JP < ∞ and, if Q1 ×· · ·×Qs is embeddable in P , then JQ1×···×Qs 6 JP . This and Lemma 3.2 yield the inequality JQ1 · · · JQs 6 JP . But from (3.1), (3.2), and the condition on Qi it follows that JQi > 2 for every i. Hence 2s 6 JP , and therefore, s 6 log2(JP ). (cid:3) Remark 3.5. The statement and the proof of Theorem 3.4 remain in effect, if in them JP is replaced by jP , and JQi by jQi. 3.2. Subgroups of the form G1 × · · ·×Gs and p-subgroups in the Cremona groups. We now apply the results from Subsection 3.1 to the Cremona groups. Theorem 3.6. Let X be a rationally connected variety X defined over k. Then there exists an integer bX , depending on X, such that every product of groups G1×· · ·×Gs, each of which contains a finite non-Abelian subgroup, is nonembeddable in the group Birk(X) if s > bX . Proof. This follows from Theorem 3.4 and the Jordan property of the group Birk(X) (see the footnote in the Introduction). (cid:3) Corollary 3.7. Let n be a positive integer. Then there exists an integer bn,k, depend- ing on n and the field k, such that every product of groups G1 × · · · × Gs, each of which contains a finite non-Abelian subgroup, is nonembeddable in the Cremona group Crn(k) if s > bn,k. Proof. This follows from Theorem 3.6 in view of rational connectedness of rational varieties. (cid:3) Remark 3.8. According to Theorem 3.4 and Remark 3.5, we can take bX = log2(jBirk(X)) in Theorem 3.6. The explicit upper bounds on jCr2(k) and jBirk(X) for rationally connected threefolds X, as well as their exact values under certain rest- rictions, are found in [Se09], [PS17]. For example, if k = k, then jCrn = 288 and 10368 respectively for n = 2 and 3. Corollary 3.9. For every prime integer p and rationally connected variety X defined over k, there exists a non-Abelian finite p-group nonembeddable in Birk(X). In parti- cular, for every integer n > 0, there exists a non-Abelian finite p-group nonem- beddable in the Cremona group Crn(k). Proof. This follows from Theorem 3.6, its Corollary 3.7, and the existence of finite non-Abelian p-groups. (cid:3) THREE PLOTS ABOUT THE CREMONA GROUPS 23 3.3. Applications: p-rank and embeddings of groups. Considering p-subgro- ups yields an obstacle to the existence of embeddings of groups. From here some applications are obtained. Namely, let p be a prime integer. Recall that a finite p-group is called elementary if it is Abelian and all its invariant factors (see above Subsection 2.10) are equal to p. Definition 3.10. For any group G and prime integer p, we call the p-rank of the group G and denote by rkp(G) the least upper bound of ranks of all elementary p-subgroups of the group G. Example 3.11. Let the group T be an n-dimensional torus in the category of either affine algebraic groups over k or real Lie groups (i.e., T is isomorphic to the product in the first case and the group {z ∈ C× z = 1} in the of n copies of the group k second case). It is not difficult to see that then rkp(T ) = n for every prime integer p. (cid:3) × Clearly, if G1 and G2 are two groups and rkp(G1) > rkp(G2) for some p, then G1 is nonembeddable in G2. The applications of this remark are based on the fact that in some cases rkp(G) can be explicitly computed or estimated. In particular, this is so for the Cremona groups: Theorem 3.12. For any integer n > 0, there exists a constant Rn, depending on n, such that rkp(H) = n for every (not necessarily closed ) subgroup H of Crn, containing an n-dimensional algebraic torus, and every p > Rn. In partucular, rkp(Crn) = rkp(Aut(An)) = n for every p > Rn. Proof. Let p be a prime integer. It follows from Example 3.11 and the condition on H that rkp(H) > n. On the other hand, combining [PS16, Thm. 1.10] with [Bi17, Cor. 1.3], we conclude that if p is bigger than a certain constant Rn, depending on n, then rkp(Crn) 6 n, and therefore, rkp(H) 6 n. This completes the proof. (cid:3) The other examples of groups G, for which one manages to compute their p-rank, are connected affine algebraic groups over k and connected real Lie groups. All the maximal tori in such a G are conjugate; let r(G) be the dimension of maximal tori of the group G. Recall that a prime integer p is called a torsion prime of the group G if G has a finite Abelian p-subgroup not lying in any maximal torus. The torsion primes of the group G divide the order of its Weyl group, so the set of all such primes is finite. Theorem 3.13. Let G be either a connected affine algebraic group over k or a connected real Lie group. Suppose that a prime integer p is not a torsion prime of the group G. Then rkp(H) = r(G) for any (not necessarily closed ) subgroup H of G, containing a maximal torus of the group G. Proof. It follows from Example 3.11 and the condition on H that rkp(H) > r(G). On the other hand, if F is a finite elementary p-subgroup of H, then F lies in a maximal torus of the group G because p is not a torsion prime of G. In view of Example 3.11, this implies that the rank of F does not exceed r(G). Therefore, rkp(H) 6 r(G). This completes the proof. (cid:3) 24 VLADIMIR L. POPOV From Theorems 3.12 and 3.13 we obtain: Corollary 3.14. Let k be an algebraically closed field of characteristic zero. We associate with each positive integer d any (not necessarily closed ) subgroup Hd from the following list: (1) a subgroup of the group Crd(k), containing a d-dimensional algebraic torus; (2) a subgroup of any connected affine algebraic group G over k with r(G) = d, containing a maximal torus of the goup G; (3) a subgroup of any connected real Lie group G with r(G) = d, containing a maximal torus of the goup G. Then the group Hn is nonembeddable in the group Hm if n > m. In particular, the following properties are equivalent: (a) the groups Hn and Hm are isomorphic; (b) n = m. Let us single out three particular cases as Corollaries 3.15, 3.16 3.17. Corollary 3.15 ([Ca14, Thm. B], [PS16, Rem. 1.11]). The group Crn is embeddable in the group Crm if and only if n 6 m. In particular, the groups Crn and Crm are isomorphic if and only if n = m. Corollary 3.16. The group Autn n 6 m. In particular, the groups Autn k and Autm k is embeddable in the group Autm k if and only if k are isomorphic if and only if n = m. Let K0 := k and Ki := k(x1, . . . , xi) for i = 1, . . . , n. For any ai, bi ∈ Ki−1, ai 6= 0, the map (1.2), where σi = aixi +bi for every i, is an element of the group Crn(k), and the set Bn(k) of all elements is a subgroup of Crn(k). According to [Po17], the group Bn := Bn(k) is a Borel subgroup of Crn; it contains the n-dimensional diagonal torus of the group GLn. Corollary 3.17. The group Bn is embeddable in the group Bm if and only if n 6 m. In particular, the groups Bn and Bm are isomorphic if and only if n = m. As another application, we get Corollary 3.18. If ϕ : Crn → Crm is a continuous epimorphism of groups endowed with the Zariski topology, then n = m and ϕ is an automorphism. Proof. In view of the topological simplicity of the group Crn (see [BZ18, Thm1. ]), the kernel of the epimorphism ϕ is trivial, and therefore it is an isomorphism of abstract groups. The statement now follows from Corollary 3.14. (cid:3) Remark 3.19. According to [Ur18], it follows from [BLZ18] the existence of an abstract group epimorphism Cr3 → Cr2. This shows that the assumption of conti- nuity in Corollary 3.18 is essential. On the other hand, Cr2 is a Hopfian abstract group, i. e., every its (not necessarily continuous) surjective endomorphism is an automorphism [D´e07]. In the following theorem is used not the exact value of the p-rank of a group, but its upper bound. THREE PLOTS ABOUT THE CREMONA GROUPS 25 Theorem 3.20. Let M be a connected compact n-dimensional topological manifold and let BM be the sum of its Betti numbers with respect to homology with coefficients in Z. If d > pn2 + 4n(n + 1)BM + n 2 + log2BM , (3.7) then the Cremons group Crd is nonembeddable in the homeomorphism group H (M) of the manifold M . Proof. Suppose that the inequality (3.7) holds. Let p > 2 be a prime integer satisfying the conditions: (i) p > Rd (see Theorem 3.12); i=0 Tors(Hi(M, Z)). It follows from [MS63, Thm. 2.5(3)] that the rank of any finite elementary p- (ii) p does not divide the order of the finite Abelian group Ln subgroup of the group H (M) does not exceed (pn2 + 4n(n + 1)BM,p + n)/2 + log2BM,p, where BM,p is the sum of Betti numbers of the manifold M with respect to homology with coefficients in Fp. It follows from (ii) and the universal coefficients theorem that BM,p = BM , whence, in view of (3.7), we obtain the inequality d > rkp(H (M)). From it, the condition (i), and Theorem 3.12 we infer that rkp(Crd) > rkp(H (M)). This completes the proof. (cid:3) According to [PS17, Thm. 1.10], [Bi17, Cor. 1.3], the constant Rd from Theorem 3.12 can be chosen so that, for any rationally connected d-dimensional variety X defined over k and any prime integer p > Rd, the inequality rkp(Birk(X)) 6 d holds. From this, another statement about nonembeddable groups follows: Theorem 3.21. Let X be a rationally connected n-dimensional variety X defined over k, and let p be a prime integer bigger than the constant Rn from Theorem 3.12 Then any product of groups G1 × · · · × Gs, each of which contains an element of order p, is nonembeddable in the group Birk(X) if s > d. 4. Connectedness of the Cremona groups 4.1. A new proof of the connectedness theorem. Two elements σ and τ ∈ Crn(k) are called linearly connected if there exist a k-defined open subset U of the affine line A1 and a k-morphism ϕ : U → Crn such that σ, τ ∈ ϕ(U(k)). It is easy to verify that the relation of being linearly connected is an equivalence relation on Crn(k) (see [Bl10, p. 363]). By definition, linear connectedness of the group Crn(k) means that there is only one equivalence class of this equivalence relation. Linear connectedness of the group Crn(k) implies its connectedness. Theorem 4.1 ([BZ18]). The Cremona group Crn(k) is liearly connected if the field k is infinite. Proof (different from the proof in [BZ18]). (a) First, id and every element σ ∈ Aff n(k) are linearly connected, because Aff n is an open subset of the (n2 + n)-dimensional affine space An of all affine maps An → An, and therefore, as ϕ one can take the identity map of the set U := ℓ ∩ Aff n, where ℓ is a line in An, containing σ and id. (b) Second, every element σ ∈ Birk(An) = Crn(k) is of the form σ = α ◦ θ ◦ τ , where α, τ ∈ Aff n(k), and θ = (θ1, . . . , θn) ∈ Crn(k) possesses the properties: 26 VLADIMIR L. POPOV (i) θ is defined at o; (ii) θ(o) = o; (iii) θ is ´etale at o, and doθ : To,An → To,An is the identity map. Indeed, since the map σ : An 99K An is k-birational, and the field k is infinite, there exists a point s ∈ An(k), at which σ is defined and ´etale (its existence is equivalent to the existence of a point in An(k) that is not zero of some nonzero polynomial from k[x1, . . . , xn]). Now, as α and τ we can take any elements from Aff n(k) such that τ −1(o) = s, α−1(σ(s)) = o, and the composition of the maps To,An doτ −1 −−−→ Ts,An dsσ−−→ Tσ(s),An dσ(s)α−1 −−−−−→ To,An is the identity map -- obviously, such elements exist. (c) We will now show that id and the element θ ∈ Crn(k) specified in (b) are linearly connected. Clearly, in view of (a) and (b), this will complete the proof of Theorem 4.1. its completion with respect to its maximal ideal. The set of functions x1, . . . , xn is a system of local parameters of the variety An at the point o. Therefore, we can Let O and bO be respectively the local ring of the variety An at the point o and (and shall) assume that bO = k[[x1, . . . , xn]] and O is the subring of bO formed by the Taylor series at the point o of all the functions from O with respect to this system of local parameters. We have Ok := O ∩ k(An) ⊂ k[[x1, . . . , xn]]. It follows from (i) that θi ∈ Ok for every i = 1, . . . , n, so we have θi = Fi(x1, . . . , xn) ∈ k[[x1, . . . , xn]] In view of (ii) and (iii), the series Fi(x1, . . . , xn) has the form Fi(x1, . . . , xn) = xi +X d>2 Fi,d(x1, . . . , xn), (4.1) (4.2) where Fi,d(x1, . . . , xn) is a form of degree d in x1, . . . , xn with the coefficients in k, so we have Fi,d(tx1, . . . , txn) = tdFi,d(x1, . . . , xn) for any t ∈ k. (4.3) From (4.1), (4.2), (4.3) it follows that, for any t ∈ k, the series txi +X d>2 tdFi,d(x1, . . . , xn) ∈ bO lies in O, and for t ∈ k, it lies in Ok. This implies that the series xi +X d>2 td−1Fi,d(x1, . . . , xn) also possesses the same properties. Therefore, for every t ∈ k, we obtain a rational map (t) : An 99K An, (t)i = xi +Pd>2 td−1Fi,d(x1, . . . , xn), In reality, (t) ∈ Crn for every t. Indeed, (4.4) yields i = 1, . . . , n. (4.4) (0) = (x1, . . . , xn) (1.2) == id ∈ Crn. (4.5) THREE PLOTS ABOUT THE CREMONA GROUPS 27 If t 6= 0 and ϑ(t) := (tx1, . . . , txn) ∈ GLn, then from (1.3), (4.1), (4.2), and (4.4) we obtain (4.6) Since the left-hand side of the equality (4.6) lies in Crn, the same is true for the right one. ϑ(t−1) ◦ θ ◦ ϑ(t) = (t). Thus, a mapping ϕ : A1 → Crn, t 7→ (t), arises. In view of (4.4), it is a k- morphism. From (4.5) and the equality (1) = θ (following from (4.4), (4.2), (4.1)) it now follows that θ and id are linearly connected. (cid:3) 4.2. The case of a finite field k. The following examples, belonging to A. Borisov [Bo17], show that the condition of infinity of k cannot be discarded in the given above proof. Examples. Let k = Fq, n = 2. Then the birational self-map τ := (x1, x2 − 1/(xq 1 − x1)) ∈ Cr2(Fq) is not defined at all points of A2(Fq), and the birational self-map τ := ((xq 1 − x1)x2) ∈ Cr2(Fq) is not ´etale at all such points. 1 − x1)x1x2, (xq References [Al23] J. W. Alexander, On the deformation of an n-cell, Proc. Nat. Acad. Sci. USA 9 (1923), 406 -- 407. C. Birkar, Birational geometry of algebraic varieties, arXiv:1801.00013 (2017). [Bi17] [BM97] E. Bierstone, P. D. Milman, Canonical desingularization in characteristic zero by blowing [Bl06] [Bl10] up the maximum strata of a local invariant, Invent. Math. 128 (1997), no. 2, 207 -- 302. J. Blanc, Finite Abelian subgroups of the Cremona group of the plane, Th`ese No. 3777, Gen`eve, 2006, arXiv:math.AG/0610368. J. Blanc, Groupes de Cremona, connexit´e et simplicit´e, Ann. Sci. ´Ec. Norm. Sup´er. (4) 43 (2010), no. 2, 357 -- 364. [BLZ18] J. Blanc, S. Lamy, S. Zimmermann, Abelian quotients of the Cremona groups in higher dimension, https://algebra.dmi.unibas.ch/blanc/articles/NonSimplicityBirPn Oberwolfach.pdf. J. Blanc, S. Zimmermann, Topological simplicity of the Cremona groups, to appear in Amer. J. Math. bf 140 (2018), no. 5, arXiv:1511.08907 (2015). J. Blanc, J.-P. Furter, Topologies and structures of the Cremona groups, Ann. of Math. 178 (2013), no. 3, 1173 -- 1198. [BF13] [BZ18] [Bo17] A. Borisov, Letters of May 1 and 2, 2017 to V. Popov. [BR97] [Ca14] [D´e07] [DI09] [DD16] J. Buhler, Z. Reichstein, On the essential dimension of a finite group, Compositio Math. 106 (1997), 159 -- 179. S. Cantat, Morphisms between Cremona groups and a characterization of rational varie- ties, Compositio Math. 150 (2014), 1107 -- 1124. J. D´eserti, Le groupe de Cremona est Hopfien, Compt. Rend. Math. 344 (2007), no. 3, 153 -- 156. I. V. Dolgachev, V. A. Iskovskikh, Finite subgroups of the plane Cremona group, in: Al- gebra, Arithmetic, and Geometry: in honor of Yu. I. Manin, Vol. I, Progress in Math., Vol. 269, 2009, Birkhauser Boston, Boston, 2009, pp. 443 -- 548. I. Dolgachev, A. Duncan, Fixed points of a finite subgroup of the plane Cremona group, Algebraic Geometry 3 (2016), no. 4, 441 -- 460. [GA16] M. Garcia-Armas, Strongly incompressible curves, Canad. J. Math. 68 (2016), 541 -- 570. [La65] [LPR06] N. Lemire, V. L. Popov, Z. Reichstein, Cayley groups, J. Amer. Math. Soc. 19 (2006), S. Lang, Algebra, Addison-Wesley, Mass., 1965. no. 4, 921 -- 967. [MS63] L. N. Mann, J. C. Su, Actions of elementary p-groups on manifolds, Trans. Amer. Math. Soc. 106 (1963), 115 -- 126. 28 VLADIMIR L. POPOV [NPT08] K. A. Nguyen, M. van der Put, J. Top, Algebraic subgroups of GL2(C), Indag. Mathem., N.S. 19 (2008), no. 2, 287 -- 297. [Po11] V. L. Popov, On the Makar-Limanov, Derksen invariants, and finite automorphism groups of algebraic varieties, in: Affine Algebraic Geometry: The Russell Festschrift, CRM Proc. and Lect. Notes 54 (2011), 289 -- 311. [Po131] V. L. Popov, Some subgroups of the Cremona groups, in: Affine Algebraic Geometry, Proceedings of the conference on the occasion of M. Miyanishi's 70th birthday, Osaka, Japan, 3 -- 6 March 2011, World Scientific Publishing Co., Singapore, 2013, 213 -- 242. [Po132] V. L. Popov, Tori in the Cremona groups, Izv. Math. 77 (2013), no. 4, 742 -- 771. [Po141] V. L. Popov, Jordan groups and automorphism groups of algebraic varieties, in: Auto- morphisms in Birational and Affine Geometry (Levico Terme, Italy, October 2012), in Springer Proceedings in Mathematics & Statistics, (Springer, Heidelberg, 2014), Vol. 79, pp. 185 -- 213. [Po142] V. L. Popov, On infinite dimensional algebraic transformation groups, Transform. Groups 19 (2014), no. 2, 549 -- 568. [Po16] V. L. Popov, Compressible finite subgroups of the Cremona groups, problem session, Cremona conference -- Basel 2016, Basel, Switzerland, September 5 -- 16, 2016, https:/ /algebra.dmi. unibas.ch/blanc/cremonaconference/index.html [Po17] V. L. Popov, Borel subgroups of Cremona groups, Mathematical Notes 102 (2017), no. 1, 60 -- 67. [PV94] V. L. Popov, E. B. Vinberg, Invariant theory, in: Algebraic Geometry IV, Encyclopaedia of Mathematical Sciences, Vol. 55, Springer-Verlag, Berlin, 1994, pp. 123 -- 284. [PS16] Yu. Prokhorov, C. Shramov, Jordan property for Cremona groups, Amer. J. Math. 138 (2016), 403 -- 418. [PS17] Yu. Prokhorov, C. Shramov, Jordan constant for Cremona group of rank 3, Mosc. Math. J. 17 (2017), no. 3, 457 -- 509. [Re001] Z. Reichstein, On the notion of essential dimension for algebraic groups, Transform. [Re04] Groups 5 (2000), no. 3, 265 -- 304. Z. Reichstein, Compression of group actions, in: Invariant Theory in All Characteristics, CRM Proceedings and Lecture Notes, Vol. 35, 2004, AMS, Providence. RI, pp. 199 -- 202. [RY002] Z. Reichstein, B. Youssin, Essential dimensions of algebraic groups and a resolution the- orem for G-varieties, Canad. J. Math. 52 (2000), no. 5, 1018 -- 1056. [Ro56] M. Rosenlicht, Some basic theorems on algebraic groups, Amer. J. Math. 78 (1956), 401 -- [Se09] [Se10] [Sh82] 443. J.-P. Serre, A Minkowski-style bound for the orders of the finite subgroups of the Cremona group of rank 2 over an arbitrary field, Mosc. Math. J. 9 (2009), no. 1, 193 -- 208. J.-P. Serre, Le groupe de Cremona et ses sous-groupes finis, S´eminaire N. Bourba- ki 2008/09, Exp. no. 1000, Ast´erisque l 332 (2010), 75 -- 100. I. R. Shafarevich, On some infinite-dimensional groups II, Math. USSR Izv. 18 (1982), 214 -- 226. I. R. Shafarevich, Basic Algebraic Geometry, 3rd edition, Springer, Heidelberg, 2013. [Sh13] [Sp87] T. A. Springer, Poincar´e series of binary polyhedral groups and McKay's correspondence, Math. Ann 287 (1987), 99 -- 116. [Ur18] Ch. Urech, Letter of October 11, 2018 to V. L. Popov. [Wa67] B. L. van der Waerden, Algebra, Springer-Verlag, Berlin, 1971. [Vi85] E. B. Vinberg, Linear Representations of Groups, Birkhauser Verlag, Basel, 1989. Steklov Mathematical Institute, Russian Academy of Sciences, Gubkina 8, Mos- cow, 119991, Russia E-mail address: [email protected]
1211.5765
1
1211
2012-11-25T14:07:24
Holography principle for twistor spaces
[ "math.AG", "math.CV", "math.DG" ]
Let $S$ be a smooth rational curve on a complex manifold $M$. It is called ample if its normal bundle is positive. We assume that $M$ is covered by smooth holomorphic deformations of $S$. The basic example of such a manifold is a twistor space of a hyperkahler or a 4-dimensional anti-selfdual Riemannian manifold $X$ (not necessarily compact). We prove "a holography principle" for such a manifold: any meromorphic function defined in a neighbourhood $U$ of $S$ can be extended to $M$, and any section of a holomorphic line bundle can be extended from $U$ to $M$. This is used to define the notion of a Moishezon twistor space: this is a twistor space $\Tw(X)$ admitting a holomorphic embedding to a Moishezon variety $M'$. We show that this property is local on $X$, and the variety $M'$ is unique up to birational transform. We prove that the twistor spaces of hyperkahler manifolds obtained by hyperkahler reduction of flat quaternionic-Hermitian spaces by the action of reductive Lie groups (such as Nakajima's quiver varieties) are always Moishezon.
math.AG
math
M. Verbitsky Holography principle for twistor spaces Holography principle for twistor spaces Misha Verbitsky1 [email protected] Abstract Let S be a smooth rational curve on a complex manifold M . It is called ample if its normal bundle is positive: N S =LO(ik), ik > 0. We assume that M is covered by smooth holomorphic deformations of S. The basic exam- ple of such a manifold is a twistor space of a hyperkahler or a 4-dimensional anti-selfdual Riemannian manifold X (not necessarily compact). We prove "a holography prin- ciple" for such a manifold: any meromorphic function de- fined in a neighbourhood U of S can be extended to M , and any section of a holomorphic line bundle can be ex- tended from U to M . This is used to define the notion of a Moishezon twistor space: this is a twistor space admitting a holomorphic embedding to a Moishezon va- riety M ′. We show that this property is local on X, and the variety M ′ is unique up to birational transform. We prove that the twistor spaces of hyperkahler manifolds obtained by hyperkahler reduction of flat quaternionic- Hermitian spaces by the action of reductive Lie groups (such as Nakajima's quiver varieties) are always Moishe- zon. Contents 1 Introduction 1.1 Quasilines on complex manifolds . . . . . . . . . . . . . . . . . . . 1.2 Holography principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Moishezon twistor spaces 2 Twistor spaces and ample rational curves 2.1 Hyperkahler manifolds . . . . . . . . . . . . . . . . . . . . . . . . . 2 2 3 4 5 5 1Partially supported by RFBR grants 12-01-00944-, 10-01-93113-NCNIL-a, and AG Laboratory NRI-HSE, RF government grant, ag. 11.G34.31.0023. -- 1 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Twistor spaces 2.3 Geometry of twistor spaces . . . . . . . . . . . . . . . . . . . . . . 2.4 Quasilines and ample curves in the twistor spaces . . . . . . . . . . 3 Holography principle for manifolds with ample rational curves 3.1 Holography principle for line bundles . . . . . . . . . . . . . . . . . 3.2 The local holography principle . . . . . . . . . . . . . . . . . . . . 3.3 Holography principle for meromorphic functions . . . . . . . . . . . 4 Moishezon twistor spaces 4.1 Vector bundles in a neighbourhood of an ample curve . . . . . . . 4.2 Algebraic dimension of the field of meromorphic functions . . . . . 4.3 Moishezon manifolds and ample rational curves . . . . . . . . . . . 4.4 Moishezon twistor spaces . . . . . . . . . . . . . . . . . . . . . . . 4.5 Twistor spaces and hyperkahler reduction . . . . . . . . . . . . . . 5 Appendix: Formal geometry and holography principle (by Dmitry Kaledin) 1 Introduction 7 8 8 9 9 10 11 12 12 14 15 16 17 21 1.1 Quasilines on complex manifolds The present paper was written as an attempt to answer the following ques- tion. Let S ⊂ M be a smooth rational curve in a complex manifold, with normal bundle NS isomorphic to O(1)n.1 Is there a notion of a normal form for a tubular neighbourhood of such a curve? When the normal bundle is O(−1)n instead of O(1)n, a tubular neigh- bourhood of the curve has a normal form, obtained by blowing down this curve to a point, and taking a sufficiently small Stein neighbourhood of this point in the corresponding singular variety. When NS = O(1)n, no such nor- mal form can be obtained. In fact, the birational type of the manifold can be reconstructed from a complex analytic (and even formal) neighbourhood of S. This was known already to Hartshorne ([Har, Theorem 6.7]). However, one can associate with a quasiline an infinite-dimensional bundle OS(M) over S ∼= CP 1, with OS(M) = lim ← OM /I n SOM , 1Such rational curves are called quasilines, see e.g. [BBI]. -- 2 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces where IS is an ideal sheaf of S. This bundle is called a formal completion of M at S. It is not hard to observe that the space of sections H 0(S, OS(M)) is finite- dimensional, and, moreover, H 0(S, OS(M)⊗OS O(i)) is finite-dimensional for each i (Proposition 4.1). For a long time, we expected that the algebra AS := Li H 0(S, OS(M) ⊗OS O(i)) would provide a sort of an algebraic "normal form" of S in M, in such a way that the algebraic structure on OS(M) can be reconstructed from this ring. In this paper, we show that this approach works when M is a Moishezon manifold, and M can be reconstructed from the ring AS, up to birational isomorphism (Subsection 4.3; this is not very surprising due to the above- mentioned theorem of Hartshorne, [Har, Theorem 6.7]). For non-Moishezon M, such as a twistor space of a simply connected, In compact hyperkahler manifold, this conjecture is spectacularly wrong. this case, AS =Li H 0(CP 1,O(i)), and this ring has no information about M whatsoever (Proposition 2.13). 1.2 Holography principle The main technical tool of the present paper is the following theorem, called the holography principle. Recall that an ample rational curve on a complex variety M is a smooth curve S ∼= CP 1 ⊂ M such that the normal bundle NS is decomposed as NS =Lk O(ik), with all ik > 0. Theorem 1.1: Let S ⊂ M be an ample curve in a simply connected complex manifold, and U its connected neighbourhood. Suppose that M is covered by smooth complex-analytic deformations of S. Then (i) For any holomorphic vector bundle B on M, the restriction map H 0(M, B) −→ H 0(U, B) is an isomorphism. (ii) Let Mer(M), Mer(U) be the fields of meromorphic functions on M and U. Then the restriction map Mer(M) −→ Mer(U) is an isomorphism. Proof: For line bundles, Theorem 1.1 (i) is implied by Theorem 3.1. For general vector bundles, an elegant argument is given by D. Kaledin in the -- 3 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces appendix to this paper (Section 5). Holography principle for meromorphic functions is proven in Theorem 3.4. The holography principle is not very surprising of one looks at the neigh- bourhood of S from the point of view of complex analysis. The normal bundle to S is obviously positive. Choose a Hermitian metric h on a neigh- bourhood of S such that the Chern connection on NS induced by h has positive curvature. Let dS : M −→ R be the Riemannian distance to S in this Hermitian metric. Since dS around S is close to the distance in NS, the form ddcdS has n − 1 positive and 1 negative eigenvalue in a sufficiently small neighbourhood of S ([D]). This means that S has a neighbourhood U with a smooth boundary ∂U such that the Levi form on ∂U has one negative and dim M − 2 positive eigenvalues. Then, a holomorphic function defined on an open subset of U and continuous on ∂U can be extended outside of a boundary; at least, this is the expectation one has from the solution of the Levi problem. 1.3 Moishezon twistor spaces Definition 1.2: Let M be a compact complex manifold. Define the alge- braic dimension as a(M) := degtr Mer(M), where degtr Mer(M) denotes the transcendence degree of the field of global meromorphic functions on M. Definition 1.3: A Moishezon variety is a compact complex variety satis- fying a(M) = dim M. The notion of a Moishezon manifold, as it is usually stated, makes no sense for non-compact varieties. Indeed, degtr Mer(M) = ∞ even if M is an open disk. However, when M contains an ample curve, the situation changes drasti- cally. Theorem 1.4: Let M be a complex manifold containing an ample rational curve. Then a(M) 6 dim M. Moreover, if a(M) = dim M, there exists an open embedding of M to a Moishezon variety M ′ which satisfies Mer(M ′) = Mer(M). Proof: See Theorem 4.8. -- 4 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces Let now M be a twistor space of X. Here, X can be either a hypercomplex (hyperkahler) manifold, a quaternionic, or quaternionic-Kahler manifold, or Riemannian anti-selfdual manifold. We are not very specific, because the only thing about M which is used is existence of a large number of quasilines. The twistor spaces are complex manifold covered by quasilines, usually non- Kahler and non-quasiprojective (see Proposition 2.13 and Theorem 4.31). Since M = Tw(X) is covered by quasilines which are by definition ample, we can apply Theorem 4.6, and obtain that degtr Mer(M) 6 dim M. We call M a Moishezon twistor space if degtr Mer(M) = dim M. This is equiva- lent to an existence of an open embedding M −→ M ′ of M to a Moishezon manifold (Theorem 1.4). Moishezon twistor spaces for compact 4-dimensional anti-selfdual man- ifolds were discovered by Y.-S. Poon in [P], and much studied since then. Structure theorems about such manifolds were obtained by F. Campana, [C], and partial classification results by N. Honda (see e.g. [Ho]). The definition given above extends the class of "Moishezon twistor man- ifolds" significantly. Let, for instance, M := V ///G be a hyperkahler man- ifold which can be obtained using the hyperkahler reduction, where V is a flat hyperkahler manifold, and G a compact Lie group (such as the Naka- jima quiver variety). Then the twistor space Tw(M) is always Moishezon (Theorem 4.29). Acknowledgements: Many thanks to Hans-Joachim Hein, Claude Le- Brun, Nobuhiro Honda and Dima Kaledin for interesting discussions on the subject of this article. 2 Twistor spaces and ample rational curves 2.1 Hyperkahler manifolds Definition 2.1: Let M be a manifold, and I, J, K ∈ End(T M) endomor- phisms of the tangent bundle satisfying the quaternionic relation I 2 = J 2 = K 2 = IJK = − IdT M . The manifold (M, I, J, K) is called hypercomplex if the almost complex structures I, J, K are integrable. If, in addition, M is equipped with a -- 5 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces Riemannian metric g which is Kahler with respect to I, J, K, the manifold (M, I, J, K, g) is called hyperkahler. Consider the Kahler forms ωI, ωJ , ωK on M: ωI(·,·) := g(·, I·), ωJ(·,·) := g(·, J·), ωK(·,·) := g(·, K·). An elementary linear-algebraic calculation implies that the 2-form Ω := ωJ + √−1 ωK is of Hodge type (2, 0) on (M, I). This form is clearly closed and non-degenerate, hence it is a holomorphic symplectic form. In algebraic geometry, the word "hyperkahler" is essentially synonymous with "holomorphically symplectic", due to the following theorem, which is implied by Yau's solution of Calabi conjecture. Theorem 2.2: Let (M, I) be a compact, Kahler, holomorphically symplectic manifold. Then there exists a unique hyperkahler metric on (M, I) with the same Kahler class. Proof: See [Y], [Bes]. Remark 2.3: The hyperkahler metric is unique, but there could be several hyperkahler structures compatible with a given hyperkahler metric on (M, I), if the holonomy of its Levi-Civita connection is strictly less than Sp(n). Definition 2.4: Let M be a hypercomplex manifold, and L a quaternion sat- isfying L2 = −1. Then L = aI +bJ +cK, a2 +b2 +c2 = 1. The corresponding complex structure on M is called an induced complex structure. The space M, considered as a complex manifold, is denoted by (M, L). The set of induced complex structures is naturally identified with S2, which we often consider as CP 1 with the standard complex structure. Definition 2.5: ([V1]) Let X ⊂ M be a closed subset of a hyperkahler manifold M. Then X is called trianalytic if X is a complex analytic subset of (M, L) for every induced complex structure L. Trianalytic subvarieties were a subject of a long study. Most importantly, consider a generic induced complex structure L on M. Then all closed com- plex subvarieties of (M, L) are trianalytic. Moreover, a trianalytic subva- riety can be canonically desingularized ([V2]), and this desingularization is hyperkahler. -- 6 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces Theorem 2.6: ([V1], [V3]) Let (M, I, J, K) be a hyperkahler manifold (not necessarily compact). Then there exists a countable subset R ⊂ CP 1, such that for any induced complex structure L /∈ R, all compact complex subva- rieties of (M, L) are trianalytic. Remark 2.7: For hypercomplex manifolds, Theorem 2.6 is (generally speak- ing) false, though for manifolds with trivial canonical bundle a weaker form of this result was obtained ([SV]). 2.2 Twistor spaces Definition 2.8: Let M be a Riemannian 4-manifold. Consider the action of the Hodge ∗-operator: ∗ : Λ2M −→ Λ2M. Since ∗2 = 1, the eigenvalues are ±1, and one has a decomposition Λ2M = Λ+M ⊕ Λ−M onto selfdual (∗η = η) and anti-selfdual (∗η = −η) forms. Remark 2.9: If one changes the orientation of M, leaving metric the same, Λ+M and Λ−M are exchanged. Therefore, their dimensions are equal, and dim Λ2M = 6 implies dim Λ±(M) = 3. Remark 2.10: Using the isomorphism Λ2M = so(T M), we interpret η ∈ Λ2 mM as an endomorphisms of TmM. Then the unit vectors η ∈ Λ+ mM correspond to oriented, orthogonal complex structures on TmM. Definition 2.11: Let Tw(M) := SΛ+M be the set of unit vectors in Λ+M. At each point (m, s) ∈ Tw(M), consider the decomposition Tm,s Tw(M) = TmM ⊕ TsSΛ+ mM, induced by the Levi-Civita connection. Let Is be the complex structure on TmM induced by s, ISΛ+ mM the complex structure on SΛ+ mM = S2 induced by the metric and orientation, and be equal to Is ⊕ ISΛ+ the twistor space of M. I : Tm,s Tw(M) −→ Tm,s Tw(M) mM . An almost complex manifold (Tw(M),I) is called Given a hyperkahler or hypercomplex manifold (Definition 2.1), one de- fines its twistor space in a similar manner. -- 7 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces Definition 2.12: A twistor space Tw(M) of a hypercomplex manifold M is S2 × M equipped with a complex structure which is defined as follows. Consider the complex structure Im : TmM → TmM on M induced by J ∈ S2 ⊂ H. Let IJ denote the complex structure on S2 = CP 1. The operator ITw = Im ⊕ IJ : Tx Tw(M) → Tx Tw(M) satisfies I 2 Tw = − Id. It defines an almost complex structure on Tw(M). The almost complex structure on the twistor space of a Riemannian 4- manifold X is integrable whenever X is anti-selfdual ([AHS]) For a hyper- complex manifold it is integrable as well ([K]). Twistor spaces are the main example of the geometries we are working with. 2.3 Geometry of twistor spaces Proposition 2.13: Let Tw(M) be a twistor space of a compact hyperkahler manifold. Then (i) Tw(M) is non-Kahler. (ii) The algebraic dimension of Tw(M) is 1. Proof of (i): Let ω be the standard Hermitian form of Tw(M). Then ddcω is a positive (2,2)-form ([KV, (8.2)]). For any Kahler form ω0 on Tw(M), this would imply ZTw(M ) d(cid:16)ωdimC M −1 0 ∧ dcω(cid:17) =ZTw(M ) ωdimC M −1 0 ∧ ddcω > 0, which is impossible by Stokes' theorem. Proof of (ii): See Theorem 4.11. 2.4 Quasilines and ample curves in the twistor spaces Definition 2.14: An ample rational curve on a complex manifold M is a smooth curve S ∼= CP 1 ⊂ M such that its normal bundle NS satisfies NS =Ln−1 k=1 O(ik), with all ik > 0 (see [Ko]). It is called a quasiline if all ik = 1. -- 8 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces Claim 2.15: Let M be a twistor space of a hyperkahler or 4-dimensional ASD manifold, m ∈ M a point, and Sm the corresponding S2 in Tw(M) = CP 1 × M. Then Sm is a quasiline. Proof: Since the claim is essentially infinitesimal, it suffices to check it when M is flat. Then Tw(M) = Tot(O(1)⊕2n) ∼= CP 2n+1\CP 2n−1, and Sm is a section of O(1)⊕2n. Existence of quasilines in twistor spaces is a very strong condition, and can be used to obtain all kinds of geometric information; for example, see [C] and [V4]. 3 Holography principle for manifolds with ample rational curves 3.1 Holography principle for line bundles Throughout this paper, all neighbourhoods and manifolds are silently as- sumed to be connected. One of the main results of the present paper is the following theorem. Theorem 3.1: (holography principle for line bundles) Let S ⊂ M be an ample rational curve in a simply connected complex manifold, which is cov- ered by smooth, ample deformations of S, and L a holomorphic line bundle on M. Consider an open neighbourhood U ⊃ S. Then the restriction map H 0(M, L) −→ H 0(U, L) is an isomorphism. We deduce Theorem 3.1 from the following local result (Proposition 3.3). Remark 3.2: Since S is an ample curve, S can be deformed in any normal direction. Therefore, there exists an open neighbourhood U ⊃ S which is contained in a union of the set S of all smooth, ample deformations of S intersecting S. Further on, we choose this neighbourhood in such a manner that any S1 ∈ S can be connected to S by a continuous family of deforma- tions intersecting S. -- 9 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces Proposition 3.3: Let S ⊂ M be an ample rational curve in a connected com- plex manifold, which is covered by deformations of S, and L a holomorphic line bundle on M. Consider a neighbourhood U ⊃ S which is is contained in a union of the set of all deformations of S intersecting S (Remark 3.2). Then for any smaller open neighbourhood V ⊂ U of S, the restriction map H 0(U, L) −→ H 0(V, L) is an isomorphism. Proof of an implication "Proposition 3.3 ⇒ Theorem 3.1". Step 1: Choose a continuous, connected family Sb of ample curves parametrized by B such that Sb∈B Sb = M, and choose a tubular neigh- bourhood Ub for each Sb, continuously depending on b. Then the inter- section Ub ∩ Ub′ for sufficiently close b, b′ always contains Sb and Sb′. By Proposition 3.3, Ub can be chosen in such a way that H 0(Ub ∩ Ub′, L) = H 0(Ub, L) = H 0(Ub′, L). Step 2: Since B is connected, all the spaces H 0(Ub, L) are isomorphic, and these isomorphisms are compatible with the restrictions to the inter- sections Ub ∩ Ub′. Let now f ∈ H 0(Ub, L), and let Mf be the domain of holomorphy for f , that is, a maximal domain (non-ramified over M) such that f admits a holomorphic extension to Mf . Since ∪Ub = M, and f can be holomorphically extended to any Ub, the domain Mf is a covering of M. Now, Theorem 3.1 follows, because M is simply connected. 3.2 The local holography principle statement of Proposition 3.3 is vacuous. Therefore, we may always assume To prove Theorem 3.1 it remains to prove Proposition 3.3. If deg L(cid:12)(cid:12)S < 0, the > 0, hence l(cid:12)(cid:12)S is generated by global sections. Let S(M) be the that deg L(cid:12)(cid:12)S space of deformations of the ample curve S which remain smooth, and SS(M) the space of pairs {(x, S1) : S1 ∈ S(M), x ∈ S1}. Consider the natural forgetful maps τ1 : SS(M) −→ M, τ2 : SS(M) −→ S(M), and let E be the bundle τ2∗τ ∗ 1 L on S(M). Denote by degS L the degree of the restriction of L to S. Since dim H 0(S1, L) = deg L + 1, E is a (deg L + 1)-dimensional vector bundle. Given a section f of L on M, denote the corresponding section of E 1 f . When degS L = d, the value of f at S1 is uniquely determined by f := τ2∗τ ∗ is d + 1-dimensional, and any section h ∈ H 0(S1, L) is uniquely determined by by the restriction of f to any d + 1 distinct points of S1. Indeed, E(cid:12)(cid:12)(cid:12)S1 -- 10 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces its values in d + 1 points. This gives a map −→ H 0(S1, L). (3.1) ϕ(cid:16)f(cid:12)(cid:12)(cid:12)z1 , f(cid:12)(cid:12)(cid:12)z2 , f(cid:12)(cid:12)(cid:12)z2 , ..., f(cid:12)(cid:12)(cid:12)zd+1(cid:17) Now, let f be a section of L on V . For any S1 ∈ S(M) intersecting V , we choose d1 distinct points z1, ..., zd+1 ∈ S1 ∩ V , and consider the section L(cid:12)(cid:12)(cid:12)z1 × L(cid:12)(cid:12)(cid:12)z2 × ... × L(cid:12)(cid:12)(cid:12)zd+1 , ..., f(cid:12)(cid:12)(cid:12)zd+1(cid:17) ∈ H 0(S1, L) defined using (3.1). When S1 ⊂ V , (z1, ..., zd+1) −→ ϕ(cid:16)f(cid:12)(cid:12)(cid:12)z1 this section is independent from the choice of z1, ..., zd+1 ∈ S1. Let RV be a connected component of the set of all S1 ∈ S(M) intersecting V and containing S. Since the map is holomorphic and independent from the choice of z1, ..., zd+1 on an open subset of RV , it is independent of z1, ..., zd+1 everywhere on RV . This gives a section f ∈ H 0(RV , E) extending the section f := τ2∗τ ∗ 1 f ∈ H 0(S(V ), E). By construction, U is contained in a connected part U1 of the union of all deformations of S intersecting V . To extend f from V to U, we use f to obtain an extension of f to U1, as follows. Any section g ∈ H 0(RV , E) gives a function ψg mapping a pair (x, S1), is independent from the choice of S1 whenever S1 lies in V . The same analytic continuation argument as above implies that ψ f (x, S1) is independent of S1 x ∈ S1 ∈ RV to g(S1)(cid:12)(cid:12)x ∈ L(cid:12)(cid:12)x . For the section f constructed above, ψ f (x, S1) everywhere. For any x ∈ U1, the set f ∈ L(cid:12)(cid:12)x equal to ψ f (x, S1), where S1 ∈ RV is an arbitrary curve passing through x. This gives an extension of f to U1. Proposition 3.3 is proven. We finished the proof of Theorem 3.1. 3.3 Holography principle for meromorphic functions The following theorem is proven in the same way as Theorem 3.1. Given a complex variety M, we denote the field of meromorphic functions on M by Mer(M). Theorem 3.4: Let S ⊂ M be an ample curve in a simply connected complex manifold, and U ⊃ S a connected neighbourhood of S. Suppose that M is covered by the union of all smooth, ample deformations of S. Then the restriction map Mer(M) −→ Mer(U) is an isomorphism. -- 11 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces Proof: The same argument as used to deduce Theorem 3.1 from Proposition 3.3 can be used to reduce Theorem 3.4 to the following statement. Proposition 3.5: Let S ⊂ M be an ample rational curve in a connected complex manifold, which is covered by smooth, ample deformations of S. Consider a neighbourhood U ⊃ S which is contained in a union S of all deformations of S intersecting S (Remark 3.2). Then for any smaller open neighbourhood V ⊂ U of S, the restriction map Mer(U) −→ Mer(V ) is an isomorphism. Mer(V ) as the degree of the pole divisor of f(cid:12)(cid:12)(cid:12)S1 Proof: Define the degree degS(f ) of a meromorphic function f ∈ for any deformation S1 of S transversal to the pole divisor of f . Denote by Merd(V ) the space mero- morphic functions of degree 6 d. To prove Proposition 3.5 it would suffice to show that the restriction map Merd(U) −→ Merd(V ) is an isomorphism, for all d. For each rational curve S1, a degree 6 d meromorphic function is uniquely determined by its values in any d + 1 distinct points on S1. Given a mero- morphic function f ∈ Merd(V ), and a deformation S1 of S intersecting V , we its values at d + 1 distinct points z1, ..., zd+1 of S1 ∩ V . Whenever S1 is in . By analytic continuation, the values of f1(z) can extend f(cid:12)(cid:12)(cid:12)S1 ∩V to a degree d meromorphic function f1 on S1 by computing V , this procedure gives f(cid:12)(cid:12)(cid:12)S1 at any z ∈ S1 are independent from the choice of zi and S1. We have shown that f1 is a well-defined meromorphic function on the union U1 of all deformations of S intersecting V . By construction, on V we have f1 = f . Since U1 contains U, this implies that f can be extended from V to U. 4 Moishezon twistor spaces 4.1 Vector bundles in a neighbourhood of an ample curve Proposition 4.1: Let S ⊂ M be an ample rational curve, U ⊃ S its neigh- bourhood, and B a holomorphic bundle on U. Then H 0(U, B) is finite- dimensional. -- 12 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces Proof. Step 1: Let IS ⊂ OU be the ideal sheaf of S, and J r S B the sheaf of r-jets of the sections of B. Since I r S(B) := S = Symr(N ∗S), is a direct sum of O(ki) with ki < −r. Therefore, S/I r+1 S/I r+1 S B/I r+1 the bundle I r SB/I r+1 H 0(I r S B) = 0 for r sufficiently big. Step 2: The sheaf I r SB admits a filtration I r with associated graded sheaves I r Therefore, H 0(I r SB) = 0 for sufficiently byg r. SB/I r+1 S B ⊃ ... S B having no sections for r ≫ 0. S B ⊃ I r+2 SB ⊃ I r+1 Step 3: If H 0(U, B) is infinite-dimensional, the map H 0(U, B) −→ H 0(B/I r S(B)) cannot be injective. Then, for each r, there exists a non-zero section with vanishing r-jet: fr ∈ H 0(U, I r+1 S B). This is impossible, as shown in Step 2. This proof is effective, and gives the following bound on the dimension of the space of sections of B. Corollary 4.2: Let S ⊂ M be an ample rational curve, U ⊃ S its neigh- bourhood, and B a holomorphic bundle on U. Then dim H 0(U, B) 6 dim H 0 S,Md Symd(N ∗S) ⊗OS B(cid:12)(cid:12)S! . The same argument can be applied to degree d meromorphic functions. Recall that the degree degS(f ) of a meromorphic function f ∈ Mer(V ) is for any deformation of S transversal to the pole divisor of f . We denote the space of meromorphic functions of degree 6 d on U by Merd(U). the degree of the pole divisor of f(cid:12)(cid:12)(cid:12)S1 Corollary 4.3: Let S ⊂ M be an ample rational curve, and U ⊃ S its neighbourhood. Then dim Merd(U) 6 dim H 0(cid:0)S, Sym6d(N ∗S) ⊗OS O(d)(cid:1) . (4.1) -- 13 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces 4.2 Algebraic dimension of the field of meromorphic functions The conormal bundle N ∗S is negative; clearly, the dimension (4.1) is maximal when S is a quasiline, and N ∗S = ⊕O(1). In this case, the bound (4.1) is realized for a rational line in CP n. Indeed, for a rational line S in CP n, the sheaf of algebraic functions in a neighbourhood of S is isomorphic to of Li Symi(N ∗S). This implies the following simple numerical result. Claim 4.4: Let S ⊂ M be an ample rational curve, and U ⊃ S its neigh- bourhood. Then dim Merd(U) 6 dim H 0(CP n,O(d)). (4.2) where n = dim M. Corollary 4.5: Let M be a complex variety containing an ample rational curve. Then the transcendence degree of Mer(M) satisfies degtr Mer(M) 6 dim M. Proof: Consider the graded ring Ld Merd(M). Since dim Merd(U) 6 dim H 0(CP n,O(d)), the Krull dimension of this ring is 6 n. Therefore, the transcendence degree of its ring of fractions is also bounded by n. This observation is not new: it was known already to Hartshorne (in the context of formal neighbourhoods). Applied to complex analytic spaces, Hartshorne's theorem can be stated as follows. Theorem 4.6: Let S ⊂ M be a connected, positive-dimensional, smooth subvariety in a complex manifold. Assume that the normal bundle of S is ample. Then the transcendence degree of the field Mer(M) of meromorphic functions is no bigger than the dimension of M: degtr Mer(M) 6 dim(M). Moreover, if equality is reached, Mer(M) is a finitely generated extension of C. Proof: [Har, Theorem 6.7]; see also [KST]. -- 14 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces 4.3 Moishezon manifolds and ample rational curves Hartshorne's methods are already sufficient to prove the following general result. Proposition 4.7: Let S ⊂ M be an ample rational curve in a simply con- nected complex manifold. Assume that degtr Mer(M) = dim(M). Then there exists a meromorphic map ϕ : M −→ Z0 to an open subset of a projective variety M, which is bijective onto its image outside of a complex analytic subset of positive codimension. Proof: By Hartshorne's theorem (Theorem 4.6), Mer(M) is a finitely generated extension of C. Let ξ1, ..., ξN be generators of Mer(M), Di their pole divisors, and Li := O(Di) the corresponding line bundles. Then ξi can be considered as sections of Li, and ξ1, ..., ξN -- as sections of L := Ni Li. Consider now the subring ofLd H 0(M, Ld) generated by ξi, and let Z be its spectre. Clearly, dim M = dim Z, Z is projective, and the natural rational map M ϕ −→ Z induces an isomorphism Mer(Z) −→ Mer(M). Let M be a resolution of the base set of ϕ, such that M −→ Z is holomorphic. If ϕ is ramified at some divisor D in ϕ( M), this divisor can be extended to Z using Theorem 3.4 applied to U = ϕ( M) ⊂ Z. Taking the corresponding ramified covering Z of Z, we obtain a holomorphic map from M to Z, which is impossible, because Mer( Z) is strictly bigger than Mer(Z), and Mer( M) = Mer(Z). Therefore, ϕ is bijective to its impage at its general point. Openness of its image is a general property of bimeromorphic maps. ϕ Theorem 4.8: Let S ⊂ M be an ample rational curve in a simply connected complex manifold. Assume that degtr Mer(M) = dim(M). Then there exists an open embedding of M to a Moishezon variety. Proof: From Proposition 4.7 we obtain a line bundle L on M inducing a bimeromorphic map ϕ : M −→ Z0 to an open subset of a projective variety Z. Resolving the base points of the inverse map if necessary, we may assume that the inverse map ψ : Z0 −→ M is holomorphic. Then, M is obtained from Z0 by blowing down a certain number of exceptional subvarieties Ei, obtained as common zero sets of a certain number of meromorphic functions. Applying Theorem 3.4 to Z0 ⊂ Z, we extend the meromorphic functions and -- 15 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces the corresponding subvarieties Ei and obtain closed exceptional subvarieties E ′ i ⊂ Z. Blowing these down, we obtain a Moishezon variety which contains M as an open subset. 4.4 Moishezon twistor spaces Definition 4.9: Let M = Tw(X) be a twistor space of a simply connected hyperkahler, hypercomplex, quaternionic or or 4-dimensional anti-selfdual manifold, not necessarily compact. We say that M is a Moishezon twistor space if degtr Mer(M) = dim(M) (see Corollary 4.5). From Theorem 4.8, we immediately obtain the following corollary. Corollary 4.10: Let M be a Moishezon twistor space. Then M admits an open embedding to a Moishezon variety M1. Moreover, M1 is unique up to a bimeromorphic equivalence. Proof: The open embedding to a Moishezon variety follows from Theorem 4.8, and its uniqueness is implied by an isomorphism Mer(M) = Mer(M1) (Theorem 3.4). It is easy to construct an example of a twistor manifold which does not belong to this class. The twistor space of a K3 surface, and, more generally, any compact hyperkahler manifold is never Moishezon. Theorem 4.11: Let M be a compact hyperkahler manifold, and Tw(M) its twistor space. Then degtr Mer(Tw(M)) = 1. Proof: Let Z ⊂ Tw(M) be any divisor, and R ⊂ CP 1 a countable subset constructed in Theorem 2.6. For any induced complex structure L /∈ R, all complex subvarieties of (M, L) are even-dimensional. Therefore, Z intersects the twistor fiber (M, L) = π−1(L) ⊂ Tw(M) non-transversally, or not at all. By Thom's transversality theorem, the intersection Z with all fibers of π except a finite number is transversal. This means that Z can intersect only finitely many of the fibers of π. However, all these fibers are irreducible divisors. Therefore, D is a union of several fibers of π. Since a meromorphic function is uniquely determined by its pole or zero divisor, all meromorphic functions on Tw(M) are pull-backs of meromorphic functions on CP 1. -- 16 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces Remark 4.12: When M is hypercomplex, no effective bounds on the tran- scendence degree degtr Mer(Tw(M)) are known. We conjecture (based on empirical evidence) that the twistor space of a compact hypercomplex man- ifold is not Moishezon, but this conjecture seems to be difficult. 4.5 Twistor spaces and hyperkahler reduction We recall the definition of hyperkahler reduction, following [HKLR] and [Nak]. This material is fairly standard. We denote the Lie derivative along a vector field as Liex : ΛiM −→ ΛiM, and contraction with a vector field by ix : ΛiM −→ Λi−1M. Recall the Cartan's formula: d ◦ ix + ix ◦ d = Liex Let (M, ω) be a symplectic manifold, G a Lie group acting on M by symplectomorphisms, and g its Lie algebra. For any g ∈ g, denote by ρg the corresponding vector field. Cartan's formula gives Lieρg ω = 0, hence d(iρg(ω)) = 0. We obtain that iρg(ω) is closed, for any g ∈ g. Definition 4.13: A Hamiltonian of g ∈ g is a function h on M such that dh = iρg (ω). Definition 4.14: (M, ω) be a symplectic manifold, G a Lie group acting on M by symplectomorphisms. A moment map µ of this action is a linear map g −→ C ∞M associating to each g ∈ g its Hamiltonian. Remark 4.15: It is more convenient to consider µ as an element of g∗ ⊗R C ∞M, or, as it is usually done, a function on M with values in g∗. Remark 4.16: Note that the moment map always exists, if M is simply connected. Definition 4.17: A moment map M −→ g∗ is called equivariant if it is equivariant with respect to the coadjoint action of G on g∗. Remark 4.18: M µ −→ g∗ is a moment map if and only if for all g ∈ g, -- 17 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces hdµ, gi = iρg (ω). Therefore, a moment map is defined up to a constant g∗- valued function. An equivariant moment map is is defined up to a constant g∗-valued function which is G-invariant. Definition 4.19: A G-invariant c ∈ g∗ is called central. Claim 4.20: An equivariant moment map exists whenever H 1(G, g∗) = 0. In particular, if G is reductive and M is simply connected, an equivariant moment map is always possible to define. Definition 4.21: Let (M, ω) be a symplectic manifold, G a compact Lie −→ g∗ an equivariant mo- group acting on M by symplectomorphisms, M ment map, and c ∈ g∗ a central element. The quotient µ−1(c)/G is called the symplectic reduction of M, denoted by M//G. µ Claim 4.22: The symplectic quotient M//G is a symplectic manifold of dimension dim M − 2 dim G. Theorem 4.23: Let (M, I, ω) be a Kahler manifold, GC a complex reductive Lie group acting on M by holomorphic automorphisms, and G is a compact form of GC acting isometrically. Then M//G is a Kahler orbifold. Remark 4.24: In such a situation, M//G is called the Kahler quotient, or GIT quotient. Remark 4.25: The points of M//G are in bijective correspondence with the orbits of GC which intersect µ−1(c). Such orbits are called polystable, and the intersection of a GC-orbit with µ−1(c) is a G-orbit. Definition 4.26: Let G be a compact Lie group, ρ its action on a hyperkahler manifold M by hyperkahler isometries, and g∗ a dual space to its Lie algebra. A hyperkahler moment map is a G-equivariant smooth map µ : M → g∗ ⊗ R3 such that hµi(v), gi = ωi(v, dρ(g)), for every v ∈ T M, g ∈ g and i = 1, 2, 3, where ωi are three Kahler forms associated with the hyperkahler structure. Definition 4.27: Let ξ1, ξ2, ξ3 be three G-invariant vectors in g∗. The quo- -- 18 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces tient manifold M///G := µ−1(ξ1, ξ2, ξ3)/G is called the hyperkahler quo- tient of M. Theorem 4.28: ([HKLR]) The quotient M///G is hyperkaehler. Proof: We sketch the proof of Hitchin-Karlhede-Lindstrom-Rocek theo- rem, because we make use of it further on. Let Ω := ωJ + √−1 ωK. This is a holomorphic symplectic (2,0)-form µC := µJ +√−1µK. Then hdµC, gi = iρg(Ω) Therefore, dµC ∈ Λ1,0(M, I)⊗g∗. on (M, I). Let µJ , µK be the moment map associated with ωJ , ωK, and This implies that the map µC is holomorphic. It is called a holomorphic moment map. By definition, M///G = µ−1 C (c)//G, where c ∈ g∗ ⊗R C is a central ele- ment. This is a Kahler manifold, because it is a Kahler quotient of a Kahler manifold. We obtain 3 complex structures I, J, K on the hyperkahler quotient M///G. They are compatible in the usual way, as seen from a simple local computa- tion. Theorem 4.29: Let V be a quaterionic Hermitian vector space, and G ⊂ Sp(V ) a compact Lie group acting on V by quaternionic isometries. Denote by M the hyperkahler reduction of V . Then Tw(M) is a Moishezon twistor space, in the sense of Definition 4.9. Proof: The holomorphic symplectic form on (V, I) depends on I holo- morphically, giving a section Ωtw ∈ Ω2 π(Tw(V )) ⊗OTw(V ) π∗O(2). Here, π : Tw(V ) −→ CP 1 is the twistor projection, and Ω2 π(Tw(V )) the sheaf of fiberwise holomorphic 2-forms. Consider the fiberwise holomor- phic moment map given by this form, µtw : Tw(V ) −→ Tot(g∗ ⊗C O(2)). Replacing the moment map by its translate, we can always assume that M///G = µ−1 C (0)//G (that is, we assume that the central vector c used to define M///G vanishes). Then Tw(M) is obtained as the space of polystable GC-orbits in µ−1 tw (0) ⊂ Tw(V ). Here, "polystability" of an orbit is under- stood as the non-emptiness of the intersection of this orbit with µ−1(0); the set of such orbits is open. -- 19 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces The space Tw(V ) = CP 2n+1\CP 2n−1 is quasiprojective. Averaging the ring of rational functions with G, we obtain that the field GC-invariant ratio- nal functions on Tw(V ) has dimension dim µ−1 tw (0) − dim GC = dim Tw(M), and hence Tw(M) is Moishezon. Corollary 4.30: Let U be an open subset of a compact, simply connected hyperkahler manifold, and U ′ an open subset of a hyperkahler manifold ob- tained as V ///G, where V is flat and G reductive. Then U is not isomorphic to U ′ as hyperkahler manifold. Proof: dimtr Tw(U ′) = dim Tw(U ′) by Theorem 4.29, and dimtr Tw(U) = 1 by Theorem 4.11. A twistor space of a manifold obtained by hyperkahler reduction is Moishe- zon, and from the above argument it is easy to see that it is Zariski open in a compact Moishezon variety. However, it is (almost) never quasiprojective. Theorem 4.31: Let M be a hyperkahler manifold such that its twistor space Tw(M) can be embedded to a projective manifold. Then, for each induced complex structure L, the complex manifold (M, L) has no compact subvarieties of positive dimension. Proof: Consider the anticomplex involution ι on Tw(M) mapping (m, L) to (m,−L). Suppose that (M, L) ⊂ Tw(M) has a compact subvariety. Since the (M, L) is quasiprojective, this would imply that (M, L) contains a com- pact curve S. The curve ι(S) is also holomorphic in (M,−L). Consider a Kahler form ω on Tw(M). Since ι is antiholomorphic, −ι(ω) is a closed, positive (1, 1)-form. Replacing ω by ω − ι(ω), we may assume that ω satisfies ι(ω) = −ω. Now, since the cohomology classes of S and ι(S) are equal, we have ω = −Zι(S) ι(ω) ZS ω =Zι(S) On the other hand,RS ω =Rι(S) ι(ω) by functorial properties of integral. This impliesRS ω = 0, giving a contradiction. Remark 4.32: As shown in [V3], a general fiber of the map Tw(M) −→ CP 1 has only even-dimensional complex subvarieties. -- 20 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces 5 Appendix: Formal geometry and hologra- phy principle (by Dmitry Kaledin) In this Appendix, we will try to explain the constructions of the paper in a slightly more general context of D-modules and "formal geometry" of Gelfand and Kazhdan [GK]. To save space, we only sketch the proofs, and we work in the algebraic setting (generalization to complex-analytic varieties is im- mediate, exactly the same arguments work). Assume given a smooth algebraic variety X over a field k of characteristic 0. We will work with coherent D-modules over X, that is, with sheaves of left modules over the algebra DX of differential operators on X which are finitely generated over D. Any coherent D-module E is also a quasicoherent sheaf of OX-modules; recall that if E is coherent over OX , then it comes from a vector bundle on X equipped with a flat connection. In particular, the structure sheaf OX is a D-module; it corresponds to the trivial line bundle. For any coherent sheaf E on X, we can consider the induced D-module DX⊗OX E; this is coherent over DX but not over OX . Coherent D-modules form an abelian category, and we can consider its derived category. Assume given an open subvariety U ⊂ X with the embedding map j : U → X, and let Z = X \ U ⊂ X be the complement U. Then the complex of quascoherent sheaves R qj∗OU on X has a natural structure of a complex of D-modules, and we have an exact triangle δZ −−−→ OX −−−→ R qj∗OX −−−→ (5.1) of complexes of D-modules on X, where δZ is supported at Z. We will need the following standard result. Lemma 5.1: In the notation above, for any coherent sheaf E on X, we have a natural identification HomOX (δZ,E) ∼= Γ(X,E ⊗ bOX,Z), where bOX,Z is the formal completion of structure sheaf OX at the closed subscheme Z ⊂ X. Sketch of a proof. Since δZ is supported at Z, the natural map HomOX (δZ,E) → HomOX (δZ,E ⊗ bOX,Z) -- 21 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces is an isomorphism. On the other hand, we obviously have HomOX (R qj∗OU , E ⊗ bOX,Z) = 0, and the claim then immediately follows from the long exact sequence associ- ated to the exact triangle (5.1). (cid:3) For any map f : X → Y of smooth algebraic varieties, the pullback functor f ∗ extends to a functor between D-modules, and its derived functor extends to a functor between the derived categories of D-modules; we will denote this last functor by f ?. We have f ? = f ∗[dim X − dim Y ], where f ∗ is the standard pullback functor for D-modules, and [−] stands for cohomological shift. If f is a smooth map, or a closed embedding, or a composition of the two, then f ? has a left-adjoint functor f? given by f? = f![dim Y − dim X], where f! is the standard functor of direct image with compact supports. Then Lemma 5.1 has the following corollary. Lemma 5.2: Assume given a map f : Y → X of smooth algebraic varieties, and assume that f factors as Y π−−−→ Z ι−−−→ X, where Z is smooth, ι is a closed embedding, and π is a smooth map with contractible fibers. Then for any vector bundle E on X, we have a natural identification HomDY (f ?(DX ⊗OX E ∗),OY ) ∼= Γ(X,E ⊗ bOX,Z), where E ∗ is the dual vector bundle, and DX ⊗OX E ∗ is the corresponding induced D-module. Proof. For any D-module F and OX-module E, we have HomDX (DX ⊗OX E,F ) ∼= HomOX (E, F ), (5.2) -- 22 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces so that Lemma 5.1 provides an identification HomDX (DX ⊗OX E ∗ ⊗OX δZ,OX) ∼= Γ(X,E ⊗ bOX,Z), and by adjunction, it suffices to construct an isomorphism f?f ?(DX ⊗OX E ∗) ∼= DX ⊗OX E ∗ ⊗OX δZ. But under our assumptions on f , we have f?OX ∼= δZ, and we are done by the projection formula. (cid:3) We note that the left-hand side of (5.2) admits a slightly different inter- pretation. Recall that for any coherent sheaf E on X, the jet bundle J ∞E of E is a (pro)coherent sheaf on X given by J ∞E = π2∗π∗ 1E, where π1, π2 : bX → X are the two natural projections of the completion bX of the product X × X near the diagonal X ⊂ X × X. In terms of D-modules, we have J ∞E ∼= HomOX (DX,E). The jet bundle J ∞E carries a canonical flat connection, and in the assump- tions of Lemma 5.2, we have HomDY (f ?(DX ⊗OX E ∗),OY ) ∼= HomDY (OY ,HomOY (f ?(DX ⊗OX E ∗),OY )) ∼= HomDY (OY , f ?(HomOX (DX ⊗OX E ∗,OX))) ∼= Γ∇(Y, f ∗J ∞E), where Γ∇(−) stands for the space of flat global sections. Then (5.2) reads as (5.3) Γ∇(Y, f ∗J ∞E) ∼= Γ(X,E ⊗ bOX,Z). Remark 5.3: The assumptions of Lemma 5.2 are in fact too strong. Firstly, it is clearly enough to require that the fibers of π are non-empty and con- nected, so that their top degree cohomology with compact supports is one- dimensional -- and under the assumptions as stated, we not only obtain an isomorphism of Hom's but also of the RHom's, so that (5.3) extends to an isomorphism H q DR(Y, f ∗J ∞E) ∼= H q(X,E ⊗ bOX,Z) -- 23 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces of cohomology groups. Secondly, one probably does not have to require that π and Z are smooth -- some assumptions are needed, but they can considerably relaxed. However, since even the stronger assumptions work for us, we did not pursue this. Assume now that we are given smooth proper algebraic varieties Y , X, and a family of closed embedding from Y to X parametrized by a smooth algebraic variety T -- that is, we have a map f : Y × T → X such that for any t ∈ T , the corresponding map ft : Y = Y × t → X is a closed embedding. Moreover, assume given a coherent sheaf E on X, denote by ρ : Y × T → T the projection, and denote ∗ f ∗J ∞E, ΦfE = ρ∇ where f ∗J ∞E is the pullback of the jet bundle J ∞E equipped with its natural flat connection ∇, and ρ∇ ∗ stands for sheaf of relative flat sections. Then ΦaE is a sheaf on T , the base of the family, and by (5.3), the fiber (ΦfE)t at a point t ∈ T is given by (ΦfE)t = Γ∇(Y, f ∗ t J ∞E) ∼= Γ(X,E ⊗ bOX,Yt), where Yt ⊂ X is the image of the closed embedding ft : Y → X. On the other hand, ΦfE carries a natural flat connection, and if we assume that f is smooth with contractible fibers, we have Γ∇(T, ΦfE) = Γ∇(T × Y, f ∗J ∞E) ∼= Γ(X,E), again by (5.3). We now note that this is exactly the situation that we have in the paper. Namely, we take X to be the twistor space of a hyperkahler manifold M, we take T = MC to be the complexification of the real-analytic manifold underlying M, we take Y = CP 1, and we let f : T × Y → X be the standard family of twistor lines (real points in this family correspond to horizontal sections of the twistor fibration X → CP 1 parametrized by -- 24 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces points of M). Then since the normal bundle to any line in our family is a sum of several copies of O(1), the family is unobstructed, and moreover, it remains unobstructed even if we fix a point at a twistor line, so that the map f is smooth. Its fibers are small polydiscs, thus contractible, and all the assumptions of Lemma 5.2 are therefore satisfied. References [BBI] Badescu, Lucian; Beltrametti, Mauro C.; Ionescu, Paltin, Almost-lines and quasi-lines on projective manifolds, Complex analysis and algebraic geome- try, 1-27, de Gruyter, Berlin, 2000. [AHS] Atiyah, M. F., Hitchin, N. J., and Singer, I. M. Self-duality in four- dimensional Riemannian geometry, Proc. Roy. Soc. London Ser. A 362 (1978), 425-461. [Bes] Besse, A., Einstein Manifolds, Springer-Verlag, New York (1987). [C] [D] F. Campana, On twistor spaces of the class C, J. Differential Geom. 33 (1991) 541-549. Demailly, Jean-Pierre, Pseudoconvex-concave duality and regularization of currents, Several complex variables (Berkeley, CA, 1995-1996), 233-271, Math. Sci. Res. Inst. Publ., 37, Cambridge Univ. Press, Cambridge, 1999. [GK] I.M. Gelfand and D.A. Kazhdan, Some problems of differential geometry and the calculation of cohomologies of Lie algebras of vector fields, Soviet Math. Dokl. 12 (1971), 1367-1370. [Har] Hartshorne, Robin, Cohomological dimension of algebraic varieties, Ann. of Math. (2) 88 1968 403-450. [HKLR] N. J. Hitchin, A. Karlhede, U. Lindstrom, M. Rocek, Hyperkahler metrics and supersymmetry, Comm. Math. Phys. 108, (1987) 535 -- 589. [Ho] Nobuhiro Honda Moishezon twistor spaces on 4CP 2, arXiv:1112.3109, 51 pages. [K] D. Kaledin, Integrability of the twistor space for a hypercomplex manifold, Sel. math., New ser. 4 (1998) 271-278. -- 25 -- version 1.0, Nov. 25, 2012 M. Verbitsky Holography principle for twistor spaces [KV] Kaledin, D., Verbitsky, M., Non-Hermitian Yang-Mills connections, Selecta Math. (N.S.) 4 (1998), no. 2, 279 -- 320. [KST] Stefan Kebekus, Luis Sola Conde, Matei Toma, Rationally connected foli- ations after Bogomolov and McQuillan, J. Algebraic Geom. 16 (2007), no. 1, 65-81. [Ko] Koll´ar, J., Rational curves on algebraic varieties, Springer, 1996, viii+320 pp.. [Nak] H. Nakajima, Lectures on Hilbert schemes of points on surfaces, Providence: American Mathematical Society, 1999. [P] Y. S. Poon, On the algebraic structure of twistor spaces, J. Diff. Geom. 36 (1992), 451-491. [SV] Andrey Soldatenkov, Misha Verbitsky, Subvarieties of hypercomplex mani- folds with holonomy in SL(n, H), Journal of Geometry and Physics, Volume 62, Issue 11 (2012), Pages 2234-2240, arXiv:1202.0222 [V1] Verbitsky M., Tri-analytic subvarieties of hyper-Kaehler manifolds, also known as Hyperkahler embeddings and holomorphic symplectic geometry II, GAFA 5 no. 1 (1995), 92-104, alg-geom/9403006. [V2] Verbitsky, M., Hypercomplex Varieties, alg-geom/9703016, Comm. Anal. Geom. 7 (1999), no. 2, 355 -- 396. [V3] Verbitsky, M., Subvarieties in non-compact hyperkahler manifolds, Math. Res. Lett., vol. 11 (2004), no. 4, pp. 413 -- 418. [V4] Verbitsky, M., Rational curves and special metrics on twistor spaces, arXiv:1210.6725, 12 pages. [Y] Yau, S. T., On the Ricci curvature of a compact Kahler manifold and the complex Monge-Amp`ere equation I. Comm. on Pure and Appl. Math. 31, 339-411 (1978). Misha Verbitsky Laboratory of Algebraic Geometry, Faculty of Mathematics, National Research University HSE, 7 Vavilova Str. Moscow, Russia -- 26 -- version 1.0, Nov. 25, 2012
1712.03801
3
1712
2018-02-16T09:26:55
On the Zariski topology of $\Omega$-groups
[ "math.AG" ]
A number of geometric properties of $\Omega$-groups from a given variety of $\Omega$-groups can be characterized using the notions of domain and equational domain. An $\Omega$-group $H$ of a variety $\Theta$ is an equational domain in $\Theta$ if the union of algebraic varieties over $H$ is an algebraic variety. We give necessary and sufficient conditions for an $\Omega$-group $H$ in $\Theta$ to be an equational domain in this variety.
math.AG
math
ON THE ZARISKI TOPOLOGY of Ω-GROUPS R. Lipyanski This article is dedicated to my teacher Prof. B. Plotkin on his 90th anniversary. Abstract. A number of geometric properties of Ω-groups from a given variety of Ω-groups can be characterized using the notions of domain and equational domain. An Ω-group H of a variety Θ is an equational domain in Θ if the union of algebraic varieties over H is an algebraic variety. We give necessary and sufficient conditions for an Ω-group H in Θ to be an equational domain in this variety. Let F = F (X) be a finitely generated by X free Ω-group in a variety of Ω- groups Θ and H be an Ω-group in Θ. One of the important questions in the algebraic geometry of varieties of Ω-groups is whether it is possible to equip the space of points Hom(F, H) with the Zariski topology, whose closed sets are precisely algebraic sets. If this is so, then the Ω-group H is called the equational domain (or stable in the terminology of the authors of the papers [P], [BPP]) in the variety Θ (see [BMR]). The same problem arises for a variety Θ(G) of Ω-groups with the given Ω-group of constants G. An important role in the study of equational domains in varieties of Ω-groups is played by the notion of domain. Necessary and sufficient conditions for linear algebras over an algebra of constants and for groups over a group of constants to be equational domains in terms of domains are given in [BMR], [BPP], [DMR] and [P]. Here we continue the study of equational domains in varieties of Ω-groups (without of the Ω-group of constants). We give necessary and sufficient conditions for an Ω-group H in Θ to be an equational domain in the variety Θ. The results presented in this paper were partially announced earlier in [L]. 1. Preliminaries 1.1. Ω-groups. Now we give the basic facts about Ω-groups (see [H]). Definition 1.1. An Ω-group (multioperator group) G is an additive group (not necessarily commutative) with some additional signature Ω, such that for ω = 0 should be every ω ∈ Ω of the arity n(ω) = n > 0, the condition 00 . . . 0 {z }n fulfilled. 1991 Mathematics Subject Classification. Primary 14A22, 16S38; Secondary 07A01, 08B20. Key words and phrases. Equational domain, variety of Ω-groups, anticommutative Ω-group, the Zariski topology. 1 2 R. LIPYANSKI A group is an Ω-group with the empty set of operations Ω; in rings the set Ω consists of a single multiplication; in Lie algebras over a commutative associative ring K with unit the set Ω consists of the Lie bracket and all elements of K belonging to the set Ω. It is clear that the class of all Ω-groups forms the variety Υ. Let FΥ = FΥ(X) be a finitely generated by X free Ω-group in the variety Υ. We use the functional notion f (x1, . . . , xn), xi ∈ X, for words in FΥ(X). For brevity we shall use symbols ¯x, ¯a, to denote finite ordered sets (x1, x2, . . . , xn), (a1, a2, . . . , am), xi ∈ X, ai ∈ G, and write f (¯x) for f (x1, x2, . . . , xn), f (¯a, ¯b) for f (a1, a2, . . . , am, b1, . . . , br). If ¯x, ¯y are ordered sets (x1, x2, . . . , xn), (y1, y2, . . . , yn) with the same number of elements, we shall denote the set x1 + y1, x2 + y2, . . . , xn + yn by ¯x + ¯y. We shall also write ¯a ∈ G when we mean a1, . . . , am ∈ G. Let f (¯x, ¯y) be a word in two disjoint sets X and Y of variables ¯x and ¯y, respectively. We shall say that the word f (¯x, ¯y) is a commutator word in ¯x and ¯y if f (¯x, ¯0) = f (¯0, ¯y) = 0. The set of all such words will be denoted by [X, Y ]. Let A and B be two subsets of an Ω-group G. The set of all elements f (¯a, ¯b) , where f (¯x, ¯y) ∈ [X, Y ] is called the commutator group of A and B and is denoted by [A, B]. Note that if A and B are Ω-subgroups, then [A, B] is also an Ω-subgroup of G. Lemma 1.2 ([H]). If [A, B] = 0, then f (¯a, ¯b) = f (¯a, ¯0) + f (¯0, ¯b) = f (¯0, ¯b) + f (¯a, ¯0) for all words f (¯x, ¯y) and all ¯a ∈ A and ¯b ∈ B. Let A and B be defined as above and ¯a = (a1, a2, . . . , an), ai ∈ A and ¯b = (b1, b2, . . . , bn), bi ∈ B. Then the element [¯a; ¯b; ω] = −¯aω − ¯bω + (¯a + ¯b)ω, ω ∈ Ω is called the ω-commutator of ¯a and ¯b. Definition 1.3. Let G be an Ω-group. A subset U of G is called an ideal in G if the following conditions are fulfilled: (1) U is closed with respect to all ω ∈ Ω. (2) U is a normal subgroup in additive group G. (3) The ω-commutator [¯a; ¯b; ω] belongs to U if ¯a ∈ U , ω ∈ Ω and ¯b ∈ G. Let A and B be two Ω-subgroups of G and let {A, B} be an Ω-subgroup of G generated by A and B. The commutator group [A, B] can be characterized as follows. Proposition 1.4 ([H]). The commutator group [A, B] is the ideal in {A, B} generated by all commutators of the kind [a, b], a ∈ A, b ∈ B, and all ω-commutators [¯a; ¯b; ω], where ¯a ∈ A and ¯b ∈ B, ω ∈ Ω. Example 1.5. Let R be an associative ring and U1, U2 subrings in R. Then it is easy to show that [U1, U2] = U1U2 + U1U2 In the case of Lie algebras (groups) we have the ordinary commutator subalgebra (commutator subgroup). Definition 1.6. An Ω-group G is called abelian if [G, G] = 0. ON THE ZARISKI TOPOLOGY OF Ω-GROUPS 3 Groups and Lie algebras are abelian in the usual sense, while for associative rings this notion means that the product of any two elements is zero. Now we turn to the property of anticommutativity for Ω-groups. Definition 1.7 ([BPP]). An Ω-group L is called anticommutative (or antia- belian in the terminology of the authors in [BMR]) if the following conditions are fullfilled: (1) L has no nontrivial abelian ideal. (2) Every two nontrivial ideals H1 and H2 in L have a nontrivial intersection. A number of interesting properties of anticommutativity are given for groups in [BMR] and for Ω-groups in [BPP], [P]. It is known that every non-abelian free group, free associative algebra, and non-abelian free Lie algebra are anticommuta- tive (see [BMR], [BPP], [L]). 1.2. Domains. Now we consider the notion of zero divisors in an Ω-group H. For each a ∈ H, denote by idhai the ideal in H generated by a and hai is an Ω-subgroup of H generated by a. Let P be an Ω- subgroup of H. We denote by idPhai the ideal in P generated by a. In our notation, we have idHhai = idhai. Definition 1.8 ([P]). A non-zero element a ∈ H is called a zero divisor if for some non-zero element b ∈ H we have (1.1) [idhai, idhbi] = 0 The Ω-group G is called a domain if G is without zero divisors, i.e., for any two elements g1 and g2 of G the following holds: [idhg1i, idhg2i] = 0 ⇒ g1 = 0 or g2 = 0, Example 1.9. Let R be an associative ring. In this case Definition 1.8 looks as follows: a non-trivial element a in A-associative ring R is a zero divisor if there exists a non-trivial element b ∈ R such that idhai · idhbi = idhbi · idhai = 0 Let L be a Lie algebra over an associative commutative ring K with unit. A non- trivial element a in L is a zero-divisor if for some non-zero element b ∈ L is fulfilled [idhai, idhbi] = 0, where [, ] is the Lie bracket. Let H be a group. A non-trivial element a in H is a zero divisor if for some non-trivial element b ∈ H we have (1.2) [g−1 1 ag1, g−1 2 bg2] = 1 for all g1, g2 ∈ H. Here [, ] is the usual commutator brackets in the group H. Remark 1.10. In [BMR]) it was proved that the condition (1.2) is equivalent to the following: a non-trivial element a in H is a zero divisor if for some non-trivial element b in H we have (1.3) for all g ∈ H. [g−1ag, b] = 1 4 R. LIPYANSKI 1.3. Algebraic varieties over groups. Let Θ be a variety of Ω-groups and F = F (X) be a finitely generated by X free group in Θ. Consider an Ω-group H in Θ. Any formula w ≡ w′, w, w′ ∈ F (X) can be treated as an equation. Denote it as w = w′. Every solution of this equation in H is a homomorphism µ : F (X) → H such that wµ = w′µ. It is possible to define a Galois correspondence ′ between subsets in Hom(F (X), H) and subsets in F (X). For a subset T in F (X) define T ′, H-closure of T T ′ = {µ ∈ Hom(F (X), H) T ⊆ Kerµ} On the other hand, for any subset A ⊆ Hom(F (X), H) define a set A′ in F (X), H-closure of A A′ = \ Kerµ The set A′ is an ideal in F (X). µ∈A Definition 1.11. A subset A ⊆ Hom(F (X), H) is called an affine algebraic variety over H if there exists a set T in F (X)) such that T ′ = A. If A is an algebraic variety, then A′ is called the ideal in F (X) corresponding to A. The intersection A∩B of algebraic varieties A and B is also an algebraic variety. The union A ∪ B of algebraic varieties is not necessarily an algebraic variety. If A = T ′ 2, then A ∪ B ⊆ (T1 ∩ T2)′. 1 and B = T ′ Definition 1.12 ([P], [DMR]). An Ω-group H is called an equational domain in Θ if for any free Ω-group F (X) and any two algebraic varieties A and B in the space Hom(F (X), H) the union A ∪ B is also an algebraic variety. Equational domains play an important role in the theory of algebraic varieties. Following [BPP], denote by AlvH (F ) the set of all algebraic varieties in Hom(F, H). The set AlvH (F ) can be considered as a lattice, where the union A ∨ B is defined by A ∨ B = (A ∪ B)′ Denote by ClH (F ) the set of all H-closed congruences in F . Lattice operations can be defined in a similar way in the set ClH (F ). The lattices ClH (F ) and AlvH (F ) are antiisomorphic. It is clear that if a Ω-group H is an equational domain, then the lattices AlvH (F ) and ClH (F ) are distributive. 2. Equational domains in a variety of Ω-groups As before, let Θ be a variety of Ω-groups. Theorem 2.1. An Ω-group H in Θ is a domain if and only if H is an equational domain in Θ. Proof. Necessity has been proved by B. Plotkin (see Theorem 1 in [P]). We present this proof for completeness. Let F (X) the free Ω-group in Θ generated by X. Suppose that the Ω-group H is a domain. Let us take two algebraic varieties A and B in V = Hom(F (X), H). Let T1 and T2 be ideals in F (X) corresponding to these varieties. We check that A ∪ B = (T1 ∩ T2)′. It is obvious that A ∪ B ⊆ (T1 ∩ T2)′ ON THE ZARISKI TOPOLOGY OF Ω-GROUPS 5 To check the inverse inclusion, it suffices to show that µ /∈ A ∪ B implies µ /∈ (T1 ∩ T2)′. Since µ /∈ A ∪ B, we have T1 /∈ Kerµ and T2 /∈ Kerµ. Hence, there exist u ∈ T1 and v ∈ T2 such that uµ = a 6= 0 and vµ = b 6= 0. Since H is the domain, [idhai, idhbi] 6= 0. The ideal [idhai, idhbi] is generated by ω-commutators c = [a1, a2, . . . , an; b1, b2, . . . , bn; ω] and ordinary commutators [a′, b′], where a′, ai ∈ idhai and b′, bi ∈ idhbi, i = 1, . . . , n. Hence, there exists a nonzero ω-commutator c = [a1, a2, . . . , an; b1, b2, . . . , bn; ω] or a nonzero commutator [a′, b′], where a′, ai ∈ idhai and b′, bi ∈ idhbi, , i = 1, . . . , n. Let us suppose that such non zero commutator is of the form c = [a1, a2, . . . , an; b1, b2, . . . , bn; ω]. It is easy to check that (idhai)µ = idhaµi. Hence, we can take u1, u2, . . . , un ∈ idGhui and v1, v2, . . . , vn ∈ idGhvi such that uµ i = bi, i = 1, . . . , n. It is clear that i = ai and vµ w = [u1, u2, . . . , un; v1, v2, . . . , vn; ω] ∈ [idhui, idhvi] and wµ = c 6= 0. We have [idhui, idhvi] ⊆ [T1, T2] ⊆ T1 ∩ T2. Finally, we have that w ∈ T1 ∩ T2 and wµ 6= 0. Hence, µ /∈ (T1 ∩ T2)′ as desired. Now we prove the sufficiency. Suppose that H is an equational domain. Assume to the contrary that H is not a domain. As a consequence, there exist two non-zero elements a, b in H such that (2.1) [idhai, idhbi] = 0 Let X = {x, y}. Take the free Ω-group F (X). The affine space V = Hom(F (X), H) over H can be identified with the set H × H. In fact, every point (h1, h2) ∈ H × H determines the homomorphism µ : F (X) → H such that µ(x) = h1 and µ(y) = h2 and vice versa. Consider two subvarieties A and B in V defined by the equations x = 0 and y = 0, respectively. Since H is an equational domain, D = A∪B is a subvariety of V . Let f (x, y) be an element in F (X) that belongs to D′. Then f (a, 0) = f (0, b) = 0. Let Z = {X1, Y1} be a set in two indeterminates X1 and Y1 and FΥ(Z) be the free algebra in the variety Υ of all Ω-groups generated by Z. Denote by f (X1, Y1) the polynomial in FΥ(Z) corresponding to f (x, y). It is clear that the values of the polynomials f (x, y) and f (X1, Y1) in H are equal. Since the formula (2.1) is valid, by Lemma 1.2 we get (2.2) f (a, b) = f (a, 0) + f (0, b) = 0 From (2.2) we obtain that the point (a, b) ∈ D′′. Since D is the algebraic variety, (a, b) ∈ D. On the other hand, since a and b are nonzero elements in H, (a, b) /∈ D. We have arrived at a contradiction. This ends the proof. (cid:3) As a consequence, Theorem 2.1 holds for the variety of all linear algebras, for the variety of all groups, and for the variety of all modules (see [P], [BPP], [DMR]). Theorem 2.1 is also true for the so-called CD-variety of Ω-groups (see Theorem 2 in [P]). Definition 2.2 ([L]). An Ω-group H is called C-anticommutative (completely anticommutative) if each of its nonzero Ω-subgroup is anticommutative. 6 R. LIPYANSKI It turns out that the concept of an equational domain and C-anticommutativity are closely related to each other. In what follows, we use the following Proposition 4 from [P]. Proposition 2.3. An Ω-group H is a domain if and only if H is anticommu- tative. We now give a useful criterion for an Ω-group H in a variety Θ to be an equational domain in this variety. Proposition 2.4. Every non-trivial Ω-group H in Θ is an equational domain in Θ if and only if H is C-anticommutative. Proof. Suppose that the Ω-group H is C-anticommutative. Let us take two nonzero elements a and b in H. Denote by P the Ω-subgroup of H generated by the elements a and b. Since H is C-anticommutative, P is anticommutative. By Proposition 2.3, P is without zero-divisors, i.e., [idPhai, idPhbi] 6= 0. From this, it follows that [hai, hbi] 6= 0. Therefore, [idhai, idhbi] 6= 0, i.e., H is a domain. By Theorem 2.1, H is an equational domain in Θ. Now suppose that the Ω-group H is an equational domain. Let us show that it is C-anticommutative. Assume to the contrary, that H is not C-anticommutative. Therefore, there exists an Ω-subgroup H1 of H which is not anticommutative. Hence, there exist two non-zero elements a, b ∈ H1 such that (2.3) [idH1 hai, idH1 hbi] = 0 Take the free Ω-group F = F (X) generated by X = {x, y}. Denote by V = Hom(F (X), H) the affine space over H. Consider two subvarieties A and B in V defined by the equations x = 0 and y = 0, respectively. Since the Ω-group H is an equational domain, D = A ∪ B is a subvariety of V . Denote by V1 = Hom(F (X), H1) the subvariety of V in the induced Zariski topology. Let A1 and B1 be the subvarieties of V1 defined by the equations x = 0 and y = 0, respectively. Then we have A1 = V1 \ A and B1 = V1 \ B. Since D is subvarieties of V , D1 = A1 S B1 is a subvariety of V1. However, the same arguments given in the proof of the sufficiency of the conditions of Theorem 2.1 show that D1 is not a variety. We have a contradiction. This ends the proof. (cid:3) Corollary 2.5. An Ω-group H in Θ is C-anticommutative if and only if for any non-zero elements a and b in H, [hai, hbi] 6= 0. Proof. Suppose that the elements a and b in H satisfy the above condition. Therefore, [idhai, idhbi] 6= 0. Hence, H is without zero divisors, i.e., H is a domain. By Theorem 2.1, it is an equational domain. According to Proposition 2.4, H is C-anticommutative. Let H be C-anticommutative. Let P be the Ω-subgroup of H generated by two non-zero elements a and b of H. Since H is C-anticommutative, P is anticommu- tative. By Proposition 2.3, P is without zero-divisors, i.e., [idPhai, idPhbi] 6= 0. It follows that [hai, hbi] 6= 0 as desired. (cid:3) Example 2.6. Now consider some examples of Ω-groups that are not equational domains in varieties related to them. ON THE ZARISKI TOPOLOGY OF Ω-GROUPS 7 (1) Every non-trivial module M over an associative commutative ring K with unit is not an equational domain in the variety of all modules over K, since it is abelian, i.e., [M, M ] = 0. (2) Every nontrivial soluble Ω-group G is not an equational domain in the variety Υ of all Ω-groups. Indeed, G has a nontrivial abelian Ω-subgroup. As a consequence, G is not C-anticommutative. The following examples were considered earlier in [DMR]. However, using Proposition 2.4 and Corollary 2.5, in contrast to the paper [DMR], we prove all statements (3)-(5) in a unified way. (3) Every non-trivial group G is not an equational domain in the variety of all groups. Indeed, G contains a non-trivial cyclic group which is not anticom- mutative. Therefore, G is not C-anticommutative. By Proposition 2.4, G is not an equational domain. (4) Every non-trivial Lie algebra L over an associative commutative ring K is not an equational domain in the variety of all Lie algebras. In fact, L contains a non-trivial cyclic subalgebra which is also not anticommutative. By Proposition 2.4, L is not an equational domain. Denote by Lring = {+, −, ·, 0} the language of associative rings. In the formulation of the following assertion, we use the language Lring. (5) An associative ring A is an equational domain if and only if A satisfies the formula: (2.4) ∀x, y((xy = yx = 0) ⇒ [(x = 0) ∨ (y = 0)]). Indeed, suppose that A is an equational domain. By Proposition 2.4, A is C- anticommutative. Let ab = ba = 0 for some elements a and b in A. From this it follows that [hai, hbi] = 0. By Corollary 2.5, a = 0 or b = 0. Conversely, assume that formula (2.4) is true in A. Then for every non-zero elements a and b in A, [hai, hbi] 6= 0. By Corollary 2.5, A is C-anticommutative. According to Proposition 2.4, A is an equational domain. 3. Acknowledgments I would like to thank Prof. B. I. Plotkin and Prof. E. B. Plotkin for useful suggestions and comments on this paper. References [BMR] G. Baumslag, A. Myasnikov, V. Remeslennikov, Algebraic geometry over groups I: Alge- braic sets an ideal theory, J. Algebra, vol. 219, 1999, pp. 16-19. [BPP] A. Berzins, B. Plotkin, E. Plotkin, Algebraic geometry in varieties of algebras with the given algebra of constants, J. Math. Sci., 3, 2000, pp. 4039-4070. [DMR] E. Daniyarova, A. Myasnikov, V, Remeslennikov, Algebraic geometry over algebraic struc- tures. IV. Equational domain and codomains, Algebra and Logic, 6, vol 49, 2011, pp. 483-508. [H] P. Higgins, Group with multiple operator, Proc. London Math. Soc., 3, 1957, pp.366-416. [L] R. Lipyanski, On stable Ω-groups, Proc. Latv. Acad. Sci. Sect. B Nat. Exact Appl. Sci. , 57, 4, 2003, pp. 102-105. [P] B.I. Plotkin, Zero-divisor in group-based algebras. Algebras without zero divisors, Buletinul A.S. a R.M., 2, 1999, pp. 67-84. 8 R. LIPYANSKI Department of Mathematics, Ben Gurion University, Beer Sheva 84105, Israel Current address: Department of Mathematics, Ben Gurion University, Beer Sheva 84105, Israel E-mail address: [email protected]
1804.03419
1
1804
2018-04-10T09:44:10
Algebraic links in the Poincar\'e sphere and the Alexander polynomials
[ "math.AG" ]
The Alexander polynomial in several variables is defined for links in three-dimensional homology spheres, in particular, in the Poincar\'e sphere: the intersection of the surface $S=\{(z_1,z_2,z_3)\in {\mathbb C}^3: z_1^5+z_2^3+z_3^2=0\}$ with the 5-dimensional sphere ${\mathbb S}_{\varepsilon}^5=\{(z_1,z_2,z_3)\in {\mathbb C}^3: \vert z_1\vert^2+\vert z_2\vert^2+\vert z_3\vert^2=\varepsilon^2\}$. An algebraic link in the Poincar\'e sphere is the intersection of a germ $(C,0)\subset (S,0)$ of a complex analytic curve in $(S,0)$ with the sphere ${\mathbb S}_{\varepsilon}^3$ of radius $\varepsilon$ small enough. Here we discuss to which extend the Alexander polynomial in several variables of an algebraic link in the Poincar\'e sphere determines the topology of the link. We show that, if the strict transform of a curve on $(S,0)$ does not intersect the component of the exceptional divisor corresponding to the end of the longest tail in the corresponding $E_8$-diagram, then its Alexander polynomial determines the combinatorial type of the minimal resolution of the curve and therefore the topology of the corresponding link. Alexander polynomial of an algebraic link in the Poincar\'e sphere coincides with the Poincar\'e series of the filtration defined by the corresponding curve valuations. We show that, under conditions similar for those for curves, the Poincar\'e series of a collection of divisorial valuations determines the combinatorial type of the minimal resolution of the collection.
math.AG
math
Algebraic links in the Poincar´e sphere and the Alexander polynomials ∗ A. Campillo, F. Delgado,† S.M. Gusein-Zade ‡ Abstract 3 = 0} with the 5-dimensional sphere S5 The Alexander polynomial in several variables is defined for links in three-dimensional homology spheres, in particular, in the Poincar´e sphere: the intersection of the surface S = {(z1, z2, z3) ∈ C3 : z5 1 + ε = {(z1, z2, z3) ∈ C3 : z3 2 + z2 z12 + z22 + z32 = ε2}. An algebraic link in the Poincar´e sphere is the intersection of a germ (C, 0) ⊂ (S, 0) of a complex analytic curve in (S, 0) with the sphere S3 ε of radius ε small enough. Here we discuss to which extend the Alexander polynomial in several variables of an algebraic link in the Poincar´e sphere determines the topology of the link. We show that, if the strict transform of a curve on (S, 0) does not intersect the component of the exceptional divisor corresponding to the end of the longest tail in the corresponding E8-diagram, then its Alexander polynomial determines the combinatorial type of the minimal resolution of the curve and therefore the topology of the corresponding link. Alexander polynomial of an algebraic link in the Poincar´e sphere coincides with the Poincar´e series of the filtration defined by the cor- responding curve valuations. We show that, under conditions similar for those for curves, the Poincar´e series of a collection of divisorial val- uations determines the combinatorial type of the minimal resolution of the collection. ∗Math. Subject Class. 14B05, 32S25, 57M25. Keywords: algebraic links, Poincar´e sphere, Alexander polynomial, Poincar´e series, topological type. †The first two authors were supported by the grant MTM2015-65764-C3-1-P (with the help of FEDER Program). ‡The work of the third author (Sections 2, 4, 6) was supported by the grant 16-11-10018 of the Russian Science Foundation. 1 1 Introduction The three-dimensional sphere is S3 ε = {(z1, z2) ∈ C2 : z12 + z22 = ε2}. An algebraic link in the three-dimensional sphere is the intersection of a germ (C, 0) ⊂ (C2, 0) of a complex analytic plane curve with the sphere S3 ε with ε small enough. The number of components of the link K = C ∩ S3 ε equals the number of the irreducible components of the curve (C, 0). A link with r components in the three-sphere has the well-known topological invariant: the Alexander polynomial in r variables: see, e. g., [8]. It is known that the Alexander polynomial in several variables determines the topological type of an algebraic link (or, equivalently, the (local) topological type of the triple (C2, C, 0)): [12], see [4] for another proof of this statement. The Alexander polynomial is defined for links in three-dimensional mani- folds which are homology spheres. The Poincar´e sphere L is the intersection of the surface S = {(z1, z2, z3) ∈ C3 : z5 3 = 0} with the 5-dimensional sphere S5 It is a three- dimensional homology sphere. This definition describes the Poincar´e sphere L as the link of a rational surface singularity of type E8. The links of other rational surface singularities are rational homology spheres, but not homology spheres. ε = {(z1, z2, z3) ∈ C3 : z12 + z22 + z32 = ε2}. 1 + z3 2 + z2 An algebraic link in the Poincar´e sphere is the intersection of a germ (C, 0) ⊂ (S, 0) of a complex analytic curve in (S, 0) with the sphere S5 ε of radius ε small enough. The number of components of the link K = C ∩ S5 ε equals the number of the irreducible components of the curve (C, 0). For a link with r components in the Poincar´e sphere L = S ∩ S5 ε one has its Alexander polynomial ∆K(t1, . . . , tr) defined in the same way as for a link in the usual three-sphere S3 ε. An irreducible curve germ (C, 0) in a germ of a complex analytic variety (V, 0) defines a valuation vC on the ring OV,0 of germ of functions on (V, 0) (called a curve valuation). Let ϕ : (C, 0) → (V, 0) be a parametrization (an uniformization) of the curve (C, 0), that is Im ϕ = (C, 0) and ϕ is an isomor- phism between punctured neighbourhoods of the origin in C and in C. For a function germ f ∈ OV,0, the value vC(f ) is defined as the degree of the leading term in the Taylor series of the function f ◦ ϕ : (C, 0) → C: f ◦ ϕ(τ ) = aτ vC (f ) + terms of higher degree, where a 6= 0; if f ◦ ϕ ≡ 0, one defines vC(f ) to be equal to +∞. A collection {(Ci, 0)} of irreducible curves in (V, 0), i = 1, . . . , r, defines the collection {vCi} of valuations. For a collection {vi} of discrete valuations on OV,0, i = 1, . . . , r, there is defined its Poincar´e series P{vi}(t1, . . . , tr) ∈ 2 Z[[t1, . . . , tr]]: [6]. In [1] it was shown that, for (V, 0) = (C2, 0), the Poincar´e series P{vCi }(t1, . . . , tr) of a collection of (different) curve valuations coincides with the Alexander polynomial ∆C(t1, . . . , tr) in r variables of the algebraic link Ci for r > 1. (For r = 1 one has PvC (t) = ∆C (t) 1−t ) In [3] it was shown that the same holds for an algebraic link in the Poincar´e sphere. defined by the curve C = rSi=1 Here we discuss to which extend the Alexander polynomial in several vari- ables of an algebraic link in the Poincar´e sphere (that is the Poincar´e series of the corresponding curve) determines the topology of the link. Two curves on (S, 0) with the same (from the combinatorial point of view) minimal resolutions define topologically equivalent links in the Poincar´e sphere. We show that two curves (even irreducible ones) on (S, 0) with combinatorially different minimal resolutions may have equal Alexander polynomials. The (infinite-dimensional) space of arcs on (S, 0) consists of 8 irreducible components. These components are in one-to-one correspondence with the components of the exceptional di- visor of the minimal (good) resolution of (S, 0). A component of the space of arcs consists of all arcs whose strict transforms intersect the corresponding component of the exceptional divisor. We show that, if the strict transform of a (possibly reducible) curve on (S, 0) does not intersect one particular com- ponent of the exceptional divisor, namely the one corresponding to the end of the longest tail in the corresponding E8-diagram, then its Poincar´e series (that is the Alexander polynomial of the corresponding link) determines the combinatorial type of the minimal resolution of the curve and therefore the topology of the corresponding link. We discuss an analogous question for a collection of divisorial valuations on the E8 surface singularity. We show that, if no divisor from the collec- tion is born by a sequence of blow-ups starting from a smooth point of the same component as above, the Poincar´e series of the collection determines the combinatorial type of the minimal resolution of the collection of valuations. The E8 surface singularity is the quotient of the plane C2 by the binary icosahedral group. Therefore the results of this paper may have an interpre- tation it terms of equivariant topology of curves and/or divisors on the plane with the binary icosahedral group action. 3 2 Poincar´e series of curve and divisorial valu- ations on the E8-singularity Let (S, 0) be a normal surface singularity of type E8 and let (Ci, 0), i = 1, 2, . . . , r, be (different) irreducible curves (branches) on (S, 0). Let (C, 0) = i=1(Ci, 0). The curves (Ci, 0) define curve valuations on the ring OS,0 of germs of functions on (S, 0) in the usual way (see Section 1). Let PC(t1, . . . , tr) be the Poincar´e series of this set of valuations. Sr Let π : (X , D) → (S, 0) be an embedded resolution of the curve C = Sr i=1 Ci. This means that: 1) X is a smooth surface, D = π−1(0); 2) π is a proper complex analytic map; 3) the total transform π−1(C) of the curve C is a normal crossing divisor on X (this implies that the exceptional divisor D is a normal crossing divisor on X as well). Let D = Sσ∈Γ Dσ be the decomposition of the exceptional divisor into irreducible components. All the components Dσ are isomorphic to the complex projective line CP1. Let (Dσ · Dδ) be the intersection matrix of the components Dσ. All the self-intersection numbers Dσ · Dσ are negative, for σ 6= δ the intersection number Dσ · Dδ equals 1 if the components Dσ and Dδ intersect each other and equals 0 otherwise. The entries of the minus inverse matrix (mσδ) = −(Dσ · Dδ)−1 are positive integers. (Let us recall that we consider the case of an E8-singularity.) Let eCi be the strict transform of the branch Ci, i. e., the closure of the preimage π−1(Ci \{0}), i = 1, 2, . . . , r, eC = Sr i=1 eCi. Let Dσ(i) be the component of the exceptional divisor D intersecting the strict transform eCi ◦ and let mσ := (mσσ(1), . . . , mσσ(r)) ∈ Zr Dσ be the "smooth part" of the component Dσ in the total transform π−1(C), i. e., the component Dσ minus the intersection points with all other components of the exceptional divisor >0. Let and with the strict transforms eCi. In [3] it was shown that ◦ PC(t) = Y σ∈Γ (1 − tmσ )−χ( Dσ) , (1) where t := (t1, . . . , tr), tm := tm1 1 the Euler characteristic. Remarks. 1. One has an essential difference between Equation 1 and the corresponding equation for all other rational surface singularities from [3]. For , for m = (m1, · · · , mr) ∈ Zr, χ(·) is · · · tmr r 4 any other rational surface singularity the numbers mσδ are, generally speaking, not integers and the Poincar´e series is obtained from a certain rational power series (somewhat similar to (1)) in variables Tσ corresponding to all the com- ponents of the exceptional divisor D by eliminating all the monomials with non-integer exponents and subsequent substitution of each variable Tσ by a product of variables t1, . . . , tr corresponding to the branches. 2. Equation 1 gives the Poincar´e series PC(t) in the form Y m∈(Z≥0)r\{0} (1 − tm)sm, (2) where sm are integers. For the E8-singularity this product has finitely many factors. This does not hold, in general, for a curve on an arbitrary rational surface singularity. Any power series in the variables t1, . . . , tr with the free term equal to 1 has a unique representation of the form 2 (generally speaking, with infinitely many factors). The dual graph of the minimal resolution of the E8-singularity has the standard E8 form (see Figure 1), all the self-intersection numbers of the com- ponents are equal to −2. An embedded resolution π : (X , D) → (S, 0) of the curve C is obtained from the minimal resolution of the singularity (S, 0) by a sequence of blow-ups made (at each step) at intersection points of the strict transform of the curve C and the exceptional divisor. Some intersection points of the strict transform of the curve C and the exceptional divisor may be at the same time intersection points of components of the exceptional divisor. Let π′ : (X ′, D′) → (S, 0) be the resolution of (S, 0) obtained only by all the blow-ups at the points of this sort. The dual graph of the resolution π′ is of the "three-tails" form and is obtained from the one of the minimal resolution by inserting some (maybe zero) new vertices between the vertices of the mini- mal one. The strict transform (π′)−1(Ci \ {0}) of the branch Ci, i = 1, . . . , r, intersects the exceptional divisor D′ at a smooth point of it, i. e., not at an intersection points of its components. Let Γ0 ⊂ Γ be the set of indices σ numbering the components of D′. A divisorial valuation v on OS,0 is defined by a component of the excep- tional divisor D = π−1(0) of a resolution π : (X , D) → (S, 0) of the singularity. For a function germ f ∈ OS,0, the value v(f ) is the multiplicity of the lifting f ◦ π of the function f to the space X of the resolution along the corresponding component. Let v1, . . . , vr be divisorial valuations on OS,0. A resolution of the collection {vi} of divisorial valuations is a resolution π : (X , D) → (S, 0) whose exceptional divisor contains all the components defining the valuations. For a Dσ be the represen- resolution π : (X , D) → (S, 0) of the set {vi}, let D = Sσ∈Γ 5 tation of the exceptional divisor as the union of its irreducible components, let • the integers mσδ be defined as above, and let Dσ be the "smooth part" of the component Dσ in the exceptional divisor D, i. e., the component Dσ minus the intersection points with all other components of the exceptional divisor. For i ∈ {1, . . . , r}, let Dσ(i) be the component of the exceptional divisor D defining the divisorial valuation vi. Just as in [3] (see also [7] and [2]), one can show that the Poincar´e series of the set {vi} of divisorial valuations is given by P{vi}(t1, . . . , tr) = Y σ∈Γ (1 − tmσ )−χ( Dσ) . • (3) For a component Dσ of the exceptional divisor D of the resolution π, let ℓσ be a germ of a smooth curve on X transversal to Dσ at a smooth point of • D (i. e., at a point of Dσ), let (Lσ, 0) ⊂ (S, 0) be the blow-down π(ℓσ) of the curve ℓσ and let Lσ be given by an equation hσ = 0 with hσ ∈ OS,0. (Let us recall that an arbitrary curve germ on the E8 surface singularity is Cartier, i. e., is defined by an equation.) The curve germ Lσ and/or the function germ hσ are called a curvette at the component Dσ. 3 The Poincar´e polynomial of an irreducible curve and the topological type For a plane valuation centred at the origin (say, for a curve or for a diviso- rial one) the Poincar´e series determines the combinatorial type of the minimal resolution: [4]. This does not hold, in general, for a valuation on a surface sin- gularity. The problem is partially related with the following one. A resolution of a valuation (a curve or a divisorial one) on a surface singularity (S, 0) is at the same time a resolution of the singularity itself. The minimal resolution of the valuation starts from a certain point on the exceptional divisor of the minimal resolution of the surface. Therefore a possibility to determine the combinatorial type of the minimal resolution of a valuation from its Poincar´e series assumes that it is possible to determine the component (or the intersec- tion of two components) of the exceptional divisor of the (minimal) resolution of the surface from which the resolution of the valuation starts (up to pos- sible symmetries of the dual graph of the minimal resolution of the surface). However, in general this is not possible. Example. The exceptional divisor of the minimal resolution of the Ak surface singularity consists of k irreducible components. One can show that a curvette at each of these components is smooth. This follows, e. g., from the results of 6 [9]. Also this can be deduced from the computation of the Poincar´e series of the curvettes using [3, Theorem 2] (which gives P (t) = 1 1−t ). It appears that the same problem can be met for valuations on the surface singularity of type E8. Examples. 1. The dual graph of the minimal resolution of the E8-singularity is shown on Figure 1. The exceptional divisor consists of 8 irreducible compo- 5 r 6 r 7 r 8 r 1 r 2 r 3 r 4 r Figure 1: The dual graph of the E8-singularity. nents D1, . . . , D8 numbered as in the order shown on Figure 1. Let C ′ be a curvette at the component D6 and let C ′′ be the blow down of the plane curve singularity of type A4 (that is with a local equation u5 + v2 = 0) transversal to the component D8 at a smooth point of the exceptional divisor (that is, not at the intersection point of D8 with D7). The minimal resolution of the curve C ′ coincides with the minimal resolution of the surface. The dual graph of the minimal resolution of the curve C ′′ is shown on Figure 2. Using (1) one can show that PC ′(t) = PC ′′(t) = . (1 − t12)(1 − t18) (1 − t4)(1 − t6)(1 − t9) Thus the Poincar´e series of a curve valuation on the E8-singularity does not determine the combinatorial type of its minimal resolution. Moreover, it does not determine the component of the exceptional divisor of the minimal reso- lution of the surface singularity intersecting the strict transform of the curve. 5 r 6 r 7 r 1 r 2 r 3 r 4 r 11 r 8 r 9 r 10 r 12 r ❅❅❘ Figure 2: The dual graph of the minimal resolution of the curve L2. 7 (We do not know whether or not the (algebraic) links in the Poincar´e sphere corresponding to the curves L1 and L2 are topologically equivalent.) 2. Let D′ be the divisor born under the blow-up of the component D12 of the resolution shown on Figure 2 (at a smooth point of the exceptional divisor) and let D′′ be the divisor born after 7 blow-ups starting at a smooth point of the component D6 and produced at each step at a smooth point of the previously born divisor. Let ν′ and ν′′ be the divisorial valuations defined by the divisors D′ and D′′ respectively. Using (3) one can show that Pν′(t) = Pν′′(t) = (1 − t12)(1 − t18) (1 − t4)(1 − t6)(1 − t9)(1 − t19) . Thus the Poincar´e series of a divisorial valuation on the E8-singularity does not determine the combinatorial type of its minimal resolution. Other examples of this sort can be obtained by applying the same additional modifications at smooth points of the divisors D′ and D′′. The examples show that one cannot restore, in general, the combinatorial type of the (minimal) resolution of a valuation (say, of an irreducible curve) on the E8 surface singularity from its Poincar´e series. However, often this is the case. According to [11] the space of arcs on the E8-singularity consists of 8 irreducible components. Each of them is the closure of the subspace of arcs whose strict transforms intersect the exceptional divisor of the minimal resolu- tion of the surface at one of the components D1, . . . , D8. Let us denote these spaces of arcs by E1, . . . , E8 respectively. One can show that only arcs from E6 and E8 may have the same Poincar´e series. Moreover, if one restricts the consideration only to the arcs not intersecting the component D8 at a smooth point of the exceptional divisor, one can determine the combinatorial type of the (minimal) resolution of the arc from its Poincar´e series. The same holds for a reducible curve: if (C, 0) = (Ci, 0) ⊂ (S, 0) and the strict transforms rSi=1 of the branches Ci do not intersect the component D8 of the exceptional di- visor at smooth points (that is, they belong to the union Ei), the Poincar´e 7Si=1 series of the curve C (in r variables) determines the combinatorial type of its minimal resolution. We shall show that analogues of these statements hold for divisorial valuations as well. Let (C, 0) ⊂ (S, 0) be an irreducible curve on the E8 surface singularity (S, 0) such that its minimal embedded resolution does not start from a smooth point of the component D8 of the minimal resolution of (S, 0). Let PC(t) be the Poincar´e series of the corresponding (curve) valuation on OS,0. Let us recall 8 that PC(t) = ∆C (t) in the Poincar´e sphere L. 1−t , where ∆C(t) is the Alexander polynomial of the knot C ∩L Theorem 1 The Poincar´e series PC(t) determines the combinatorial type of the minimal embedded resolution of the irreducible curve C ⊂ S (and therefore the topological type of the knot (L, C ∩ L), where L is the link of the E8 surface singularity (S, 0), i. e., the Poincar´e sphere). Proof. Let π : (X , D) → (S, 0) be the minimal embedded resolution of the curve (C, 0) and let π′ : (X ′, D′) → (S, 0) be the resolution of the surface Dσ. The resolution π′ singularity (S, 0) described in Section 2, D′ = Sσ∈Γ0 either is the minimal resolution of the singularity (S, 0), or is obtained from the minimal one by blow-ups points inbetween two particular components of it. In the latter case the dual graph of the resolution π′ is obtained from the one of the minimal resolution by inserting several vertices inbetween two neighbouring vertices Di and Dj of the E8-graph. Each component Dσ0 (σ0 ∈ Γ0) inbetween Di and Dj is characterized by a pair of coprime positive integers s1 and s2 so that one has mkσ0 = s1mki + s2mkj for 1 ≤ k ≤ 8. Remark. If one knows the numbers i and j of the components and the ratio s1/sj for the component Dσ0, one knows the resolution π′ itself. Either the resolution π coincides with the resolution π′ (then C is a curvette at a component Dσ0 of D′) or it is obtained from π′ by a sequence of blow-ups starting at a smooth point of a component Dσ0. The matrix (mij) = −(Di · Dj)−1 (minus the inverse of the intersection ma- trix of the components of the minimal resolution of the E8 surface singularity) is equal to 8   5 4 8 6 10 7 6 14 20 10 16 12 4 2 7 4 10 20 30 15 24 18 12 6 10 15 5 3 8 16 24 12 20 15 10 5 4 12 18 6 3 12 8 4 2 4 6 2 15 12 8 10 5 4 8 6 3 9 6 3 8 12 9   (4) For 1 ≤ i ≤ 8, let Qi = (mi8, mi1, mi4) ∈ R3. For i, j such that the components Di and Dj intersect, let Iij be the segment between the points Qi and Qj (that is, the set of points of the form λQi + (1 − λ)Qj with 0 ≤ λ ≤ 1). Let us consider the one-dimensional simplicial complex G in R3 with the vertices Qi and the edges Iij. (As an abstract graph G is isomorphic to the E8-graph.) 9 mk4/mk1 ✻ 1.7 1.5 6 r r 7 r 8 r r4 5 r3 ☞☞ r2 ☞ ☞ ☞ ☞ r1 1.0 0.0 0.5 1.0 ✲ mk8/mk1 Figure 3: The graph G. Lemma 1 The image of the graph G in RP2 (under the natural quotient map R3 \ {0} → RP2) is a planar graph, i. e. images in RP2 of different points of G are different. A proof is obtained by drawing the image of G in an affine chart of RP2: Figure 3. Remark. One can see that the graph embedded into the projective plane consists of straight lines in between the rupture points and the deadends. This is a general property for the image in the projective space of the dual resolution graph of a surface singularity under the map which sends a vertex σ to the ratio of the "multiplicities" mσiσ (that is of the elements of the minus inverse of the intersection matrix) for deadends σi of the graph. Lemma 1 says that the ratios of the three coordinates of different points of G never coincide. Lemma 2 The Poincar´e series PC(t) of the curve C determines the resolution π′ and the component Dσ0 in D′ intersecting the strict transform of the curve C. Proof. Let us write the Poincar´e series PC(t) in the form qY i=1 (1 − tmi)−1 Y m>0 (1 − tm)sm, 10 (5) where m1 ≤ m2 ≤ . . . ≤ mq (thus the first product may have repeated factors) and in the second product the (integer) exponents sm are non-negative and are equal to zero for m = mi, i = 1, . . . , q. Let us recall that the representation of the Poincar´e series in this form is unique. This follows from the fact that any series from 1 + tZ[[t]] has a unique representation in the form of a (finite or infinite) product (1 − tm)sm with integer exponents sm. ∞Qm=1 The representation (5) may not have less than two binomial factors with the exponent (−1). As a rational function the Poincar´e series PC(t) has the form of a polynomial (in fact the Alexander polynomial of the corresponding algebraic link) divided by (1 − t). One cannot have a degree of the binomial (1 − t) in (5) since all the entries of the matrix (4) are greater than 1. If there is one factor (1 − tm1)−1 with m1 > 1, its poles at the degree m1 roots (1 − tm)sm. of 1 different from 1 have to be zeroes of the second product Qm This implies that this product contains a binomial (1 − tkm1) with a non-zero exponent skm1 and therefore the series itself is a polynomial. the component D1, then the ratio m2/m1 is greater than 2. This follows from the fact that the exponent m1 is equal to ℓm18 = 2ℓ, where ℓ is the intersection If the strict transform eC (in the space X ′ of the resolution π′) intersects number eC ·D1, and the exponent m2 is either equal to ℓm14 = 5ℓ or corresponds If the strict transform eC does not intersect the component D1 (and inter- m2 = ℓmσ01, where ℓ = eC · Dσ0. (This follows from the fact that in the ma- to a divisor born from D1 under some blow-ups. In the last case it is greater that ℓm11 = 4ℓ. The ratio m2/m1 > 2 cannot be met in other cases: see below. sects a component Dσ0 of the exceptional divisor D), one has m1 = ℓmσ08, trix (4) one has mi8 ≤ mi1 ≤ mik for 1 < k < 8 and all i.) From Figure 3 one can see that 1 < m2/m1 < 2. Moreover, if m2/m1 6= 5/3, the ratio m2/m1 determines the component Dσ0 (that is the components Di and Dj from the minimal resolution graph of the E8 surface singularity and the corresponding ratio s1/s2). The ratio m2/m1 is equal to 5/3 if and only if the strict transform eC intersects either D3, or D4, or a component inbetween D3 and D4 in D′. In this case if the strict transform eC does not intersect D4, there are at least three binomial factors with the exponent −1 and one has m1 = ℓmσ08, m2 = ℓmσ01, m3 = ℓmσ04. Lemma 1 implies that the ratio m3 : m2 : m1 determines the component Dσ0. If there are less than three binomial factors with the exponent −1 or the ratio m3 : m2 : m1 is different from 8 : 5 : 3, the strict transform eC intersects the component D4. (cid:3) Remark. One can avoid arguments from the first paragraph of the proof formulating the criterium for the strict transform eC not to intersect the com- 11 ponent D1 in the following form: this holds if and only if the Poincar´e se- ries PC(t) contains at least two binomial factors with the exponent (−1) and m2/m1 ≤ 2. The formal negation says that the strict transform eC intersects the component D1 if and only if either the Poincar´e series PC(t) contains less than two binomial factors with the exponent (−1) or m2/m1 > 2. The first option does not take place, but this does not contradict the statement. This sort of formulation can be useful for the proof of a version of this Lemma for a divisorial valuation in Section 4. Lemma 2 says that the Poincar´e series PC(t) determines the (minimal) modification π′ : (X ′, D′) → (S, 0) and the component Dσ0 containing the In particular, one knows the multiplicity mσ0σ0. In terms of the decomposition (5) one has m1 = ℓm8σ0 and point P = eC ∩ D′ (a smooth point of D′). therefore the intersection number ℓ = eC · Dσ0 is determined by the Poincar´e series PC(t). Let D(t) := (1 − tℓm1σ0 )−1(1 − tℓm4σ0 )−1(1 − tℓm8σ0 )−1(1 − tℓm3σ0 ) . Let π′′ : (X , D′′) → (X ′, P ) be the minimal embedded resolution of the Fδ. The dual graph Γ1 (with an germ (eC ∪ Dσ0, P ) ⊂ (X ′, D′), D′′ = Sδ∈Γ1 arrow representing the strict transform of the curve eC by π′′) is of the form shown on Figure 4. Here τ0 marks the divisor corresponding to the first blow- r r r τ0 r r ♣ ♣ ♣ r r r r δ1 r r r r τ1 δ2 r r r r r τ2 r r δg ✒ r r r r r r τg Figure 4: The graph Γ1. is the last component of D′′ (i. e., the component with self-intersection −1): the graph Γ1. Let (ℓτi, P ) ⊂ (X ′, P ) be a curvette at the component Dτi and up at the point P , g is the number of Puiseux pairs of the curve eC and Fδg the strict transform of eC intersects Fδg ; τi, i = 0, 1, . . . , g, are the deadends of let βi be the intersection number eC · ℓτi. One has β0 < β1 < · · · < βg and of the curve germ (eC, P ) (in particular β0 is the multuiplicity of eC). Moreover, {βii = 0, 1, . . . , g} is the minimal set of generators of the semigroup of values the sequence β0, . . . , βg determines the graph Γ1. Let Q(t) := sY j=0 (1 − tmδg τj )−1 · sY j=1 (1 − tmδg δj ) . 12 If eC is a curvette at the component Dσ0 one defines Q(t) to be equal to 1 and Γ1 is only an arrow without any vertex. If eC is smooth but eC · Dσ0 = ℓ > 1 (that is eC is tangent to Dσ0), we put Q(t) = (1 − tmδg τ0 )−1 (the dual graph is shown in Figure 5). Pay attention that r✒δ rσ0 r8 r1 r r τ0 r4 Figure 5: Case 2 mδg τ0 < mδg τ1 < . . . < mδg τg , mδgτj < mδgδj < mδgτj+1 . The modification π = π′ ◦ π′′ : (X ′, D) → (S, 0) is the minimal resolution of the curve C. The dual graph of this resolution (i. e., the one of the modification π with an arrow corresponding to C added) is obtained by joining the graphs Γ0 and Γ1 by an edge between σ0 and a vertex δ ∈ Γ1. If eC is a curvette at the component Dσ0, the graph Γ is obtained from the graph Γ0 by attaching an arrow to the vertex σ0 ∈ Γ0. This case (we refer to it as Case 1 in the sequel) is characterized by the condition ℓ = 1. One has PC(t) = D(t)(1 − tmσ0σ0 ) . We will consider several cases corresponding to essentially different possi- bilities for the position of the vertex δ in Γ1. Taking into account the fact that mδg σ = ℓmσσ for all σ ∈ Γ0, one can show that in all these cases one has PC(t) = D(t)Q(t)(1 − tmδg δ)(1 − tmδg σ0 ) (see the discussion of the cases below). The exponent mδg σ0 is equal to ℓmσ0σ0 and the series D(t) is known. Therefore the remaining part of the proof consists in the computation of δ, mδgδ and β0, . . . , βg from the (known) series B(t) = Q(t)(1 − tmδg δ). Let us write the series B(t) in the form B(t) = rY k=1 (1 − tmk )−1 · 13 rY k=1 (1 − tnk) with m1 < · · · < mr and n1 < · · · < nr. In the following µ will denote the integer µ = m1 − mδgσ0. Case 2. The curve eC is smooth but ℓ > 1. In this case one has δ = δg, mδg τ0 = ℓmσ0σ0 + 1 and mδgδ = ℓ(mδgσ0 + 1) > mδg τ0. The graph Γ1 is shown in Figure 5 and the series B(t) is equal to B(t) = (1 − tmδg τ0 )−1 · (1 − tℓmδg τ0 ) . Note that in this case µ = 1. Case 3. The curve eC is non smooth but it is transversal to the component Dσ0. In this case δ = τ0 and mδg δ = mδg τ0. The binomial factor (1 − tmδg δ) = (1 − tmδg τ0 ) does not participate in the decomposition of B(t). The integer ℓ = β0 coincides with the multiplicity of eC at P and one has m1 = mδg τ1 = ℓmσ0σ0 + β1 = mδg σ0 + β1 . For µ = m1 − mδg ,σ0 one has µ > ℓ > 1. Comparing with all the other cases below, one can see that these conditions characterize the case 3. One has β0 = ℓ, β1 = µ and the graph Γ is shown on Figure 6. Simple computations δ = τ0 r δ1 r ♣ ♣ ♣ r τ1 r 8 r1 rσ0 r4 Figure 6: Case 3 r r r✒δg r τg using the well-known description of the resolution process of a plane curve singularity give us mδg τj = Nj−1 · · · N1mδg σ0 + βj , mδg δj = Nj · · · N1mδg σ0 + Njβj = Njmδgτj for j = 1, . . . , g. Here Nk = ek−1/ek, where ek = gcd(β0, . . . , βk) for k = 0, 1, . . . , g. Note that Nk depends only on the numbers β0, . . . , βk. The equations above permit to compute the sequence β0, . . . , βg starting from the (known) sequence β0 < mδg τ1 < · · · < mδgτg . The expression for B(t) implies that g = r, mj = mδgτj for j ≥ 1. 14 Case 4. The component Dσ0 has the maximal contact with eC, i. e., ℓ = eC · Dσ0 = β1. In this case δ = τ1 and mδg δ = mδg τ1. As a consequence the binomial factor (1 − tmδg δ ) = (1 − tmδg τ1 ) does not participate in the decomposition of the series B(t). In this case one has m1 = mδg τ0 = ℓmσ0σ0 + β0 = mδg σ0 + β0 . This case is characterized by the conditions µ = m1 − mδgσ0 < ℓ and µ does not divide ℓ. The graph Γ is shown in Figure 7. As in the previous case one r r r✒δg r τg rτ0 δ1 r δ2 r ♣ ♣ ♣ δ = τ1 r rσ0 r τ2 r 8 r1 r4 Figure 7: Case 4 mδg τj = Nj−1 · · · N2 β1 e1 mδg σ0 + βj , mδg δj = Nj · · · N2 β1 e1 mδg σ0 + Njβj = Njmδgτj has for j = 1, . . . , g. These equations permit to compute the sequence β0, . . . , βg starting from the sequence mδg τ2 < · · · < mδg τg and the (known) integers µ and ℓ. The expression for B(t) implies that g = r, mj = mδg τj for j ≥ 2, nj = mδg δj for j ≥ 1. Thus the integers m1, . . . , mr permit to compute the integers β0, . . . , βg. Case 5. The component Dσ0 is tangent to eC but does not have the maximal contact with it. This case is equivalent to the condition ℓ = k · β0 for some integer k with 1 < k < β1/β0. The vertex δ is the k-th vertex of the geodesic in Γ1 from τ0 to δ1. In this case one has m1 = mδg τ0 = ℓmσ0σ0 + β0 = mδg σ0 + β0 . The case is characterized by the conditions µ = m1 − mδg σ0 < ℓ and µ(= β0) divides ℓ(= kβ0). The graph Γ is shown in Figure 8. As in the previous case 15 rτ0 δ r δ1 r ♣ ♣ ♣ r τ1 r 8 r1 rσ0 r4 Figure 8: Case 5 r r r✒δg r τg mδg δ = kmδg τ0 = k(mδg σ0 + β0) , mδg τj = Nj−1 · · · N1kmδg σ0 + βj , mδg δj = Nj · · · N1kmδg σ0 + Njβj = Njmδgτj one has for j = 1, . . . , g. The expression for B(t) implies that r = g + 1, mj = mδg τj for j ≥ 0; n1 = mδg δ and nj = mδgδj−1 for j ≥ 2. Thus the integers m1, . . . , mr permit to compute the intergers β0, . . . , βg. Remark. The characterizations of the different possibilities for the location of the vertex δ are already made in the course of the analysis of the different cases above. For a convenience of understanding let us summarize these char- acterizations. Let ℓ = eC · Dσ0 be the intersection multiplicity between eC and Dσ0 and let µ := m1 − mδg σ0. Then one has Case 1 2 3 4 5 ℓ 1 µ 1 > 1 > 1 > 1 µ < ℓ and µ 6 ℓ µ < ℓ and µℓ > 1 µ > ℓ β0 β1 1 1 ℓ µ µ µ ℓ (cid:3) Remark. The Poincar´e series of a curve (reducible or irreducible) on the E8 surface singularity being either the Alexander polynomial of the corresponding link or the Alexander polynomial divided by 1 − t, is a topological invariant of the curve. Therefore Theorem 1 means that for i 6= j, i, j ≤ 7, all arcs from the component Ei are not topologically equivalent to arcs from the component Ej (except those from the intersection of the components). 16 4 The Poincar´e polynomial of one divisorial valuation and the topological type Let v be a divisorial valuation defined by a component Dσ∗ of the exceptional divisor D of a resolution π : (X , D) → (S, 0) of the E8 surface singularity (S, 0). We assume that π is the minimal modification containing the compo- nent defining the valuation, i. e., the minimal resolution of the valuation v. Let π′ : (X ′, D′) → (S, 0) be the resolution of the surface described above and let Dσ0 be the component of the exceptional divisor D′ which either coincides with Dσ∗ or is such that Dσ∗ is born by a sequence of blow-ups starting at a point of Dσ0 (smooth in D′). Assume that σ0 6= 8. This means that the component Dσ∗ does not originate from a point of D8 (smooth in D′). Theorem 2 In the described situation the Poincar´e series Pv(t) of the divi- sorial valuation v determines the combinatorial type of the minimal resolution of the valuation. Proof. One has the following analogue of Lemma 2. Lemma 3 The Poincar´e series Pv(t) of the divisorial valuation v determines the resolution π′ and the component Dσ0 in D′. Proof. The proof is essentially the same as of Lemma 2. The component Dσ0 coincides with D1 if and only if either the Poincar´e series Pv(t) contains less than two binomial factors with the exponent (−1) or m2/m1 > 2. (Again, as in the curve case, the first option does not take place, but it is easier to If Dσ0 6= D1, the Poincar´e series Pv(t) contains at avoid a proof of that.) least two binomial factors with the exponent (−1) and one has m1 = ℓmσ08, m2 = ℓmσ01, where ℓ is the intersection number of the strict transform in X ′ of a curvette at the component defining the valuation with the component Dσ0. One has 1 < m2/m1 < 2. If m2/m1 6= 5/3, the ratio m2/m1 determines the component Dσ0. The ratio m2/m1 is equal to 5/3 if and only if the component Dσ0 either coincides with D3, or coincides with D4, or is produced by blow-ups inbetween D3 and D4. If Dσ0 6= D4, the Poincar´e series contains at least three binomial factors with the exponent −1 and the ratio m3 : m2 : m1 determines the component Dσ0 (Lemma 1). If the Poincar´e series contains less than three binomial factors with the exponent −1 or the ratio m3 : m2 : m1 is different from 8 : 5 : 3, one has Dσ0 = D4. (cid:3) The vertices σ∗ and σ0 coincide if and only if Pv(t) = (1 − tm1σ0 )−1(1 − tm4σ0 )−1(1 − tm8σ0 )−1(1 − tm3σ0 ) . 17 Let σ∗ 6= σ0 and let π′′ : (X , D′′) → (X ′, P ) be the minimal modification of X ′ containing the component Dσ∗ defining the valuation v (P ∈ Dσ0). Let (ℓσ∗, P ) ⊂ (X ′, P ) be a curvette at the component Dσ∗. The dual graph of the modification π′′ differs from the dual graph of the minimal resolution of the curve eC = ℓσ∗ by a tail of length k ≥ 0 attached to the vertex δg corresponding to the curve eC: see Figure 9. The only difference with the case δg δ1 r r r r τ0 r r r δ2 r r r r τ1 r r r r τ2 r r ♣ ♣ ♣ r r r r r r r❡ σ∗ r r r r r τg Figure 9: Resolution graph for a divisorial valuation. of an irreducible curve treated in Theorem 1 above (and applied to the curve C = π′(eC)) consists in the necessity to find the length k of the tail. In the case k = 0 the vertex σ∗ just coincides with δg. In this case the Poincar´e series Pv(t) does not contain the factor (1 − tmδg δg ) (since now there is no arrow at the vertex δg). If k > 0 then, in order to obtain the Poincar´e series Pv(t), one has to add the factor (1 − tmσ∗ σ∗ )−1 to the decomposition of the Poincar´e series PC(t). In any case one has Pv(t) = PC(t)(1 − tmσ∗σ∗ )−1. The intersection number ℓ = eC · Dσ0 can be determined from the Poincar´e series Pv(t) just in the same way as for a curve valuation. As a consequence, the proof for a divisorial valuation almost repeats the one of Theorem 1 for an irreducible curve with the additional duty to determine the component (vertex) σ∗ and the multiplicity mσ∗σ∗. The analogue of the series B(t) considered in the proof of Theorem 1 is the series Bv(t) = B(t)(1 − tmσ∗σ∗ )−1 = Q(t)(1 − tmδg δ)(1 − tmσ∗σ∗ )−1. (Notice that for all σ not in the tail one has mσ∗σ = mδgσ.) Let us write the series Bv(t) in the form Bv(t) = rY k=1 (1 − tmk )−1 · with m1 < · · · < mr and n1 < · · · < nr−1. 18 r−1Y k=1 (1 − tnk) Case 2. The component of the exceptional divisor corresponding to the vertex σ∗ is produced by k supplementary blow-ups at smooth points starting at a point of δg. If k = 0, i. e., if σ∗ = δg = δ, the series Bv(t) consists only of one term (1 − tmσ∗τ0 )−1. Otherwise Bv(t) = (1 − tm1)−1(1 − tm2)−1(1 − tn1) and k = m2 − n1. Cases 3, 4, 5. From the proof of Theorem 1, it follows that the inte- gers m1, . . . , mi alongside with β0 and β1 permit to determine the numbers β0, . . . , βi. For i = 1, . . . , r let εi := gcd(β0, β1, m1, . . . , mi). One can see that k = 0 if and only if εr < εr−1. In this case σ∗ = δg. Otherwise one has k = mr − nr−1, mσ∗σ∗ = mδg δg + k. (cid:3) 5 The Poincar´e polynomial of a collection of divisorial valuations and the topological type Let vi, i = 1, . . . , r, be divisorial valuations defined by components of the exceptional divisor D of a resolution π : (X , D) → (S, 0) of the E8 surface singularity (S, 0). We assume that π is the minimal modification containing the components defining the valuations, i. e., the minimal resolution of the col- lection {vi} of valuations. The resolution π can be obtained from the minimal resolution of the E8 surface singularity (S, 0) by a sequence of blow-ups such that at first some of them are made at intersection points of the components of the exceptional divisor (and produce a resolution π′ : (X ′, D′) → (S, 0) with a "three tails" dual graph) and later additional blow-ups do not touch the intersection points of the components of D′, but start from smooth points of D′. Assume that the modification π of π′ does not include blow-ups of smooth (in D′) points of the component D8. Theorem 3 In the described situation the Poincar´e series P{vi}(t), t = (t1, . . . , tr), of the collection {vi} of divisorial valuations determines the combinatorial type of the minimal resolution of collection. Proof. Equation 3 implies the following projection formula: if {i1, . . . , iℓ} ⊂ {1, . . . , r}, then the Poincar´e series of the ℓ-index filtration corresponding to the divisorial valuations vi1,. . . , viℓ is obtained from the Poincar´e series P{vi}(t1, . . . , tr) by substituting the variables ti with i /∈ {i1, . . . , iℓ} by 1. Remark. The last property does not hold for the Poincar´e series of the filtra- tion defined by a collection of curve valuations. This makes the proof of the 19 corresponding statement for curves valuations (Theorem 4 below) somewhat more complicated. The dual graph of the minimal resolution of a set of divisorial valuations is determined by the dual graph of the minimal resolution for each divisor plus the deviation points of the resolutions for each pair of divisors. The projection formula alongside with Theorem 2 imply that the Poincar´e series P{vi}(t) determines the minimal resolution graph of each valuation from the collection and, in particular, the component of the exceptional divisor D′ of the modification π′ which the resolution of the divisorial valuation start from. If, for two divisorial valuations from the collection, these starting components are different, one does not need to find the deviation point. Assume that the starting components coincide. In order to find the deviation point, without lost of generality (due to the projection formula) one may assume that r = 2, i. e., that the collection consists of these two valuations: v1 and v2. In this situation one has the following picture. In the dual resolution graph on the geodesic inbetween the vertices σ1 and σ2 (defining the divisorial valuation) the ratio mσσ1/mσσ2 (as a function on σ) is strictly monotonous being maximal at the vertex σ1 and minimal at σ2; on the components of the closure of the complement to this geodesic in the dual graph this ratio is constant: see [4, page 43], see also Proposition 1 below for the same statemant in a somewhat different setting. One of the components includes all the vertices corresponding to the components of the exceptional divisor D′ and, in particular, the vertex 8. The factor (1 − tm8σ1 )−1 participates in the decomposition of the Poincar´e series P{v1,v2}(t1, t2). Moreover, its exponent (m8σ1, m8σ2) is the minimal one in it. The splitting point of the resolutions of the valuations v1 and v2 also participates in the decomposition and is maximal in the described components, i. e., among those with mσσ1/mσσ2 = m8,σ1/m8σ2. This determines the splitting point between the resolutions of the valuations v1 and v2. (cid:3) tm8σ2 2 1 6 The Poincar´e polynomial of a reducible curve and the topological type Let C = Sr i=1 Ci be a (reducible: r > 1) curve germ on the surface (S, 0) and let π : (X , D) → (S, 0) be the minimal embedded resolution of the curve (C, 0) ⊂ (S, 0). Let us assume that the resolution process does not contain a blow-up of a smooth (in the exceptional divisor) point of the component D8 of the minimal resolution of the surface (S, 0). 20 Theorem 4 In the described situation the Poincar´e series PC(t), t = (t1, . . . , tr), i=1 Ci determines the combinatorial type of the minimal res- of the curve C = Sr olution of the curve. Proof. We have to show that the Poincar´e series PC(t) determines the minimal resolution graph Γ of C. In the case under consideration one has a projection formula different of the one for divisorial valuations. In what follows, let us denote mσσ(i) (σ(i) is the vertex of Γ such that the σ. Therefore one has mσ = (m1 component Dσ(i) of the exceptional divisor D intersects the strict transform eCi of the curve Ci) by mi σ). The reason (somewhat psychological) for that is the fact that, for a multi-exponent of a term of the Poincar´e series PC(t1, . . . , tr) or of a factor of its decomposition, one knows its components m1 σ, but does not know the vertex σ. One can say that our aim is to find vertices σ(i) corresponding to the curve. σ, . . . , mr σ, . . . , mr Let i0 ∈ {1, . . . , r}. The A'Campo type formula (1) for PC(t) implies that PC(t)ti0 =1 = PC\{Ci0 }(t1, . . . , ti0−1, ti0+1, . . . , tr) · (1 − tmσ(i0))ti0 =1 . (6) Applying (6) several times one gets PC(t)tj =1 for j6=i0 = PCi0 (ti0) · Y i6=i0 (1 − t i0 σ(i) m i0 ) . (7) Pay attention to the fact that mi0 (ti0) can be determined from the Poincar´e series PC(t) if one knows the multiplicity mσ(i0). The strategy of the proof follows the steps from [4] (see also [5]): σ(i0) and therefore the series PCi0 σ(i) = mi 1) To detect an index i0 for which one can find the corresponding multiplic- ity mσ(i0) from the A'Campo type formula for PC(t). Then Theorem 2 and equation (7) permit to recover the minimal resolution graph Γi0 of the curve Ci0. Equation (6) gives the possibility to compute the Poincar´e series PC\{Ci0 }(t1, . . . , ti0−1, ti0+1, . . . , tr) of the curve C \{Ci0}. By induc- tion one can assume that the resolution graph Γi0 of the curve C \ {Ci0} is known. 2) To determine the separation vertex of the curves Ci0 and Cj for j 6= i0 in order to join the graphs Γi0 and Γi0 to obtain the resolution graph Γ. Once we finish the first step, the second one almost repeats the same steps in the proof of Theorem 3 (for divisorial valuations). Therefore we omit the analysis of 2). 21 Let [σ(j), σ(i)] ⊂ Γ be the (oriented) geodesic from σ(j) to σ(i) and let {∆p}, p ∈ Π, be the connected components of Γ \ [σ(j), σ(i)]. For each p ∈ Π there exists an unique ρp ∈ [σ(j), σ(i)] connecting ∆p with [σ(j), σ(i)], i. e., such that ∆∗ p = ∆p ∪ {ρp} is connected. Proposition 1 Let q : Γ → Q be the function defined by q(α) = mj α ∈ Γ. Then one has: α/mi α for 1. The function q is strictly decreasing along the geodesic [σ(j), σ(i)]. 2. For each p ∈ Π, the function q is constant on ∆∗ p. Proof. Let C k (k = 1, . . . , r) be the total transform of the curve Ck in X . One has C k = eCk +X σ∈Γ mk σDσ , where eCk is the strict transform of the curve Ck. For each component Dα, α ∈ Γ, one has C k · Dα = 0 and therefore eCk · Dα +X σ∈Γ mk σ Dσ · Dα = 0. (8) This equation is a consequence of the Mumford formula (see [10, Equation (1)]) applied to the function defining the curve Ck. (Let us recall that on the E8 surface singularity (S, 0) all divisors are Cartier ones.) Dα = 0 and let {ρ1, . . . , ρs} ⊂ Γ be the set of all vertices connected by an edge Lemma 4 Let Dα be a component of the exceptional divisor D such that eCi · with α. Let us assume that either eCj intersects Dα or there exists ρi0 such that q(ρi0) > q(α). Then there exists ρk such that q(α) > q(ρk). Proof. Assume that q(ρk) ≥ q(α) for any k = 1, . . . , s. Applying (8) to Cj and Ci one gets: k=1 sX sX k=1 mj ρk ≥ q(α)mi ρk ≥ α + αD2 0 = eCj · Dα + mi ≥ eCj · Dα + mi = eCj · Dα + q(α)(mj αD2 α + αD2 α + sX k=1 mi ρk ) = eCj · Dα ≥ 0 22 The inequality is strict if eCj ·Dα > 0 or if there exists i0 such that q(ρi0) > q(α). This implies the statement. (cid:3) Let α and β be two vertices of Γ connected by an edge and let q(α) > q(β). Lemma 4 permits to construct a maximal sequence α0, α1, . . . , αk of consecutive vertices starting with α and β (i. e., α0 = α, α1 = β) such that q(αi) > q(αi+1). (We will call a sequence of this sort a decreasing path. If the inequality is in the other direction, the path will be called increasing.) The maximality means that either αk is a deadend of Γ or eCi · Dαk 6= 0. If αk is a deadend, αk−1 is the only vertex connected with αk and Lemma 4 implies that q(αk) = q(αk−1). Therefore the constructed path finishes by the vertex αk = σ(i). Note that, if α ∈ [σ(j), σ(i)] and β /∈ [σ(j), σ(i)], the end of a maximal decreasing (or increasing) path has to finish at a deadend and therefore q(α) = q(β). In particular, this implies that the function q is constant on each connected set ∆∗ p. Assume that σ(i) 6= σ(j). Lemma 4 implies that there exists a vertex α1 connected with σ(j) such that q(σ(j)) > q(α1). Therefore the maximal de- creasing path starting with σ(j) and α1 coincides with the geodesic [σ(j), σ(i)]. (cid:3) Proposition 1 implies that, for any fixed i0 and for any j 6= i0 and σ ∈ Γ, one has mj σ/mi0 σ ≥ mj σ(i0)/mi0 σ(i0). Therefore one has 1 mi0 σ mσ(i0) . 1 mi0 mσ ≥ σ(i0) Let PC(t) = Qp sk 6= 0 for all k. For i ∈ {1, . . . , r} let k = k(i) be such that k=1(1 − tnk)sk be the Poincar´e series of the curve C, where 1 ni j nj ≥ 1 ni k nk for all j. Let E ⊂ {1, . . . , p} be the set of indices k such that k = k(i) for some i ∈ {1, . . . , r} and for k ∈ E let A(k) ⊂ {1, . . . , r} denote the set of indices i such that k = k(i). Note that A(k) contains all the indices i ∈ {1, . . . , r} such that nk = mσ(i). Let B(k) be the subset of such indices. Our aim is to show that one can find k ∈ E such that B(k) 6= ∅. Let j ∈ A(k), j /∈ B(k). One has 1 nj k nk > 1 mj σ(j) mσ(j) and therefore χ( ◦ Dσ(j)) = 0. This implies that σ(j) is connected with only one vertex in Γ (plus the arrow corresponding to eCj), i. e., σ(j) is a deadend of the 23 resolution graph of the curve C \ {Cj}. In particular, there are at most two indices i, j ∈ A(k) such that mσ(i) and mσ(j) are different from nk. Moreover, if there are two indices of this sort, the vertex σ ∈ Γ such that nk = mσ is the vertex 3 corresponding to the divisor D3 of the minimal resolution of (S, 0). In D1 and D4. Therefore, if #A(k) ≥ 3, there exists i0 ∈ B(k). fact in this case the strict transforms eCi and fCj are curvettes at the divisors Let k ∈ E be such that B(k) = ∅ and let us assume that nk = m3 (i. e., that the multiplicity nk is the multiplicity of the divisor D3). Let m8 = (m1 8, . . . , mr 8) be the multiplicity of the divisor D8. Notice that m8 is determined by the Poincar´e series PC(t) because the decomposition of the Poincar´e series contains the factor (1 − tm8)−1 and, moreover, the multiplicity m8 is the smallest one appearing in it. 8 = 3. If there exists i ∈ A(k) such that eCi is a curvete at D1, then mi 8 = 2 implies that eCi is a curvette at D1). 8 = 2 (note that mi In this case, if A(k) 6= {1, . . . , r}, one can choose any other k′ ∈ E instead of k. If A(k) = {1, . . . , r}, then r = 2 and the branches C1 and C2 are curvettes at the divisors D1 and D4 (see Figure 10). This situation is equivalent to have the Poincar´e series of the form PC(t1, t2) = (1 − t(2,3))−1(1 − t(10,15)), what gives the statement in this case. Note also that in this case m2 a curvette at the divisor D4 then one has mi does not characterize completely the situation described: 8 = 3. However this condition for nk′ = m7 and 8 = 3. If the both multiplicities appear simultaneously, one can distinguish the first one because nk′ = m7 is always a multiple of m8 (see Proposition 1) but nk is not (in the presence of k′). This permits to determine the index k in this case from the information given by the series PC(t). A(k′) = {j} with eCj a curvette at D7 one has also that mj If A(k) = {i} and eCi is 1 ✒ ✉ r σ = 3 ✉ r r r 8 r ✒ ✉ 4 Figure 10: The case nk = m3, r = 2, B(k) = ∅ Let us now consider the case when one has B(k) = ∅ and nk = mσ 6= m3 for some σ ∈ Γ. In this case one has A(k) = {i} and σ(i) is a deadend of the dual resolution graph of the curve C \{Ci}. In particular, the vertex σ appears after σ(i) in the resolution process of a certain branch Cj, j 6= i, which is not a curvette at Dσ. It is clear that in this case nk < mσ(j) and also nk < nk(j). 24 Thus in this case we take k′ = k(j) and nk′ instead of k and nk. Iterating this procedure one gets k′ such that B(k′) 6= ∅. Note that this situation can be determined from PC(t) taking k ∈ E such that nk is maximal among the elements nk for k ∈ E not excluded on the previous stages. Once we have an index k ∈ E such that B(k) 6= ∅ we have to choose an k = nj index i0 ∈ B(k). Since ni k for i, j ∈ B(k), for the role of i0 one can take an index from A(k) such that ni0 k is the maximal one in {ni k : i ∈ A(k)}. This finishes the step 1) of the proof and thus the proof itself. (cid:3) k for i ∈ B(k) and j ∈ A(k) \ B(k) and ni k > nj References [1] Campillo A., Delgado F., Gusein-Zade S.M. The Alexander polynomial of a plane curve singularity via the ring of functions on it. Duke Math. J., v.117, no.1, 125–156 (2003). [2] Campillo A., Delgado F., Gusein-Zade S.M. Poincar´e series of a rational surface singularity. Invent. Math., v.155, no.1, 41–53 (2004). [3] Campillo A., Delgado F., Gusein-Zade S.M. Poincar´e series of curves on rational surface singularities. Comment. Math. Helv., v.80, no.1, 95–102 (2005). [4] Campillo A., Delgado F., Gusein-Zade S.M. The Poincar´e series of divi- sorial valuations in the plane defines the topology of the set of divisors. Funct. Anal. Other Math., v.3, no.1, 39–46 (2010). [5] Campillo A., Delgado F., Gusein-Zade S.M. On the topological type of a set of plane valuations with symmetries. Math. Nachr., v.290, no.13, 1925–1938 (2017). [6] Campillo A., Delgado F., Kiyek, K. Gorenstein property and symmetry for one-dimensional local Cohen-Macaulay rings. Manuscripta Math., v.83, no.3-4, 405–423 (1994). [7] Delgado F., Gusein-Zade S.M. Poincar´e series for several plane divisorial valuations. Proc. Edinb. Math. Soc. (2), v.46, no.2, 501–509 (2003). [8] Eisenbud D., Neumann W. Three-dimensional link theory and invariants of plane curve singularities. Annals of Mathematics Studies, 110, Prince- ton University Press, Princeton, NJ, 1985. 25 [9] Gonzalez-Sprinberg G., Lejeune-Jalabert M. Families of smooth curves on surface singularities and wedges. Ann. Polon. Math., v.67, no.2, 179–190 (1997). [10] Mumford D. The topology of normal singularities of an algebraic surface and a criterion for simplicity. Publ. Math. de l'IH´ES, v.9, 5–22 (1961). [11] Pe Pereira M. Nash problem for quotient surface singularities. J. Lond. Math. Soc. (2), v.87, no.1, 177–203 (2013). [12] Yamamoto M. Classification of isolated algebraic singularities by their Alexander polynomials. Topology, v.23, no.3, 277–287 (1984). Addresses: A. Campillo and F. Delgado: IMUVA (Instituto de Investigaci´on en Matem´aticas), Universidad de Valladolid, Paseo de Bel´en, 7, 47011 Valladolid, Spain. E-mail: [email protected], [email protected] S.M. Gusein-Zade: Moscow State University, Faculty of Mathematics and Mechanics, Moscow, GSP-1, 119991, Russia. E-mail: [email protected] 26
1101.5462
2
1101
2011-12-27T02:53:40
Arithmetic linear series with base conditions
[ "math.AG" ]
In this note, we study the volume of arithmetic linear series with base conditions. As an application, we consider the problem of Zariski decompositions on arithmetic varieties.
math.AG
math
ARITHMETIC LINEAR SERIES WITH BASE CONDITIONS ATSUSHI MORIWAKI Abstract. In this note, we study the volume of arithmetic linear series with base conditions. As an application, we consider the problem of Zariski decompositions on arithmetic varieties. Introduction Let X be a projective and flat integral scheme over Z. We assume that X is normal and the generic fiber of X → Spec(Z) is a d-dimensional smooth variety over Q. Let Div(X) be the group of Cartier divisors on X and let Div(X)R := Div(X)⊗ZR. A pair D = (D, g) is called an arithmetic R-Cartier divisor of C0-type if D ∈ Div(X)R (i.e. i=1 aiDi for some D1, . . . , Dr ∈ Div(X) and a1, . . . , ar ∈ R), g : X(C) → R∪{±∞} is a locally integrable function invariant under the complex conjugation map F∞ : X(C) → X(C) and, for any point x ∈ X(C), there are an open neighborhood Ux of x and a continuous function ux over Ux such that D =Pr g = ux + rXi=1 (−ai) log fi2 (a.e.) on Ux, where fi is a local equation of Di on Ux for each i. We denote the vector space consisting of arithmetic R-Cartier divisors of C0-type bydDivC0(X)R. R := Rat(X)× ⊗Z R and let Let Rat(X)× b( )R : Rat(X)× R → dDivC0(X)R be the natural extension of the homomorphism Rat(X)× → dDivC0(X)R given by φ 7→ c(φ). Let D be an arithmetic R-Cartier divisor of C0-type on X. We define H0(X, D) and H0 R(X, D) to be For ξ ∈ X, the R-asymptotic multiplicity of D at ξ is given by H0 H0(X, D) :=nφ ∈ Rat(X)× D + c(φ) ≥ (0, 0)o ∪ {0}, R(X, D) :=nφ ∈ Rat(X)× R D + c(φ)R ≥ (0, 0)o ∪ {0}. infnmultξ(D + (φ)R) φ ∈ H0 R(X, D) \ {0}o ∞ R(X, D) , {0}, if H0 otherwise,  µR,ξ(D) := where multξ is the multiplicity of the local ring given by a local equation (for details, see [2, Section 2.8] or [7, SubSection 6.5]). Moreover, for ξ1, . . . , ξl ∈ X Date: 24/December/2011, 12:30 (Kyoto), (Version 1.5). 1991 Mathematics Subject Classification. Primary 14G40; Secondary 11G50. 1 2 ATSUSHI MORIWAKI and µ1, . . . , µl ∈ R≥0, we define the arithmetic linear series of D with base conditions µ1ξ1, . . . , µlξl to be H0(X, D; µ1ξ1, . . . , µlξl) :=nφ ∈ H0(X, D) \ {0} multξi(D + (φ)) ≥ µi (∀i)o ∪ {0}. In addition, its volume is given by cvol(D; µ1ξ1, . . . , µlξl) := lim sup n→∞ log # H0(X, nD; nµ1ξ1, . . . , nµlξl) nd+1/(d + 1)! . The main result of this paper is the following theorem: Theorem 0.1. If ξ1, . . . , ξl ∈ XQ, D is big and µi > µR,ξi(D) for some i, then cvol(D; µ1ξ1, . . . , µlξl) < cvol(D). Let us introduce a Zariski decomposition on X. A decomposition D = P + N is called a Zariski decomposition of D if the following conditions are satisfied (for the positivity of arithmetic R-Cartier divisors of C0-type, see Conventions and terminology 3): (1) P is a nef arithmetic R-Cartier divisor of C0-type. (2) N is an effective arithmetic R-Cartier divisor of C0-type. (3) cvol(P) = cvol(D). It is easy to see that the existence of a Zariski decomposition of D implies the pseudo-effectivity of D (cf. SubSection 4.1). Let Υ(D) be the set of all nef arithmetic R-Cartier divisors M of C0-type with M ≤ D. Note that Υ(D) is not empty if and only if there is a decomposition D = P+N with the conditions (1) and (2). For a non- big pseudo-effective arithmetic R-Cartier divisor D of C0-type, the non-emptyness of Υ(D) is a non-trivial problem. It is closely related to the fundamental question raised in the paper [9]. Further, in the case where d = 1 and X is regular, once we can see Υ(D) , ∅, the greatest element of Υ(D) is ensured by the main theorem of the paper [7, Theorem 9.2.1], and it turns out to be the actual positive part of D (cf. Remark 4.1.2). In this sense, the above definition has a meaning even for a non-big pseudo-effective arithmetic R-Cartier divisor. Of course, in this case, the uniqueness of the decomposition is not guaranteed. We would like to apply the above theorem to the problem of Zariski decompo- sitions on arithmetic varieties (cf. Conventions and terminology 1). The first one is an estimation of the asymptotic multiplicity. Theorem 0.2. We assume that D is big. If D = P + N is a Zariski decomposition of D, then µR,ξ(D) = multξ(N) for all ξ ∈ XQ. In the paper [8], we considered a decomposition D = P + N such that µR,Γ(D) = multΓ(N) for any horizontal prime divisor Γ on X. The above theorem means that a Zariski decomposition of a big arithmetic R-Cartier divisor of C0-type yields the decomposition treated in [8, Section 5] (cf. Remark 4.1.2). Thus, as a corollary, we have the following variant of the impossibility of Zariski decompositions. The condition µR,Γ(D) = multΓ(N) is rather technical, so that this form seems to be more acceptable than [8, Theorem 5.6]. ARITHMETIC LINEAR SERIES WITH BASE CONDITIONS Theorem 0.3. We suppose that d ≥ 2 and X = Pd addition, we assume that D is given by Z(:= Proj(Z[T0, T1, . . . , Td])). 3 In (cid:16)H0, log(a0 + a1z12 + · · · + adzd2)(cid:17) , where H0 := {T0 = 0}, zi := Ti/T0 (i = 1, . . . , d) and a0, a1, . . . , ad ∈ R>0. If D is big and not nef (i.e. a0 + · · · + ad > 1 and ai < 1 for some i), then, for any birational morphism f : Y → Pd Z of generically smooth, normal and projective arithmetic varieties (cf. Conventions and terminology 1), there is no Zariski decomposition of f ∗(D) on Y. The third application is the unique existence of Zariski decompositions of big arithmetic R-Cartier divisors on arithmetic surfaces. The new point is the unique- ness of the Zariski decomposition in the sense of this paper, which gives a char- acterization of the Zariski decompositions. Theorem 0.4. We assume that d = 1 and X is regular. If D is big, then there exists a unique Zariski decomposition of D. Namely, the positive part of the Zariski decomposition of D is the greatest element of Υ(D). Finally I would like to give my hearty thanks to the referee for pointing out inadequate parts of this paper. Conventions and terminology. Here we fix several conventions and the termi- nology of this paper. Let K be either Q or R. For details of 2 and 3, see [7]. 1. An arithmetic variety means a quasi-projective and flat integral scheme over Z. An arithmetic variety is said to be generically smooth if the generic fiber over Z is smooth over Q. 2. Let X be a generically smooth and normal arithmetic variety. Let Div(X) be the group of Cartier divisors on X and let Div(X)K := Div(X) ⊗Z K, whose element is called a K-Cartier divisor on X. A pair D = (D, g) is called an arithmetic K-Cartier divisor of C∞-type (resp. of C0-type) if the following conditions are satisfied: (a) D is a K-Cartier divisor on X, that is, D = Pr Div(X) and a1, . . . , ar ∈ K. i=1 aiDi for some D1, . . . , Dr ∈ (b) g : X(C) → R ∪ {±∞} is a locally integrable function and g ◦ F∞ = g (a.e.), where F∞ : X(C) → X(C) is the complex conjugation map. (c) For any point x ∈ X(C), there are an open neighborhood Ux of x and a C∞-function (resp. continuous function) ux on Ux such that g = ux + rXi=1 (−ai) log fi2 (a.e.) on Ux, where fi is a local equation of Di over Ux for each i. If ux can be taken as a continuous plurisubharmonic function over Ux for all x ∈ X(C), then the pair D is called an arithmetic K-Cartier divisor of (C0 ∩ PSH)-type. Let C be either C∞ or C0 or C0 ∩ PSH. The set of all arithmetic K-Cartier divisors of C-type is denoted bydDivC(X)K. Moreover, the set n(D, g) ∈ dDivC(X)K D ∈ Div(X)o 4 ATSUSHI MORIWAKI is denoted by dDivC(X). An element of dDivC(X) is called an arithmetic Cartier divisor of C-type. Note thatdDivC∞(X)K anddDivC0(X)K are vector spaces over K and dDivC0∩PSH(X)K forms a cone in dDivC0(X)K. For D = (D, g), E = (E, h) ∈ dDivC0(X)K, we define relations D = E and D ≥ E as follows: D = E D ≥ E def ⇐⇒ D = E, g = h (a.e.), def ⇐⇒ D ≥ E, g ≥ h (a.e.). 3. Let X be a generically smooth, normal and projective arithmetic variety. Let D be an arithmetic R-Cartier divisor of C0-type on X. The effectivity, bigness, pseudo-effectivity and nefness of D are defined as follows: • D is effective def • D is big def ⇐⇒ D ≥ (0, 0). ⇐⇒ cvol(D) > 0. • D is pseudo-effective divisor A of C0-type. def ⇐⇒ D + A is big for any big arithmetic R-Cartier • D = (D, g) is nef def ⇐⇒ (a) ddeg(D(cid:12)(cid:12)(cid:12)C) ≥ 0 for all reduced and irreducible 1-dimensional closed subschemes C of X. (b) D is of (C0 ∩ PSH)-type. The interrelations of the various types of positivity as above can be summarized as follows: effective ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ 3 pseudo-effective 8✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ big nef 1. Generalizations of Boucksom-Chen's results to R-Cartier divisors In this section, we will give generalizations of Boucksom-Chen's results [1] to arithmetic R-Cartier divisors. All results in this section can be proved in the similar way as the paper [1]. 1.1. Geometric case. First of all, let us review the geometric case. The contents of this subsection are generalizations of the works due to Okounkov [10], Lazarsfeld- Mustat¸a [5] and Kaveh-Khovanskii [3], [4] to R-Cartier divisors. Let T be a d-dimensional, geometrically irreducible, normal and projective variety over a field F. Let F be an algebraic closure of F and let TF := T ×Spec(F) Spec(F). Let P ∈ T(F) be a regular point and let zP = (z1, . . . , zd) be a local system of parameters of OTF,P. Then bOTF,P = F[[z1, . . . , zd]], & . + 0 ARITHMETIC LINEAR SERIES WITH BASE CONDITIONS 5 where bOTF,P is the completion of OTF,P with respect to the maximal ideal of OTF,P. Thus, for f ∈ OTF,P, we can put f = X(a1,...,ad)∈Zd ≥0 c(a1,...,ad)za1 1 · · · zad d , where c(a1,...,ad) ∈ F. Note that Zd has the lexicographic order <lex, that is, (a1, . . . , ad) <lex (b1, · · · , bd) def ⇐⇒ a1 = b1, . . . , ai−1 = bi−1, ai < bi for some i. We define ordzP( f ) to be ordzP( f ) := min<lexn(a1, . . . , ad) c(a1,...,ad) , 0o ∞ if f , 0, otherwise, which gives rise to a rank d valuation, that is, the following properties are satisfied: (i) ordzP( f g) = ordzP( f ) + ordzP(g) for f, g ∈ OTK,P. (ii) ordzP( f + g) ≥ min{ordzP( f ), ordzP(g)} for f, g ∈ OTF,P. By the property (i), ordzP : OTF,P \ {0} → Zd has the natural extension ordzP : Rat(TF)× → Zd given by ordzP( f /g) = ordzP( f ) − ordzP(g). As ordzP(u) = (0, . . . , 0) for all u ∈ O × ordzP induces Rat(TF)×/O × → Zd. The composition of homomorphisms TF,P, TF,P Div(TF) αP−→ Rat×(TF)/O × TF,P ordzP−→ Zd is denoted by multzP, where Div(TF) is the group of Cartier divisors on TF and αP : Div(TF) → Rat(TF)×/O × TF,P is the natural homomorphism. Moreover, the homomorphism multzP : Div(TF) → Zd yields the natural extension Div(TF) ⊗Z R → Rd over R. By abuse of notation, the above extension is also denoted by multzP. For D ∈ Div(T)R := Div(T) ⊗Z R, let H0(T, D) be a vector space over F given by H0(T, D) := {φ ∈ Rat(T)× (φ) + D ≥ 0} ∪ {0}. In the same way as [5, Lemma 1.3] or [1, (1.1)]], we can see dimF V = #nmultzP((φ) + DF) φ ∈ V ⊗F F \ {0}o for a subspace V of H0(T, D). We set R(D) := Lm≥0 H0(T, mD), which forms a graded algebra in the natural way. Let V• be a graded subalgebra of R(D). We say V• contains an ample series if Vm , {0} for m ≫ 1 and there is an ample Q-Cartier divisor A with the following properties: (• A ≤ D. • There is a positive integer m0 such that H0(T, mm0A) ⊆ Vmm0 for all m ≥ 1. 6 ATSUSHI MORIWAKI We set Γ(V•) = [Vm,{0},m≥0n(multzP((φ) + mDF), m) ∈ Rd ≥0 × Z≥0 φ ∈ Vm ⊗F F \ {0}o . Let v : Rd+1 → Rd and h : Rd+1 → R be the projections given by v(x1, . . . , xd, xd+1) = (x1, . . . , xd) and h(x1, . . . , xd, xd+1) = xd+1. Let Θ be an effective R-Cartier divisor such that D + Θ ∈ Div(T). We assume that V• contains an ample linear series. Then, in the same way as [5, Lemma 2.2], we can see the following: (1) If we set θ = multzP(ΘF) and Γ′(V•) = {γ + h(γ)(θ, 0) γ ∈ Γ(V•)}, then Γ′(V•) ⊆ Zd+1 ≥0 and Γ′(V•) generates Zd+1 as a Z-module. (2) [m>0 1 m Γ(V•)m is bounded in Rd, where Γ(V•)m := v(cid:16)Γ(V•) ∩ (Rd ≥0 × {m})(cid:17) = v(cid:0){γ ∈ Γ(V•) h(γ) = m}(cid:1) . Let ∆(V•) be the closed convex hull of Sm>0 In the case where Vm = H0(T, mD) for all m ≥ 0, ∆(V•) is denoted by ∆(D). In the same arguments as [5, Proposition 2.1] by using the above properties (1) and (2), we can see that Γ(V•)m. 1 m vol(∆(V•)) = lim m→∞ dimF Vm md . 1.2. Arithmetic case. Let X be a (d + 1)-dimensional, generically smooth, normal and projective arithmetic variety (cf. Conventions and terminology 1). Let X → Spec(OK) be the Stein factorization of X → Spec(Z), so that the generic fiber of X → Spec(OK) is geometrically irreducible. Let D = (D, g) be an arithmetic R-Cartier divisor of C0-type (cf. Conventions and terminology 2). We define H0(X, D) to be wherebΓ×(X, D) := nφ ∈ Rat(X)× D + c(φ) ≥ (0, 0)o (for details, see Section 2). Let V• be a graded subalgebra ofLm≥0 H0(XK, mDK) over K. Using X and D, we can define the natural filtration F⋆ D H0(X, D) :=bΓ×(X, D) ∪ {0}, of V• given by Ft D Vm = hVm ∩ H0(X, mD + (0, −2t))iK for t ∈ R. Note that we use (0, −2t) to ensure consistency with the notation in [1, Definition 2.3] . It is easy to see that Ft D Vm+m′. Thus, if we set Vm′ ⊆ Ft+t′ D Vm · Ft′ D Vm, Vt m = Ftm D m forms a subalgebra of V•. For each m, we define emin(D; Vm) then Vt and emax(D; Vm) to be • :=Lm≥0 Vt  emin(D; Vm) := infnt ∈ R Ft emax(D; Vm) := supnt ∈ R Ft Vm , Vmo , Vm , {0}o . D D Then, in the similar way as [1, Section 2], we can see the following: (1) −∞ < emin(D; Vm) for each m. ARITHMETIC LINEAR SERIES WITH BASE CONDITIONS 7 (2) There is a constant C such that emax(D; Vm) ≤ Cm. (3) We set emax(D; V•) = lim supm→∞ emax(D; Vm)/m. If V• contains an ample series, then Vt • also contains an ample series for t < emax(D; V•). We assume that V• contains an ample series. As in [1, Definition 1.8], we define G(D;V•) : ∆(V•) → R ∪ {−∞} andb∆(D; V•) to be  G(D;V•)(x) := sup(cid:8)t ∈ R x ∈ ∆(Vt •)(cid:9) , b∆(D; V•) :=n(x, t) ∈ ∆(V•) × R 0 ≤ t ≤ G(D;V•)o . Note that G(D;V•) : ∆(V•) → R∪{−∞} is an upper semicontinuous concave function (cf. [1, SubSection 1.3]). In the case where Vm = H0(XK, mDK) for all m ≥ 0, G(D;V•) and b∆(D; V•) are denoted by GD and b∆(D) respectively. Moreover, we define cvol(D; V•) to be # log(cid:16)Vm ∩ H0(X, mD)(cid:17) md+1/(d + 1)! . cvol(D; V•) := lim sup m→∞ Then, in the similar way as [1, Theorem 2.8], we have the following theorem: Theorem 1.2.1. cvol(D; V•) = (d + 1)![K : Q] vol(b∆(D; V•)), that is, cvol(D; V•) = (d + 1)![K : Q]ZΘ(D;V•) where Θ(D; V•) is the closure ofnx ∈ ∆(V•) G(D;V•)(x) > 0o. G(D;V•)(x)dx, 2. Asymptotic multiplicity Let X be a (d + 1)-dimensional, generically smooth, normal and projective arith- metic variety (cf. Conventions and terminology 1). Let K be either Q or R. Let Rat(X)× K := Rat(X)× ⊗Z K, and let ( )K : Rat(X)× be the natural extensions of the homomorphisms K → Div(X)K and b( )K : Rat(X)× K → dDivC∞(X)K given by φ 7→ (φ) and φ 7→ c(φ) respectively. Let D be an arithmetic R-Cartier divisor of C0-type (cf. Conventions and terminology 2). We definebΓ×(X, D) and bΓ× K(X, D) to be Rat(X)× → Div(X) and Rat(X)× → dDivC∞(X) bΓ×(X, D) :=nφ ∈ Rat(X)× D + c(φ) ≥ (0, 0)o , K(X, D) :=nφ ∈ Rat(X)× K D + c(φ)K ≥ (0, 0)o . bΓ× Q(X, D) = S∞ K(X, D) :=bΓ× H0(X, D) :=bΓ×(X, D) ∪ {0} n=1bΓ×(X, nD)1/n. Moreover, Note that bΓ× defined by H0(X, D) and H0 K(X, D) ∪ {0}. K(X, D) are and H0 8 ATSUSHI MORIWAKI For ξ ∈ X, we define the K-asymptotic multiplicity of D at ξ to be infnmultξ(D + (φ)K) φ ∈ Γ× K(X, D)o ∞ K(X, D) , ∅, if Γ× otherwise. µK,ξ(D) := First let us observe the elementary properties of the K-asymptotic multiplicity (cf. [7, Proposition 6.5.2 and Proposition 6.5.3]). Proposition 2.1. Let D and E be arithmetic R-Cartier divisors of C0-type. Then we have the following: (1) µK,ξ(D + E) ≤ µK,ξ(D) + µK,ξ(E). (2) If D ≤ E, then µK,ξ(E) ≤ µK,ξ(D) + multξ(E − D). (3) µK,ξ(D + c(φ)K) = µK,ξ(D) for φ ∈ Rat(X)× (4) µK,ξ(aD) = aµK,ξ(D) for a ∈ K>0. (5) 0 ≤ µR,ξ(D) ≤ µQ,ξ(D). (6) If D is nef and big, then µK,ξ(D) = 0. K. K(X, E) = ∅, so that we K(X, D) , ∅ K(X, D + E) for K(X, D + E) = ∅, then eitherbΓ× K(X, D) = ∅ orbΓ× Proof. (1) IfbΓ× may assume thatbΓ× andbΓ× all φ ∈bΓ× (2) is derived from (1). (3) The assertion follows from the following: K(X, D + E) , ∅. Thus we may also assume thatbΓ× K(X, E) , ∅. Therefore, the assertion follows because φψ ∈bΓ× K(X, D) and ψ ∈bΓ× ψ ∈bΓ× K(X, D) ⇐⇒ ψφ−1 ∈bΓ× K(X, D) if and only if ψa ∈bΓ× (4) Note that ψ ∈bΓ× multξ(aD + (ψa)K) = a multξ(D + (ψ)K), K(X, D + c(φ)K). K(X, aD), and that K(X, E). which implies (4). (5) is obvious. (6) follows from (5) and [7, Proposition 6.5.3] (cid:3) Remark 2.2. Theorem 2.5 says that if D is big, then µR,ξ(D) = µQ,ξ(D). In general, it does not hold. Let P1 Z = Proj(Z[T0, T1]) be the projective line over Z. We set D := {T0 = 0} and z := T1/T0. Let a0, a1 ∈ R>0 such that a0 + a1 = 1 and a0 < Q. Let D be an arithmetic divisor of C∞-type on P1 Z given by Then it is easy to see that D :=(cid:16)D, log(a0 + a1z2)(cid:17) . Γ× Q(X, D) = ∅ and Γ× R(X, D) ∋ za1 (for details, see [8, (6) in Theorem 2.3]). Thus µQ,ξ(D) = ∞ for all ξ ∈ P1(Q) and µR,ξ(D) ≤ a0 ≤ a1 = 0 if ξ = (0 : 1), if ξ = (1 : 0), if ξ ∈ P1(Q) \ {(0 : 1), (1 : 0)}. ARITHMETIC LINEAR SERIES WITH BASE CONDITIONS 9 Next we consider the following lemmas, which will be important for the proof of Theorem 2.5. Lemma 2.3. We assume that D is big. Let a = inf{x ∈ R cvol(D + (0, x)) > 0} and let f : (a, ∞) → R be the function given by f (x) = µK,ξ(D + (0, x)). Then f is a monotone decreasing continuous function. Proof. For x, y ∈ (a, ∞) with x ≤ y, we have D + (0, x) ≤ D + (0, y), and hence f (x) ≥ f (y) by (2) in Proposition 2.1. Here let us see that f is a K-convex function on (−∞, a) ∩ K, that is, f (tx + (1 − t)y) ≤ t f (x) + (1 − t) f (y) holds for all x, y ∈ (a, ∞) ∩ K and t ∈ [0, 1] ∩ K. Indeed, by using (1) and (4) in Proposition 2.1, f (tx + (1 − t)y) = µK,ξ(cid:16)t(D + (0, x)) + (1 − t)(D + (0, y))(cid:17) ≤ µK,ξ(cid:16)t(D + (0, x))(cid:17) + µK,ξ(cid:16)(1 − t)(D + (0, y))(cid:17) = t f (x) + (1 − t) f (y). The continuity of an R-convex function on an open interval of R is well-known (cf. [11, Theorem 5.5.1]), so that we assume K = Q. We check the continuity of f at x ∈ (a, ∞). By [6, Proposition 1.3.1], there are positive numbers ǫ and L such that (x − ǫ, x + ǫ) ⊆ (a, ∞) and 0 ≤ f (v) − f (u) ≤ L(u − v) for all u, v ∈ (x − ǫ, x + ǫ) ∩ Q with u ≥ v. Let y, z ∈ (x − ǫ, x + ǫ) with y ≥ z. Here we choose arbitrary rational numbers u, v such that x − ǫ < v ≤ z ≤ y ≤ u < x + ǫ. Then 0 ≤ f (z) − f (y) ≤ f (v) − f (u) ≤ L(u − v), and hence 0 ≤ f (z) − f (y) ≤ L(y − z) holds. Therefore, the lemma follows. (cid:3) Lemma 2.4. We assume that D is effective. Let φ1, . . . , φr ∈ Rat(X)× Q and a1, . . . , ar ∈ R with a1d(φ1) + · · · + ard(φr) + D ≥ 0. Then there is a subspace W of Qr over Q with the (1) dimQ W = dimQha1, . . . , ariQ, where ha1, . . . , ariQ is the subspace of R generated following properties: by a1, . . . , ar over Q. (2) (a1, . . . , ar) ∈ WR := W ⊗Q R. (3) For positive numbers ǫ and ǫ′, there is a positive number δ such that for any (c1, . . . , cr) ∈ WR with c1d(φ1) + · · · + crd(φr) + D + (0, ǫ′) ≥ 0 k(c1, . . . , cr) − (a1/(1 + ǫ), . . . , ar/(1 + ǫ))k ≤ δ, where k · k is the standard L2-norm on Rr. Proof. First we assume that a1, . . . , ar are linearly independent over Q, that is, Replacing φ1, . . . , φr, a1, . . . , ar by φn r , a1/n, . . . , ar/n respectively for some n ∈ Z>0, we may assume that φ1, . . . , φr ∈ Rat(X)×. The set of all prime divisors on 1, . . . , φn dimQha1, . . . , ariQ = r. 10 ATSUSHI MORIWAKI X is denoted by I. Moreover, for x ∈ X(C), the set {B ∈ I x ∈ B(C)} is denoted by Ix. For B ∈ Ix, let B(C)x = B1 + · · · + BnBx be the irreducible decomposition of B(C) at x on X(C), that is, B1, . . . , BnBx are irreducible components of B(C) on X(C) passing through x. Note that ordB(φ) = ordB j(φ) for φ ∈ Rat(X)× and j = 1, . . . , nBx. We set D = (D, g) and D = PB∈I dBB. By our assumption, dB ≥ 0 for all B ∈ I and g ≥ 0. For ccc = (c1, . . . , cr) ∈ Rr, we define φccc, Dccc, gccc and Dccc to be  Note that φccc := φc1 1 · · · φcr r , i=1(−ci) log φi2 + g, Dccc := (φccc)R + D =Pr gccc :=Pr Dccc := (Dccc, gccc) = d(φccc)R + D. i=1 ci(φi) + D, Dccc =XB∈I (ordB(φccc) + dB)B, where ordB : Rat(X)× Rat(X)× → Z given by ψ 7→ ordB(ψ). Around x ∈ X(C), we set R → R is the natural extension of the homomorphism φi = ρiYB∈Ix nBxYj=1 f ordB(φi) B j , where fB j is a local equation of B j around x and ρi ∈ O × X(C),x. Then φccc = ρc1 1 · · · ρcr r YB∈Ix nBxYj=1 f ordB(φccc) B j . g =XB∈Ix nBxXj=1 −dB log fB j2 + ux nBxXj=1 (− ordB(φccc)) log fB j2 + (−ci) log ρi2 −(dB + ordB(φccc)) log fB j2 + (−ci) log ρi2 + ux. rXi=1 rXi=1 Thus, if we set around x, then gccc = g +XB∈Ix nBxXj=1 =XB∈Ix (2.4.1) We put S =Sr Claim 2.4.2. i=1 Supp((φi)) and aaa = (a1, . . . , ar). (i) [(φaaa/(1+ǫ))R + D ≥ 0. In particular, gaaa/(1+ǫ) ≥ 0. (ii) ordB(φaaa/(1+ǫ)) + dB > 0 for all B ∈ I with B ⊆ S. In particular, we can find δ0 > 0 such that (φccc)R + D ≥ 0 for any ccc ∈ Rr with kccc − aaa/(1 + ǫ)k ≤ δ0. Proof. (i) The assertion follows from the following: [(φaaa/(1+ǫ))R + D = 1 1 + ǫ (d(φaaa)R + D) +(cid:18)1 − 1 1 + ǫ(cid:19) D. ARITHMETIC LINEAR SERIES WITH BASE CONDITIONS 11 (ii) It is sufficient to show that ordB(φaaa) + (1 + ǫ)dB > 0 for all B ∈ I with B ⊆ S. First of all, note that ordB(φaaa) + dB ≥ 0. If either ordB(φaaa) > 0 or dB > 0, then the assertion is obvious, so that we assume ordB(φaaa) ≤ 0 and dB = 0. Then ordB(φaaa) = a1 ordB(φ1) + · · · + ar ordB(φr) = 0, which yields ordB(φ1) = · · · = ordB(φr) = 0 by the linear independency of a1, . . . , ar over Q. This is a contradiction because B ⊆ S. (cid:3) Claim 2.4.3. For each x ∈ X(C), there are δx > 0 and an open neighborhood Ux of x such that gccc + ǫ′ ≥ 0 on Ux for any ccc ∈ Rr with kccc − aaa/(1 + ǫ)k ≤ δx. Proof. First we assume x ∈ S(C). For B ∈ I with B ⊆ S, we set We choose δ′ > 0 such that B = ordB(φaaa/(1+ǫ)) + dB > 0. d′ 1 2 B ≤ dB + ordB(φccc) ≤ d′ 3 2 d′ B for all ccc ∈ Rr and B ∈ I with kccc − aaa/(1 + ǫ)k ≤ δ′ and B ⊆ S. Note that there are an open neighborhood Ux and a constant M such that rXi=1 (−ci) log ρi2 + ux ≥ M over Ux for all ccc ∈ Rr with kccc−aaa/(1+ǫ)k ≤ δ′. Moreover, shrinking Ux if necessarily, we may assume that fB j ≤ 1 for all B j with either B ⊆ S or dB > 0 because the set {B ∈ I B ⊆ S or dB > 0} is finite and fB j(x) = 0. Thus, by using (2.4.1), −(dB + ordB(φccc)) log fB j2 + M nBxXj=1 gccc ≥XB∈Ix =XB∈Ix −(dB + ordB(φccc)) log fB j2 + XB∈Ix nBxXj=1 nBxXj=1 ≥ XB∈Ix,B⊆S B log fB j2 + M. d′ − 1 2 B⊆S B*S,dB>0 nBxXj=1 −dB log fB j2 + M Note that limy→x(−d′ smaller neighborhood Ux. B) log fB j(y)2 = ∞. Thus, the assertion follows if we take a Next we consider the case where x < S(C). Then, by (i) in Claim 2.4.2, (−ci) log ρi2 + ǫ′ = gaaa/(1+ǫ) + ǫ′ + rXi=1 (ai/(1 + ǫ) − ci) log ρi2 gccc + ǫ′ = g + ≥ ǫ′ + rXi=1 rXi=1 (ai/(1 + ǫ) − ci) log ρi2. Thus the assertion follows. (cid:3) 12 ATSUSHI MORIWAKI As X(C) =Sx∈X(C) Ux and X(C) is compact, there are x1, . . . , xl ∈ X(C) such that X(C) = Ux1 ∪ · · · ∪ Uxl. Therefore, if we set δ1 = {δx1, . . . , δxl}, then gccc + ǫ′ ≥ 0 for all ccc ∈ Rr with kccc − aaa/(1 + ǫ)k ≤ δ1, and hence, if we put δ = min{δ0, δ1}, then, by (ii) in Claim 2.4.2, for all ccc ∈ Rr with kccc − aaa/(1 + ǫ)k ≤ δ. Dccc + (0, ǫ′) ≥ 0 Finally we consider the lemma without the linear independency of a1, . . . , ar over Q. We set s = dimQha1, . . . , ariQ. If s = 0 (i.e. a1 = · · · = ar = 0), then we can take W as {(0, . . . , 0)}, so that we may assume s ≥ 1. Renumbering a1, . . . , ar, we j=1 eija j (j = 1, . . . , s). Note that eij ∈ Q, and hence may further assume that a1, . . . , as are linearly independent. We set ai =Ps (i = 1, . . . , r) and ψj = Qr Q. Let α : Rs → Rr be the homomorphism given by ψj ∈ Rat(X)× ei j i=1 φ i α(x1, . . . , xs) = (α1(x1, . . . , xs), . . . , αr(x1, . . . , xs)) and αi(x1, . . . , xs) = sXj=1 eijx j. As the rank of (eij) is s, α is injective. In addition, (a1, . . . , ar) = α(a1, . . . , as) and x1d(ψ1) + · · · + xsd(ψs) = α1(x1, . . . , xs)d(φ1) + · · · + αr(x1, . . . , xs)d(φr). for (x1, . . . , xs) ∈ Rs. Therefore, if we put W = α(Qs) ⊆ Qr, then the assertion follows from the previous observation. (cid:3) The following theorem is the main result of this section. Theorem 2.5. If D is big, then µQ,ξ(D) = µR,ξ(D). Proof. First of all, by (3) in Proposition 2.1, we may assume that D is effective. Moreover, by (5) in Proposition 2.1, µR,ξ(D) ≤ µQ,ξ(D), so that we consider the converse inequality. For this purpose, it is sufficient to show that µQ,ξ(D) ≤ multξ(D + (ψ)R) R(X, D). We choose φ1, . . . , φr ∈ Rat(X)× and aaa = (a1, . . . , ar) ∈ Rr such that a1, . . . , ar are linearly independent over Q and ψ = φa1 r . Let ǫ be a 1 positive number. Applying Lemma 2.4 to the case ǫ = ǫ′, we can find a sequence {cccn}∞ Q(X, D + (0, ǫ)) for all n. Thus we have for all ψ ∈ bΓ× n=1 in Qr such that limn→∞ cccn = aaa/(1 + ǫ) and φcccn ∈bΓ× · · · φar µQ,ξ(D + (0, ǫ)) ≤ multξ(Dcccn) for all n, and hence, µQ,ξ(D + (0, ǫ)) ≤ multξ(Daaa/(1+ǫ)). Therefore, by Lemma 2.3, µQ,ξ(D) = lim ǫ↓0 µQ,ξ(D + (0, ǫ)) ≤ lim ǫ↓0 multξ(Daaa/(1+ǫ)) = multξ(D + (ψ)R). (cid:3) ARITHMETIC LINEAR SERIES WITH BASE CONDITIONS 13 3. Proof of Theorem 0.1 In this section, we give the proof of Theorem 0.1. Since it is sufficient to show the following: cvol(D; µ1ξ1, . . . , µlξl) ≤ cvol(D; µiξi), (3.1) If D is big and µ > µR,ξ(D) for ξ ∈ XQ, then cvol(D; µξ) < cvol(D). Let B be the Zariski closure of {ξ} in X. Let us begin with the following claim: Claim 3.2. We may assume that B is a prime divisor. Proof. Let νB : XB → X be the blowing-up along B. As XQ is regular, we can find a unique prime divisor EB on XB such that νB(EB) = B. Let ν′ : X′ → XB be a projective birational morphism such that X′ is normal and ν yields a resolution of singularities on the generic fiber. Let B′ be the strict transform of EB and let ν : X′ → X be the composition of ν′ If ξ′ is the generic point of B′, then it is easy to see that ordξ( f ) = ordξ′(ν∗( f )) for all f ∈ Rat(X)×, and hence multξ(L) = multξ′(ν∗(L)) for all L ∈ Div(X)R. Moreover, the natural homomorphism ν∗ : Rat(X) → Rat(X′) yields a bijection H0(X, nD) → H0(X′, ν∗(nD)). Therefore, we have : X′ → XB and νB : XB → X. # H0(X, nD; nµξ) = # H0(X′, nν∗(D); nµξ′), which implies cvol(D; µξ) = cvol(ν∗(D); µξ′), as required. From now on, we assume that B is a prime divisor. Let µ0 = µR,ξ(D) and let (cid:3) X → Spec(OK) be the Stein factorization of X → Spec(Z). Claim 3.3. There is a positive number ǫ0 such that D − (µ0 + ǫ)B is big on XK for all 0 ≤ ǫ ≤ ǫ0. Proof. Let A be a big arithmetic Cartier divisor of C0-type on X such that A ≥ (0, 0) and B * Supp(A). We can choose a sufficiently small positive number a such that cvol(D − aA) > 0. In particular, there is φ ∈ Rat(X)× By (2) in Proposition 2.1, Q such that D − aA + c(φ)Q ≥ 0. µ0 = µR,ξ(D) ≤ µR,ξ(D − aA) + multξ(aA) = µR,ξ(D − aA) ≤ multξ(D − aA + (φ)Q). Thus D − aA + (φ)Q ≥ µ0B, and hence D − µ0B ≥ aA − (φ)Q. In particular, D − µ0B is big on XK, and hence the assertion follows. (cid:3) It is sufficient to show (3.1) in the case where µ = µ0 + ǫ with 0 < ǫ < ǫ0. We set Vm = H0(XK, mDK − mµBK). Note that cvol(D; V•) = cvol(D; µξ). By Claim 3.3, V• We choose P ∈ X(K) and a local system of parameters zP = (z1, . . . , zd) at P such contains an ample series. that P is a regular point of BK and z1 is a local equation of B at P. Claim 3.4. If we set multzP(L) = (x1, . . . , xd) for L ∈ Div(X)R, then x1 = multξ(L). 14 ATSUSHI MORIWAKI Proof. First we assume that L ∈ Div(X) and L is effective. Let f be a local equation 1 in K[[z1, . . . , zd]], where a ∈ Z≥0, fi ∈ K[[z2, . . . , zd]] and fa , 0. Then a = multξ( f ). Moreover, the lowest term with respect to the lexicographical order must appear in faza 1. Thus x1 = a, as required. i=a fizi of L around P. We set f =P∞ In general, we set L =Pl i=1 aiLi, where a1, . . . , al ∈ R and Li's are effective divisors. Moreover, if we set multzP(Li) = (xi1, . . . , xid), then xi1 = multξ(Li) by the previous observation. On the other hand, as multzP(L) = lXi=1 ai multzP(Li), the first entry of multzP(L) is equal to lXi=1 aixi1 = lXi=1 ai multξ(Li) = multξ(L), as desired. (cid:3) By Claim 3.4, ∆(V•) ⊆ ∆(DK) ∩ {x1 ≥ µ} and GD ≥ G(D;V•). As in Theorem 1.2.1, let Θ(D) and Θ(D; V•) be the closures of nx ∈ ∆(DK) GD(x) > 0o and nx ∈ ∆(V•) G(D;V•)(x) > 0o respectively. Clearly Θ(D; V•) ⊆ Θ(D) ∩ {x1 ≥ µ}. Claim 3.5. Θ(D) ∩ {x1 < µ} , ∅. Proof. As D is big, µ0 = µQ,ξ(D) by Theorem 2.5, and hence, by Lemma 2.3, there is a positive number t0 such that µ0 ≤ µQ,ξ(D + (0, −2t0)) < µ0 + ǫ/2. Q(X, D + (0, −2t0)) such that µQ,ξ(D + (0, −2t0)) ≤ multξ(D + (φ)Q) ≤ µ0 + ǫ/2 < µ. Thus we can find φ ∈bΓ× Moreover, as φ ∈bΓ× and hence x ∈ Θ(D). Q(X, D + (0, −2t0)), if we set x = multzP(D + (φ)Q), then GD(x) > 0, (cid:3) Here we fix notation. Let T be a topological space and S a subset of T. The set of all interior points of S is denoted by S◦. Claim 3.6. Let C be a closed convex set in Rd. For a ∈ R, we set C(a) = {x ∈ C p(x) < a}, where p : Rd → R is the projection to the first factor, that is, p(x1, . . . , xd) = x1. If C◦ , ∅ and C(a) , ∅ for some a ∈ R, then C(a)◦ , ∅. Proof. Let us choose x ∈ C(a). We assume that C(a)◦ = ∅. Then, as C(a) is a convex set, by [11, Corollary 2.3.2], there is a hyperplane H such that C(a) ⊆ H. Moreover, as C◦ , ∅, there is y ∈ C \ H. Note that p(y) ≥ a, so that 0 < (a − p(x))/(p(y) − p(x)) ≤ 1. Here we choose t ∈ R with 0 < t < (a − p(x))/(p(y) − p(x)). Then (1 − t)x + ty ∈ C \ H and p((1 − t)x + ty) < a. This is a contradiction. (cid:3) ARITHMETIC LINEAR SERIES WITH BASE CONDITIONS 15 As cvol(D) > 0, Θ(D)◦ , ∅ by Theorem 1.2.1. Therefore, by Claim 3.5 and Claim 3.6, Θ(D)◦ ∩ {x1 < µ} , ∅, and hence (Θ(D) \ Θ(D; V•))◦ , ∅ because (Θ(D) ∩ {x1 < µ})◦ ⊆ (Θ(D) \ Θ(D; V•))◦. Moreover, note that Θ(D)◦ =nx ∈ ∆(D) GD(x) > 0o◦ ⊆nx ∈ ∆(D) GD(x) > 0o GD(x)dx and vol(b∆(D; V•)) =ZΘ(D;V•) G(D;V•)(x)dx. GD(x)dx +ZΘ(D;V•) GD(x)dx > vol(b∆(D; V•)). (cf. [11, Corollary 2.3.9]). Further, Therefore, vol(b∆(D)) =ZΘ(D) vol(b∆(D)) =ZΘ(D)\Θ(D;V•) Thus (3.1) follows from Theorem 1.2.1. 4. Applications of Theorem 0.1 In this section, let us study several applications of Theorem 0.1. 4.1. Impossibility of Zariski decomposition. Let X be a (d + 1)-dimensional, generically smooth, normal and projective arithmetic variety (cf. Conventions and terminology 1). A Zariski decomposition of D is a decomposition D = P + N such that (1) P is a nef arithmetic R-Cartier divisor of C0-type, (2) N is an effective arithmetic R-Cartier divisor of C0-type, and that (3) cvol(P) = cvol(D). The arithmetic R-Cartier divisor P (resp. N) is called a positive part (resp. a negative part) of D. As a nef arithmetic R-Cartier divisor of C0-type is pseudo-effective (cf. [7, Proposition 6.2.1, Proposition 6.2.2 and Proposition 6.3.2]), if D has a Zariski decomposition, then D is pseudo-effective. Let Υ(D) be the set of all nef arithmetic R(X, D) , ∅ implies Υ(D) , ∅. In the paper [7], we proved that if d = 1, X is regular and Υ(D) , ∅, then a Zariski decomposition of D exists. Moreover, by [9, Theorem 3.5.3], if D is pseudo-effective and D is numerically trivial on XQ, then a Zariski decomposition of D exists in the above sense. R-Cartier divisors M of C0-type with M ≤ D. Note thatbΓ× Theorem 4.1.1. Let D be a big arithmetic R-Cartier divisor of C0-type on X. If there is a Zariski decomposition D = P + N of D, then µR,ξ(D) = multξ(N) for all ξ ∈ XQ. Proof. Since D ≥ P and µR,ξ(P) = 0 (cf. Proposition 2.1), we have µR,ξ(D) ≤ µR,ξ(P) + multξ(N) = multξ(N). Here we assume that µR,ξ(D) < multξ(N). If we set µ = multξ(N), then cvol(D; µξ) < cvol(D) 16 ATSUSHI MORIWAKI by Theorem 0.1. On the other hand, if φ ∈bΓ×(X, nP), then multξ(nD + (φ)) = multξ(nP + (φ) + nN) ≥ multξ(nN) = nµ. Therefore, H0(X, nP) ⊆ H0(X, nD; nµξ). Thus cvol(P) ≤ cvol(D; µξ) < cvol(D). This is a contradiction. Remark 4.1.2. In the papers [8] and [7], we gave the different kinds of definitions of Zariski decompositions. Let us recall their definitions. Let D be an arithmetic R-Cartier divisor of C0-type. (cid:3) (a) A decomposition D = P + N is called a Zariski decomposition in the sense of R(X, D) , ∅, P is a nef arithmetic R-Cartier divisor of C0-type, N is an [8] if Γ× effective arithmetic R-Cartier divisor of C0-type, and µR,Γ(D) = multΓ(N) for any horizontal prime divisor Γ on X. (b) In the case where d = 1 and X is regular, we say a decomposition D = P+N is a Zariski decomposition in the sense of [7] if Υ(D) , ∅ and P yields the greatest element of Υ(D). In this case, cvol(D) = cvol(P) by [7, Theorem 9.3.4], so that it is a Zariski decomposition in the sense of this paper. The interrelations of these definitions can be described as follows: (i) If D is big, then a Zariski decomposition in the sense of this paper is a Zariski decomposition in the sense of [8] (cf. Theorem 4.1.1). (ii) We assume that d = 1 and X is regular. A Zariski decomposition in the sense of this paper gives rise to a Zariski decomposition in the sense of [7] without the bigness of D, that is, a Zariski decomposition in the sense of this paper implies Υ(D) , ∅, so that, by [7, Theorem 9.2.1], we can find the greatest element of Υ(D), which turns out to be the positive part of the Zariski decomposition in the sense [7]. Moreover, if D is big, then a Zariski decomposition in the sense of this paper coincides with a Zariski decomposition in the sense of [7] (cf. Theorem 4.2.1). By the above remark, we have the following corollary. Corollary 4.1.3. We suppose that d ≥ 2 and X = Pd Z(= Proj(Z[T0, T1, . . . , Td])). We set Hi := {Ti = 0} and zi := Ti/T0 for i = 0, . . . , d. For a sequence aaa = (a0, a1, . . . , ad) of positive numbers, we define an H0-Green function gaaa of (C∞ ∩ PSH)-type on Pd(C) to be gaaa := log(a0 + a1z12 + · · · + adzd2). We assume that D is given by (H0, gaaa). If D is big and not nef (i.e., a0 + · · · + ad > 1 and ai < 1 for some i), then, for any birational morphism f : Y → Pd Z of generically smooth, normal and projective arithmetic varieties, there is no Zariski decomposition of f ∗(D) on Y. Remark 4.1.4. For a non-big pseudo-effective arithmetic R-Cartier divisor, to find a Zariski decomposition in the sense of this paper is a non-trivial problem. This is closely related to the fundamental question raised in the paper [9]. Here let us consider an example. We use the same notation as in [9]. We assume that Daaa ARITHMETIC LINEAR SERIES WITH BASE CONDITIONS 17 Z)× is pseudo-effective and not big. Then, by [9, Corollary 3.6.4, Proposition 3.6.7, Example 3.6.8], we can find φ ∈ Rat(Pn R such that Daaa + c(φ)R ≥ 0. Thus, if we set P = [(φ−1)R and N = Daaa +c(φ)R, then the decomposition Daaa = P + N yields a Zariski 1 and P = −a1d(z1). Moreover, −a1d(z1) is the greatest element of Υ(Daaa) (cf. [8, decomposition. Note that P is not necessarily an arithmetic Q-Cartier divisor of C0-type. For example, in the case where d = 1, a0 + a1 = 1 and a1 < Q, φ is given by za1 Section 4]). 4.2. Characterization of Zariski decompositions on arithmetic surfaces. Let X be a regular projective arithmetic surface and let π : X → Spec(OK) be the Stein factorization of X → Spec(Z). In this subsection, we study the following charac- terizations of the Zariski decomposition of big arithmetic R-Cartier divisors on X. Theorem 4.2.1. Let D be a big arithmetic R-Cartier divisor of C0-type on X and let D = P + N be a Zariski decomposition of D, where P is a positive part of D. Then P gives the greatest element of Υ(D) =nM M is a nef arithmetic R-Cartier divisor of C0-type on X with M ≤ Do . Proof. Let us begin with the following claim: Claim 4.2.1.1. Let P and Q be nef arithmetic R-Cartier divisors of C0-type. We assume the following: (1) There are an effective vertical R-Cartier divisor E and an F∞-invariant non- negative continuous function u on X(C) such that Q = P + (E, u). (2) cvol(P) = cvol(Q) > 0. Then P = Q. Proof. First of all, note that the degree of P on XQ is positive and u ∈ hQPSH(X(C)) ∩ C0(X(C))iR 2 (for details, see [9, SubSection 1.2]). Moreover, by [7, Proposition 6.4.2], cvol(P) = ddeg(P ), and hence ddeg(P ) +ddeg(Q · (E, u)) +ddeg(P · (E, u)), ) = ddeg(Q ). As 2 2 2 2 2 ) and cvol(Q) = ddeg(Q ddeg(Q ) = ddeg(P ddeg(Q · (E, u)) ≥ 0, ddeg(P · (E, u)) ≥ 0,  we have which yields On the other hand, by virtue of [9, Proposition 2.1.1], ddeg(Q · (E, u)) = ddeg(P · (E, u)) = 0, ddeg(P · (E, u)) = ddeg((E, u)2) = 0. 2ZX(C) 1 ddeg(P · (E, u)) = ddeg(P · (E, 0)) + 2ZX(C) ddeg((E, u)2) = ddeg((E, 0)2) + 1 c1(P)u, uddc([u]).  18 ATSUSHI MORIWAKI Therefore, by using Zariski's lemma and [9, Proposition 1.2.4, (3)], ZX(C) c1(P)u =ZX(C) uddc([u]) = 0 and ddeg(P · (E, 0)) = ddeg((E, 0)2) = 0. By the equality condition of [9, Proposition 1.2.4, (3)], u is locally constant, and hence 0 =ZX(C) uc1(P) = (uX1 + · · · + uX[K:Q]) degQ(PQ) [K : Q] , where X1, . . . , X[K:Q] are connected components of X(C). Thus u = 0 on X(C). Moreover, by the equality condition of Zariski's lemma, there are p1, . . . , pk ∈ Spec(OK) \ {0} and a1, . . . , ak ∈ R≥0 such that E = a1π−1(p1) + · · · + akπ−1(pk). Thus 0 = ddeg(P · (E, 0)) = (a1 log #(OK/p1) + · · · + ak log(OK/pk)) and hence a1 = · · · = ak = 0, that is, E = 0, as desired. degQ(PQ) [K : Q] , (cid:3) Let us go back to the proof of Theorem 4.2.1. As D is big, Υ(D) , ∅, and hence we can find the greatest element PZar of Υ(D) by [7, Theorem 9.2.1]. Then P ≤ PZar and cvol(P) = cvol(PZar) = cvol(D). Thus, if we set NZar = D − PZar, then, by Theorem 4.1.1, µR,ξ(D) = multξ(N) = multξ(NZar) for all ξ ∈ XQ. Therefore, there are an effective vertical R-Cartier divisor E and an F∞-invariant non-negative continuous function u on X(C) such that PZar = P + (E, u). Note that P is big. Thus, P = PZar by Claim 4.2.1.1. (cid:3) As a corollary of Theorem 4.2.1, we have the following stronger version of Claim 4.2.1.1. Corollary 4.2.2. Let P and Q be nef arithmetic R-Cartier divisors of C0-type. If P ≤ Q and 0 < cvol(P) = cvol(Q), then P = Q. Proof. If we set N = Q − P, then Q = P + N is a Zariski decomposition of Q. Therefore, by Theorem 4.2.1, P = Q. (cid:3) Theorem 0.1 still holds for the regular projective arithmetic surface X without the assumption ξ1, . . . , ξl ∈ XQ. Namely we have the following theorem: Theorem 4.2.3. Let D be a big arithmetic R-Cartier divisor of C0-type on X and let Proof. As in the proof of Theorem 0.1, it is sufficient to see the following: ξ1, . . . , ξl ∈ X. If µi > µR,ξi(D) for some i, then cvol(D; µ1ξ1, . . . , µlξl) < cvol(D). If D is big and µ > µR,ξ(D) for ξ ∈ X, then cvol(D; µξ) < cvol(D). By Theorem 0.1, we may assume that the characteristic of the residue field at ξ is positive. Let B be the Zariski closure of {ξ} in X. In the same way as Claim 3.2, we (4.2.3.1) If D−µ(B, 0) is not big, then the assertion is obvious, so that we may further assume may also assume that B is a prime divisor. Note that cvol(D; µξ) = cvol(D − µ(B, 0)). that D − µ(B, 0) is big. We suppose cvol(D − µ(B, 0)) = cvol(D). Let D = P + N and ARITHMETIC LINEAR SERIES WITH BASE CONDITIONS 19 ′ ′ +N be the Zariski decompositions of D and D−µ(B, 0) respectively. D−µ(B, 0) = P ′ As P ≤ D − µ(B, 0) ≤ D, we have P′ ≤ P. Moreover, Thus, by Corollary 4.2.2, we obtain P ′ cvol(P ) = cvol(D − µ(B, 0)) = cvol(D) = cvol(P). = P, which implies ′ ′ N + µ(B, 0) = N. In particular, multB(N) ≥ µ. On the other hand, by [7, Claim 9.3.5.1], µR,ξ(D) = multB(N), and hence µR,ξ(D) ≥ µ. This is a contradiction. (cid:3) References [1] S. Boucksom and H. Chen, Okounkov Bodies of filtered linear series, preprint. [2] J. Koll´ar, Lectures on resolution of singularities, Annals of Mathematics Studies 166, Princeton Univ. Press, (2007). [3] K. Kaveh and A. Khovanskii, Convex bodies and algebraic equations on affine varieties, preprint. [4] K. Kaveh and A. Khovanskii, Newton convex bodies, semigroups of integral points, graded algebras and intersection theory, preprint. [5] R. Lazarsfeld and M. Mustat¸a, Convex bodies associated to linear series, Ann. Scient. ´Ec. Norm. Sup. 42 (2009), 783 -- 835. [6] A. Moriwaki, Estimation of arithmetic linear series, Kyoto J. Math. 50 (2010), 685-725. [7] A. Moriwaki, Zariski decompositions on arithmetic surfaces, preprint (arXiv:0911.2951v3 [math.AG]). [8] A. Moriwaki, Big arithmetic divisors on Pn [9] A. Moriwaki, Toward Dirichlet's unit Z, Kyoto J. Math. 51 (2011), 503 -- 534. theorem on arithmetic varieties, preprint (arXiv:1010.1599v2 [math.AG]). [10] A. Okounkov, Brunn-Minkowski inequality for multiplicities, Invent. Math. 125 (1996), 405 -- 411. [11] R. Webster, Convexity, Oxford University Press, (1994). Department of Mathematics, Faculty of Science, Kyoto University, Kyoto, 606-8502, Japan E-mail address: [email protected]
1312.7066
3
1312
2015-12-18T09:10:35
On the automorphism of a smooth Schubert variety
[ "math.AG" ]
Let $G$ be a simple algebraic group of adjoint type over the field $\mathbb{C}$ of complex numbers. Let $B$ be a Borel subgroup of $G$ containing a maximal torus $T$ of $G$. Let $w$ be an element of the Weyl group $W$ and let $X(w)$ be the Schubert variety in $G/B$ corresponding to $w$. Let $\alpha_{0}$ denote the highest root of $G$ with respect to $T$ and $B.$ Let $P$ be the stabiliser of $X(w)$ in $G.$ In this paper, we prove that if $G$ is simply laced and $X(w)$ is smooth, then the connected component of the automorphism group of $X(w)$ containing the identity automorphism equals $P$ if and only if $w^{-1}(\alpha_{0})$ is a negative root ( see Theorem 4.2 ). We prove a partial result in the non simply laced case ( see Theorem 6.6 ).
math.AG
math
ON THE AUTOMORPHISM GROUP OF A SMOOTH SCHUBERT VARIETY S. SENTHAMARAI KANNAN Abstract. Let G be a simple algebraic group of adjoint type over the field C of complex numbers. Let B be a Borel subgroup of G containing a maximal torus T of G. Let w be an element of the Weyl group W and let X(w) be the Schubert variety in G/B corresponding to w. Let α0 denote the highest root of G with respect to T and B. Let P be the stabiliser of X(w) in G. In this paper, we prove that if G is simply laced and X(w) is smooth, then the connected component of the automorphism group of X(w) containing the identity automorphism equals P if and only if w−1(α0) is a negative root ( see Theorem 4.2 ). We prove a partial result in the non simply laced case ( see Theorem 6.6 ). Keywords: Automorphism group, Schubert varieties, Tangent bundle. 1. Introduction Recall that if X is a smooth projective variety over C, the connected component of the group of all automorphisms of X containing the identity automorphism is an algebraic group ( see [15, Theorem 3.7, p.17] and [7, p.268] ), which deals also with the case when X may be singular or it may be defined over any field. Further, the Lie algebra of this automorphism group is isomorphic to the space of all tangent vector fields on X, that is the space H 0(X, TX ) of all global sections of the tangent bundle TX of X ( see [15, Lemma 3.4, p.13] ). The aim of this paper is to study the connected component, containing the identity auto- morphism of the group of all automorphisms of a smooth Schubert variety X(w) in the flag variety G/B associated to a simple algebraic group G ( see notation in section 2 ). We give a fairly precise description of the connected component, containing the identity automorphism of the group of all automorphisms of a smooth Schubert variety X(w) in the flag variety G/B associated to a simple simply laced algebraic group G of adjoint type over C, that is of type A, D, E; in particular, we give a description of a smooth Schubert variety for which this automorphism group is a subgroup of G. More precisely, we prove the following results: Let w ∈ W be such that X(w) is smooth. Let Aw denote the connected component of the group of all automorphisms of X(w) containing the identity automorphism. Let T be a maximal torus of G contained in B and B+ ⊃ T be the Borel subgroup of G opposite to B determined by T. Let α0 denote the highest root of G with respect to T and B+. For the left action of G on G/B, let Pw denote the stabiliser of X(w) in G. Let φw : Pw −→ Aw be the homomorphism of algebraic groups induced by the action of Pw on X(w). For precise notation, see section 2 and section 3. Theorem 1.1. (see Theorem 4.2 ) 1 Assume that G is a simple, simply laced algebraic group of adjoint type over C. Then we have (1) φw : Pw −→ Aw is surjective. (2) φw : Pw −→ Aw is an isomorphism if and only if w−1(α0) is a negative root. Theorem 1.2. ( see Theorem 6.6 ) Assume that G is a simple algebraic group of adjoint type which is not simply laced. Then, φw : Pw −→ Aw is injective if and only if w−1(α0) is a negative root. Recall that the vanishing results of the cohomology groups of the restrictions of the homo- geneous vector bundles to Schubert varieties have been an important area of research in the theory of algebraic groups ( see [1], [3], [4], [6], [11], [14], [16] and [18] ). In this paper, we prove the following vanishing results of the cohomology groups: By abuse of notation, we denote by TG/B also the restriction of the tangent bundle of G/B to X(w). Theorem 1.3. ( see Theorem 4.1 ) Assume that G is a simple, simply laced algebraic group of adjoint type over C. Then we have (1) H i(X(w), TG/B) = 0 for every i ≥ 1. (2) H 0(X(w), TG/B) is the adjoint representation g of G if and only if w−1(α0) is a negative root. Theorem 1.4. ( see Theorem 6.5 ) Assume that G is a simple algebraic group of adjoint type over C which is not simply laced. Then we have (1) H i(X(w), TG/B) = 0 for every i ≥ 1. (2) The adjoint representation g of G is a B-submodule of H 0(X(w), TG/B) if and only if w−1(α0) is a negative root. The results in this paper play an important role in the study of the connected component, containing the identity automorphism of the group of all automorphisms of a Bott-Samelson- Demazure-Hansen variety ( see [5] ). The organisation of the paper is as follows: Section 2 consists of preliminaries from [4], [9], [10] and [11]. The strategy of the proof of the results in this paper uses the induction on the dimension of Schubert varieties, using their Bott-Samelson-Demazure-Hansen desingularisations and the structure of the indecomposable representations of a Borel subgroup of SL(2, C) (see [2. p.130, Corollary 9.1]). In section 3 and section 4, we assume that G is simply laced. We first describe the B- module of the global sections of the homogeneous vector bundles associated to all those B-submodules V of the adjoint representation of G which contain the adjoint representation of B ( see Lemma 3.3 ). Using this result, we prove that all higher cohomology groups 2 H i(X(w), L(V )) vanish for any such B-module V, where L(V ) is the homogeneous vector bundle associated to V ( see Lemma 3.4 ). In section 4, we prove the main results in the simply laced case applying Lemma 3.4 to a long exact sequence of cohomology groups associated to a certain short exact sequence of B-modules. In section 5, we prove some vanishing results in the non simply laced case. These results are similar to those in section 3. In section 6, we prove the main results in the non simply laced case. The proofs of these results are similar to those of the main results in the simply laced case. 2. Notation and Preliminaries In this section, we set up some notation and preliminaries. We refer to [9] and [10] for preliminaries in Lie algebras and algebraic groups. Let G be a simple algebraic group of adjoint type over C and T a maximal torus of G. Let W = NG(T )/T denote the Weyl group of G with respect to T and we denote the set of roots of G with respect to T by R. Let B+ be a Borel subgroup of G containing T . Let B be the Borel subgroup of G opposite to B+ determined by T . That is, B = n0B+n−1 0 , where n0 is a representative in NG(T ) of the longest element w0 of W . Let R+ ⊂ R be the set of positive roots of G with respect to the Borel subgroup B+. Note that the set of roots of B is equal to the set R− := −R+ of negative roots. Let S = {α1, . . . , αn} denote the set of simple roots in R+. For β ∈ R+, we also use the notation β > 0. The reflection in W corresponding to a root α ( respectively, a simple root αi ) is denoted by sα ( respectively, si ). Let g be the Lie algebra of G. Let h ⊂ g be the Lie algebra of T and b ⊂ g be the Lie algebra of B. Let X(T ) denote the group of all characters of T . We have X(T ) ⊗ R = HomR(hR, R), the dual of the real form of h. The positive definite W -invariant form on HomR(hR, R) induced by the Killing form of g is denoted by ( , ). We use the notation h , i to denote hµ, αi = 2(µ,α) (α,α) , for every µ ∈ X(T ) ⊗ R and α ∈ R. Let {xα, hβ : α ∈ R, β ∈ S} denote the Chevalley basis of g corresponding to R. For α ∈ R, we denote by gα the one dimensional root subspace of g spanned by xα. Let sl2,α denote the 3 dimensional Lie subalgebra of g generated by xα and x−α. Let ≤ denote the partial order on X(T ) given by µ ≤ λ if λ − µ is a non negative integral linear combination of simple roots. We denote by X(T )+ the set of dominant characters of T with respect to B+. Let ρ denote the half sum of all positive roots of G with respect to T and B+. For any simple root α, we denote the fundamental weight corresponding to α by ωα. For w ∈ W, let l(w) denote the length of w. We define the dot action of W on X(T ) ⊗ R by w ◦ λ = w(λ + ρ) − ρ, where w ∈ W and λ ∈ X(T ) ⊗ R. We set R+(w) := {β ∈ R+ : w(β) ∈ −R+}. For w ∈ W , let X(w) := BwB/B denote the Schubert variety in G/B corresponding to w. For a simple root α, we denote by Pα the minimal parabolic subgroup of G generated by B and nα, where nα is a representative of sα in NG(T ). Let Lα denote the Levi subgroup of Pα containing T . Note that Lα is the product of T and the homomorphic image Gα of SL(2, C) via a homomorphism ψ : SL(2, C) −→ Lα ( see [11, II, 1.3] ). We denote the intersection of Lα and B by Bα. We note that the morphism Lα/Bα ֒→ Pα/B induced by the inclusion Lα ֒→ Pα is an isomorphism. Therefore, to compute the cohomology modules 3 H i(Pα/B, L(V )) ( 0 ≤ i ≤ 1 ) for any B-module V, we treat V as a Bα-module and we compute H i(Lα/Bα, L(V )). Now, we recall some preliminaries on Bott-Samelson-Demazure-Hansen varieties which we call BSDH-varieties and some application of Leray spectral sequence to compute the cohomology of line bundles on Schubert varieties. We refer to [4] and [11] for preliminaries related to BSDH varieties. We refer to [19] for spectral sequences. We recall that the BSDH-variety corresponding to a reduced expression i = (i1, i2, · · · , ir) of w = si1si2 · · · sir is defined by Z(w, i) = Pαi1 × · · · × Pαir × Pαi2 B × · · · × B , where the action of B × · · ·× B on Pαi1 (p1 · b1, b−1 1 Definition 2.2.1] ). · p2 · b2, . . . , b−1 r−1 · pr · br), pj ∈ Pαij × Pαi2 × · · ·× Pαir is given by (p1, . . . , pr)(b1, . . . , br) = , bj ∈ B and i = (i1, i2, . . . , ir) (see [4, p.64, We note that for each reduced expression i of w, Z(w, i) is a smooth projective variety. We denote both the natural birational surjective morphism from Z(w, i) to X(w) and the composition map Z(w, i) −→ X(w) ֒→ G/B by φw. We denote the restriction of the homoge- neous vector bundle L(V ) to X(w) by L(w, V ) and its pull back to Z(w, i) via the birational morphism φw by L(w, i, V ). Let fr : Z(w, i) −→ Z(wsir , i′) denote the map induced by the projection Pαi1 × Pαi2 × · · · × Pαir −→ Pαi1 × Pαi2 × · · · × Pαir−1 , where i′ = (i1, i2, . . . , ir−1). We note that fr is a Pαir /B ≃ P1-fibration. Then for j ≥ 0, we have the following isomorphism of B-linearized sheaves ( see [11, II, p.366, section 14.1, equation (4)] and [8, Theorem 12.11, p.290] ): Rjfr ∗L(w, i, V ) = L(wsir, i′, H j(Pαir /B, L(sir, V ))). (Iso) We use the following ascending 1-step construction as a basic tool in computing cohomology modules. Let γ be a simple root such that l(w) = l(sγw) + 1. Let Z(w, i) be a BSDH-variety corresponding to a reduced expression w = si1si2 · · · sir , where αi1 = γ. Then we have an induced morphism g : Z(w, i) −→ Pγ/B ≃ P1, with fibres Z(sγw, i′′), where i′′ = (i2, i3, . . . , ir). We note that Pγ acts on both Z(w, i) and on Pγ/B by the left and that the map g : Z(w, i) −→ Pγ/B is Pγ-equivariant. By an application of the Leray spectral sequence together with the fact that the base is P1, for every B-module V, we obtain the following exact sequence of Pγ-modules: 0 → H 1(Pγ/B, Rj−1g∗L(w, i, V )) → H j(Z(w, i), L(w, i, V )) → H 0(Pγ/B, Rjg∗L(w, i, V )) → 0. 4 Since for any B-module V, the vector bundle L(w, i, V ) on Z(w, i) is the pull back of the homogeneous vector bundle L(w, V ) from X(w), we conclude that the cohomology modules H j(Z(w, i), L(w, i, V )) ∼= H j(X(w), L(w, V )) (see [4, Theorem 3.3.4 (b)] ), and are independent of the choice of the reduced expression i. Hence we denote H j(Z(w, i), L(w, i, V )) by H j(w, V ). For a character λ of B, we denote the one dimensional B-module corresponding to λ by Cλ. Further, we denote the cohomology modules H j(Z(w, i), L(w, i, Cλ)) by H j(w, λ). Recall the following isomorphism of B-linearized sheaves ( see [11, II, p.366, section 14.1, equation (4) ] and [8, Theorem 12.11, p.290] ): Rjg∗L(w, i, V ) = L(sγ, H j(Z(sγw, i′′), L(sγw, i′′, V ))) (j ≥ 0). We use the simple notation for the cohomology modules and apply the above isomorphism in the above short exact sequence to obtain the following short exact sequence of Pγ-modules: 0 → H 1(sγ, H j−1(sγw, V )) → H j(w, V ) → H 0(sγ, H j(sγw, V )) → 0. In this paper, we call the above short exact sequence of B-modules SES when ever we use it. Let α be a simple root and λ ∈ X(T ) be such that hλ, αi ≥ 0. Here, we recall the following result due to Demazure ( see [6, Page 1] ) on a short exact sequence of B-modules: Lemma 2.1. Let K denote the kernel of the surjective evaluation map H 0(sα, λ) −→ Cλ. Then we have (1) The sequence 0 −→ K −→ H 0(sα, λ) −→ Cλ −→ 0 (2) The sequence 0 −→ Csα(λ) −→ K −→ H 0(sα, λ − α) −→ 0 of B-modules is exact, of B-modules is exact. whenever hλ, αi ≥ 1. (3) If hλ, αi = 1 then H 0(sα, λ − α) = 0. (4) If hλ, αi = 0, then K = 0 and hence (2) does not hold. We use the following lemma to compute cohomology groups. The following lemma is due to Demazure ( see [6, page 1] ). He used this lemma to prove Borel-Weil-Bott's theorem. Lemma 2.2. Let w = τ sα, l(w) = l(τ ) + 1. Then we have (1) If hλ, αi ≥ 0 then H j(w, λ) = H j(τ, H 0(sα, λ)) for all j ≥ 0. (2) If hλ, αi ≥ 0, then H j(w, λ) = H j+1(w, sα ◦ λ) for all j ≥ 0. (3) If hλ, αi ≤ −2, then H j+1(w, λ) = H j(w, sα ◦ λ) for all j ≥ 0. (4) If hλ, αi = −1, then H j(w, λ) vanishes for every j ≥ 0. Proof. Choose a reduced expression for w = si1si2 · · · sir with αir = α. Hence τ = si1si2 · · · sir−1 is a reduced expression for τ . Let i = (i1, i2, · · · , ir) and i′ = (i1, i2, · · · , ir−1). Therefore, we have the morphism fr : Z(w, i) −→ Z(τ, i′) defined as above. Now, the proof of the lemma follows from the fact that the functor H 0(w, −) is the composite of H 0(τ, −) and H 0(sα, −) ( together with the well-known properties of the cohomology groups of line bundles over P1 ). For instance, see [8, p.252, III, Ex 8.1] and [11, p.218, Proposition 5.2(b)]. (cid:3) 5 The following consequence of Lemma 2.2 will be used to compute the cohomology modules in this paper. Let π : G −→ G be the simply connected covering of G. Let Lα ( respectively, Bα ) be the inverse image of Lα ( respectively, of Bα ) in G. Note that Lα/ Bα is isomorphic to Lα/Bα. We make use of this isomorphism to use the same notation for the vector bundle on Lα/Bα associated to a Bα-module. Lemma 2.3. Let V be an irreducible Lα-module and λ be a character of Bα. Then we have (1) If hλ, αi ≥ 0, then the Lα-module H 0(Lα/Bα, V ⊗ Cλ) is isomorphic to the tensor product of V and H 0(Lα/Bα, Cλ). Further, we have H j(Lα/Bα, V ⊗ Cλ) = 0 for every j ≥ 1. (2) If hλ, αi ≤ −2, then we have H 0(Lα/Bα, V ⊗ Cλ) = 0. Further, the Lα-module H 1(Lα/Bα, V ⊗ Cλ) is isomorphic to the tensor product of V and H 0(Lα/Bα, Csα◦λ). (3) If hλ, αi = −1, then H j(Lα/Bα, V ⊗ Cλ) = 0 for every j ≥ 0. Proof. By [11, p.53, I, Proposition 4.8] and [11, p.77, I, Proposition 5.12], for all j ≥ 0, we have the following isomorphism as Lα-modules: H j(Lα/Bα, V ⊗ Cλ) ≃ V ⊗ H j(Lα/Bα, Cλ). Now, the proof of the lemma follows from Lemma 2.2 by taking w = sα and the fact that Lα/Bα ≃ Pα/B. (cid:3) We now state the following Lemma on indecomposable Bα ( respectively, Bα ) modules which will be used in computing the cohomology modules ( see [2, p.130, Corollary 9.1] ). Lemma 2.4. (1) Any finite dimensional indecomposable Bα-module V is isomorphic to V ′ ⊗ Cλ for some irreducible representation V ′ of Lα, and some character λ of Bα. (2) Any finite dimensional indecomposable Bα-module V is isomorphic to V ′ ⊗ Cλ for some irreducible representation V ′ of Lα, and some character λ of Bα. Proof. The proof of part 1 follows from [2, p.130, Corollary 9.1]. The proof of part 2 follows from the fact that every Bα-module can be viewed as a Bα- (cid:3) module via the natural homomorphism. The following lemmas on the evaluation map are known. For the sake of completeness, we provide a sketch of a proof. For a B-module V, the evaluation map ev : H 0(w, V ) −→ V is given by ev(s) = s(idB). Lemma 2.5. Let V be a finite dimensional rational G-module, and let w ∈ W . Then we have (1) The evaluation map ev : H 0(w, V ) −→ V is an isomorphism of B-modules. (2) H i(w, V ) = 0 for every i ≥ 1. Proof. Since V is a G-module, the homogeneous vector bundle L(V ) on G/B is trivial. Therefore, its restriction L(w, V ) to X(w) is also trivial. Hence, the assertions of both parts of the lemma follow immediately. (cid:3) 6 Lemma 2.6. Let w ∈ W, and V be a B-submodule of a G-module V ′. Then the evaluation map ev : H 0(w, V ) −→ V is injective. Proof. Since V is a B-submodule of V ′, H 0(w, V ′). Hence, we have the following commutative diagram of B-modules: it follows that H 0(w, V ) is a B-submodule of H 0(w, V ) / H 0(w, V ′) V / V ′ Here, both the horizontal maps are the canonical inclusions and both the vertical maps are evaluation maps. Since the B-module V ′ is the restriction of a G-module, by using Lemma 2.5, we see that the evaluation map ev′ : H 0(w, V ′) −→ V ′ on the right hand side is an isomorphism of B-modules. Since the first horizontal map H 0(w, V ) −→ H 0(w, V ′) is injective, the composition H 0(w, V ) −→ V ′ is also injective. Thus, we conclude that the vertical map ev : H 0(w, V ) −→ V on the left hand side is injective. This completes the proof of the lemma. (cid:3) 3. Cohomology modules in the simply laced case In this section, we prove some preliminary results for the simply laced case. We use these results in section 4 to prove the main results. Through out this section, we assume that G is simply laced. We use the following lemma whose proof is well known. Lemma 3.1. Let α ∈ S and β be a root different from both α and −α. Then we have hβ, αi ∈ {−1, 0, 1}. From now on, for any T -module and a character µ of T , we denote the set of all vectors v in V such that tv = µ(t)v for every t ∈ T by Vµ. That is, Here tv denotes the action of t on v. Vµ := {v ∈ V : tv = µ(t)v f or all t ∈ T }. Notation. We set up a notation for some indecomposable Bγ-summand of g. Let γ be a simple root. We recall that sl2,γ is the simple Lie algebra corresponding to γ. We first note that sl2,γ is an indecomposable Bγ-summand of g. We also note that for β ∈ R such that hβ, γi = 1, the T -submodule gβ ⊕ gβ−γ is an indecomposable Bγ-summand of g. We denote it by gβ,β−γ. Indeed, this is an irreducible Lγ-submodule of g. In this paper, we denote by the one dimensional complex vector space generated by a non zero vector v by Cv. The following lemma gives a description of indecomposable Bγ-summands of g. Lemma 3.2. Every indecomposable Bγ-summand V of g is one of the following: (1) V = Ch for some non zero vector h ∈ h such that γ(h) = 0. (2) V = gβ for some root β such that hβ, γi = 0. (3) V = gβ,β−γ for some root β such that hβ, γi = 1. 7   /   / (4) V = sl2,γ, the three dimensional irreducible Lγ-module with highest weight γ. Proof. Let V be an indecomposable Bγ-summand of g. Let λ be a maximal weight of V . Then the direct sum ⊕r∈Z≥0Vλ−rγ is a Bγ-summand of V . Hence, we have V = ⊕r∈Z≥0Vλ−rγ. Note that V is a Lγ-module. Therefore, using Lemma 3.1, we see that the dimension of V must be at most two unless V = sl2,γ. Further, if the dimension of V is one, either V = Ch for some non zero vector h ∈ h such that γ(h) = 0, or V = gβ for some root β such that hβ, γi = 0. Also, if the dimension of V is two, then we must have V = gβ,β−γ for some root β such that hβ, γi = 1. This completes the proof of the lemma. (cid:3) First, we note that a decomposition of a Bγ-module into a direct sum of indecomposable submodules is not necessarily unique. The following lemma is crucial in the proof of the main results in section 3 and section 4. We make a remark on some notation used in the statement of Lemma 3.3 and its proof. Notation. Let h(γ) ∈ h be the fundamental dominant coweight corresponding to γ. That is, h(γ) satisfies γ(h(γ)) = 1 and ν(h(γ)) = 0 for every simple root ν different from γ. Then the smallest Bγ-submodule of g containing h(γ) is the two dimensional vector subspace Ch(γ) ⊕ g−γ. Note that this is an indecomposable Bγ-submodule of g. In the context of Lemma 2.4, we have Ch(γ) ⊕ g−γ = V1 ⊗ C−ωγ , where V1 is the standard two dimensional irreducible Lγ-module. We denote this indecomposable Bγ-submodule by g0,−γ. Further, g0,−γ is an indecomposable Bγ-direct summand of any Bγ-submodule V ′ of g such that g0,−γ ⊂ V ′ and gγ T V ′ = 0. We use this two dimensional Bγ-submodule in type (2) of the statement of Lemma 3.3 and its proof. Lemma 3.3. Let w ∈ W and γ ∈ S. Let V be a B-submodule of g containing b. Then there is a decomposition of the Bγ-module H 0(w, V ) such that every indecomposable Bγ-summand V ′ of H 0(w, V ) is one of the following: (1) V ′ = Ch for some non zero vector h ∈ h such that γ(h) = 0. (2) V ′ = g0,−γ. (3) V ′ = gβ for some root β such that hβ, γi ∈ {−1, 0}. (4) V ′ = gβ,β−γ for some root β such that hβ, γi = 1. (5) V ′ = sl2,γ, the restriction of the three dimensional irreducible Lγ-module with highest weight γ. Proof. First note that in view of Lemma 2.6, H 0(w, V ) is a B-submodule of V . Let V ′ be an indecomposable Bγ-summand of H 0(w, V ). Note that there is a non zero vector v ∈ V ′ and a µ ∈ X(Bγ) = X(T ) such that bv = µ(b)v for every b ∈ Bγ. If µ 6= −γ, then using the arguments similar to the proof of Lemma 3.2, we see that V ′ must be one of the types (1), (3) or (4). Therefore, we may assume that g−γ is a Bγ-submodule of V ′. To complete the proof of the lemma, we need to show that either sl2,γ or g0,−γ is an indecomposable Bγ-summand of H 0(w, V ). The proof is by induction on l(w). If l(w) = 0, then we have w = 1. First note that since S is a basis of the complex vector space HomC(h, C), it follows that h(γ) ∈ h. Further, since h is a complex vector subspace of V, it follows that h(γ) ∈ V . Thus, type (2) indecomposable 8 Bγ-module g0,−γ is a submodule of V . Now, if gγ ⊂ V, then it is easy to see that V ′ = sl2,γ is an indecomposable Bγ-summand of V containing g−γ. Otherwise, the Bγ-submodule g0,−γ is a summand of V . Therefore, we may choose a simple root α such that l(w) = 1 + l(sαw). By induction on l(w), we assume that for any simple root ν such that g−ν ⊂ H 0(sαw, V ), there is a decomposition of the Bν-module H 0(sαw, V ) such that the indecomposable Bν-summand V ′ of H 0(sαw, V ) containing g−ν is either of the form V ′ = g0,−ν or of the form V ′ = sl2,ν. In particular, it follows that either g0,−γ or sl2,γ is a Bγ-summand of H 0(sαw, V ). We divide the proof into three different cases as follows. Case 1: We first assume that γ = α. By using SES, we have H 0(sα, H 0(sαw, V )) = H 0(w, V ). Note that in view of Lemma 2.4, we have g0,−γ = V1 ⊗ C−ωγ , where V1 is the standard two dimensional irreducible Lγ-module. Hence by Lemma 2.3, we have H 0(sα, g0,−γ) = H 0(sγ, g0,−γ) = 0 ( since α = γ ). Therefore, if g0,−γ is a Bγ-summand of H 0(sαw, V ), then by using Lemma 2.6, we see that g−γ can not be a subspace of H 0(sα, H 0(sαw, V )) = H 0(w, V ). This is a contradiction to the above observation. Hence, we conclude that sl2,γ is a Bγ-summand of H 0(sαw, V ). Further, by Lemma 2.5, we have H 0(Lα/Bα, sl2,γ) = H 0(Lγ/Bγ, sl2,γ) = sl2,γ ( since α = γ ). This completes the proof for the case when γ = α. Case 2 : We assume that α is different from γ and hγ, αi 6= 0. By using Lemma 3.1, we have hγ, αi = −1. Sub case 1: Assume that g0,−γ is a Bγ-summand of H 0(sαw, V ). Note that h(γ) ∈ h T g0,−γ. Since α 6= γ, we have α(h(γ)) = 0. Hence, Ch(γ) is a Bα-direct summand of H 0(sαw, V ). Hence, Ch(γ) must be a Bα-submodule of H 0(sα, H 0(sαw, V )) = H 0(w, V ). Since g−γ ⊂ H 0(w, V ), it follows that g0,−γ is a Bγ-summand of H 0(w, V ). Sub case 2: Assume that sl2,γ is a Bγ-summand of H 0(sαw, V ). Now, if gα+γ is a subspace of H 0(sαw, V ), then gα+γ,γ is an indecomposable Bα-summand of H 0(sαw, V ). Since gα+γ,γ is a Lα-module, using Lemma 2.5, we see that H 0(sα, gα+γ,γ) = gα+γ,γ ⊂ H 0(sα, H 0(sαw, V )). Thus, we see that the Bγ-span sl2,γ of gγ must be a Bγ-summand of H 0(sα, H 0(sαw, V )) = H 0(w, V ). On the other hand, if gα+γ is not a subspace of H 0(sαw, V ), then gγ is an indecomposable Bα-direct summand of H 0(sαw, V ). Since hγ, αi = −1, by Lemma 2.2 , we have H i(sα, γ) = 0 for every i ∈ Z≥0. In particular, gγ can not be a subspace of H 0(sα, H 0(sαw, V )). Hence, sl2,γ is not a Bγ-summand of H 0(sα, H 0(sαw, V )) = H 0(w, V ). We now show that g0,−γ is a Bγ-summand of H 0(sα, H 0(sαw, V )). Let S1 be the set of all simple roots β such that sl2,β is a Bβ-summand of H 0(sαw, V ). By the hypothesis, we have γ ∈ S1. Since S1 is a linearly independent subset of HomC(⊕β∈S1 Chβ, C), there is a Chβ such that γ(h1) = 1 and β(h1) = 0 for every simple root β in S1 different h1 ∈ ⊕β∈S1 9 from γ. If ν(h1) = 0 for every ν ∈ S \ S1, we are done. Otherwise, let S2 be the set of all simple roots ν ∈ S \ S1 such that ν(h1) 6= 0. Let ν ∈ S2. Then there is a β ∈ S1 such that ν(hβ) = −1. Since β ∈ S1, we have hβ ∈ H 0(sαw, V ). Hence, the Lie bracket x−ν = [x−ν, hβ] is in H 0(sαw, V ). Hence, g−ν must be a subspace of H 0(sαw, V ). Therefore, by induction applied to the simple root ν, we see that either g0,−ν or sl2,ν is a Bν-summand of H 0(sαw, V ). Since ν /∈ S1, we conclude that g0,−ν is an indecomposable Bν-summand of H 0(sαw, V ). Note that the fundamental dominant coweight h(ν) is in h T g0,−ν. Let h2 = Pν∈S2 ν(h1)h(ν). Then we have h(γ) = h1 − h2. Therefore, Ch(γ) is a Bα- summand of H 0(sαw, V ). Hence, we see that H 0(sα, Ch(γ)) = Ch(γ) is a subspace of H 0(sα, H 0(sαw, V )). Thus, we conclude that g0,−γ is a Bγ-summand of H 0(sα, H 0(sαw, V )) = H 0(w, V ). Case 3: We assume that hγ, αi = 0. If g0,−γ is a Bγ-summand of H 0(sαw, V ), using hγ, αi = 0 we see that it is also a Bα-summand of H 0(sα, H 0(sαw, V )). For the same reason, the vector bundle on Lα/Bα associated to the Bα-module g0,−γ is trivial. Thus, g0,−γ is a Bγ-summand of H 0(sα, H 0(sαw, V )) = H 0(w, V ). The proof of the case when sl2,γ is a Bγ-summand of H 0(sαw, V ) is similar. (cid:3) We now deduce the following lemma as a consequence of Lemma 3.3. Lemma 3.4. Let w ∈ W and V be a B-submodule of g containing b. Then we have H i(w, V ) = 0 for every i ≥ 1. Proof. The proof is by induction on l(w). If l(w) = 0, we are done. Otherwise, we choose a simple root γ ∈ S be such that l(sγw) = l(w) − 1. By Lemma 3.3, there is a decomposition of the Bγ-module H 0(sγw, V ) such that every indecomposable Bγ-summand V ′ of H 0(sγw, V ) must be one of the 5 types given in Lemma 3.3. In view of Lemma 2.4, any such V ′ is of the form V ′ = V ′′ ⊗ Caωγ for some irreducible Lγ-module V ′′ and an integer a ∈ {−1, 0}. Hence, using Lemma 2.3, we conclude that H i(Lγ/Bγ, V ′) is zero for every indecomposable Bγ-summand V ′ of H 0(sγw, V ) and for every i ≥ 1. Thus, we see that H i(Pγ/B, H 0(sγw, V )) = 0 (1) for all i ≥ 1. By induction on l(w), we have H i(sγw, V ) is zero for all i ≥ 1. Now, using (1) and using the short exact sequence SES of B-modules, we conclude that H i(w, V ) is zero for all i ≥ 1. This completes the proof of lemma. (cid:3) We now prove the following. Lemma 3.5. Let w ∈ W . Let V1 be a B-submodule of g containing b and V2 be a B-submodule of V1 containing b. Then we have (1) H i(w, V1/V2) = 0 for every i ≥ 1. (2) The homomorphism Πw : H 0(w, V1) −→ H 0(w, V1/V2) of B-modules induced by the natural homomorphism Π : V1 −→ V1/V2 is surjective and kernel of Πw is H 0(w, V2). (3) The restriction map r : H 0(w0, g/b) −→ H 0(w, g/b) is surjective. 10 Proof. Proof of (1): We have the short exact sequence 0 −→ V2 −→ V1 −→ V1/V2 −→ 0 of B-modules. Applying H 0(w, −) to this short exact sequence of B-modules, we obtain the following long exact sequence of B-modules: · · · H i(w, V2) −→ H i(w, V1) −→ H i(w, V1/V2) −→ H i+1(w, V2) · · · By Lemma 3.4, H i(w, V1) and H i+1(w, V2) are zero for every i ≥ 1. Thus, we conclude that H i(w, V1/V2) = 0 for every i ≥ 1. This proves (1). Proof of (2): Taking i = 0 in the above long exact sequence of B-modules and using H 1(w, V2) = 0 ( see Lemma 3.4 ), we obtain the following short exact sequence of B-modules: 0 −→ H 0(w, V2) −→ H 0(w, V1) −→ H 0(w, V1/V2) −→ 0. This proves (2). Proof of (3): We have the following commutative diagram of B-modules: H 0(w0, g) res / H 0(w, g) Πw0 Πw H 0(w0, g/b) r / H 0(w, g/b). By (2), Πw : H 0(w, g) −→ H 0(w, g/b) is surjective. By Lemma 2.5(1), the restriction map res : H 0(w0, g) −→ H 0(w, g) is an isomorphism. Thus, r : H 0(w0, g/b) −→ H 0(w, g/b) is surjective. This completes the proof of the lemma. (cid:3) The following is a useful consequence of Lemma 3.5. Corollary 3.6. Let w ∈ W and α ∈ R+. Then we have H i(w, α) = 0 for every i ≥ 1. Proof. Let V1 := ⊕µ≤αgµ denote the direct sum of the weight spaces of g of weights µ satisfying µ ≤ α and let V2 := ⊕µ<αgµ denote the direct sum of the weight spaces of g of weights µ satisfying µ < α. It is clear that V2 is a B-submodule of g containing b and V1 is a B-submodule of g containing V2. Since the B-module V1/V2 is isomorphic to Cα, we have H i(w, α) = H i(w, V1/V2) for every i ≥ 1. Hence, by Lemma 3.5(1), H i(w, α) = 0 for every i ≥ 1. This completes the proof of the corollary. (cid:3) 4. Main Results in the simply laced case In this section, we prove the main results in the simply laced case. Through out this section, we assume that G is simply laced. We first prove the following: Theorem 4.1. Let w ∈ W and let α0 denote the highest root of G with respect to T and B+. Then we have (1) H i(X(w), TG/B) = 0 for every i ≥ 1. 11   /   / (2) H 0(X(w), TG/B) is the adjoint representation g of G if and only if w−1(α0) is a negative root. Proof. Since the tangent space of G/B at the point idB is g/b, the tangent bundle TG/B is the homogeneous vector bundle L(g/b) on G/B associated to the B-module g/b. Hence, it is sufficient to prove the following: (1) H i(w, g/b) = 0 for every i ≥ 1. (2) H 0(w, g/b) is the adjoint representation g of G if and only if w−1(α0) < 0. The proof of (1) follows from Lemma 3.5(1). To prove (2), we first note that the natural projection Π : g −→ g/b of B-modules induces a homomorphism Πw : H 0(w, g) −→ H 0(w, g/b) of B-modules. Since the evaluation map ev : H 0(w, g) −→ g is an isomorphism ( see Lemma 2.5(1) ), using Lemma 3.5(2), we have the following short exact sequence of B-modules: 0 −→ H 0(w, b) −→ g −→ H 0(w, g/b) −→ 0. Taking −α0-weight spaces, we obtain the following short eact sequence of vector spaces: 0 −→ H 0(w, b)−α0 −→ g−α0 −→ H 0(w, g/b)−α0 −→ 0. Since g−α0 is one dimensional, H 0(w, b)−α0 is zero if and only if H 0(w, g/b)−α0 is non-zero. Since g is an irreducible G-module of highest weight α0, the B-stable line in g is unique and it is the one dimensional subspace g−α0. Therefore, it follows that H 0(w, b) is zero if and only if H 0(w, b)−α0 is zero. We now show that H 0(w, g/b)−α0 is non-zero if and only if w−1(−α0) ∈ R+. For this, we first note that the B-module g/b has a composition series of B-modules with each successive simple quotient is isomorphic to Cα, where α is running over positive roots. By taking global sections H 0(X(w), L(w, Cα)) and applying Corollary 3.6, we see that the B-module H 0(w, g/b) has a filtration of B-modules with each successive quotient is isomorphic to H 0(w, α) for some positive root α. Therefore, it follows that H 0(w, g/b)−α0 is non-zero if and only if H 0(w, α)−α0 is non-zero for some positive root α. On the other hand, for any positive root α, there is a v ∈ W such that v(α0) = α. Without loss of generality, we may assume that v is of minimum length among such elements. It follows from the Demazure character formula that if H 0(w, α0)µ 6= 0, then µ is in the convex hull of the set {x(α0) : x ≤ wv} ( see [4, Theorem 3.3.8, p.97, equation (3)] and [11, Proposition 14.18(b), p.379] ). Note that the convention for the signature of the weights in the Demazure character formula in this paper is the same as the one in [11]. Further, using the above arguments, we see that Cα is a B-submodule of H 0(v, α0). Therefore, by using SES , we see that H 0(w, α) is a B-submodule of H 0(wv, α0). Hence, every weight µ of H 0(w, α) satisfies µ ≥ w(α). Clearly, w(α) ≥ −α0. Hence, H 0(w, α)−α0 is non zero if and only if w(α) = −α0. Thus, we conclude that H 0(w, g/b)−α0 is non-zero if and only if w−1(−α0) ∈ R+. Summarising the above arguments, we conclude that H 0(w, b) is zero if and only if w−1(α0) is a negative root. Since Ker(Πw) = H 0(w, b) ( see Lemma 3.5(2) ), we see that Ker(Πw) is zero if and only if w−1(α0) is a negative root. This completes the proof of (2). (cid:3) 12 The following theorem describes the automorphism group of a smooth Schubert variety in the simply laced case. Theorem 4.2. Let w ∈ W be such that X(w) is smooth. Let Aw denote the connected component of the group of all automorphisms of X(w) containing the identity automorphism. For the left action of G on G/B, let Pw denote the stabiliser of X(w) in G. Then we have (1) The homomorphism φw : Pw −→ Aw induced by the action of Pw on X(w) is surjec- tive. (2) φw : Pw −→ Aw is an isomorphism if and only if w−1(α0) is a negative root. Proof. Let Tw denote the tangent bundle of X(w). By [15, Theorem 3.7, p.17], we see that Aw is an algebraic group. Further, by [15, Lemma 3.4, p.13], it follows that the Lie algebra of Aw is isomorphic to the space of all global sections H 0(X(w), Tw). Since Tw is a vector subbundle of TG/B, we have an injective homomorphism i : H 0(X(w), Tw) ֒→ H 0(X(w), TG/B) = H 0(w, g/b) of B-modules. We first note that φw induces a homomorphism ψw : pw −→ H 0(X(w), Tw) of Lie algebras, where pw is the Lie algebra of Pw. We now prove (1). By Lemma 3.5(3), the restriction map r : g = H 0(w0, g/b) −→ H 0(w, g/b) is surjective. Since Pw is a parabolic subgroup of G containing B, the Lie algebra pw is a Lie subalgebra of g = H 0(w0, g/b) containing b. Further, since X(w) is a Pw-stable subvariety for the left action of Pw on G/B, we have the following commutative diagram of B-modules: pw ψw / H 0(X(w), Tw) H 0(w0, g/b) r / H 0(w, g/b) i Now, let q = r−1(H 0(X(w), Tw)). Note that since q is a B-submodule of g containing pw, q is a parabolic subalgebra of g containing pw. We denote the restriction of r to q also by r. We now show that pw = q. Since g = H 0(w0, g/b), every element x ∈ q ⊂ g is a tangent vector field on G/B. Further, by the definition of q, the ideal sheaf of X(w) is x-stable for every x ∈ q. Therefore, q is contained in the Lie algebra pw of the stabiliser Pw of X(w) in G. Thus, we have pw = q. Clearly, r : q −→ H 0(X(w), Tw) is a homomorphism of Lie algebras. Therefore ψw : pw −→ H 0(X(w), Tw) is surjective. By counting the dimensions of the images both at the level of algebraic groups and at the level of their Lie algebras, we conclude that φw : Pw −→ Aw is surjective. We now prove (2). Assume that w−1(α0) is a negative root. Then by the proof of Theorem 4.1(2), the ho- momorphism Πw : H 0(w, g) = g −→ H 0(w, g/b) = g of B-modules induced by the natural homomorphism g −→ g/b is an isomorphism. Therefore, H 0(w, g/b) has a unique B- stable line, namely g−α0. Note that since w−1(α0) < 0, we have w 6= 1. Therefore, the action of Pw on X(w) is non trivial. Hence, the homomorphism ψw : pw −→ H 0(X(w), Tw) of B-modules 13 _   /  _   / is non-zero. Therefore, the B-stable line H 0(w, g/b)−α0 is in the image ψw(pw) ⊂ H 0(X(w), Tw) ⊂ H 0(w, g/b). Hence, we have g−α0 T ker(ψw) = 0. Thus, ψw : pw −→ H 0(X(w), Tw) is injective. Since the base field is C, it follows that the kernel of ψw is the Lie algebra of the kernel of φw. Therefore, φw : Pw −→ Aw is injective. Now, that φw is an isomorphism follows from (1). This is again because the base field is C. Conversely, if φw : Pw −→ Aw is an isomorphism, then the induced homomorphism ψw : pw −→ H 0(X(w), Tw) ⊂ H 0(w, g/b) is injective. In particular, the −α0-weight space H 0(w, g/b)−α0 is non-zero. By using arguments similar to the proof of Theorem 4.1, we conclude that w−1(α0) is a negative root. (cid:3) In the following Corollary, we describe the kernel of φw : Pw −→ Aw. Let Jw := {α ∈ S : sα ≤ w} and let T (w) := Tα∈Jw Ker(α) = {t ∈ T : α(t) = 1 f or α ∈ Jw}. Let U ( respectively, U + ) denote the unipotent radical of B ( respectively, B+ ) and let U−α ( respectively, U + α ) be the the root subgroup of U ( respectively, U + ) normalised by T corresponding to α ∈ R+. Further, let U≤w be the subgroup of U generated by {U−α : α ∈ R+ \ (Sv≤w R+(v−1))}. Then we have Corollary 4.3. The kernel Kw of φw : Pw −→ Aw is generated by T (w) and U≤w. Proof. Let R+(w−1) := {β1, β2, · · · , βl(w)}. Note that for every 1 ≤ j ≤ l(w), we have βj = Pα∈S mαα where each mα ∈ Z≥0 and mα = 0 unless sα ≤ w. Therefore, it follows that T (w) ⊂ Ker(βj) for every 1 ≤ j ≤ l(w). Hence the action of T (w) on BwB/B = Πl(w) j=1U−βj wB/B is trivial. Here, we note that product is independent of the ordering of R+(w−1) ( see [11, II, p.354] ). Since BwB/B is an open dense subset of X(w), we conclude that the action of T (w) on X(w) is trivial. If the action of U−β on X(w) is trivial for some positive root β, then it fixes vB/B for every v ≤ w. Hence, v−1(β) ∈ R+ for every v ≤ w. Conversely, if β ∈ R+ is such that v−1(β) > 0 for every v ≤ w, then we show that the action of U−β on X(w) is trivial. We show this by induction on l(w). So, fix such a β. If l(w) = 1, then we have w = sα for some simple root α. For any u1 ∈ U−α and u2 ∈ U−β, the commutator u−1 1 u2u1 ∈ ΠγU−γ, where the product is taken over all positive roots γ of the form γ = iα + jβ with i, j ∈ N (see, [10, p.203] and [10, p.209 - p.215] ). First note that α 6= β, since sα(β) ∈ R+. Further, if u1 and u2 commute with each other, then u1wB/B is fixed by u2. Therefore, we may assume that they do not commute with each other. 2 u−1 Since G is simply laced, we have hα, βi = hβ, αi = −1. Therefore, it follows that α + β = 1 u2u1 ∈ U−(α+β). It is easy to see that the subgroup U−(α+β) fixes the sα(β) ∈ R+ and u−1 point sαB/B. Hence, by the above arguments, it follows that 2 u−1 u2u1wB/B = u1u2(u−1 2 u−1 1 u2u1)wB/B = u1u2wB/B = u1wB/B. We may assume that l(w) ≥ 2 and choose a simple root α such that l(sαw) = l(w) − 1. Let v = sαw and let nα be a representative of sα in NG(T ). We have BwB/B = U−αnαUvvB/B, where Uv = Πγ∈R+(v−1)U−γ ( see [11, II, p.354] ). Note that by induction, the action of U−β on UvvB/B is trivial. Therefore, by the above arguments we may assume that the subgroups 14 U−α and U−β do not commute with each other. By the proof in the case of length one, for every u2 ∈ U−β, u1 ∈ U−α and u ∈ Uv, we see that u2 · (u1nαuvB/B) = u1nα · (u′ · uvB/B), for some u′ ∈ U−sα(β)U−β. Note that for every v1 ≤ v, both v−1 1 (sα(β)) are positive roots. Therefore, by the induction hypothesis, u′ fixes the point uvB/B. Thus, we conclude that the action of U−β on the Schubert cell BwB/B is trivial. Since, BwB/B is an open dense subset of X(w), it follows that the action of U−β on X(w) is trivial. 1 (β) and v−1 Finally, we show that U + β T Kw = {1} for every positive root β such that U + β is a subgroup of Pw. First note that if U + β T Kw is non trivial, then by conjugating by T , we could prove that U + β ⊂ Kw. Further, since G is simply laced, if β is not a simple root, then there is a simple root α such that hβ, αi = 1. Therefore, by using similar arguments as above, we conclude that 1 6= u−1 β . Hence, U + β−α T Kw is non trivial. Therefore, U + β−α ⊂ Kw. Proceeding this way, we are able to find a simple root γ such that U + γ ⊂ Kw. Hence, by using the arguments similar to the proof of Theorem 4.2, we conclude that Kw contains a representative nγ of sγ in NG(T ). Thus, we have l(sγw) = l(w) − 1. This is a contradiction to the fact that if α is a simple root such that l(sαw) = l(w) − 1, then the action of U + β−α, for some u1 ∈ U−α and some u2 ∈ U + α on X(w) is non trivial. 2 u−1 1 u2u1 ∈ U + This completes the proof . (cid:3) Again we assume that X(w) is smooth. Let Lα0 denote the line bundle on X(w) associated to α0. Consider the left action of T on G/B. Note that X(w) is stable under T . Let λ be a dominant character of T . In the following Corollary, we use the notion of semi-stable points introduced by Mumford ( see [17] ). We denote by X(w−1)ss T (Lα0) the set of all semi-stable points of X(w−1) with respect to the T -linearised line bundle Lα0. Then we have Corollary 4.4. Let Aw, Pw and φw : Pw −→ Aw be as in the hypothesis of Theorem 4.2. Then φw : Pw −→ Aw is an isomorphism if and only if the set X(w−1)ss T (Lα0) of semi-stable points is non-empty. Proof. By Theorem 4.2, we see that φw : Pw −→ Aw is an isomorphism if and only if w−1(α0) is a negative root. By [13, Lemma 2.1], we note that w−1(α0) < 0 if and only if X(w−1)ss T (Lα0) is non empty. This completes the proof of the corollary. (cid:3) The following Corollary connects the set X(w−1)ss T (Lα0) of semi-stable points with the vanishing of all cohomology modules of the vector bundle L(w, b) on X(w). Corollary 4.5. The set X(w−1)ss H i(w, b) = 0 for every i ∈ Z≥0. T (Lα0) of semi-stable points is non-empty if and only if Proof. By Lemma 3.4, H j(w, b) = 0 for every j ≥ 1. By using the proof of Theorem 4.1, we see that H 0(w, b) is zero if and only if w−1(α0) is a negative root. Therefore, by using Theorem 4.2, it follows that φw : Pw −→ Aw is an isomorphism if and only if H 0(w, b) is zero. Now, the proof of the corollary follows from Corollary 4.4. (cid:3) 15 Remark 4.6. We recall the restriction map r : H 0(w0, g/b) −→ H 0(w, g/b) which is used in the proof of Theorem 4.2. Let q = r−1(H 0(X(w), Tw)). The proof of the statement q = pw is due to the referee. We are very grateful to him/her for the proof. Remark 4.7. Corollary 4.3 does not hold in the non simply laced case. For instance, let G be of type B2, and w = s2s1. Here, we follow the convention with hα1, α2i = −2 and hα2, α1i = −1. In this case, we have β = α1 + α2 ∈ R+ \ (Sv≤w R+(v−1)), and the natural action of U−β on X(w) is not trivial. 5. Cohomology modules in the non simply laced case Throughout this section, we assume that G is a simple algebraic group of adjoint type over C which is not simply laced. Since G is not simply laced, there is a highest long root and there is a highest short root. We denote the highest long root by α0 and the highest short root by β0. We first prove the following. Lemma 5.1. There is a positive root β and a simple root α such that sα ◦ β = β0. Proof. Since β0 < α0, there is a simple root α such that β0 + α is a positive root. Since β0 + α 6= α, it follows that β = sα ◦ β0 = sα(β0 + α) is a positive root. (cid:3) Let α and β be as in Lemma 5.1. Let w ∈ W be such that sα ≤ w. Let V1 := ⊕µ≤βgµ be the direct sum of all weight spaces of weights µ satisfying µ ≤ β. Also, let V2 := ⊕µ<βgµ be the direct sum of all weight spaces of weights µ satisfying µ < β. We note that V2 is a B- submodule of g containing b and V1 is a B-submodule of g containing V2. Then we have Lemma 5.2. H 1(w, V1/V2) is non-zero. Proof. Note that β0 = sα ◦ β is a dominant character of T . Hence, by the Borel-Weil- Bott's theorem [3], H 1(w0, β) is an irreducible representation of G with highest weight sα ◦ β. Also, by Lemma 2.2, H 1(sα, β) is non-zero. Also, by [12, Corollary 4.3], the restriction map H 1(w0, β) −→ H 1(sα, β) is surjective. Thus, the restriction map H 1(w, β) −→ H 1(sα, β) is also surjective. Hence, H 1(w, β) is non-zero. Since the quotient B-module V1/V2 is isomorphic to Cβ, we have H 1(w, V1/V2) = H 1(w, β). Hence H 1(w, V1/V2) is non-zero. This completes the proof. (cid:3) Let V (β0) be the irreducible G-module of highest weight β0. Notation: (1) Let γ be a short simple root. Since V (β0)γ is non zero, there is a v ∈ V (β0)0 such that x−γ · v is non zero and x−ν · v is zero for every simple root ν different from γ. The cyclic Bγ-submodule of V (β0) generated by v is a two dimensional indecomposable Bγ-submodule of type similar to that of type (2) in the statement of Lemma 3.3. We denote it by V (β0)0,−γ. 16 (2) Let γ be a simple root not necessarily a short root. If β is a short root such that hβ, γi = 1, then the cyclic Bγ-submodule of V (β0) generated by the one dimensional subspace V (β0)β is a two dimensional indecomposable Bγ-submodule of type similar to that of type (4) in the statement of Lemma 3.3. We denote it by V (β0)β,β−γ. Let V be a B-submodule of V (β0) such that for every short simple root α, there is a decomposition of the Bα-module V such that every indecomposable Bα-summand V ′ of V is one of the following: (1) V ′ = Cv for some non zero vector v ∈ V (β0)0 such that x−α · v = 0. (2) V ′ = V (β0)0,−α. (3) V ′ = V (β0)β for some short root β such that hβ, αi ∈ {−1, 0}. (4) V ′ = V (β0)β,β−α for some short root β such hβ, αi = 1. (5) V ′ = sl2,α, the three dimensional irreducible Lα-module with highest weight α. The following lemma is useful to prove the main results in section 5 and section 6. Lemma 5.3. Let φ ∈ W and γ ∈ S. Then we have (1) If γ is a short root, then there is a decomposition of the Bγ-module H 0(φ, V ) such that every indecomposable Bγ-summand of H 0(φ, V ) is one of the five types as in the hypothesis above. (2) If γ is a long root, then there is a decomposition of the Bγ-module H 0(φ, V ) such that every indecomposable Bγ-summand of H 0(φ, V ) is one of the types (1), (3) or (4) as in the hypothesis above. Proof. Proof of (2): If γ is a long simple root, then V (β0)−γ = 0. The proof of (2) follows by using similar arguments in the proof of Lemma 3.3. The proof of (1) is essentially the same as that of Lemma 3.3. (cid:3) Corollary 5.4. Let φ ∈ W . (1) Let V be a B-submodule of V (β0) satisfying the hypothesis of Lemma 5.3. Then we have H i(φ, V ) = 0 for every i ≥ 1. (2) Let V be a B-submodule of V (β0) such that Vµ = 0 unless µ ∈ −(R+ \ S). Then we have H i(φ, V ) = 0 for every i ≥ 1. Proof. The proof is by induction on l(φ). If l(φ) = 0, then φ = 1 and so we are done. Otherwise, choose a simple root α such that l(φ) = 1 + l(sαφ). By induction, we have H i(sαφ, V ) = 0 for every i ≥ 1. Applying this to SES, we see that H i(φ, V ) = 0 for every i ≥ 2 and H 1(φ, V ) = H 1(sα, H 0(sαφ, V )). Therefore, it is sufficient to show that H 1(sα, H 0(sαφ, V )) = 0. Proof of 1: If α is a long root, then by Lemma 5.3(2), every indecomposable Bα-summand V ′ of H 0(sαφ, V ) is one of the types (1), (3) or (4) as in the hypothesis of Lemma 5.3. In all these cases, in view of Lemma 2.4, we see that there is an irreducible Lα-module V ′′ and an integer a ∈ {−1, 0} such that V ′ = V ′′ ⊗ Caωα. Hence, by using Lemma 2.3, we see that H i(sα, V ′) = 0 for every i ≥ 1. Thus, we have H 1(sα, H 0(sαφ, V )) = 0. If α is a short root, using Lemma 5.3(1), we see that every indecomposable Bα-summand V ′ of H 0(sαφ, V ) is one of the five types as in the hypothesis of Lemma 5.3. Therefore, in 17 view of Lemma 2.4, we see that every indecomposable Bα-summand V ′ of H 0(sαφ, V ) is of the form V ′ = V ′′ ⊗ Caωα for some irreducible Lα-module V ′′ and an integer a ∈ {−1, 0}. We apply Lemma 2.3 to conclude that H 1(sα, H 0(sαφ, V )) = 0. This completes the proof of (1). Proof of 2: By Lemma 2.6, H 0(sαφ, V ) is a B-submodule of V and therefore it satisfies the hypothesis of (2). Let V ′ be an indecomposable Bα-summand of H 0(sαφ, V ). In view of Lemma 2.4, we have V ′ = V ′′ ⊗ Caωα for some irreducible Lα-module V ′′ and an integer a. In particular, the set of all weights of V ′ is equal to {λ + aωα, λ − α + aωα, · · · , sα(λ) + aωα}, where λ is the highest weight of the irreducible Lα-module V ′′. Note that by the hypothesis, we have λ − iα + aωα ∈ −(R+ \ S), for every 0 ≤ i ≤ hλ, αi. Further, λ − iα + aωα is a short root for every 0 ≤ i ≤ hλ, αi. Therefore, we have −1 ≤ hsα(λ) + aωα, αi ≤ a ≤ hλ + aωα, αi ≤ 1. Thus, we have a ∈ {−1, 0, 1}. Hence, by using Lemma 2.3, we see that H i(sα, V ′) = 0 for every i ≥ 1. Thus, we have H 1(sα, H 0(sαφ, V )) = 0. This completes the proof of (2). (cid:3) Corollary 5.5. Let φ ∈ W . Let V1 be a B-submodule of V (β0) and V2 be a B-submodule of V1 . Assume that both V1 and V2 satisfy either the hypothesis of Corollary 5.4(1) or the hypothesis of Corollary 5.4(2). Then we have H i(φ, V1/V2) = 0 for every i ≥ 1. Proof. By Corollary 5.4, we see that H i(φ, Vj) = 0 for every i ≥ 1 and j = 1, 2. Now, the proof follows by using the arguments similar to the proof of Lemma 3.5(1). (cid:3) Corollary 5.6. Let φ ∈ W and α be a short root such that −α /∈ S. Then we have H i(φ, α) = 0 for every i ≥ 1. Proof. Take V1 := ⊕µ≤αV (β0)µ and V2 := ⊕µ<αV (β0)µ. If α ∈ R+, then ⊕µ≤0V (β0)µ is a subspace of V2. Therefore, imitating the proof of Lemma 3.3, we see that both the B-modules V1 and V2 satisfy the hypothesis of Lemma 5.3 ( the only difference here is that we have to work with the zero wight space of V (β0) instead of h and the set of all short simple roots instead of the set of all simple roots ). Therefore, both V1 and V2 satisfy the hypothesis of Corollary 5.4(1). Otherwise, both the B-modules V1 and V2 satisfy the hypothesis of Corollary 5.4(2). In both cases, the proof of the corollary follows by using Corollary 5.5 and imitating the proof of Corollary 3.6. (cid:3) 6. Main results in the non simply laced case In this section, we prove the main results in the non simply laced case. Through out this section, we assume that G is not simply laced. The following lemma is useful to prove the main results in this section. 18 Lemma 6.1. Let V be a B-module. Let w ∈ W and let γ be a simple root such that l(wsγ) = l(w) − 1. Let φ = wsγ. Then we have the following long exact sequence of B- modules: H 1(φ, H 0(sγ, V )) −→ H 1(w, V ) −→ H 0(φ, H 1(sγ, V )) −→ · · · · · · −→ H i(φ, H 0(sγ, V )) −→ H i(w, V ) −→ H i−1(φ, H 1(sγ, V )) −→ · · · Proof. Fix a decomposition of V = ⊕r j=1Vj into a direct sum of indecomposable Bγ-modules. By using Lemma 2.4, for each 1 ≤ j ≤ r, there exists an irreducible Lγ-module V ′ j and an integer mj such that Vj = V ′ j ⊗ Cmj ωγ . Now, let J1 := {j ∈ {1, 2, · · · , r} : mj ≥ 0} and J2 := {1, 2, · · · , r} \ J1. By using Lemma 2.3, we see that the image V+ of the evaluation map H 0(sγ, V ) −→ V is equal to ⊕j∈J1Vj. Let V− := ⊕j∈J2Vj. Clearly, V+ is a B-submodule of V and V− is a Bγ-submodule of V . Note that the restriction to V− of the natural map V −→ V /V+ is an isomorphism of Bγ-modules. Further, we have H i(sγ, V+) = 0 for every i ≥ 1 and H i(sγ, V /V+) = 0 for every i 6= 1 ( see Lemma 2.3 ). Also, the inclusion V+ ⊂ V of B-modules induces an isomorphism H 0(sγ, V+) −→ H 0(sγ, V ) of B-modules. Similarly, the natural map V −→ V /V+ induces an isomorphism H 1(sγ, V ) −→ H 1(sγ, V /V+) of B-modules. Let w = si1si2 · · · sir be a reduced expression for w with sir = sγ. Let i = (i1, i2, · · · , ir) and i′ = (i1, i2, · · · , ir−1) Recall the morphism fr : Z(w, i) −→ Z(φ, i′) from section 2. Using (Iso), we see that Rjfr ∗L(w, i, V ) = L(φ, i′, H j(Pαir /B, L(sir, V ))) (j ≥ 0). Using this isomorphism and the fact that H j(sγ, V+) = 0 for j ≥ 1, it follows that the B-modules H i(φ, H 0(sγ, V+)) and H i(w, V+) are isomorphic for every i ≥ 0. Similarly, the B-modules H i−1(φ, H 1(sγ, V /V+)) and H i(w, V /V+) are isomorphic for every i ≥ 1. Sum- marising these observations, we see that the B-modules H i(φ, H 0(sγ, V )) and H i(w, V+) are isomorphic for every i ≥ 0. Similarly, we also see that H i−1(φ, H 1(sγ, V )) and H i(w, V /V+) are isomorphic for every i ≥ 1. Applying H 0(w, −) to the short exact sequence 0 −→ V+ −→ V −→ V /V+ −→ 0 of B-modules, we obtain the following long exact sequence of B-modules: · · · H i−1(w, V /V+) −→ H i(w, V+) −→ H i(w, V ) −→ H i(w, V /V+) −→ H i+1(w, V+) · · · By using the above isomorphisms in this long exact sequence, we obtain the following long exact sequence of B-modules: H 1(φ, H 0(sγ, V )) −→ H 1(w, V ) −→ H 0(φ, H 1(sγ, V )) −→ · · · · · · −→ H i(φ, H 0(sγ, V )) −→ H i(w, V ) −→ H i−1(φ, H 1(sγ, V )) −→ · · · This completes the proof of the lemma. (cid:3) We now deduce the following. 19 Lemma 6.2. Let w ∈ W and V be a B-submodule of g. Then we have H i(w, V ) = 0 for every i ≥ 2. Proof. The proof is by induction on l(w). If l(w) = 0, we are done. Otherwise, choose a simple root γ ∈ S such that l(wsγ) = l(w) − 1. Step1. We show that H 1(sγ, V ) is either zero or is a direct sum of the trivial B-module C0 with multiplicity at most one and a B-module V ′ having a composition series V ′ = Vm ⊃ Vm−1 ⊃ · · · V1 ⊃ V0 such that each subquotient Vi+1/Vi ( 0 ≤ i ≤ m − 1 ) is isomorphic to Cβi for some short root βi which is not in −S. Assume that H 1(sγ, V )µ 6= 0. Then there exists an indecomposable Bγ-direct summand V1 of V such that H 1(sγ, V1)µ 6= 0. By Lemma 2.4, V1 = V ′ ⊗ Caωγ for some irreducible Lγ-module V ′ and an integer a. Since H 1(sγ, V1) 6= 0, by Lemma 2.3, we have a ≤ −2 and H 1(sγ, V1) = V ′ ⊗ H 0(sγ, aωγ − (a + 1)γ). Therefore, any weight µ′′ of H 1(sγ, V1) is in the γ-string from µ1 + γ = µ1 + ρ − sγ(ρ) = sγ(sγ ◦ µ1) to sγ ◦ µ1, where µ1 is the lowest weight of V1. Thus, there is an integer 1 ≤ t ≤ −(hµ1, γi + 1) such that µ = µ1 + tγ. We now analyse the cases. If µ = 0, then µ1 = −γ, and the indecomposable Bγ-summand V1 is equal to g−γ. Further, by Lemma 2.2(3), it follows that the trivial B-module C0 is a B-summand of H 1(sγ, V ). Also, the zero weight space H 1(sγ, V )0 of H 1(sγ, V ) has multiplicity one in this case. Otherwise, by using the above discussion, we see that µ1 is a root different from −γ In particular, µ is a short root. Further, such that hµ1, γi ∈ {−2, −3}, and V1 = gµ1. if µ ∈ −S, there is an integer 1 ≤ t ≤ −(hµ1, γi + 1) such that µ = µ1 + tγ ∈ −S. If hµ1, γi = −2, then t = 1 and so the simple roots −µ and γ are orthogonal with µ1 = µ − γ is a root, contradiction to the fact that sum of two orthogonal simple roots is not a root. If hµ1, γi = −3, then t = 1, 2. The proof of the case when t = 1 follows by similar arguments as above. If t = 2, then −µ is a simple root such that h−µ, γi = −1. This is a contradiction to the assumption that −µ + 2γ = −µ1 ∈ R ( For instance, since both −µ and γ are simple roots, sγ(−µ1) = −µ − γ /∈ R ). This completes the proof of Step 1. Now, let φ = wsγ. By Corollary 5.6, we see that H i(φ, H 1(sγ, V )) = 0 for every i ≥ 1. On the other hand, using Lemma 2.6, we see that H 0(sγ, V ) is a B-submodule of V . Hence, by using induction on l(w), we see that H i(φ, H 0(sγ, V )) = 0 for every i ≥ 2. Now, the proof of the lemma follows by applying the above observations in the following long exact sequence of B-modules ( see Lemma 6.1 ): H 1(φ, H 0(sγ, V )) −→ H 1(w, V ) −→ H 0(φ, H 1(sγ, V )) −→ · · · · · · −→ H i(φ, H 0(sγ, V )) −→ H i(w, V ) −→ H i−1(φ, H 1(sγ, V )) −→ · · · (cid:3) The following is a consequence of Lemma 6.2. Lemma 6.3. Let w ∈ W . Let V2 be a B-submodule of g and V1 be a B-submodule of g containing V2. Then we have H i(w, V1/V2) = 0 for all i ≥ 2. 20 Proof. The proof is similar to that of Lemma 3.5(1). We use Lemma 6.2 instead of Lemma 3.4. (cid:3) Corollary 6.4. Let w ∈ W and let α ∈ R. Then we have H i(w, α) = 0 for every i ≥ 2. Proof. Let V1 := ⊕µ≤αgµ denote the direct sum of the weight spaces of g of weights µ satisfying µ ≤ α. Let V2 := ⊕µ<αgµ denote the direct sum of the weight spaces of g of weights µ satisfying µ < α. It is clear that V2 is a B-submodule of g and V1 is a B-submodule of g containing V2. Since the B-module V1/V2 is isomorphic to Cα, we have H i(w, α) = H i(w, V1/V2) for every (cid:3) i ≥ 2. Now, the proof of the corollary follows from Lemma 6.3. The following theorem is a main result in the non simply laced case. Theorem 6.5. Let w ∈ W . Then we have (1) H i(X(w), TG/B) = 0 for every i ≥ 1. (2) The adjoint representation g of G is a B-submodule of H 0(X(w), TG/B) if and only if w−1(α0) is a negative root. Proof. The proof is similar to that of Theorem 4.1. We provide a proof here for completeness. As in the proof of Theorem 4.1, it is sufficient to prove the following: (1) H i(w, g/b) = 0 for every i ≥ 1. (2) The adjoint representation g of G is a B-submodule of H 0(w, g/b) if and only if w−1(α0) is a negative root. Proof of (1): As in the proof of Theorem 4.1, let V1 := g and V2 := b. The natural projec- tion Π : g −→ g/b of B-modules induces a homomorphism Πw : H 0(w, g) −→ H 0(w, g/b) of B-modules. We have the short exact sequence 0 −→ b −→ g −→ g/b −→ 0 of B-modules. Applying H 0(w, −) to this short exact sequence of B-modules, we obtain the following long exact sequence of B-modules: · · · H i(w, b) −→ H i(w, g) −→ H i(w, g/b) −→ H i+1(w, b) · · · On the other hand, by Lemma 2.5(2), we have H i(w, g) = 0 for every i ≥ 1. Further, by Lemma 6.2, we have H i+1(w, b) = 0 for every i ≥ 1. Applying this in the above long exact sequence of B-modules, we conclude that H i(w, g/b) = 0 for every i ≥ 1. This proves (1). The proof of (2) is similar to that of Theorem 4.1. (cid:3) The following result is the analogue of Theorem 4.2 in the non simply laced case. Theorem 6.6. Let w ∈ W be such that X(w) is smooth. Let Aw denote the connected com- ponent of the group of all automorphisms of X(w) containing the identity automorphism. For the left action of G on G/B, let Pw denote the stabiliser of X(w) in G. The homomorphism 21 φw : Pw −→ Aw of algebraic groups induced by the action of Pw on X(w) is injective if and only if w−1(α0) is a negative root. Proof. The proof is similar to that of Theorem 4.2(2). (cid:3) The following remark illustrates the difference between the main results in the simply laced case and those in the non simply laced case. Remark 6.7. The adjoint representation g of G could be a proper B-submodule of the space H 0(X(w), TG/B) of global sections. Let G be of type B2, and w = s2s1s2. Here, we follow the convention with hα1, α2i = −2 and hα2, α1i = −1. In this case, H 0(X(w), TG/B) is a direct sum of the B-modules g and C−(α1+α2). Acknowledgements We are very grateful to Professor M. Brion for pointing out this problem. We would like to thank the referee for useful comments to improve the exposition. We would like to thank Professor C. S. Seshadri, A. J. Parameswaran and D. S. Nagaraj for useful discussions. References [1] H.H. Andersen, Schubert varieties and Demazure's character formula, Invent. Math. 79 (1985), no. 3, 611-618. [2] V. Balaji, S. Senthamarai Kannan, K.V. Subrahmanyam, Cohomology of line bundles on Schubert varieties-I, Transformation Groups 9 (2004), no.2, 105-131. [3] R. Bott, Homogeneous vector bundles, Annals of Math., (2) 66 (1957), 203-248. [4] M. Brion, S. Kumar, Frobenius splitting methods in geometry and representation theory, Progress in Mathematics, Vol.231, Birkhauser, Boston, Inc., Boston, MA, 2005. [5] B. Narasmha Chary, S. Senthamarai Kannan, A.J. Parameswaran, Automorphism group of a Bott- Samelson-Demazure-Hansen variety, Transformation Groups 20 (2015), no.3, 665-698. [6] M. Demazure, A very simple proof of Bott's theorem, Invent. Math. 33 (1976), 271-272. [7] A. Grothendieck, Techniques de construction et th´eor`emes d'existence en g´eom´etrie alg´ebrique IV , les sch´emas de Hilbert, S´eminaire Bourbaki 5 (1960-1961), Expos´e no. 221, 28 pp. [8] R. Hartshorne, Algebraic Geometry , Graduate Texts in Mathematics, Springer Verlag, Berlin Heidel- berg, 1977. [9] J.E. Humphreys, Introduction to Lie algebras and Representation theory, Springer-Verlag, Berlin Hei- delberg, New York, 1972. [10] J.E. Humphreys, Linear Algebraic Groups, Springer-Verlag, Berlin Heidelberg, New York, 1975. [11] J.C. Jantzen, Representations of Algebraic Groups, ( Second Edition ), Mathematical Surveys and Monographs, Vol. 107, 2003. [12] S. Senthamarai Kannan, Cohomology of line bundles on Schubert varieties in the Kac-Moody setting, J. Algebra 310 (2007), no. 1, 88-107. [13] S.S. Kannan, S.K. Pattanayak, Torus quotients of homogeneous spaces-minimal dimensional Schubert varieties admitting semi-stable points, Proc. Indian Acad. Sci.(Math. Sci.) 119 (2009), no.4, 469-485. [14] P. Littelmann, Contracting modules and standard monomial theory for symmetrizable Kac-Moody al- gebras. J. Amer. Math. Soc. 11 (1998), no. 3, 551-567. [15] H. Matsumura, F. Oort, Representability of group functors, and automorphisms of algebraic schemes, Invent. Math. 4 (1967), 1-25. [16] V. B. Mehta, A. Ramanathan, Frobenius splitting and cohomology vanishing for Schubert varieties. Ann. of Math. (2) 122 (1985), no. 1, 27-40. [17] D. Mumford, J. Fogarty and F. Kirwan, Geometric Invariant Theory, (Third Edition), Springer-Verlag, Berlin Heidelberg, New York, 1994. 22 [18] C. S. Seshadri, Introduction to the theory of Standard Monomials. With notes by Peter Littelmann and Pradeep Shukla. Appendix A by V. Lakshmibai. Revised reprint of lectures published in the Brandeis Lecture Notes series. Texts and Readings in Mathematics, 46. Hindustan Book Agency, New Delhi, 2007. [19] C. Weibel, An introduction to homological algebra, Cambridge University Press, 1994. Chennai Mathematical Institute, Plot H1, SIPCOT IT Park, Siruseri, Kelambakkam 603103, Tamilnadu, India. E-mail address: [email protected]. 23
1412.3720
2
1412
2017-11-30T04:02:17
Note on MacPherson's local Euler obstruction
[ "math.AG" ]
This is a note on MacPherson's local Euler obstruction, which plays an important role recently in Donaldson-Thomas theory by the work of Behrend. We introduce MacPherson's original definition, and prove that it is equivalent to the algebraic definition used by Behrend, following the method of Gonzalez-Sprinberg. We also give a formula of the local Euler obstruction in terms of Lagrangian intersections. As an application, we consider a scheme or DM stack $X$ admitting a symmetric obstruction theory. Furthermore we assume that there is a $\CC^*$ action on $X$, which makes the obstruction theory $\CC^*$-equivariant. The $\CC^*$-action on the obstruction theory naturally gives rise to a cosection map in the sense of Kiem-Li. We prove that Behrend's weighted Euler characteristic of $X$ is the same as Kiem-Li localized invariant of $X$ by the $\CC^*$-action.
math.AG
math
NOTE ON MACPHERSON'S LOCAL EULER OBSTRUCTION YUNFENG JIANG Abstract. This is a note on MacPherson's local Euler obstruction, which plays an important role recently in Donaldson-Thomas theory by the work of Behrend. We introduce MacPherson's original definition, and prove that it is equiv- alent to the algebraic definition used by Behrend, following the method of Gonzalez-Sprinberg. We also give a formula of the local Euler obstruction in terms of Lagrangian intersections. As an application, we consider a scheme or DM stack X admitting a symmetric obstruction theory. Furthermore we assume that there is a C∗ action on X, which makes the obstruction theory C∗-equivariant. The C∗-action on the obstruction theory naturally gives rise to a cosection map in the sense of Kiem-Li. We prove that Behrend's weighted Euler characteristic of X is the same as Kiem-Li localized invariant of X by the C∗-action. 1. Introduction 1.1. Local Euler obstruction. Local Euler obstruction of a scheme or a variety was originally introduced by MacPherson [14] to study Chern class of singular alge- braic varieties. The definition of the local Euler obstruction is given by topological and geometrical method in Section 3 of [14], by using the Nash blow-up. In [16], Gonzalez-Sprinberg proves an algebraic formula using Segre class of normal cone of a closed subscheme inside a scheme. Let Z be a prime cycle in a scheme or DM stack X, the local Euler obstruc- tion Eu(Z) is a constructible function on X, and all the local Euler obstructions form a basis in the group F (X) of constructible functions on X. Furthermore, MacPherson proves that there is a unique transformation functor from the functor of constructible functions to the functor of homology groups satisfying the push- forward property, such that the homology class of the constant function 1X for a smooth scheme X is c(X) ∩ [X]. A global characteristic class for the local Euler obstruction Eu(Z) is the Chern- Mather class cM (Z) also introduced in Section 2 of [14]. This class is the pushfor- ward ν∗(c(T Z)∩ [bZ]) ∈ A∗(Z) where ν : bZ → Z is the Nash blow-up and T Z is the Nash tangent bundle. Global index theorem of MacPherson says that cM (Z) = χ(X, Eu(Z)), ZX if the scheme X is proper, where χ(X, Eu(Z)) is the weighted Euler characteristic weighted by the constructible function Eu(Z). In the case that Z = X and X is smooth, Chern-Mather class cM (Z) is just the Chern class and χ(X, Eu(Z)) is just the topological Euler characteristic of X. This is the Gauss-Bonnet theorem. 2010 Mathematics Subject Classification. Primary 14N35; Secondary 14A20. 1 2 JIANG In this paper we review the definition of MacPherson on Euler obstruction using topological method, and we prove that it is equivalent to the algebraic formula in the Theoreme of [16] following the arguments of Gonzalez-Sprinberg. Note that [16] was written in French, and the translation proof may be helpful for English readers. 1.2. Donaldson-Thomas theory. The local Euler obstruction recently becomes very important in the study of Donaldson-Thomas theory by the work of Behrend [2]. In [2], Behrend defines a canonical cycle cX ∈ Z∗(X) for a scheme or DM stack X, which Behrend calls the sign support of the intrinsic normal cone cX in [3]. The Euler obstruction νX := Eu(cX ) of this cycle is called the "Behrend function" of X. The weighted Euler characteristic χ(X, νX ) is well defined. If the scheme or DM stack X admits a symmetric obstruction theory EX intro- duced in [2], then the virtual fundamental class [X]vir ∈ A0(X) is zero dimensional. In the case that X is proper, the virtual count (Donaldson-Thomas type invariant) #vir(X) is defined as Z[X]vir 1. Behrend proves that it is the same as the weighted Euler characteristic χ(X, νX ). The Donaldson-Thomas moduli space of stable coherent sheaves on Calabi-Yau threefolds admits a symmetric obstruction theory, and the Donaldson-Thomas in- variant constructed in [17] is the weighted Euler characteristic weighted by the Behrend function of the moduli space. If a moduli space X is not proper so that the integration does not makes sense, the weighted Euler characteristic χ(X, νX ) is defined as the Donaldson-Thomas invariant. Examples include an amount of moduli spaces of quiver representations with potentials. 1.3. Lagrangian intersection. Let X ֒→ M be an embedding into a smooth DM stack M . Then the symmetric obstruction theory EX on X induces a cone C inside ΩMX , which Behrend calls the obstruction cone. The key observation of [2] (suggested by R. Thomas) is that the obstruction cone C is a Lagrangian cone inside ΩM . The obstruction cone C, ´etale locally on a chart of X, is given by the normal cone of a closed immersion. Z/M , where Z is a closed substack of X and N ∗ There is an isomorphism between the group of integral cycles Z∗(X) of X and the group LX (ΩM ) of conic Lagrangian cycles inside ΩM supported on X. The Lagrangian cone C is a linear combination of conic Lagrangian cycles of the form N ∗ Z/M is the closure inside ΩM of the conormal bundle of smooth locus of Z. Note that N ∗ Z/M gives all the irreducible components of the cone C, and its image under the morphism π : C → X gives the integral cycle cX . Ginzburg's theory [9] tells us that the Lagrangian intersection I(N ∗ Z/M with the zero section is the weighted Euler characteristic χ(X, Eu(Z)) of the Euler obstruction Eu(Z) up to a sign when Z is proper. On the other hand, the Lagrangian intersection also gives the degree zero Chern-Mather class of Z. So the Lagrangian intersection I(C, [M ]) gives the weighted Euler char- acteristic χ(X, νX ), hence the Donaldson-Thomas invariant #vir(X). Z/M , [M ]) of N ∗ The Lagrangian intersection I(C, [M ]) can be explained using another (real) analytic section of ΩM as studied in Theorem 9.5.3 and Theorem 9.7.11 of [11], and the Appendix in [7]. By fixing a Hermitian metric on ΩM , we may choose a small NOTE ON MACPHERSON'S LOCAL EULER OBSTRUCTION 3 perturbation Γ of the zero section of ΩM . Then I(C, [M ]) = I(C, Γ), see Theorem 4.4. 1.4. Applications. The idea of Lagrangian intersection above has applications in Donaldson-Thomas theory if the scheme (the moduli space) X admits a C∗ action. Let F ⊂ X be the fixed point locus of the C∗ action. Then the weighted Euler characteristic χ(X, νX ) = χ(F, νXF ), where νXF is the restriction of νX to F . The C∗ action on X naturally gives rise to a cosection morphism σ : ΩX → OX in the sense of [7] such that the degenerate locus is exactly the fixed point locus F . The virtual dimension of X with symmetric obstruction theory EX has dimension zero. Hence the Kiem-Li localized virtual cycle [F ]vir loc,KL ∈ A0(F ) is a dimension zero class. We prove that the Kiem-Li localized invariant Z[F ]vir loc,KL 1 = χ(F, νXF ), see Theorem 5.20. We should remark that in the case that X is proper, this is a direct application of Theorem 4.18 of Behrend [2] and Theorem 1.1 of Kiem-Li [7]. So the key point of this paper is to prove Theorem 5.20 in the case that X is nonproper, but the C∗ fixed locus F is proper. Since one can do C∗ localization on the Kiem-Li localized virtual cycle, it is hoped that Kiem-Li calculation is easier than the calculation of the Behrend function. If the C∗ action only has isolated fixed points, we recover Theorem 3.4 of Behrend and Fantechi [4]. This note is organized as follows. Section 2 introduces MacPherson's original definition of local Euler obstruction. In Section 3 we prove the algebraic formula of local Euler obstruction following the method in [16]. In Section 4 the global index theorem and the Lagrangian intersection are discussed, and in Section 5 we give the application of Euler obstruction and Lagrangian intersection to Behrend theory and Kiem-Li cosection localization. Acknowledgement. The author would like to thank Professors Kai Behrend, Jun Li and Richard Thomas for valuable discussions. The last section of this paper was motivated by a discussion with Kai Behrend when the author was visiting UBC in November 2012, a discussion with Jun Li when the author was visiting Stanford in December 2012, and the work with Richard Thomas about virtual signed Euler characteristics. He thanks Jim Bryan, Kai Behrend (UBC), and Jun Li (Stanford) for hospitality. This work is partially supported by Simons Foundation Collaboration Grant 311837, and NSF Grant DMS-1600997. 2. Local Euler obstruction of MacPherson 2.1. Nash blow-up. Let Z be a d-dimensional scheme or DM stack, which is embedded into a smooth scheme or DM stack A. Let Grd(T A) be the Grassmannian bundle of d-dimensional subspaces of the bundle T A over A. There is a section s : Z ◦ → Grd(T A) 7→ Tz(Z ◦) z 4 JIANG of smooth locus Z ◦ ⊂ Z into the Grassmannian bundle, where Tz(Z ◦) is the tangent Grassmannian bundle. Denote by space of Z ◦ at the smooth point z. Let bZ be the closure of the image s inside the ν : bZ → Z the map given by the restriction of the projection of Grd(T A). There is a vector bundle T Z over bZ, which is the restriction of the tautological bundle over the Grassmannian Grd(T A). The map ν : bZ → Z is called the Nash blow-up of Z. Remark 2.1. The construction of Nash blow-up of Z is independent to the embed- ding. 2.2. Local Euler obstruction. We give MacPherson's original local Euler ob- struction for a scheme or DM stack Z. Let Z ֒→ A be an embedding. Let P ∈ Z. We choose a local coordinates (z1,··· , zn) of A at the point P , such that zi(P ) = 0. Let kzk2 = √z1z1 + ··· + z1zn. Then kzk2 is a real-valued function, and the differential dkzk2 may be considered as a section of T ∗A, where T ∗A is the cotangent bundle of A but viewed as a real vector bundle. The section dkzk2 pulls back and restricts to a section r of T ∗Z, which is the dual of T Z. Remark 2.2. Here we use T ∗A to represent the real cotangent bundle of A. Later on in Section 5 we use ΩA as the cotangent bundle of a scheme or DM stack A. Lemma 2.3. For small enough ǫ > 0, the section r is nonzero over ν−1(z) where 0 < kzk ≤ ǫ. Proof. Let bZP = ν−1(P ) be the fibre over P , and K be the set of zeros of r. Then bZP ⊂ K. Suppose by contradiction there exists x ∈ bZP ∩ K − bZP , where K − bZP (closure of K−bZP ) is real sub-analytic. So by Bruhat-Whitney Lemma, there exists an analytic curve eC : [0, t] → K − bZP such that eC(0) = x and eC((0, t]) ⊂ K − bZP . Projecting on Z, an analytic curve C passing through P is obtained. Let S be a Whitney stratification of Z. For t 6= 0 sufficiently small, C((0, t]) is fully contained in a stratum S. Let P ′ 6= P be a point on the curve C. Then the right secant P P ′ is orthogonal to the limit tangent space TP ′ at P ′ (i.e. a point of bZP ′ ∩ eC which projects to P ′). However from Condition A of Whitney, TP ′ contains the tangent space to the stratum at P ′, that is , it contains the right tangent space to C at P ′. So P P ′ is orthogonal to the tangent to C at P ′ for all P ′ ∈ C, and P ′ 6= P , which is impossible. (cid:3) Let Bǫ be the ǫ-ball {zkzk ≤ ǫ} and Sǫ be the ǫ-sphere {zkzk = ǫ}. The obstruction to extending r as a nonzero section of T ∗Z from ν−1Sǫ to ν−1Bǫ, which MacPherson denoted it by Eu(T ∗Z, r), lies in H 2d(ν−1Bǫ, ν−1Sǫ; Z). Let O ∈ H2d(ν−1Bǫ, ν−1Sǫ; Z) be the orientation class. Definition 2.4. The local Euler obstruction of Z at P is defined as: Eu(Z)(P ) = hEu(T ∗Z, r),Oi. The local Euler obstruction has the following properties, see Section 3 in [14]: (1) Eu(Z)(P ) = 1 if Z is nonsingular at P ; (2) If Z is a curve, then Eu(Z)(P ) is the multiplicity of Z at P . If Z is the cone on a nonsingular plane curve of degree d and P is the vertex, Eu(Z)(P ) = 2d − d2; NOTE ON MACPHERSON'S LOCAL EULER OBSTRUCTION 5 (3) Eu(Z1 × Z2)(P1 × P2) = Eu(Z1)(P1) · Eu(Z2)(P2); (4) If we have a scheme Z or DM stack which is reducible at the point P , and let Zi are the irreducible components, then Eu(Z)(P ) =P Eu(Zi)(P ). The local Euler obstruction Eu(Z) is a constructible function, which was proved in algebraic sense by Kennedy in Lemma 4 of [10]. Let F (Z) be the group of constructible functions with integer values if Z is a scheme and rational values if Z is a DM stack. Then there is a mapping (2.1) given by: T : Z∗(Z) → F (Z) X aiZi 7→X ai Eu(Zi). MacPherson proved that T is an isomorphism of abelian groups in Lemma 2 of [14]. The proof is true for DM stacks by taking Q coefficients for the constructible function. 3. Algebraic formula of local Euler obstruction 3.1. Algebraic formula. In this section we give the algebraic explanation of the local Euler obstruction. We mainly follow the proof of Gonzalez-Sprinberg [16]. of Z. Let Z be a prime cycle in a DM stack X. Recall ν : bZ → Z is the Nash blow-up Theorem 3.1. We have: Eu(Z)(P ) =Zν−1(P ) c(T Z) ∩ s(ν−1(P ),bZ), where s(ν−1(P ),bZ) is the Segre class of the normal cone of ν−1(P ) in bZ. Remark 3.2. This formula is used by Behrend [2] in Section 1.2 of [2] as the defini- tion of local Euler obstruction. 3.2. Proof of Theorem 3.1. The local Euler obstruction Eu(Z)(P ) is local, then we may suppose that Z ⊂ Cn and take P to be the origin. The Nash tangent diagram: bundle is T Z and bZ0 = ν−1(0) is the fibre over 0 ∈ Z ⊂ Cn. Consider the following (3.1) bZ0 bZ ν Grd(T Cn) 0 / Z / Cn Step 1: Let T Cn = Cn × Cn be the trivial tangent bundle over Cn, and let E := ν∗(T CnZ ) be the pullback by ν to bZ. Then we have the following diagram: T Z ❆ E ❆ ❆ ❆ ❆ ❆ ❆ ❆ bZ where the Nash tangent bundle T Z can be taken as a subbundle of E. There is a canonical section Cn → T Cn, by P 7→ −−→OP (the vector of P ), then through the map ν, we have an induced section ρ of E over bZ. By choosing a Hermitian metric on E / /   / /     / / / /   6 JIANG (denoted it by a Hermitian form s), then we can project ρ to T Z and get a section σs of T Z over bZ. The Hermitian form s induces an isomorphism T Z ∼= T ∗Z, which transforms the section σs to r we defined before. Hence Eu(T ∗Z, r) ∼= Eu(T Z, σs). Remark 3.3. Note that the orientation of T ∗Z is the one from the identification between complex cotangent bundle and real cotangent bundle, and the orientation of T Z is the one coming from the complex structure. Step 2: Let ǫ > 0 be sufficiently small, such that σs does not vanish on ν−1Sǫ. Set V = ν−1Bǫ, ∂V = ν−1Sǫ. In order to calculate Eu(T Z, σs), we consider the class ω of universal obstruction of T Z, determined by the canonical section of p∗ T ZT Z, which is v 7→ (v, v), where v ∈ T Z, and pT Z : T Z → bZ is the projection. The class after identifying bZ as the zero section of T Z, and then it restricts to a class in ω ∈ H 2d(T Z, T Z − bZ) = H 2d H 2d V (T ZV ), still denoted by ω. The section σs of T Z induces: bZ (T Z) (V, V − bZ0) → (T ZV , T ZV − V ), (V ). σ∗ s : H 2d V (T ZV ) → H 2d bZ0 and induces: Then Let ǫ > 0 be small enough so that V is a neighbourhood of bZ0 and V −bZ0 retracts to ∂V . So H ∗ bZ0 (V ) = H ∗(V, V − bZ0) ∼= H ∗(V, ∂V ). σ∗ s (ω) ∈ H 2d (V ) ∼= H 2d(V, ∂V ) bZ0 is the class Eu(T Z, σs) by universal property of ω. So we have H 2d bZ0 and σ∗ s (ω) can be considered in H 2d bZ0 (V ) ∼= H 2d bZ0 (bZ), (3.2) Eu(Z)(0) = deg(σ∗ (bZ). So s (ω) ∩ [V, ∂V ]) = deg(σ∗ s (ω) ∩ [bZ]). where Q = E/T Z, and rk(T Z) = d, rk(E) = n. that we have an algebraic section of T Z on it, since σs is analytic. Our method is to split the bundle E as the sum of T Z and its complement without the aid of Hermitian form, and do it algebraically. More precisely, let us consider the exact Step 3: The next step is to construct a scheme or variety associated with bZ such sequence of vector bundles on bZ: Let Grn−d(E) be the Grassmannian bundle over bZ, whose fibre over x ∈ bZ U ⊂ Grn−d(E), whose fibre over x ∈ bZ is the complement of T Zx of Ex, i.e. a subspace a : Q ⊂ E such that j ◦ a = idQ. Let p : U → bZ be the projection, whose fibre over x ∈ bZ is an affine space of dimension d(n − d). This U is a principal is the space of (n − d)-dimensional subspaces in Ex. Consider the open subset 0 → T Z −→ E −→ Q → 0, j NOTE ON MACPHERSON'S LOCAL EULER OBSTRUCTION 7 homogeneous space. On U consider the pullback bundle p∗E, its section ρ and σs of T U = p∗T Z (we use the same notations). Let S be the complement of T U , obtained by restriction on U of the tautological fibre of the Grassmannian. Let σ (reap. η) be the projections of ρ on T U (resp. S). Then and E = T U ⊕ S, ρ = σ + η. Let U0 = p−1(bZ0). Then we have the following diagram: U0 U Proposition 3.4. Let u ∈ U0, and ǫ > 0 sufficiently small. Then there exists a neighbourhood W of u in U such that for all w ∈ W , bZ0 / bZ kη(w)k < ǫkσ(w)k. the origin, and an additional space S in E. Fix a Hermitian form s, then it gives the decomposition of E, i.e. S = T ⊥. Then s induces a section of U passing It suffices to prove the proposition on this section, because U is a through u. Proof. A point u ∈ U0 represents a point in bZ0, i.e. a limit tangent space T at principal homogeneous space, and we can identify bZ with this section. So we shall be induced to prove on bZ the following statement: Let x ∈ bZ0, ǫ > 0, then there exists a neighbourhood bV of x in bZ, such that for all v ∈ bV , we have kηs(v)k < ǫkσs(v)k (3.3) where σs and ηs are the components of ρ determined by s. Let {Mα} be a Whitney stratification of Z. So it suffices to show that there exists a neighbourhood V of 0 in Z such that for all α and for all z ∈ Mα ∩ V , one has tg2(oz, TMα,z) < ǫ2, where TMα,z is the tangent space of Mα at the point z. (This is because tg(oz, TZ) = kηs(v)k kσs(v)k .) Here tg(oz, TMα,z) and tg(oz, TZ) are the trigonometric tangent functions of the angles between oz and the tangent planes. We prove this by contradiction. Suppose that there is an α such that for every neighbourhood V of 0 there exists z ∈ Mα ∩ V and tg2(oz, TMα,z ) ≥ ǫ2. Consider all real semi-analytic sets R = {z ∈ Mαtg2(oz, TMα,z ) ≥ ǫ2}. Then 0 ∈ R, the closure of R. By the Bruhat-Whitney Lemma, there is a real analytic curve C : [0, t] → R, such that C(0) = 0, and C((0, t]) ⊂ R. So this curve would have the following property: at any point P 6= 0, P ∈ C, tg2(OP, tgP ) ≥ ǫ2. This is impossible because the tg tends to zero when P → 0. / /     / 8 JIANG As a result we have (3.3). ν(v) ∈ Z and a limit tangent space Tz of the tangent spaces to smooth points in a neighbourhood of z. So by Whitney condition A, we have: TMα,z ⊂ Tz and In fact, giving v ∈ bZ is equivalent to giving z = tg(oz, TMα,z) ≥ tg(oz, Tz) = kηs(v)k kσs(v)k . (cid:3) Corollary 3.5. The set U0 is open and closed in the set K of zeroes of σ. Proof. Indeed, U0 is the set of zeros of ρ, and ρ = 0 implies σ = 0. We show U0 is open by contradiction. Suppose there exists a sequence {ui} contained in K − U0 which converges to u ∈ U0. By the previous proposition, for i sufficiently large, σ(ui) = 0 implies η(ui) = 0. Since ρ = σ + η, ρ(ui) = 0, that is to say, ui ∈ U0 for i large enough and we have a contradiction. (cid:3) Step 4: We continue the proof of the theorem. Let K ′ = K − U0 be the union of connected components of K (the zeros of σ) disjoint to U0. Let W := U − K ′; which is an open neighbourhood of U0 in U , according to Corollary 3.5. Note also the restriction to W of p : U → bZ and consider the following diagram: p∗T Z WV U0 W where V = ν−1Bǫ and WV = p−1(V ). Then we have: π T Z p / bZ and So: / V bZ0 (WV , WV − U0) p∗(σ∗ p∗σ∗ s (ω) ∈ H 2d s (ω) ∩ [bZ]) = p∗σ∗ p → (V, V − bZ0) σ→ (T Z, T Z − bZ) U0 (WV ) ∼= H 2d U0 (W ). s ω ∩ [W ] ∈ H2d(n−d)(U0). In addition, we have morphisms of pairs of spaces: (W, W − U0) σ→ (p∗T ZW , p∗T ZW − W ) π→ (T Z, T Z − bZ) U0 (W ). where σ∗π∗ω ∈ H 2d We prove that p∗σ∗ the Hermitian form and sV is the restriction to V . It induces the isomorphism (sV )∗ : H 2d (V ), because we have the following commutative diagram: s ω = σ∗π∗ω. Note that s : bZ → U is the section induced from ∼=→ H 2d U0 (W ) bZ0 H 2d U0 (W ) (sV )∗ H 2d bZ0 (V ) ∼= ∼= where the horizontal arrows are excisions. H 2d bZ0 (U ) s∗ / H 2d bZ0 (bZ) / /   / /     o o   / / o o / /     / W σ / p∗T Z sV V σs π T Z p∗σ∗ s ω ∩ [W ] = σ∗π∗ω ∩ [W ] As a consequence we have: (3.4) and by projection: (3.5) NOTE ON MACPHERSON'S LOCAL EULER OBSTRUCTION 9 In fact, (sV )∗ is the inverse of p∗, so p∗σ∗ (sV )∗σ∗π∗ω. The latter is true by the following commutative diagram: s ω = σ∗π∗ω is equivalent to σ∗ s ω = σ∗(σ∗π∗ω ∩ [W ]) = π∗ω ∩ [σ(W )]. The latter class is the product of the fundamental class of W with fundamental homology class of σ(W ), so this is the class of intersection cycle [W ·σ W ] in A∗(W ∩ σ(W )). So (3.6) π∗ω ∩ [σ(W )] = [W ·σ W ]. Step 5: There is an algebraic formula to calculate the above intersection cycle due to Fulton [8]. The variety W is embedded as zero section into p∗T Z, which is locally complete intersection. The scheme or DM stack intersection W ∩ σ(W ) is defined by σ = 0 in W , which we will denote it by W0. The dimension of W is d + d(n − d) and its codimension in p∗T Z is d. Look at the following diagram: W σ / p∗T Z Then W0 W [W ·σ W ] = (c(p∗T ZW0) ∩ s(W0, W ))d(n−d) (3.7) i.e. the component of dimension d(n− d) in A∗(W0) is the cap product of the Chern class of p∗T ZW0 with the Segre class of W0 in W , see Chapter 6 in [8]. Corollary 3.6. With the above notation, we have s(W0, W ) = s(U0, U ). Proof. Let I = (σ1,··· , σn) (resp. J = (ρ1,··· , ρn)) be the ideal of definition of W0 (reap. U0), where σi (resp. ρi) is the i-th component of σ (resp. ρ). Let f : W → W be the blow up of W along W0. Let I = f ∗I = (σ1,··· , σn) (resp. J = f ∗J = (ρ1,··· , ρn)), where σi = f ∗σi, (resp. ρi = f ∗ρi). We show that I = J . On one hand, we have: I ⊂ J , because ρ = 0 implies σ = 0. So we have: I ⊂ J . On the other hand, since ρ = σ + η and kηk < ǫkσk in a neighbourhood of U0, then ρi < (1 + ǫ)Pn j=1 σj for i = 1,··· , n. Let g be a local generator of I (which is invertible). Then ρi g is a function locally bounded for all i, so holomorphic because W is normal. As a result, we have J ⊂ I. So J = I. By invariance of Segre class under birational map, s(U0, W ) = f∗s(U 0, W ) = s(W0, W ), where U 0 = W 0. Finally, since U0 is closed in U , we have s(U0, W ) = s(U0, U ). (cid:3) /   / / O O / / / O O O O 10 So (3.8) JIANG The projection p is flat, p∗s(bZ0, Z) = s(U0, U ). So [W ·σ W ] = (c(p∗T ZU0) ∩ s(U0, U ))d(n−d) [W ·σ W ] =(cid:16)c(p∗T Z bZ0 (3.9) So: This proves Theorem 3.1. Eu(Z)(P ) = deg(c(T Z bZ0 ) ∩ s(bZ0,bZ)(cid:17)0 ) ∩ s(bZ0,bZ)). 3.3. Another formula. We make the following diagram: D bZ0 0 eZ bZ b ν / Z where b : eZ → bZ is the blow-up of bZ along bZ0, and D = b−1(bZ0) is the exceptional divisor. Let ξ be the normal bundle of D ⊂ eZ. Corollary 3.7. Eu(Z)(P ) =ZD cd−1(T Z − ξ) ∩ [D], where c(T Z − ξ) = c(T Z) c(ξ) . Proof. This is from the definition of Segre class: So s(bZ0,bZ) = b∗(c(ξ)−1 ∩ [D]). deg(c(T Z) ∩ s(bZ0,bZ)) = deg(c(T Z) ∩ b∗(c(ξ)−1 ∩ [D])) = deg(c(T Z) ∪ c(ξ)−1 ∩ [D]) = deg(cd−1(T Z − ξ) ∩ [D]). (cid:3) Example 3.8. Let Z = Spec C[x, y, u, v]/(yu − xv). Then Z is smooth except the origin P , which is called conifold singularity. The natural embedding Z ⊂ C4 gives the Nash blow-up which is OP1×P1 (−1,−1). The Nash cotangent bundle is T ∗Z = PT ∗C4/O(1). is O(1, 1). Let c1(O(1, 1)) = x + y. Then from Corollary 3.7 Note that the line bundle O(1), when restrict to the exceptional divisor P1× P1, ν : bZ → Z Eu(Z)(P ) = deg(c(T Z) ∩ s(P1 × P1,bZ)) = deg( 1 1 + x + y · c(O(−1,−1))−1 ∩ [P1 × P1]) = 2. / /     / /     / NOTE ON MACPHERSON'S LOCAL EULER OBSTRUCTION 11 4. Global index theorem and Lagrangian Intersection 4.1. Chern-Mather class. Fix a DM stack X throughout this section. Let Z ⊂ X are defined in Section 2.1. be a prime cycle. The Nash blow-up ν : bZ → Z and the Nash tangent bundle T Z Definition 4.1. The Chern-Mather class cM (Z) is: where cM ([Z]) ∈ A∗(X). cM ([Z]) = ν∗(c(T Z) ∩ [bZ]), Let cM 0 (Z) be the degree zero part cM 0 : Z∗(X) → A0(X). The local Euler obstruction Eu(Z) is a constructible function X → Z. The weighted Euler charac- teristic χ(X, Eu(Z)) is given by Xi ni · χ(Eu(Z)−1(i)). In Proposition 1.12 of [2], Behrend proves Theorem 4.2. ZX cM (Z) = χ(X, Eu(Z)), if X is a proper scheme or global finite group quotient stack, or a grebe over a scheme. Proof. The scheme case is MacPherson's index theorem in [14]. Other cases were proved by Behrend [2] using properties of the local Euler obstruction. (cid:3) In ([5]), the result is proved for proper DM stacks: Theorem 4.3. if X is a proper DM stack. ZX cM (Z) = χ(X, Eu(Z)), Proof. In [5], the author defines ProChow group and classes for not necessarily proper DM stacks, generalizing the result in [1]. There is a natural transformation functor from the group of constructible functions on DM stacks to the ProChow groups. Every ProChow group class of a DM stack X gives a degree, which is the weighted Euler characteristic of the corresponding constructible function. In the case X is proper, the ProChow group of X coincides with the Chow group A∗(X), and the Euler obstruction Eu(Z) of the prime cycle Z in X is a constructible function, hence the degree of the corresponding Chow group class (Chern-Mather class) gives its weighted Euler characteristic. The formula in terms of Lagrangian intersection for orbifolds was addressed by Maulik and Treumann in [15]. (cid:3) 4.2. Lagrangian Intersection. The degree zero Chern-Mather class cM 0 (Z) and the weighted Euler characteristic can be interpreted by Lagrangian intersections. We mainly follow Behrend's proof in Section 4.1 of [2], only the last part involving small perturbation of zero section of vector bundles is not included in [2]. 12 JIANG Fix an embedding X → M of the DM stack X into a smooth DM stack M . We will explain the following diagram due to Behrend in Diagram (2) of [2]. (4.1) ∼= Z∗(X) Eu $■■■■■■■■■ cM 0 F (X) LX (ΩM ) Ch ∼= cSM 0 ysssssssss I(·,[M]) A0(X) where Z∗(X) is the group of integral cycles of X, F (X) is the group of constructible functions on X, and LX (ΩM ) is the subgroup of Zn(ΩM ) generated by the conic Lagrangian prime cycles supported on X. The maps cM and I(·, [M ]) are degree zero Chern-Mather class, degree zero Chern-Schwartz-Mather class and the Lagrangian intersection with zero section of ΩM , respectively. Note that in [2], the notation of Lagrangian intersection with zero section is denoted by 0! 0 , cSM 0 We briefly explain the horizontal morphisms in the diagram. The first map is the local Euler obstruction Eu and it gives an isomorphism from Z∗(X) to F (X), which is given by (2.1). ΩM (·). Behrend (Section 4.1, [2]) defined the following isomorphism of groups: (4.2) which is given by L : Z∗(X) → LX (ΩM ) where N ∗ Conversely there is an isomorphism: Z/M is the closure of the conormal bundle of smooth part of Z inside M . Z 7→ (−1)dim(Z)N ∗ Z/M , (4.3) which is given by π : LX (ΩM ) → Z∗(X) where π : V → X is the projection. Then the morphism Ch is defined by the isomorphism Eu and the morphism L defined above. V 7→ (−1)dim(π(V ))π(V ), Proof of Diagram 4.1: Let Z ⊂ X be a prime cycle. We may take the Nash blow-up ν : bZ → Z by embedding Z ⊂ M into the smooth DM stack M . Let be the Grassmannian of rank d quotient of ΩM . The Nash blow-up bZ is the closure inside fM of the section Z ֒→ fM for the smooth locus of Z. Then there is an exact µ : fM := Grd(ΩM ) → M sequence of vector bundles: 0 → N bZ −→ µ∗ΩM bZ −→ T ∗Z → 0 where N bZ is the kernel of the surjective map of the right arrow. We have the following commutative diagram: bZ [M ] N bZ / µ∗ΩM bZ. / / $ / /   y / /     / NOTE ON MACPHERSON'S LOCAL EULER OBSTRUCTION 13 So from Fulton Chapter 6 of [8], I(N bZ, [M ]) = c(ΩM bZ) · s(bZ, N bZ ) = c(ΩM bZ) · c(N bZ)−1 · [bZ] = c(T ∗Z) · [bZ]. From Proposition 4.6 in [2], we have diagram: (4.4) ν η bZ 0 N bZ M Z 0 / N ∗ Z/M / ΩM Z/M is a proper birational map of integral stacks, where N ∗ where η : N bZ → N ∗ is the closure inside ΩM of the conormal bundle of the smooth part of Z. Hence Z/M cM 0 (Z) = (−1)dim(Z)ν∗(c(T ∗Z ∩ [bZ])) = (−1)dim(Z)ν∗(I(N bZ , [M ])) = (−1)dim(Z)I(η∗N bZ , [M ]) = (−1)dim(Z)I(N ∗ Z/M , [M ]). (cid:3) 4.3. Analytic version of the Lagrangian intersection. In practice, it is often useful to do analytic or real version of Lagrangian intersection as in Theorem 9.5.3 and Theorem 9.7.11 of [11]. We choose a Hermitian metric for the cotangent bundle ΩM such that it is naturally isomorphic to the tangent bundle T M . Let be a real C2-function. Then the graph Γ = dψ is a real Lagrangian cycle in ΩM . Suppose that ψ : M → R (1) {x ∈ M : ψ(x) ≤ t} is compact for every t; (2) Γ ∩ N ∗ Z/M is compact. Then Theorem 4.4. The Lagrangian intersection: I(N ∗ Z/M , Γ) = I(N ∗ Z/M , [M ]). Proof. As a cycle, the cycle N ∗ in [11], Z/M is Lagrangian. So from Theorem 9.5.3 and 9.7.11 I(Γ, N ∗ Z/M ) = I(N ∗ Z/M , [M ]). (cid:3) / /   / /     / / 14 JIANG 5. Application to Donaldson-Thomas invariants 5.1. Symmetric obstruction theory. The notion of symmetric perfect obstruc- tion theory was defined by Behrend in [2], and its main application is to Donaldson- Thomas theory. The Donaldson-Thomas type moduli space of stable simple com- plexes over a smooth Calabi-Yau threefold Y admits a symmetry obstruction theory. Let X be a DM stack and EX an object in the derived category Db(Coh(X)) of coherent sheaves on X with amplitude contained in [−1, 0]. From [3] and [12], EX is a perfect obstruction theory on X if there exists a morphism ϕ : EX → LX in the derived category of coherent sheaves on X such that h0(ϕ) is an isomorphism and h−1(ϕ) is surjective. Here LX is the truncated cotangent complex of X. From [3], the intrinsic normal cone cX is a subcone stack inside the cone stack h1/h0(L∨ X ). → E0] of EX , which We assume that there exists a global resolution E = [E−1 φ means there is a quasi-isomorphism in the derived category Db(Coh(X)) of coherent sheaves on X and Ei are vector bundles for i = −1, 0. Let E → EX E∨ = [φ∨ : E0 → E1] be the dual of of E, where E0 = (E0)∗, and E1 = (E−1)∗. The perfect obstruction theory gives a closed immersion of cone stacks h1/h0(L∨ X ) = [E1/E0]. The latter Artin stack h1/h0(EX )∨ = [E1/E0] is called the bundle stack and the intrinsic normal cone cX is a subcone stack in it. X ) ֒→ h1/h0(E∨ The perfect obstruction theory EX is symmetric if there is a bilinear form Θ : EX ∼=→ E∨ X [1] which is non-degenerate and symmetric. Hence this implies that rk EX = rk(E∨ − rk E∨ X = − rk EX , so rk EX = 0. The obstruction sheaf ob = h1(E∨ X [1]) = h0(EX ) = ΩX . h0(E∨ X [1]) = X ) = We fix an embedding i : X ֒→ M into a smooth DM stack M . Then Φ : ΩMX → ob = ΩX is an epimorphism of coherent sheaves. Let cv be the coarse moduli space of the intrinsic normal cone cX . Then there is a Cartesian diagram C cv ΩMX / ob . The cone C is called the obstruction cone of the symmetric perfect obstruction theory EX . Proposition 5.1. (Theorem 4.9 in [2]) The cone C ⊂ ΩM is conic Lagrangian. Definition 5.2. The virtual fundamental class [X]vir is given by [X]vir = 0! ΩM ([C]) ∈ A0(X). / /     / NOTE ON MACPHERSON'S LOCAL EULER OBSTRUCTION 15 The Donaldson-Thomas type invariant of X is defined by #vir(X) =Z[X]vir 1 when X is proper. 5.2. Weighted Euler characteristic. The DM stack X has a canonical integral cycle cX ∈ Z∗(X) in Section 1.1 of [2]. This cycle is defined as follows: on a local chart U → X, which is ´etale, and an embedding U ֒→ M , the cycle is cXU =Xi (−1)dim(π(Ci)) · mult(Ci) · π(Ci), where the sum is over all irreducible components Ci of the normal cone CU/M , and π : CU/M → U is the projection. The set π(Ci) is an irreducible closed subscheme or substack in U and mult(Ci) is the multiplicity of the component Ci at the generic point. Behrend proves that cXU is independent to the embedding and these local data glue to give the canonical integral cycle cX . Definition 5.3. The Behrend function νX is defined by: which is the local Euler obstruction of the cycle cX . νX = Eu(cX ), Look at Diagram (4.1), when applying the degree zero Chern-Mather class to the Behrend function we get cM 0 (νX ) ∈ A0(X) and also Ch(νX ) = C. From the above section, the virtual fundamental class [X]vir is the Lagrangian intersection of the cone C with the zero section of ΩM . Hence Behrend proves that (5.1) So: [X]vir = cM 0 (νX ). Theorem 5.4. If X is proper and admits a symmetric obstruction theory EX , then χ(X, νX ) =Z[X]vir 1. This was proved in Theorem 4.18 of [2]. 5.3. C∗-equivariant obstruction theory. In this section we assume that there is a C∗ action on the scheme or DM stack X. We talk about C∗-equivariant obstruction theory on X. The general G-equivariant obstruction theory and G- equivariant symmetric obstruction theory for an algebraic group G has been dis- cussed by Behrend and Fantechi in [4]. Let Db(Coh(X))C∗ be the derived category of C∗ equivariant coherent sheaves over X. From Section 2.2 in [4], Definition 5.5. A C∗-equivariant perfect obstruction theory on X is a morphism EX → LX in the category Db(Coh(X))C∗ . As mentioned by Behrend, this is origi- nally sue to Graber-Pandharipande. A C∗-symmetric equivariant obstruction theory is given by (EX → LX , Θ : , such that EX → LX is an equivari- X [1] is an isomorphism satisfying EX → E∨ ant perfect obstruction theory, and Θ : EX → E∨ Θ∨[1] = Θ. X [1]) of morphisms in Db(Coh(X))C∗ 16 JIANG We include an example of Behrend here so that an equivariant symmetric ob- struction theory locally always has the following form. → ΩMX ] gives the symmetric obstruction theory EX . theory on X locally is given by an almost closed one form. Let ω =Pn Example 5.6. In Section 3.4 of [2], Behrend proves that a symmetric obstruction i=1 fidxi be an almost one form on M ∼= An so that X = Z(ω) is the zero locus of ω. Then H(ω) = [TMX Let C∗ act on An by setting the degree of xi to be ri ∈ Z. Each fi is homogeneous with respect to these degrees and denote the degree of fi by ni ∈ Z. Then the zero locus X admits a C∗ action. If we let C∗ act on the tangent TM by letting the degree to be ni, then H(ω) → LX defines a C∗ perfect equivariant obstruction of ∂xi theory. If ni = −ri, then the form ω is an invariant element of Γ(M, ΩM ). Then we have an "equivariant symmetric" obstruction theory. ▽ω ∂ The following result is due to Proposition 2.6 of Behrend-Fantechi [4]. Proposition 5.7. Let X be an affine C∗-scheme with a fixed point P and let n = dim TXP . If X is endowed with a symmetric equivariant obstruction theory EX → LX . Then there exists an invariant affine open neighbourhood U of P in X, an equivariant closed embedding U ֒→ M into a smooth C∗-scheme M of dimension n and an invariant almost closed one form ω on M such that X = Z(ω). We actually don't need X to admit an equivariant "symmetric" obstruction theory, only a C∗-equivariant obstruction theory EX → LX . Theorem 5.8. Let X be a scheme which admits a symmetric obstruction theory EX . Furthermore assume that there is a C∗ action on X with fixed point scheme F ⊂ X, such that EX → LX is a C∗-equivariant obstruction theory. Then χ(X, νX ) = χ(F, νXF ), where νXF is the restriction of νX to F . Moreover if X is proper, thenR[X]vir 1 = χ(X, νX ) = χ(F, νXF ). Proof. From the property of the Behrend function, νX is constant on the nontriv- ial C∗ orbits, and the Euler characteristic of a C∗-scheme without fixed points is zero. So the only contribution comes from the fixed point locus and χ(X, νX ) = If X is proper, the last statement is Behrend's theorem Theorem χ(F, νXF ). 5.4. (cid:3) 5.4. Kiem-Li construction. We denote the C∗ action on X by: Let λ be the parameter of the group C∗ and consider the following vector field µ : C∗ × X → X. given by v = d fixed point locus. The vector field defines a cosection dλ (µ(λ · x))λ=1. The zero locus of such vector field is F ⊂ X, the v : X → TX by taking dual in the sense of [7]. The degenerate locus D(σ) is exactly the fixed point locus F ⊂ X. σ : ΩX → OX NOTE ON MACPHERSON'S LOCAL EULER OBSTRUCTION 17 The obstruction theory EX defines a bundle stack E := h1/h0((EX )∨) in the sense of Behrend-Fantechi in Section 2 of [3], such that h1(E) = ΩX . Let U := X\F . Then U is an open subset of X. The cosection σ gives a surjective morphism (5.2) This surjective morphism (5.2) induces a morphism from the bundle stack EU to CU . Let ΩXU = h1((EX )∨)U → OU . E(σ) := EF ∪ ker(EU → CU ). Let cX be the intrinsic normal cone of X. As proved in [7], the intrinsic normal cone cX ⊂ E(σ). Applying the localized Gysin map in Section 3 of [7], we get the zero dimensional localized virtual cycle s! E,σ([cX ]) = [F ]vir loc,KL ∈ A0(F ). Let ι : F → X be the inclusion. This cycle satisfies ι∗([F ]vir projective. loc,KL) = [X]vir if X is Definition 5.9. If the scheme or DM stack F is proper, then the virtual localized invariant of X is defined by Z[F ]vir loc,KL 1. 5.5. Cone decomposition. Recall that we fix an embedding i : X ֒→ M of X to a smooth scheme M . Hence there is a surjective morphism: φ : ΩMX → ΩX . From Section 5.1, the obstruction cone C ⊂ ΩMX is a conic Lagrangian cycle in ΩM , which is proved in Theorem 4.9 of [2]. Definition 5.10. Define and as cycles. C2 := CX\F . [C1] := [C] − [C2] Remark 5.11. The cone C1 is supported on the fixed point locus F of X. 5.6. Kiem-Li localized invariant via bundle stack. First we prove the follow- ing: Proposition 5.12. The surjective morphism φ : ΩMX → ΩX induces a surjective morphism Φ : ΩMX → E e∗EX ∼= [F −1 → F 0] to the bundle stack E. Proof. Let e : X → X be an ´etale open covering. The the pullback has a locally free resolution, and F −1, F 0 are locally free coherent sheaves. Then e∗E∨ X ∼= [F0 → F1], with F0 = (F 0)∨, F1 = (F −1)∨ and e∗E = [F1/F0]. Hence we have the following exact sequence Suppose that the ranks of F0 and F1 are r0 and r1. Let g be the number of generators of the sheaf ob on one local chart around a point P . F0 → F1 → e∗ ob → 0. 18 JIANG Locally we can take r1 = g without loss of generality. (This is because the rank of the first arrow is r1 − g at P , so we may choose a rank r1 − g subbundle S on which the map is of full rank (possibly after shrinking our open set). Then dividing out by the acyclic complex S → S we get the claim.) Now our vector bundle ΩMX surjects onto ob. Since it is free, we can lift to a map ΩMX → F1. Then the composition ΩMX → F1 → ob is onto, so in particular the first map must have rank ≥ g, so it must be onto, so ΩMX → F1 is onto in the local chart and e∗ΩMX maps onto the stack e∗E = [F1/F0]. The construction is canonical so it induces a surjective morphism Φ : ΩMX → E. (cid:3) We construct the following diagram: (5.3) ΩMX Φ E #❋❋❋❋❋❋❋❋ σ !❈❈❈❈❈❈❈❈❈ / OX σ ΩX where σ is the cosection map. To simplify notation, we denote by V := ΩMX . Let V (σ) := V F ⊕ ker(V U=X\F → OXU ). The Lagrangian cone C lies in V (σ). Kiem-Li's localized virtual cycle [F ]vir given by the localized Gysin map KL is (5.4) V (σ),σ([C]) = [F ]vir s! loc,KL constructed in Section 3 of [7]. The following result is easily from the property of localized Gysin map. Proposition 5.13. V (σ),σ([C]) = s! s! V (σ),σ([C1]) + s! V (σ),σ([C2]). 5.7. Kiem-Li localized invariant via Lagrangian intersection. 5.7.1. Type one cone: The cone C1 inside ΩMX supports on the fixed point loci F ⊂ X. So C1 lies inside V F . The scheme F is closed and proper. Hence from Kiem-Li [7], the localized Gysin map is just the usual Gysin map and (5.5) V (σ),σ([C1]) = s! s! V F ([C1]). 5.7.2. Type two cone: Each irreducible component of the cone C2 supports on a C∗-invariant subscheme of X. Using idea in the Appendix in [7], we prove that the localized Gysin map s! V (σ),σ([C2]) is still a Lagrangian intersection of two La- grangian cycles inside ΩMX . vector bundle ΩM , and choose a smooth function We use a similar idea as in Section 4 of [6]. Choose a Hermitian metric on the ψ : M → R such that {x ∈ M : ψ(x) ≤ t} is compact for t ∈ R. We also choose a smooth function ρ : R → R≥0 such that ρ(r) =(0, ǫr, r ≤ 1; r ≥ 2 / / #   ! / NOTE ON MACPHERSON'S LOCAL EULER OBSTRUCTION 19 Let Γ := Γ(dρψ) be the graph of the differential of the function ρψ. Then Γ is a small perturbation of the zero section of ΩM . The graph Γ is a real Lagrangian cycle inside ΩM . From Theorem 9.5.3 and Theorem 9.7.11 of [11], the Lagrangian intersection I(C2, Γ) makes sense. Theorem 5.14. The Lagrangian intersection C2 ∩ Γ supports on the fixed point loci F ⊂ X and the intersection number I(C2, Γ) is the same as Kiem-Li localized invariant s! V (σ),σ([C2]). Proof. We can choose the smooth function ψ : M → R such that ψ is nonzero on X \ F . The composition ΩMX → ΩX → OX is the contraction with the Euler vector field associated with the C∗ action. When restricting to the graph Γ, this gives a function ǫψ outside the neighbourhood {ψ ≤ 2}. This is nonzero, hence the intersection of C2 with Γ can not support on the kernel of ΩMX → OX of the cosection on the locus X \ F . But the cone C2 lies in the kernel ΩMX → OX . Hence C2 ∩ Γ must support on the locus F ⊂ X. V (σ),σ([C2]) is a global result of Proposition A.1 of the Appendix in [7]. (cid:3) The intersection number I(C2, Γ) = s! 5.8. Behrend VS Kiem-Li. Recall that in Section 5.2 there is a canonical integral cycle cX for the scheme or DM stack X. The cycle cX has a decomposition: cX = c1 + c2, where c2 = cXX\F and c1 = cX \ c2. By the uniqueness of the integral cycle cX , the integral cycles c1 and c2 are unique. Recall that the obstruction cone C ⊂ ΩMX is a conic Lagrangian cycle and we have the cone decomposition C = C1 + C2 in Definition 5.10. Let LX (ΩM ) be the subgroup of Zn(ΩM ) generated by conic Lagrangian prime cycles which support on X, where n = dim(M ). Proposition 5.15. We have: L(c1) = C1, L(c2) = C2. Proof. The subgroup LX (ΩM ) is generated by conic Lagrangian prime cycles, and from [9], also [2], every irreducible conic Lagrangian cycle in ΩMX is of the form N ∗ Z/M . Our cones C1, C2 are conic Lagrangian in ΩMX . So each irreducible com- ponent of C1 (resp. C2) is of the form N ∗ Z/M is supported on Z ⊂ X \ F , and in the case of C1, N ∗ Z/M is supported on Z ⊂ M . Thus the results follows from the isomorphisms in (4.2) and (4.3). Z/M , where in the case of C2, N ∗ (cid:3) Definition 5.16. Define ν1 = Eu(c1), ν2 = Eu(c2). Remark 5.17. From the property of Euler obstruction, Eu(cN ) = Eu(c1) + Eu(c2). Hence the Behrend function νX = ν1 + ν2. 20 JIANG Theorem 5.18. We have: χ(F, ν1F ) = s! V (σ),σ([C1]). Proof. The Lagrangian cycle [C1] only supports on F . From (5.5), s! V (σ),σ([C1]) = s! ([C1]), the usual Lagrangian intersection of C1 with the zero section of V F . V F In the case that M is a scheme, or a finite group quotient stack, or a finite gerbe over a scheme, this is Behrend's Proposition 4.6 in [2]. In the DM stack case, it is due to Maulik and Treumann in [15], where the main result is the generalization of Kashiwara index theorem to orbifolds. (cid:3) Recall our Lagrangian cycle Γ ⊂ ΩM , which is a small perturbation of the zero section. Theorem 5.19. We have: χ(F, ν2F ) = I(C2, Γ). Proof. In the case that M is a scheme, this is the global index theorem due to Theorem 9.7.11 of Kashiwara-Schapira [11] since the characteristic cycle of ν2 is C2. In the DM stack case, it is still due to Maulik and Treumann in [15]. (cid:3) So we finally have: Theorem 5.20. If the C∗ fixed locus F is proper, then we have: χ(F, νXF ) =Z[F ]vir loc,KL 1. Proof. Combining Theorem 5.14, Theorem 5.18 and Theorem 5.19, the result fol- lows. (cid:3) Remark 5.21. Actually to prove this main result we don't need to split cones as in the proof in Section 4 in [6]. To make things clearer, we do it here. 5.9. C∗-action with Isolated fixed points. In this section we assume that the C∗-action on the scheme X has isolated fixed point set. In Theorem 3.4 of [3], Behrend and Fantechi proved the following result: Proposition 5.22. Let X be a scheme with a C∗ action so that (EX → LX , Θ) is a C∗-equivariant symmetric obstruction theory. Suppose that the C∗ action only has isolated fixed point set F . Let P ∈ F . Then νX (P ) = (−1)dim(TX P ), where TXP is the tangent space of X at P . Proof. The formula in the theorem is a local issue question. We prove the result using Theorem 5.20. The symmetric obstruction theory EX = [E−1 → E0] is given by a global resolution. We may use C∗ equivariant embedding P ֒→ X ⊂ M, where we assume that the dimension of M is the same as dim(TXP ). Using defor- mation to normal cone CX/M and the deformation invariance of virtual class, we may use the zero section [M ] to do the Kiem-Li localization. So in this case we have a vector bundle ΩM and a cosection map σ : ΩM → OM such that σ−1(0) = P . Then the result comes from Example 2.4 in [7]. (cid:3) NOTE ON MACPHERSON'S LOCAL EULER OBSTRUCTION 21 References [1] Paolo Aluffi, Limits of Chow groups, and a new construction of Chern-Schwartz-MacPherson classes, Pure Appl. Math. Q. 2 (2006), no. 4, 915-941. [2] Kai Behrend, Donaldson-Thomas type invariant via microlocal geometry, Ann. Math. 170 (2009), no. 3, 1307-1338. [3] Kai Behrend and Barbara Fantechi, The intrinsic normal cone, Invent. Math. (1997), 1337– 1398. [4] , Symmetric obstruction theories and Hilbert scheme of point on threefolds, Algebra Number Theory 2 (2008), no. 3, 313-345. [5] Yunfeng Jiang, The Pro-Chern-Schwartz-MacPherson class for DM stacks, Pure and Applied Mathematics Quarterly 11 (2015), no. 1, 87-114, arXiv:1412.3724. [6] Yunfeng Jiang and Richard P Thomas, Virtual signed Euler characteristics, Journal of Al- gebraic Geometry 26 ((2017)), 379-397. [7] Young-Hoon Kiem and Jun Li, Localizing virtual cycles by cosection, J. Amer. Math. Soc. 26 (2013), 1025-1050. [8] William Fulton, Intersection theory, 2nd ed., Ergebnisse der Mathematik und ihrer Grenzge- biete. 3. Folge (1998), A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics], vol. 2, Springer- Verlag, Berlin, 1998. [9] Victor Ginzburg, Characteristic varieties and vanishing cycles, Invent. Math. 84 (1986), 327-402. [10] Gary Kennedy, Macpherson's chern classes of singular algebraic varieties, Communications in Algebra 18:9 (1990), 2821-2839. [11] Masaki Kashiwara and Pierre Schapira, Sheaves on manifolds, Grundlehren der Mathema- tischen Wissenschaften. 292 (Verlag, Berlin, 1990 Springer). [12] Jun Li and Gang Tian, The virtual fundamental class for algebraic varieties, J. Amer. Math. Soc. (1998). [13] Weiping Li and Zhenbo Qin, Donaldson-Thomas invariants of certain Calabi-Yau 3-folds, Commu. in Anal. and Geom. 21 (2013), 541-578. [14] Robert MacPherson, Chern class for singular algebraic varieties, Ann. Math. 100 (1974), no. 2, 423-432. [15] Devash Maulik and David Treumann, Constructible functions and Lagrangian cycles on orb- ifolds, arXiv:1110.3866. [16] G. Gonzalez-Sprinberg, L'obstruction locale d'Euler et le theoreme de MacPherson, Aster- isque 82-83 (1981), 7-32. [17] Richard P Thomas, A holomorphic Casson invariant for Calabi-Yau 3-folds, and bundles on K3 fibrations, J. Differential Geom. 54 (2000), 367-438. Department of Mathematics, University of Kansas, 405 Snow Hall, 1460 Jayhawk Blvd, Lawrence 66045, USA E-mail address: [email protected]
1202.3958
4
1202
2012-10-03T07:28:49
The projective translation equation and unramified 2-dimensional flows with rational vector fields
[ "math.AG", "math.AP", "math.NT" ]
Let X=(x,y). Previously we have found all rational solutions of the 2-dimensional projective translation equation, or PrTE, (1-z)f(X)=f(f(Xz)(1-z)/z); here f(X)=(u(x,y),v(x,y)) is a pair of two (real or complex) functions. Solutions of this functional equation are called projective flows. A vector field of a rational flow is a pair of 2-homogenic rational functions. On the other hand, only special pairs of 2-homogenic rational functions give rise to rational flows. In this paper we are interested in all non-singular (satisfying the boundary condition) and unramified (without branching points, i.e. single-valued functions in C^2\{union of curves}) projective flows whose vector field is still rational. We prove that, up to conjugation with 1-homogenic birational plane transformation, these are of 6 types: 1) the identity flow; 2) one flow for each non-negative integer N - these flows are rational of level N; 3) the level 1 exponential flow, which is also conjugate to the level 1 tangent flow; 4) the level 3 flow expressable in terms of Dixonian (equianharmonic) elliptic functions; 5) the level 4 flow expressable in terms of lemniscatic elliptic functions; 6) the level 6 flow expressable in terms of Dixonian elliptic functions again. This reveals another aspect of the PrTE: in the latter four cases this equation is equivalent and provides a uniform framework to addition formulas for exponential, tangent, or special elliptic functions (also addition formulas for polynomials and the logarithm, though the latter appears only in branched flows). Moreover, the PrTE turns out to have a connection with Polya-Eggenberger urn models. Another purpose of this study is expository, and we provide the list of open problems and directions in the theory of PrTE; for example, we define the notion of quasi-rational projective flows which includes curves of arbitrary genus.
math.AG
math
THE PROJECTIVE TRANSLATION EQUATION AND UNRAMIFIED 2-DIMENSIONAL FLOWS WITH RATIONAL VECTOR FIELDS GIEDRIUS ALKAUSKAS Abstract. Let x = (x, y). Previously we have found all rational solutions of the 2-dimensional projective translation equation, or PrTE, (1 − z)φ(x) = φ(φ(xz)(1 − z)/z); here φ(x) = (u(x, y), v(x, y)) is a pair of two (real or complex) functions. Solutions of this functional equa- tion are called projective flows. A vector field of a rational flow is a pair of 2-homogenic rational functions. On the other hand, only special pairs of 2-homogenic rational functions give rise to rational flows. In this paper we are interested in all non-singular (satisfying the boundary condition) and unramified (without branching points, i.e. single-valued functions in C2\{union of curves}) projective flows whose vector field is still rational. If an orbit of the flow is given by homogeneous rational function of degree N , then N is called the level of the flow. We prove that, up to conjugation with 1-homogenic birational plane transformation, these are of 6 types: 1) the identity flow; 2) one flow for each non-negative integer N - these flows are rational of level N ; 3) the level 1 exponential flow, which is also conjugate to the level 1 tangent flow; 4) the level 3 flow expressable in terms of Dixonian (equianharmonic) elliptic functions; 5) the level 4 flow expressable in terms of lemniscatic elliptic functions; 6) the level 6 flow express- able in terms of Dixonian elliptic functions again. This reveals another aspect of the PrTE: in the latter four cases this equation is equivalent and provides a uniform framework to addition formulas for exponential, tangent, or special elliptic functions (also addition formulas for poly- nomials and the logarithm, though the latter appears only in branched flows). Moreover, the PrTE turns out to have a connection with P´olya-Eggenberger urn models. Another purpose of this study is expository, and we provide the list of open problems and directions in the theory of PrTE; for example, we define the notion of quasi-rational projective flows which includes curves of arbitrary genus. This study, though seemingly analytic, is in fact algebraic and our method reduces to algebraic transformations in various quotient rings of rational function fields. Contents Introduction 1. 2. Auxiliary functions 3. The results 4. The proofs. Further properties 5. Open directions, further advances References 2 7 10 19 26 33 Date: September 15, 2012. 2010 Mathematics Subject Classification. Primary 39B12, 33E05, 35F05; Secondary 14H52, 14H05, 14E05. Key words and phrases. Projective translation equation, flows, rational vector fields, iterative functional equation, elliptic curves, elliptic functions, Dixonian elliptic functions, linear PDE's, finite group representations, hypergeometric functions. The author gratefully acknowledges support by the Lithuanian Science Council whose postdoctoral fellowship is being funded by European Union Structural Funds project "Postdoctoral Fellowship Implementation in Lithuania". 1 Projective translation equation 2 1. Introduction For convenience reasons, we write F (x, y) • G(x, y) instead of (cid:0)F (x, y), G(x, y)(cid:1). We also write x = x • y. This paper is a continuation of [Al10, Al12]. It is completely independent from the first, and also mostly independent from the second paper, apart from the ([Al12], Section 4, steps I and III), which are crucial for the current work. The main needed steps will be summarized here in the Subsection 1.3. The current paper is an introduction to the prospective paper [Alpr2] where general plane vector fields are treated in the framework of quasi-rational flows; see the Subsection 5.4. 1.1. Background. The general "affine" translation equation is a functional equation of the type F (x, s + t) = F (F (x, s), t) x ∈ Cn, s, t ∈ C, where F : Cn × C 7→ Cn. Note that this is not the most general form of the translation equa- tion; to get the feeling of the variety of structure and methods that underlies this equation, the reader may consult [Ac66, Mo73, Mo95, FR08, FR10]. Our research concentrates on the special case of this equation, where F (x, t) is of the form φ(xt)t−1; this choice is not accidental, since it exhibits several fascinating features not encountered in the general affine solutions ("affine flows"). This special case was first introduced in [Al10] where we considered the 2-dimensional equation from the topologic point of view in the simplest case of the sphere S2. In the paper [Al12] we solved the following problem: find all rational solutions of the 2- dimensional projective translation equation (PrTE for short) (1 − z)φ(x) = φ(cid:16)φ(xz) 1 − z z (cid:17). (1) Here φ(x, y) = u(x, y) • v(x, y) is a pair of rational functions in two real or complex variables. It appears that, up to conjugation with a 1-homogenic birational plane transformation (1-BIR in short, for the structure of these see the Appendix in [Al12]), all rational solutions of this equation are as follows: the zero flow 0 • 0, two singular flows 0 • y and 0 • y y+1 , an identity flow x • y, and one non-singular flow for each non-negative integer N, called the level of the flow; this solution is given by φN below. The non-singular solution of this equation is called the projective flow. Non-singularity means that a flow satisfies the boundary condition φ(xz) z lim z→0 = x. (2) Thus, each rational projective flow u(x, y)• v(x, y) is automatically a birational plane transfor- mation whose inverse is given by −u(−x,−y) • −v(−x,−y). The main object associated with a solution is its vector field given by v(φ; x, y) = (x, y) • (x, y) = d dz φ(xz, yz) z . (3) Vector field is necessarily a pair of 2-homogenic functions. For smooth functions, the functional equation (1) implies the PDE (cid:12)(cid:12)(cid:12)z=0 ux(x, y)((x, y) − x) + uy(x, y)((x, y) − y) = −u(x, y), (4) 3 G. Alkauskas and the same PDE for v, with boundary conditions u(xz, yz) z lim z→0 = x, lim z→0 v(xz, yz) z = y. In principle, the PDE (4) with the above boundary conditions is equivalent to (1); but the proof of this equivalence is possible only in each case separately, since there arises the complication to determine the definition domain of the flow - for example, explore φlog below. Thus, not all functions in two complex variables can be iterated without restrictions - this restriction is explicitly present in (1), but only implicitly present in (4), and only a posteriori, after we have found the solution, whether in analytic form in terms of known special functions, or via an independent analysis. Each point under a flow possesses the orbit, which is either a curve (when we deal with flows over R) or a surface (complex curve over C), or a single point; the orbit is defined by V (x) =n φ(xz) z : z ∈ R or Co. For rational flows of level N, there exists an Nth degree homogenic rational function W (x, y) such that the orbits are the curves W (x, y) = c, c ∈ C∪{∞}. We will also refer to non-rational flows as also being of level N if the orbits are given by Nth degree homogeneous rational function. It appears that the vector field of a rational flow is a pair of 2-homogenic rational functions. The main theorem in [Al12] gives the first structural result for the rational flows; many more properties of the flows which take into account not only the level but the birational conjugation itself will be treated in [Alpr1]. The main idea of the proof (which is rather lengthy) is that using conjugation with 1-BIR, the vector field can be reduced step by step, leading to rational 2-homogenic functions with numerators of smaller degree, and eventually, provided we do not encounter an obstruction, to quadratic forms. The obstruction arises when we hit the vector field whose both coordinates are proportional. So, we are left to find all rational flows whose vector field is given by a pair of two quadratic forms in two variables, or whose one of the coordinates vanish (this is equivalent, after a linear conjugation, to the obstruction). The first case is a rewarding part of the proof. For example, we find that some of these pairs of quadratic forms lead to rational solutions, while others should be discarded since they arise from non-rational flows. In this paper we take a closer look at non-rational solutions of (1) whose vector field is a pair of quadratic forms. In general, every vector field whose orbit is given by a homogeneous rational function of degree N ∈ N0 gives rise, generally, not to a rational flow but to a quasi-rational flow; see the Subsection 5.4 and [Alpr2]. 1.2. Addition formulas. Motivation. Consider the following two-variable two-dimensional functions: φexp(x) = xey • y φtan(x) = xy + y2 tan y y − x tan y • y (y + 1)[y + x log(y + 1)] • φN (x) = x(y + 1)N−1 • xy φlog(x) = y y + 1 = xf (y) • y; = xy + y2f (y) y − xf (y) • y; = xf (y) • = y ; y + 1 xy y y + 1 (y + 1)[y + xf (y)] • y y + 1 . Projective translation equation 4 (Here in each case f (y) stands for the corresponding function, ey, tan(y), and so on). We can check directly that these functions satisfy the PrTE (1). The vector fields and the equations for the orbits of these flows are given by xy • 0, exp: tan: N: log: −x2 − xy • (−y2), W (x, y) = exp(yx−1)y. (N − 1)xy • (−y2), W (x, y) = xyN−1; W (x, y) = y; W (x, y) = y; x2 + y2 • 0, (In the last case the orbits are non-algebraic curves). In fact, replace the known function in the expression of φ with the unknown f (this is shown above). Now require that the so obtained function satisfies (1). Eventually, this turns out to be equivalent to nothing else but the standard addition formulas: exp : f (x + y) = f (x)f (y); tan : f (x + y) = f (x) + f (y) 1 − f (x)f (y) ; N : f (xy − 1) = f (x − 1)f (y − 1); log : f (xy − 1) = f (x − 1) + f (y − 1). The first two can be restated in a symmetric way; that is, if u + v + w = 0, then exp u · exp v · exp w = 1, tan u · tan v · tan w = tan u + tan v + tan w. These are superior to the former, since if we know additionally that exp(0) = 1 and tan(0) = 0, then these identities also encode the symmetry properties exp(u) exp(−u) = 1 and tan(u) + tan(−u) = 0. In [Al12] we found out that the k-dimensional PrTE is directly related to bira- tional transformations of P k−1(C), as opposed to general "affine" translation equation which is tied to birational affine transformations of Ck, albeit this dependency is of different nature, the first case being much more involved. Now we see another fascinating feature of (1); namely, the PrTE provides a uniform framework for addition formulas for certain functions: abelian - exp(y), tan(y); algebraic - (y + 1)N−1, if N ∈ Q; integrals over algebraic - log(y + 1). This is the simplest, 2-dimensional case of the theory. As we will see in Theorem 1, the above list will not include trigonometric functions sin(y) and cos(y), but will include rather special elliptic functions related to regular hexagonal and square lattices. The higher dimensional case even in the setting of [Al12] (that is, classification of higher dimensional rational projective flows) is open and very promising; see the Subsection 5.1. Now we formulate the main problem of this paper. Problem 1. Find all flows, that is, bivariate functions satisfying (1), which are defined for x ∈ C2 \ {union of isolated curves}, whose vector field is rational, and which are single-valued functions, i.e. without branching points. Thus, the flow φlog has an infinite branching point at y = −1. The flow φN has an infinite branching if N /∈ Q and a finite one if N ∈ Q\ Z. So, our chief interest is only the case N ∈ Z, and this was dealt with in [Al12]. On the other hand, the flows φexp and φtan are single-valued functions for all x ∈ C2. With the help of linear conjugation we can force the exponential and tangent flows to be symmetric with respect to the linear involution (x, y) 7→ (y, x). Thus, we summarize this as 5 G. Alkauskas 1 2(cid:0)(x − y)ex+y + x + y(cid:1), and (x + y)x + (x + y)y tan(x + y) x + y + (y − x) tan(x + y) . Proposition 1. Let φe = ue(x, y) • ue(y, x), φt = ut(x, y) • ut(y, x), ue(x, y) = ut(x, y) = These functions are symmetric projective flows. Both ue and ut satisfy the PDE (4), where (x, y) = 1 2(x2 − y2) = (y, x) in the exponential case, and (x, y) = x2 + y2 = (y, x) in the tangential case, respectively. The projective flow property is equivalent to addition formulas for corresponding functions. Note however that these two cases are conjugate and will fall under the same item in the classification (see Theorem 1). In fact, let us define the 1-BIR by ℓ(x, y) = x(x + y) y • (x + y). Then ℓ−1 ◦ φexp ◦ ℓ(x, y) = x(x + y)ex+y xex+y + y y(x + y) xex+y + y , • and the latter projective flow is linearly conjugate (over C) to φtan. 1.3. Quadratic forms as vector fields of projective flows. We will see later that along with (y + 1)N−1, tan(y) and exp(y), there are three more pairs of functions which complete the picture. As a crucial part of our investigations, let us make a brief summary of the Step II of the proof of the main theorem in [Al12]. Let the vector field of the projective flow φ = u • v be given by (x, y) • (x, y), where both coordinates are 2-homogenic rational functions, and the common denominator has a degree d ≥ 1. We found that, unless and are proportional, there exists a 1-BIR ℓ such that the vector field of the flow ℓ−1 ◦ φ ◦ ℓ is a pair of rational functions with lowered degree in the common denominator. Thus, we are left to consider cases where either and are proportional, or they both are quadratic forms. If the first statement holds, then we see that (after a linear conjugation) one can confine to the case (x, y) ≡ 0. This implies v(x, y) = y. If φ were a rational function, then this would necessarily mean that u is a Jonqui`eres transformation; this would imply that is in fact a quadratic form, and calculations in ([Al12], Step II) provide the complete solution. The flows φexp and φtan arise exactly from this analysis in cases (x, y) = xy and (x, y) = x2 + y2, respectively. In the setting of the Problem 1, the pair u(x, y) • y with the vector field (x, y) • 0 ought no longer be rational, the implication that u is a Jonqui`eres transformation is irrelevant, and we need to provide an independent analysis of the solution of (4) in case is any 2-homogenic rational function, and ≡ 0. This is accomplished in the Subsection 4.4. Suppose now that both and are quadratic forms, see ([Al12], Step III). We found that if x− y is a cube of a linear polynomial, then this always leads to the flow with a ramification of the type log(y + 1). If x − y is not a cube, then the vector field, with the help of linear conjugation, can be transformed into (x, y) = ax2 + bxy, (x, y) = cxy + dy2. Projective translation equation 6 The cases b = 0 or c = 0 lead to rational solutions or flows with ramification of the type (y +1)r, r /∈ Z. So, let b, c 6= 0. If d = 0 or a = 0, this again leads to a ramification of the type log(y +1). So, after a linear conjugation, we may assume that (x, y) = x2 + Bxy, (x, y) = Cxy + y2. The flow with this vector field has ramifications of both types (1 − y)−B and (1 − x)−C, so, if there is none, we get the arithmetic condition If B = 1, then again, C = 1, otherwise the flow is ramified, and B = C = 1 gives exactly the rational flow of level 0. Assume B, C 6= 0, 1. Then the vector field • with the help of linear conjugation can be transformed into the vector field Since this is also unramified, we get another arithmetic condition x2 + B + C − 2 BC − 1 xy • Cxy + y2. (5) (6) (7) (8) (9) B, C ∈ Z. B + C − 2 BC − 1 ∈ Z. The case B + C = 2 leads to a rational or algebraically ramified flow. It is fascinating that if B, C 6= 0, 1, and B + C 6= 2, then all pairs (B, C) which satisfy as simple arithmetic conditions as (5) and (7) are encoding elliptic unramified flows! More precisely, there are exactly 10 such pairs (standard exercise): (−2,−1) ↔ (−5,−1), (−1,−3) ↔ (−3,−3), (−2,−2) (cid:9) . (−1,−2) ↔ (−5,−2), (−3,−1) (cid:9); (−2,−5) ↔ (−1,−5); The symbol " ↔ " means that the two pairs a linearly conjugate via (6), and " (cid:9) " means that the flow is self conjugate. The pairs (B, C) and (C, B) are also linearly conjugate with the help of the involution i(x, y) = (y, x). So, there are three equivalence classes of flows (shown as rows above), consisting of 6, 3 and 1 pairs respectively; in each class any two flows are linearly conjugate, and we will show that all three arise form elliptic flows. We will choose such representatives: (B, C) = (−2,−2), (B, C) = (−3,−3), and (B, C) = (−1,−2). Most of this paper deals with the first case, the fascinating vector field which is the class on its own and thus it has exactly the 6-fold symmetry: for every γ ∈ Σ, (x, y) = x2 − 2xy, (x, y) = y2 − 2xy, γ−1 ◦ (, ) ◦ γ = (, ); for the definitions, see the property (SYMM), Subsection 3.2. In the setting of [Al12], all the tricks which ruled out other vector fields as arising from non-rational flows (as a rule, these tricks constituted in showing that corresponding flows have branching points, and rational flows, obviously, cannot have these) were not applicable in all these exceptional cases of (B, C), and it was still not clear why the solution of (4) in case (8), for example, which is exactly the function λ(x, y) (see the Subsection 3.2), cannot be a rational function. And it appears that it is not; since, if we put f (z) = λ(z,−z)/z, then, as the property (ELL), Subsection 3.2, implies, one has f (z)f (−z)[f (z) + f (−z)] ≡ 2, 7 G. Alkauskas and thus (f (z), f (−z)) parametrizes the elliptic curve XY (X + Y ) = 2 and f (z) cannot be In this case λ(x, y) is not rational itself. Two other vector fields with a rational function. (B, C) = (−3,−3) and (B, C) = (−1,−2), whose orbits are also elliptic curves, are examined in the Subsection 4.5. 2. Auxiliary functions Now we make an interlude and introduce functions which will be crucial in our study of the vector field (8). The material is just a collection of various facts from the literature; our contribution to this topic is an introduction of special elliptic functions sp(u) and cp(u) for which we prove Proposition 3. 2.1. The special hypergeometric series. Let us introduce our first auxiliary function [BAT] W (x) = 1 3 1Z0 dt [(1 − t)(1 − xt)]2/3 = 2F1(cid:16)2 3 , 1; 4 3 ; x(cid:17), −∞ < x < 1. It satisfies W (0) = 1, and the linear ODE The derivative of this gives the second order ODE 3x(1 − x)W ′(x) + (1 − 2x)W (x) = 1. (10) (11) 3x(1 − x)W ′′(x) + (4 − 8x)W ′(x) − 2W (x) = 0, which coincided with Euler's hypergeometric differential equation for (a, b; c) = ( 2 3). Since c = 2a, the function W (x) is invariant under Pfaff's transformation and thus it satisfies the functional equation [BAT] 3 , 1; 4 W (x) = x < 0; 1 1 − x x − 1(cid:17), W(cid:16) x this can be verified easily using the integral representation (10). Of course, this functional equation is satisfied by all hypergeometric functions of the form 2F1(a, 1; 2a; x). For example, when a = 1, this hypergeometric function reduces to −x−1 log(1 − x). The change of variables 1 − xt = (1 − x)T in (10) gives W (x) = 1 3[x(1 − x)]1/3 1 1−xZ1 dt [t(t − 1)]2/3 , x < 1. (12) The appearance of the symmetric group S3 in our investigations - see (SYMM), the Subsection 3.2 - can be explained from several points of view; here is one of them. The special case of Kummer's theory for hypergeometric series [BAT] is the following fact, which is easily checked in our case: the differential equation (11) has the following three solu- tions: W0(x) = W (x) W1(x) = −W (1 − x) W (x) = 1 x x(cid:17) W(cid:16) 1 ∞ for −∞ < x < 1, for 0 < x < ∞, for x ∈ (−∞, 0) ∪ (1,∞), W ∞ W0(0) = 1; W1(1) = −1; (∞) = 0. Projective translation equation 8 This can be treated as the complete description of the differential equation (11) on the line P 1(R). The three singular points 0, 1,∞ divide this circle into three parts. For each interval (a, b), a, b ∈ {0, 1,∞}, a 6= b, the general solution of (11) in the interval (a, b) with a floating boundary condition is given by pWa(x) + (1 − p)Wb(x), p ∈ R. The symmetric group S3 acts on the set {0, 1,∞} by permutations. This corresponds to the action of Mobius transformations on W0, W1 and W as follows. ∞ f (x) f (x) f (x) 1 x 7→ 7→ −f (1 − x) corresponds to (0 1)(∞), 7→ x(cid:17) corresponds to (0 ∞)(1), f(cid:16) 1 x − 1(cid:17) corresponds to (1 ∞)(0). f(cid:16) x 1 − x 1 The first entry, for instance, means the following: the map under consideration interchanges W0 and W but leaves the function W1 intact. Other two elements of S3 (the last one is the identity) are obtained from the above. So, for example, the cycle (0 1 ∞) = (0 ∞)(1)· (1 ∞)(0) corresponds to the transformation ∞ f (x) 7→ − 1 x x (cid:17). f(cid:16)x − 1 These correspondences should find their analogues in the setting of the Subsection 5.1, at least in case of symmetric groups SN , N ≥ 4. 2.2. The Dixonian elliptic functions. For the general theory of elliptic functions we may refer to the classical book (in Russian) [Ah48]. Let ω = e2πi/3. The functions sm(u), cm(u) were introduced by Dixon [Di90] as a pair of functions which parematrize the Fermat cubic X 3 + Y 3 = 1. These are in fact special elliptic functions satisfying the following Proposition 2. [Di90, FCF05] The Dixonian elliptic functions have these properties sm(0) = 0, cm(0) = 1, sm′(u) = cm2(u), sm(u) cm(u) sm(−u) = − , sm(ωu) = ω sm(u), cm(ωu) = cm(u), sm(u + v) = cm(u + v) = here s1 = sm(u), c1 = cm(u), 1 cm(u) sm3(u) + cm3(u) ≡ 1, cm′(u) = − sm2(u), , cm(−u) = sm(cid:16)π3 3 − u(cid:17) = cm(u), 1 − s2 2 + s2c2 1s2 1 − s3 1s3 2 c1c2 − s1s2(s1c2 2 + s2c2 1) 1 − s3 1s3 2 2c1c2 s1c2 , , s2 = sm(v), c2 = cm(v), 9 G. Alkauskas u4 sm(u) = u − 4 4! u3 cm(u) = 1 − 2 3! 1 + 160 + 40 u7 7! − 20800 u9 u6 6! − 3680 9! u10 10! + 647680 + 880000 u12 12! + · · · , u3 13! + · · · , (−x)1/3 (1 − x)1/3 ; cm(cid:16)x1/3(x − 1)1/3W (x)(cid:17) = (1 − x)1/3 , sm(cid:16)x1/3(x − 1)1/3W (x)(cid:17) = the last holds for −∞ < x < 1. Here the contant π3 (the period, according to the notion of Zagier-Kontsevich) is given by π3 = B(cid:16)1 3 , 1 3(cid:17) = Γ( 1 3)2 Γ( 2 3) = √3 2π 3(cid:17)3 Γ(cid:16) 1 = 5.299916250856+, and Π = π3 3 27 = 5.513701576710+. The last two constants are of great importance in the current paper. The can be numerically calculated by the fast converging series 3 = 8π6 1 − π6 ∞Xn=1 504n5 (−1)ne√3πn − 1! = 820.824437079556+. The full lattice of periods for both sm(u) and cm(u) is given by Zπ3 ⊕ Zπ3ω. Let F = {π3s + π3ωt : s, t ∈ [0, 1)} be the fundamental parallelogram. Consider 9 special points qa,b = aπ3 3 + bπ3ω 3 , 0 ≤ a, b ≤ 2. In terms of the theory of elliptic functions, both sm(u) and cm(u) are order 3 elliptic functions, and so each of them attains every value in C∪{∞} in F exactly thrice, counting multiplicities. The simple poles of both sm(u) and cm(u) are q2,0, q1,1 and q0,2. The function sm(u) has simple zeros at q0,0, q2,1 and q1,2, while the values at q1,0, q0,1 and q2,2 are, respectively, 1, ω, ω2. Likewise, cm(u) has simple zeros at q1,0, q0,1 and q2,2, and the values at q0,0, q1,2 and q2,1 are, respectively, 1 (triple value), ω, ω2. These properties follow from Proposition 2. The authors in [FCF05], for the convenience reasons, introduce the hyperbolic versions of these functions, given by smh(u) = −sm(−u), cmh(u) = cm(−u). These functions parametrize the "Fermat hyperbola" y3 − x3 = 1. For our purposes, we will need yet another version of these functions, and here we introduce sp(u) = − sm2(u) cm(u) , cp(u) = cm2(u) sm(u) . These are order 6 elliptic function with the same period lattice Zπ3 ⊕ Zπ3ω. Moreover, we have Projective translation equation 10 Proposition 3. The functions sp(u) and cp(u) have these properties: 1 ≡ sp(u) cp(u)[ sp(u) − cp(u)], sp′(u) = − sp2(u) + 2 sp(u) cp(u), cp′(u) = − cp2(u) + 2 sp(u) cp(u), sp(ωu) = ω2 sp(u), cp(ωu) = ω2 cp(u), sp(−z) = sp(z), cp(−z) = 1 sp(z) cp(z) , cp(cid:16) π3 3 − u(cid:17) = − sp(u), sp(u + v) = cp(u + v) = (C1 + C2 − S1S2C1C2)2S1S2 1 S2 (1 − S1S2C1 − S1S2C2)2C1C2 1 S2 2C1C2)(S1S2C1 + S1S2C2 − 1) ; 2C1C2)(C1 + C2 − S1S2C1C2) , (1 − S2 (1 − S2 here S1 = sp(u), C1 = cp(u), S2 = sp(v), C2 = cp(v). These are verified using Proposition 2. 3. The results 3.1. Classification. The first main result of this paper is the complete solution of the Problem 1. Theorem 1. Let φ(x) = u(x, y) • v(x, y) be a smooth projective flow such that (2) holds, and its vector field, defined by (3), is a pair of 2-homogenic rational functions. Suppose that both u and v are defined on C2, except for a countable set of isolated curves each, where they might have poles, and that u, v in their definition domains are single-valued analytic functions. Then there exists a 1-BIR ℓ, such that ℓ−1 ◦ φ ◦ ℓ(x) is one of the following canonic projective flows: 1) x • y; 2) φN for N ∈ N0, the level N flow, whose orbits are given by xyN−1 = c; only in the latter 3) φe(x), the level 1 flow whose orbits are given by x + y = c; it is conjugate to the flow φt(x), also the level 1 flow and also with orbits given by x + y = c (for these two, see Proposition 1); two cases the flow is rational; 4) Λ(x) = λ(x, y) • λ(y, x), the level 3 flow whose vector field is x2 − 2xy • y2 − 2xy, orbits are given by xy(x−y) = c, and this flow is algebraically expressable in terms of Dixonian elliptic functions (see Theorem 2); 5) Ψ(x) = ψ(x, y)• ψ(y, x), the level 4 flow whose vector field is x2 − 3xy • y2 − 3xy, orbits are given by xy(x − y)2 = c, and this flow is expressable in terms of lemniscatic elliptic functions (with quadratic period lattice); 6) ∆(x) = α(x, y) • β(x, y), the level 6 flow whose vector field is x2 − xy • y2 − 2xy, orbits are given by (3x− 2y)x3y2 = c, and this flow is expressable in terms of Dixonian elliptic functions again (for the last two, see the Subsection 4.5). 11 G. Alkauskas Our main concern of this paper is the case 4); the last two cases will be covered in the Subsection 4.5. The case 4) is particularly interesting since it has an additional 6-fold symmetry, and now we will concentrate on it. 3.2. The S3-superflow. (For the explanation of the title, see the Subsection 5.1). So, we investigate the fascinating function Λ(x, y) = λ(x, y)• λ(y, x) which possesses these main prop- erties. • (PDE). The first coordinate satisfies the partial differential equation λx(x, y)(x2 − 2xy − x) + λy(x, y)(y2 − 2xy − y) = −λ(x, y) with the boundary condition λ(xz, yz) z lim z→0 = x. (13) (14) 1 − z • (FLOW). The function Λ satisfies the iterative functional equation of GL2(R), call it Σ ∼= S3, whose elements are (1 − z)Λ(x) = Λ(cid:16)Λ(xz) c d(cid:1), let γ(x, y) = (ax+ by, cx+ dy). Consider the 6 element subgroup • (SYMM). If γ =(cid:0) a b i =(cid:18) 1 0 0 1 (cid:19) , στ σ =(cid:18) −1 0 −1 1 (cid:19) , z (cid:17), x = (x, y) ∈ C2, z ∈ C. σ =(cid:18) 0 1 τ =(cid:18) 1 −1 1 0 (cid:19) , 0 −1 (cid:19) , στ =(cid:18) 0 −1 τ σ =(cid:18) −1 1 1 −1 (cid:19) , −1 0 (cid:19) ; and then (στ )3 = (τ σ)3 = i, σ2 = τ 2 = (στ σ)2 = i. The function Λ(x, y) possesses the 6-fold symmetry: for every γ ∈ Σ, we have γ−1 ◦ Λ ◦ γ(x, y) = Λ(x, y). Of course, involutions σ and τ generate the whole group Σ, so only two of these in- variance properties are independent. The invariance under σ tells us that Λ(x, y) = (λ(x, y), λ(y, x)), that is, the second coordinate is just the flip of the first, and the invariance under τ implies the identities (cid:26) λ(x, y) + λ(−y, x − y) + λ(y − x,−x) = 0, λ(x, y) + λ(−x, y − x) = 0. (15) • (ELL). In case x, y are fixed, xy(x − y) 6= 0, the pair of functions z (X(z), Y (z)) =(cid:16)λ(xz, yz) ·(cid:16) λ(xz, yz) · z z , − z parametrize the elliptic curve XY (X − Y ) = xy(x − y); thus, λ(xz, yz) λ(yz, xz) λ(yz, xz) λ(yz, xz) (cid:17) = (λz(x, y), λz(y, x)) (cid:17) ≡ xy(x − y), z z x, y, z ∈ C. It turns out that three exceptional lines y = 0, x− y = 0, and x = 0 correspond to three ramification points of the algebraic function [z(1 − z)]−2/3; namely, z = 0, 1, and ∞. Projective translation equation 12 y -- 1 -- 0.8 -- 0.6 -- 0.4 -- 0.2 1 0.8 0.6 0.4 0.2 0 -- 0.2 -- 0.4 -- 0.6 -- 0.8 -- 1 0.2 0.4 0.6 0.8 1 x Figure 1. The vector field of the flow Λ(x) is given by x2− 2xy• y2− 2xy. Here we show the normalized vector field, meaning that all arrows are adjusted to have the same length. A selected orbit shown is the elliptic curve xy(x−y) = −0.04. A point travels this orbit with a convention that, for example, if it goes up via the left side of the top-right branch, it reappears on the left side of the bottom-right branch, and so on. So, we always mind the asymptote. 3.3. Basic properties of λ(x, y). The PDE (13) with the boundary condition (14) has, as already mentioned, the unique solution λ(x, y). In this subsection we will derive few computa- tional results. As was proved in [Al12], we formally have λ(xz, yz) = xz + zi(i)(x, y), ∞Xi=2 and the homogenic functions (i)(x, y) can be recurrently calculated by (i+1)(x, y) = 1 i [(i) x (x, y)(x, y) + (i) y (x, y)(x, y)], i ≥ 2; (16) (17) here, as before, (x, y) = x2 − 2xy, (x, y) = y2 − 2xy. This recursion is essentially equivalent to the PDE (13). The above also holds for i = 1 if we make a natural convention that (1)(x, y) = x. Thus, we can calculate these polynomials, as presented in the Table 1. The two pairs of functions (x, y) and (x2 − 2xy, y2 − 2xy) are invariant under conjugation with two independent linear involutions (x, y) 7→ (y, x) and (x, y) 7→ (x − y,−y). We thus get 13 G. Alkauskas 7 8 6 5 4 3 2 t3 2 t4 + 5 i (i)(t, 1) = wi(t). 1 2 t t2 − 2t t3 − t2 + t t4 − 2t3 t5 − 5 t6 − 3t5 + 3t4 − 2t3 t7 − 7 2 t6 + 9 t8 − 4t7 + 43 t9 − 9 2 t8 + 225 10 t10 − 5t9 + 285 11 t11 − 11 2 t10 + 88 12 t12 − 6t11 + 213 13 t13 − 13 2 t12 + 507 14 t14 − 7t13 + 85 15 t15 − 15 2 t14 + 345 2 t5 − 2t4 + t3 7 t6 − 32 7 t5 + 5 28 t7 − 199 28 t8 − 75 7 t9 − 29 14 t10 − 295 28 t11 − 1573 4 t12 − 73 14 t13 − 325 9 28 t6 + 51 28 t4 + 1 28 t3 7 t7 + 165 7 t4 − 2 7 t3 14 t5 − 3 28 t6 − 33 14 t5 28 t7 − 99 7 t8 − 117 14 t9 + 120 28 t8 + 297 28 t6 + 33 28 t5 56 t10 + 1475 2 t11 + 3527 14 t7 + 3 2 t6 − 3 7 t5 56 t7 − 3 56 t8 + 309 13 t8 − 285 2548 t10 + 53905 56 t9 − 843 91 t10 − 4741 182 t9 + 138 2548 t11 − 108123 7 t6 + 3 28 t5 364 t6 − 3 182 t8 + 3855 182 t5 91 t7 + 27 2548 t9 − 1095 7 t12 + 140325 Table 1. Polynomials wi(t). 2548 t7 − 15 2548 t6 + 3 2548 t5. that polynomials wn(t) = (n)(t, 1) possess the same 6-fold symmetry:  1 − t(cid:17) = 0, 1 wn(t) + (−t)n wn(cid:16)1 − wn(t) + (1 − t)n wn(cid:16) t wn(t) ≡ 0 mod t(n), t(cid:17) + (t − 1)n wn(cid:16) 1 t − 1(cid:17) = 0. (n) = 2jn + 2 6 k, Moreover, and this congruence is exact, meaning that the next higher power of t is the smallest power present in the polynomial wn(t). Computer calculations show that this fact uniquely charac- terizes wn(t) among monic polynomials with 6-fold symmetry only for small n. Let λ(x, y) = ∞Xn=0 xnfn(y), fn(y) = 1 n! ∂n ∂xn λ(x, y)(cid:12)(cid:12)(cid:12)x=0 ∈ C[[y]]. Projective translation equation 14 The functions fn(y) are in fact all polynomials: 5 2 f1(y) = 1 − 2y + y2, f2(y) = 1 − y, f3(y) = 1 − 2y + f4(y) = 1 − f5(y) = 1 − 3y + f6(y) = 1 − y + 5 2 5 y + 3y2 − 2y3 + 7 y3 + 2 y2 − 2y3 + y4 − 7 y4 − 51 14 y3 + 9 32 y2 − 7 2 43 199 y2 − 28 7 3 28 y4 − 165 28 7 2 y5, 33 14 y4 − y5 + 1 28 y6, y5 + 33 28 y5 + 3 7 y6 − y6 − 3 2 y7 + 3 28 y7 + y8 − 27 364 3 182 y8 − y9 + 15 2548 3 2548 y9, y10, 3 7 99 28 and so on. By the direct calculation, using (17), we get (computationally now, which is auto- matic a posteriori we finish the proof of the Theorem 2) that λ(0, x) = 0, λ(x, x) = λx(x, x) = 1 2 (x + 1)−2 + but , λ(x, 0) = x 1 + x 1 2 (x + 1)2, x 1 − x 1 λy(x, x) = 2 , (18) (x + 1)−2 − 1 2 (x + 1)2, λ(x,−x) = x + 3x2 + 3x3 + 3x4 + 6x5 + 9x6 + 12x7 + x8 + x11 + x12 + x13 + x14 + 708 7 495 7 x17 + 117 7 13140 91 542532 171 7 x9 131076 x15 637 1130958 266670 380403 637 209391300 637 x22 + 84721 x18 + x19 + 298661544 84721 637 x23 + 931 425993769 84721 x24 + · · · . (19) x20 + + + x10 + 246 7 186903 348 7 x16 + 637 20971530 12103 x21 + The orbits of the flow Λ are given by W (x, y) = xy(x − y) = c ∈ C. Consider the plane cubic (in affine coordinates) XY (X − Y ) = c. When c 6= 0, this is a non-singular cubic and thus an elliptic curve. If 2p we get it in Weierstrass form (also in affine coordinates) X = c − q 2p , Y = −c − q , (p : q : 1) ∈ P 2(C) : q2 = 4p3 + c2. 3.4. Hypergeometric approach. The function λ(x, y) can be explored in two different (mu- tually inverse) ways - either using hypergeometric function W (x), or using Dixonian elliptic functions sm(u) and cm(u). We choose the second way, but will now exhibit how the first approach does work. Both methods reduce in fact to algebraic manipulations in quotient rings of rings of rational functions. Let us define the curve C0 ⊂ R2 parametrically by (see Figure 2) C0 =n(cid:16)xW (x), W (x)(cid:17) : −∞ < x < 1o. Proposition 4. The function λ(x, y) vanishes on the curve C0: λ(xW (x), W (x)) ≡ 0 for − ∞ < x < 1. 15 G. Alkauskas Proof. Let P (x) = λ(xW (x), W (x)). When x = 0, we have P (0) = 0. Let A = xW (x), B = W (x). Then P ′(x) = λx(A, B)[W (x) + xW ′(x)] + λy(A, B)W ′(x). (20) Note that the differential equation (11) for W (x) implies A2 − 2AB − A = [W (x) + xW ′(x)] · W (x)3x(x − 1), B2 − 2AB − B = W ′(x) · W (x)3x(x − 1). Thus, let us multiply (20) by W (x)3x(x − 1). Then the PDE for λ(x, y) implies P ′(x) ·(cid:16)W (x)3x(x − 1)(cid:17) = λx(A, B)(A2 − 2AB − A) + λy(A, B)(B2 − 2AB − B) = −λ(A, B) = −P (x), for − ∞ < x < 1. This is the differential equation for P (x), and, minding P (0) = 0, its only solution is P (x) ≡ 0. (cid:3) The curve C0 has the asymptote x = y. More precisely, it is much closer to the cubic (cid:16)x, x + Π x2(cid:17), x > 1. Let E(c) = {(x, y) ∈ C2 : xy(x − y) = c}, and E R(c) = {(x, y) ∈ R2 : xy(x − y) = c}. Not every orbit (elliptic curve) E R(c) for c 6= 0 intersects the curve C0, but only those with c ∈ (−Π, Π) \ {0}. This follows from the inspection of the function W 3(x)x(1 − x), which is monotonically increasing from −Π to Π in the interval (−∞, 1); see the representation (12). The Figure 2 shows the boundary case E R(Π), where this elliptic curve and C0 touch at infinity. On the curve the function λ(x, y) attains the value ∞, and the same holds on the curve C ∞ =n(cid:16)W (x), xW (x)(cid:17) : −∞ < x < 1o C1 =n(cid:16)(x − 1)W (x),−W (x)(cid:17) : −∞ < x < 1o. The curve C0 as a whole remains intact under the linear involution (x, y) 7→ (−x, y − x), while the other two curves interchange. This and similar properties follow from the Subsection 2.1. This shows how the action of the group S3 on the function W (x) can be interpreted as its action on λ(x, y). Despite the possibility to develop all properties of λ(x, y) over C2 (not just over R2) in the framework of W (x), henceforth we choose the elliptic function setting. 3.5. Analytic formulas. Our second main result of this paper reads as follows. Theorem 2. The function λ(x, y) can be given the analytic expression: λ(x, y) = λ(y, x) = ς(cid:0)c ς 2 − sc2y ς + s2xy(cid:1)2 ς(cid:0)c2 ς 2 − sx ς + s2cxy(cid:1)2 y(cid:0)x − c3y(cid:1)(cid:0)c2 ς 2 − sx ς + s2cxy(cid:1), x(cid:0)x − c3y(cid:1)(cid:0)c ς 2 − sc2y ς + s2xy(cid:1); here ς = ς(x, y) = [xy(x−y)]1/3, and s = sm( ς), c = cm( ς) are the Dixonian elliptic functions. This function satisfies all the above properties. In particular, λ(x, y) is a single-valued 2-variable analytic function. Projective translation equation - 16 y l(0,x)=0 + double zero l ( , ) = x y E - y 1 l ( , ) = x y E - y -1 1 l( ,0) = x x 1-x x pole -1 pole l( , ) = x x x 1+x l ( , ) = x y E - y + - Figure 2. The function λ(x, y) for (x, y) ∈ R2; it is a real section of the bivariate analytic function. The grey and white regions show where this function is negative and, respectively, positive. The double zero is the curve C0 parametrized by (xW (x), W (x)), where W (x) is the hypergeometric function; see the Subsection 2.1. Three branches of the elliptic curve xy(x − y) = π3 3/27 = 5.513+ are shown; on this curve the expression for λ(x, y) is particularly simple. Note however that there is a countable number of curves on which λ(x, y) has double-zeros or poles, but these fall out of the borders of the picture. Of course, one expression for λ is sufficient, but interchanging x and y changes ς into − ς, and we rather rewrote cm(− ς) and sm(− ς) in terms of cm( ς) and sm( ς) using Proposition 2. The function λ is unramified because of Proposition 3. Indeed, for given (x, y) ∈ C2, fix one of the three values for ς = [xy(x − y)]1/3, and use this value consistently in the above formula for λ(x, y). Then we get that the value of λ(x, y) does not depend on our choice of the cubic root: we can rewrite the expression for λ as λ(x, y) = x(x − y)(cid:0)c − (s ς−1)c2y + (s ς−1)2xy(cid:1)2 (cid:0)x − c3y(cid:1)(cid:0)c2 − (s ς−1)x + (s ς−1)2cxy(cid:1), and both series cm( ς) and sm( ς) ς−1 contain only integral powers of x, y. This function does satisfy the property (PDE) by the very solution; see the Subsection 4.1. The property (FLOW), as in the cases of φe and φt, is equivalent to addition formulas for the Dixonian functions given by Proposition 2. This is, to our understanding, a remarkable fact! We note that it is 17 G. Alkauskas considerably more natural to write down the analytic expression of λ in terms of A = sp( ς) ς and B = cp( ς) ς, where sp and cp are other two elliptic functions introduced in the Subsection 2.2. This expression is given by (23), and the rational function in the middle of (23) is 1-homogenic and defines the quasi-rational projective flow of genus 1; see the Subsection 5.4. We chose to present the results in terms of Dixonian functions sm and cm solely because these are more conventional functions and there are few results already available in the literature. In fact, the expression for λ(x, y) in the Theorem 2 is also 1-homogenic, if x, y, ς are given weight 1, and cm( ς) and sm( ς) are both of weight 0. 3.6. Values on special curves. Now we are able to say more about the Taylor coefficients of the function λ(x,−x)x−1, which were computed in ([Al12], Section 4, Step III); see (19). Corollary 1. Let (x, y) = (z,−z), choose ς = −z 3√2. We have c = cm(−z 3√2), Further, λ(z,−z)z−1 is an elliptic function. In general, for fixed x, y, xy(x − y) 6= 0, the function λ(xz, yz)z−1 is an elliptic function in z. On the other hand, fix c 6= 0. On the elliptic curve E(c) the function λ(x, y) is a rational function in x, y with degree 6 numerator and degree 3 denominator: 3√2(cid:0)c 3√4 − sc2 3√2 − s2(cid:1)2 (cid:0)1 + c3(cid:1)(cid:0)c2 3√4 + s 3√2 − s2c(cid:1), s = sm(−z 3√2). λ(z,−z) = z λ(x, y)(cid:12)(cid:12)(cid:12)E(c) λ(y, x)(cid:12)(cid:12)(cid:12)E(c) = = x(x − y)(B − AB2y + A2xy)2 (x − B3y)(B2 − Ax + A2Bxy) y(x − y)(B2 − Ax + A2Bxy)2 (x − B3y)(B − AB2y + A2xy) = R(A, B; x, y), = R(cid:16) A B , 1 B ; y, x(cid:17). here A = sm( ς) ς−1, B = cm( ς). The power series at the origin for cm(−z 3√2) and sm(−z 3√2) 3√4 contain only rational coef- ficients, so Taylor coefficients of λ(z,−z)z−1 belong to Q, and they are given by (19). In fact, we can use the Taylor series in Proposition 2 to calculate the expansion of λ(z,−z)z−1; these series were borrowed from [FCF05], formula (6), only note the typo "8880000" instead of the correct value "880000". This and our first method of calculations, i.e. (19), which is done using the recursion (17), match perfectly. For some special c the function λ(x, y) has a particularly simple expression. For example, 3 + π3k) for any k ∈ Z, c = ς 3. (see the Subsection 2.2). Then B = 0, A = ς−1, and let ς = ( π3 then λ(x, y)(cid:12)(cid:12)E(c) = − x(x − y)y2 ς 3 = −y, λ(y, x)(cid:12)(cid:12)E(c) = x − y. Also, for example, when ς is a zero of sm equivalent to the point q2,1 (see the Subsection 2.2 as well), we get A = 0, B = ω2, and then To summarize, we have observed a different behaviour of λ(x, y) on three kinds of curves: lines through the origin; elliptic curves; transcendental curves parametrized by hypergeometric λ(x, y)(cid:12)(cid:12)E(c) = x, λ(y, x)(cid:12)(cid:12)E(c) = y. Projective translation equation 18 function W (x). In the end of the Subsection 4.2 we will see that the function λ(x, y) is algebraic function on another transcendental curve 3.7. The symmetry property. Now we will turn our attention to the property (SYMM). Note that if xy(x − y) = c, then (−y, x − y), (y − x,−x) ∈ E(c), and so the first equality in (15) means that the following as if holds on the elliptic curve E(c): sp(u)u ≡ 1. R(A, B; x, y) + R(A, B;−y, x − y) + R(A, B; y − x,−x) ?= 0. In fact, this is NOT the identity satisfied by the rational function R(A, B; x, y) for arbitrary free A, B, x, y, as computer calculations show: if we calculate the l.h.s. for arbitrary unspecified variables A, B, x, y, we get a complicated and lengthy expression, and it involves high powers of A and B. But in our case the four variables are related via the identity A3xy(x − y) + B3 ≡ 1. Consider now the field of rational functions C(A, B, x, y). What we really get is R(A, B; x, y) + R(A, B;−y, x − y) + R(A, B; y − x,−x) ≡ 0 mod(cid:0)A3xy(x − y) + B3 − 1(cid:1), and this is satisfied for free variables: the numerator of the l.h.s. is divisible by A3xy(x−y)+B3− 1, what MAPLE really does confirm, too. Likewise, if (x, y) ∈ E(c), then (−x, y − x) ∈ E(−c). So, the second identity of (15), if we use Proposition 2, gives This is also easily reverified using computer algebra engine. ;−x, y − x(cid:17) ≡ 0 mod(cid:0)A3xy(x − y) + B3 − 1(cid:1). R(A, B; x, y) + R(cid:16) A B , 1 B In the Subsection 4.3 we will see that the Symmetry property can be described in terms of the group structure of the elliptic curve E(c). 3.8. Relation to P´olya-Eggenberger urn model. The PrTE (1) was first introduced in [Al10]. This is a new and fascinating object with many ramifications; see the Section 5. While working on the paper [Al12] we were unaware of any research related to the algebraic theory of the PDE (4). However, in the course of the evolvement of the current work the importance of the constant π3 emerged. This later led us to the paper [BF10], and then to many other works, including [Du79, Du86, BF11, FCF05, FGP05, FDP06]. It appeared that a theory related to the PDE similar to (4) was already well investigated with relation to the P´olya-Eggenberger urn models. We will shortly describe the setting and the reasons why both directions (projective flows and urns) have a certain intersection in common. Suppose there is an urn containing a finite number of balls of two types: x and y. At each tick of a time we choose one ball at random, observe its color, put it back to the urn, and then add or subtract balls from the urn according to the result of this random pick. If the ball x is picked, we add balls of type x, y in quantities α and β, respectively. If the ball is y, we add balls of type x, y in quantities γ and δ, and negative numbers are allowed and stand for a subtraction. The urn is said to be balanced if increase of balls at each step is deterministic, i.e. α + β = γ + δ = s. The task is to describe all histories of any given length; a history is any legal sequence of steps that lead from the initial urn to any specific one; say, the one where all balls are of type x. (Of course, certain simple arithmetic conditions, called tenability conditions, on coefficients α, β, γ, δ, and the initial urn composition are needed to ensure that at each step the action of subtraction can be performed). It appears that histories can be investigated via analysis of the differential operator [FGP05] 19 G. Alkauskas Υ = x1+αyβ ∂ ∂x + xγy1+δ ∂ ∂y . This is completely analogous to the recurrence (17) where (x, y) • (x, y) is an arbitrary 2- homogenic vector field. So, it appears that the property of the urn to be balanced corresponds to projectivity in our setting. Nevertheless, these fields of research have many differences. While we deal with arbitrary 2-homogenic rational functions in C(x, y), the research in P´olya urns concentrates and has a combinatoric meaning only for the monomials x1+αyβ • xγy1+δ. On the other hand, our vector fields are 2-homogenic, which corresponds to the balance 1 in the urn setting; but other balances are also investigated and are of great importance; balance s 6= 1 does not seem to correspond to a flow or other geometric object. So, the two theories are different though have an non-empty intersection and there are many similarities! For example, a relative of the function λ(x, y) has already appeared in the literature, though in a disguise which at first seem to be unrecognisable ([FCF05], Proposition 3): H(x, z) = (1 − x3)1/3 smh (1 − x3)1/3z +Z x(1−x3)−1/3 0 ds (1 + s2)2/3!; here, as mentioned in the Subsection 2.2, smh(u) = − sm(−u). Still, this function is closely related to λ(x, y), which can be seen if we go a few step further: that is, apply the addition formulas for the Dixonian elliptic functions and use their relation with hypergeometric series. Then the above complicated formula would turn into a formula similar to that in our Theorem 2. Philippe Flajolet and many other mathematicians have developed powerful analytic and combinatoric techniques to investigate (not necessarily balanced) urns; see, for example, the excellent study [BF11]. In our paper we did not touch the combinatoric aspects at all. Possibly, these techniques will be of big help in the prospective work on quasi-rational flows [Alpr2]. 4. The proofs. Further properties 4.1. The analytic expression for λ(x, y). With all tools at hand, we can solve analytically the PDE (13) and prove the Theorem 2. Proof. Let us introduce Ω(u, ς) = λ(cid:16) sp(u) ς, cp(u) ς(cid:17). Each pair of finite complex numbers x, y can be represented as (x, y) = ( sp(u) ς, cp(u) ς) for ς ∈ C, u ∈ F (fundamental parallelogram of our elliptic functions), except when xy(x− y) = 0. The boundary condition requires that Ω(u, ς) ς lim ς→0 = sp(u). (21) Projective translation equation 20 Then, by a direct calculation and Proposition 3, Ωu(u, ς) = λx(cid:16) sp(u) ς, cp(u) ς(cid:17) sp′(u) ς + λy(cid:16) sp(u) ς, cp(u) ς(cid:17) cp′(u) ς = λx(cid:16) sp(u) ς, cp(u) ς(cid:17)(cid:16) − sp2(u) + 2 sp(u) cp(u)(cid:17) ς + λy(cid:16) sp(u) ς, cp(u) ς(cid:17)(cid:16) − cp2(u) + 2 sp(u) cp(u)(cid:17) ς; Ω ς(u, ς) = λx(cid:16) sp(u) ς, cp(u) ς(cid:17) sp(u) + λy(cid:16) sp(u) ς, cp(u) ς(cid:17) cp(u). Therefore, according to (13), we have ςΩu(u, ς) + ςΩ ς (u, ς) = Ω(u, ς). The general solution of this linear PDE is given by H(u − ς) ς, for any H, which is smooth in C, except possibly at the poles. The boundary condition (21) requires that H(u) = sp(u). So, Ω(u, ς) = λ(cid:16) sp(u) ς, cp(u) ς(cid:17) = sp(u − ς) ς. Recall that x = sp(u) ς, y = cp(u) ς. Then xy(x − y) = ς 3, and thus ς = [xy(x − y)]1/3. Let S = sp(u), C = cp(u), a = sp(− ς), b = cp(− ς). Then using Proposition 3, we get sp(u − ς) = (b + C − abSC)2aS (1 − a2bS2C)(aSC + abS − 1) . Replace in the latter expression S = x ς−1, C = y ς−1. We get (b ς 2 + y ς − abxy)2ax . sp(u − ς) = ( ς 3 − a2bx2y)(axy + abx ς − ς 2) Further, let A = sp( ς), B = cp( ς). Again, from Proposition 3 we know that a = A, b = (AB)−1. Rewriting the above in terms of A, B, we get sp(u − ς) = ( ς 2 + ABy ς − Axy)2x A(B ς 3 − Ax2y)(ABxy + x ς − B ς 2) . (22) Finally, let sm( ς) = s, cm( ς) = c. So, A = −s2c−1, B = c2s−1. This gives sp(u − ς) ς = (c ς 2 − sc2y ς + s2xy)2x ς (c3 ς 3 + s3x2y)(s2cxy − sx ς + c2 ς 2) . Since x2y − xy2 = ς 3, the last simplifies to λ(x, y) = sp(u − ς) ς = ς(c ς 2 − sc2y ς + s2xy)2 y(x − c3y)(c2 ς 2 − sx ς + s2cxy) . This is exactly the statement of Theorem 2. (cid:3) 4.2. Rational and algebraic interpretation of λ. We now proceed in showing that the function λ(x, y) can be described purely algebraically without appeal to elliptic functions, which, most importantly, leads us to the definition of projective quasi-rational flows of arbitrary genus; see the Subsection 5.4. Let ς be as in the previous Subsection, but let us rewrite the formula (22) not in terms of A and B as given there, but (as mentioned in the end of the Subsection 3.5) in terms of A = sp( ς) ς, B = cp( ς) ς, which is the most natural. We thus have 21 G. Alkauskas λ(x, y) = [x(x − y) + AB − Ax]2y(x − y) A[B(x − y) − Ax][AB + x(x − y) − B(x − y)] = T (A, B; x, y). (23) Note that if A, B, x, y are all given a weight 1, then T (A, B; x, y) is a 1-homogenic function. The boundary condition property (14) reads as The PDE (13) can be rewritten in terms of the rational function T . Indeed, T (−z3xy(x − y), 1; xz, yz) z = x. lim z→0 ςx = 2xy − y2 Ax = [ sp′( ς) ς + sp( ς)] ςx, Ay = [ sp′( ς) ς + sp( ς)] ςy, , 3 ς 2 ςy = x2 − 2xy 3 ς 2 , Bx = [ cp′( ς) ς + cp( ς)] ςx, By = [ cp′( ς) ς + cp( ς)] ςy. Thus, x ςx + y ςy = ς, The PDE (13) now read as ςx(x2 − 2xy) + ςy(y2 − 2xy) = 0. (24) (25) 0 = λx(x2 − 2xy − x) + λy(y2 − 2xy − y) + λ(x, y) = TA[ sp′( ς) ς + sp( ς)] ςx(x2 − 2xy − x) + TB[ cp′( ς) ς + cp( ς)] ςx(x2 − 2xy − x) + Tx(x2 − 2xy − x) + TA[ sp′( ς) ς + sp( ς)] ςy(y2 − 2xy − y) + TB[ cp′( ς) ς + cp( ς)] ςy(y2 − 2xy − y) + Ty(y2 − 2xy − y) + T = −TA[ sp′( ς) ς + sp( ς)] ς − TB[ cp′( ς) ς + cp( ς)] ς + Tx(x2 − 2xy − x) + Ty(y2 − 2xy − y) + T = TA(A2 − 2AB − A) + TB(B2 − 2AB − B) + Tx(x2 − 2xy − x) + Ty(y2 − 2xy − y) + T ; in the penultimate equality we used (25), and in the last one Proposition 3 was used. We get that the PDE for the rational function T is just an amalgam of two identical PDE's. Each of them separately is solvable in terms of elliptic functions, while their amalgam gives a rational solution! Since T is 1-homogenic, we seem to obtain the 4-variable PDE TA(A2 − 2AB) + TB(B2 − 2AB) + Tx(x2 − 2xy) + Ty(y2 − 2xy) ?= 0. (26) However, computer calculations show that this is NOT the identity satisfied by the rational function T - that is the meaning of the question mark. In fact, the four variables A, B, x, y are not independent but satisfy AB(A − B) = xy(x − y). Now, calculations prove that this PDE is indeed satisfied modulo AB(A − B) − xy(x − y). We thus get the needed property: for free variables A, B, x, y, one has l.h.s. of (26) ≡ 0 mod(cid:16)AB(A − B) − xy(x − y)(cid:17). This shows that the first coordinate of the flow Λ, now in the avatar T , can be described purely algebraically as follows: it is 1-homogenic, it satisfies the boundary condition (24), and Projective translation equation 22 the 4-variable PDE in the quotient ring. This leads us to the definition of quasi-rational flows in the Subsection 5.4. Moreover, we can give (26) another appearance by eliminating one variable; say, A. First, by the homogeneity property, Now plug this into (26). The derivative with respect to the variable A is not present any more, and we can freely choose A = 1. We thus obtain the equation ATA = T − BTB − xTx − yTy. TB(3B2 − 3B) + Tx(x2 − 2xy − x + 2Bx) + Ty(y2 − 2xy − y + 2By) = (2B − 1)T , which is valid in the quotient ring C(B; x, y)/(cid:0)B(1 − B) − xy(x − y)(cid:1). So, let B = B(x, y) be the solution of B(1 − B) = xy(x − y), which we can express in quadratic radicals: (27) B(x, y) = Let therefore 1 2 + 1 2p1 − 4xy(x − y). be an algebraic function of two free variables. This also satisfies the PDE, as we will now see. First, we have E(x, y) = T (1, B(x, y); x, y) Bx = 2xy − y2 1 − 2B , By = x2 − 2xy 1 − 2B . Ex = TB 2xy − y2 1 − 2B + Tx, Ey = TB x2 − 2xy 1 − 2B + Ty. Therefore, So, Ex(x2 − 2xy) + Ey(y2 − 2xy) = Tx(x2 − 2xy) + Ty(y2 − 2xy), Exx(1 − 2B) + Eyy(1 − 2B) = Txx(1 − 2B) + Tyy(1 − 2B) + TB(3x2y − 3xy2). Subtract now second from the first. This, using (27), implies (in an unconventional form) Ex(x2 − 2xy) + Ey(y2 − 2xy) E − xEx − yEy = 2B − 1 =p1 − 4xy(x − y). The validity of this identity was also double-checked on MAPLE, and it holds true; both numerator and denominator on the left side are rather complicated expressions. 4.3. The group structure of E(c) and its effect on the flow Λ. Let, as before, ω = e2πi/3. If c 6= 0, the projective transformation X = cr − q, Y = −cr − q, Z = 2p maps the curve XY (X − Y ) = cZ 3 to the elliptic curve [Kn92] bE(c) = {(p : q : r) ∈ P 2(C) : q2r = 4p3 + c2r3}. If P = (X : Y : 1) ∈ E(c), then −P = (−Y : −X : 1). Further, let P1 = (X1 : Y1 : 1) and P2 = (X2 : Y2 : 1) be two (finite) points on E(c). The standard addition formulas for the curve 23 G. Alkauskas bE(c) translate onto E(c) as follows. Let P3 = P1 + P2 = (X3 : Y3 : Z3). Then 1 + X2Y2 − X 2 X3 = (−X1Y1 + X 2 Y3 = (X1Y1 − Y 2 Z3 = (X1 − X2)(Y1 − Y2)(X1 − Y1 + Y2 − X2). 2 )(Y1 − Y2)2, 2 )(X1 − X2)2, 1 − X2Y2 + Y 2 This can be given an alternative expression. Using X1Y1(X1 − Y1) = c and X2Y2(X2 − Y2) = c, we can rewrite this as X3 = (Y1 − Y2)3X1X2, Y3 = (X1 − X2)3Y1Y2, Z3 = (X1 − X2)(Y1 − Y2)(X1Y1 − X2Y2). The duplication formula read as follows. If P = (X : Y : 1), then 2P = (X1 : Y1 : Z1), where X1 = (2X − Y )3Y, Y1 = (2Y − X)3X, Z1 = (X + Y )(2X − Y )(2Y − X). Now we will introduce few special points, which give the cyclic 6-group on the curve E(c), as presented in the Table 2. Note that for any c 6= 0 these 6 points are different. The addition 1 2 3 3 6 6 (p : q : r) O (1 : 1 : 0) E(c) (X : Y : Z) Order bE(c) bO (0 : −1 : 0) bQ2 bQ3 2bQ3 bQ6 5bQ6 (− 3pc2/4 : 0 : 1) Q2 (0 : c : 1) (0 : −c : 1) ( 3√2c2 : 3c : 1) ( 3√2c2 : −3c : 1) (− 3pc/2 : 3pc/2 : 1). (− 3pc/2 : − 3√4c : 1). ( 3√4c : 3pc/2 : 1). Table 2. Finite order points on elliptic curves bE(c) and E(c). (0 : 1 : 0) (1 : 0 : 0). Q3 2Q3 Q6 5Q6 formula below give Q6 + Q2 = 2Q3, 5Q6 + Q2 = Q3, Q3 = 2Q6. So, we deduce that Q2 = 3Q6. Let this group be C6. When c is complex, we fix the cubic root 3√c. We can also consider a wider group. Let Qω 6 = 6 = (−ω 3pc/2 : −ω 3√4c : 1), Qω2 (−ω2 3pc/2 : −ω2 3√4c : 1). The addition formulas give Q6 + Qω 6 + Qω2 Q6 − Qω Q6 − Qω2 6 = O, 6 = Qω2 6 = Qω 2 = (−ω2 3pc/2 : ω2 3pc/2 : 1), 2 = (−ω 3pc/2 : ω 3pc/2 : 1). 7→ (4, 1), Qω2 7→ (0, 1), Qω 2 So, these points produce the group C12 which is isomorphic to Z6 × Z2 via the following (non- canonical) isomorphism: Q6 7→ (1, 0), Qω 6 7→ (1, 1), Qω2 2 7→ (3, 1). 6 Projective translation equation 24 From the property (ELL) we know that if P = (x : y : 1) ∈ E(c), then Λz(P) = (λz(x, y) : λz(y, x) : 1) ∈ E(c). Further, for the three points at infinity - O, Q3 and 2Q3 - the action of Λ can be calculated using the Corollary 1. Here we use its notation. Indeed, assume A, B 6= 0, A3c + B3 = 1. Let now (x, y) ∈ E(c), and x, y → ∞ while remaining on the curve (in case c > 0 is real, this corresponds to the left-bottom asymptote in the Figure 2). Then (x : y : 1) → O. Thus, using the identity (x − y) = c(xy)−1, we can calculate , Λ(O) = lim 1 AB • B2 A = Fc ∈ E(c), R(A, B; x, y) • R(cid:16) A B 1 B ; y, x(cid:17) = x∼ y→∞ x−y∼ cx−2 where Fc is the special point on the curve E(c) as definition of A and B shows. In the same manner, Λ(Q3) = − cA2 B • − 1 AB , Λ(2Q3) = − B2 A • cA2 B . In case A = 0 the calculations give Λ(O) = O, Λ(Q3) = Q3, Λ(2Q3) = 2Q3, and in case B = 0 we have Λ(O) = 2Q3, Λ(Q3) = O, Λ(2Q3) = Q3. The order 3 linear map (x, y) 7→ (y − x,−x) corresponds to P 7→ P + Q3. If we know already that Λ(x) = (λ(x, y), λ(y, x)), the symmetry property (SYMM) now can be checked to be equivalent to the following nice identity: Λ(P + Q3) = Λ(P) + Q3 for every P ∈ E(c). 4.4. The case of a vector field (x, y) • 0. In this subsection we will show how to find all unramified flows in case in the PDE (4) we have (x, y) ≡ 0; see the first paragraph of the Subsection 1.3 for the explanation. It appears that no new unramified flows appear, apart from those found in [Al12], Chapter 4, Step II. It was shown there that the solution u(x, y) to (4) in case (x, y) ≡ 0 satisfies y/xZy/u(x,y) dt (1, t) = y. Put y = 1, and let q(x) = find all rational functions (1, t) such that the function q(x), x ∈ C, defined by u(x−1,1). Then (after a change x 7→ x−1) we get that one needs to 1 dt (1, t) = 1, xZq(x) BZA is single-valued. This is equivalent to the following: for all A, B ∈ C satisfying dt (1, t) = 0, (28) 25 G. Alkauskas this condition forces A = B; the integral is taken via any path only avoiding singularities. Suppose (1, t) = k(t)ℓ(t)−1, where k(t), ℓ(t) are polynomials and the fraction is irreducible over C. If deg(ℓ) > 0, let t0 be the root of ℓ(t), say, of multiplicity n0 ≥ 1. Let us define T (x) = xZt0 dt (1, t) , where x − t0 < ǫ, ǫ is sufficiently small such that (1, t)−1 has no poles or zeros inside this disc except at the center, and the path of integration is a segment. Then the function T (x) is well defined and, most importantly, it has a zero at x = t0 of multiplicity n0 + 1 ≥ 2. So, we know that there exists a small open set U ⊂ {x : x − t0 < ǫ} containing t0 such that in U the function T (x) attains each value exactly n0 + 1 times, T (t0) = 0 being the only multiple value [LS87]. In particular, if T (A) = T (B), A 6= B, A, B ∈ U \ {t0}, then (28) holds; the path of integration consists of a junction of two segments [A, t0] and [t0, B]. We get a contradiction. Suppose now ℓ(t) ≡ 1, and so (1, t) is a polynomial. If we perform the change of variables t 7→ 1 t in (28), we would obtain (by what was just proved) the neccessary condition that (1, 1 t )t2 is also a polynomial. So, deg (1, t) ≤ 2. In [Al12] we have explored all these cases; in particular, (1, t) = t and (1, t) = t2 + 1 produce the unramified flows φexp and φtan, the case (1, t) = t + 1 produces the flow linearly conjugate to φtan, while (1, t) = 1 and (1, t) = t2 give rational flows. 4.5. Other two elliptic flows. It can be seen that all the results, apart from the Λ(x, y)- specific symmetry property (SYMM), can be restated in terms of the flows with the vector fields x2 − 3xy • y2 − 3xy and x2 − xy • y2 − 2xy, respectively; see Theorem 1, items 5) and 6), and also the Subsection 1.3. Consider the first vector field. Let, as already defined in Theorem 1, the solution to (4) with the boundary condition (2) be given by ψ(x, y) • ψ(y, x). The orbits of this flow are the curves xy(x − y)2 = c. Consider the projective curve xy(x − y)2 = cz4, c 6= 0. If we make a birational change (x : y : z) 7→ ( z2 x + y : y : z), we see that these curves are birationally equivalent to elliptic curves. So, let us introduce two new elliptic functions p(u) and q(u) with such properties. The Taylor series for both these functions start at p(u) = u3 + · · · and q(u) = u−1 + · · · , and they contains only the powers u4n−1, n ≥ 1 and n ≥ 0, respectively. Further, we require 1 ≡ p(u)q(u)[p(u) − q(u)]2, p′(u) = −p2(u) + 3p(u)q(u), q′(u) = −q2(u) + 3p(u)q(u).  Using these properties, one can recurrently calculate the unique Taylor coefficients, and we do it with the help of MAPLE: p(u) = u3 + 1 5 q(u) = u−1 + u7 + 2 25 u11 + 3 5 u3 + 17 75 u7 + 127 4875 126 1625 u27 + · · · , u23 + · · · . u15 + 246 27625 u19 + 1246 414375 u23 + 1234412 1212046875 u11 + 32639 1243125 u15 + 6138 690625 u19 + 42898 14259375 Projective translation equation 26 Let x = p(u) ς, y = q(u) ς, ς = [xy(x − y)2]1/4. Similarly as in the Subsection 4.1, we find that the solution of (4) in case of the vector field x2 − 3xy • y2 − 3xy satisfies ψ(x, y) = ψ(cid:16)p(u) ς, q(u) ς(cid:17) = p(u − ς) ς. Let A = p( ς) ς, B = q( ς) ς. We can act completely analogously as in the Subsection 4.1, i.e. one can derive addition formulas for the elliptic functions p and q and then use them in the above expression. This gives the analogue of Theorem 2 in this case, and we omit the details. Equally, the vector field x2 − xy • x2 − 2xy has projective curves (3x − 2y)x3y2 = cz6 as orbits. The projective birational change (x : y : z) 7→ ( z2 y : z) transforms this curve into 3z2y − 2x3 = cy3, and thus for c 6= 0 this is an elliptic curve, and so we can derive analogous results as in the other two cases. x : x2 5. Open directions, further advances We finish this study with listing the variety of ways and points of view the multivariate PrTE can be explored from. If we pose a problem, it does not necessarily mean that this problem is hard - though it might be - or has a positive answer; the aim of this section is mainly expository: we just wish to exhibit the richness and structural variety that underlies (1). 5.1. Higher dimensional generalization of λ. We will now define an analogue of the flow Λ(x) in higher dimensions. Let N ∈ N, N ≥ 2. The symmetric group SN +1 has the standard (N + 1)-dimensional permutation representation inside the group GL(CN +1). The invariant subspace of this representation is the line µ·(1, 1, . . . , 1), µ ∈ C. So, SN +1 acts on the orthogonal complement A ⊂ CN +1 of this line which is given by A =n(v1, v2, . . . , vN +1) ∈ CN +1 : N +1Xi=1 vi = 0o. Let π : SN +1 7→ GL(A) be this representation. It is well known that it is irreducible and exact. Let us choose the basis of A as follows: qi = (0, . . . , 0, 1 i , 0, . . . , 0,−1), 1 ≤ i ≤ N. Each transposition of the form (ij) ∈ SN +1, 1 ≤ i < j ≤ N, acts on vectors qi, 1 ≤ i ≤ N, as the transposition (qiqj). On the other hand, if η ∈ SN +1 is the transposition (1(N + 1)), then the matrix representation of the linear map π(η) in basis (q1, . . . , qN )T is as follows: κ = −1 −1 1 −1 ... −1 ;  1 . . . 1 (matrices act on vector-columns by multiplication from the left). For example, when N = 2, this corresponds to the matrix στ σ, see (SYMM). Let ΣN +1 = π(SN +1). For each γ ∈ ΣN +1, we henceforth consider γ as a matrix in the fixed basis {q1, . . . , qN}. Now, we want to find an N-tuple of quadratic forms Q(x) =(cid:0)Q1(x), Q2(x), . . . , QN (x)(cid:1), (cid:26) Qi(x1, . . . , xi, . . . , xN ) = Q1(xi, . . . , x1, . . . , xN ), 1 ≤ i ≤ N, Qj(x1, . . . , xi, . . . , xN ) = Qj(xi, . . . , x1, . . . , xN ), for j 6= 1, i. (30) such that if Γ = ΣN +1, we have 27 G. Alkauskas (29) for each γ ∈ Γ. It is clear from the above that the vector Q should be symmetric with respect to all coordinates. This means γ−1 ◦ Q ◦ γ(x) = Q(x) Since transpositions (ij), 1 ≤ i < j ≤ N, together with the transposition (1(N + 1)) generate the whole group SN +1, it is enough to find a quadratic form Q1(x1, . . . , xN ) such that if Qi are defined by the first entry of (30) and they all satisfy the second entry, then the vector Q satisfies (29) in a special case γ = κ. This is a linear algebra task, and we find that the solution is given by So, we pose the following Q1(x) = x2 1 − 2 N − 1 · x1 xi. NXi=2 Problem 2. Let Q1 be as above, and Qi are given by (30). Let ΛN (x) = (λ(1), . . . , λ(N )), where λ(i)(x) is the solution of the PDE with the boundary condition NXi=1 fxi(x)[Qi(x) − xi] = −f (x) (Note that only the boundary condition depends on i). Describe the algebraic and analytic properties of the flow ΛN (x). f (xz) z = xi. lim z→0 We call the function ΛN (x) the ΣN +1-superflow. The solution of this problem might reveal deeper connections with various objects encountered in algebraic geometry and number theory. More generally, we may pose the same problem not only for ΣN +1 ⊂ GL(A), but also for any finite (sufficiently large, as compared to the degree of representation) subgroup of the linear group. The results in the Subsecion 3.7 show that this might be interpreted as symmetry identities for certain rational functions in the quotient rings. Definition 1. Let N ∈ N, N ≥ 2, and Γ ֒→ GL(CN ) be an exact representation of a finite group, and we identify Γ with the image. We call the flow φ(x) the Γ-superflow, if there exists a vector field Q(x) 6= (0, 0, . . . , 0) whose components are 2-homogenic rational functions and which is exactly the vector field of the flow φ(x), such that (29) is satisfied for all γ ∈ Γ, every other vector Q′ which satisfies (29) for all γ ∈ Γ is either a scalar multiple of Q, or its degree of a common denominator is higher. Thus, Q is uniquely defined up to conjugation with a linear map x 7→ zx. Note that we do not require the components of Q(x), which is the vector field of the flow φ, to be quadratic forms, they are just 2-homogenic rational functions. The main reason is obvious - the flow φ might possess lots of symmetries while ℓ−1 ◦ φ ◦ ℓ might not, even if we confine to special 1-BIRs ℓ. Another reason for this is that a priori there is no reason why for Projective translation equation 28 every vector field Q there must exists an N-dimensional 1-BIR ℓ such that the vector field of ℓ−1 ◦ φ ◦ ℓ has only quadratic forms as its components. We know that even in a 2-dimensional case the reduction is rather complicated, and there exists a particular obstruction where such ℓ does not exist ([Al12], Section 4). Problem 3. Describe all groups Γ inside GL(CN ) for which there exists a Γ-superflow. Describe the properties of such a flow φ(x) = φΓ(x) algebraically and analytically. For example, the group S4 has two non-equivalent exact irreducible 3-dimensional represen- tations. The first is constructed in the beginning of this subsection, while the second one, call it Σ′4, arises as the group of rotations of a cube, and it is given by the embedding (12) 7→ 0 0 0 1 1 0 0 0 −1 , (13) 7→ 0 0 1 0 −1 0 1 0 0 , (14) 7→ −1 0 0 0 0 1 0 1 0 . The Young diagram of this representation is dual to the Young diagram of the first represen- tation. Unfortunately, the vector Q of quadratic forms which satisfies (29) for all γ ∈ Σ′4, is trivial. But there exists a unique, up to scalar multiple, Q, which is invariant under even permutations, that is, the image of the group A4. This vector field is given by Q(x) = (yz, xz, xy). (31) To check the invariance we note that this group contains the order 4 subgroup consisting of the elements I, diag(−1,−1, 1), diag(−1, 1,−1), and diag(1,−1,−1) (the fourth Klein group). The invariance under conjugation with these transformations are immediate, and we only need to check invariance under one element of order three, say, the image of (12)(13) = (123). Nevertheless, this vector field is also invariant under the matrix 0 1 0 1 0 0  0 0 1 , and adjoining this matrix to tha image of A4 we thus obtain the S4-superflow, which is linearly conjugate to the flow Λ3, as given above. So, for the contragradient representation Σ′4 we obtain no new superflow, if the vector field is given by the collection of quadratic forms. Such negative outcome happens for majority of finite subgroups of the full linear group, provided these groups are large enough or have a particular structure. For example, already one condition (29) for γ = −I forces Q to be trivial. However, this is not the end of the story in the case of Σ′4. One can check directly the the vector field y3z − yz3 x2 + y2 + z2 • z3x − zx3 x2 + y2 + z2 • x3y − xy3 x2 + y2 + z2 Q(x) = is invariant under conjugation with all matrices from Σ′4. Up to the constant factor, this is the only vector field whose common numerator is of degree at most 2; so we get a Σ′4-superflow, essentially different form Σ4-superflow. We leave this very promising side of investigations of the projective translation equation for the future. Concerning the vector field (31), we could perform similar analysis as in the case of Λ = Λ2. In fact, few initial steps are already contained in the literature - this vector field lies exactly in the intersection of P´olya urns (all components are monomials) and projective flows (they are 2-homogenic). This urn model is called the pelican's urn, it can be analytically solved in terms of Jacobian elliptic functions, and the 29 G. Alkauskas combinatoric and analytic theory is very well developed and is profound [Du79, Du81, Du86, FDP06, Sch76, Vie80]. So, our contribution to this theory is the remark that this flow has a 24-fold symmetry, and the next step would be to rewrite the analytic and symmetry results of the Subsections 3.5 and 3.7 in the case of the vector field (31) purely algebraically. This seems to be very promising. 5.2. Topology. Continuous flows on the sphere Sk, k ≥ 3. Consider the functional equation (1) over R in dimension k ≥ 3, and suppose that a solution extends as a continuous function φ :cRk 7→cRk, where the latter is a single point compactification of Rk, this point being denoted by "∞"; we make no further regularity assumptions on φ. Problem 4. Is the following true? If the above assumption holds, then such φ is either given by φ(x) = x, φ(x) = 0, φ(x) = ∞, or there exists a 1-homogenic continuous function ℓ : Rk 7→ Rk such that ℓ−1 ◦ φ ◦ ℓ is either φ1 or φ are given, respectively, for j = 1, . . . , k, by , where the j-th coordinates of φ1 and φ ∞ ∞ (φ1(x))j = x2 i + k · xj (xi + 1)2 , (φ ∞ (x))j = dj(cid:16) kXi=1 xi(cid:17)2 + xj; kPi=1 kPi=1 here d = (d1, d2, . . . , dk) is a fixed in advance non-zero vector such thatPk The main result in [Al10] claims that this is true when k = 2. i=1 di = 0. 5.3. Birational geometry. Rational flows in dimension k ≥ 3. As mentioned in ([Al12], Subsection 5.6), in order to understand the algebro-geometric nature of projective flows, one needs to investigate the structure of rational solutions of (1) in dimension k ≥ 3 over C or any other algebraically closed field of characteristic 0 (this is the easiest case of the theory). In particular, we pose Problem 5. Let φ(x, y, w) = u(x, y, w)• v(x, y, w)• t(x, y, w) 6= x• y • w be a triplet of rational functions such that φ satisfies (1) and the boundary condition (2). Is it true that then there exists a 1-homogenic birational transformation of C3, call it ℓ, such that ℓ−1 ◦ φ ◦ ℓ(x, y, w) = x(w + 1)N−1 • y w + 1 • w w + 1 for a certain non-negative integer N? This is certainly a hard problem, unless some more powerful techniques than the ones used in [Al12] are applied. Moreover, this problem, if solved, would exhibit only basic invariants of a k-dimensional rational projective flows. For example, even in dimension 2 there are many interesting characteristics of projective flows, such as surface (compact or non-compact) on which a flow is naturally defined; symmetries; limits under conjugation; orbits; fixed points; homotopy of flows, and many more [Alpr1]. As a continuation of the previous problem and as a generalizaion of our Theorem 1, we pose Problem 6. Let k ≥ 3. Classify all k-dimensional smooth projective flows such that they are non-singular, i.e. satisfy (2), they have rational vector field, and are unramified. Projective translation equation 30 5.4. Quasi-rational flows of arbitrary genus. The results of Subsections 3.6 and 4.2 suggest that projective rational flows can be generalized purely algebraically without appeal to elliptic, abelian or other transcendental functions. We will define the notion of 2-dimensional projective quasi-rational flow of genus g, quasi-flow in short. Suppose, given a pair of 2-homogenic rational functions (x, y) and (x, y). Suppose there exists a positive integer N such that the ODE Nf (x)(x, 1) + f ′(x)[(x, 1) − x(x, 1)] = 0 (32) has a rational solution f (x). Let N be the smallest such integer, and let W (x, y) = yN f ( x y ) = P (x, y)Q(x, y)−1, where P and Q are homogenic polynomials of degrees N + d and d for some d ≥ 0, and their ratio is irreducible over C. The orbits of the flow φ = u • v defined by (4) and below are then the curves W (x, y) = c, c ∈ C ∪ {∞}. Of course, φ itself may turn out to be multi-valued analytic function with many branching points. Suppose the genus of the projective curve P (x, y) = zN Q(x, y) is g ≥ 0. Definition 2. Given a pair of 2-homogenic rational functions • such that the above assump- tion concerning ODE (32) does hold, N being the smallest such positive integer. The rational function U(A, B; x, y) is called a pre-quasi-flow of genus g with vector field • , if (i) U is 1-homogenic: (ii) The function U satisfies the PDE in the quotient ring, i.e. the following holds: U(zA, zB; zx, zy) = zU(A, B; x, y); UA(A, B) + UB(A, B) + Ux(x, y) + Uy(x, y) ≡ 0 mod(cid:16)P (A, B)Q(x, y) − P (x, y)Q(A, B)(cid:17). (33) To define a quasi-flow, not just a pre-quasi-flow, we need to specify the boundary conditions exactly, to define a pair of such functions U(A, B; x, y) • V (A, B; x, y), both of which satisfy the above definition but which are subject to different boundary conditions. For this purpose we need to delve deeper into a certain system of ODE's. This topic is the central topic of the prospective paper [Alpr2]. Here we confine with the example T given in the Subsection 4.2 and the examples to follow now. For example, it turns out that any rational flow gives rise to a quasi-flow of genus 0. We will show this in case of the canonical flow φN , see the Subsection 1.1; the general case will be treated in [Alpr1]. Since the vector field of the flow φN is (N − 1)xy • (−y2), and W (x, y) = xyN−1, the general strategy of the Subsection 4.2 shows that we need to solve the system a(u)b(u)N−1 = 1, a′(u) = −(N − 1)a(u)b(u), b′(u) = b2(u).  So, we can choose a = −uN−1, b = −u−1. In our case ς = (xyN−1) −xyN−1, B = b( ς) ς = −1. Thus, A(y − B)N−1 = U(A, B; x, y); = 1 N . So, let A = a( ς) ς = = V (A, B; x, y). x(y + 1)N−1 = − yN−1 y y + 1 yB B − y We check that now the l.h.s. of (33) in both cases is identically 0, so the congruence holds all the more. The bondary conditions read as U(−zN xyN−1,−1; xz, yz) z lim z→0 = x, lim z→0 V (−zN xyN−1,−1; xz, yz) z = y. 31 G. Alkauskas Note that the function U, even if we enforce three conditions - (i), (ii), and the above boundary condition - is not uniquely defined. For example, all three are also satisfied by x(y + 1)N−1 = x(B − y)N−1 BN−1 = U (A, B; x, y) = −U(x, y; A, B). Note also that V (A, B; x, y) = −V (x, y; A, B). Thus, rational flows give rise to quasi-flows of genus 0. The converse is not true. We will show this on the example of the flow φe, as given by the Proposition 1. Let therefore N = 1, (x, y) = (x2 − y2)/2, (x, y) = (y2 − x2)/2. Then f (x) = x + 1, and P (x, y) = x + y, Q(x, y) = 1. The curve x + y = z is a line and thus is of genus 0. The system has a solution a(u) = 1 2 − e−u, b(u) = 1 2 + e−u. In this case ς = x + y, so let a(u) + b(u) = 1, 2a′(u) = b2(u) − a2(u), 2b′(u) = a2(u) − b2(u),  A = a( ς) ς = (x + y)(cid:16)1 B = b( ς) ς = (x + y)(cid:16)1 2 2 − e−(x+y)(cid:17), + e−(x+y)(cid:17). So, according to Proposition 1, ue(x, y) = and 1 2(cid:0)(x − y)ex+y + x + y(cid:1) = x2 − y2 B − A + x + y 2 = U(A, B; x, y) = V (A, B; y, x), UA(A, B) + UB(A, B) + Ux(x, y) + Uy(x, y) = (x2 − y2)(x + y − A − B) B − A , = y; and so (33) holds. The boundary conditions now read as V (− 1 2 (x + y)z, 3 2(x + y)z; xz, yz) z U(− 1 = x, lim z→0 lim z→0 2 (x + y)z, 3 2(x + y)z; xz, yz) z (since both functions are 1-homogenic, we have an identity and not just the limit). Nevertheless, this quasi-flow arises from the flow φe which is not rational, though the orbits are lines. From the other side, given a projective curve W (x, y) = zN . There are infinitely many pairs of 2-homogenic rational functions and such that So, there are potentially many flows having the same orbits. Wx(x, y)(x, y) + Wy(x, y)(x, y) = 0. Problem 7. Develop a theory of quasi-flows. In particular, find all quasi-flows with orbits given by W (x, y) = c for any curve of arbitrary genus. Find a condition on and which, if satisfied, leads to a quasi-flow. As was proved in [Al12], the pair • leads to a rational flow if and only if yy(x, y) − xy(x, y) yx(x, y) − xx(x, y) = ax + by cx + dy Projective translation equation 32 for certain a, b, c, d ∈ C satisfying a 6= d, and (a + d)2 − 4bc (a − d)2 = M 2, for a certain M ∈ N. In this case, the integer M is exactly the level of a flow. We therefore ask for a similar criterio for quasi-flow. 5.5. Projective flows over finite fields. For simplicity, consider a finite field Fp, p being a a + ∞ = ∞, a ∞ prime. Let bFp = Fp ∪ {∞}, and extend the sum and the multiplication operations, initially defined on Fp, to bFp using the natural definition: that is, if a ∈ Fp \ {0}, then a · ∞ = ∞, find all 1−dimenional projective flows over finite fields. Suppose, f :bFp 7→bFp satisfies = 0, 0 + ∞ = ∞, ∞ · ∞ = ∞, but 0 · ∞, ∞ + ∞ are undefined. We will now 1 − z (1 − z)f (x) = f(cid:16)f (xz) z (cid:17), Then f (x) ≡ 0, f (x) ≡ ∞, or there exists a ∈ Fp, such that f (x) = x(ax + 1)−1. Indeed, assume f (∞) = b, b 6= 0,∞. Then substitution x = ∞ into the functional equation, for z 6= 0, gives (1 − z)b = f (b(1 − z)z−1), and the answer follows for a = b−1. The cases f (∞) = 0 or ∞ are dealt similarly. So, there are exactly p + 2 projective 1-dimensional flows over the finite field Fp, p of them are non-singular. Now consider the dimension 2. For further convenience we note that (1) implies the identity x, z ∈bFp. (34) n φ ◦ · · · ◦ φ (x) = φ(nx), n ∈ N; n {z } this is valid in any dimension over any field provided that n is not a multiple of the characteristic of this field. Returning to finite fields and dimension 2, we encounter the interesting problem to investigate where the projective flow is defined. For example, the flows (x − y)2 + x • (x − y)2 + y, are defined and take values on spaces F2 respectively. In fact, these maps are automorphisms of corresponding spaces (in this case this simply means they are bijections). On the other hand, let p be an odd prime. The nature of the flow p, (bFp)2 and P 2(Fp), of cardinalities p2, (p+1)2, p2+p+1, x + y + 1 • x + y + 1 y + 1 , x x + 1 • y x y φp(x, y) = x2 + y2 + 2x (x + 1)2 + (y + 1)2 • x2 + y2 + 2y (x + 1)2 + (y + 1)2 depends on the arithmetic of p. If(cid:16)−1 Consider now the case (cid:16) −1 p(cid:17) = −1, this flow is the bijection from F2 p (cid:17) = 1; for example, take p = 5. For (x, y) ∈ F2 p ∪ {∞} onto itself, where φ(−1,−1) = ∞, φ(∞) = 1 • 1; this space ("sphere") is of cardinality p2 + 1. 5, the function φ is defined for 16 pairs, for 9 pairs it is undefined, and it takes also 16 and leaves 9 values. It appears that the only meaningful completion is to add 11 additional points. For example, since φ(3 • 1) is undefined, let φ(3 • 1) = η. Further, by the projective flow property (34) φ(3 • 1) = φ ◦ φ(4 • 3) = 2−1φ(2(4 • 3)) = 2−1φ(3 • 1), so 2η = η. So, η can be formally denoted by ∞ • 0. Further, let φ(0 • 1) = α. Then φ(α) = φ ◦ φ(0 • 1) = 2−1φ(2(0 • 1)) = 2−1φ(0 • 2) = 3φ(0 • 2). More calculations confirm that α can be formally denoted by 1 • ∞. Once again, φ(η) = φ ◦ φ(3 • 1) = 2−1φ(2(3 • 1)) = 2−1φ(1 • 2) = 2−1(4 • 3) = 2 • 4. Also, φ ◦ φ ◦ φ(3 • 1) = 3−1φ(3(3 • 1)) = 3−1φ(4 • 3) = 1 • 2. By direct similar calculations, we thus complete the Table 3. The symbols 1•∞ and 0•∞, for example, obey the rules 2(1•∞) = 2•∞ 33 G. Alkauskas 1 2 x\y 0 ∞ 3 0 • 0 2 • 0 1 • ∞ ∞ • 1 0 4 • 2 1 ∞ • 2 3 • 3 4 • 3 0 • ∞ 1 • 0 ∞ • 3 2 3 • ∞ 3 • 4 4 • 4 ∞ • 4 3 • 0 3 2 • 3 0 • 2 ∞ • 0 2 • ∞ 2 • 2 0 • 1 3 • 1 ∞ • ∞ 0 • 4 2 • 1 4 • 1 4 ∞ 2 • 4 4 • ∞ 4 • 0 1 • 1 3 • 2 Table 3. The flow φp over the finite field F5 4 1 • 4 1 • 2 1 • 3 0 • 3 and 3(0 • ∞) = 0 • ∞. So, the case(cid:16) −1 p (cid:17) = 1 corresponds a bijection of (bFp)2 onto itself and thus to torus rather than a sphere. On the other hand, the analysis of the flow x(x2 + y2) x2y + xy2 + x2 + y2 • x(x2 + y2) x2y + xy2 + x2 + y2 leads to investigations of the arithmetic of genus 0 cubic x2y + xy2 + x2 + y2 = 0 over Fp. Further, it is obvious that the number of projective flows over a finite field in any dimension is finite. Finally, note that the exponential flow φexp has no analogue in the finite field setting; for example, the flow x2y • y is not property defined on (Fp)2 for p odd. So, we formulate the following Problem 8. Let k ≥ 2. Describe all projective flows over (Fp)k. What is the total number of flows? How they distribute among different completions of (Fp)k? Do all these flows arise as reductions mod p of rational flows over Q (like φp above)? References [Ac66] J. Acz´el, Lectures on functional equations and their applications, Mathematics in Science and Engi- neering, Vol. 19 Academic Press, New York-London 1966. [Ah48] N.I. Ahiezer, N. I. `Elementy teorii `ellipticeskih funkcii. (Russian) [Elements of the Theory of Elliptic Functions] Gosudarstv. Izdat. Tehn.-Teor. Lit., Moscow-Leningrad, 1948. 291 pp. [Al10] G. Alkauskas, Multi-variable translation equation which arises from homothety. Aequationes Math. 80 (3) (2010), 335 -- 350; arxiv.org/abs/0911.1513. [Al12] G. Alkauskas, The projective translation equation and rational plane flows. I (submitted); arxiv.org/abs/1201.0894. [Alpr1] G. Alkauskas, The projective translation equation and rational plane flows. II (in preparation) [Alpr2] G. Alkauskas, The projective translation equation: rational vector fields and quasi-flows (tentative title). [Di90] A.C. Dixon, On the doubly periodic functions arising out of the curve x3 + y3 − 3αxy = 1, Quart. J. [Du79] D. Dumont, A combinatorial interpretation for the Schett recurrence on the Jacobian elliptic functions. XXIV (1890). 167 -- 233. Math. Comp. 33 (1979), 1293 -- 1297. [Du81] D. Dumont, Une approche combinatoire des fonctions elliptiques de Jacobi. Adv. Math. 1 (1981), 1 -- 39. [Du86] D. Dumont, Grammaires de William Chen et d´erivations dans les arbres et arborescences, S´em. Lothar. Combin. 37 (1996), Art. B37a, 21 pp. (electronic). [BAT] A. Erd´elyi, W. Magnus, F. Oberhettinger, F.G. Tricomi, Francesco G., Higher transcen- dental functions. Vol. I. Based on notes left by Harry Bateman. With a preface by Mina Rees. With a Projective translation equation 34 foreword by E. C. Watson. Reprint of the 1953 original. Robert E. Krieger Publishing Co., Inc., Mel- bourne, Fla., 1981. [BF10] R. Bacher, Ph. Flajolet, Pseudo-factorials, elliptic functions, and continued fractions. Ramanujan J. 21 (1)(2010), 71 -- 97; arxiv.org/abs/0901.1379. [BF11] P. Blasiak, Ph. Flajolet, Combinatorial models of creation-annihilation. S´em. Lothar. Combin. 65 (2011), Art. B65c; arxiv.org/abs/1010.0354. [FGP05] Ph. Flajolet, J. Gabarr´o, H. Pekari, Analytic urns. Ann. Probab. 33 (3) (2005), 1200 -- 1233; arxiv.org/abs/math/0407098. [FCF05] E.van Fossen Conrad, Ph. Flajolet, The Fermat cubic, contin- ued fractions, and a combinatorial excursion. S´em. Lothar. Combin. 54 (2005/07), Art. B54g; arxiv.org/abs/math/0507268. elliptic functions, [FDP06] Ph. Flajolet, Ph. Dumas, V. Puyhaubert, Some exactly solvable models of urn process theory. Fourth Colloquium on Mathematics and Computer Science Algorithms, Trees, Combinatorics and Prob- abilities, Discrete Math. Theor. Comput. Sci. Proc., AG, Assoc. Discrete Math. Theor. Comput. Sci., Nancy, (2006), 59 -- 118. [FR08] H. Fripertinger, L. Reich, The formal translation equation and formal cocycle equations for iteration groups of type I. Aequationes Math. 76 (1-2) (2008), 54 -- 91. [FR10] H. Fripertinger, L. Reich, The formal translation equation for iteration groups of type II. Aequa- tiones Math. 79 (1-2) (2010), 111 -- 156. [Kn92] A.W. Knapp, Elliptic curves, Mathematical Notes, 40. Princeton University Press, Princeton, NJ, 1992. [LS87] M.A. Lavrent'ev, B.V. Shabat, Metody teorii funktsii kompleksnogo peremennogo. (Russian) [Meth- ods of the theory of functions in a complex variable] Fifth edition. Nauka, Moscow, 1987. 688 pp. [Sch76] A. Schett, Properties of the Taylor series expansion coefficients of the Jacobian elliptic functions. Math. Comp. 30 (1976), 143 -- 147. [Mo73] Z. Moszner, The translation equation and its application. Demonstratio Math. 6 (1973), 309 -- 327. [Mo95] Z. Moszner, General theory of the translation equation. Aequationes Math. 50(1-2) (1995), 17 -- 37. [Vie80] G. Viennot, Une interpretation combinatoire des coefficients des d´eveloppments en s´erie enti´ere des fonctions elliptiques de Jacobi. J. Combin. Theory Ser. B. 29 (1980), 121 -- 133. Vilnius University, Department of Mathematics and Informatics, Naugarduko 24, LT-03225 Vilnius, Lithuania E-mail address: [email protected]
1701.00511
3
1701
2018-08-07T19:26:53
Frobenius-Seshadri constants and characterizations of projective space
[ "math.AG", "math.AC" ]
We introduce higher-order variants of the Frobenius-Seshadri constant due to Musta\c{t}\u{a} and Schwede, which are defined for ample line bundles in positive characteristic. These constants are used to show that Demailly's criterion for separation of higher-order jets by adjoint bundles also holds in positive characteristic. As an application, we give a characterization of projective space using Seshadri constants in positive characteristic, which was proved in characteristic zero by Bauer and Szemberg. We also discuss connections with other characterizations of projective space.
math.AG
math
FROBENIUS -- SESHADRI CONSTANTS AND CHARACTERIZATIONS OF PROJECTIVE SPACE TAKUMI MURAYAMA Abstract. We introduce higher-order variants of the Frobenius -- Seshadri constant due to Mustat¸a and Schwede, which are defined for ample line bundles in positive characteristic. These constants are used to show that Demailly's criterion for separation of higher-order jets by adjoint bundles also holds in positive characteristic. As an application, we give a characterization of projective space using Seshadri constants in positive characteristic, which was proved in characteristic zero by Bauer and Szemberg. We also discuss connections with other characterizations of projective space. 1. Introduction Let L be an ample line bundle on a smooth projective variety X defined over an algebraically closed field k. Demailly in [Dem92, §6] introduced the Seshadri constant ε(L; x), which measures the local positivity of L at a closed point x ∈ X. If µ : X ′ → X is the blow-up of X at x with exceptional divisor E, then the Seshadri constant is Seshadri constants have received much attention since their inception: see [Bau+09] and [Laz04, Ch. 5]. ε(L; x) := sup(cid:8)t ∈ R≥0(cid:12)(cid:12) µ∗(L)(−tE) is nef(cid:9). Part of this interest in Seshadri constants stems from the fact that they give effective positivity statements for adjoint bundles. We will be particularly interested in how Seshadri constants can determine when adjoint bundles separate higher-order jets. Recall that we say L separates ℓ-jets if the restriction map H 0(X, L) −→ H 0(X, L ⊗ OX /m ℓ+1 x ) (1) is surjective, where mx ⊂ OX is the ideal defining x. Algebraically, this says ℓth-order Taylor polynomials at x are restrictions of global sections, and geometrically, this says L separates ℓth- order tangent directions at x. We can now state our first main result. Let ωX denote the canonical bundle on X. Theorem A. Let L be an ample line bundle on a smooth projective variety X of dimension n defined over an algebraically closed field of positive characteristic. If ε(L; x) > n + ℓ at a closed point x ∈ X, then ωX ⊗ L separates ℓ-jets at x. Demailly showed this result in characteristic zero using the Kawamata -- Viehweg vanishing theo- rem [Dem92, Prop. 6.8(a)]. Our contribution is that the same result holds in positive characteristic. As an application of this result, we prove the following: Theorem B. Let X be a smooth Fano variety of dimension n defined over an algebraically closed field of positive characteristic. If there exists a closed point x ∈ X with ε(ω−1 X ; x) ≥ n + 1, then X is isomorphic to the n-dimensional projective space Pn. 2010 Mathematics Subject Classification. Primary 14C20; Secondary 13A35, 14E25, 14M20. Key words and phrases. Seshadri constants, Frobenius, separation of jets, projective space. This material is based upon work supported by the National Science Foundation under Grant No. DMS-1265256. 1 2 TAKUMI MURAYAMA Bauer and Szemberg showed the analogous statement in characteristic zero as an application of the proof of Demailly's result [BS09, Thm. 2]. One interesting feature of this theorem is that it only requires a positivity condition on the anti-canonical bundle ω−1 X at one point x ∈ X. In characteristic zero, Liu and Zhuang in [LZ18, Thm. 2] generalized [BS09, Thm. 2] to Q-Fano varieties; see Remark 4.10 for a comparison between their result and Theorem B. There is an interesting connection between Theorem B and the Mori -- Mukai conjecture, which states that if X is a smooth Fano variety of dimension n such that the canonical divisor KX satisfies (−KX · C) ≥ n + 1 for all rational curves C ⊆ X, then X is isomorphic to Pn. In characteristic zero, Cho, Miyaoka, and Shepherd-Barron's proof of the conjecture [CMSB02, Cor. 0.4] implies [BS09, Thm. 2]; see Proposition 4.6(i). In positive characteristic, the conjecture is still open, so instead one must assume the lower bound in Theorem B holds at all closed points in order to use a weaker result due to Kachi and Koll´ar [KK00, Cor. 3]; see Proposition 4.6(ii). On the other hand, this connection raises the question of whether the opposite relationship holds. More precisely, we ask the following: Question. Let X be a smooth Fano variety of dimension n defined over an algebraically closed field. If (−KX · C) ≥ n + 1 for every rational curve C ⊆ X, then is there a closed point x ∈ X with (−KX · C) ≥ (multx C) · (n + 1) for all reduced and irreducible curves C ⊆ X passing through x? This latter condition is equivalent to having ε(ω−1 In characteristic zero, [CMSB02, Cor. 0.4] answers this question affirmatively. If one could do this independently of their result, then [BS09, Thm. 2] would give an alternative proof of the Mori -- Mukai conjecture in characteristic zero, and Theorem B would resolve the conjecture in positive characteristic. See §4.1 for further discussion. X ; x) ≥ n + 1 by [Laz04, Prop. 5.1.5]. The proofs of Theorems A and B use a new variant of the Seshadri constant. To motivate our definition, we recall the following alternative characterization of the Seshadri constant in terms of if for each integer m, we denote by s(Lm; x) the largest separation of jets [Laz04, Prop. 5.1.17]: integer ℓ such that Lm separates ℓ-jets, then we have the equalities ε(L; x) = sup m≥1 s(Lm; x) m = lim m→∞ s(Lm; x) m . In positive characteristic, Mustat¸a and Schwede [MS14] defined the Frobenius -- Seshadri constant essentially by replacing ordinary powers of mx in the definition of separation of jets (1) with Frobenius powers, in order to take advantage of the Frobenius morphism. Using their definition, Mustat¸a and Schwede were able to recover Theorem A when ℓ = 0, and deduce global generation and very ampleness results for adjoint bundles [MS14, Thm. 3.1]. However, the full statement of Theorem A remained out of reach; see Remark 3.1. Our solution is to introduce higher-order variants of the Frobenius -- Seshadri constant, which mix both ordinary and Frobenius powers of mx. This allows us to more directly deduce separation of higher-order jets. For each integer ℓ ≥ 0 and m ≥ 1, let sℓ F (Lm; x) be the largest integer e ≥ 0 such that the restriction map H 0(X, Lm) −→ H 0(cid:0)X, Lm ⊗ OX /(m )[pe] denotes the eth Frobenius power of mℓ+1 )[pe](cid:1) ℓ+1 x . Then, the ℓth Frobenius -- x is surjective, where (mℓ+1 Seshadri constant of L at x is x εℓ F (L; x) := sup m≥1 psℓ F (Lm;x) − 1 m/(ℓ + 1) = lim sup m→∞ psℓ F (Lm;x) − 1 m/(ℓ + 1) . These constants are related to the ordinary Seshadri constant in the following manner (Proposition 2.9): ℓ + 1 ℓ + n · ε(L; x) ≤ εℓ F (L; x) ≤ ε(L; x). (2) FROBENIUS -- SESHADRI CONSTANTS AND CHARACTERIZATIONS OF PROJECTIVE SPACE 3 Note that the ℓth Frobenius -- Seshadri constant εℓ ε(L; x) as ℓ → ∞. F (L; x) converges to the ordinary Seshadri constant Using the ℓth Frobenius -- Seshadri constant, we prove the following statement en route to proving Theorem A: Theorem C. Let L be an ample line bundle on a smooth projective variety X defined over an algebraically closed field of positive characteristic. If εℓ F (L; x) > ℓ + 1 at a closed point x ∈ X, then ωX ⊗ L separates ℓ-jets at x. By the comparison (2), Theorem C immediately implies Theorem A. Our proof of Theorem C also works for singular varieties that are F -injective; see Remark 3.3. F -injective varieties are related to varieties with Du Bois singularities in characteristic zero [Sch09b]. Given Theorem C, it would be very interesting to have non-trivial lower bounds for any of the aforementioned versions of the Seshadri constant at very general points of X. In characteristic zero, it is conjectured that ε(L; x) ≥ 1 at all very general x ∈ X. This is known if n = dim X = 2 [EL93]; if n ≥ 3, then only the lower bound 1/n is known [EKL95, Thm. 1]. However, the proofs of both results rely heavily on the characteristic zero assumption, and in arbitrary characteristic, we are only aware of the lower bound ε(L; x) ≥ 2/(cid:0)1 +p4σ(L) + 13(cid:1) for arbitrary points on surfaces, where is the canonical slope of L [Bau99, Thm. 3.1]. σ(L) := inf(cid:8)s ∈ R(cid:12)(cid:12) sL − KX is nef(cid:9) Outline. Our paper is structured as follows: In §2, we define the ℓth Frobenius -- Seshadri constant, and prove its basic properties. Most of what we prove is modeled after [MS14, §2], which studies what would be the zeroth Frobenius -- Seshadri constant in our notation. In §3, we prove Theorem C. The main technical tool is the trace map T : F∗ωX → ωX associated to the (absolute) Frobenius morphism. Finally, in §4, we prove Theorem B, following [BS09]. Notation. A variety is a reduced and irreducible separated scheme of finite type defined over an algebraically closed field k of characteristic p > 0, unless stated otherwise. We denote by X a positive-dimensional projective variety, and denote by F : X → X the (absolute) Frobenius morphism, which is given by the identity map on points, and the p-power map OX (U ) F∗OX (U ) f f p on structure sheaves, where U ⊆ X is an open set. If a ⊆ OX is a coherent ideal sheaf, we define the eth Frobenius power a[pe] to be the inverse image of a via the eth iterate of the Frobenius morphism. Locally, if a is generated by (hi)i∈I , then a[pe] is generated by (hpe i )i∈I . If X is smooth, we denote by ωX the canonical bundle on X and KX the canonical divisor on X. 2. Definitions and preliminaries We start by recalling the definition of the (ordinary) Seshadri constant of a line bundle L at a point. We adopt the "separation of jets" description of the Seshadri constant as our definition, which is equivalent to the other definitions when the line bundle L is ample and the closed point x is smooth [Laz04, Prop. 5.1.17]. Definition 2.1. Let L be a line bundle on a projective variety X, and let x ∈ X be a closed point with defining ideal mx ⊂ OX . For all integers ℓ ≥ 0 and m ≥ 1, we say that Lm separates ℓ-jets at x if the restriction map ρℓ,0 Lm : H 0(X, Lm) −→ H 0(X, Lm ⊗ OX/m ℓ+1 x ) 4 TAKUMI MURAYAMA is surjective. Let s(Lm; x) be the largest integer ℓ ≥ 0 such that Lm separates ℓ-jets at x; if no such ℓ exists, set s(Lm; x) = −∞. The Seshadri constant of L at x is ε(L; x) := sup m≥1 s(Lm; x) m . (3) We now give our main definition, which is modeled after the above interpretation of the Seshadri constant in terms of separation of jets. This definition combines both ordinary and Frobenius powers of the ideal mx. Compared to the Frobenius -- Seshadri constant defined in [MS14, Def. 2.4], our definition has the advantage of directly encoding information about higher-order jets; see Remark 3.1. Definition 2.2. Let L be a line bundle on a projective variety X, and let x ∈ X be a closed point with defining ideal mx ⊂ OX . For all integers ℓ, e ≥ 0 and m ≥ 1, we say that Lm separates pe-Frobenius ℓ-jets at x if the restriction map ρℓ,e Lm : H 0(X, Lm) −→ H 0(cid:0)X, Lm ⊗ OX /(m ℓ+1 x )[pe](cid:1) is surjective. Let sℓ at x; if no such e exists, set sℓ F (Lm; x) be the largest integer e ≥ 0 such that Lm separates pe-Frobenius ℓ-jets F (Lm; x) = −∞. The ℓth Frobenius -- Seshadri constant of L at x is εℓ F (L; x) := sup m≥1 psℓ F (Lm;x) − 1 m/(ℓ + 1) . (4) We refer to the constants εℓ Seshadri constant ε0 F (L; x) as Frobenius -- Seshadri constants. Note that the zeroth Frobenius -- F (L; x) is the Frobenius -- Seshadri constant defined in [MS14, Def. 2.4]. In the rest of this section, we will prove basic formal properties about Frobenius -- Seshadri con- stants, following [MS14, §2]. The only statements used explicitly in later sections are Lemmas 2.4 and 2.7, and Propositions 2.5(i) and 2.9. 2.1. Separation of Frobenius jets under tensor powers. For the ordinary Seshadri constant, the supremum in (3) is actually a limit when L is ample [Dem92, p. 97]. This property follows from Fekete's lemma [PS98, Pt. I, no 98], since for all positive integers m and n, the sequence s(Lm; x) satisfies the superadditivity property (see, e.g., [Ito13, Lem. 3.7] for a proof) s(Lm+n; x) ≥ s(Lm; x) + s(Ln; x). (5) For Frobenius -- Seshadri constants, we cannot have an analogous property, since the supremum in (4) may not be a limit; see Example 2.6. Our first goal is to find a replacement for this superadditivity property. This will allow us to show that the supremum in (4) is actually a limit supremum. We start with the following observation about ideals in a ring of characteristic p > 0, which will also be useful later. Note that the first inclusion in (6) is a slight improvement on [Sch09a, Lem. 4.6]. Lemma 2.3. Let R be a commutative ring of characteristic p > 0. Then, for any ideal a generated by n elements and for any non-negative integers e and ℓ, we have the sequence of inclusions ℓpe+n(pe−1)+1 ⊆ (a ℓ+1)[pe] ⊆ a a (ℓ+1)pe . (6) Moreover, if R is a regular local ring of dimension n, and a is the maximal ideal of R, then ℓpe+n(pe−1) 6⊆ (a ℓ+1)[pe]. a Proof. The second inclusion in (6) is clear; we want to show the first inclusion. Let y1, y2, . . . , yn be a set of generators for a. The ideal aℓpe+n(pe−1)+1 is generated by all elements of the form yai i nYi=1 such that nXi=1 ai = ℓpe + n(pe − 1) + 1, (7) FROBENIUS -- SESHADRI CONSTANTS AND CHARACTERIZATIONS OF PROJECTIVE SPACE 5 and the ideal (aℓ+1)[pe] is generated by all elements of the form ypebi i such that bi = ℓ + 1. (8) nYi=1 nXi=1 We want to show that the elements (7) are divisible by some elements of the form (8). By the division algorithm, we may write ai = ai,0 + pea′ i such that 0 ≤ ai,0 ≤ pe − 1. Then, i for some non-negative integers ai,0 and a′ yai i = nYi=1 and since ai,0 ≤ pe − 1, we have thatPn which implies ℓ + p−e ≤Pn that ℓ + 1 ≤Pn i, i.e., the element Qn i=1 a′ i=1 a′ ℓpe + n(pe − 1) + 1 = nYi=1 nXi=1 yai,0 i · pea′ i i y , nYi=1 ai ≤ n(pe − 1) + pea′ i, nXi=1 i=1 ai,0 ≤ n(pe − 1). Thus, we have the inequality i. Since the right-hand side of this inequality is an integer, we have is divisible by one of the form (8). Thus, each i=1 y pea′ i i element of the form in (7) is divisible by one of the form in (8). Now suppose R is a regular local ring of dimension n, and a is the maximal ideal of R. Let y1, y2, . . . , yn be a regular system of parameters. Then, we have yℓpe i0 · ype−1 i ℓpe+n(pe−1) ∈ a nYi=1 for any i0 ∈ {1, 2, . . . , n}. This monomial does not lie in (aℓ+1)[pe] since its image is not in the extension of (aℓ+1)[pe] in the completion of R at a, which is isomorphic to a formal power series ring with variables y1, y2, . . . , yn by the Cohen structure theorem. (cid:3) Lemma 2.4 (cf. [MS14, Lem. 2.5]). Let L be a line bundle on a projective variety X, and let x ∈ X be a closed point. Suppose Lm separates pe-Frobenius ℓ-jets at x for some integers m ≥ 1, e ≥ 1, and ℓ ≥ 0, and denote dr = pre − 1 pe − 1 for each positive integer r. Then, Lmdr separates pre-Frobenius ℓ-jets at x for all r ≥ 1. Proof. We want to show the restriction maps are surjective for all positive integers r. We prove this by induction on r. The case r = 1 is true by assumption, so we consider the case when r ≥ 2. ϕr : H 0(X, Lmdr ) −→ H 0(cid:0)X, Lmdr ⊗ OX/(m )[pre](cid:1) Let ey1,ey2, . . . ,eyn generate mx · OX,x. After choosing an isomorphism Lm )[pre] ≃ OX /(m Lmdr ⊗ OX /(m )[pre]. ℓ+1 x ℓ+1 x ℓ+1 x x ≃ OX,x, we can make the identification Then, Lmdr ⊗ OX /(mℓ+1 x )[pre] is generated as a vector space over k by the residue classes proof of Lemma 2.3, write ai = ai,0 + pea′ 1 eya2 of the monomials eya1 2 · · ·eyan 2 · · · yan ya1 1 ya2 n ∈ Lmdr x n ∈ Lmdr ⊗ OX /(m , where the ai can be any non-negative integers. As in the (9) ℓ+1 x )[pre] i for each i, where 0 ≤ ai,0 ≤ pe − 1, so that ya1 1 ya2 2 · · · yan n = yai,0 i · nYi=1 pea′ i i y . nYi=1 6 TAKUMI MURAYAMA We will show that the elements in (9) lie in the image of ϕr by descending induction on S :=Pn By Lemma 2.3, if S ≥ (ℓ+n)p(r−1)e, then ya1 to show. It therefore suffices to consider the inductive case, when S ≤ (ℓ + n)p(r−1)e − 1. i=1 a′ i. )[pre], and so there is nothing n ≡ 0 mod (mℓ+1 2 · · · yan 1 ya2 x By the assumption that ϕ1 is surjective, we know that there exists t1 ∈ H 0(X, Lm) such that its germ t1,x ∈ Lm x satisfies i − t1,x ∈ (m ℓ+1 x )[pe] ⊗ Lm x . (10) By the inductive hypothesis with respect to r, we know that ϕr−1 is surjective, so there exists t2 ∈ H 0(X, Lmdr−1 ) such that its germ t2,x ∈ Lmdr−1 satisfies x a′ i − t2,x ∈ (m i ℓ+1 x )[p(r−1)e] ⊗ Lmdr−1 x . Now consider the composition below: H 0(X, Lm) ⊗ H 0(X, Lmdr−1 ) 1⊗(F e)∗ H 0(X, Lm) ⊗ H 0(X, Lmpedr−1) mult H 0(X, Lmdr ) t1 ⊗ t2 t1 ⊗ tpe 2 t1tpe 2 Note that t1tpe 2 restricts to t1,xtpe 2,x in the stalk Lmdr x , and that 2,x ∈ (m ℓ+1 x )[pre] ⊗ Lmpedr−1 x . (11) Then, after passing to the stalk Lmdr , we have 2,x = i − t1,x pea′ i + t1,x i i − t1,xtpe nYi=1ey i − t1,x! · nYi=1ey i + t1,x nYi=1ey nYi=1ey nYi=1eyai To show that Qn First, for each monomial µ in the eyi that appears in the differenceQn some n-tuple (bi)1≤i≤n where Pn equation is in the image of ϕr modulo (mℓ+1 to zero modulo (mℓ+1 (mℓ+1 corresponding monomial that appears in the product i=1 yai )[pre]. pea′ i x x i pea′ i − t1,xtpe i 2,x pea′ i − tpe i 2,x!. is in the image of ϕr, it suffices to show that the right-hand side of this )[pre]. By (11), we know the second term is congruent )[pre], and so it remains to show the first term is in the image of ϕr modulo x i − t1,x, there exists divides µ by (10). Thus, the i=1eyai,0 nYi=1eyai,0 nYi=1ey x pea′ i − tpe i nYi=1ey nYi=1eyai = nYi=1eyai,0 i i=1eypebi i − t1,x! · i=1 bi = ℓ + 1 such that Qn nYi=1eyai,0 nYi=1ey nXi=1 . Since pe(a′ i i+bi) (a′ i + bi) = S + ℓ + 1 > S, pea′ i i is divisible by the productQn i=1ey each monomial that appears in the product (12) is therefore in the image of ϕr modulo (mℓ+1 by the inductive hypothesis on S. x This allows us to show that the supremum in (4) can actually be computed as a limit supremum. (12) )[pre] (cid:3) FROBENIUS -- SESHADRI CONSTANTS AND CHARACTERIZATIONS OF PROJECTIVE SPACE 7 Proposition 2.5 (cf. [MS14, Prop. 2.6]). Let ℓ ≥ 0 be fixed, and let L be an ample line bundle on a projective variety X. Let x ∈ X be a closed point. (i) The line bundle Lm separates pe-Frobenius ℓ-jets at x for some positive integers m and e. (ii) We have εℓ F (L; x) = sup m,e pe − 1 m/(ℓ + 1) , where the supremum is taken over all positive integers m and e such that Lm separates pe-Frobenius ℓ-jets at x. (iii) Given any δ > 0, there is a positive integer e0 such that for every positive integer e divisible by e0, there is a positive integer m such that Lm separates pe-Frobenius ℓ-jets at x and (iv) We have pe − 1 m/(ℓ + 1) > εℓ F (L; x) − δ. εℓ F (L; x) = lim sup m→∞ psℓ F (Lm;x) − 1 m/(ℓ + 1) . (13) Proof. For (i), let m ≥ 1 be such that Lm is very ample, and let n be the number of generators of mx · OX,x. Then, Lm separates tangent directions (i.e., 1-jets) and so Lm(ℓpe+n(pe−1)) separates (ℓpe + n(pe − 1))-jets by the superadditivity property (5). Thus, Lm(ℓpe+n(pe−1)) separates pe- Frobenius ℓ-jets at x by the inclusion m )[pe] in Lemma 2.3. ℓpe+n(pe−1)+1 x ⊆ (mℓ+1 x Assertion (ii) follows by (i) and the definition of the ℓth Frobenius -- Seshadri constant, since sℓ F (Lm; x) is defined as the maximum e ≥ 0 such that Lm separates pe-Frobenius ℓ-jets at x. For (iii), there exist positive integers m0 and e0 such that the inequality (13) holds by (i) and pe−1 pe0 −1 . Then, by Lemma 2.4, the definition of εℓ we have that Lm separates pe-Frobenius ℓ-jets at x. The inequality (13) then follows, since F (L; x). For each multiple e = re0 of e0, let m = m0 pe − 1 m/(ℓ + 1) = pe0 − 1 m0/(ℓ + 1) > εℓ F (L; x) − δ. For (iv), let m0 and e0 be as in (iii) for δ = 1. We inductively choose an increasing sequence of positive integers (mr)r≥0 as follows: having chosen mr, we choose mr+1 such that (13) holds with δ = 1/(r + 1), and such that mr < mr+1. Note that this increasing property can be ensured by the fact that the m's in (iii) increase as e increases. Then, we have F (Lmr ;x) − 1 mr/(ℓ + 1) εℓ F (L; x) = lim r→∞ psℓ , which implies (iv). (cid:3) We now give a calculation of both the ordinary Seshadri constant and the ℓth Frobenius -- Seshadri constants on projective space, which shows that the limit supremum in Proposition 2.5(iv) cannot be computed as a limit. k , and let L = OX (1) be the line bundle associated to a hyperplane. For k , we claim that the restriction map Example 2.6. Let X = Pn every closed point x ∈ Pn ρℓ,e OX (m) : H 0(X, OX (m)) −→ H 0(cid:0)X, OX (m) ⊗ OX /(m ℓ+1 x )[pe](cid:1) is surjective if and only if m ≥ ℓpe + n(pe − 1). This claim follows from Lemma 2.3 after choosing local affine coordinates in OX,x, and observing that H 0(X, OX (m)) maps onto all monomials of degree ≤ m in OX,x. Note that the inequality m ≥ ℓpe + n(pe − 1) is equivalent to m + n ℓ + n ≥ pe. 8 TAKUMI MURAYAMA Letting e = 0 gives ε(OX (m); x) = 1. For Frobenius -- Seshadri constants, the equality in Proposition 2.5(iv) implies εℓ F (OX (1); x) ≤ lim sup m→∞ m+n ℓ+n − 1 m/(ℓ + 1) = ℓ + 1 ℓ + n · lim sup m→∞ m − ℓ m = ℓ + 1 ℓ + n , and the reverse inequality holds by computing a lower bound for the limit supremum using the sequence me = ℓpe + n(pe − 1). On the other hand, we see that the limit supremum is not a limit, since the sequence m′ e = ℓpe + n(pe − 1) − 1 gives the limit lim sup e→∞ e ;x) − 1 F (Lm′ psℓ m′ e/(ℓ + 1) = Note that in this example, we have the equality ℓ+1 2.9 that in fact, the inequality ≤ always holds. ℓ + 1 ℓ + n · lim e→∞ pe−1 − 1 pe − n+1 n+ℓ ℓ+n · ε(L; x) = εℓ = ℓ + 1 (ℓ + n)p . F (L; x). We will see in Proposition We can use Lemma 2.4 and Proposition 2.5(iv) to prove that Frobenius -- Seshadri constants are also well-behaved with respect to taking tensor powers. While we will not need this result in the sequel, the proof will use the following lemma, which will also be useful in the proof of Theorem C. Lemma 2.7 (cf. [MS14, Rem. 2.3]). Let X be a projectivty variety, and let x ∈ X be a closed point. Let L be a line bundle on X, and let M be another line bundle that is globally generated at x. If for some integers e ≥ 1 and ℓ ≥ 0, we have that L separates pe-Frobenius ℓ-jets at x, then L ⊗ M also separates pe-Frobenius ℓ-jets at x. Proof. Since M is globally generated at x, there is a global section t ∈ H 0(X, M ) that does not vanish at x. The commutative diagram H 0(X, L) t⊗− H 0(X, L ⊗ M ) ρℓ,e L ρℓ,e L⊗M x H 0(cid:0)X, L ⊗ OX /(mℓ+1 )[pe](cid:1) )[pe](cid:1) H 0(cid:0)X, L ⊗ M ⊗ OX /(mℓ+1 tx⊗− ∼ x shows that the restriction map ρℓ,e ℓ-jets. L⊗M is surjective, i.e., we have that L ⊗ M separates pe-Frobenius (cid:3) Proposition 2.8 (cf. [MS14, Prop. 2.8]). Let L be an ample line bundle on a projective variety X, and let x ∈ X be a closed point. Then, for all integers ℓ ≥ 0 and s > 0, we have εℓ F (Lr; x) = r · εℓ F (L; x). Proof. First, we have F (Lr; x) = r · sup εℓ m≥1 F (Lrm;x)−1 psℓ rm/(ℓ + 1) ≤ r · sup m′≥1 ;x)−1 F (Lm′ psℓ m′/(ℓ + 1) = r · εℓ F (L; x) by running through all tensor powers of L instead of just the powers that are divisible by r. It therefore remains to show the opposite inequality. We will fix an integer j > 0 such that Lj is globally generated. Let δ > 0 be given, and let m be a positive integer such that Lm separates pe-Frobenius ℓ-jets for some integers e, ℓ ≥ 0 such that Now let i be a positive integer. Denoting di = pie−1 pe−1 , we have pe − 1 m/(ℓ + 1) > εℓ F (L; x) − δ r . sF (Lmdi+j; x) ≥ sF (Lmdi; x) ≥ ie (14) FROBENIUS -- SESHADRI CONSTANTS AND CHARACTERIZATIONS OF PROJECTIVE SPACE 9 by Lemmas 2.4 and 2.7. Now denoting ai =(cid:24) mdi + j r (cid:25) , we have that rai ≥ mdi + j, hence psℓ F (Lrai ;x) − 1 ai/(ℓ + 1) ≥ psℓ F (Lmdi+j;x) − 1 ai/(ℓ + 1) ≥ pie − 1 ai/(ℓ + 1) = pe − 1 m/(ℓ + 1) · dim (cid:6)(mdi + j)/r(cid:7) , where the second inequality is by (14). Taking limit suprema as i → ∞ and using Proposition 2.5(iv) gives εℓ F (Lr; x) ≥ lim sup i→∞ psℓ F (Lrai ;x) − 1 ai/(ℓ + 1) ≥ r · εℓ F (L; x) − δ. Since δ > 0 was arbitrary, we have the inequality εℓ F (Lr; x) ≥ r · εℓ F (L; x). (cid:3) 2.2. A comparison with the ordinary Seshadri constant. In order to use the ℓth Frobenius -- Seshadri constant to prove Theorem A, we require a comparison with the ordinary Seshadri constant. This will allow us to deduce positivity properties of adjoint bundles in §3. Proposition 2.9 (cf. [MS14, Prop. 2.12]). If L is an ample line bundle on a projective variety X of dimension n, then for every smooth point x ∈ X and integer ℓ ≥ 0, we have the sequence of inequalities ℓ + 1 ℓ + n · ε(L; x) ≤ εℓ F (L; x) ≤ ε(L; x). In particular, εℓ F (L; x) → ε(L; x) as ℓ → ∞. Proof. Since mx · OX,x is generated by n elements, we have the sequence of inclusions by Lemma 2.3. The right inclusion in (16) implies m ℓpe+n(pe−1)+1 x ⊆ (m ℓ+1 x )[pe] ⊆ m (ℓ+1)pe x (15) (16) s(Lm; x) ≥ (ℓ + 1)psℓ F (Lm;x) − 1 ≥ (ℓ + 1)(psℓ F (Lm;x) − 1), and so the right inequality in (15) follows after dividing by m throughout, and taking limit suprema as m → ∞. For the left inequality in (15), let δ > 0 be given, and let m0 be a positive integer such that s(Lm0; x) m0 > ε(L; x) − ℓ + n ℓ + 1 · δ. Given any non-negative integer e, denote de =(cid:24) ℓpe + n(pe − 1) s(Lm0; x) (cid:25) =(cid:24) (ℓ + n)pe − n s(Lm0; x) (cid:25) . By the superadditivity property (5), we have s(Lm0de; x) ≥ de · s(Lm0; x) ≥ ℓpe + n(pe − 1). By the left inclusion in (16), this inequality implies sℓ F (Lm0de; x) ≥ e, and therefore εℓ F (L; x) ≥ F (Lm0 de ;x) − 1 psℓ m0de/(ℓ + 1) ≥ (ℓ + 1)(pe − 1) m0(cid:6)(cid:0)(ℓ + n)pe − n(cid:1)/s(Lm0 ; x)(cid:7) . 10 TAKUMI MURAYAMA As e → ∞, the right-hand side converges to hence we have the inequality ℓ + 1 ℓ + n · s(Lm0; x) m0 > ℓ + 1 ℓ + n · ε(L; x) − δ, εℓ F (L; x) > ℓ + 1 ℓ + n · ε(L; x) − δ for all δ > 0. Since δ > 0 was arbitrary, we obtain the left inequality in (15). (cid:3) In light of Example 2.6 and Theorems A and C, it seems more accurate to think of the ℓth Frobenius -- Seshadri constant as being closer to ℓ+1 ℓ+n · ε(L; x) than to ε(L; x), just as for the zeroth Frobenius -- Seshadri constant [MS14, p. 869]. We also observe that Example 2.6 shows that the lower bound in (15) is optimal. We can also compare different Frobenius -- Seshadri constants: Corollary 2.10. If L is a line bundle on an n-dimensional projective variety X, then for every smooth point x ∈ X and integers ℓ > m ≥ 0, we have ℓ + 1 ℓ + n · εm F (L; x) ≤ εℓ F (L; x) ≤ ℓ + 1 m + 1 · εm F (L; x) Proof. If Lr separates pe-Frobenius ℓ-jets at x, then it separates pe-Frobenius m-jets at x, giving the right inequality. The left inequality follows by using Proposition 2.9 for different values of ℓ. (cid:3) 2.3. Numerical invariance. We now prove that Frobenius -- Seshadri constants only depend on the numerical equivalence class of a line bundle. This fact will not be used in the sequel. Regularity in the proof below is in the sense of Castelnuovo and Mumford; see [Laz04, Def. 1.8.4] for the definition. Proposition 2.11 (cf. [MS14, Prop. 2.14]). Let X be a projective variety, and let x ∈ X be a closed point. If L1 and L2 are numerically equivalent ample line bundles on X, then εℓ F (L2; x) for all integers ℓ ≥ 0. F (L1; x) = εℓ Proof. We first claim that if A is a globally generated ample line bundle, then there exists m0 such that Am ⊗ N is globally generated for all integers m ≥ m0 and nef line bundles N . First, by Fujita's vanishing theorem [Fuj83, Thm. 5.1], there exists an integer m1 such that for all integers m ≥ m1 and nef line bundles N , we have H i(X, Am ⊗ N ) = 0 for all i > 0. Thus, if m ≥ m1 + dim X, then the line bundle Am ⊗ N is 0-regular with respect to A, hence is globally generated by [Laz04, Thm. 1.8.5(i)]. It therefore suffices to set m0 = m1 + dim X. We now prove the proposition. By hypothesis, there exists a numerically trivial line bundle P such that L2 ≃ L1 ⊗ P . Applying the result of the previous paragraph where A is a large enough power of L1, we see that there exists a positive integer j such that Lj 1 ⊗ N is globally generated for all nef line bundles N , hence in particular, Lj 1 ⊗ P i is globally generated for all integers i. Now by Proposition 2.5(iv), there exists an increasing sequence (mr)r≥0 of positive integers such that For each integer r ≥ 0, since Lmr+j see that 2 ≃ Lmr 1 ⊗ Lj 1 ⊗ P mr+j is globally generated, we εℓ F (L1; x) = lim r→∞ F (Lmr 1 psℓ ;x) − 1 . mr/(ℓ + 1) 1 ⊗ P mr+j and Lj F (Lmr sℓ 1 ; x) ≤ sℓ F (Lmr+j 2 ; x) by Lemma 2.7. We therefore have that psℓ ;x) − 1 F (Lmr psℓ 1 F (Lmr +j 2 ;x) − 1 mr/(ℓ + 1) mr/(ℓ + 1) ≤ = 2 F (Lmr +j psℓ ;x) − 1 (mr + j)/(ℓ + 1) · mr + j mr . FROBENIUS -- SESHADRI CONSTANTS AND CHARACTERIZATIONS OF PROJECTIVE SPACE 11 Since the limit of the left-hand side is εℓ F (L1; x) by choice of the sequence (mr)r≥0, taking limit suprema as r → ∞ throughout this inequality yields the inequality εℓ F (L2; x) by Proposition 2.5(iv). Finally, repeating the argument above after switching the roles of L1 and L2, we have the equality εℓ (cid:3) F (L1; x) ≤ εℓ F (L1; x) = εℓ F (L2; x). 3. Frobenius -- Seshadri constants and adjoint bundles We now turn to the proofs of Theorems A and C, which we restate below. Recall our standing assumption that our ground field k is algebraically closed and of characteristic p > 0. Theorem A. Let L be an ample line bundle on a smooth projective variety X of dimension n. Let x ∈ X be a closed point. If the inequality holds, then ωX ⊗ L separates ℓ-jets at x. ε(L; x) > n + ℓ Theorem C. Let L be an ample line bundle on a smooth projective variety X. Let x ∈ X be a closed point. If the inequality holds, then ωX ⊗ L separates ℓ-jets at x. εℓ F (L; x) > ℓ + 1 As we mentioned in §1, Theorem A is an immediate consequence of Theorem C: Proof of Theorem A. The inequality ε(L; x) > n + ℓ implies the inequality εℓ Proposition 2.9, hence the assertion immediately follows from Theorem C. F (L; x) > ℓ + 1 by (cid:3) Demailly proved the analogue of Theorem A in characteristic zero by using the Kawamata -- Viehweg vanishing theorem, but only assuming that L is big and nef [Dem92, Prop. 6.8(a)]. We do not know if this assumption suffices in positive characteristic. Remark 3.1. It is possible to obtain a version of Theorem A using only the zeroth Frobenius -- Seshadri constant by naıvely inducing on the order of jets in the proof of [MS14, Lem. 3.3]. How- ever, the hypothesis needed for separation of ℓ-jets using this method is the stronger lower bound ε(L; x) > n + nℓ. The proof is also more technical and relies on Castelnuovo -- Mumford regularity. There are two main ingredients in the proof of Theorem C. The first is the following reformulation of our results from §2. Proposition 3.2. Let L be an ample line bundle on a projective variety X, and let x ∈ X be a closed point. If εℓ F (L; x) > α for some real number α > 0, then we can find positive integers m and e satisfying pe − 1 m > α ℓ + 1 (17) such that Lm separates pe-Frobenius ℓ-jets at x. Furthermore, we may take m and e so that the quantity is arbitrarily large. pe − 1 − α ℓ + 1 m (18) Proof. By Proposition 2.5(i) and the definition of εℓ F (L; x), we know there exist m, e ≥ 1 such that the inequality (17) is satisfied and Lm separates pe-Frobenius ℓ-jets at x. Moreover, by applying Lemma 2.4, we may make the replacements e 7−→ re and m 7−→ m(pre − 1) pe − 1 12 TAKUMI MURAYAMA and not change the inequality (17) or the condition on separation of jets. Thus, by applying these replacements for integers r ≥ 1, the quantity in (18) satisfies pre − 1 − α ℓ + 1 · m(pre − 1) pe − 1 = (pre − 1)(cid:18)1 − α ℓ + 1 · m pe − 1(cid:19) −→ ∞ as r → ∞ by the inequality (17). We can therefore assume that the quantity (18) is arbitrarily large. (cid:3) As in the proof of [MS14, Thm. 3.1], the other main ingredient in the proof of Theorem C is the Cartier operator or the trace map T : F∗(ωX ) −→ ωX, which is a morphism of OX-modules. Here, F : X → X denotes the (absolute) Frobenius morphism. See [BK05, §1.3] for the definition and basic properties of the map T . Briefly, it can be defined as the trace map for relative duality for the finite flat morphism F as in [Har66, Ch. III, §6]. We note that F is finite since k is perfect, and F is flat by Kunz's theorem [BK05, Lem. 1.1.1] since X is smooth. The trace map satisfies the following key properties needed for our proof: (a) The trace map T and its iterates T e : F e (b) If a ⊆ OX is a coherent ideal sheaf, then T e satisfies the equality ∗ (ωX) → ωX are surjective [BK05, Thm. 1.3.4]; This follows from (a) by considering the OX -module structure on F e ∗ (ωX). ∗ (a T e(cid:0)F e [pe] · ωX)(cid:1) = a · T e(cid:0)F e ∗ (ωX )(cid:1) = a · ωX. (19) Remark 3.3. The surjectivity of the trace map in (a) is part of the definition for what are called F -injective varieties [Sch14, Def. 2.10(iv)], as long as we interpret ωX as the cohomology sheaf h− dim Xω• X. F -injective varieties are related to varieties with Du Bois singularities in characteristic zero [Sch09b]. Since the justification for (19) still works in this generality, our proof of Theorem C still works for F -injective varieties. X of the dualizing complex ω• We are now ready to prove Theorem C. The proof closely follows that of [MS14, Thm. 3.1(i)]. The idea is the following: We can increase powers on L freely so that Lm separates pe-Frobenius ℓ-jets, and then tensor by an appropriate product of the form ωX ⊗Lpe−m that is globally generated. Then, ωX ⊗ Lpe separates pe-Frobenius ℓ-jets. The eth iterate T e of the trace map T allows us to take out these factors of pe, and thereby deduce that ωX ⊗ L separates ℓ-jets. Proof of Theorem C. Let mx denote the defining ideal of x. By Proposition 3.2, we can find m and e such that m < pe − 1 and the restriction map ρℓ,e Lm : H 0(X, Lm) −→ H 0(cid:0)X, Lm ⊗ OX /(m ℓ+1 x )[pe](cid:1) is surjective; moreover, we may assume that pe − 1 − m is arbitrarily large. In particular, we may assume that pe − m is arbitrarily large, so that ωX ⊗ Lpe−m is globally generated. By Lemma 2.7, we then have that ωX ⊗ Lpe separates pe-Frobenius ℓ-jets, i.e., the restriction map ϕ : H 0(X, ωX ⊗ Lpe is surjective. ) −→ H 0(cid:0)X, ωX ⊗ Lpe )[pe](cid:1) ⊗ OX/(m ℓ+1 x (20) We now use the surjectivity of the eth iterate T e : F e ∗ (ωX ) → ωX of the trace map T . By (19), the map T e induces a surjective morphism Tensoring this by L and applying the projection formula yields a surjective morphism m ℓ+1 x · ωX ⊗ L. (21) m ℓ+1 x · ωX. F e ℓ+1 x ∗(cid:0)(m )[pe] · ωX(cid:1) )[pe] · ωX ⊗ Lpe(cid:1) ℓ+1 x F e ∗(cid:0)(m FROBENIUS -- SESHADRI CONSTANTS AND CHARACTERIZATIONS OF PROJECTIVE SPACE 13 Since the Frobenius morphism F is affine, the pushforward functor F e the exactness of the left column in the following commutative diagram: ∗ is exact, hence we obtain 0 0 )[pe] · ωX ⊗ Lpe(cid:1) mℓ+1 x · ωX ⊗ L F e ∗(cid:0)(mℓ+1 x ∗ (ωX ⊗ Lpe F e ) ωX ⊗ L F e ∗(cid:0)ωX ⊗ Lpe ⊗ OX /(mℓ+1 x )[pe](cid:1) ωX ⊗ L ⊗ OX /mℓ+1 x 0 0 The top horizontal arrow is the map in (21); the middle horizontal arrow is obtained analogously from T e by tensoring with L, and is therefore surjective. The surjectivity of the middle horizontal arrow also implies the bottom horizontal arrow is surjective. Finally, by taking global sections in the bottom square, we obtain the following commutative square: H 0(X, ωX ⊗ Lpe ) ϕ H 0(cid:0)X, ωX ⊗ Lpe ⊗ OX /(mℓ+1 x )[pe](cid:1) H 0(X, ωX ⊗ L) ρℓ,0 ωX ⊗L ψ H 0(X, ωX ⊗ L ⊗ OX/mℓ+1 x ) Note that ψ is surjective because the kernel of the corresponding morphism of sheaves is a skyscraper sheaf supported at x. We have already shown that the restriction map ϕ is surjective in (20), hence ρℓ,0 ωX ⊗L is necessarily surjective. This shows ωX ⊗ L indeed separates ℓ-jets at x. (cid:3) Remark 3.4. It is possible to define a multi-point version of εℓ F (L; x) following [MS14], which would capture how ωX ⊗ L simultaneously separates higher-order jets at different points. This method does not improve the result of [MS14, Thm. 3.1(iii),(iv)], which says the following: (a) If ε0 (b) If ε0 F (L; x) > 2 at some closed point x ∈ X, then ωX ⊗ L is very big, i.e., the rational map defined by ωX ⊗ L is birational onto its image; F (L; x) > 2 at all closed points x ∈ X, then ωX ⊗ L is very ample. 4. Characterizations of projective space We now give an application of our result on separation of jets. As far as we know, this is the first application of the methods of [MS14]. Recall that a Fano variety is a projective variety whose anti-canonical bundle ω−1 X is ample. Using Seshadri constants, Bauer and Szemberg showed the following characterization of projective space amongst smooth Fano varieties: Theorem 4.1 [BS09, Thm. 2]. Let X be a smooth Fano variety of dimension n defined over an algebraically closed field of characteristic zero. If there exists a closed point x ∈ X with then X is isomorphic to the n-dimensional projective space Pn. ε(ω−1 X ; x) ≥ n + 1, Our goal in this section is to prove the following positive characteristic version of this result. 14 TAKUMI MURAYAMA Theorem B. Let X be a smooth Fano variety of dimension n defined over an algebraically closed field of positive characteristic. If there exists a closed point x ∈ X with then X is isomorphic to the n-dimensional projective space Pn. ε(ω−1 X ; x) ≥ n + 1, 4.1. Comparison with other results and a weaker statement. Before moving on to the proof of Theorem B, we compare our result to other characterizations of projective space. As a consequence of these other characterizations, we also prove a weaker statement (Proposition 4.6) to illustrate why we might expect lower bounds on Seshadri constants to give characterizations of projective space. Let KX denote the canonical divisor on X. Theorem B can be restated as follows: Theorem 4.2. Let X be a smooth Fano variety of dimension n defined over an algebraically closed field of positive characteristic. If there exists a closed point x ∈ X with for all reduced and irreducible curves C ⊆ X passing through x, then X is isomorphic to the n-dimensional projective space Pn. (−KX · C) ≥ (multx C) · (n + 1) Proof. This follows from Theorem B by using [Laz04, Props. 5.1.5, 5.1.17], which say ε(ω−1 X ; x) = inf multx C (cid:27), x∈C⊆X(cid:26) (−KX · C) (22) where the infimum is taken over all reduced and irreducible curves C ⊆ X passing through x. (cid:3) This formulation is reminiscent of the following conjecture due to Mori and Mukai, which we mentioned in §1: Conjecture 4.3 [Kol96, Conj. V.1.7]. Let X be a smooth Fano variety of dimension n defined over an algebraically closed field. If the inequality (−KX · C) ≥ n + 1 holds for every rational curve C ⊆ X, then X is isomorphic to the n-dimensional projective space Pn. By using results of Kebekus [Keb02] on families of singular rational curves, Cho, Miyaoka, and Shepherd-Barron proved this conjecture in characteristic zero. More precisely, they showed the following stronger statement: Theorem 4.4 [CMSB02, Cor. 0.4]. Let X be a smooth projective variety of dimension n defined over an algebraically closed field of characteristic zero. If X is uniruled, and the inequality (−KX · C) ≥ n + 1 holds for every rational curve C ⊆ X passing through a general point x0, then X is isomorphic to the n-dimensional projective space Pn. In arbitrary characteristic, as far as we know the only result in this direction is the following: Theorem 4.5 [KK00, Cor. 3]. Let X be a smooth projective variety of dimension n defined over an algebraically closed field of arbitrary characteristic. Suppose KX is not nef. If (a) (−KX · C) ≥ n + 1 for every rational curve C ⊆ X; and (b) (−KX)n ≥ (n + 1)n, then X is isomorphic to the n-dimensional projective space Pn. FROBENIUS -- SESHADRI CONSTANTS AND CHARACTERIZATIONS OF PROJECTIVE SPACE 15 Given the similarity between the Mori -- Mukai conjecture 4.3 and Theorem 4.2, we asked the following question in §1: Question. Let X be a smooth Fano variety of dimension n defined over an algebraically closed field of arbitrary characteristic. If the inequality (−KX · C) ≥ n + 1 holds for every rational curve C ⊆ X, then does there exist a closed point x ∈ X with for all reduced and irreducible curves C ⊆ X passing through x? (−KX · C) ≥ (multx C) · (n + 1) As mentioned in §1, the answer to this question is "yes" in characteristic zero by using Theorem 4.4, since Theorem 4.4 implies X ≃ Pn, and therefore ε(ω−1 X ; x) ≥ n + 1 for all closed points x ∈ X by Example 2.6. If one could answer this question affirmatively independently of Theorem 4.4, then Theorem 4.1 would give an alternative proof of the Mori -- Mukai conjecture 4.3 in characteristic zero, and Theorem 4.2 would resolve their conjecture in positive characteristic. Returning to Seshadri constants, we can show the following statement as a consequence of the characterizations of projective space given above. The statement in characteristic zero gives a different proof of [BS09, Thm. 2]. Proposition 4.6. Let X be a smooth Fano variety of dimension n defined over an algebraically closed field k. Consider the inequality ε(ω−1 X ; x) ≥ n + 1 (23) for each closed point x ∈ X. Suppose one of the following is satisfied: (i) We have char k = 0 and the inequality (23) holds for a single closed point x ∈ X; or (ii) We have char k = p > 0 and the inequality (23) holds for all closed points x ∈ X. Then, X is isomorphic to the n-dimensional projective space Pn over k. Proof. For (i), we use Theorem 4.4. Since Fano varieties are uniruled [Kol96, Cor. IV.1.15], it suffices to verify the condition (−KX · C) ≥ n + 1. First, note that ε(ω−1 X ; x) > n at the given point x ∈ X, and since the locus (cid:8)x ∈ X ε(ω−1 ε(ω−1 in terms of curves in (22), we have the chain of inequalities X ; x) > n(cid:9) is open [MS14, Rem. 2.15], we have X ; x0) > n at a general point x0 ∈ X. By the alternative characterization of Seshadri constants n < ε(ω−1 X ; x0) ≤ (−KX · C) multx0 C ≤ (−KX · C) for any rational curve C containing x0. Since (−KX · C) is an integer, we have (−KX · C) ≥ n + 1. For (ii), we use Theorem 4.5. The verification of condition (a) proceeds as in (i) by applying (22) to a closed point x ∈ C contained in a given rational curve C ⊆ X. For condition (b), we use the inequality ε(ω−1 X ; x) ≤ np(−KX )n, which is [Laz04, eq. 5.2]. The inequality ε(ω−1 X ; x) ≥ n + 1 then implies condition (b). (cid:3) 4.2. Proof of Theorem B. We now turn to the proof of Theorem B. The main technical tool is the notion of bundles of principal parts, which are also known as jet bundles in the literature. See [LT95, §4] for a detailed discussion. 16 TAKUMI MURAYAMA Definition 4.7. Let X be a variety defined over an algebraically closed field k of arbitrary charac- teristic. Denote by p and q the projections X × X p q X X Let I ⊂ OX×X be the ideal defining the diagonal, and let L be a line bundle on X. For each integer ℓ ≥ 0, the ℓth bundle of principal parts associated to L is the sheaf Note that P 0(L) ≃ L, since the diagonal in X × X is isomorphic to X. P ℓ(L) := p∗(q∗L ⊗ OX×X /I ℓ+1). We will use the following facts about these sheaves from [LT95, §4], assuming X is smooth: (a) There exists a short exact sequence [LT95, no 4.2] 0 −→ Symℓ(ΩX) ⊗ L −→ P ℓ(L) −→ P ℓ−1(L) −→ 0, (24) where ΩX denotes the cotangent bundle on X. By using induction and this short exact sequence, it follows that the sheaf P ℓ(L) is a vector bundle for all integers ℓ ≥ 0. (b) There exists an identification P ℓ(L) ≃ q∗(q∗L ⊗ OX×X /I ℓ+1), and by applying adjunction to the map q∗L → q∗L ⊗ OX×X /I ℓ+1, there is a morphism of sheaves [LT95, no 4.1], such that the diagram dℓ : L −→ P ℓ(L) H 0(dℓ) H 0(X, L) ρℓ,0 L H 0(X, L ⊗ OX/mℓ+1 x ) ∼ H 0(cid:0)X, P ℓ(L)(cid:1) ρ0,0 Pℓ(L) H 0(cid:0)X, P ℓ(L) ⊗ OX/mx(cid:1) commutes for all closed points x ∈ X [LT95, Lem. 4.5(1)]. Thus, if L separates ℓ-jets at x, then P ℓ(L) is globally generated at x. We will also use the following description of the determinant of the ℓth bundle of principal parts. This description is stated in [DRS01, p. 1660]. Lemma 4.8. Let X be a smooth variety of dimension n, and let L be a line bundle. Then, for each ℓ ≥ 0, we have an isomorphism Proof. We proceed by induction on ℓ ≥ 0. If ℓ = 0, then P 0(L) ≃ L, so we are done. det(P ℓ(L)) ≃(cid:0)ωℓ X ⊗ Ln+1(cid:1) 1 n+1 (n+ℓ n ). Now suppose ℓ > 0. Since X is smooth, the cotangent bundle ΩX has rank n, and we have isomorphisms det(cid:0)Symℓ(ΩX ) ⊗ L(cid:1) ≃ det(cid:0)Symℓ(ΩX)(cid:1) ⊗ L (n+ℓ−1 n−1 ) ≃ ω (n+ℓ−1 n ) X (n+ℓ−1 n−1 ) . ⊗ L By induction and taking top exterior powers in the short exact sequence (24), we obtain det(P ℓ(L)) ≃ ω ≃ ω (n+ℓ−1 n ) X (n+ℓ−1 n ) X ⊗ L ⊗ L (n+ℓ−1 n−1 ) (n+ℓ−1 n−1 ) ≃(cid:0)ωℓ X ⊗ Ln+1(cid:1) 1 n+1 (n+ℓ n ). ⊗ det(P ℓ−1(L)) ⊗(cid:0)ωℓ−1 X ⊗ Ln+1(cid:1) 1 n+1 (n+ℓ−1 n ) FROBENIUS -- SESHADRI CONSTANTS AND CHARACTERIZATIONS OF PROJECTIVE SPACE 17 Note that the last isomorphism holds because of the identities (cid:18)n + ℓ − 1 (cid:19) + n n ℓ − 1 n + ℓ n + 1(cid:18)n + ℓ − 1 (cid:19) = (cid:18)n + ℓ − 1 n − 1 (cid:19) +(cid:18)n + ℓ − 1 n + 1(cid:18)n + ℓ − 1 (cid:19) = (cid:19) =(cid:18)n + ℓ n (cid:19) n n ℓ n + 1(cid:18)n + ℓ n (cid:19) (cid:3) involving binomial coefficients. We now return to the setting where our ground field k is an algebraically closed field of charac- teristic p > 0. We begin with the following key chain of inequalities. Note that our statement is weaker than [BS09, Prop. 1.1], but it still suffices for our purposes. Lemma 4.9. Let X be a smooth Fano variety of dimension n, and let x ∈ X be a closed point. Denote ε = ε(ω−1 X ; x). For every integer m ≥ 1, we have the chain of inequalities (m + 1)ε − (n + 1) ≤ s(ω−m X ; x) ≤ mε. (25) In particular, ε(ω−1 X ; x) ≤ n + 1. Proof. We have the inequality s(ω−m X ; x) m ≤ ε by the definition of the ordinary Seshadri constant in (3). We can then multiply by m throughout to obtain the right inequality in (25). For the left inequality in (25), we know that if ω−m X does not separate ℓ-jets, then ε(ω−(m+1) ; x) = (m + 1) · ε(ω−1 by the contrapositive of Theorem A applied to L = ω−(m+1) . Note that the equality in (26) holds by [Laz04, Ex. 5.1.4]. By the definition of s(ω−m X ; x) + 1, hence the left inequality in (25) follows. The last assertion follows by rearranging (25) for m = 1. (cid:3) X ; x), the inequality in (26) holds for ℓ = s(ω−m X ; x) ≤ n + ℓ (26) X X We now prove Theorem B. Our proof follows that of [BS09, Thm. 1.7], although we must be more careful with tensor operations in positive characteristic. Proof of Theorem B. We first claim that P n+1(ω−1 we know that at the given point x ∈ X, we have the equality ε(ω−1 X ; x) ≤ ε(ω−1 X ; x) − (n + 1) ≤ s(ω−1 n + 1 = 2 · ε(ω−1 X ) is a trivial bundle. By Lemma 4.9 for m = 1, X ; x) = n + 1, and moreover X ; x) = n + 1, hence equality holds throughout. By property (b) of bundles of principal parts, we therefore have that P n+1(ω−1 X ) is globally generated at x. On the other hand, by Lemma 4.8 applied to L = ω−1 X , we have an isomorphism det(P n+1(ω−1 X ) is a trivial bundle, consider the following diagram: X )) ≃ OX . Now to show that P n+1(ω−1 X )(cid:1) det(cid:0)P n+1(ω−1 ∼ OX det(cid:0)P n+1(ω−1 Suppose the isomorphism in the top row is given by a non-vanishing global section X ) ⊗ OX /mx(cid:1) X )(cid:1)(cid:1). s ∈ H 0(cid:0)X, det(cid:0)P n+1(ω−1 X )⊗OX /mx(cid:1), which gives the isomorphism Let s1,x∧s2,x∧· · ·∧sr,x be the image of s in det(cid:0)P n+1(ω−1 X ) is globally generated at x, each si,x can be lifted to in the bottom row. Then, since P n+1(ω−1 OX /mx ∼ 18 TAKUMI MURAYAMA vanish at x, this exterior product does not vanish anywhere, since H 0(X, OX ) = k. Thus, the X )(cid:1). Because the exterior productes1 ∧es2 ∧ · · · ∧esr does not X ), and therefore P n+1(ω−1 X ) is a trivial bundle. a global section esi ∈ H 0(cid:0)X, P n+1(ω−1 global sectionsesi give a frame for P n+1(ω−1 To show X ≃ Pn suffices to show that for every nonconstant morphism f : P1 line bundles of positive degree. Write k , we use Mori's characterization of projective space [Kol96, Thm. V.3.2]. It k → X, the pull back f ∗TX is a sum of f ∗(TX) ≃ O(ai) and f ∗(ω−1 X ) ≃ O(b), nMi=1 where b is positive since ω−1 X is ample. We want to show that each ai is positive. Note that f ∗(ΩX) ≃ f ∗(TX )∨ ≃ O(−ai). nMi=1 Dualizing the short exact sequence (24), we have the short exact sequence 0 −→ P n(ω−1 X )∨ −→ P n+1(ω−1 X )∨ −→ (Symn+1 ΩX )∨ ⊗ ωX −→ 0. The quotient on the right is globally generated because it is a quotient of the trivial bundle P n+1(ω−1 X )∨. We have isomorphisms f ∗(cid:0)(Symn+1 ΩX)∨ ⊗ ωX(cid:1) ≃(cid:0)Symn+1 f ∗(ΩX )(cid:1)∨ ⊗ f ∗(ωX) ≃(cid:18)Symn+1 O(−ai)(cid:19)∨ nMi=1 ⊗ O(−b), and this bundle is globally generated since it is the pullback of a globally generated bundle. By expanding out the symmetric power on the right-hand side, we have a surjection f ∗(cid:0)(Symn+1 ΩX)∨ ⊗ ωX(cid:1) nMi=1 O(cid:0)(n + 1)ai − b(cid:1), hence the direct sum on the right-hand side is also globally generated. Finally, this implies (n + 1)ai − b ≥ 0, and therefore since b > 0, we have that ai > 0 as required. (cid:3) Remark 4.10. Liu and Zhuang's characteristic zero statement in [LZ18, Thm. 2] is stronger than Theorem B: it only assumes that X is Q-Fano, and in particular that X is not necessarily smooth. While Theorem C holds for a large class of singular varieties (see Remark 3.3), the rest of our approach does not generalize to the non-smooth setting, since Mori's characterization of projective space uses bend and break techniques. On the other hand, Liu and Zhuang's methods do not seem to work in positive characteristic without very strong assumptions on dimension and F - singularities since, in particular, they use the Kawamata -- Shokurov basepoint-freeness theorem and the Kawamata -- Viehweg vanishing theorem. Acknowledgments I am grateful to my advisor Mircea Mustat¸a for his support and for numerous fruitful conver- sations, and to Rankeya Datta, Emanuel Reinecke, and Matthew Stevenson for helpful comments on drafts of this paper. I would also like to thank J´anos Koll´ar and Ziquan Zhuang for insights on their results on characterizations of projective space. Finally, I am indebted to the anonymous referee for useful suggestions that improved the quality of this paper. FROBENIUS -- SESHADRI CONSTANTS AND CHARACTERIZATIONS OF PROJECTIVE SPACE 19 [Bau99] T. Bauer. "Seshadri constants on algebraic surfaces." Math. Ann. 313.3 (1999), pp. 547 -- 583. doi: 10. 1007/s002080050272. mr: 1678549. 3 References [Bau+09] T. Bauer, S. Di Rocco, B. Harbourne, M. Kapustka, A. Knutsen, W. Syzdek, and T. Szemberg. "A primer on Seshadri constants." Interactions of classical and numerical algebraic geometry. Contemp. Math., Vol. 496. Providence, RI: Amer. Math. Soc, 2009, pp. 33 -- 70. doi: 10.1090/conm/496/09718. mr: 2555949. 1 M. Brion and S. Kumar. Frobenius splitting methods in geometry and representation theory. Progr. Math., Vol. 231. Boston, MA: Birkhauser Boston, 2005. doi: 10.1007/b137486. mr: 2107324. 12 T. Bauer and T. Szemberg. "Seshadri constants and the generation of jets." J. Pure Appl. Algebra 213.11 (2009), pp. 2134 -- 2140. doi: 10.1016/j.jpaa.2009.03.005. mr: 2533311. 1, 2, 3, 13, 15, 17 [BK05] [BS09] [Har66] [Ito13] [EL93] [Fuj83] [Dem92] [DRS01] [EKL95] [CMSB02] K. Cho, Y. Miyaoka, and N. I. Shepherd-Barron. "Characterizations of projective space and applications to complex symplectic manifolds." Higher dimensional birational geometry (Kyoto, 1997). Adv. Stud. Pure Math., Vol. 35. Tokyo: Math. Soc. Japan, 2002, pp. 1 -- 88. url: http://www.mathbooks.org/aspm/ aspm35/aspm35.pdf. mr: 1929792. 2, 14 J.-P. Demailly. "Singular Hermitian metrics on positive line bundles." Complex algebraic varieties (Bay- reuth, 1990). Lecture Notes in Math., Vol. 1507. Berlin: Springer-Verlag, 1992, pp. 87 -- 104. doi: 10.1007/ BFb0094512. mr: 1178721. 1, 4, 11 S. Di Rocco and A. J. Sommese. "Line bundles for which a projectivized jet bundle is a product." Proc. Amer. Math. Soc. 129.6 (2001), pp. 1659 -- 1663. doi: 10.1090/S0002-9939-00-05875-5. mr: 1814094. 16 L. Ein, O. Kuchle, and R. Lazarsfeld. "Local positivity of ample line bundles." J. Differential Geom. 42.2 (1995), pp. 193 -- 219. doi: 10.4310/jdg/1214457231. mr: 1366545. 3 L. Ein and R. Lazarsfeld. "Seshadri constants on smooth surfaces." Ast´erisque 218 (1993): Journ´ees de g´eom´etrie alg´ebrique d'Orsay (Orsay, 1992), pp. 177 -- 186. mr: 1265313. 3 T. Fujita. "Semipositive line bundles." J. Fac. Sci. Univ. Tokyo Sect. IA Math. 30.2 (1983), pp. 353 -- 378. url: http://hdl.handle.net/2261/6381. mr: 0722501. 10 R. Hartshorne. Residues and duality. Lecture notes of a seminar on the work of A. Grothendieck, given at Harvard 1963 -- 1964. With an appendix by P. Deligne. Lecture Notes in Math., Vol. 20. Berlin: Springer- Verlag, 1966. doi: 10.1007/BFb0080482. mr: 0222093. 12 A. Ito. "Okounkov bodies and Seshadri constants." Adv. Math. 241 (2013), pp. 246 -- 262. doi: 10.1016/j. aim.2013.04.005. mr: 3053712. 4 S. Kebekus. "Families of singular rational curves." J. Algebraic Geom. 11.2 (2002), pp. 245 -- 256. doi: 10. 1090/S1056-3911-01-00308-3. mr: 1874114. 14 Y. Kachi and J. Koll´ar. "Characterizations of Pn in arbitrary characteristic." Asian J. Math. 4.1 (2000): Kodaira's issue, pp. 115 -- 121. doi: 10.4310/AJM.2000.v4.n1.a8. mr: 1802915. 2, 14 J. Koll´ar. Rational curves on algebraic varieties. Ergeb. Math. Grenzgeb. (3), Vol. 32. Berlin: Springer- Verlag, 1996. doi: 10.1007/978-3-662-03276-3. mr: 1440180. 14, 15, 18 R. Lazarsfeld. Positivity in algebraic geometry. I. Classical setting: line bundles and linear series. Ergeb. Math. Grenzgeb. (3), Vol. 48. Berlin: Springer-Verlag, 2004. doi: 10.1007/978-3-642-18808-4. mr: 2095471. 1, 2, 3, 10, 14, 15, 17 D. Laksov and A. Thorup. "Weierstrass points on schemes." J. Reine Angew. Math. 460 (1995), pp. 127 -- 164. doi: 10.1515/crll.1995.460.127. mr: 1316575. 15, 16 Y. Liu and Z. Zhuang. "Characterization of projective spaces by Seshadri constants." Math. Z. 289.1 -- 2 (2018), pp. 25 -- 38. doi: 10.1007/s00209-017-1941-9. mr: 3803780. 2, 18 M. Mustat¸a and K. Schwede. "A Frobenius variant of Seshadri constants." Math. Ann. 358.3 -- 4 (2014), pp. 861 -- 878. doi: 10.1007/s00208-013-0976-4. mr: 3175143. 2, 3, 4, 5, 7, 8, 9, 10, 11, 12, 13, 15 G. P´olya and G. Szego. Problems and theorems in analysis. I. Series, integral calculus, theory of functions. Translated from the German by D. Aeppli. Reprint of the 1978 English translation. Classics Math. Berlin: Springer-Verlag, 1998. doi: 10.1007/978-3-642-61983-0. mr: 1492447. 4 [LT95] [LZ18] [Keb02] [KK00] [Kol96] [Laz04] [MS14] [PS98] [Sch09a] K. Schwede. "F -adjunction." Algebra Number Theory 3.8 (2009), pp. 907 -- 950. doi: 10.2140/ant.2009. 3.907. mr: 2587408. 4 [Sch09b] K. Schwede. "F -injective singularities are Du Bois." Amer. J. Math. 131.2 (2009), pp. 445 -- 473. doi: 10. [Sch14] 1353/ajm.0.0049. mr: 2503989. 3, 12 K. Schwede. "A canonical linear system associated to adjoint divisors in characteristic p > 0." J. Reine Angew. Math. 696 (2014), pp. 69 -- 87. doi: 10.1515/crelle-2012-0087. mr: 3276163. 12 Department of Mathematics, University of Michigan, Ann Arbor, MI 48109-1043, USA E-mail address: [email protected] URL: http://www-personal.umich.edu/~takumim/
1907.04792
1
1907
2019-07-10T15:27:50
Deformation classes of real Cayley M-octads
[ "math.AG" ]
We study 8-point configurations in the real projective space forming an intersection locus of three quadrics and containing no coplanar quadruples. We found that there exists precisely 8 mirror-pairs of deformation classes of such configurations. We describe also the mutual position of these 8 pairs and find the real monodromy groups acting on the 8-point configurations, for each deformation class.
math.AG
math
DEFORMATION CLASSES OF REAL CAYLEY M-OCTADS Sergey Finashin We study 8-point configurations in the real projective space forming an intersection locus of three quadrics and containing no coplanar quadruples. We found that there exists precisely 8 mirror-pairs of deformation classes of such configurations. We describe also the mutual position of these 8 pairs and find the real monodromy groups acting on the 8-point configurations, for each deformation class. 9 1 0 2 l u J 0 1 ] . G A h t a m [ 1 v 2 9 7 4 0 . 7 0 9 1 : v i X r a 1. Introduction 1.1. Cayley octads. A Cayley octad X ⊂ P3 is an 8-point configuration obtained as the intersection locus of three quadric surfaces X = Q0 ∩ Q1 ∩ Q2. We allow multiple points of X, which appear if the intersection is not transverse, in which case 8 is the sum of multiplicities. A simple analysis shows that for any Cayley octad X the net of quadrics N = {Qt = t0Q0 + t1Q1 + t2Q2}, t = [t0 : t1 : t2] ∈ P 2, is the complete linear system of quadrics passing through X, and so, N and X determine each other. Singular quadrics Qt ∈ N are parameterized by a quartic curve H ⊂ N ∼= P 2 called the Hessian curve (Hessian quartic): it is defined by the determinant of the symmetric 4 × 4-matrix defining Qt. The following conditions are known to be equivalent (cf., [Do, Sect. 6.3.2], or [GH, Lemma 6.4]): (1) The Hessian curve H associated to a Cayley octad X is non-singular. (2) The Cayley octad X contains neither multiple points, nor coplanar subsets of more than 3 points. (3) Each quadric Qt from the net N associated to X is irreducible, and the points of X are non-singular on Qt for all t ∈ P 2. A Cayley octad X is called regular if these equivalent conditions are satisfied, and singular otherwise. The reality condition (invariance under the complex conjugation) for Cayley octad X and for the net of quadrics N are obviously equivalent and imply reality of the Hessian curve H. We say that a real Cayley octad X is maximal or M-octad if its eight points are all real. If X contains k > 0 pairs of conjugate complex points and 8 − 2k real ones, we call it (M − k)-octad. 2010 Mathematics Subject Classification. Primary 14P25, 14M10, 14C21, 14N20. 1 2 S. FINASHIN 1.2. Principal Results. Our principal aim is to enumerate the deformation classes that is path-connected components in the space of regular M-octads and to describe the mutual position of these classes. Our first principal result says that there exist precisely 16 deformation classes of regular M-octads, which can be grouped into 8 pairs of mirror partner classes (see Theorems 3.4.1 and 4.1.2). Such pairs of classes, [X] and [ ¯X], are respresented by a regular M-octad X and its image ¯X under an orientation reversing projective transformation of RP3. The union [X] ∪ [ ¯X] is called the coarse deformation class of an M-octad X. The 8 pairs of deformation classes are named (O+ α β, O− α β), where (α, β) ∈ {0, 2, 4}× {0, 3, 4} r {(4, 3)}, and the course deformation classes are Oα β = O+ α β. The meaning of indices α and β can be explained in terms of decorated graph ΓX related to some degenerations of X. This graph has the vertex set X and its edges are line segments joining the pairs of vertices, which can be merged by a real variation of X formed by regular M-octad. More precisely, among the two line segments in RP3 joining a pair of points of X we select the one homotopic to the trace of these points under the merging variation. It follows that an edge with endpoints x, y ∈ X cannot cross any of the 20 planes passing though the triples of points of X r {x, y}. α β ∪ O− The combinatorial types of such graphs are presented on Fig. 1, Table 1. Fig. 1. Graphs ΓX and monodromy groups AutR(X) of regular M-octads X ∈ Oα β Table 1. Combinatorial types of ΓX for X ∈ Oα β Table 2. Monodromy groups D4 Z2 ⊕ Z2 D4 S3 Z2 β ⑦⑦⑦⑦ • α • β ❅❅❅❅ • α • O4 4 • α • β❅❅❅❅ • α • β⑦⑦⑦⑦ • α • β ⑦⑦⑦⑦ • O2 4 • β ❅❅❅❅ • α • • ⑦⑦⑦⑦ β • β❅❅❅❅ • • β⑦⑦⑦⑦ • β ❅❅❅❅ • • β ❅❅❅❅ • α • • α • β • ⑦⑦⑦⑦ • • β ❅❅❅❅ • • ⑦⑦⑦⑦ β • O0 4 • β❅❅❅❅ • • β⑦⑦⑦⑦ • • β ❅❅❅❅ • O0 3 • β ❅❅❅❅ • • • • • • O0 0 • • • • • O2 3 α • S4 Z2 ⊕ Z2 S4 • • O2 0 • α • • • • • α • α • O4 0 • α • α • • Table 3. Number of orbits 1 2 1 2 4 1 2 1 The edges of graph ΓX split into two types: oval-type labeled by "α" on Table 1, and bridge-type labeled by "β". These types indicate the degeneration of the Hessian curve (collapsing of an oval or a bridge) as the endpoints of an edge are merging. The numbers of edges of the corresponding types are denoted by α(X) and β(X) and called respectively the oval-index and the bridge-index of X. Our second principal result (see Theorems 3.2.1 and 3.4.1) shows that the pair α(X), β(X) is a complete invariant of coarse deformation equivalence, and any DEFORMATION CLASSES OF REAL CAYLEY M-OCTADS 3 combination (α, β) ∈ {0, 2, 4} × {0, 3, 4} except (4, 3) is realizable by some Cayley M-octad X. Our next result is description of the mutual position (adjacency) of the corresponding coarse deformation components Oα β: in Fig. 1, Table 1 adjacent components stand next to each other (in the same row or column). After proving chirality of all Cayley octads (see Theorem 4.1.2), we deduce adjacency of the pure deformation components O± α β. Finally, in Theorem 5.2.1 we described the real monodromy groups Aut(X) ⊂ S8 of M-octads X: they are indicated in Table 2 (the 8 cells of this table correspond to the cells of Table 1). Here group D4 is dihedral of order 8. These groups act on the graphs ΓX preserving the decoration of edges. Table 3 gives the number of orbits of Aut(X) acting on X, whose meaning is clarified in Sect. 6.1. 2. Preliminaries In this Section, we outline some essentials on the Cayley octads and their Hessian curves with spectral theta-charactristics (for more details see [Do] and [GH]) and recall a few well-known facts on real plane quartics. 2.1. Degeneration of Cayley octads and nets of quadrics. The variety of Cayley octads form a Zariski-open subset of the Grassmannian of 2-planes in the projective space of quadrics, N ⊂ P (Sym2(C4)), namely, the nets of quadrics N that have purely zero-dimensional basepoint locus, X = X(N ). The Hessian curve H of a Cayley octad X is interpreted as the intersection H = N ∩∆, where N is the associated net of quadrics and ∆ ⊂ P (Sym2(C4)) the discriminant hypersurface. In the variety of Cayley octads regular ones form a Zariski-open subset and singular ones form a certain discriminant locus, whose principal stratum (of codi- mension 1) is characterized in terms of nets N and their Hessian curves H = N ∩ ∆ as follows (cf. [GH], Lemma 6.4). 2.1.1. Lemma. A singular Cayley octad X represents a non-singular points of the discriminant locus of the variety of octads if and only if the H has one node and no other singular points. Such a node may appear in two ways. (1) At the point of (simple) tangency of N with ∆. Then the net N contains one and only one cone with the vertex at some point of X, and the Cayley octad X has one double point, whereas the other 6 points are ordinary and no 4 of them are coplanar. (2) At the point of (generic) intersection of N with the singular locus Sing(∆) of ∆. Then N contains a reducible quadric (a pair of distinct planes), and the 8 points of X are distinct and admit one and only one splitting into two coplanar quadruples. Proof. The claim on the non-singular points is straightforward, and the description how a node on H = N ∩ ∆ may appear is also trivial. The locus Sing(∆) param- eterizes reducible quadrics, which proves the interpretation formulated in the case (2). In the remaining case then the description (1) follows. (cid:3) In the case (1), we say that Cayley octad X experiences a 2-collision, and in the case (2) (two coplanar quadruple of points), a 4-collision. The space of real octads is the real locus of the space of complex octads, that is the corresponding Zariski-open subset in the real Grassmannian of 2-planes in 4 S. FINASHIN P (Sym2(R4)). We will be interested in the part of this space represented by M- octads. Namely, we denote by O∗, ∆2 O the strata formed by regular M- octads, by singular ones experiencing 2-collision and 4-collision respectively. To- gether they give space O = O∗ ∪ ∆2 O and ∆4 O ∪ ∆4 O. 2.1.2. Lemma. Space O is a manifold with boundary ∆2 in the interior of O and separates the connected components of O∗. O. The stratum ∆4 O lies O as well as ∆4 Proof. As it follows from Lemma 2.1.1, ∆2 O are strata of codimension 1 in the variety of real Cayley octads, so, it is enough to check what lies from the two sides of them. The double point of an M-octad X ∈ ∆2 O can be perturbed into a pair of real points, or a pair of conjugate imaginary ones by a small real variation of X, which implies that ∆2 O is bounded by O from one side. On the contrary, the number 8 of real points of X ∈ ∆4 O is preserved after any real variation, so X is adjacent to ∆4 O from both sides. (cid:3) 2.2. The spectral correspondence. For a non-singular curve C, let Θ(C) be the set of theta-characteristics on C. Consider function h : Θ(C) → Z/2, θ 7→ dim H 0(C; θ) mod 2, and let Θi(C) = h−1(i), i = 0, 1, which gives a partition Θ(C) = Θ0(C) ∪ Θ1(C) of theta-characteristics into even and odd, respectively. Given a regular Cayley octad X with the Hessian curve H = N ∩ ∆, there is a natural line bundle LH → H, whose fiber over Qt ∈ H ⊂ N is the null-spaces of the degenerate quadratic form representing Qt. This bundle represents linear system KH + θX , where KH is the canonical class of H and θX is an even theta-characteristic called spectral. This linear system embeds H to P 3 as a sextic, and the image, eH, is called the Steinerian curve. More precisely, the map H → eH associates to a singular quadric Qt (a cone) from the net N the vertex of Qt, and so, eH lies in the original space P 3 containing X (see [Do] for details). This correspondence X 7→ (H, θX ) is bijective up to projective equivalence. 2.2.1. Theorem. (Hesse-Dixon) For any non-singular plane quartic C and even theta-characteristic θ on C there exists a unique up to projective transformation of P 3 regular Cayley octad X, such that C is projectively equivalent to its Hessian curve C and θ is identified by this equivalence with the spectral theta-characteristic of X. (cid:3) We outline the proof in Sect. 2.3: it goes back to O.Hesse, who explained how X is reconstructed from (C, θ), then A.Dixon [Di] elaborated and generalized it by proving a similar correspondence for nets of quadrics in any dimension. These arguments are applicable also in the real setting (cf. [DIK], Th. 3.3.4). Recall that for a real curve C reality of θ ∈ Θ(C) means that it is preserved invariant under the involution on Θ(C) induced by the complex conjugation. 2.2.2. Theorem. For any real quartic C and an even real theta-characteristic θ, there exists a unique, up to real projective transformation of P 3 R, real net of quadrics in P 3 R such that C is its Hessian curve and θ its spectral theta-characteristic. (cid:3) Pairs (C, θ), with a non-singular quartic C ⊂ P 2 and an even theta-characteristic θ will be called even spin quartics. The spectral correspondence is not only a bi- jective map between the projective classes of regular Cayley octads and projective DEFORMATION CLASSES OF REAL CAYLEY M-OCTADS 5 classes of non-singular even spin quartics, but is an isomorphism of the correspond- ing moduli spaces, see [Do], Remark 6.3.1 and [GH] Sect. 6. Restricting it to the real loci we can conclude in particular that their connected components are in one-to-one correspondence. 2.2.3. Corollary. The spectral correspondence induces a bijective correspondence between the coarse deformation classes of real regular Cayley octads and the defor- mation classes of real even spin quartics. (cid:3) 2.3. Points of a Cayley octad as Aronhold sets for its Hessian quar- tics. The line xixj connecting a pair of points of a regular Cayley octad X = {x0, . . . , x7} determines a line Bij in the plane N that is a pencil of quadrics whose base-locus is the union of line xixj with a twisted cubic passing through the other six points of X r{xi, xj}. Line Bij is bitangent to the Hessian quartic H and the di- visor formed by the two tangency points defines an odd theta-characteristic denoted θij (see [Do], Theorem 6.3.2). This gives one-to-one correspondence between the set Θ1(H) of 28 odd theta-characteristics on H and the 28 pairs {i, j} ⊂ {0, . . . , 7}. Any quadruple {i, j, k, l} ⊂ {0, . . . , 7} defines an even theta-characteristic θijkl = θij + θik + θil − KH. The choice of a distinguished index i here is not essential, and moreover, the com- plementary quadruple {0, . . . , 7} r {i, j, k, l} gives the same theta-characteristic as θijkl. This describes 35 = 1 The remaining even theta-characteristic is 2(cid:0)8 4(cid:1) even theta-characteristics of the set Θ1(H) r {θX}. θX = θ01 + θ02 + · · · + θ07 − 3KH. The set of 7 bitangents B0i corresponding to the 7 summands θ0i in θX form an Aronhold set (see [Do], Sect.6.3.3 for details). There exist precisely 288 = 8 × 36 Aronhold sets representing 36 even theta-characteristics, so that every chosen θ ∈ Θ0(H) is represented precisely by 8 Aronhold sets, and each Aronhold set match precisely to one vertex of the Cayley octad X defined by (H, θ), θ ∈ Θ0(H). Namely, given such (H, θ), we embed H to P 3 by the linear system KH + θ0. Each binangent Bi of a fixed Aronhold set B1, . . . , B7 representing θ, gives a pair of points on eH. The seven lines passing through such pairs of points must be concurrent and intersect at the point of X corresponding to the fixed Aronhold set. This is how X is reconstructed from a given even spin quartic. 2.4. Theta-characteristics as quadratic functions. With θ we associate qθ : H1(C; Z/2) → Z/2, x 7→ h(θ) + h(θ + x∗) mod 2, where x∗ ∈ H 1(C; Z/2) is Poincare dual to x ∈ H1(C; Z/2) (see [At] and [Mu]). It is a quadratic function in the sense that (1) qθ(x + y) = qθ(x) + qθ(y) + x · y mod 2, This gives the well-known identification of the set Θ(C) with the set of quadratic functions on H1(C; Z/2), compatible with the action of H 1(C; Z/2) on the both 6 S. FINASHIN sets. The function h is identified then with the Arf invariant of quadratic functions, Arf(qθ) ∈ Z/2. Note that by definition of qθ (2) Arf(qθ+x∗ ) = h(θ + x∗) = h(θ) + qθ(x) = Arf(qθ) + qθ(x) mod 2 In the case of Cayley octad X and its Hessian quartic H and x ∈ H1(H; Z/2) we obtain (3) qθX (x) = Arf(qθX +x∗) = h(θX + x∗) mod 2 2.5. Bipartitions of Cayley octads. For a a regular Cayley octad X = {x0, . . . , x7} we consider its even bipartitions X = A ∪ B, A ∩ B = ∅, into subsets of even car- dinalities A and B, and denote by Λ(X) the set formed by 64 such unordered pairs {A, B}. Then the assignments {xi, xj } 7→ θij , {xi, xj , xk, xl} 7→ θijkl, X 7→ θX induces a natural one-to-one correspondence ΦΘ : Λ(X) → Θ(H). Note that Λ(X) is a Z/2-vector space as a subquotient of the power set P(X), with the sum operation induced by the symmetric difference △: {A, B} + {C, D} = {A △ C, A △ D} = {B △ D, B △ C} and is endowed with the non-degenerate Z/2-valued inner product {A, B} · {C, D} = A ∩ B mod 2. The difference θ − θ′ for θ, θ′ ∈ Θ(H) is a 2-torsion element of Pic0(H) and such elements are identified with the classes in H 1(H; Z/2), so, we obtain the associated one-to-one correspondence ΦH : Λ(X) → H 1(H; Z/2) which is the composition of ΦΘ and the map θ 7→ θ − θX . Composing ΦH with the Poincare duality we obtain also a map ΦH : Λ(X) → H1(H; Z/2). 2.5.1. Proposition. (1) The maps ΦH and ΦH are linear isomorphisms of Z/2-vector spaces. (2) Isomorphism ΦΘ identifies h : Θ(H) → Z/2, with the map Λ(X) → Z/2, {A, B} 7→ 1 2 A mod 2. (3) If x = ΦH ({A, B}) ∈ H1(H; Z/2), then qθX (x) = h(x) = 1 (4) ΦH preserves the inner product, namely, for all x, y ∈ H1(H; Z/2) 2 A mod 2. Φ−1 H (x) · Φ−1 H (y) = x · y, where x · y is the homology intersection index. Proof. (1) Linearity of ΦH (and thus, ΦH ) follows from the relation θij +θjk +θik = θX + KH, or equivalently, θik = θij + θjk − θX (cf. [GH, Sect. 6] or [Do, Theorem 6.3.3]). For instance, we can deduce from it ΦH ({xi, xj}, X r {xi, xj}) + ΦH ({xj , xk}, X r {xj, xk}) = (θij − θX ) + (θjk − θX ) = θik − θX = ΦH ({xi, xk}, X r {xi, xk}), and similarly, linearity in the other cases. Bijectivity of ΦΘ implies that the linear maps ΦH and ΦH are isomorphisms. Part (2) holds by definition of ΦΘ, since θij are odd and θijkl are even. Part (3) follows from relation (3) of Sect. 2.4, which taking into account also relation (1) applied to qθX implies part (4). (cid:3) DEFORMATION CLASSES OF REAL CAYLEY M-OCTADS 7 2.6. Picard-Lefschetz monodromy transformation in Θ(H). 2.6.1. Lemma. The Picard-Lefschetz monodromy transformation associated with the vanishing class v ∈ H1(H; Z/2) induces map T Θ v : Θ(H) → Θ(H) (3) θ 7→ θ + (qθ(v) + 1)v∗. In particular, θ is preserve invariant if and only if qθ(v) = 1. Proof. As is well-known, the monodromy action in H1(H; Z/2) associated with v is the Picard-Lefschetz transformation x 7→ x + (x · v)v, so, the corresponding monodromy transforms in Θ(H), qθ 7→ q′ θ(x) = qθ(x + (x · v)v) = qθ(x) + (x · v)qθ(v) + (x · v)2 = qθ(x) + (qθ(v) + 1)v∗(x). (cid:3) q′ 2.6.2. Proposition. The Picard-Lefschetz transformation T Θ v preserves the pair- ity of theta-characteristics θ (Arf-invariant of qθ). The action induced by T Θ in v Λ(X) via ΦΘ depends on the bipartition Φ−1 H (v) = {Av, Bv} ∈ Λ(X) as follows. θ should satisfy modulo 2 relation (1) If one of the sets Av or Bv is a 2-elements set {xi, xj} ⊂ X then the action of T Θ v in Λ(X) is induced by the transposition of xi and xj. (2) If Av and Bv are 4-element sets, then T Θ v replaces 2-element subsets {xi, xj} ⊂ Av (or {xi, xj} ⊂ Bv ) by their complements Ax r {xi, xj} (respectively Bv r {xi, xj } ) and keep {xi, xj} unchanged if it has one common point with Av and Bv. Proof. Applying formula (3) and using relation (1), (2) we obtain Arf(q′ θ) = Arf(qθ) + qθ((qθ(v) + 1)v) = Arf(qθ) mod 2, since qθ((qθ(v) + 1)v) = (qθ(v) + 1)qθ(v), which means preserving of pairity of θ. The action on {A, B} ∈ Λ(X) induced by T Θ v due to Lemma 2.6.1, in terms of {Av, Bv} is as follows: {A, B} 7→ {A, B} + ε{Av, Bv}, where ε = Av 2 + A ∩ Av + 1 mod 2, which gives ε = A ∩ Av mod 2 if Av = {xi, xj }, and so, in the case (1) T Θ is v induced by transposition of xi and xj. In the case (2) ε = A ∩ Av + 1 mod 2 and thus, T Θ v acts also as is described. (cid:3) Transformations part (2) of Proposition 2.6.2 were named bifid substitutions by Cayley, who found them as the monodromy of 4-collisions in the corollary below. 2.6.3. Corollary. Let H be the Hessian quartic of a regular Cayley octad X with the spectral theta-characteristic θ ∈ Θ(H) and associated quadratic function qθ. Assume that v ∈ H1(H; Z/2) is a vanishing class associated with some nodal degen- eration of H, v = ΦH ({Av, Bv}). Then qθ(v) = A 2 mod 2 and the degeneration of X corresponding to the nodal degeneration of H is (1) a 2-collision of xi, xj ∈ X that form 2-element set, Av or Bv, in the case of qθ(v) = 1; (2) a 4-collision involving Av and Bv, in the case of qθ(v) = 0. Proof. It follows from the description of the Picard-Lefschetz transformation cor- responding to the vanishing class v in terms of X given in Proposition 2.6.2. (cid:3) 8 S. FINASHIN 2.7. Real non-singular quartics. The real deformation classification of non- singular real plane quartics C goes back to F.Klein: there exist 6 real deformation classes characterized by the number 0 6 k 6 4 of components of CR called ovals and in the case k = 2 by their mutual position: in one class two ovals bound disjoint pair of discs in P 2 R, and in the other two ovals are nested: one oval bounds a disc containing another oval. By M-quartics we mean real non-singular quartics with the maximal number 4 of ovals. The ovals of such quartics can be arbitrarily permuted by the monodromy. 2.7.1. Proposition. (Klein) M-quartics form one real deformation class. The ovals of an M-quartic can be arbitrarily permuted by the real deformation mon- odromy: the corresponding monodromy group is the symmetric group S4. (cid:3) 2.7.2. Proposition. A real regular Cayley octad X has 8 − 2d real points, 0 6 d 6 3, if and only if its Hessian quartic H has 4 − d real ovals, which (for d = 2) are not nested. Proof. The real structure from H can be lifted to the del Pezzo surface Z obtained by double covering of P 2 branched along H, so that the real locus ZR is projected to the non-orientable region of P 2 R) − k) = 2 − 2k, where k is the number of components of HR. On the other hand, if X has 8 − 2d > 0 real points, then in the blowup model of Z, its real locus ZR is obtained by blowing up P 2 R at 8 − 2d − 1 points, thus, χ(ZR) = 1 − (8 − 2d − 1) = 2 − 2(4 − d). Thus, k = 4 − d. (cid:3) R bounded by HR. Then χ(ZH) = 2(χ(P 2 3. Real Hessian curve with real θ-characteristics 3.1. The ovals and bridges of M-quartics. Given an M-quartic C ⊂ P 2 we consider real vanishing cycles, which are conj-invariant simple closed curves on C that can be contracted to the nodal point by some real degeneration of C. Such cycles include the the four ovals, a0, . . . , a3 ⊂ CR: contraction of an oval yields a node of solitary type. Another kind of vanishing cycles are bridges bij, 0 6 i < j 6 3, connecting ovals ai and aj. Namely, the quartic splitting into four real lines in general position in P 2 can be perturbed into M-quartic, and the six intersection points of the lines yield the six bridges bij, which are real vanishing cycles having precisely two real points at the intersection with ovals ai and aj. Connectedness of the space of M-quartics yields such bridges for any given M-quartic C and Proposition 3.1.1 below shows that the classes [bij] ∈ H1(C; Z/2) are well-defined (unique). Consider ±1-eigengroups H ± 1 (C) = {x ∈ H1(C) c(x) = ±x} along with their images H ± 1 (C; Z/2) ⊂ H1(C; Z/2) under the modulo 2 reduction homomorphism. It is a well-known fact that the action of c in H1(C; Z/2) is trivial for M-curves, which leads to the direct sum decompositions H1(C) = H + 1 (C) ⊕ H − 1 (C), H1(C; Z/2) = H + 1 (C; Z/2) ⊕ H − 1 (C; Z/2). 3.1.1. Proposition. (1) Classes [ai], 0 6 i 6 3 span Z/2 vector space H + 1 (C; Z/2) with the only relation [a0] + · · · + [a3] = 0. In particular, any three of these classes form a basis. (2) The homology classes [bij] ∈ H1(C; Z/2), 0 6 i < j 6 3 span H − 1 (C; Z/2) and satisfy relations [bij ]+[bjk]+[bik] = 0 for any triple of indices 0 6 i < j < k 6 3 In particular, classes [b0i], 1 6 i 6 3 form a basis of H − 1 (C; Z/2). DEFORMATION CLASSES OF REAL CAYLEY M-OCTADS 9 (3) Any real nodal degeneration which leads to merging of ovals ai and aj at a cross-like node is described by the vanishing class [bij] ∈ H1(C; Z/2). In particular, bridge classes are independent of the choice of a degeneration of an M-quartic C into four real lines. Proof. If C is an M-quartic, then the complement C r CR splits into two connected components, C1, C2, topological spheres with 4 holes permuted by the complex conjugation, which yield the homology relation in (1) and trivially implies that classes [ai] span H + 1 (C; Z/2). Since bridge bij is the union of a path connecting ai with aj in C1 with the conj-symmetric path in C2, we can also trivially deduce (2). Finally, [bij] is determined by the modulo 2 intersection indices bij ·ai = bij ·aj = 1, bij · ak = 0, for k 6= i, j, which gives (3). (cid:3) 3.2. The real monodromy action on theta-characteristics. Consider the variety C of spin M-quartics, (C, θ), θ ∈ Θ(C). It splits into two components, C = C ev ∪ C odd, where for C ev and C odd θ is respectively even and odd. Each of these two components is furthermore partitioned as C ev = [α,β C ev α,β, C odd = [α,β C odd α,β , where subscript α in C ev qθ([ai]) = 1, while β counts the number of bridge classes [bij] with qθ([bij]) = 1. α,β stands for the numbers of ovals ai for which α,β and Codd In this Subsection we will prove the following result. 3.2.1. Theorem. Space C has precisely 11 connected components: 8 components in Cev and 3 components in Codd. The components of Cev are Cev α,β where a ∈ {0, 2, 4}, b ∈ {0, 3, 4}, with exception of the pair (α, β) = (2, 4), and the components of Codd are Codd α,β where (α, β) is (2, 4), (2, 3), or (4, 3). We start proving it with two lemmas. 3.2.2. Lemma. The 64 theta-characterisitics θ ∈ Θ(C) on an M-quartic C are classified by the three Z/2-values qθ([ai]) and three Z/2-values qθ([b0i]), i = 1, 2, 3. The Arf-invariant of qθ is equal to Arf (qθ) = 3X1=3 qθ([ai])qθ([b0i]). Proof. It follows trivially from Proposition 3.1.1 and from the definition of the Arf-invariant. (cid:3) So, we can encode the theta-characteristics θ ∈ Θ(C) by binary 2 × 3-matrices (cid:20) qθ([a1]) qθ([b01]) qθ([a2]) qθ([b02]) qθ([a3]) qθ([b03])(cid:21). Symmetric group S4 permuting the ovals (or in the other words, changing ordering a0, . . . , a3) acts on the set Θ(C). Namely, σ ∈ S4 induces an action on the bridges classes sending [bij ] to [bσ(i)σ(j)], which determines the induced action on the set of 64 binary 2 × 3-matrices. We will list the orbits of this actions. 10 S. FINASHIN 3.2.3. Lemma. There exist precisely 11 orbits of the S4-action on Θ(C): eight in Θ0(C) and three in Θ1(C). In terms of binary matrices they are represented by 1 0 (cid:20) 0 (cid:20) 0 (cid:20) 0 (cid:20) 0 0 1 0 1 0 0 1(cid:21), (cid:20) 0 1(cid:21), (cid:20) 0 0(cid:21), (cid:20) 0 1(cid:21), (cid:20) 0 0 0 0 1 0 1 0 1 0 1 0 0 0 0 0 0 0 1 0 (1) eight matrices for θ ∈ Θ0(C) 1 0 1 0 1(cid:21), (cid:20) 1 1(cid:21), 0(cid:21), (cid:20) 1 1(cid:21), (cid:20) 1 0 0 0 1 0 1 0 1 0 1(cid:21), 1 0(cid:21), 1(cid:21). (2) three matrices for θ ∈ Θ1(C) 1 1 1 Proof. Enumeration of orbits is a straightforward exercise. The pairity of θ ∈ Θ(C) represented by the matrices listed are determined by Lemma 3.2.2. (cid:3) Proof of Theorem 3.2.1. Connectedness of the space of M-quartics implies imme- diately that the set of connected components of the space C is in one-to-one corre- spondence with the set of orbits of the real deformation monodromy action on the set Θ(C) for any chosen M-quartic C. Since the real monodromy group acting on the ovals of C is S4 (see 2.7.1), Lemma 3.2.3 implies the first part of the Theorem. For the second part we need just to determine the values of α and β (numbers of non-zero values of qθ on the 4 ovals and 6 bridges) for the 11 matrices in Lemma 3.2.3. It is straightforward because both subgroups H ±(C; Z/2) are isotropic with respect to the intersection form, and thus, the restriction of qθ to each of them is linear. Namely, in the list (1) values of α are respectively 0, 2 and 4 in the first, the second and the third columns, while the values of β are respectively 4, 3 and 0 in the first, second and third rows. In part (2) the three matrices represent pairs (α, β) which are, respectively, (2, 4), (2, 3), and (4, 3). (cid:3) 3.3. Theta-diagrams of M-quartics. We give below a simple graphical inter- pretation of the 2 × 3-matrices listed in Lemma 3.2.2, which reveals the geometry behind the enumeration of orbits in Lemma 3.2.3 and simplifies it. The four ovals ai of an M-quartic C are represented by four vertices depicted as small circles on the plane and the six bridges are represented as edges depicted as line segments connecting these circles pairwise: bij connects circle ai with aj in RP2, so that two of the six bridges pass "through infinity", as it is shown on Fig. 2. One can view this diagram as a complete graph K4 embedded in RP2. A theta-characteristic θ ∈ Θ(C) can be described as a coloring of such diagram: black color for an oval ai or bridge bij means that qθ takes value 0 on the cor- responding class [ai] or [bij], and white means values 1. Black colored ovals are depicted as filled circles, and black bridges are usual edges, while white ovals and bridges look like empty circles and dotted edges. A diagram decorated in this way will be called the theta-diagram of θ ∈ Θ(C). Such diagram with even θ will be denoted Dαβ, where α and β are respectively the numbers of white vertices (ovals) and white edges (bridges) in it. 3.4. Coarse Deformation Components. DEFORMATION CLASSES OF REAL CAYLEY M-OCTADS 11 Fig. 2. Ovals and bridges of M-quartics Even θ Odd θ β = 4 β = 3 β = 0 ❄❄❄ • • ⑧⑧⑧⑧ • • ⑧⑧⑧⑧ ❄❄❄ • • ⑧⑧⑧⑧ • • • • • • ⑧⑧⑧ ❄❄❄❄ ⑧⑧⑧ ⑧⑧⑧ ❄❄❄❄ ❄❄❄ • ◦ ⑧⑧⑧⑧ • ◦ ⑧⑧⑧⑧ ❄❄❄ • ◦ ⑧⑧⑧⑧ ◦ • ◦ • ◦ • ⑧⑧⑧ ❄❄❄❄ ⑧⑧⑧ ⑧⑧⑧ ❄❄❄❄ ❄❄❄ ◦ ◦ ⑧⑧⑧⑧ ◦ ◦ ⑧⑧⑧ ❄❄❄❄ ❄❄❄ • ◦ ⑧⑧⑧⑧ • ◦ ⑧⑧⑧⑧ ⑧⑧⑧ ❄❄❄❄ ⑧⑧⑧ • ◦ • ◦ ⑧⑧⑧ ◦ ◦ ◦ ◦ ⑧⑧⑧⑧ ❄❄❄ ◦ ◦ ⑧⑧⑧⑧ ◦ ◦ ⑧⑧⑧ ❄❄❄❄ • black oval: ◦ white oval: qθ([ai]) = 0 qθ([ai]) = 1 black bridge: qθ([bij]) = 0 white bridge: qθ([bij]) = 1 α = 0 α = 2 α = 4 3.4.1. Theorem. There exist eight coarse deformation classes of regular Cayley M-octads which are in spectral correspondence with the eight components Cev αβ of C ev listed in Theorem 3.2.1. Proof. Theorem 2.2.2 with Corollary 2.2.3 and Proposition 2.7.2 (in the case of d = 0) give a bijective correspondence between the set of coarse deformation classes of real regular M-octads and the set of deformation classes of non-singular real even spin M-quartics. So, it is left to apply Theorem 3.2.1 that enumerates them. (cid:3) Let us denote by Oαβ the coarse deformation class corresponding to C ev αβ by Theorem 3.4.1. 3.5. Adjacency of the coarse deformation components. Two real deforma- tion classes of M-octads (connected components of O∗) lying on the opposite sides from a wall of ∆4 O (formed by M-octad with a 4-collision) are said to be adja- cent. A pair of coarse deformation classes Oαβ containing such an adjacent pair of components will be called adjacent too. Adjacency of classes Oαβ are analyzed below through adjacency of the corre- sponding components Cev αβ. Namely, let us fix an M-quartic C with the ovals ai, i = 0, . . . , 3 and bridges bij, 0 6 i < j 6 3. We say that theta-characteristics θ and θ′ in Θ0(C) are adjacent theta-characteristics if the pairs (C, θ) and (C, θ′) represent adjacent components C ev α′β ′ (corresponding to adjacent classes Oαβ and Oα′β ′). αβ and C ev 3.5.1. Proposition. (1) Group H1(C; Z/2) contains precisely 10 real vanishing classes: four oval- classes [ai] and six bridge-classes [bij ]. (2) A pair of theta-characteristics θ, θ′ ∈ Θ0(C) are adjacent if and only if θ′ = θ + v∗, where v∗ is Poincare dual to some real vanishing class v ∈ H1(C; Z/2) satisfying the condition qθ(v) = 0. Proof. Any nodal degeneration of C is either a contraction of some oval ai with the vanishing class [ai], or merging of two ovals, ai and aj with the vanishing class [bij] 12 S. FINASHIN (see Proposition 3.1.1(3)). This proves part (1). By Lemma 2.1.2 and Corollary 2.6.3, the condition qθ(v) = 0 in part (2) means that the corresponding to this nodal degeneration wall is internal (deformation components of spin M-quartics lie on the both sides of it). For proving the wall-crossing formula θ′ = θ + v∗, recall that the forgetful map (C, θ) 7→ C from the variety of spin quartics to the space of quartics is unramified near nodal quartic if for the vanishing class v we have qθ(v) = 1 and has ramification of index 2 along the stratum of nodal quartics with qθ(v) = 0. It follows for instance from Lemma 2.6.1. The Picard-Lefschetz transformation in Θ(C), as it was observed in Lemma 2.6.1, is qθ 7→ qθ + (qθ(v) + 1)v∗, which in the case of qθ(v) = 0 is just just adding v∗, so, it implies qθ ′(v) = qθ(v) + v∗(v) = qθ(v). Finally, it is left to notice that a loop around a wall in the space of spin quar- tics after lifting to the covering space with the ramification of index 2 becomes a path into the adjacent deformation component and the Picard-Lefschetz formula becomes the wall-crossing formula for the markings that define a given branched covering. (cid:3) In terms of M-quartic theta-diagrams Dαβ and Dα′β ′ of adjacent deformation classes this proposition means that Dα′β ′ is obtained from Dαβ after one of the following two modifications (see Fig.3): Fig. 3. Black edge and vertex modifications ◦ • Black edge modification • ◦ • Black vertex modification ❄❄❄❄❄ • (1) Black edge modification: each endpoint, ai and aj, of some black edge βij changes its color, while the other 2 vertices and all 6 edges preserve. (2) Black vertex modification: each edge bij incident to some black vertex ai changes its color, while the other 3 edges and all 4 vertices preserve. The following theorem shows that adjacency of coarse deformation classes Oα β corresponds to vertical and horizontal adjacency of theta-diagrams Dαβ on Fig. 2. 3.5.2. Theorem. Two different coarse deformation classes of M-octads Oα β and Oα′ β ′ are adjacent if and only if one of the following holds (1) α = α′ and numbers in the pair {β, β′} ⊂ {0, 3, 4} have odd difference, (2) β = β′ and α − α′ = 2. Besides, a coarse deformation component Oα β is adjacent to itself if and only if (α, β) is either (2, 3) or (2, 0). Proof. If the vanishing cycle v representing the wall between adjacent deformation classes is a bridge-class bij, then according to Lemma 3.5.1 the value of qθ changes on the two oval-classes ai and aj and is preserved on the other oval classes and on all bridge-classes. Then β = β′ and α − α′ is either 2 or 0. Moreover, 0 may appear only if some black edge has endpoints of different colour, which is possible only for O2 3 and O2 0. If v is an oval-class ai, then for the same reason qθ changes on the three bridge- classes [bij], 0 6 j 6 3, j 6= i and so, α = α′ and β − β′ is either 1 or 3. (cid:3) DEFORMATION CLASSES OF REAL CAYLEY M-OCTADS 13 3.5.3. Corollary. Coarse deformation classes of regular M-octads are related through a finite number of wall-crossings (passing to an adjacent coarse deformation class of M-octads). (cid:3) 3.6. Graphs ΓX . Assume that X is a regular M-octad and (H, θ) is the associ- ated even spin Hessian quartic. Consider a pair of real vanishing cycles v1, v2 ∈ H1(H; Z/2) representing 2-collisions of X, and let e1, e2 be the corresponding edges of ΓX . 3.6.1. Lemma. (1) Edges e1 and e2 are different for different v1 and v2. (2) The intersection index v1 (cid:5) v2 is 1 if and only if e1 and e2 have a common vertex; in particular, edges e1 and e2 are disjoint if v1, v2 are both oval-cycles, or both bridge-cycles. Proof of Lemma 3.6.1. It follows immediately from that ΦH preserves Z/2-inner product by Proposition 2.5.1. (cid:3) 3.6.2. Proposition. The graphs ΓX for X ∈ Oα β have combinatorial types as presented in Table 1 of Fig. 1. Proof. By Proposition 2.5.1, white ovals and bridges of H represent edges of graph ΓX , so that an incident pair of oval end bridge represents a pair of adjacent edges. Note that white ovals and edges shown on each of the theta-diagrams of Fig. 2 split into several chains formed by consecutively incident ovals and bridges which alternate in the chain. The corresponding edges in ΓX then split into chains of consecutively adjacent edges. Applying it to each of the eight theta-diagrams, we obtain the graphs ΓX presented in Table 1 of Fig. 1. For instance, in the (least trivial) case of X ∈ O4 4, there is one chain forming a cycle, a0, b01, a1, b12, . . . , a7, b07 (if the ovals are numerated cyclically). This cycle should represent an cycle on ΓX forming an octagon shown in the top-right cell in Table 1 of Fig. 1. Analysis of the other cases is similar. (cid:3) 4. Pure deformation classification 4.1. Chirality of configurations in RP3. By a simple n-configuration in RP3, n > 4, we mean an n-point subset A ⊂ RP3 not containing coplanar quadruples of points. A pair of such configurations are said to be deformation equivalent if they can be connected by a continuous family of simple n-configurations. A mirror partner of A is a configuration ¯A obtained from A by an orientation reversing projective transformation (for instance, reflection across a plane). If a simple n- configuration A is deformation equivalent to ¯A, it is called achiral and otherwise chiral. The following observation belongs to O.Viro and V.Kharlamov (see [VV]). 4.1.1. Proposition. All simple configurations of 6 and 7 points in RP3 are chiral. Proof. Following [VV], with a triple of skew lines {ℓ1, ℓ2, ℓ3} in RP3 we associate sign lk(ℓ1, ℓ2, ℓ3) = lk(¯ℓ1, ¯ℓ2)lk(¯ℓ1, ¯ℓ3)lk(¯ℓ2, ¯ℓ3) ∈ {+1, −1} where ¯ℓi is the line ℓi endowed with an arbitrary orientation and lk(¯ℓi, ¯ℓj) = ±1 is the normalized linking number in RP3 (normalization means multiplication by 2, since the usual linking number is ± 1 2 ). This triple index is clearly independent of the choice of the order of lines ℓ1, . . . ,ℓ3 and of their orientations, but alternates if we change the orientation of RP3, or reflect a triple of lines across a plane. 14 S. FINASHIN Given a simple 6-configuration A, we consider a triple of skew lines connecting the points of A pairwise. We can obtain 15 such triples of lines corresponding to 15 splitting of 6 points into 3 pairs, and the product of the corresponding 15 signs lk(ℓ1, ℓ2, ℓ3) is a deformation invariant of simple 6-configuration A, denoted sign6(A). Then sign6(A′) = −sign6(A) if A′ is a mirror of A, and so, A is chiral. For a simple 7-configuration B we obtain seven signs for its 6-subconfigurations, and the product of these signs, sign7(B) ∈ {+1, −1}, is a deformation invariant, such that sign7(B′) = −sign7(B) for a mirror image B′ of B. (cid:3) Each point x of a simple 8-configuration X can be also equipped with a sign, sign(X, x) = sign7(X r {x}). Then, if X ′ is the mirror partner of X and x′ ∈ X ′ the vertex corresponding to x, we have sign(X ′, x′) = −sign(X, x). 4.1.2. Theorem. All points of a regular Cayley M-octads are equipped with the same sign. In particular, any regular Cayley M-octad is chiral. Thus, there exist precisely 16 pure deformation classes of them. We complete its prove towards the end of section 4. 4.2. Plane 4-moves for 6 and 7-configurations. We say that two deformation classes of simple n-configurations, n > 4, are adjacent if they lie from the opposite sides of the codimension 1 stratum formed by n-configurations with a coplanar quadruple of points (a deformation class can be adjacent to itself if this stratum is one-sided or the same deformation class lie from the both sides). 4.2.1. Lemma. Assume that simple n-configurations A0 and A1 represent adja- cent deformation classes. Then signn(A0) = −signn(A1) if n = 6 or n = 7. In particular, for such n self-adjacent deformation classes do not exist. Proof. Since A0 and A1 are adjacent, they can be connected with a path At = {x1(t), . . . , xn(t)}, t ∈ [0, 1], in which configurations At are simple for all t except one value t = t0, and for this value precisely one quadruple of points xi(t) become coplanar. These four points can be connected pairwise by two lines in three ways. The linking number between the corresponding pairs of lines (with some auxiliary orientations) for t 6= t0 alternates as we cross the wall at t = t0. Thus, in the case of n = 6, for 3 of the 15 pairwise matchings the triple linking number alternates. For the remaining 12 matchings such number is obviously preserved, so, sign6(A0) = −sign6(A1). In the case of n = 7, among the seven signs of 6-subconfigurations exactly three will change: if the point that we drop is not among the four points which become coplanar in the process of deformation At. Thus, sign7(A0) = −sign7(A1). (cid:3) 4.2.2. Lemma. For any Cayley M-octad X and any point x ∈ X the sign sign(X, x) alternates after X experiences a 4-collision. Proof. It follows from Lemma 4.2.1 applied to the residual 7-configuration X r{x}, because it contains precisely one of the two complementary quadruple of points of X involved into a 4-collision. (cid:3) 4.2.3. Lemma. If in a Cayley M-octad X points xi, xj ∈ X are adjacent in the graph ΓX , then sign(X, xi) = sign(X, xj). Proof. This follows from that residual 7-configurations Xi = X r {xi}, i = 1, 2 are deformation equivalent through simple 7-configurations, in which x1 moves DEFORMATION CLASSES OF REAL CAYLEY M-OCTADS 15 towards x2 along the corresponding edge of ΓX and the other 6 points of X remain constant. (cid:3) Proof of Theorem 4.1.2. The signs sign(X, x) are the same for all x ∈ X if X ∈ O4 4 due to Lemma 4.2.3, because the graph ΓX is connected. Since by Corollary 3.5.3 we may reach any coarse deformation classes Oα β through wall-crossing beginning from O4 4, Lemma 4.2.2 implies that for any regular M-octad X the signs sign(X, x) are the same for all eight points of X. (cid:3) By Theorem 4.1.2 each coarse deformation class Oα β is the union of two pure deformation classes distinguished by the common sign of vertices. According to this sign, we will denote the corresponding pair of deformation classes O+ α β. α β and O− 4.2.4. Corollary. Any pair of deformation classes O± crossings. 2 3 and O− Proof. It follows from Corollary 3.5.3 and adjacency between classes O+ 2 3 realized by a black-edge-move in the theta-diagram D23, for a black edge whose endpoints are of different colors. (cid:3) α β are related by several wall- 5. Real monodromy groups 5.1. The Complex monodromy. Consider a regular Cayley Octad X with its Hessian quartic H and spectral theta-characteristic θ ∈ Θ0(H). Isomorphism ΦH : Λ(X) ∼= H1(H; Z/2) due to Proposition 2.5.1 and relation (3) in Sect. 2.4 induces a homomorphism Φaut : S(X) → Aut(H1(H; Z/2), qθ) from the permuta- tion group S(X) ∼= S8 acting on X to the group of automorphisms of H1(H; Z/2) preserving the quadratic function qθ associated to θ (and thus, preserving Z/2- valued intersection form in H1(H; Z/2)). 5.1.1. Proposition. (cf. [GH], Proposition 2.1) Φaut is a group isomorphism. Proof. Proposition 2.5.1 immediately implies that Φaut is monomorphic, so, it is left to to verify that Aut(H1(H; Z/2), qθ) = 8!. First, we find the order of the automor- phism group Aut(H1(H; Z/2)) preserving just the intersection form, by counting the number of symplectic bases e1, f1, e2, f2, e3, f3 in H1(H; Z/2). This number is (63 · 32)(15 · 8)(3 · 2) = 36(8!), where, 63 · 32 counts the number of pairs e1, f1, etc. Group Aut(H1(H; Z/2)) acts transitively on the set Θ0(H) (identified with the set of quadratic function with even Arf invariant). Since Aut(H1(H; Z/2), qθ) is the stabilizer of θ, its order is 8!, since Θ0(H) = 36. (cid:3) 5.2. The monodromy groups of regular M-octads. By the real monodromy group, AutR(X), of a real regular Cayley M-octad X we mean the subgroup of S(X) that is the image under the monodromy homomorphism π1(O± α β, X) → S(X), where O± α β is the deformation component of M-octads containing X. 5.2.1. Theorem. For any regular Cayley M-octad X the group AutR(X) is iso- morphic to a subgroup of S4 formed by the permutations of the ovals ai, 0 6 i 6 3, of the associated Hessian quartic H which preserve their colors as well as the colors of bridges bij, 0 6 i < j 6 3 under the induced permutation of them. Proof. By Proposition 3.1.1 a permutation of ovals of M-quartic H determines the corresponding permutation of bridges and thus, an automorphism of H1(H; Z/2) which is necessarily real and non-trivial for a non-trivial permutation of ovals. By 16 S. FINASHIN Proposition 5.1.1, an automorphism of H1(H; Z/2) determines a permutation of X provided it preserves the quadratic function qθ associated to the spectral theta- characteristic θ, which is equivalent to preserving the coloring of ovals and edges on the theta-diagram of (H, θ). The permutation of X that we obtain is represented by a real monodromy, as it follows from Theorem 2.2.2. (cid:3) 5.2.2. Corollary. The list of groups AutR(X) for X ∈ Oα β, a ∈ {0, 2, 4} and b ∈ {0, 3, 4}, (α, β) 6= (2, 3) is as presented in the Table 2 of Fig. 1. Proof. Theorem 5.2.1 reduces finding of AutR(X) to an trivial analysis of symme- tries of the eight theta-diagrams on Fig. 2. (cid:3) 6. Concluding remarks 6.1. Deformation classes of marked Cayley M-octads. As an application of analysis of the real monodromy action in the last Section, we can now easily upgrade our deformation classification of Cayley M-octad to more refined classification of M-octads with a marked point. Namely, each coarse deformation class Oα β gives a number of coarse deformation classes or marked M-octad which is obviously equal to the number of orbits of the real monodromy action on X ∈ Oα β. The numbers of orbits are indicated in the Table 3 on Figure 1 and there sum, 14, is the number of classes of marked M-octads. Furthermore, dropping the chosen point in a marked regular M-octad gives a configuration of 7 points in P 3 R. On the other hand, the central projection from this marked point send the others into a configuration of 7 points in P 2 R, none 3 of which is collinear and no 6 is coconic. Following [GH] we call such 7-configurations typical. These two 7-configurations, planar and spacial, are related by the Gale duality (see [GH], Sect. 7), and knowing one of them (up to projective equivalence), one can recover the whole Cayley octad (cf., Sect. 2.3). In particular the 14 coarse deformation classes of marked regular Cayley M- octads correspond to the 14 deformation classes of planar typical 7-configurations that were analyzed in [FZ]. 6.2. Theta-characteristics for real (M-1)-quartics. In the case of regular (M- 1)-octads the corresponding Hessian curves H are (M-1)-quartics, by Proposition 2.7.2. The set of ovals a1, a2, a3 of H is connected pairwise by bridges: each pair ai, aj by two bridges denoted b± ij: in accord with two kinds of line segments connecting a pair of points in RP2, see Fig. 4 for the corresponding theta-diagrams. It is not difficult to show that [b+ ij] = [HR], and so, qθ takes opposite values on [b+ ij] and [b− ij]. A further analysis shows that there exist two deformation classes of theta- diagrams for even spin (M-1)-quartics, which we presented on Fig. 4 together with the corresponding graphs ΓX (which have six vertices for (M-1)-octads) and real monodromy groups AutR(X) (the definitions and proofs are analogous to the case of M-octads). ij] + [b− In particular, we see that there exist precisely two coarse deformation classes of regular (M-1)-octads described by theta-diagrams and graphs ΓX on Fig. 4. By Proposition 4.1.1, pure deformation classification of regular (M-1)-octads gives then 4 deformation classes. 6.3. Odd spin quartics. An odd theta-characteristic, θ, on a non-singular quar- tic C defines a cubic surface Z with a point z ∈ Z, so that C is projectively DEFORMATION CLASSES OF REAL CAYLEY M-OCTADS 17 Fig. 4. Theta-diagrams and graphs ΓX for (M-1)-octads X ⑧⑧⑧ ◦ ◦ • ⑧⑧⑧⑧ ⑧⑧⑧ • ◦ ◦ ⑧⑧⑧⑧ • α • β ⑦⑦⑦⑦ • β❅❅❅❅ • • β • Groups AutR(X): Z/2 β❅❅❅❅ • α⑦⑦⑦⑦ • α • β ⑦⑦⑦⑦ • α ❅❅❅❅ • • β S3 equivalent to the critical locus of the central projection of Z to P 2 from the center z, and θ is represented by the tangent plane to Z at z which is projected to a bitangent to C. The three deformation classes of odd spin M-quartics (C, θ) found in Proposition 3.2.3 corresponds to the three types of real points on a real nonsingular M-cubic surface Z ⊂ P 3. Recall that Z is M-cubic if and only if all 27 lines on it are real, and the complement of these lines in ZR split into polygonal regions, which can be triangular, quadrilateral, or pentagonal. These three types of regions correspond to the above three types of (C, θ), according to the total number of black ovals and bridges on an odd theta-diagram, which can be respectively 3,4 or 5 (see the three rightmost diagrams of Fig. 3). References [At] M.Atiyah, Riemann surfaces and spin structures, Ann. Sci. ENS 4 (1971), 47 -- 62. [DIK] A.Degtyarev, I.Itenberg, V.Kharlamov, On the number of components of a complete inter- section of real quadrics, Perspectives in Analysis, Geometry, and Topology, Progr. Math., 296, Birkhauser/Springer, New York, 2012, pp. 81-107. A.C. Dixon, Note on the reduction of a ternary quantic to symmetrical determinant, Proc. Cambridge Philos. Soc. 5 (1902), 305 -- 351. I.Dolgachev, Classical algebraic geometry. A modern view. Cambridge University Press, Cambridge, 2012, pp. 639pp. [Do] [Di] [Ed] W. L. Edge, The Jacobian Curve of a Net of Quadrics, Proc. Edinburgh Math. Soc. 3, [FZ] no. 4, 259-268. S. Finashin, A. Zabun, Deformation classification of typical configurations of 7 points in the real projective plane, Topol. Appl. 194, (2015), 358 -- 385. [GH] B.H. Gross, J. Harris, On some geometric constructions related to Theta characteristics, Contributions to automorphic forms, geometry and number theory John Hopkins Press, 2004, pp. 279 -- 311. [Mu] D. Mumford, Theta characteristics of an algebraic curve, Ann. Sci. ENS Ser. 4 (2) (1971), [VV] 181 -- 192. J. Viro, O. Viro, Configurations of skew lines, Leningrad Mathematical Journal 1:4 (1990), 1027-1050 updated: arXiv:math/0611374. Middle East Technical University, Department of Mathematics Ankara 06800 Turkey
1803.07664
3
1803
2019-12-01T19:09:42
Order of tangency between manifolds
[ "math.AG", "math.DG" ]
We study the order of tangency between two manifolds of same dimension and give that notion three quite different geometric interpretations. Related aspects of the order of tangency, e.g., regular separation exponents, are also discussed.
math.AG
math
Order of tangency between manifolds Wojciech Domitrz, Piotr Mormul and Piotr Pragacz Abstract We study the order of tangency between two manifolds of same dimen- sion and give that notion three quite different geometric interpretations. Related aspects of the order of tangency, e. g., regular separation expo- nents, are also discussed. Keywords and Phrases. Order of tangency. Order of contact. Taylor polynomial. Higher jets. Tower of Grassmannians. Regular separation exponent. Lojasiewicz exponent. 2010 Mathematics Subject Classification. Primary 14C17, 14M15, 14N10, 14N15, 14H99, 14P10, 14P20, 32B20, 32C07. 1 Introduction In the present paper we discuss the order of tangency (or that of contact) be- tween manifolds and its relation to enumerative geometry started with classical Schubert calculus. Two plane curves, both sufficiently smooth and nonsingular at a point x0, are said to have a contact of order at least k at x0 if, in properly chosen regular parametrizations, those two curves have identical Taylor polynomials of degree k about the respective preimages of x0.1 Alternatively, those curves have such contact when their minimal regular separation exponent at x0, cf. [11], is strictly bigger than k or is not defined. Formulas enumerating contacts have been widely investigated. For example in [3] the authors derive a formula for the number of contacts of order n between members of specified (n−1)-parameter family of plane curves and a generic plane curve of a sufficiently high degree. Contact problems of this sort have been of both old and new interests, par- ticularly in the light of Hilbert's 15th problem to make rigorous the classical calculations of enumerative geometry, especially those undertaken by Schubert [16]. The situation regarding ordinary (i. e., second -- order) contacts between families of varieties is now well understood thanks in large measure to the con- tact formula of Fulton, Kleiman and MacPherson [6]. The above mentioned formula in [3] generalises that given by Schubert in [17] for the number of triple 1 Some authors prefer to use at this place Taylor polynomials of degree k − 1 instead, see for instance [3]. 1 contacts between a given plane curve and a specified 2-parameter family of curves. Schubert made his computations through the use of what has come to be known as "Schubert triangles". This theory has been made completely rigorous by Roberts and Speiser, see, e. g., [15], and independently by Collino and Fulton [2]. Apart from contact formulas, an important role is played by the "order of tangency". Let us discuss this notion for Thom polynomials. Among important properties of Thom polynomials we record their positivity closely related to Schubert calculus (see, e. g., [12] and [14] for a survey). Namely, the order of tangency allows one to define the jets of Lagrangian submanifolds. The space of these jets is a fibration over the Lagrangian Grassmannian and leads to a positive decomposition of the Lagrangian Thom polynomial in the basis of Lagrangian Schubert cycles. In this paper, we give three approaches to the order of tangency. The first one (in section 2) is by the Taylor approximations of local parametrizations of manifolds. The second one (a mini-max procedure in section 3) makes use of curves sitting in the relevant manifolds. The third approach (in section 4) is by Grassmann bundles. We show that these three approaches are equivalent. We basically work with manifolds over the reals (of various classes of smoothness), but the results carry over -- in the holomorphic category -- to complex manifolds. In the last two sections, we discuss some issues related to the "closeness" of pairs of geometric objects: branches of algebraic sets and relations with contact geometry. In fact, in section 5 discussed are the regular separation exponents of pairs of semialgebraic sets, sometimes called Lojasiewicz exponents (not to be mixed with the by now classical exponents in the renowned Lojasiewicz inequality for analytic functions). Then, in section 6 we report on an unexpected application of a modification of tangency order in 3D which yields an elegant criterion for a rank-2 distribution on a 3-manifold to be contact. These concluding sections are not less important than the preceding ones. They show that the precise measurement of closeness is sometimes more demanding -- and giving more -- than merely bounding below tangency orders. In the case of singular varieties different approaches to tangency orders lead to different notions. In this respect we refer the reader to [5] where compared were two discrete symplectic invariants of singular curves: the Lagrangian tan- gency order and index of isotropy. We firstly thank the anonymous referee of [14] for the report which was very stimulating for our present studies. Also, we thank Tadeusz Krasi´nski for informing us about the regular separation exponents of pairs of sets, a notion due to Lojasiewicz. Lastly, we thank anonymous referees of the present work for their meticulous inspection and advice. 2 2 By Taylor One situation that is frequently encountered at the crossroads of geometry and analysis deals with pairs of manifolds which are the graphs of functions of the same number of variables. Such graphs can intersect, or touch each other, at a prescribed point, with various degrees of closeness. Our departing point is a definition of such proximity going precisely in the spirit of a benchmark reference book [9], p. 18, although not formulated expressis verbis there. Definition. Two manifolds M and fM embedded in Rm, both of class Cr, r ≥ 1, and the same dimension p, intersecting at x0 ∈ M ∩ fM , for k ≤ r, have at x0 the order of tangency at least k, when there exist a neighbourhood U ∋ u0 in Rp and parametrizations2 (diffeomorphisms onto the image) q :(cid:0)U, u0(cid:1) →(cid:0)M, x0(cid:1), q :(cid:0)U, u0(cid:1) →(cid:0)fM , x0(cid:1) (cid:16)q − q(cid:17)(u) = o(cid:16)(cid:12)(cid:12)u − u0(cid:12)(cid:12)k(cid:17) of class Cr such that when U ∋ u → u0. (1) (We underline the existence clause in this definition. Supposing having already such a couple of local parametrizations q and q, there is an abundance of other pairs of Cr parametrizations serving the vicinities of x0 in M and fM , respec- tively, and not satisfying the condition (1). Note also that in this definition the order of tangency is automatically at least 0.) Below in •• in section 3, and also in section 4 we restrict ourselves to parametriza- tions of very specific type -- just the graphs of Cr mappings going from p di- mensions to m − p dimensions. This appears to be possible while not violating the key condition (1). Naturally enough, the notion of the order of tangency not smaller than . . . is invariant under the local Cr diffeomorphisms of neighbourhoods in Rm of the tangency point x0. Attention. In the real C∞ category it is possible for the order of tangency to be at least k for all k ∈ N. In other words -- be infinite even though {x0} = M ∩fM . The rest of this paper is to be read with this remark in mind. As a matter of record, basically the same definition is evoked in Proposition on page 4 in [8]. In [8] there is also proposed the following reformulation of (1). Proposition 1 The condition (1) is equivalent to T k u0(cid:0)q(cid:1) = T k u0(cid:0)q(cid:1) , where T k u0(cid:0) ·(cid:1) means the Taylor polynomial about u0 of degree k. 2 the standard topology language adopted, among many other sources, in [13] (2) 3 .(1) ⇒ (2). o(cid:16)(cid:12)(cid:12)u − u0(cid:12)(cid:12)k(cid:17) = q(u) − q(u) =(cid:0)q(u) − T k u0(cid:0)q(cid:1)(u − u0)(cid:1) +(cid:0)T k u0(cid:0)q(cid:1)(u − u0) − T k + (cid:0)T k u0(cid:0)q(cid:1)(u − u0)(cid:1) u0(cid:0)q(cid:1)(u − u0) − q(u)(cid:1) , where the first and last summands on the right hand side are o(cid:16)(cid:12)(cid:12)u − u0(cid:12)(cid:12)k(cid:17) by Taylor. Under (1), so is the middle summand (3) T k u0(cid:0)q(cid:1)(u − u0) − T k u0(cid:0)q(cid:1)(u − u0) = o(cid:16)(cid:12)(cid:12)u − u0(cid:12)(cid:12)k(cid:17) and (2) follows from the following general result. Lemma 1 Let w ∈ R[u1, u2, . . . , up] , deg w ≤ k, w(u) = o(cid:0)uk(cid:1) when u → 0 in Rp. Then w is identically zero. The proof goes by induction on k ≥ 0, with an obvious start for k = 0. Then, assuming this for the polynomials of degrees smaller than k ≥ 1 and taking a polynomial w of degree k as in the wording of the lemma, we can assume without loss of generality that w is homogeneous of degree k (the terms of lower degrees vanish altogether by the inductive assumption). Let u ∈ Rp, u = 1,be otherwise arbitrary. Then tkw(u) = w(tu) = o(cid:0)tuk(cid:1) = o(cid:0)tk(cid:1) when t → 0 . Hence w(u) = 0 and the vanishing of w follows. .(1) ⇐ (2). This implication is obvious, because now the middle term on the right hand side of (3) vanishes, so that the right hand side is automatically o(cid:16)(cid:12)(cid:12)u − u0(cid:12)(cid:12)k(cid:17). (cid:3) 3 By curves In the discussion in this section important will be the quantity s : = sup{k ∈ N : the order of tangency ≥ k} . (4) (Note that an additional restriction here on k is k ≤ r, cf. Definition above.) If the class of smoothness r = ∞, then, by the very definition, the condition (1) holds for all k if and only if s = ∞. Is it possible to ascertain something similar in the finite-order-of-tangency case? With an answer to this question in view, we stick in the present section to the notation introduced in section 2, but assume additionally that s < r . 4 (5) (Reiterating, the quantity s is defined in (4) above, and r is the assumed class of smoothness of the underlying manifolds, finite or infinite when the category is real. When r = ∞, the condition (5) simply says that s is finite.) min Theorem 1 Under (5), Our second approach uses pairs of curves lying, respectively, in M and fM . We naturally assume that Tx0M = Tx0fM . Our actual objective is to show that γ, γ(cid:0) max(cid:8)l ∈ {0} ∪ N : γ(t) − γ(t) = o(cid:0)tl(cid:1) when t → 0(cid:9)(cid:1)(cid:17) = s . (6) v (cid:16)max The minimum is taken over all 0 6= v ∈ Tx0M = Tx0fM . The outer maxi- mum is taken over all pairs of Cr curves γ ⊂ M , γ ⊂ fM such that γ(0) = x0 = γ(0), and -- both non-zero! -- velocities γ(0), γ(0) are both parallel to v. Attention. In this theorem the assumption (5) is essential; our proof would not work in the situation s = r. Proof of Theorem 1. It is quick to show that the integer on the left hand side of equality (6) is at least s. Indeed, for every fixed vector v as above, v = dq(u0)u (without loss of generality, u is like in the proof of Lemma 1), one can take δ(t) = q(u0 + tu) and δ(t) = q(cid:0)u0 + tu(cid:1). Then δ(t) − δ(t) = o(cid:0)tus(cid:1) = o(cid:0)ts(cid:1) and so max γ, γ(cid:0) max(cid:8) l : γ(t) − γ(t) = o(cid:0)tl(cid:1) when t → 0(cid:9)(cid:1) ≥ s . In view of the arbitrariness in our choice of v, the same remains true after taking the minimum over all admissible v's which is actually done on the left hand side of (6). •• To show the opposite non-sharp inequality in (6) is more involved. It is pre- cisely in this part that the additional assumption s ≤ r − 1 is needed. We study the two manifolds in the vicinity of x0 via an appropriate local Cr diffeomor- phism of the ambient space, after which (cid:0)M, x0(cid:1) =(cid:16)(cid:8)xp+1 = xp+2 = · · · = xm = 0(cid:9), 0(cid:17) and (fM , x0) =(cid:16)(cid:8)xj = F j(x1, x2, . . . , xp) , j = p + 1, p + 2, . . . , m(cid:9), 0(cid:17) for some Cr functions F j. Having the manifolds so neatly (graph-like) posi- tioned, we take the most adapted parametrizations q(u1, u2, . . . , up) =(cid:0)u1, u2, . . . , up, 0, 0, . . . , 0(cid:1) , q(cid:0)u1, u2, . . . , up(cid:1) =(cid:16)u1, u2, . . . , up, F(cid:0)u1, u2, . . . , up(cid:1)(cid:17) , 5 extra technical work. Firstly the initial couple of parametrizations satisfying where F = (cid:0)F p+1, F p+2, . . . , F m(cid:1). This -- important -- necessitates some (1) is being straightened simultaneously with manifolds M and fM . Naturally enough, the resulting parametrizations keep satisfying (1), but are not yet of the above-desired form. So the parametrizations and manifolds are to be ad- ditionally slightly upgraded via another local Cr ambient diffeomorphism so as (a) to keep the simple description of manifolds and (b) to have the eventual parametrizations adapted as desired above. Given the definition (4) of s, there hold that is, u0(q) = T s T s u0(cid:0)q(cid:1) T s+1 u0 (q) 6= T s+1 u0 (cid:0)q(cid:1) , T s u0(F ) = 0 T s+1 u0 (F ) 6= 0 . and and and (7) (8) It follows that there exist an integer j ∈ {p + 1, p + 2, . . . , m} and a vector w ∈ Rp such that T s u0(F j(cid:1)(w) = 0 T s+1 u0 (cid:0)F j(cid:1)(w) 6= 0 . Now let u and u be two Cr curves in Rp passing at t = 0 through u0 and such that the vectors u(0) and u(0) are both non-zero and parallel to w. These curves in parameters give rise to Cr curves δ(t) = q(u(t)) and δ(t) = q(u(t)) in the manifolds, both having at t = 0 non-zero speeds parallel to the vector v : = dq(u0)w = dq(u0)w. We will now estimate from above (by s) the left hand side of the equality (6) using, no wonder, v, δ, and δ: δ(t) − δ(t) =qu(t) − u(t)2 + F(cid:0)u(t)(cid:1)2 ≥ F(cid:0)u(t)(cid:1) ≥ F j(cid:0)u(t)(cid:1) 6= o(cid:0)ts+1(cid:1) , u0 (cid:0)F j(cid:1)(w) 6= o(cid:0)ts+1(cid:1) when t → 0 . where the last inequality necessitates an explanation. In fact, by (7) and for every c 6= 0 T s+1 u0 (cid:0)F j(cid:1)(tcw) = (ct)s+1T s+1 as well. Also, just by Peano in the class of smoothness s + 1 ≤ r, cf. (5), T s+1 But u(t) − u(0) = ctw + o(cid:0)t(cid:1) for some non-zero c, hence u0 (cid:0)F j(cid:1)(cid:0)u(t) − u(0)(cid:1) 6= o(cid:0)ts+1(cid:1) when t → 0 u0 (cid:0)F j(cid:1)(cid:0)u − u0(cid:1) + o(cid:16)(cid:12)(cid:12)u − u0(cid:12)(cid:12)s+1(cid:17) F j(u) = T s+1 when u → u0 in Rp. Therefore, F j(cid:0)u(t)(cid:1) 6= o(cid:0)ts+1(cid:1) as written in (8). Now it is important to note that the pair of curves δ and δ produced by us above is completely general in the category Cr for that chosen vector v. Hence it follows that -- for this precise vector v ! -- the quantity max γ, γ(cid:0) max(cid:8) l ∈ {0} ∪ N : γ(t) − γ(t) = o(cid:0)tl(cid:1) when t → 0(cid:9)(cid:1) 6 does not exceed s. Understandingly, so does the minimum of such quantities (cid:3) over all v's in Tx0M = Tx0fM . Theorem 1 is now proved. 4 By Grassmannians Our third approach is based on the introductory pages of [8] where a natural tower of consecutive Grassmannians is being attached to every given local Cr parametrization q as used by us in the preceding sections. However, to allow for a recursive definition of tower's members, a more general framework is needed. Namely, to every C1 immersion H : N → N ′, N -- an n-dimensional man- ifold, N ′ -- an n′-dimensional manifold (manifolds not necessarily embedded in Euclidean spaces!), we attach the so-called image map GH : N → Gn(N ′) of the tangent map d H: for s ∈ N , GH(s) = dH(s)(TsN ) , (9) where Gn(N ′) is the total space of the Grassmann bundle, with base N ′, of all n-planes tangent to N ′. That is, Gn(N ′) is a new manifold, much bigger than N ′ (whenever n′ > n), of dimension n′ + n(n′ − n). We stick in the present section to the notation from section 2 and invariably use the pair of parametrizations q and q. So we are given the mappings G q : U −→ Gp(cid:0)Rm(cid:1) , G q : U −→ Gp(cid:0)Rm(cid:1) . Upon putting M (0) = Rm, G(1) = G, there emerge two sequences of recursively defined mappings. Namely, for l ≥ 1, G(l)q : U −→ Gp(cid:0)M (l−1)(cid:1) , G(l) q : U −→ Gp(cid:0)M (l−1)(cid:1) , G(l+1)q = G(cid:16)G(l)q(cid:17) G(l+1) q = G(cid:16)G(l) q(cid:17), and ing. where, naturally, M (l) = Gp(cid:0)M (l−1)(cid:1). Now our objective is to show the follow- Theorem 2 Cr manifolds M and fM have at x0 the order of tangency at least x0 in, respectively, M and fM , such that k (1 ≤ k ≤ r) iff there exist Cr parametrizations q and q of the vicinities of G(k)q (u0) = G(k) q (u0) . (10) (Observe that, in (10), there is clearly encoded that q(u0) = x0 = q(u0).) 7 4.1 Proof of Theorem 2 Then (9) assumes by far more precise form In what follows, of interest for us will be the situations when H in (9) above is locally (and all is local in tangency considerations!) the graph of a C1 mapping h : Rp ⊃ U → Rt. That is, for u ∈ U , H(u) = (cid:0)u, h(u)(cid:1) ∈ Rp+t = Rp × Rt. GH(u) =(cid:16)u, h(u) ; d(cid:0)u, h(u)(cid:1)(u)(cid:17) =(cid:16)u, h(u); span{∂j+hj(u) : j = 1, 2, . . . , p}(cid:17) derivative of the vector mapping (ι, h) : U → Rp(cid:0)u1, . . . , up(cid:1) × Rt with respect (11) where the symbol hj means the partial derivative of a vector mapping h with respect to the variable uj (j = 1, . . . , p), and ∂j + hj(u) denotes the partial to uj, where ι : U ֒→ Rp is the inclusion. Now observe that the expression for GH(u) on the right hand side of (11) is still not quite useful. Yet there are charts in each newly appearing Grassmannian (see, for instance, [8] or p. 46 in [1])! The chart in a typical fibre Gp over a point in the base Rp+t, good for (11), consists of all the entries in the bottommost rows (indexed by numbers p + 1, p + 2, . . . , p + t) in the (p + t) × p matrices (cid:2) v1 v2 . . . vp (cid:3) with non-zero upper p × p minor, after multiplying the matrix on the right by the inverse of that upper p × p submatrix. That is to say, taking as the local coordinates all the entries in rows (p + 1)-st,. . . , (p + t)-th of the matrix That is, these coordinates are all t × p entries of the matrix . 1≤j≤p (cid:18)(cid:2)v i ji1≤i≤p+t hv i 1≤j≤p (cid:18)(cid:2)v i jip+1≤i≤p+t hv i GH(u) =(cid:16)u, h(u) ; 1≤j≤p(cid:19)−1 j(cid:3)1≤i≤p 1≤j≤p(cid:19)−1 j(cid:3)1≤i≤p (u)(cid:17), ∂h ∂u . In these, extremely useful, glasses the description (11) gets stenographed to (12) where, under the symbol ∂h ∂u (u) understood are all the entries of this Jacobian (t × p)-matrix written in row and separated by commas. This technical simpli- fication is central for a proof that follows. After this, basically algebraic, preparation we come back to Theorem 2. The order of tangency between M and fM at x0 being at least k precisely means (Proposition 1) the existence of local Cr parametrizations q and q satisfying 8 (2). So we are just going to show that (10) ⇐⇒ (2). Moreover, we assume without loss of generality -- much like it has been the case of parametrizations q and q, respectively. Which, at the same time, satisfy in the part •• of the proof of Theorem 1 -- that M and fM are locally the graphs (2). So (2) holds for q(u) = (u, f (u)), f : U → Rm−p(cid:0)yp+1, . . . , ym(cid:1) and for q(u) = (u, f (u)), f : U → Rm−p(cid:0)yp+1, . . . , ym(cid:1), x0 =(cid:0)u0, f (u0)(cid:1) =(cid:0)u0, f (u0)(cid:1). .(2) ⇒ (10). We will derive such expressions for G(k)q (u) and G(k) q (u), u ∈ U , that the use of the condition (2) will just prompt by itself. An added value of this derivation will be the control over the sets of natural local coordinates in the Grassmannians in question. (With this information at hand the opposite implication (2) ⇐ (10) will follow in no time.) Our main technical tool for the ⇒ implication is Lemma 2 For 1 ≤ l ≤ k there exists a local chart on the Grassmannian space Gp(cid:0)M (l−1)(cid:1) in which the mapping G(l)q evaluated at u assumes the form l(cid:19) × f[l](u)(cid:19) , 2(cid:19) × f[2](u), . . . ,(cid:18)l 1(cid:19) × f[1](u), (cid:18) l (cid:18)u, f (u);(cid:18)l where f[ν](u) is a shorthand notation for the aggregate of all the partials of the ν-th order at u, of all the components of f , which are in the number (m−p)×p ν , and the symbol N × (∗) stands for the N copies going in row and separated by commas, of an object (∗). Attention. In this lemma we purposefully distinguish mixed derivatives taken in different orders, simply disregarding the Schwarz symmetricity discovery. Proof. l = 1. We note that G(1)q (u) =(cid:16)u, f (u); span{∂j + fj(u) : j = 1, 2, . . . , p}(cid:17) , in the relevant Jensen-Borisenko-Nikolaevskii chart, is nothing but (cid:0)u, f (u); f[1](u)(cid:1) = (cid:18)u, f (u);(cid:18)l 1(cid:19) × f[1](u)(cid:19) . The beginning of induction is done. l ⇒ l + 1, l < k. The mapping G(l)q : U → M (l), evaluated at u, is already written down, in appropriate local chart assumed to exist in M (l), as u, f (u),(cid:18) l 1(cid:19) × f[1](u),(cid:18) l 2(cid:19) × f[2](u), . . . ,(cid:18)l l(cid:19) × f[l](u)!. We work with G(l+1)q = G(cid:16)G(l)q(cid:17). Now, (13) being clearly of the form H(u) = (cid:0)u, h(u)(cid:1) in the previously introduced notation, the mapping h reads l(cid:19) × f[l](u)(cid:17). 2(cid:19) × f[2](u), . . . ,(cid:18)l h(u) = (cid:16)f (u),(cid:18) l 1(cid:19) × f[1](u),(cid:18) l (13) 9 In order to have GH(u) written down, in view of (12), one ought to write in row: u, then h(u), and then all the entries of the Jacobian matrix ∂h ∂u (u), also written in row and separated by commas. The latter, in our shorthand notation, are computed immediately. Namely ∂h ∂u 0(cid:19) × f[1](u),(cid:18) l 1(cid:19) × f[2](u),(cid:18) l 2(cid:19) × f[3](u), . . . ,(cid:18)l These entries on the right hand side are to be juxtaposed with the former entries (u) =(cid:16)(cid:18)l l(cid:19) × f[l+1](u)(cid:17). (cid:0)u, h(u)(cid:1). For better readability, we put together the groups of same partials the lemma). In view of the elementary identities(cid:0) l ν (cid:1), we get in (cid:16)u, f (u), (cid:18)l + 1 2 (cid:19)×f[2](u), . . . , (cid:18)l + 1 l + 1(cid:19)×f[l+1](u)(cid:17). ν−1(cid:1) +(cid:0) l ν(cid:1) =(cid:0)l+1 l (cid:19)×f[l](u), (cid:18)l + 1 (a yet another permutation of Grassmann-type coordinates, cf. the wording of 1 (cid:19)×f[1](u), (cid:18)l + 1 the outcome Lemma 2 is now proved by induction. We now take l = k in Lemma 2 and get, for arbitrary u ∈ U , two similar visualisations of G(k)q (u) and G(k) q (u). At that, the equality (2) holds true at u = u0. As a consequence, (10) follows. .(2) ⇐ (10). With the information on superpositions of the mappings G, gathered in the course of proving the implication (2) ⇒ (10), this opposite implication is clear. Theorem 2 is now proved. (cid:3) 5 Algebraic geometry examples and regular sep- aration exponents In the present section, we shall work with the regular separation exponents of pairs of sets, a notion due to Lojasiewicz [11]. We shall compute these exponents in several natural examples, and compare the results with the information (when available) about the relevant orders of tangency. These examples deal with branches of algebraic sets which often happen to be tangent one to another, with various degrees of closeness. Example 1. In the work [4] (Figure 2 on page 37 there) analyzed is the following algebraic set in R2(x, y) C = {(x, y) : (y − x2)2 = x5} . (14) The two branches of C issuing from the point (0, 0), C− = {y = x2 − x5/2 , x ≥ 0} and C+ = {y = x2 + x5/2 , x ≥ 0} , could be naturally extended to one-dimensional manifolds D− and D+, both of class C2 -- the graphs of functions y−(x) = x2 − x5/2 and y+(x) = x2 + x5/2 , 10 respectively. The Taylor polynomials of degree 2 about x = 0 of y− and y+ coincide. Hence D− and D+ have at (0, 0) the order of tangency at least 2 (cf. section 2), and clearly not at least 3. This example clearly suggests that, in real algebraic geometry, it would be pertinent to use non-integer measures of closeness. For instance, for the above sets y−(x) and y+(x), we may take sup{α > 0 : y+(x) − y−(x) = o(xα) when x → 0} . This kind of a generalized order of tangency would be 5/2 in the Colley-Kennedy example. In the local analytic geometry there is a precise name for this notion -- the minimal regular separation exponent for two (semialgebraic) branches, say X and Y , of an algebraic set. That is, here the minimal exponent for X = C− and Y = C+. Some authors working after [11] used to call such a quantity the Lojasiewicz exponent of, in this case, C−, C+ at (0, 0). And denoted it -- when specialized to the present situation -- by L(0,0)(cid:0)C−, C+(cid:1).3 Remark 1. (i) Therefore, in Example 1, the rational number 5/2 is the minimal regular separation exponent of the semialgebraic sets C− and C+ which touch each other at (0, 0). (ii) Example 1 quickly generalizes, by means of the equation (y − xN )2 = x2N +1 with N arbitrarily large, to yield a pair of CN manifolds having the order of tangency at least N and not at least N + 1, and having the minimal regular separation exponent N + 1 2 . It has not been difficult in Example 1 to discern the pair of branches C− and C+, initially slightly hidden in a synthetic equation (14). But it can happen considerably worse in this respect. Consider, for instance (a) an algebraic set in the plane R2(x, y) defined by a single Example 2. equation (xy)2 = . (15) 1 4(cid:16)x2 + y2(cid:17)3 This set possesses a pair (even more than one such pair) of semialgebraic branches touching each other at the point (0, 0). Yet it is not so immediate to ascertain their minimal regular separation exponent. Only after recognizing in (15) the classical quatrefoil x = cos(ϕ) sin(2ϕ), y = sin(ϕ) sin(2ϕ), it becomes quick to compute the relevant minimal regular separation exponent equal to 2. (b) It is even more interestingly with another algebraic set in 2D given by the equation (cid:16)x2 + y2 − 1 2 x(cid:17)2 = 1 4(cid:0)x2 + y2(cid:1). (16) 3 However, this terminology is not yet definitely settled, as shown in a recent work [10]. The authors of the latter speak just descriptively about 'the Lojasiewicz exponent for the regular separation of closed semialgebraic sets'. 11 This set possesses as well a pair of semialgebraic branches {y ≤ 0} and {y ≥ 0} touching each other at the point (0, 0). Yet it takes some time to find their minimal regular separation exponent. In fact, after discovering in (16) the classical cardioid r = 1 2 (1 + cos ϕ), that exponent turns out to be -- one more time -- a non-integer (3/2, in the occurrence). (c) It is worthy of note that for both explicitly defined algebraic sets (15) and (16) one can apply general type bounds above, for the minimal separation exponent, produced in [10]. Yet the estimations got in that way are unrealistically (by factors of thousands) high. Returning to the notion of the order of tangency, in the realm of algebraic geometry the distinction between that order and minimal regular separation exponent sometimes happens to be fairly clear, with both discussed quantities effectively computable. An instructive instance of such a situation occurs in [18] (Example 3.5 there). Example 3. The author of [18] deals there with a pair of one-dimensional algebraic manifolds N and Z in R2(x, y) intersecting at (0, 0). The manifold N = {y = 0} is already utmostly simplified, whereas Z = {yd + yxd−1 + xs = 0} depends on two integer parameters d and s, 1 < d < s, d odd. What we are going to discuss here is a kind of reworking of Tworzewski's original approach, see also Attention below. These manifolds have at (0, 0) the order of tangency at least s − d, and not at least s − d + 1, while their minimal regular separation exponent at (0, 0) is s − d + 1. Indeed -- to justify this one tries to present Z as the graph of a function y = y(x). Clearly, y(0) = 0 and a function y(x) could not be divisible by, for instance, xs+1. So, with no loss of generality, y(x) = xkz(x) − xs−d+1 for certain integer k ≥ 1 and another function z(x) such that z(0) 6= 0. • The possibility k < s − d + 1 boils rather quickly down to k = 1, and then to the relation(cid:0)z − xs−d(cid:1)d d being odd. •• So k ≥ s − d + 1 and now + z = 0, impossible at x = 0, for(cid:0)z(0)(cid:1)d + z(0) 6= 0, y(x) = xs−d+1z(x) − xs−d+1 for a certain function z(x). Upon substituting this y(x) to the defining equation of Z and simplifying, (z − 1)d x(d−1)(s−d) + z = 0. Hence z(0) = 0. The Implicit Function Theorem is applicable here around (0, 0), because ∂ ∂ z(cid:16)(z − 1)dx(d−1)(s−d) + z(cid:17)(cid:12)(cid:12)(cid:12)(cid:12)(0,0) = 1 . One gets a locally unique C∞ function z(x), z(0) = 0, hence also a locally unique function y(x) = xs−d+1z(x) − xs−d+1 whose graph is Z. Because the 12 function z vanishes at 0, the minuend in this expression for y is an 'o'of the subtrahend when x → 0. So the statements about the order of tangency and minimal regular separation exponent follow immediately. Attention. When 2d, the above-found resolving function y(x) is not the only + solution to the defining equation of Z. Namely, the necessary equality(cid:0)z(0)(cid:1)d z(0) = 0, z(0) 6= 0, is then possible with z(0) = −1 and = d(−1)d−1 + 1 = 1 − d 6= 0 . ∂ ∂ z(cid:16)(cid:0)z − xs−d(cid:1)d + z(cid:17)(cid:12)(cid:12)(cid:12)(cid:12)(0,−1) Hence the Implicit Function Theorem gives this time a locally unique (for that k = 1) C∞ function z(x), z(0) = −1. Then the graph of y(x) = x z(x) − xs−d+1 = −x + (higher powers of x) is a second branch of Z passing through (0, 0) ∈ R2, transversal to N , in a stark distinction to the previously found, tangent to N , branch. Remark 2. More generally, one could not hope to get a precise information regarding the minimal regular separation exponent for the pair of manifolds seen, r = 2 and the quantity s defined in (4) is also 2, while the minimal regular separation exponent is but s + 1 under (5). Despite the inequality (8), that exponent need not necessarily be s + 1. Following Example 1 earlier in this section, one could just take the curves C− and C+ as the curves δ and δ, respectively, in the proof of Theorem 1. That M, fM on the sole basis of the assumptions in Theorem 1. That is, basically, is, to take M = C− = δ and fM = C+ = δ. Then, as the reader has lately Even restricting oneself to a benchmark setting dim M = dimfM , an addi- M ∩ fM might be a topologically highly nontrivial set (think about the C∞ category). And it is precisely M ∩ fM which enters the definition of regular separation exponents for the pair M, fM . tional enormous complication could come from the fact that the intersection 2 ( = 5 2 ). 6 Relation with contact topology Unsurprisingly, the notion of order of contact proves useful not only in algebraic geometry (cf. Introduction), but also in geometry tout court. One not so obvious application in the real category deals with the real contact structures in three dimensions. Our summarising it here follows closely Section 1.6 in [7]. The author considers there a couple Σ ⊂ M , where M is a contact 3-dimensional manifold and Σ -- a fixed embedded surface in it. Contact means M being endowed with a contact structure, say ξ, in T M . When one approaches a given point p ∈ Σ by points q staying within Σ, a natural question is about the order of smallness of the angle ∠(cid:0)TqΣ, ξq(cid:1). If that angle is an 'O' of the distance of q to p taken to power k (the distance measured 13 in any chosen, and hence every, set of smooth local coordinates about p), then it is said that ξ has the order of contact at least k with Σ at p. (Therefore, what is discussed in this section differs a little from the notion of closeness of a pair of manifolds investigated in the preceding sections. Yet the added value is substantial.) That is, to say that the new order of contact is at least 1 at a given point p is tantamount to saying that ξp = TpΣ. And it is exactly 0 at p whenever ξp 6= TpΣ. So it comes as a not small surprise that this elementary notion allows one to characterise the contact structures as such! Namely, a theorem proved in [7] asserts that a rank-2 tangent distribution ξ on a 3-dimensional M is contact iff ξ has the new order of contact at most 1 with every surface Σ embedded in M , and this at every point of Σ. The next natural question in this direction is whether it is possible to simi- larly characterise contact structures on (2n + 1)-dimensional manifolds, n ≥ 2. The author of [7] says nothing in this respect. References [1] A. A. Borisenko, Yu. A. Nikolaevskii. Grassmannian manifolds and Grass- mann image of submanifolds, Russian Math. Surveys 46 (1991), 45 -- 94 (English translation). [2] A. Collino, W. Fulton. Intersection rings of spaces of triangles, M´em. Soc. Math. France (N.S.) 38 (1989), 75 -- 117. [3] S. J. Colley, G. Kennedy. A Higher-order contact formula for plane curves, Comm. Algebra 19(2) (1991), 479 -- 508. [4] S. J. Colley, G. Kennedy. Triple and quadruple contact of plane curves, Con- temp. Math. 123, AMS 1991, 31 -- 59. [5] W. Domitrz, Z. Tr¸ebska. Symplectic T7, T8 singularities and Lagrangian tan- gency orders, Proc. of the Edinbourgh Math. Soc. 55 (2012), 657 -- 683. [6] W. Fulton, S. L. Kleiman, R. MacPherson. About the enumeration of con- tacts, Lect. Notes Math. 997, Springer 1983, 156 -- 196. [7] H. Geiges. An Introduction to Contact Topology, Cambridge University Press, Cambridge 2008. [8] G. R. Jensen. Higher Order Contact of Submanifolds of Homogeneous Spaces, Lect. Notes Math. 610, Springer 1977. [9] I. S. Krasil'shchik, V. V. Lychagin, A. M. Vinogradov. Geometry of Jet Spaces and Nonlinear Partial Differential Equations, Nauka, Moscow 1986 (in Russian). 14 [10] K. Kurdyka, S. Spodzieja, A. Szlachci´nska. Metric properties of semialge- braic mappings, Discrete Comput. Geom. 55 (2016), 786 -- 800. [11] S. Lojasiewicz. On semi-analytic and subanalytic geometry, Banach Center Publ. 34 (1995), 89 -- 104. [12] M. Mikosz, P. Pragacz, A. Weber. Positivity of Thom polynomials II: the Lagrange singularities, Fund. Math. 202 (2009), 65 -- 79. [13] J. W. Milnor. Topology From the Differentiable Viewpoint, The University Press of Wirginia, Charlottesville 1965. [14] P. Pragacz. Positivity of Thom polynomials and Schubert calculus, Adv. Stud. in Pure Math. 71 (2016) "Schubert Calculus -- Osaka 2012", 419 -- 451. [15] J. Roberts, R. Speiser. Enumerative geometry of triangles I, Comm. Algebra 12 (1984), 1213 -- 1255. [16] H. C. H. Schubert. Kalkul der abzahlenden Geometrie, Teubner, Leipzig, 1879; reprinted by Springer-Verlag, Berlin 1979. [17] H. C. H. Schubert. Anzahlgeometrische Behandlung des Dreiecks, Math. Ann. 17 (1880), 153 -- 212. [18] P. Tworzewski. Isolated intersection multiplicity and regular separation of analytic sets, Ann. Polon. Math. 58 (1993), 213 -- 219. Wojciech Domitrz Faculty of Mathematics and Information Science Warsaw University of Technology Koszykowa 75, 00-662 Warszawa, Poland E-mail: [email protected] Piotr Mormul Institute of Mathematics, University of Warsaw Banacha 2, 02-097 Warszawa, Poland E-mail: [email protected] Piotr Pragacz Institute of Mathematics, Polish Academy of Sciences ´Sniadeckich 8, 00-656 Warszawa, Poland E-mail: [email protected] 15
1309.6028
1
1309
2013-09-24T02:29:49
The space of complete quotients
[ "math.AG" ]
We introduce complete quotients over the projective line and prove that they form smooth projective varieties. The resulting parameter spaces coincide with the varieties constructed in [HLS11] and [Shao11]. Hence they provide modular smooth compactifications with normal crossing boundaries of the spaces of algebraic maps from the projective line to Grassmannian varieties.
math.AG
math
THE SPACE OF COMPLETE QUOTIENTS YI HU AND YIJUN SHAO Abstract. We introduce complete quotients over the projective line and prove that they form smooth projective varieties. The resulting parameter spaces coincide with the varieties constructed in [HLS11] and [Shao11]. Hence they provide modular smooth compactifications with normal crossing boundaries of the spaces of algebraic maps from the projective line to Grassmannian varieties. 1. Introduction We work throughout over an algebraically closed field k of characteristic 0. Fix the Grassmannian Gr(k, V ) of k-dimensional subspace in a vector space V ∼= kn. The set Mord(P1, Gr(k, V )) of degree d algebraic maps from P1 to Gr(k, V ) has a natural It comes with a natural compactification, structure of a smooth quasi-projective variety. the Grothendieck Quot scheme Qd := Quotd,n−k VP1 /P1/k, parametrizing all equivalence classes [VP1 ։ F ] of quotients of degree d and rank n − k, where VP1 := V ⊗k OP1. The subset Qd := {[VP1 ։ F ] ∈ Qd F is locally free} is open in Qd and can be identified with the variety Mord(P1, Gr(k, V )). The compactification Qd is smooth but the boundary Qd \ Qd has rather intricate singu- larities. It comes equipped with a natural filtration by closed subsets Zd,0 ⊂ Zd,1 ⊂ · · · ⊂ Zd,d−1 = Qd \ Qd where Zd,r = {[VP1 ։ F ] ∈ Qd the torsion of F has degree ≥ d − r}, r = 0, . . . , d − 1. The subsets Zd,r admit natural subscheme structures (cf. Section 3, [Shao11]). Theorem 1.1 ([HLS11, Shao11]). The singularities of the subscheme Zd,r can be resolved by repeatedly blowing up Zd,0, Zd,1, · · · , Zd,r−1. Consequently, by iteratively blowing-up the Quot scheme Qd along Zd,0, Zd,1, · · · , Zd,d−1, we obtain a smooth compactification eQd such that the boundary eQd \ Qd is a simple normal crossing divisor. The authors of [HLS11] proved the case when k = 1. In his Ph.D thesis [Shao11], the second named author proved it for all Grassmannians Gr(k, V ), k ≥ 1. 1 2 YI HU AND YIJUN SHAO The main purpose of this paper is to provide eQd the following modular interpretation. For any coherent sheaf X over P1, we let X t denote the torsion subsheaf of X and X f = X/X t denote the free part of X. For any coherent sheaves A and B on P1, an extension of A by B is a short exact sequence 0 → B → X → A → 0, which will also be shorthanded by B ֌ X ։ A. Two extensions B ֌ X1 ։ A and B ֌ X2 ։ A differ by a scalar multiple λ if there is an isomorphism X1 ≃ X2 such that the following diagram 0 0 / B λ· / B X1 ≃ / X2 A / 0 / A / 0 commutes, where λ· stands for the multiplication by the scalar λ. We use [B ֌ X ։ A] to denote the equivalence class of the extension B ֌ X ։ A modulo scalar multiplication. Definition 1.2. A complete quotient of VP1 of degree d and rank n − k on P1 is either • a quotient [VP1 ։ X1] ∈ Qd such that X1 is locally free; or, • a sequence ([VP1 ։ X1], [X f 1 m]) with m ≥ 1 such that [VP1 ։ X1] ∈ Qd, for every 1 ≤ i ≤ m, Xi+1 is a non-split extension of X t i by X f i , and further, the last sheaf Xm+1 is the unique one that is locally free. ֌ Xm+1 ։ X t ֌ X2 ։ X t 1], . . . , [X f m It follows from the definition that all the sheaves X1, . . . , Xm+1 are coherent and have degree d and rank n − k. The main theorem of this paper is Theorem 1.3. The projective variety eQd parameterizes all complete quotients of VP1 of degree d and rank n − k. To prove this theorem, we calculate the normal bundle of the blowup center in each blowup of eQd −→ Qd as presented in [HLS11, Shao11], and modularly interpret the points in the normal bundle as the desired extensions. This work is the genus zero case of a larger project. The higher genus case will appear in forthcoming papers. Acknowledgement. While this paper being prepared, the first named author was partially supported by NSF DMS 0901136. He would also like to dedicate this work to the memory of Andrey Todorov who passed away in Jerusalem in the spring of 2012. The notations in this paper closely follow those in [Shao11] with occasional modifications. / / /   / /   / / / / / THE SPACE OF COMPLETE QUOTIENTS 3 2. The space of relative extensions 2.1. The space of non-split extensions. Let F and T be two coherent sheaves over P1 with F being locally free and T torsion. The vector space E = Ext1 P1(T, F ) can be identified with the set of all (isomorphism classes of) extensions F ֌ X ։ T of T by F . The zero element corresponds to the split extension F ֌ F ⊕ T ։ T while the remainders correspond to non-split ones. Therefore the projective space P(E) := Proj(Sym∗(E∨)), where E∨ is the linear dual of the vector space E, parametrizes the set of all non-split extensions of T by F up to scalar multiplication. Here again, as in the introduction, two extensions F ֌ X1 ։ T and F ֌ X2 ։ T differ by a nonzero scalar multiple λ if and only if there is an isomorphism X1 ≃ X2 that makes the following diagram commute: 0 0 / F λ· / F X1 ≃ / X2 T / T / 0 / 0 where λ· is the scalar multiplication by λ. We denote by [F ֌ X ։ T ] the point of P(E) corresponding to a non-split extension F ֌ X ։ T . One checks that an extension F ֌ X ։ T splits if and only if deg X t = deg T . For a discussion on universal extensions, see Example 2.1.12, [HL97] . We need to introduce a relative version of the space of non-split extensions. For this, we begin with a lemma. Lemma 2.1. Let S be a noetherian scheme, 0 → E1 → E0 → T → 0 a short exact sequence of coherent sheaves on S with T torsion and E0 and E1 locally free, and F a coherent torsion-free sheaf on S. Then (1) H 0(S, Ext 1(T , F)) = Ext1(T , F); (2) for any morphism f : R → S, f ∗ Ext 1 S(T , F) = Ext 1 R(f ∗T , f ∗F). Proof. (1) Since T is torsion and F is torsion-free, we have Hom(T , F) = 0 and Hom(T , F) = 0. Applying Hom(−, F) and Hom(−, F) to the locally free presentation of T , we obtain a long exact sequence 0 → Hom(E0, F) → Hom(E1, F) → Ext1(T , F) → Ext1(E0, F) → Ext1(E1, F) and a short exact sequence 0 → Hom(E0, F) → Hom(E1, F) → Ext 1(T , F) → 0 / / /   / /   / / / / / 4 YI HU AND YIJUN SHAO Taking global sections of the above short exact sequence, we obtain a long exact sequence: 0 → H 0(Hom(E0, F)) → H 0(Hom(E1, F)) → H 0(Ext 1(T , F)) → H 1(Hom(E0, F)) → H 1(Hom(E1, F)) We have identifications Hom(Ei, F) = H 0(Hom(Ei, F)), i = 0, 1 and a natural homomorphism Ext1(T , F) → H 0(Ext 1(T , F)) Using the Grothendieck spectral sequence for the composition of the two functors H 0 and Hom 1(Ei, −), we obtain exact sequences of low degrees: 0 → H 1(Hom(Ei, F)) → Ext1(Ei, F) → H 0(Ext 1(Ei, F)) → · · · , i = 0, 1. Since Ei are locally free, we have H 0(Ext 1(Ei, F)) = 0. Therefore we obtain identifications: H 1(Hom(Ei, F)) = Ext1(Ei, F), i = 0, 1 Thus we have a commutative diagram Hom(E0, F) / Hom(E1, F) / Ext1(T , F) Ext1(E0, F) / Ext1(E1, F) H 0(Hom(E0, F)) / H 0(Hom(E1, F)) / H 0(Ext 1(T , F)) / H 1(Hom(E0, F)) / H 1(Hom(E1, F)) By the Five Lemma, we obtain the identification H 0(Ext 1(T , F)) = Ext1(T , F) The proof of (2) is parallel to that of [Shao11], Proposition 2.3. So, we omit the details. (cid:3) 2.2. The space of relative non-split extensions. To proceed, we fix a set of notations. Notation. For any k-schemes R and S, (1) we denote the projection P1 × R → R by πR or simply π; (2) for any coherent sheaf H over R, we let H∨ denote the dual sheaf Hom R(H, OR); (3) for any coherent sheaf F on P1 × R, we set Fx := FP1×{x} for any point x ∈ R; (4) for any morphism f : R → S, we set ¯f := 1 × f : P1 × R → P1 × S. / / / /   / / / / / THE SPACE OF COMPLETE QUOTIENTS 5 We now introduce a relative version of the space of non-split extensions. Let S be a noetherian scheme over k and let F and T be two coherent sheaves on P1 × S, both flat over S, with F locally free and T torsion (i.e., having rank 0). Let πS : P1 × S → S be the projection, and set E := πS∗ Ext 1 P1×S(T , F). As in [Shao11], Proposition 2.4 (1), one checks that Ext 1(T , F) is a torsion sheaf and is flat over S. By Cohomology and Base Change, we can show E is locally free and for any point s ∈ S, Es = H 0(Ext 1(T , F)s) = H 0(Ext 1(Ts, Fs)) = Ext1(Ts, Fs). where the second equality holds by Lemma 2.1 (2) and the third holds by Lemma 2.1 (1). So the projective bundle P(E) := Proj(Sym∗(E ∨)) over S is a family of spaces of non-split extensions: for each point s ∈ S, the fiber of P(E) over s is P(Es) = P(Ext1(Ts, Fs)). 2.3. The universal extension. Notation. For any coherent sheaf H on P1 × P(E), we denote by H(m, n) the sheaf H ⊗ p∗OP1(m) ⊗ π∗ P(E)OP(E)(n) where p : P1 × P(E) → P1 and πP(E) : P1 × P(E) → P(E) are the two projections. Let a : P(E) → S be the structure morphism. The space P(E) comes equipped with a universal quotient a∗E ∨ ։ OP(E)(1). Dualizing the universal quotient and tensoring the result with OP(E)(1), we obtain a line subbundle OP(E) ֒→ a∗E ⊗ OP(E)(1) = πP(E)∗¯a∗ Ext 1(T , F) ⊗ OP(E)(1) = πP(E)∗(Ext 1(¯a∗T , ¯a∗F)(0, 1)) = πP(E)∗ Ext 1(¯a∗T , ¯a∗F(0, 1)) Here the first equality holds because the morphism a is flat, and the second holds by Lemma 2.1 (2). This line subbundle corresponds to a nonzero element of Γ(P(E), πP(E)∗ Ext 1(¯a∗T , ¯a∗F(0, 1))) = Γ(P1 × P(E), Ext 1(¯a∗T , ¯a∗F(0, 1))) = Ext1(¯a∗T , ¯a∗F(0, 1)) where the second equality holds by Lemma 2.1 (1). We can write this element as an extension (2.1) 0 → ¯a∗F(0, 1) → X → ¯a∗T → 0. 6 YI HU AND YIJUN SHAO which we call the universal extension. Note that the universal extension is nowhere-split, which means that, for each k-point s ∈ S and each k-point x ∈ a−1({s}) = P(Ext1(Ts, Fs)), the extension (2.2) 0 → Fs → Xx → Ts → 0 obtained by pulling back the universal extension to P1 × {x} ≃ P1 is non-split. Conversely, the universal quotient can be recovered from the universal extension by re- versing the above process. In fact, the universal extension itself is a nonzero element of Ext1(¯a∗T , ¯a∗F(0, 1)) = Γ(P1 × P(E), Ext 1(¯a∗T , ¯a∗F(0, 1)) = Γ(P(E), πP(E)∗ Ext 1(¯a∗T , ¯a∗F(0, 1))) = Γ(P(E), a∗E(1)) hence determines a line subbundle OP(E) ֒→ a∗E(1). Tensoring the line subbundle with OP(E)(−1) and taking the dual, we obtain the universal quotient. Thus, based on the universal property of the universal quotient, we have Theorem 2.2. Let R be an S-scheme with structure morphism ρ : R → S, L a line bundle on R. If 0 → ¯ρ∗F ⊗ π∗ RL → Y → ¯ρ∗T → 0 is a nowhere-split extension on P1 × R, then there g ≃ f ∗OP(E)(1), is a unique S-morphism f : R → P(E) such that there are isomorphisms L h ≃ ¯f ∗X that make the following diagram commute: Y 0 0 / ¯ρ∗F ⊗ π∗ RL ≃ 1⊗π∗ Rg / ¯f ∗¯a∗F ⊗ π∗ Rf ∗OP(E)(1) Y ≃h / ¯f ∗X ¯r∗T / ¯f ∗¯a∗T / 0 / 0 where the second row is the pullback of the universal extension via ¯f . 3. A flattening stratification In this section, we continue to follow the previous notations, and we impose an additional assumption on the sheaf F of Theorem 2.2 (also Lemma 2.1): (3.1) R1πS∗(F ⊗ p∗OP1(−1)) = 0, where p : P1 × S → P1 is the projection. We also need the following notation (cf. Introduction). Notation. For any coherent sheaf H on P1, we denote by H t the torsion subsheaf of H. The quotient H/H t is locally free and we will denote it by H f . Since H ≃ H t ⊕ H f on P1, we call H t and H f the torsion part and the locally free part of H, respectively. / / /   / /   / / / / / THE SPACE OF COMPLETE QUOTIENTS 7 Let d1 and d2 be the relative degrees of F and T over S, respectively. Then the universal extention X (see (2.1)) is of relative degree d := d1 + d2 over S. The set P(E) := {x ∈ P(E) Xx is locally free} is an open subset of P(E). Its complement P(E)\P(E) has a sequence of nested closed subsets: ∅ = Yd1 ⊂ Yd1+1 ⊂ · · · ⊂ Yd−1 = P(E) \ P(E) where d is the relative degree of T over S, and Yr = {x ∈ P(E) deg((Xx)t) ≥ d − r}, r = d1, . . . , d − 1. Note that an extension 0 → Fs → X → Ts → 0 is split if and only if the torsion part of X is of degree d2. Hence Yd1 = ∅. For m ≥ 0, applying Hom(−, O(m)) to the extension (2.2), we obtain an exact sequence 0 → Hom(Xx, O(m)) → Hom(Fs, O(m)) δx,m→ Ext1(Ts, O(m)) → Ext1(Xx, O(m)) → 0 where δx,m denotes the connecting homomorphism. Then we have Proposition 3.1. rank δx,m = d2 − deg(Xx)t for any m ≥ d. Proof. Suppose Fs has rank k. Then Xx has rank k as well. We have non-canonical isomor- phisms Fs ≃ O(ai), Xx ≃ (Xx)t ⊕ O(bi) where Pr i=1 ai = d1 and Pr H 1(P1, Fs(−1)) = 0 and also H 1(P1, Xx(−1)) = 0. all i. Hence we have ai ≤ d1 < d and bi ≤ d for all i. So i=1 bi = d − deg(Xx)t. By the assumption (3.1), we have It follows that ai ≥ 0 and bi ≥ 0 for Hom(Fs, O(m)) ≃ Hom(O(ai), O(m)) = H 0(O(m − ai)), and rMi=1 kMi=1 kMi=1 rMi=1 kMi=1 kMi=1 kXi=1 Hom(Xx, O(m)) ≃ Hom(O(bi), O(m)) = H 0(O(m − bi)). For m ≥ d, we have ai ≤ m and bi ≤ m, therefore rank δx,m = dim Hom(Fs, O(m)) − dim Hom(Xx, O(m)) = (m − ai + 1) − (m − bi + 1) = kXi=1 bi − kXi=1 ai = d − deg(Xx)t − d1 = d2 − deg(Xx)t 8 YI HU AND YIJUN SHAO (cid:3) Corollary 3.2. For any integers m, r with m ≥ d and d1 ≤ r ≤ d, we have deg(Xx)t ≥ d − r if and only if rank δx,m ≤ r − d1. Notation. For any scheme R and any coherent sheaf H on R, we will use the following abbreviations: Hε := Ext 1 R(H, OR) Corollary 3.2 suggests us that we can define the scheme structure of Yr in the following way. Let m ≫ 0. Applying Hom(−, OP1×P(E)(m, 1)) to the universal extension (2.1) to obtain an exact sequence 0 → X ∨(m, 1) → ¯a∗F ∨(m, 0) δm→ ¯a∗T ε(m, 1) → X ε(m, 1) → 0 where δm is the connecting homomorphism. Next, applying πP(E)∗ to δm and using identifi- cations π∗(¯a∗F ∨(m, 0)) = π∗¯a∗(F ∨(m)) = a∗(π∗F ∨(m)), π∗(¯a∗T ε(m, 1)) = π∗¯a∗(T ε(m))(1) = a∗(π∗T ε(m))(1) we obtain a nowhere-vanishing homomorphism (3.2) π∗δm : a∗(π∗F ∨(m)) → a∗(π∗T ε(m))(1) Note that both π∗F ∨(m) and π∗T ε(m) are locally free sheaves on S for m ≫ 0. Applying the exterior powerVl+1 (l ≥ 0) to π∗δm, we obtain l+1^ a∗(π∗F ∨(m)) → By Corollary 3.2, Yd1+l is exactly the (set-theoretic) zero locus ofVl+1 π∗δm, for each l with 0 ≤ l < d2. The sectionVl+1 π∗δm induces a homomorphism l+1^ a∗(π∗T ε(m))(l + 1) l+1^ π∗δm : Hom(cid:18) l+1^ a∗(π∗F ∨(m)), l+1^ a∗(π∗T ε(m))(l + 1)(cid:19)∨ → OP(E) The image, which we denote by Il,m, is an ideal sheaf. In a similar way as in [Shao11], Proposition 3.4, we can prove that Proposition 3.3. There exists an integer N > 0 such that Il,m = Il,N as subsheaves of OP(E) for all m > N and for all l with 0 ≤ l < d2. THE SPACE OF COMPLETE QUOTIENTS 9 Obviously, the subscheme defined by the ideal Il,N is supported on the closed subset Yd1+l. For simplicity, we denote this subscheme still by Yd1+l, that is, (3.3) Yr = the closed subscheme defined by the ideal Ir−d1+1,N , r = d1 + 1, . . . , d − 1 In addition, we set Yr := Yr \ Yr−1 for d1 < r < d. Note that Yd1+1 = Yd1+1 since Yd1 = ∅. Again, in a similar way as in [Shao11], Theorem 3.5, we can prove that Proposition 3.4. The locally closed subschemes Yd1+1, Yd1+2, · · · , Yd−1 of P(E) form the flattening stratification of P(E) by the sheaf X ε, which means that, for any noetherian k- scheme R and any morphism f : R → P(E), the sheaf ¯f ∗X ε on P1 × R is flat over R with relative degree d − r if and only if f factors through the inclusion Yr ֒→ P(E). In particular, the restriction of X ε to P1 × Yr is flat over Yr with relative degree d − r. 4. The normal bundles: a first case Let V be a vector space of dimension n over k, and Gr(k, V ) be the Grassmannian parametrizing all the k-dimensional subspaces of V . For any d ≥ 0, the space Mord(P1, Gr(k, V )) of degree d maps from P1 to Gr(k, V ) is a nonsingular quasi-projective variety. A smooth compactification of Mord(P1, Gr(k, V )) is given by the Quot scheme Qd := Quotn−k,d VP1 /P1/k, parametrizing all rank-(n − k), degree-d quotients VP1 ։ F of the trivial vector bundle VP1 of rank n on P1. It comes with a universal exact sequence of sheaves on P1 × Qd: 0 → Ed → VP1×Qd → Fd → 0 Here Fd is flat over Qd with rank n − k and relative degree d. It follows that Ed is locally free with rank k and relative degree −d over Qd. The open subvariety Qd := {x ∈ Qd (Fd)x is locally free} coincides with Mord(P1, Gr(k, V )). Recall from the introduction that for any d > r ≥ 0, we have the closed subscheme Zd,r = {[VP1 ։ F ] ∈ Qd deg(F t) ≥ d − r}. We refer the reader to Section 3 in [Shao11] for the details on the subscheme structure of Zd,r. These are the subschemes that are blown up to yield the variety eQd. Below we analyze the normal bundle of the locally closed subsets Zd,r := Zd,r \ Zd,r−1. The subscheme Zd,r is closely related to the following relative Quot scheme over Qr: Qd,r := Quot0,d−r Er/P1×Qr/Qr If a point of Qr is represented by the exact sequence E ֌ VP1 ։ F , then the fiber of Qd,r over the point consists of points represented by the quotient E ։ T with T torsion of degree 10 YI HU AND YIJUN SHAO d − r. Let θd,r : Qd,r → Qr be the structure morphism. (Note that the notation for this morphism is simply θ in [Shao11]. We add sub-index in this paper because we will deal with multiple Qd,r's together with their structure morphisms simultaneously. The same reason applies to the other similar situations below.) In [Shao11], we showed that Qd,r is relatively smooth over Qr, hence is a nonsingular variety. It comes equipped with a universal exact sequence on P1 × Qd,r: (4.1) 0 → Ed,r → θ∗ d,rEr → Td,r → 0 Here Td,r is flat over Qd,r with rank 0 and relative degree d − r. Since θ∗ d,rEr is locally free of rank k and of relative degree −r over Qd,r, it follows that Ed,r is locally free of rank k and d,r(Qr) and write θd,r : Qd,r → Qr for the of relative degree −d over Qd,r. We set Qd,r := θ−1 restriction of θd,r on Qd,r. We form a commutative diagram on P1 × Qd,r: (4.2) 0 / Ed,r / ¯θ∗ d,rEr 0 0 0 Td,r 0 0 / Ed,r / VP1×Qd,r Fd,r ¯θ∗ d,rFr ¯θ∗ d,rFr 0 0 where the middle column is the pullback of the universal exact sequence of Qr via ¯θd,r, and Fd,r is defined to be the cokernel of the composite map Ed,r → ¯θ∗ d,rEr → VP1×Qd,r . One checks that Fd,r is flat over Qd,r of rank n − k and relative degree d. By the universal property of Qd, the middle row determines a morphism φd,r : Qd,r → Qd (it is denoted by φ in [Shao11]) such that the following diagram commutes: (4.3) 0 0 / Ed,r / VP1×Qd,r / Fd,r ≃ / ¯φ∗ d,rEd / ¯φ∗ d,rVP1×Qd / ¯φ∗ d,rFd 0 / 0 The morphism φd,r maps Qd,r onto Zd,r (cf. Proposition 4.6 of [Shao11]). We denote by φd,r : Qd,r → Qd the restriction of φd,r on Qd,r. Note that φd,0 = φd,0 since Qd,0 = Qd,0. We showed that φd,r maps Qd,r into Zd,r ⊂ Qd in [Shao11], where Zd,r = Zd,r \ Zd,r−1, and we set     / / / /   / /   / / / /   / /       / / / / /   / / / / THE SPACE OF COMPLETE QUOTIENTS 11 Zd,−1 := ∅. We denote by ϕd,r : Qd,r → Zd,r the map obtained by restricting the codomain of φd,r to Zd,r. In [Shao11], we showed that ϕd,r : Qd,r → Zd,r is in fact an isomorphism of schemes, hence φd,r : Qd,r → Qd is an embedding. (cf. Proposition 4.8, [Shao11]) Proposition 4.1 ([Str87], Theorem 7.1). The tangent bundle TQd of Qd is naturally isomor- phic to π∗ Hom(Ed, Fd). The relative tangent bundle TQd,r/Qr of Qd,r over Qr is naturally isomorphic to π∗ Hom(Ed,r, Td,r). Proof. The first assertion is proved in [Str87], Theorem 7.1. The second, which is a relative version of the first, can be proved by slightly modifying the proof of [Str87], Theorem 7.1. We omit the details. (cid:3) The tangent bundle TQd,r of Qd,r fits into the following exact sequence (4.4) 0 → TQd,r/Qr → TQd,r → θ∗ d,rTQr → 0 The morphism φd,r : Qd,r → Qd is an embedding and is factored as Qd,0 Proposition 4.8 of [Shao11]). Let N Qd,r/Qd is nonsingular, N Qd,r/Qd ϕd,r ∼→ Zd,r ֒→ Qd (cf. denote the normal sheaf of Qd,0 in Qd. Since Qd,r is locally free. Proposition 4.2. We have a natural identification N Qd,r/Qd particular, letting r = 0, we have NQd,0/Qd = π∗ Ext 1(Td,0, ¯θ∗ d,0F0). = π∗ Ext 1(Td,r, ¯θ∗ d,rFr) Qd,r . In Proof. Since N Qd,r/Qd is locally free, we have the following exact sequence of sheaves on Qd,r: 0 → T Qd,r → φ∗ d,rTQd → N Qd,r/Qd → 0 Note that the restriction of the exact sequence (4.4) to Qd,r gives an exact sequence 0 → T Qd,r/ Qr → T Qd,r → θ∗ d,rTQr → 0. 12 YI HU AND YIJUN SHAO Combining the above two sequences, we can form a commutative diagram of sheaves on Qd,r: 0 / T Qd,r/ Qr / T Qd,r 0 θ∗ d,rTQr / T Qd,r/ Qr / φ∗ d,rTQd π∗ Hom(Ed,r, ¯θ∗ d,rFr) Qd,r 0 0 0 0 N Qd,r/Qd N Qd,r/Qd 0 0 where the middle row comes from the natural identifications T Qd,r/ Qr = π∗ Hom(Ed,r, Td,r) Qd,r , φ∗ d,rTQd = φ∗ d,rπ∗ Hom(Ed, Fd) = π∗ Hom(Ed,r, Fd,r) Qd,r by Proposition 4.1 and the exact sequence 0 → π∗ Hom(Ed,r, Td,r) → π∗ Hom(Ed,r, Fd,r) → π∗ Hom(Ed,r, ¯θ∗ d,rFr) → 0 obtained by applying π∗ Hom(Ed,r, −) to the third column of the diagram (4.2). The dotted arrows in the third column are induced maps on the quotients. Since all rows and the middle column are exact, the third column is forced to be exact as well. Using the identification d,rTQr = θ∗ θ∗ d,rπ∗ Hom(Er, Fr) = π∗ Hom(¯θ∗ d,rEr, ¯θ∗ d,rFr) Qd,r and comparing the third column with the short exact sequence 0 → π∗ Hom(¯θ∗ d,rEr, ¯θ∗ d,rFr) → π∗ Hom(Ed,r, ¯θ∗ d,rFr) → π∗ Ext 1(Td,r, ¯θ∗ d,rFr) → 0 obtained by applying π∗ Hom(−, ¯θ∗ identification d,0F0) to the exact sequence (4.1), we obtain a natural N Qd,r/Qd = π∗ Ext 1(Td,r, ¯θ∗ d,rFr) Qd,r . (cid:3) 5. The normal bundles: the general case In this technical section, we introduce and analyze the properties of a set of auxiliary schemes. These will be used in the final section to identify with and to derive the desired modular properties of the exceptional divisors created in the sequence of blowups eQd → Qd.     / / / /   / /   / / / /   / /       THE SPACE OF COMPLETE QUOTIENTS 13 5.1. The schemes Qd,r,l and morphisms ψd,r,l. Let d > r > l ≥ 0, and we consider : Qd,r,l → Qr,l be the two := Qd,r ×Qr Qr,l. Let ψd,r,l : Qd,r,l → Qd,r and ψd,r,l Qd,r,l projections. (Here the underscored subscript l in the map ψd,r,l indicates that the subscript l shows up in the source (Qd,r,l) but not in the target (Qd,r). Ditto for d, and for r below.) First of all, we see that Qd,r,l is smooth over Qr,l because Qd,r is smooth over Qr. It follows that Qd,r,l is a nonsingular variety. Next, we will define a finite morphism ψd,r,l : Qd,r,l → Qd,l based on the morphisms in the following diagram (5.1) Qr,l θr,l φr,l 8qqqqqqq f▲▲▲▲▲▲ ψd,r,l Qr 1(cid:13) Qd,r,l θd,r f▼▼▼▼▼▼▼ 8qqqqqq ψd,r,l Qd,r φd,r Qd Ql 2(cid:13) f▼▼▼▼▼▼▼ θd,l 3(cid:13) ψd,r,l 8qqqqqqq φd,l Qd,l The parallelogram 1(cid:13) is commutative by the definition of Qd,r,l. On Pr × Qd,r,l we have two short exact sequences: 0 → ¯ψ∗ 0 → ¯ψ∗ d,r,lEr,l → ¯ψ∗ d,r,lEd,r → ¯ψ∗ d,r,l d,r,l r,lEl → ¯ψ∗ ¯θ∗ d,rEr → ¯ψ∗ ¯θ∗ d,r,lTr,l → 0 d,r,lTd,r → 0 which are pullbacks of the universal exact sequences of Qr,l and Qd,r via ¯ψd,r,l and ¯ψd,r,l respectively. Note that ¯ψ∗ r,lEr = ¯ψd,r,lEr,l. Putting the two sequences ¯φ∗ d,rEr = ¯ψ∗ ¯θ∗ d,r,l d,r,l together, we can form a commutative diagram as follows: (5.2) 0 0 0 0 / ¯ψ∗ d,r,lEd,r / ¯ψ∗ d,r,lEr,l ¯ψ∗ d,r,lTd,r / ¯ψ∗ d,r,lEd,r / ¯ψ∗ d,r,l ¯θ∗ r,lEl T 0 0 ¯ψ∗ d,r,lTr,l ¯ψ∗ d,r,lTr,l 0 0 8     f f f   f f f 8   f f f 8     / / / /   / /   / / / /   / /       14 YI HU AND YIJUN SHAO d,r,l d,r,lEd,r → ¯ψ∗ d,r,lEr,l → ¯ψ∗ where T = Coker( ¯ψ∗ ¯θ∗ r,lEl), and the dotted arrows are the induced maps on quotients. Since the upper two rows and the middle column are exact, the last column are forced to be exact as well. Note that ¯ψ∗ d,r,lTr,l are both flat over Qd,r,l with rank 0 but with relative degree d − r and r − l respectively. Hence T is flat over ¯θ∗ Qd,r,l as well and is of rank 0 and of relative degree d − l. Thus the quotient ¯ψ∗ r,lEl ։ T from the middle row determines a Ql-morphism d,r,lTd,r and ¯ψ∗ d,r,l such that the pullback of the universal exact sequence of Qd,l is the middle row of (5.2), i.e., ψd,r,l : Qd,r,l → Qd,l we have the following identifications: (5.3) 0 0 / ¯ψ∗ d,r,lEd,r / ¯ψ∗ d,r,l ¯θ∗ r,lEl / T / ¯ψ∗ d,r,lEd,l / ¯ψ∗ d,r,l ¯θ∗ d,lEl / ¯ψ∗ d,r,lTd,l / 0 / 0 In particular, the identification T = ¯ψ∗ d,r,lTd,l allows us to rewrite the third column of the diagram (5.2) as (5.4) 0 → ¯ψ∗ d,r,lTd,r → ¯ψ∗ d,r,lTd,l → ¯ψ∗ d,r,lTr,l → 0 By the definition of ψd,r,l, the parallelogram 2(cid:13) in the diagram (5.1) automatically commutes. We also have Proposition 5.1. The parallelogram 3(cid:13) in diagram (5.1) commutes. Proof. We need to show φd,rψd,r,l = φd,lψd,r,l. Because both φd,rψd,r,l and φd,lψd,r,l map into Qd, by the universal property of the Quot scheme Qd (see [Shao11], Theorem 2.1), it suffices to show that there is an isomorphism (φd,rψd,r,l)∗Fd ≃ (φd,lψd,r,l)∗Fd that makes the following diagram commute: VP1×Qd,r,l (φd,rψd,r,l)∗VP1×Qd / (φd,rψd,r,l)∗Fd ≃ VP1×Qd,r,l (φd,lψd,r,l)∗VP1×Qd / (φd,lψd,r,l)∗Fd By the diagram (4.3), we have an exact sequence 0 → Ed,r → VP1×Qd,r → ¯φ∗ d,rFd → 0 Replacing r with l, we obtain another one 0 → Ed,l → VP1×Qd,l → ¯φ∗ d,lFd → 0 / / / / / / / / /   / THE SPACE OF COMPLETE QUOTIENTS 15 Applying ¯ψ∗ d,r,l and ¯ψ∗ d,r,l to the above two exact sequences respectively, we obtain two exact sequences 0 → ¯ψ∗ 0 → ¯ψ∗ d,r,lEd,r → VP1×Qd,r,l → ¯ψ∗ d,r,lEd,l → VP1×Qd,r,l → ¯ψ∗ d,r,l d,r,l ¯φ∗ d,rFd → 0 ¯φ∗ d,lFd → 0 Using the identification ¯ψ∗ d,r,lEd,r = ¯ψ∗ d,r,lEd,l from the diagram (5.3), we obtain a commutative diagram 0 0 / ¯ψ∗ d,r,lEd,r / VP1×Qd,r,l / ¯ψ∗ d,r,l / ¯ψ∗ d,r,lEd,l / ¯ψ∗ d,r,l / VP1×Qd,r,l d,rFd ≃ ¯ψ∗ ¯φ∗ d,r,l ¯φ∗ d,rFd ≃ ¯φ∗ d,lFd 0 / 0 which induces an isomorphism ¯ψ∗ d,r,l phism is the desired one. ¯φ∗ d,lFd in the third column. This isomor- (cid:3) The map ψd,r,l is both proper and quasi-finite, hence it is a finite morphism. Notation. For any integer m ≥ 1, let Σm be the set of all strictly decreasing sequences of nonnegative integers of length m: Σm = {(r1, · · · , rm) r1 > · · · > rm} and let Σ be the set of all strictly decreasing sequences of nonnegative integers of any finite length: Σ = Σm ∞[m=1 For any sequence σ = (r1, · · · , rm) ∈ Σ, the first term r1 is called the leading term and is denoted as lt(σ). For any integer r > lt(σ), by (r, σ) we mean the new sequence (r, r1, · · · , rm). The length of a sequence σ is denoted by σ. When we use a sequence of Σ as sub-index, we would omit the parentheses. For example, if σ = (8, 5, 3, 1, 0), τ = (5, 3, 1, 0) and λ = (3, 1, 0), then the notations Qσ, Q8,τ and Q8,5,λ all mean the same thing: Q8,5,3,1,0. 5.2. The schemes Pσ and their properties. We now introduce a set of spaces Pσ together with a set of coherent sheaves Xσ on P1×Pσ, indexed by σ ∈ Σ. We will need a set of auxiliary spaces Rσ indexed by σ ∈ Σ with σ ≥ 2. First, we define Pσ for σ ∈ Σ1. Suppose σ = (d). In this case, we set Pσ = Pd := Qd, Pσ = Pd := Qd, Xσ = Xd := Fd / / / / /   / / / / 16 YI HU AND YIJUN SHAO We will denote by Xd the restriction of Xd on P1 × Pd. We know that Xd is locally free. Next we define Pσ for σ ∈ Σ2. Suppose σ = (d, r), d > r. We set Rd,r := Qd,r ×Qr Pr, Rd,r := Qd,r ×Qr Pr ⊂ Rd,r Since Pr = Qr and Pr = Qr, we actually have Rd,r = Qd,r and Rd,r = Qd,r. Let qd,r and pd,r be the projections from Rd,r to Qd,r and to Pr, respectively, and let qd,r and pd,r be the projections from Rd,r to Qd,r and to Pr, respectively. We define Pd,r to be a (relative) space of non-split extensions: Pd,r = P(π∗ Ext 1(¯q∗ d,rTd,r, ¯p∗ d,r Xr)) and let ad,r : Pd,r → Rd,r be the structure morphism. We denote the universal extension on P1 × Pd,r by 0 → (¯a∗ d,r ¯p∗ d,r Xr)(0, 1) → Xd,r → ¯a∗ d,r ¯q∗ d,rTd,r → 0 Let Pd,r ⊂ Pd,r be the open subset defined as Pd,r := {x ∈ Pd,r (Xd,r)x is locally free} We denote by Xd,r the restriction of Xd,r to P1 × Pd,r. Then Xd,r is locally free, and Pd,r can be considered as a Qd-scheme through the composition Pd,r → Rd,r → Qd,r → Qd. So we have defined Pσ, Xσ, Rσ, etc., for any σ ∈ Σ with σ = 2. In the following, we will define Pσ, Xσ, etc., for any σ ∈ Σ with σ ≥ 3 inductively. Assume that, for each σ ∈ Σm for some m ≥ 2, the space Pσ of non-split extensions is defined and the sheaf Xσ is the middle term from the universal extension on P1 ×Pσ . Assume also that a morphism Pσ → Ql (l = lt(σ)) has been specified so that Pσ can be considered as a Ql-scheme. Let σ ∈ Σm+1, d = lt(σ), τ be the sequence formed from σ by removing the leading term d, and r = lt(τ ). So σ = (d, τ ) = (d, r, · · · ). By induction hypothesis, the space Pτ of non-split extensions is defined and is a Qr-scheme. We set Rσ := Qd,r ×Qr Pτ , Rσ := Qd,r ×Qr Pτ ⊂ Rσ Let qσ and pσ be the two projections from Rσ to Qd,r and to Pτ , respectively, and let qσ and pσ be the two projections from Rσ to Qd,r and to Pτ , respectively. We define Pσ = Pd,τ to be a space of non-split extensions over Rσ by Pσ := P(π∗ Ext 1(¯q∗ σTd,r, ¯p∗ σ Xτ )) THE SPACE OF COMPLETE QUOTIENTS 17 and let aσ : Pσ → Rσ be the structure morphism. We denote the universal extension on P1 × Pσ by 0 → (¯a∗ σ ¯p∗ σ Xτ )(0, 1) → Xσ → ¯a∗ σ ¯q∗ σTd,r → 0 Pσ can be considered as a Qd-scheme through Pσ → Rσ → Qd,r → Qd. We define the open subset Pσ ⊂ Pσ as Pσ := {x ∈ Pσ (Xσ)x is locally free} Then, Xσ, the restriction of Xσ on P1 × Pσ is locally free. By induction, we have defined Pσ, Xσ, Rσ, etc., for all σ ∈ Σ. Lemma 5.2. For each σ, (1) Xσ is flat over Pσ with relative degree lt(σ). (2) R1π∗(Xσ(−1)) = 0. The closed subset Pσ \ Pσ of Pσ has a sequence of nested closed subschemes ∅ = Yd,r,τ ⊂ Yd,r+1,τ ⊂ · · · ⊂ Yd,d−1,τ = Pσ \ Pσ where Yd,e,τ = {x ∈ Pσ deg((Xσ)t x) ≥ d − e} and the subscheme structure on Yd,e,τ is defined by (3.3). We set Yd,e,τ := Yd,e,τ \ Yd,e−1,τ for e = r + 1, · · · , d − 1. That is, Yd,e,τ = {x ∈ Pσ deg((Xσ)t x) = d − e}. Lemma 5.3. The space Pσ is nonsingular for any σ ∈ Σ. Proof. We prove it by induction on σ. When σ = 1, say σ = (d) for some d, we have Pσ = Pd = Qd. So this case is obvious since Qd is nonsingular. Assume that the statement holds true for all σ ∈ Σm for some m. Let σ ∈ Σm+1 and suppose σ = (d, τ ) where τ ∈ Σm. Pτ where r = lt(τ ). Hence Pσ is By definition, Pσ = Pd,τ is a projective bundle over Qd,r ×Qr Pτ is smooth smooth over Qd,r ×Qr over Pτ . Since smoothness is transitive, Pσ is smooth over Pτ . By induction hypothesis, Pτ is nonsingular, and so is Pτ and hence Pσ is nonsingular. This completes the proof. (cid:3) Pτ . We know that Qd,r is smooth over Qr, hence Qd,r ×Qr 18 YI HU AND YIJUN SHAO Let σ ∈ Σ, l = lt(σ) and let d > r > l. We define a morphism φd,r,σ : Rd,r,σ → Pd,σ using the following commutative diagram (5.5) Pr,σ ar,σ pd,r,σ Qd,r ×Qr Pr,σ = Rd,r,σ φd,r,σ 1×ar,σ Pd,σ ad,σ Qr,l ×Ql Pσ ψd,r,l×1 Qd,r ×Qr Qr,l ×Ql Pσ ψd,r,l×1 Qd,l ×Ql Pσ qr,σ Qr,l ψd,r,l 1×qr,σ Qd,r ×Qr Qr,l ψd,r,l qd,σ / Qd,l By the base change property, Rd,r,σ is a projective bundle over Qd,r ×Qr Qr,l ×Ql Pσ Rr,σ = Qd,r ×Qr Rd,r,σ = (Qd,r ×Qr Rr,σ) ×Rr,σ Pr,σ = P((ψd,r,l × 1)∗π∗ Ext 1(¯q∗ r,σTr,l, ¯p∗ r,σ Xσ)) with ORd,r,σ (1) = p∗ commutative diagram of sheaves d,r,σOPr,σ (1) and structure morphism 1 × ar,σ. On P1 × Rd,r,σ, we have a (5.6) 0 0 ¯q∗ d,r,σTd,r 'ar,σ ∗'qr,σ ∗ ¯ψ∗ d,r,lTd,r / ('ar,σ ∗ψ′ d,r,l ∗ ¯p∗ d,σ Xσ)(0, 1) / X 'ar,σ ∗ψ′ d,r,l ∗ ¯q∗ d,σTd,l / ¯p∗ d,r,σ((¯a∗ r,σ ¯p∗ r,σ Xσ)(0, 1)) / ¯p∗ d,r,σXr,σ d,r,σ¯a∗ ¯p∗ r,σ ¯q∗ r,σTr,l 0 0 0 0 0 0 where • for short notations, we have set 'ar,σ := 1 × ar,σ, 'qr,σ := 1 × qr,σ, ψ′ d,r,l := ψd,r,l × 1, • the last row is the pullback of the universal extension of Pr,σ via ¯pd,r,σ : P1 × Rd,r,σ → P1 × Pr,σ, • the last column is the pullback of the exact sequence of torsion sheaves via 'ar,σ 'qr,σ : P1 × Rd,r,σ → P1 × (Qd,r ×Qr Qr,l), and d,r,σXr,σ) ×(¯p∗ • X is the fiber product (¯p∗ d,r,σ¯a∗ r,σ gory of coherent sheaves. r,σTr,l) ('ar,σ ¯q∗ ∗ψ′ d,r,l ∗ ¯q∗ r,σTd,l) in the cate-   o o / /       o o / /     o o /         / / / /   / /   / / / /   / /   THE SPACE OF COMPLETE QUOTIENTS 19 Recall that the universal extension of Pd,σ is the exact sequence (5.7) 0 → (¯a∗ d,σ ¯p∗ d,σ Xσ)(0, 1) → Xd,σ → ¯a∗ d,σ ¯q∗ d,σTd,l → 0 on P1 × Pd,σ. By Theorem 2.2, the middle row determines a (Qd,l ×Ql Pσ)-morphism φd,r,σ : Rd,r,σ → Pd,σ such that there are isomorphisms ORd,r,σ (1) ≃ φ∗ the following diagram commute: d,r,σOPd,σ (1) and X ≃ ¯φ∗ d,r,σXd,σ that make 0 0 / ('ar,σ ∗ψ′ d,r,l ∗ ¯p∗ d,σ Xσ)(0, 1) ≃ X ≃ 'ar,σ ∗ψ′ d,r,l ∗ ¯q∗ d,σTd,l / ¯φ∗ d,r,σ((¯a∗ d,σ ¯p∗ d,σ Xσ)(0, 1)) / ¯φ∗ d,r,σXd,σ / ¯φ∗ d,r,σ¯a∗ d,σ ¯q∗ d,σTd,l / 0 / 0 where the first row is the middle row of diagram (5.6) and the second row is the pullback of the universal extension (5.7) of Pd,σ via ¯φd,r,σ : P1 × Rd,r,σ → P1 × Pd,σ. For simplicity, we make identifications (5.8) ORd,r,σ (1) = φ∗ d,r,σOPd,σ (1), X = ¯φ∗ d,r,σXd,σ. Let φd,r,σ : Rd,r,σ → Pd,σ be the restriction of φd,r,σ to Rd,r,σ. The restriction of the middle column of diagram (5.6) to P1 × Rd,r,σ is the exact sequence (5.9) 0 → ¯q∗ d,r,σTd,r → ¯φ∗ d,r,σXd,σ → ¯p∗ d,r,σ Xr,σ → 0 Proposition 5.4. The map φd,r,σ factors through the inclusion Yd,r,σ ֒→ Pd,σ. Proof. Taking the dual of the sequence (5.9), we obtain ( ¯φ∗ d,r,σXd,σ)ε = ( d − r. By Proposition 3.4, the map φd,r,σ factors through the inclusion Yd,r,σ ֒→ Pd,σ. d,r,σTd,r)ε. Since d,σ) is flat over Rd,r,σ with relative degree d,σ), we have that (cid:3) ¯φ∗ d,r,σ(X ε ¯φ∗ d,r,σ(X ε ¯φ∗ d,r,σXd,σ)ε ≃ (¯q∗ We denote by ϕd,r,σ : Rd,r,σ → Yd,r,σ the map factored out from φd,r,σ. Proposition 5.5. The morphism ϕd,r,σ : Rd,r,σ → Yd,r,σ is an isomorphism. Proof. We prove by constructing an inverse of ϕd,r,σ. Let i : Yd,r,σ ֒→ Pd,σ be the inclusion map. The pullback of the universal extension (5.7) via i is the exact sequence 0 → ¯i∗((¯a∗ d,σ ¯p∗ d,σ Xσ)(0, 1)) → ¯i∗Xd,σ → ¯i∗¯a∗ d,σ ¯q∗ d,σTd,l → 0 on P1 × Yd,r,σ. Taking dual, we obtain a long exact sequence 0 → (¯i∗Xd,σ)∨ → (¯i∗((¯a∗ d,σ ¯p∗ d,σ Xσ)(0, 1)))∨ → (¯i∗¯a∗ d,σ ¯q∗ d,σTd,l)ε → (¯i∗Xd,σ)ε → 0 / / /   / /   / / / / / 20 YI HU AND YIJUN SHAO We break it into two short exact sequences 0 → (¯i∗Xd,σ)∨ → (¯i∗((¯a∗ d,σ ¯p∗ d,σ Xσ)(0, 1)))∨ → T → 0 0 → T → (¯i∗¯a∗ d,σ ¯q∗ d,σTd,l)ε → (¯i∗Xd,σ)ε → 0 d,σ d,σ) and (¯i∗¯a∗ ¯q∗ d,σTd,l)ε = ¯i∗¯a∗ d,σ ¯q∗ We have (¯i∗Xd,σ)ε = ¯i∗(X ε d,σ(T ε d,l). By Proposition 3.4, d,σ) is flat over Yd,r,σ with relative degree d − r. Since T ε ¯i∗(X ε d,l is also flat over Qd,l with relative degree d − l, we have T is flat over Yd,r,σ with relative degree r − l. We also know ¯q∗ d,σTd,l)ε and (¯i∗Xd,σ)ε are torsion, hence T is also torsion. Since the middle that both (¯i∗¯a∗ term of the first sequence is locally free and the last term is flat over Yd,r,σ, the first term, (¯i∗Xd,σ)∨, is locally free as well. d,σ Now dualizing both of the above sequences, we obtain another two exact sequences (5.10) (5.11) 0 → ¯i∗((¯a∗ d,σ ¯p∗ d,σ Xσ)(0, 1)) → (¯i∗Xd,σ)∨∨ → T ε → 0 0 → (¯i∗Xd,σ)εε → ¯i∗¯a∗ d,σ ¯q∗ d,σTd,l → T ε → 0 The Quot scheme Qd,l has a universal quotient ¯θ∗ quotient, we obtain a quotient ¯i∗¯a∗ quotient ¯i∗¯a∗ ¯q∗ d,σ ¯θ∗ d,lEl ։ ¯i∗¯a∗ ¯q∗ d,σTd,l ։ T ε from the sequence (5.11) yields a quotient ¯q∗ d,σTd,l. d,σ d,σ d,lEl ։ Td,l. Applying ¯i∗¯a∗ d,σ ¯q∗ d,σ to this Its composition with the d,σ We form a commutative diagram / ¯i∗¯a∗ d,σ ¯q∗ d,σEd,l 0 / E 0 (¯i∗Xd,σ)εε / ¯i∗¯a∗ d,σ ¯q∗ d,σEd,l / ¯i∗¯a∗ d,σ ¯q∗ d,σ ¯θ∗ d,lEl ¯i∗¯a∗ d,σ ¯q∗ d,σTd,l 0 0 0 0 T ε 0 T ε 0 where the last column is the sequence (5.11), the middle row is the pullback of the universal exact sequence of Qd,l via ¯qd,σ¯ad,σ¯i : P1 × Yd,r,σ → P1 × Qd,l, and E is defined to be the ¯q∗ kernel of the composition ¯i∗¯a∗ d,σTd,l ։ T ε. The dotted arrows in the first row are the induced maps. Since T ε is torsion and flat over Yd,r,σ with relative degree ¯θ∗ r − l, the quotient ¯i∗¯a∗ d,lEl ։ T ε from the middle column induces a Ql-morphism ¯θ∗ d,lEl ։ ¯i∗¯a∗ d,σ ¯q∗ d,σ ¯q∗ d,σ d,σ d,σ     / / / /   / /   / / / /   / /       THE SPACE OF COMPLETE QUOTIENTS 21 λ : Yd,r,σ → Qr,l such that the pullback of the universal exact sequence of Qr,l via ¯λ is the same as the middle column: (5.12) 0 0 / ¯λ∗Er,l / ¯λ∗ ¯θ∗ r,lEl / ¯λ∗Tr,l / E / ¯i∗¯a∗ d,σ ¯q∗ d,σ ¯θ∗ d,lEl / T ε / 0 / 0 We have a morphism φr,l : Qr,l → Qr and an identification Er,l = ¯φ∗ Thus we have an identification r,lEr as in diagram (4.3). and we can rewrite the first row as E = ¯λ∗Er,l = ¯λ∗ ¯φ∗ r,lEr 0 → ¯i∗¯a∗ d,σ ¯q∗ d,σEd,l → ¯λ∗ ¯φ∗ r,lEr → (¯i∗Xd,σ)εε → 0 Since (¯i∗Xd,σ)εε is torsion and flat over Yd,r,σ with relative degree d − r, the above sequence induces a morphism µ : Yd,r,σ → Qd,r such that the pullback of the universal exact sequence of Qd,r via ¯µ is the same as the above sequence: 0 0 / ¯µ∗Ed,r / ¯µ∗ ¯θ∗ d,rEr / ¯µ∗Td,r / ¯i∗¯a∗ d,σ ¯q∗ d,σEd,l / ¯λ∗ ¯φ∗ r,lEr / (¯i∗Xd,σ)εε / 0 / 0 Using the identification T ε = ¯λ∗Tr,l from sequence (5.12), we can rewrite the sequence (5.10) as 0 → ¯i∗((¯a∗ d,σ ¯p∗ d,σ Xσ)(0, 1)) → (¯i∗Xd,σ)∨∨ → ¯λ∗Tr,l → 0 By Theorem 2.2, the above exact sequence determines a morphism ν : Yd,r,σ → Pr,σ such that the pullback of the universal extension of Pr,σ is the same as the above sequence: 0 0 / ¯ν∗((¯a∗ r,σ ¯p∗ r,σ Xσ)(0, 1)) / ¯ν∗Xr,σ / ¯ν∗¯a∗ r,σ ¯q∗ r,σTr,l / ¯i∗((¯a∗ d,σ ¯p∗ d,σ Xσ)(0, 1)) / (¯i∗Xd,σ)∨∨ / ¯λ∗Tr,l / 0 / 0 But because the middle term (¯i∗Xd,σ)∨∨ is locally free, the map ν actually maps Yd,r,σ into Pr,σ. So we obtain a morphism µ × ν : Yd,r,σ → Qd,r ×Qr Pr,σ := Rd,r,σ It is now routine to check that µ× ν is the inverse of ϕd,r,σ : Rd,r,σ → Yd,r,σ, and this complete the proof. (cid:3) / / / / / / / / / / / / / / / / / / / / / / / / 22 YI HU AND YIJUN SHAO The above proposition shows that φd,r,σ is an embedding. So we can identify Rd,r,σ with the ¯φ∗ d,r,σXd,σ = subscheme Yd,r,σ of Pd,σ and identify φd,r,σ with the inclusion map. We also have XRd,r,σ . So the exact sequence (5.9) can be rewritten as: 0 → ¯q∗ d,r,σTd,r → ¯φ∗ d,r,σXd,σ → ¯p∗ d,r,σ Xr,σ → 0 Since Rd,r,σ and Pd,σ are both nonsingular, we can talk about their tangent bundles as well as the normal bundle of the embedding φd,r,σ : Rd,r,σ ֒→ Pd,σ. Proposition 5.6. The normal bundle NRd,r,σ/Pd,σ isomorphic to π∗ Ext 1(¯q∗ d,r,σTd,r, ¯p∗ Xr,σ). d,r,σ of the embedding φd,r,σ : Rd,r,σ ֒→ Pd,σ is Proof. Since Pd,σ is a projective bundle over Rd,σ, the relative cotangent bundle ΩPd,σ/Rd,σ fits into the following exact sequence 0 → ΩPd,σ/Rd,σ → (a∗ d,σπ∗ Ext 1(¯q∗ d,σTd,l, ¯p∗ d,σ Xσ))∨(−1) → OPd,σ → 0 Pulling the sequence back to Rd,r,σ via φd,r,σ, we obtain an exact sequence (5.13) 0 → φ∗ d,r,σΩPd,σ/Rd,σ → (φ∗ d,r,σa∗ d,σπ∗ Ext 1(¯q∗ d,σTd,l, ¯p∗ d,σ Xσ))∨(−1) → φ∗ d,r,σOPd,σ → 0 We can rewrite φ∗ d,r,σa∗ d,σπ∗ Ext 1(¯q∗ d,σTd,l, ¯p∗ d,σ Xσ) as φ∗ d,r,σa∗ d,σTd,l, ¯p∗ d,σπ∗ Ext 1(¯q∗ d,σ ¯q∗ d,σTd,l, ¯φ∗ d,r,σ¯a∗ d,σ d,r,σ¯a∗ d,σ Xσ) = π∗ Ext 1( ¯φ∗ ¯p∗ d,σ Xσ) = π∗ Ext 1( ¯f ∗ ¯ψ∗ d,r,lTd,l, ¯g∗ Xσ) where f := (1 × qr,σ)(1 × ar,σ), g := pd,σad,σφd,r,σ. Set QQP := Qd,r ×Qr Qr,l ×Ql cotangent bundle ΩRd,r,σ/QQP fits into the following exact sequence (5.14) Pσ. Since Rd,r,σ is a projective bundle over QQP , the relative 0 → ΩRd,r,σ/QQP → ((1 × ar,σ)∗(ψd,r,l × 1)∗π∗ Ext 1(¯q∗ r,σTr,l, ¯p∗ r,σ Xσ))∨(−1) → ORd,r,σ → 0 We have (1 × ar,σ)∗(ψd,r,l × 1)∗π∗ Ext 1(¯q∗ r,σTr,l, ¯p∗ r,σ Xσ) = π∗ Ext 1( ¯f ∗ ¯ψ∗ d,r,lTr,l, ¯g∗ Xσ) based on the equalities qr,σ(ψd,r,l × 1)(1 × ar,σ) = ψd,r,lf and pr,σ(ψd,r,l × 1)(1 × ar,σ) = g. THE SPACE OF COMPLETE QUOTIENTS 23 We make a diagram (5.15) 0 K : 0 0 / φ∗ d,r,σΩPd,σ/Rd,σ / ΩRd,r,σ/QQP 0 (π∗ Ext 1( ¯f ∗ ¯ψ∗ d,r,lTd,r, ¯g∗ Xσ))∨(−1) (π∗ Ext 1( ¯f ∗ ¯ψ∗ (π∗ Ext 1( ¯f ∗ ¯ψ∗ d,r,lTd,l, ¯g∗ Xσ))∨(−1) 1(cid:13) d,r,lTr,l, ¯g∗ Xσ))∨(−1) φ∗ d,r,σOPd,σ ORd,r,σ / 0 / 0 0 0 where • the last row is the exact sequence (5.13), • the middle row is the exact sequence (5.14), and • the middle column is obtained by applying π∗ Ext 1( ¯f ∗(−), ¯g∗ Xσ) to the exact sequence (5.4), then taking dual, and lastly twisting by ORd,r,σ (−1). The commutativity of the rectangle 1(cid:13) in the diagram (5.15) follows from diagram (5.6) and the identifications (5.8). Thus, we have induced maps (the dotted arrows) in the first column and the first column is exact. Restricting the first column to Rd,r,σ, we obtain an exact sequence 0 → K → φ∗ d,r,σΩPd,σ/Rd,σ → ΩRd,r,σ/QQP → 0 where K := KRd,r,σ . We have K = (π∗ Ext 1( ¯f ∗ ¯ψ∗ d,r,lTd,r, ¯g∗ Xσ))∨(−1) where f and g are the restrictions of f and g to Rd,r,σ. Recall that ψd,r,l : Qd,r,l → Qd,l is a finite morphism. Since k is assumed to be of characteristic 0, the function field extension K(Qd,r,l)/K(Qd,l) is separable. Therefore, we have an exact sequence of relative cotangent sheaves 0 → ψ∗ d,r,lΩQd,l → ΩQd,r,l → ΩQd,r,l/Qd,l → 0 where both ΩQd,l and ΩQd,r,l are locally free of the same rank while ΩQd,r,l/Qd,l is torsion.         / / /   / /   / / / /   / /   / 24 YI HU AND YIJUN SHAO We form the following diagram 0 0 0 (π∗ Hom( ¯ψ∗ d,r,lEd,r, ¯ψ∗ d,r,lTr,l))∨ ◗◗◗◗◗◗◗ ◗◗◗◗◗◗◗ / ψ∗ ψ∗ d,r,lθ∗ r,lΩQl 0 0 d,r,lθ∗ d,lΩQl ψ∗ d,r,lΩQd,l ψ∗ d,r,lΩQd,l/Ql / ψ∗ d,r,lΩQr,l ΩQd,r,l ΩQd,r,l/Qr,l ψ∗ d,r,lΩQr,l/Ql ΩQd,r,l/Qd,l 0 0 0 0 0 Applying the Snake Lemma to the third row and the fourth row of the above diagram, we obtain an exact sequence by connecting the second row with the last row: 0 → (π∗ Hom( ¯ψ∗ d,r,lEd,r, ¯ψ∗ d,r,lTr,l))∨ → ψ∗ d,r,lΩQr,l/Ql → ΩQd,r,l/Qd,l → 0 This sequence fits into the following commutative diagram 0 0 (π∗ Hom( ¯ψ∗ d,r,lEd,r, ¯ψ∗ d,r,lTr,l))∨ ψ∗ d,r,lΩQr,l/Ql ΩQd,r,l/Qd,l 0 (π∗ Hom( ¯ψ∗ d,r,lEd,r, ¯ψ∗ d,r,lTr,l))∨ (π∗ Hom( ¯ψ∗ d,r,lEr,l, ¯ψ∗ d,r,lTr,l))∨ (π∗ Ext 1( ¯ψ∗ d,r,lTd,r, ¯ψ∗ d,r,lTr,l))ε 0 where the identification in the second row follows from the canonical identification TQr,l/Ql = π∗ Hom( ¯ψ∗ d,r,lTr,l). So we obtain an identification of the quotients d,r,lEd,r, ¯ψ∗ ΩQd,r,l/Qd,l = (π∗ Ext 1( ¯ψ∗ d,r,lTd,r, ¯ψ∗ d,r,lTr,l))ε Since Qd,r,l is smooth over Ql, QQP = Qd,r,l ×Ql Pσ is smooth over Pσ. The morphism ψd,r,l × 1 : QQP → Rd,σ is obtained from ψd,r,l by the base change qd,σ : Rd,σ → Qd,l, hence     / /     / / /   / /   / /   / / /   / /   / /   / /   / /                   THE SPACE OF COMPLETE QUOTIENTS 25 we have the following commutative diagram 0 0 / (ψd,r,l × 1)∗Ω Pd,σ / ΩQQP / ΩQQP/ Pd,σ / (1 × qd,σ)∗ψ∗ d,r,lΩQd,l / (1 × qd,σ)∗ΩQd,r,l / (1 × qd,σ)∗ΩQd,r,l/Qd,l / 0 / 0 The conormal bundle N ∨ Rd,r,σ/Pd,σ fits into the following commutative diagram 0 0 N ∨ Rd,r,σ/Pd,σ 0 K 0 0 / φ∗ d,r,σa∗ d,σΩRd,σ φ∗ d,r,σΩPd,σ φ∗ d,r,σΩPd,σ/Rd,σ / (1 × ar,σ)∗ΩQQP ΩRd,r,σ ΩRd,r,σ/QQP f ∗ΩQd,r,l/Qd,l 0 0 0 0 0 Applying the Snake Lemma to the third and fourth rows, we obtain an exact sequence by connecting the second row with the fifth row: 0 → N ∨ Rd,r,σ/Pd,σ → K → f ∗ΩQd,r,l/Qd,l → 0 The above sequence fits into the following commutative diagram 0 0 N ∨ Rd,r,σ/Pd,σ K f ∗ΩQd,r,l/Qd,l (π∗ Ext 1( ¯f ∗ ¯ψ∗ d,r,lTd,r, ¯p∗ d,r,σ Xr,σ))∨ (π∗ Ext 1( ¯f ∗ ¯ψ∗ d,r,lTd,r, ¯g∗ Xσ))∨(−1) (π∗ Ext 1( ¯f ∗ ¯ψ∗ d,r,lTd,r, ¯f ∗ ¯ψ∗ d,r,lTr,l))ε 0 0 / / / / / / / /     / /   / /     / / /   / /   / /   / / /   / /   / /   / /                   26 YI HU AND YIJUN SHAO where the second column is obtained as follows: first we can rewrite the last row of diagram (5.6) as 0 → (¯g∗ Xσ)(0, 1) → ¯p∗ d,r,σXr,σ → ¯f ∗ ¯ψd,r,lTr,l → 0; next the restriction to P1 × Rd,r,σ is 0 → (¯g∗ Xσ)(0, 1) → ¯p∗ d,r,σ Xr,σ → ¯f ∗ ¯ψd,r,lTr,l → 0; and next applying π∗ Hom( d,r,lTd,r, −) to the above sequence, and lastly taking dual, we obtain the second column in the diagram. Thus there is an induced identification as in the ¯f ∗ ¯ψ∗ first row, which gives the natural identification NRd,r,σ/Pd,σ = π∗ Ext 1( ¯f ∗ ¯ψ∗ d,r,lTd,r, ¯p∗ d,r,σ Xr,σ) = π∗ Ext 1(¯q∗ d,r,σTd,r, ¯p∗ d,r,σ Xr,σ) (cid:3) 6. The modular interpretation The compactification eQd is obtained by successively blowing up the Quot scheme Qd along Zd,0, · · · , Zd,d−1. We illustrate the process in the following diagram Z d−1 d,0 ↓ ... ↓ Z 1 d,0 ↓ Z 0 d,0 ↓ d = eQd ... · · · d,d−1 ⊂ Qd−1 Z d−1 Z d−1 Z d−1 d,2 d,1 ↓ ↓ ↓ ↓ ... ... ... ... ↓ ↓ ↓ ↓ Q1 d,2 ⊂ · · · ⊂ Z 1 Z 1 Z 1 d,d−1 ⊂ d d,1 ↓ ↓ ↓ ↓ Q0 d,2 ⊂ · · · ⊂ Z 0 d,1 ⊂ Z 0 Z 0 d,d−1 ⊂ d ↓ ↓ ↓ ↓ ← along Z d−2 d,d−1 ← along Z 1 d,2 ← along Z 0 d,1 ← along Zd,0 Zd,0 ⊂ Zd,1 ⊂ Zd,2 ⊂ · · · ⊂ Zd,d−1 ⊂ Qd inductively, Z j d,0, · · · , Z j d → Qd (0 ≤ j ≤ d − 1); the nested subschemes Z j Here, blowups Qj proper transforms of the subschemes Z j−1 d,j are the exceptional divisors created by the sequence of d,d−1 are the d,d−1 (respectively); these are the sub- schemes lining up to be blown-up in the next steps. Below, we will provide modular meanings to the points of Z j d, j = 0, . . . , d − 1, also compactification of Qd, admits parameter space interpretation (Proposition. 6.4). The case Qd−1 d,j for all j. Thus every intermediate space Qj d,j+1 ⊂ · · · ⊂ Z j−1 d,j+1 ⊂ · · · ⊂ Z j d,0, · · · , Z j d is our final space eQd (Corollary 6.5). THE SPACE OF COMPLETE QUOTIENTS 27 To proceed, we introduce the following d,r :=( ∅, Z l Z l d,r, if r = l if l < r < d , Z l d,r := Z l d,r \ Z l d,r−1, r = l + 1, . . . , d − 1 Let d > r ≥ 0. For each subsequence σ ⊂ [r] := (r, r − 1, · · · , 1, 0) ∈ Σ, we set Er d,σ :=\i∈σ Z r d,i \ [i∈[r]\σ Z r d,i. Then there is a stratification of Qr d: Qr d = (Qr d \ [i∈[r] Z r d,i) ⊔ Gσ⊂[r] Er d,σ = (Qd \ Zd,r) ⊔ Gσ⊂[r] Er d,σ. Lemma 6.1. Let τ ∈ Σ, t = lt(τ ). We have d,τ = Em (1) Ej (2) if l > t and σ = (l, τ ), then El d,τ \ Z m d,j, for any m and j with t ≤ m < j < d. d,σ is the exceptional divisor of the blowup of Et d,τ \ Z t d,l−1 along Et d,τ ∩ Z t d,l. Proof. (1) Consider the composite blowup b : Qj i = 0, . . . , m, and b−1(Z m d → Qm d,i for d,i. Moreover, b is an isomorphism away from Z m d,j. d , m < j. We have b−1(Z m d,i) = Z j i=m+1 Z j These facts give us identifications d,j) =Sj Qj d \ b−1(Z m d,j) = Qm d \ Z m d,j, and Z j d,i \ b−1(Z m d,j) = Z m d,i \ Z m d,j, for i = 0, . . . , m Therefore d,i \ [i∈[m]\τ Z j d,i(cid:19) \ j[i=m+1 Z j d,i Z j Z j d,i \ [i∈[j]\τ d,i \ [i∈[m]\τ d,i \ [i∈[m]\τ Z m Z j Z j d,i =(cid:18)\i∈τ d,i(cid:19) \ b−1(Z m d,i(cid:19) \ Z m Z m d,j) d,j = Em d,τ \ Z m d,j Ej Z j d,τ =\i∈τ =(cid:18)\i∈τ =(cid:18)\i∈τ d,σ =\i∈σ El (2) By definition, We know Z l d,l is the exceptional divisor of the blowup Ql (Ti∈τ Z l d,i \Si∈[l]\σ Z l d,i) ⊂ Ql Z l d,i Z l Z l d,i = Z l Z l d,l ∩\i∈τ d,i \ [i∈[l]\σ d,i \ [i∈[l]\σ d is exactly the preimage of Ti∈τ Z l−1 d → Ql−1 d , which is along Z l−1 d,i \Si∈[l]\σ Z l−1 d,l . Since d,i ⊂ Ql−1 d 28 YI HU AND YIJUN SHAO d,σ is the exceptional divisor of the (induced) blowup of under the blowup, we have that El d,i along Z l−1 d,i \Si∈[l]\σ Z l−1 Ti∈τ Z l−1 d,i \ [i∈[l]\σ \i∈τ Z l−1 Z l−1 d,i =\i∈τ d,l ∩Ti∈τ Z l−1 d,i \(cid:18) [i∈[t]\τ d,i \Si∈[l]\σ Z l−1 l−1[i=t+1 d,i ∪ Z l−1 Z l−1 d,i . Note that Z l−1 d,i (cid:19) = El−1 d,τ \ l−1[i=t+1 Z l−1 d,i d,σ is the exceptional divisor of the blowup of Et If l − 1 = t, then El done. Now suppose l − 1 > t. Then under the identification Ql−1 we have identifications El−1 d d,l−1 and i=t+1 Z l−1 d,i = Et d,τ \ Z t \Sl−1 d,τ along Z t i=t+1 Z l−1 d,t+1, and we are d,i = Qt d,l−1, d \ Z t d,τ \Sl−1 l−1[i=t+1 d,l \ Z l−1 Z l−1 d,i = Z t d,l \ Z t d,l−1 = Z t d,l. Proposition 6.2. There is a collection of isomorphisms id,σ : Pd,σ ∼→ El d,σ, one for each σ ∈ Σ with l := lt(σ) < d, such that the following properties hold: (cid:3) (1) id,σ maps Yd,r,σ onto El (2) id,σ maps Yd,r,σ isomorphically onto El (3) The following diagram commutes: d,σ ∩ Z l d,r for all r, l < r < d;; d,σ ∩ Z l d,r for all r, l < r < d; Pd,σ id,σ ∼ / El d,σ Rd,l,τ ∼ / Yd,l,τ ∼ id,τ / Et d,τ ∩ Z t d,l where σ = (l, τ ) and t = lt(τ ); (4) Let ed,σ : Pd,σ ֒→ Ql d be the embedding obtained through the composition Pd,σ ∼→ El d,σ ֒→ Ql d. Then the following diagram commutes Qd,r ×Qr Pr,σ  1×er,σ Qd,r ×Qr Ql r φd,r,σ Pd,σ ed,σ φl d,r / Ql d,σ for all r, l < r < d.   /   / /  / /       / THE SPACE OF COMPLETE QUOTIENTS 29 Proof. We prove by constructing the isomorphisms id,σ explicitly, and this is done by induc- tion on the length of σ. We first deal with the base case: σ ∈ Σ1 or σ = (l). In this case, we construct isomorphisms Pd,l ∼→ El d,l = Z l d,l \ [0≤i≤l−1 Z l d,i which map Yd,r,l onto El l such that l < r < d. d,l ∩ Z l d,r and map Yd,r,l isomorphically onto El d,l ∩ Z l d,r for all d, r and Recall that Z l d,l is the exceptional divisor in the blowup b : Ql d → Ql−1 d along Z l−1 d,l . We have b−1(Z l−1 d,i ) = Z l d,i for i = 0, . . . , l − 1, hence b : Ql d \ l−1[i=0 d,i → Ql−1 Z l d \ Z l−1 d,i l−1[i=0 is the blowup along Z l−1 that Ql−1 El \Sl−1 i=0 Z l−1 d d,l is the projective bundle P(NZd,l/Qd d,i with exceptional divisor Z l d,l. Note d,i = Zd,l \ Zd,l−1 = Zd,l. Therefore ) over Zd,l. Since Zd,l ≃ Qd,l and by Proposition 4.2 d,l \Sl−1 i=0 Z l−1 d,i = El i=0 Z l d,l \Sl−1 i=0 Z l−1 d,l \Sl−1 d,i = Qd \ Zd,l−1 and Z l−1 N Qd,l/Qd = π∗ Ext 1(Td,l, ¯θ∗ d,lFl) Qd,l = π∗ Ext 1(¯q∗ d,lTd,l, ¯p∗ d,l Xl), we know El d,l is isomorphic to the projective bundle Pd,l = P(π∗ Ext 1(¯q∗ d,lTd,l, ¯p∗ d,l Xl)) over Qd,l. So we obtain an embedding ed,l : Pd,l ֒→ Ql d through the composition Pd,l ∼→ El d,l ֒→ Ql d. For each r with l < r < d, we have a commtative diagram Qd,r ×Qr Pr,l φd,r,l Pd,l 1×er,l ed,l Qd,r ×Qr Ql r φl d,r / Ql d We have that Im(φd,r,l) = Yd,r,l, that Im(1 × er,l) = Qd,r ×Qr El d,r. It follows that ed,l maps Yd,r,l onto El Qd,r ×Qr El r,l onto El d,l ∩ Z l r,l, and that φl d,l ∩ Z l d,r. d,r maps   / /       / 30 YI HU AND YIJUN SHAO Next, we show that ed,l : Pd,l → Ql d maps Yd,r,l isomorphically onto El d,l ∩ Z l d,r for l < r < d. We have a commutative diagram Qd,r ×Qr Pr,l 1×er,l Qd,r ×Qr (Ql r \ Z l r,r−1) φd,r,l φl d,r Pd,l \ Yd,r−1,l ed,l / Ql d \ Z l d,r−1 Qd,r ×Qr (Qr−1 r \ r−1[i=l+1 Z r−1 r,i ) φr−1 d,r r−1[i=l+1 Z r−1 d,i Qr−1 d \ We see that er,l maps Pr,l isomorphically onto El be identified with Er−1 isomorphically onto Er−1 r,r−1, which can d,r maps Qd,r ×Qr Er−1 d,l ∩ Z r−1 d,r. On the other hand, Pr,l isomorphically onto Yd,r,l ⊂ Pd,l \ Yd,r−1,l. It follows that ed,l maps r,r−1, El r,l\Z l (by Lemma 6.1), and φr−1 d,r , which is identified with El \Sr−1 r,l ⊂ Qr−1 i=l+1 Z r−1 r,i r,r−1 in Ql d,l ∩ Z l r,l\Z l r\Z l r,l r φd,r,l maps Qd,r ×Qr Yd,r,l isomorphically onto El d,l ∩ Z l d,r. Thus the case that σ ∈ Σ1 is constructed. Suppose we have constructed the isomorphisms for all σ ∈ Σm for some m. We now induction hypothesis, for any d > t, we have an isomorphism Pd,τ d,τ ∩ Z t onto Et 6.1, El construct the isomorphisms for σ ∈ Σm+1. Now write σ = (l, τ ) and let t = lt(τ ). By d,τ which maps Yd,r,τ d,r. Let d > l. By Lemma d,l. Since the d,l, we have an isomorphism d,τ ∩ Z t d,σ is the exceptional divisor of the blowup of Et d,τ \ Z t d,τ maps Yd,l,τ isomorphically onto Et d,r and maps Yd,r,τ isomorphically onto Et d,l−1 along Et d,τ ∩ Z t isomorphism Pd,τ d,τ ∩ Z t ∼→ Et ∼→ Et BlYd,l,τ (Pd,τ \ Yd,l−1,τ ) ∼→ BlEt d,τ ∩Z t d,l (Et d,τ \ Z t d,l−1) d,l (Et d,τ ∩Z t d,τ \ Z t d,σ of BlEt d,σ of the blowup BlYd,l,τ which maps the exceptional divisor El onto the exceptional divisor El (Pd,τ \ Yd,l−1,τ ) isomorphically d,l−1). On the other hand, the excep- (Pd,τ \ Yd,l−1,τ ) is isomorphic to the projective normal bundle of Yd,l,τ Pl,τ Pl,τ in Pd,τ Xl,τ ). Hence the projective normal bundle of Yd,l,τ in Pd,τ is iso- d,l,τ Td,l, ¯p∗ Xl,τ )) = Pd,l,τ = Pd,σ. Thus we obtain an isomorphism d,r and maps tional divisor of BlYd,l,τ in Pd,τ \ Yd,l−1,τ or just in Pd,τ . We know that φd,l,τ : Qd,l ×Ql Pl,τ → Pd,τ maps Qd,l ×Ql isomorphically onto Yd,l,τ , and by Proposition 5.6 the normal bundle of Qd,l ×Ql is π∗ Ext 1(¯q∗ d,l,τ morphic to P(π∗ Ext 1(¯q∗ Pd,σ Yd,r,σ isomorphically onto Ed,σ ∩ Z l d,σ. Next we show that this isomorphism maps Yd,r,σ onto El d,l,τ Td,l, ¯p∗ d,σ ∩ Z l ∼→ El d,l,τ d,r.   / /  _      _     / THE SPACE OF COMPLETE QUOTIENTS Let ed,σ : Pd,σ ֒→ Ql d,σ for each d > l. Let d > r > l. Then we have a commutative diagram d denote the embedding obtained from the composition Pd,σ Ql 31 ∼→ El d,σ ֒→ Qd,r ×Qr Pr,σ  1×er,σ Qd,r ×Qr Ql r φd,r,σ Pd,σ ed,σ φl d,r / Ql d,σ We have that Im(φd,r,σ) = Yd,r,σ, that Im(1 × er,σ) = Qd,r ×Qr El Qd,r ×Qr El d,r. It follows that ed,σ maps Yd,r,σ onto El r,σ onto El d,σ ∩ Z l r,σ, and that φl d,r maps d,σ ∩ Z l d maps Yd,r,σ isomorphically onto El d,r. d,σ ∩ Z l d,r for any Next, we show that ed,σ : Pd,σ ֒→ Ql r, l < r < d. We have a commutative diagram Qd,r ×Qr Pr,σ 1×er,σ Qd,r ×Qr (Ql r \ Z l r,r−1) φd,r,σ φl d,r Pd,σ \ Yd,r−1,σ ed,σ / Ql d \ Z l d,r−1 Qd,r ×Qr (Qr−1 r \ r−1[i=l+1 Z r−1 r,i ) φr−1 d,r r−1[i=l+1 Z r−1 d,i Qr−1 d \ r,σ \ Z l r,σ ⊂ Qr−1 We see that er,σ maps Pr,σ isomorphically onto El can be identified with Er−1 Qd,r ×Qr Er−1 r,l the other hand, φd,r,σ maps Qd,r ×Qr follows that ed,σ maps Yd,r,σ isomorphically onto El constructed. r,σ \ Z l r,r−1 d,r maps d,l ∩ Z r−1 d,r. On Pr,σ in isomorphically onto Yd,r,σ ⊂ Pd,σ \ Yd,r−1,σ. It d,r. Thus the case that σ ∈ Σm+1 is (cid:3) (by Lemma 6.1), and that φr−1 d,l ∩ Z l d,r , which is identified with El isomorphically onto Er−1 \Sr−1 r,r−1, that El i=l+1 Z r−1 r,i r,r−1 in Ql d,σ ∩ Z l r \ Z l r Proposition 6.3. Let σ ∈ Σ. Then for any d and l, d > l ≥ lt(σ), there is an isomorphism Pd,σ \ Yd,l,σ ∼→ El d,σ which maps Yd,r,σ isomorphically onto El d,σ ∩ Z l d,r for any r, l < r < d; Proof. The case that l = lt(σ) is proved in the above proposition. We now prove the case that l > lt(σ). Let t = lt(σ). Then by the above proposition, we have an isomorphism Pd,σ which maps Yd,l,σ onto Et Pd,σ \ Yd,l,σ is identified with El d,σ d,l for each r. This isomorphism restricts to an isomorphism d,σ \ Z t d,l d,r for any r, l < r < d. d,σ which maps Yd,r,σ isomorphically onto d \ Z t d,r is identified with El d,l. Under the identification Qt d,l = Ql d,σ ∩ Z l d \Sl d,σ, and Et d,σ ∩ Z t d,σ ∩ Z t i=t+1 Z l d,σ \ Z t d,i, Et ∼→ Et ∼→ El Hence we obtain an isomorphism Pd,σ \ Yd,l,σ d,σ ∩ Z l El d,r for any r, l < r < d. (cid:3) ∼→ Et  / /       /  / /  _      _     / 32 YI HU AND YIJUN SHAO Let σ = (lm, · · · , l1) ∈ Σ and d > lm. We have the following commutative diagram Elm d,σ ≃ 7♣♣♣♣♣♣♣♣♣♣ d,τ ∩ Z lm−1 Elm−1 d,lm Pd,σ ≃ 8qqqqqqq Rd,lm,τ / Pd,τ ≃ 7♣♣♣♣♣♣♣♣♣♣ / Elm−1 d,τ . . .  / Pd,l3,l2,l1 ∩ Z l2 d,l3 d,l2,l1 / El3 d,l3,l2,l1 ≃ . . .  8qqqqqqqqq El2 7♣♣♣♣♣♣♣ ≃ Rd,l3,l2,l1 Pd,l2,l1 Rd,l2,l1 / El2 d,l2,l1 ≃ 7♥♥♥♥♥♥♥♥♥♥ d,l1 ∩ Z l1 El1 ≃ 6 6♠♠♠♠♠♠♠♠ / El1 d,l1 d,l2 ≃ 9rrrrrrr Pd,l1 Rd,l1 ≃ 8rrrrrrr φd,l1 / Qd ⑤ ⑤⑤ ⑤⑤ ⑤⑤ ⑤⑤ ⑤ Zd,l1 / Pd where τ = (lm−1, · · · , l1, l0). Thus we obtain a sequence of canonical identifications: Rd,l0 = Zd,l0, Pd,l0 = El0 d,l0 , Rd,l1,l0 = El0 d,l0 ∩ Z l0 d,l1 , · · · , Pd,σ = Elm d,σ. Recall that Qd parametrizes quotient of the form VP1 ։ X1 with deg X1 = d, and the subset Zd,l1 of Qd parametrizes such quotients with deg X t 1 = d − l1. Let x1 = [VP1 ։ X1] ∈ Zd,l1 = Rd,l1. Using the identification, d,l1 = Pd,l1 = P(π∗ Ext 1(¯q∗ El1 d,l1Td,l1, ¯p∗ d,l1 Xl1)), we see that the fiber of El1 d,l1 → Zd,l1 over x1 is P(Ext1((¯q∗ d,lTd,l)x1, (¯p∗ d,l Xl)x1)). On the other hand, we have an exact sequence 0 → ¯q∗ d,l1Td,l1 → ¯φ∗ d,l1Xd → ¯p∗ d,l1 Xl1 → 0 whose restriction to P1 × {x1}, 0 → (¯q∗ d,l1Td,l1)x1 → ( ¯φ∗ d,l1Xd)x1 → (¯p∗ d,l1 Xl1)x1 → 0 is also an exact sequence. Since φd,l1 is an inclusion map, we have that ( Hence (¯q∗ Xl1)x1 = X f 1 , and the fiber over x1 is P(Ext1(X t d,l1Td,l1)x1 = X t 1 and (¯p∗ ¯φ∗ d,l1 Xd)x1 = X1. 1 )), 1, X f d,l1   7     /   8   / 7    /    /   8   /     / / 7   7   /     / /   9   / 8   / THE SPACE OF COMPLETE QUOTIENTS 33 which parametrizes non-split extensions of the form [X f 1 sequences of the form ֌ X2 ։ X t 1]. Thus El1 d,l1 parametrizes ([VP1 ։ X1], [X f 1 ֌ X2 ։ X t 1]). with deg X t d,l1 ∩ Z l1 El1 deg X t 1 = d − l1 and deg X t d,l1 ∩ Z l1 d,l2, we see that El1 2 < d − l1. Using the identification Rd,l2,l1 = Yd,l2,l1 = d,l2 parametrizes such sequences with deg X t 1 = d − l1 and Let x2 = ([VP1 ։ X1], [X f 1 2 = d − l2, by the definition of Yd,l2,l1. ֌ X2 ։ X t d,l2,l1Td,l2, ¯p∗ over the point x2 is P(Ext 1((¯q∗ = Pd,l2,l1 = P(Ext 1(¯q∗ ∩ Z l1 d,l2 d,l2,l1 El2 El1 d,l1 d,l2,l1 1]) ∈ El1 d,l1 ∩ Z l1 d,l2 Xl2,l1)), we have that the fiber of El2 = Rd,l2,l1. Using the identification → Xl2,l1)x2)). On the other hand, d,l2,l1 d,l2,l1Td,l2)x2, (¯p∗ d,l2,l1 we have an exact sequence 0 → ¯q∗ d,l2,l1Td,l2 → ¯φ∗ d,l2,l1Xd,l1 → ¯p∗ d,l2,l1 Xl2,l1 → 0 whose restriction to P1 × {x2}, 0 → (¯q∗ d,l2,l1Td,l2)x2 → ( ¯φ∗ d,l2,l1Xd,l1)x2 → (¯p∗ d,l2,l1 Xl2,l1)x2 → 0 is also an exact sequence. Since Hence (¯q∗ 2, (¯p∗ which parametrizes non-split extensions of the form [X f 2 ¯φ∗ d,l2,l1 is an inclusion map, we have ( Xl2,l1)x2 = X f d,l2,l1Td,l2)x2 = X t d,l2,l1 ֌ X3 ։ X t ¯φ∗ d,l2,l1Xd,l1)x2 = X2. 2 )), 2, X f 2]. Thus El2 d,l2,l1 2, and the fiber over x2 is P(Ext1(X t parametrizes sequences of the form: ([VP1 ։ X1], [X f 1 ֌ X2 ։ X t 1], [X f 2 ֌ X3 ։ X t 2]). with deg X t d,l2,l1 ∩ Z l2 El2 deg X t 1 = d − l1, deg X t d,l3, by the definition of Yd,l3,l2,l1, El2 2 = d − l2 and deg X t d,l2,l1 ∩ Z l2 3 < d − l2. Since Rd,l3,l2,l1 = Yd,l3,l2,l1 = d,l3 parametrizes such sequences with 1 = d − l1, deg X t 2 = d − l2 and deg X t 3 = d − l3. Continuing this argument, we eventually obtain the parameter-space interpretation for d,σ: Elm Elm d,σ parametrizes sequences of the form ([VP1 ։ X1], [X f 1 ֌ X2 ։ X t 1], · · · , [X f m ֌ Xm+1 ։ X t m]) i = d − li, (i = 1, · · · , m) and deg X t d,r parametrizes such sequences with deg X t m+1 < d − lm. For any r with d > r > lm, m+1 = i = d − li, (i = 1, · · · , m) and deg X t with deg X t d,σ ∩ Z lm Elm d − r. Next, we deal with the modular interpretation of Er lemma before, we have Er parametrizes sequences as above with deg X t d,σ = Elm d,σ \Z lm d,r. Since Z lm d,r =Sr d,σ for any r, d > r > lm. By the d,σ ∩Z lm d,r m+1 ≥ d − r. Z lm d,i , we know that Elm i=lm+1 i = d − li, (i = 1, · · · , m) and deg X t 34 YI HU AND YIJUN SHAO Therefore, Er (i = 1, · · · , m) and deg X t d,σ, which equals Elm m+1 < d − r. d,σ \ Z lm d,r, parametrizes such sequences with deg X t i = d − li, In summary, we have Proposition 6.4. Let d > r. For any σ = (lm, · · · , l2, l1) ∈ Σ with lm < r, the stratum Er of Qr d parametrizes the sequences of the form d,σ ([VP1 ։ X1], [X f 1 ֌ X2 ։ X t 1], · · · , [X f m ֌ Xm+1 ։ X t m]) with deg X t i = d − li, (i = 1, · · · , m) and deg X t m+1 < d − r. Corollary 6.5. The boundary eQd \ Qd comes equipped with a natural stratification with strata Ed,σ indexed by σ = (lm, · · · , l2, l1) with d > lm > · · · > l1 ≥ 0,. The stratum Ed,σ parametrizes the sequences of the form ([VP1 ։ X1], [X f 1 ֌ X2 ։ X t 1], · · · , [X f m ֌ Xm+1 ։ X t m]) such that deg X t i = d − li, i = 1, · · · , m, and the last sheaf Xm+1 is the unique one that is locally free. This proves Theorem 1.3. References [HL97] D. Huybrechts and M. Lehn. The Geometry of Moduli Spaces of Sheaves. Vieweg & Sohn Verlagsge- sellschaft mbH, Braunschweig/Wiesbaden, 1997. [HLS11] Y. Hu, J. Lin, and Y. Shao, A Compactification of the Space of Algebraic Maps from P1 to Pn. Communications in Analysis and Geometry 19 (2011), no. 1, 1-30. [Shao11] Y. Shao. A compactification of the space of parametrized rational curves in Grassmannians. arXiv:1108.2299 [Str87] S. Strømme. On Parametrized Rational Curves in Grassmann Varieties. Lecture Notes in Mathematics 1266, Springer-Verlag, Berlin, New York, 1987. [Vain82] I. Vainsencher. Schubert calculus for complete quadrics. In Enumerative Geometry and Classical Algebraic Geometry (Nice 1981), volume 24 of Progress in Mathemtics, pages 199 -- 235. Birkhauser, 1982. [Vain84] I. Vainsencher. Complete collineations and blowing up determinantal ideals. Math. Ann., 267, 1984. Department of Mathematics, University of Arizona, USA. E-mail address: [email protected] E-mail address: [email protected]
0906.0553
6
0906
2010-02-01T09:15:43
One base point free theorem for weak log Fano threefolds
[ "math.AG" ]
Let $(X,D)$ be log canonical pair such $\dim X = 3$ and the divisor $-(K_X + D)$ is nef and big. For a special class of such $(X,D)$'s we prove that the linear system $|-n(K_{X}+D)|$ is free for $n \gg 0$.
math.AG
math
ONE BASE POINT FREE THEOREM FOR WEAK LOG FANO THREEFOLDS ILYA KARZHEMANOV Abstract. Let (X, D) be log canonical pair such dim X = 3 and the divisor −(KX + D) is nef and big. For a special class of such (X, D)'s we prove that the linear system − n(KX + D) is free for n ≫ 0. 1. Introduction Let X be algebraic variety1 of dimension > 2 with a Q-boundary D such that the pair (X, D) is log canonical and the divisor −(KX + D) is nef and big. Then one may consider the following Conjecture 1.1 (M. Reid (see [5], [11])). The linear system − n(KX + D) is free for n ≫ 0. According to [10, Proposition 11.1] (see also [11]), Conjecture 1.1 is true when dim X = 2. Unfortunately, it is false when dim X > 3: Example 1.2 (see [3]). Let Z be a smooth elliptic curve and E indecomposable rank 2 vector bundle over Z with deg(E) = 0 (see [1]). Put S = PZ(E) and let C be the tautological section on S. Then we have C 2 = 0 and KS = −2C. Let F be a fibre of the P1-bundle S → P1. Then the cone N E(S) is generated by two rays R1 = R>0[C], R2 = R>0[F ], and there is no curve C ′ In particular, the linear system − nKS is not free for any n. Consider the cone X over S ⊂ PN with respect to some projective embedding and the blow up σ : Y −→ X of the vertex on X with exceptional divisor E. We have −(KX + E) = σ∗(OX (1)) + π∗(−KS), where π : Y → S is the natural projection, which implies that −(KX + E) is nef and big. On the other hand, (X, E) is purely 6= C with [C ′] ∈ R1 (see [11, Example 1.1]). (X, E + π∗(C)) is log canonical, −(KX + E + π∗(C)) = σ∗(OX (1)) + π∗(C) is nef and big, but log terminal and − n(KX + E)(cid:12)(cid:12)E = − nKE is not free for any n because E ≃ S. Moreover, again − n(KX + E + π∗(C))(cid:12)(cid:12)E = (−n/2)KE is not free. The latter shows that Conjecture 1.1 is not true also for strictly log canonical pairs (the special case of xDy = 0 and dim X 6 4 was treated in [3]). It follows from Example 1.2 that the case when D contains a reduced part is far from being trivial. The present paper aims to correct the main result of [4]. We shall consider in some sense the simplest case when Conjecture 1.1 is true for dim X > 3 and xDy 6= 0: Theorem 1.3. Let (X, D) be as above. Suppose that • X is a smooth 3-fold and D = S is a smooth surface; • S · Z > 0 for every curve Z on S with KS · Z = 0. Then the linear system − n(KX + D) is free for n ≫ 0. It follows from Example 1.2 that the assertion of Theorem 1.3 is false without the additional assumption on KS-trivial curves. This suggests the following Conjecture 1.4. Let (X, D) be as above. Suppose that X is Q-factorial and (KX + D) · Sdim X−1 > 0 for every irreducible component S ⊆ xDy. Then the linear system − n(KX + D) is free for n ≫ 0. I would like to thank Y. Gongyo for pointing out the mistake in [4]. The work was partially supported by RFFI grant No. 08-01-00395-a and grant N.Sh.-1987.2008.1. 1All algebraic varieties are assumed to be projective and defined over C. 1 2. Preliminaries We use standard notation, notions and facts from the theory of minimal models and singu- larities of pairs (see [6], [9], [7], [8]). In what follows, (X, S) is the pair from Theorem 1.3. In order to prove Theorem 1.3, we assume that Bs( − n(KX + S)) 6= ∅,2 where n ≫ 0. Proposition 2.1. We have Proof. Consider the exact sequence Bs( − n(KX + S)) ∩ S = Bs( − n(KX + S)(cid:12)(cid:12)S) = Bs( − nKS) 6= ∅. 0 → OX (−n(KX + S) − S) → OX(−n(KX + S)) → OS(−n(KX + S)(cid:12)(cid:12)S) → 0. We have H 1(X, OX (−n(KX + S) − S)) = H 1(X, OX (KX − (n + 1)(KX + S))) = 0 by Kawamata -- Viehweg Vanishing Theorem. This gives the exact sequence which implies that H 0(X, OX (−n(KX + S)) → H 0(S, OS (−n(KX + S)(cid:12)(cid:12)S)) → 0, Bs( − n(KX + S)) ∩ S = Bs( − n(KX + S)(cid:12)(cid:12)S) = Bs( − nKS). Finally, if Bs( − n(KX + S)) ∩ S = ∅, then Bs( − n(KX + S)) = ∅ (see the proof of the Basepoint-free Theorem in [9])), a contradiction. (cid:3) From Proposition 2.1 we get the following Corollary 2.2. Equality K 2 S = 0 holds. Proof. Since −(KX + S) is nef, we have S = (KX + S)2 · S > 0 K 2 (see [9, Theorem 1.38]). Now, if K 2 S > 0, then −KS is nef and big, and the Basepoint-free Theorem implies that Bs( − nKS) = ∅, a contradiction. (cid:3) 3. Reduction to the non-complementary case We use notation and conventions of Section 2. Let us show that the surface S does not have Q-complements. Assume the contrary. Then we have the following Lemma 3.1. S is a rational surface. Proof. Suppose that S is non-rational. Then it follows from the proof of [2, Theorem 1.3], [11, Corollary 2.2] and [11, Example 2.1] that Bs(−nKS) = ∅, which contradicts Proposition 2.1. (cid:3) Thus, there exists a birational contraction χ : S −→ eS, where either χ is the blow up of eS = P2 at some points p1, . . . , p9, or χ is the blow up of eS = Fm, m ∈ N, at some points q1, . . . , q8 (see Corollary 2.2). To simplify the notation, in what follows we assume that all pi (respectively, all qi) are distinct. Further, by our assumption the equivalence KS + ∆ ∼ 0 holds for some effective Q-divisor ∆ such that the pair (S, ∆) is log canonical. Then we have −KS ∼ NX i=1 ∆i, N ∈ N, where ∆i are reduced and irreducible curves such that ∆i ∩ ∆j 6= ∅ for all i 6= j and the intersection is transversal. Lemma 3.2. Equality h0(S, OS(−nKS)) = 1 holds. 2Bs(M) denotes the base locus of the linear system M. 2 Proof. We have h0(S, OS (−nKS)) > 0. Suppose that h0(S, OS(−nKS)) > 2. Then, since the sum PN S = 0, − nKS is a free pencil, which (cid:3) i=1 ∆i is connected and −KS is nef with K 2 contradicts Proposition 2.1. Proposition 3.3. If N = 1, then eS 6= P2. Proof. Suppose that eS = P2. Let us consider two cases: 9X Case (1). The curve C = ∆1 is smooth. Write S(cid:12)(cid:12)S = χ∗(aL) + aiEi, i=1 where L is a line on P2, a and ai ∈ Z, Ei = χ−1(pi). Let ϕ : Y −→ X be the blow up of X at C with exceptional divisor E. For SY = ϕ−1 ∗ (S), ϕ induces an isomorphism ϕS : SY ≃ S such that ϕS(SY ∩ E) = C and ϕS is identical out of CY = SY · E, which implies that ϕS is an automorphism of S, identical on Pic(S). In particular, we have E · CY = E · E · SY = C 2 Y = 0, which together with equalities KY + SY = ϕ∗(KX + S) and SY = ϕ∗(S) − E implies that to prove Proposition 3.3 we may pass from (X, S) to the pair (Y, SY ). Moreover, we have SY (cid:12)(cid:12)SY = χ∗(aY LY ) + 9X i=1 ai,Y Ei,Y , where LY = ϕ−1 ∗ (L), Ei,Y = ϕ−1 ∗ (Ei), aY and ai,Y ∈ Z, which implies that −ai,Y = SY · Ei,Y = S · Ei − E · Ei,Y 6 S · Ei − 1 = −ai − 1, and hence ai,Y > ai. Thus, applying the above arguments to (Y, SY ), after a number of blow ups we obtain that to prove Proposition 3.3 we may assume that ai > 0 for all i. In particular, we have (KX + S) · E1 = KS · E1 < 0 and S · E1 = −a1 < 0, and it follows from the Cone Theorem that equality E1 ≡ Pi Ri +Pj Cj holds on X, where Ri are (KX + S)-negative extremal rays and (KX + S) · Cj = 0 for all j. Moreover, by assumption on KS-trivial curves we have S · Cj > 0 for all j, which implies that there exists a (KX + S)- negative extremal ray R on X such that S · R < 0. In particular, we have R ⊂ S and the extremal contraction contR : X −→ W is birational. Lemma 3.4. contR is not a divisorial morphism. Proof. Assume the contrary. Then the image of S is either a point or a curve. But the first case is impossible because (KX + S) · C = 0. Thus, contR(S) is a curve. Then there exists a 9 on P2, with birational contraction χ′ : S −→ P2, which is the blow up at some points p′ exceptional curves E′ 1, . . . , p′ 1, . . . , E′ 9 such that 1 · R = 1 and E′ • E′ • R = R>0[E′ i] for all i > 2. 1 · Z = 0 for some curve Z on S such that R = R>0[Z]; Let ϕ : Y −→ X be the blow up of X at E′ (KY + SY )-negative extremal ray RY = ϕ−1 numerical class of a fibre of ϕ. Note that α 6 0, which implies that 1 with exceptional divisor E. Then Y possesses a ∗ (S) and e is the ∗ (R) + αe, where α ∈ Q, SY = ϕ−1 SY · RY = S · R + α < 0 and hence RY ⊂ SY . In particular, RY = ϕ−1 ZY = ϕ−1 ∗ (Z) and E′ i), i > 2, we have i,Y = ϕ−1 ∗ (E′ ∗ (R). Moreover, for the curves E′ 1,Y = SY · E, 3 1,Y · RY = 1 and E′ • E′ • RY = R>0[E′ 1,Y · ZY = 0; i,Y ] = R>0[ZY ] for all i > 2. On the other hand, we get 0 = E′ 1,Y · ZY = E · ZY = E · RY = E′ 1,Y · RY = 1, a contradiction. (cid:3) Thus, contR is a small contraction. Then R is generated by a (−1)-curve on S. Consider the (KX + S)-flip: X /_______ τ X + AAAAAAAA contR }{{{{{{{{ cont+ R W, so that the map τ is an isomorphism in codimension 1, for every curve R+ ⊂ X +, which is contracted by cont+ R , we have (KX + + S+) · R+ > 0, where S+ = τ∗(S), 3-fold X + is Q-factorial and the pair (X +, S+) is purely log terminal (see [8] and [9, Proposition 3.36, Lemma 3.38]). Let T f + !BBBBBBBB f ~~~~~~~~ X /_______ τ X + i=1 Gi) is a union of all cont+ be resolution of indeterminacies of τ over W . Then f is a sequence of blow ups at smooth centers over R with exceptional divisors G1, . . . , Gs ⊂ T such that Gi constitute the f +-exceptional locus and Z = f +(Ps Lemma 3.5. We have Z ⊆ Bs( − n(KX + + D+)) and R+ 6⊂ S+ for every R+ ⊆ Z. Proof. The statement follows from conditions KX + + S+ = τ∗(KX + S), R 6⊂ Supp(−KS), (KX + + S+) · R+ > 0 for every R+ ⊆ Z and the fact that f −1 ∗ (S · (−n(KX + S))) (the latter holds because f is a sequence of blow ups at smooth centers). (cid:3) ∗ (−n(KX + S)) = f −1 R -exceptional curves. ∗ (S) · f −1 It follows from Lemma 3.5 that S+ ≃ contR(S) and τ induces contraction τS : S −→ S+ of the (−1)-curve in R. Then, since KX + + S+ = τ∗(KX + S), the divisor −(KX + + S+)(cid:12)(cid:12)S+ ≡ −KS+ = τS ∗(C) is nef and big. Moreover, the surface S+ has only log terminal singularities by the Inversion of adjunction, which implies that Bs( − nKS+) = ∅ by the Basepoint-free Theorem. Lemma 3.6. Inequality h0(S, OS (−nKS)) > 2 holds. Proof. We have R1(contR)∗(−n(KX + S)− S) = 0 by the relative Kawamata -- Viehweg Vanishing Theorem. This and the isomorphism S+ ≃ S∗ = contR(S) imply that the push-forwards to W of exact sequences 0 → OX (−n(KX + S) − S) → OX (−n(KX + S)) → OS(−n(KX + S)(cid:12)(cid:12)S) → 0 and 0 → OX +(−n(KX + + S+) − S+) → OX + (−n(KX + + S+)) → OS+(−n(KX + + S+)(cid:12)(cid:12)S+) → 0 coincide with exact sequence 0 → OW (−n(KW + S∗) − S∗) → OW (−n(KW + S∗)) → OS ∗ (−n(KW + S∗)(cid:12)(cid:12)S ∗) → 0. Then it follows from Bs( − nKS+) = ∅ that h0(S, OS(−nKS)) > 2. (cid:3) From Lemma 3.6 we get contradiction with Lemma 3.2. Thus, Case (1) is impossible, and we pass to Case (2). The curve C = ∆1 is singular. Since C ∼ −KS and the pair (S, C) is log canonical, we have pa(C) = 1 and the only singular point on C is an ordinary double point O. 4 / }  ! / Let ϕ : Y −→ X be the blow up of X at C with exceptional divisor E. Locally near O there is an analytic isomorphism (X, S, ∆) ≃ (cid:0)C3 x,y,x, {x = 0}, {yz = 0}(cid:1). Then locally over O we have the following representation for Y : Y = {yzt0 = xt1} ⊂ C3 x,y,z × P1 t0,t1, which implies that the only singular point on Y is a non-Q-factorial quadratic singularity. Then, since KY + ϕ−1 ∗ (S) = ϕ∗(KX + S), after a small resolution ψ : eY −→ Y we may pass from (X, S) to the pair (eY , ψ−1 above and apply the arguments from Case (1). ∗ (ϕ−1 ∗ (S))) as (cid:3) Applying the same arguments as in the proof of Proposition 3.2 to eS = Fm, we see that the case N = 1 is impossible. Finally, in the case when N > 2 we apply the same arguments as in Thus, the surface S does not have Q-complements, and we get the following the proof of Proposition 3.2, replacing the curve ∆1 with the cycle PN Corollary 3.7. In the notation of Example 1.2, we have S = PZ(E) and Supp(−n(KX +S)(cid:12)(cid:12)S) = C. In particular, Bs( − n(KX + S)) ∩ S = C. i=1 ∆i. Proof. The statement follows from [2, Theorem 1.3] and Proposition 2.1. (cid:3) Let F be a fibre of the P1-bundle S → P1. Write where a, b ∈ Z. Note that b < 0 because C is KS-trivial and hence 0 < S · C = S(cid:12)(cid:12)S · C = −b by assumption on KS-trivial curves. S(cid:12)(cid:12)S = aC − bF, Lemma 3.8. Equality deg(NC/X ) = −b holds. Proof. Since C is a smooth elliptic curve, we have deg(NC/X ) = −KX · C = ((2 + a)C − bF ) · C = −b. (cid:3) Put Ln = − n(KX + S). Then for the general element Ln ∈ Ln we have Ln = M + X ri,SBi,S + X riBi, where Bi, Bi,S are the base components of Ln, ri, ri,S > 0 the corresponding multiplicities, Bi∩S = ∅, Bi,S∩S 6= ∅ for all i, and the linear system M is movable. According to Corollary 3.7, we have Bs(−n(KX +D))∩S = C and Bi,S ∩S = C for all i, which implies that Bs(M )∩S = C or ∅. In what follows, we assume that Bs(M ) = Bs(M )∩S (see the proof of the Basepoint-free Theorem in [9]). Furthermore, arguing exactly as in the proof of Proposition 3.2, we can replace X by its blow up at the curve C. Then, after applying Corollary 3.7 and a number of blow ups, in what follows we assume that the following conditions are satisfied: • ri,S = r > 0 and Bi,S = B for all i, where B = PC(NC/X) with B3 = − deg(NC/X ) = b (see Lemma 3.8); • S · B = C; • the linear system M is free and M ∩ B = ∅; • Bj ∩ B 6= ∅ for exactly one j and the intersection is transversal, rj = r, B2 j · B = b. 5 4. Exclusion of the non-complementary case We use notation and conventions of Section 3. Let ϕ : Y −→ X be the blow up of X at the curve C with exceptional divisor E. Put SY = ϕ−1 ∗ (S), BY = ϕ−1 ∗ (B), MY = ϕ−1 ∗ (M ), Bi,Y = ϕ−1 ∗ (Bi). Then for m ≫ 0, 0 < δ1, δ2 ≪ 1 and 0 < c 6 1 we write (4.1) R = ϕ∗(−(KX + S) + mLn − cLn) + cMY + δ1SY + δ2E = = ϕ∗(mLn) + (−1 + δ1)SY + (δ2 − cr)E − criBi,Y − KY . −crBY − crBj,Y − X i6=j Proposition 4.2. The divisor R is nef and big for δ1 > δ2. Proof. Since the divisors −(KX + S) and MY are nef and big, it suffices to prove that the divisor R = ϕ∗(−(KX + S) + mLn − cLn) + cMY + δ1SY + δ2E intersects every curve on the surfaces SY and E non-negatively. Lemma 4.3. Inequality R · Z > 0 holds for every curve Z on SY . Proof. Since SY ≃ S, the cone N E(SY ) is generated by the classes [CY ] = [SY · E] and [FY ] = [ϕ−1 ∗ (F )]. Thus, it is enough to consider only the cases when Z = C or F . We have E · CY = 0 and which implies that R · CY > 0. Furthermore, we have SY · CY = S · C = −b > 0, R · FY ≫ ϕ∗(Ln) · FY = Ln · F > nC · F = n ≫ 0, and the assertion follows. (cid:3) Lemma 4.4. Inequality R · Z > 0 holds for every curve Z on E and δ1 > δ2. Proof. Let FE be a fibre of the P1-bundle E = PC(NC/X). We have (BY (cid:12)(cid:12)E)2 = (ϕ∗(B) − E)2 · E = 2B · C + E3 = 2C 2 + E3 = − deg(NC/X ) = b < 0, which implies that the cone N E(E) is generated by the classes [−E(cid:12)(cid:12)E] = [BY (cid:12)(cid:12)E] and [FE ] (see [9, Lemma 1.22]). Thus, it is enough to consider only the cases when Z = −E(cid:12)(cid:12)E or FE. We have SY · (−E(cid:12)(cid:12)E) = −SY · E2 = 0, R · (−E(cid:12)(cid:12)E) > δ2E · (−E(cid:12)(cid:12)E) = −bδ2 > 0. SY · FE = 1 and E · FE = −1, R · FE = δ1 − δ2 > 0, which implies that Furthermore, we have which implies that and the assertion follows. Lemmas 4.3 and 4.4 prove Proposition 4.2. 6 (cid:3) (cid:3) Take c = 1/r in (4.1). Then we obtain pRq = ϕ∗(mLn) − BY − Bj,Y + X p−criqBi,Y − KY , i6=j and Proposition 4.2 and Kawamata -- Viehweg Vanishing Theorem imply that H i(Y, OY (ϕ∗(mLn) − BY − Bj,Y + X p−criqBi,Y )) = 0 i6=j (4.5) for all i > 0. Lemma 4.6. Inequality H 0(BY , OBY ((ϕ∗(mLn) − Bj,Y + X p−criqBi,Y )(cid:12)(cid:12)BY )) 6= 0 holds. i6=j Proof. Note that (Pi6=j p−criqBi,Y )(cid:12)(cid:12)BY = 0. Let us prove that H 0(BY , OBY ((ϕ∗(mLn) − Bj,Y )(cid:12)(cid:12)BY )) 6= 0. We have ϕ∗(mLn) = mMY + mrBY + mrBj,Y + mrE + X mriBi,Y , i6=j which implies that ϕ∗(mLn)(cid:12)(cid:12)BY = mrBY(cid:12)(cid:12)BY + mrBj,Y(cid:12)(cid:12)BY + mrE(cid:12)(cid:12)BY . Further, since BY = ϕ∗(B) − E and ϕ∗(B) · E2 = −B · C = 0, we obtain (E(cid:12)(cid:12)BY )2 = E2 · BY = −E3 = −b (Bj,Y(cid:12)(cid:12)BY )2 = B2 j · B = b, + bFBY . Furthermore, we have )2 = B3 Y = ϕ∗(B)3 − E3 = 0 and which implies that E(cid:12)(cid:12)BY fibre FBY , and Bj,Y(cid:12)(cid:12)BY ∼ E(cid:12)(cid:12)BY and which implies that BY (cid:12)(cid:12)BY (BY (cid:12)(cid:12)BY BY(cid:12)(cid:12)BY · E(cid:12)(cid:12)BY ϕ∗(mLn)(cid:12)(cid:12)BY − Bj,Y(cid:12)(cid:12)BY and hence H 0(BY , OBY ((ϕ∗(mLn) − Bj,Y )(cid:12)(cid:12)BY ϕ∗(mLn)(cid:12)(cid:12)BY which implies that ∼ bFBY . Thus, we get = B2 Y · E = E3 = b, , ∼ 2mrBj,Y(cid:12)(cid:12)BY ∼ (2mr − 1)Bj,Y (cid:12)(cid:12)BY )) 6= 0. is the tautological section of the P1-bundle BY = PC(NC/X) with a (cid:3) From (4.5) and the exact sequence 0 → OY (ϕ∗(mLn) − BY − Bj,Y + X → OY (ϕ∗(mLn) − Bj,Y + X i6=j p−criqBi,Y ) → p−criqBi,Y ) → → OBY ((ϕ∗(mLn) − Bj,Y + X i6=j p−criqBi,Y )(cid:12)(cid:12)BY ) → 0 i6=j 7 we get exact sequence 0 → H 0(Y, OY (ϕ∗(mLn) − BY − Bj,Y + X → H 0(Y, OY (ϕ∗(mLn) − Bj,Y + X i6=j p−criqBi,Y )) → p−criqBi,Y )) → → H 0(BY , OBY ((ϕ∗(mLn) − Bj,Y + X i6=j p−criqBi,Y )(cid:12)(cid:12)BY i6=j )) → 0, which implies, since −ri 6 p−criq 6 0 and BY , Bj,Y , Bi,Y are the base components of the linear system ϕ∗(mLn), that H 0(Y, OY (ϕ∗(mLn) − BY − Bj,Y + X p−criqBi,Y )) ≃ ≃ H 0(Y, OY (ϕ∗(mLn) − Bj,Y + X i6=j i6=j p−criqBi,Y )) ≃ H 0(Y, OY (ϕ∗(mLn))) and H 0(BY , OBY ((ϕ∗(mLn) − Bj,Y + X p−criqBi,Y )(cid:12)(cid:12)BY i6=j )) = 0, a contradiction with Lemma 4.6. Theorem 1.3 is completely proved. References [1] Atiyah M. Vector bundles over an elliptic curve // Proc. Lond. Math. Soc. 1957. V. 7. P. 414 -- 452. [2] Fedorov I. Yu, Kudryavtsev S. A. Q-Complements on log surfaces // Proc. of the Steklov Institute. 2004. V. 246. P. 181 -- 182. [3] Gongyo Y. On semiampleness of anti-canonical divisors of weak Fano varieties with log canonical singularities // arXiv: math. AG0911.0974. 2009. P. 1 -- 12. [4] Karzhemanov I. V. Semiampleness theorem for weak log Fano varieties // Russ. Acad. Sci. Sb. Math. 2006. V. 197 (10). P. 57 -- 64. [5] Kawamata Y. The crepant blowing-up of 3-dimensional canonical singularities and its application to the degeneration of surfaces // Ann. of Math. 1988. V. 127(2). P. 93 -- 163. [6] Kawamata Y., Matsuda K., Matsuki K. Introduction to the Minimal Model Problem // Advanced Studies in Pure Math. 1987. V. 10. P. 283 -- 360. [7] Koll´ar J. Singularities of pairs // Proc. Symp. Pure Math. 1997. V. 62. P. 221 -- 287. [8] Koll´ar J. et al. Flips and Abundance for Algebraic Threefolds // Ast´erisque. 1992. V. 211. [9] Koll´ar J., Mori S. Birational geometry of algebraic varieties // Cambridge Univ. Press. 1998. [10] Prokhorov Yu. G. Lectures on complements on log surfaces // MSJ Memoirs. V. 10. 2001. [11] Shokurov V. V. Complements on surfaces // J. Math. Sci. (New York). 2000. V. 102(2). P. 3876-3932. 8
1503.03103
1
1503
2015-03-10T21:41:52
Transposing Noninvertible Polynomials
[ "math.AG" ]
Landau-Ginzburg mirror symmetry predicts isomorphisms between graded Frobenius algebras (denoted $\mathcal{A}$ and $\mathcal{B}$) that are constructed from a nondegenerate quasihomogeneous polynomial $W$ and a related group of symmetries $G$. Duality between $\mathcal{A}$ and $\mathcal{B}$ models has been conjectured for particular choices of $W$ and $G$. These conjectures have been proven in many instances where $W$ is restricted to having the same number of monomials as variables (called $\textit{invertible}$). Some conjectures have been made regarding isomorphisms between $\mathcal{A}$ and $\mathcal{B}$ models when $W$ is allowed to have more monomials than variables. In this paper we show these conjectures are false; that is, the conjectured isomorphisms do not exist. Insight into this problem will not only generate new results for Landau-Ginzburg mirror symmetry, but will also be interesting from a purely algebraic standpoint as a result about groups acting on graded algebras.
math.AG
math
Transposing Noninvertible Polynomials Nathan Cordner July 25, 2021 Abstract Landau-Ginzburg mirror symmetry predicts isomorphisms between graded Frobenius algebras (denoted A and B) that are constructed from a nondegenerate quasihomogeneous polynomial W and a related group of symmetries G. Duality between A and B models has been conjectured for particular choices of W and G. These conjectures have been proven in many instances where W is restricted to having the same number of monomials as variables (called invertible). Some conjectures have been made regarding isomorphisms between A and B models when W is allowed to have more monomials than variables. In this paper we show these conjectures are false; that is, the conjectured isomorphisms do not exist. Insight into this problem will not only generate new results for Landau- Ginzburg mirror symmetry, but will also be interesting from a purely algebraic standpoint as a result about groups acting on graded algebras. 1 Introduction Physicists conjectured some time ago that that to each quasihomogeneous (weighted homogeneous) polynomial W with an isolated singularity at the origin, and to each admissible group of symmetries G of W , there should exist two different physical "theories," (called the Landau-Ginzburg A and B models, respectively) consisting of graded Frobenius algebras (algebras with a nondegenerate pairing that is compatible with the multiplication). The B-model theories have been constructed [6, 7, 8, 9, 10] and correspond to an "orbifolded Milnor ring." The A-model theories have also been constructed [4] and are a special case of what is often called "FJRW theory." We will not address these in this paper, but in many cases, these theories can be extended to whole families of Frobenius algebras, called Frobenius manifolds. For a large class of these polynomials (called invertible) Berglund-Hubsch [3], Henningson [2], and Krawitz [10] described the construction of a dual (or transpose) polynomial W T and a dual group GT . The Landau-Ginzburg mirror symmetry conjecture states that the A-model of a pair W, G should be isomor- phic to the B-model of the dual pair W T , GT . This conjecture has been proved in many cases in papers such as [10] and [5], although the proof of the full conjecture remains open. It has been further conjectured that the Berglund-Hubsch-Henningson-Krawitz duality transform should extend to large classes of noninvertible polynomials and that Landau-Ginzburg mirror symmetry should also hold for these polynomials. In this paper we investigate some candidate mirror pairs of noninvertible polynomials and show that many obvious candidates for mirror duality cannot satisfy mirror symmetry. To approach this problem, we study the A and B models as graded vector spaces and inspect how the symmetry groups act on these spaces. Insight into this problem will not only generate new results for Landau-Ginzburg mirror symmetry, but will also be interesting from a purely algebraic standpoint as a result about groups acting on graded algebras. One case of mirror symmetry that has been verified for all invertible polynomials is when the A-model is constructed from an invertible polynomial W with its maximal group of symmetries and the B-model is constructed from the corresponding transpose polynomial with the trivial group of symmetries. This is ∼= BW T ,{0}. This intuition stemming from invertible polynomials motivated sometimes denoted AW,Gmax two conjectures about isomorphisms between A and B models built from noninvertible polynomials. We often refer to polynomials for which the A and B models exist as admissible. W Conjecture 1. For any admissible (not necessarily invertible) polynomial W in n variables, there exists a corresponding admissible polynomial W T in n variables satisfying AW,Gmax ∼= BW T ,{0}. W 1 Note that this conjecture includes the collection of noninvertible polynomials, which are allowed to have more monomials than variables. In Section 3.1 we show that this conjecture is false. By relaxing the restriction on the number of variables that W T is allowed to have, we obtain a second conjecture. Conjecture 2. For any admissible W , there is a corresponding admissible W T satisfying AW,Gmax BW T ,{0}. In Section 3.2 we look at an example of a particular noninvertible polynomial, and expand our search space for finding a suitable W T . We develop some formulas and show that they rule out the existence of W T in a few more cases that were not considered in Conjecture 1. Thereby we also establish that Conjecture 2 is unlikely to be true in general. ∼= W 2 Preliminaries Here we will introduce some of the concepts needed to explain the theory of this paper. 2.1 Admissible Polynomials Definition. For a polynomial W ∈ C[x1, . . . , xn], we say that W is nondegenerate if it has an isolated critical point at the origin. Definition. Let W ∈ C[x1, . . . , xn]. We say that W is quasihomogeneous if there exist positive rational numbers q1, . . . , qn such that for any c ∈ C, W (cq1x1, . . . , cqnxn) = cW (x1, . . . , xn). We often refer to the qi as the quasihomogeneous weights of a polynomial W , or just simply the weights of W , and we write the weights in vector form J = (q1, . . . , qn). Definition. W ∈ C[x1, . . . , xn] is admissible if W is both nondegenerate and quasihomogeneous, with the weights of W being unique. We will use the following result about admissible polynomials later in the paper. Proposition 2.1.6 of [4]. If W ∈ C[x1, . . . , xn] is admissible, and contains no monomials of the form xixj for i (cid:54)= j, then the qi are bounded above by 1 2 . Because the construction of AW,G requires an admissible polynomial, we will only be concerned with admissible polynomials in this paper. In order for a polynomial to be admissible, it needs to have at least as many monomials as variables. Otherwise its quasihomogeneous weights cannot be uniquely determined. We now state the main subdivision of the admissible polynomials. Definition. Let W be an admissible polynomial. We say that W is invertible if it has the same number of monomials as variables. If W has more monomials than variables, then it is noninvertible. Admissible polynomials with the same number of variables as monomials are called invertible since their associated exponent matrices (which we define in the next section) are square and invertible. We will now introduce the idea of the transpose operation for invertible polynomials. 2.2 Dual Polynomials Definition. Let W ∈ C[x1, . . . , xn]. If we write W =(cid:80)m (cid:81)n matrix is defined to be A = (aij). i=1 ci j=1 xaij j , then the associated exponent From this definition we notice that n is the number of variables in W , and m is the number of monomials in W . A is an m × n matrix. Thus when W is invertible, we have that m = n which implies that A is square. One can show, without much work, that this square matrix is invertible if the polynomial W is quasihomogeneous with unique weights. When W is noninvertible, m > n. A then has more rows than columns. Observe that if a polynomial is invertible, then we may rescale all nonzero coefficients to 1. So there is effectively a one-to-one correspondence between exponent matrices of invertible polynomials and the polynomials themselves. 2 Definition. Let W be an invertible polynomial. If A is the exponent matrix of W , then we define the transpose polynomial to be the polynomial W T resulting from AT . By the classification in [11], W T is again a nondegenerate, invertible polynomial. We now have reached our fundamental problem. When a polynomial W is noninvertible, its exponent matrix A is no longer square. Taking AT yields a polynomial with fewer monomials than variables, which is not admissible. Therefore, we will require a different approach to define what the transpose polynomial should be for noninvertibles. 2.3 Symmetry Groups and Their Duals Definition. Let W be an admissible polynomial. We define the maximal Abelian symmetry group of W to be Gmax W = {(ζ1, . . . , ζn) ∈ (C×)n W (ζ1x1, . . . , ζnxn) = W (x1, . . . , xn)}. The proofs of Lemma 2.1.8 in [4] and Lemma 1 in [1] observe that Gmax of every group element is a root of unity. The group operation ◦ in Gmax That is, W is finite and that each coordinate W is coordinate-wise multiplication. (e2πiθ1 , . . . , e2πiθn ) ◦ (e2πiφ1 , . . . , e2πiφn ) = (e2πi(θ1+φ1), . . . , e2πi(θn+φn)). Equivalently, in additive notation we can write (θ1, . . . , θn)+(φ1, . . . , φn) = (θ1+φ1, . . . , θn+φn) mod Z. The map (e2πiθ1, . . . , e2πiθn ) (cid:55)→ (θ1, . . . , θn) mod Z gives a group isomorphism. Using additive notation, we will often write Gmax W = {g ∈ (Q/Z)n Ag ∈ Zm}, where A is the m × n exponent matrix of W . Definition. In this notation, Gmax For g ∈ Gmax are called the phases of g. W is a subgroup of (Q/Z)n with respect to coordinate-wise addition. W , we write g = (g1, . . . , gn) where each gi is a rational number in the interval [0,1). The gi The following definition of the transpose group is due to Krawitz and Henningson [10, 2]. Definition. Let W be an invertible polynomial, and let A be its associated exponent matrix. The transpose group of a subgroup G ≤ Gmax Since this relies on knowing what W T is, this definition currently does not extend to noninvertible polynomials. The following is a list of common results for the transpose group. Proposition 2 of [1]. Let W be an invertible polynomial with weights vector J, and let G ≤ Gmax W . W T gAhT ∈ Z for all h ∈ G}. W is the set GT = {g ∈ Gmax (1) (GT )T = G, (2) {0}T = Gmax (3) (cid:104)J(cid:105)T = Gmax (4) if G1 ≤ G2, then GT W )T = {0}, W T and (Gmax W T ∩ SL(n, C) where n is the number of variables in W , 2 ≤ GT ∼= GT 2.4 Some Notes on A and B Models Landau-Ginzburg A and B models are algebraic objects that are endowed with many levels of structure. In this paper, we will chiefly be concerned with their structure as graded vector spaces, although we will also occasionally consider their Frobenius algebra structure. For the benefit of the reader, we will give a formal definition of a Frobenius algebra. Definition. An algebra is a vector space A over a field of scalars F (in our case it is C), together with a multiplication · : A × A → A that satisfies for all x, y, z ∈ A and α, β ∈ F 1 and G2/G1 1 /GT 2 . • Right distributivity: (x + y) · z = x · z + y · z, • Left distributivity: x · (y + z) = x · y + x · z, • Compatability with scalars: (αx) · (βy) = (αβ)(x · y). We further require the multiplication to be associative and commutative, and for A to have a unity e such that e · x = x for all x ∈ A. We also define a pairing operation (cid:104)·,·(cid:105) : A × A → F that is • Symmetric: (cid:104)x, y(cid:105) = (cid:104)y, x(cid:105), • Linear: (cid:104)αx + βy, z(cid:105) = α(cid:104)x, z(cid:105) + β(cid:104)y, z(cid:105), • Nondegenerate: for every x ∈ A there exists y ∈ A such that (cid:104)x, y(cid:105) (cid:54)= 0. If the pairing further satisfies the Frobenius property, meaning that (cid:104)x· y, z(cid:105) = (cid:104)x, y · z(cid:105) for all x, y, z ∈ A, then we call A a Frobenius algebra. 3 ∂x1 (cid:17) − 1 , . . . , ∂W ∂xn ) is called the Milnor ring of W (or local algebra of W ). We will only develop the theory needed for the proofs in Section 3. We refer the interested reader to [4] for more details on the construction of the A-model. [5], [10], and [12] also contain more information on constructing A and B models, and related isomorphisms. We will start by discussing the B-model. Definition. QW = C[x1, . . . , xn]/( ∂W Definition. We define the unorbifolded B-model to be BW,{0} = QW . We will think of the unorbifolded B-model as a graded vector space over C. The degree of a monomial i aiqi. This defines a grading on the basis of QW . We note in QW is given by deg(xa1 the following: Theorem 2.6 of [12]. If W is admissible, then QW is finite dimensional. We will need two results about the unorbifolded B-model. First, dim(BW,{0}) =(cid:81)n the highest degree of its graded pieces is 2(cid:80)n i=1 (1 − 2qi). (See Section 2.1 of [10]) n ) = 2(cid:80)n 2 . . . xan (cid:16) 1 . Second, 1 xa2 We will now develop some needed ideas about A-models. Definition. Let W be an admissible polynomial with weights vector J = (q1, . . . , qn), and let G ≤ Gmax W . Then G is admissible if J ∈ G. We note that since W is quasihomogeneous, we have that AJ T = (1, . . . , 1)T ∈ Zm. Thus J ∈ Gmax W . The construction of the A-model requires that G be an admissible group. From parts (3) and (4) of the proposition in Section 2.3, the corresponding condition for the B-model is that GT ≤ Gmax W T ∩ SL(n, C). Definition. Let W ∈ C[x1, . . . , xn] be admissible, and let g = (g1, . . . , gn) ∈ Gmax W . The fixed locus of the group element g is the set fix(g) = {xi gi = 0}. We now state how G acts on the Milnor ring. W . We define the map g∗ : QW → QW Definition. Let W be an admissible polynomial, and let g ∈ Gmax by g∗(m) = det(g)m ◦ g. (Here we think of g as being a diagonal map with multiplicative coordinates) Definition. Let W be an admissible polynomial, and let G ≤ Gmax W = {m ∈ QW g∗(m) = m for each g ∈ G}. QW is defined to be QG Definition. Let W be an admissible polynomial, and G an admissible group. We define AW,G = W . Then the G-invariant subspace of i=1 qi (cid:77) (cid:16)QWfix(g) (cid:17)G , where (·)G denotes all the G-invariants. This is called the A-model state space. g∈G We further note that the state space of the orbifolded B-model BW,G is constructed similarly, but with W ∩ SL(n, C). If we let G = {0}, then the formula yields the Milnor ring of W the condition that G ≤ Gmax as expected. The grading on the A-model, which will will define in a moment, differs from the B-model grading; but as graded vector spaces, the A and B models are very much related. We will not discuss many details of constructing the state space here. For further treatment of this topic, we refer the reader to Section 2.4 of [12]. A brief comment on notation: we represent basis elements of AW,G in the form (cid:98)m; g(cid:101), where m is a monomial and g is a group element. Definition. The A-model degree of a basis element (cid:98)m; g(cid:101) is defined to be deg((cid:98)m; g(cid:101)) = dim(fix(g)) + i=1(gi − qi), where g = (g1, . . . , gn) with the gi chosen such that 0 ≤ gi < 1 and J = (q1, . . . , qn) is 2(cid:80)n the vector of quasihomogeneous weights of W . (See Section 2.1 of [10]) Finally, we state one important theorem for A-model isomorphisms. Theorem in Section 7.1 of [12] (Group-Weights). Let W1 and W2 be admissible polynomials which have the same weights. Suppose G ≤ Gmax Note that one can give the A-model a product and pairing such that A is a Frobenius algebra. The above is then an isomorphism of Frobenius algebras, not just graded vector spaces. . Then AW1,G and G ≤ Gmax ∼= AW2,G. W1 W2 4 2.5 Properties of Invertible Polynomials Our initial intuition tells us that some of the properties of invertible polynomials should extend to the noninvertible case. For example, we'd like to keep the results of the following proposition. Proposition. Let W be an invertible polynomial. Then (1) W and W T have the same number of variables. (2) (Gmax ∼= BW T ,{0}, as graded vector spaces. (3) AW,Gmax W )T = {0}. W Proof. (1) follows from noticing that the exponent matrix of W is square. Hence its transpose is also square and of the same size, so W and W T have the same number of variables. (2) was stated previously in Section 2.3. (3) is a special case of the mirror symmetry conjecture that has been verified. Reference Theorem 4.1 in [10]. (cid:3) Part (3) of the proposition is especially important, and will be what we use to look for candidate transpose polynomials. In other words, given a noninvertible polynomial W , we would like to identify a candidate (cid:17)Gmax W ∼= QW T . Though we would like this isomorphism (cid:16)QWfix(g) polynomial W T that satisfies (cid:77) g∈Gmax W to hold for all levels of algebraic structure, we will mainly investigate it on the level of graded vector spaces. For the benefit of the reader, we will restate the first conjecture. Conjecture 1. For any admissible polynomial W in n variables, there exists a corresponding admissible polynomial W T in n variables satisfying AW,Gmax ∼= BW T ,{0}. W 3 Results 3.1 Disproving Conjecture 1 with weight system J =(cid:0) 1 To disprove Conjecture 1, we prove a related nonexistence result. Note that this theorem is about any W,(cid:104)J(cid:105), whereas Conjecture 1 is about W, Gmax W . Theorem. For any n ∈ N, n > 3, let W be an admissible but noninvertible polynomial in two variables variables satisfying AW,G Before proving this theorem, we will demonstrate the hypothesis by exhibiting a few examples of such admissible polynomials for small values of n. (cid:1), and let G = (cid:104)J(cid:105). Then there does not exist a corresponding W T in two n , 1 ∼= BW T ,{0}. n n 4 6 J 4 , 1 4 (cid:0) 1 (cid:0) 1 6 , 1 6 Some Examples x4 + y4 + x3y (cid:1) x4 + x2y2 + xy3 (cid:1) x5y + x4y2 + y6 x4 + xy3 x6 + y6 + x5y n 5 7 J 5 , 1 5 (cid:0) 1 (cid:0) 1 7 , 1 7 Some Examples x5 + y5 + x4y (cid:1) x4y + xy4 + x3y2 + x2y3 (cid:1) x6y + x5y2 + y7 x5 + x2y3 + xy4 x7 + y7 + x6y x6y + xy6 x6 + x2y4 + xy5 + y6 compute some formulas for its A-model using the group (cid:104)J(cid:105), and show that there is no corresponding isomorphic unorbifolded B-model. Then, under the Group-Weights isomorphism for A-models, we will be able to generalize the result for any admissible polynomial with the same weights. Proof. The idea of this proof is to choose an admissible polynomial with weight system J =(cid:0) 1 To start, we need an admissible polynomial in two variables with weight system J = (cid:0) 1 For the unorbifolded B-model, we know that dim(BW T ,{0}) =(cid:81)n of its graded pieces is given by 2(cid:80)n (cid:17) − 1 ∼= BW T ,{0}, we need the degrees W (cid:48) = xn + yn + xn−1y, and let G = (cid:104)J(cid:105). Certainly W (cid:48) has weight system J, and G fixes W (cid:48). (cid:1), (cid:1). Let i=1 (1 − 2qi). In order to have AW,G and that the highest degree (cid:16) 1 n , 1 n , 1 of the vector spaces and the degrees of each of the graded pieces to be equal. Therefore we now need i=1 qi n n 5 corresponding formulas for the dimension of the A-model vector space and the degree of the highest degree piece of the A-model. Lemma. As a graded vector space, dim (AW (cid:48),G) = 2n − 2, and the highest degree of any element is 2(2n−4) . (n ∈ N, n ≥ 3). n Proof of Lemma. Recall that AW (cid:48),G = {(0, 0),(cid:0) 1 n , 1 n (cid:1), . . . ,(cid:0) n−1 n , n−1 n (cid:77) (cid:16)QW (cid:48)fix(g) (cid:1)(cid:105) = (cid:1)}. Then W (cid:48)fix(g) = W (cid:48) only for g = (0, 0). Otherwise W (cid:48)fix(g) is trivial. . Notice that in our case G = (cid:104)(cid:0) 1 (cid:17)G n , 1 g∈G n Case 1 When W (cid:48)fix(g) is trivial, we get n − 1 basis elements of the form (cid:98)1; g(cid:101). Case 2 W (cid:48)fix(g) = W (cid:48). Then g = (0, 0). The basis elements we get in this case are of the form (cid:98)xayb; (0, 0)(cid:101) where a + b ≡ n − 2 mod n and a, b ∈ {0, 1, . . . , n − 2}. So we have (a, b) = (0, n − 2), (1, n − 3), . . . , (n − 3, 1), (n − 2, 0). Hence there are n − 1 basis elements of this type. The total dimension of AW (cid:48),G is therefore (n − 1) + (n − 1) = 2n − 2. Now we will consider the degree of each basis element. Recall that deg((cid:98)m; g(cid:101)) = dim(fix(g)) + 2 (gi − qi), n(cid:88) i=1 where g = (g1, . . . , gn) and J = (q1, . . . , qn) is the vector of quasihomogeneous weights. For g = (0, 0), the degree is 2 +(cid:0)− 2 deg(cid:0)(cid:98)1;(cid:0) n−1 (cid:1)(cid:101)(cid:1) > deg(cid:0)(cid:98)1;(cid:0) m n , n−1 n (cid:1) = 2(n−2) (cid:1) +(cid:0)− 2 (cid:1)(cid:101)(cid:1) for all m ∈ {1, . . . , n − 2}. Compute deg(cid:0)(cid:98)1;(cid:0) n−1 (cid:16) 2(n−2) (cid:17) n n n n , and notice that 2(2n−4) 2(2n−4) degree part of AW,G is 2(2n−4) . (cid:3) From the lemma, we now have the following system of equations for the possible weights q1, q2 for a candidate W T : for all n ≥ 3. Hence the degree of the highest > 2(n−2) n n n n (cid:1)(cid:101)(cid:1) = n , n−1 n . Also notice by the above equation that n , m = 2 n (cid:18) 1 q1 (cid:19)(cid:18) 1 q2 (cid:19) − 1 − 1 = 2n − 2, 2 ((1 − 2q1) + (1 − 2q2)) = 2(2n − 4) n . Solving for q1 in the second equation, we have q1 = 2 yields n − q2. Substituting back into the first equation n(2n − 3)q2 2 + 2(3 − 2n)q2 + n − 2 = 0. We now have a quadratic equation in q2. Consider the discriminant D = −4(2n3 − 11n2 + 18n − 9). Weights theorem, for any admissible polynomial W with weights(cid:0) 1 When D < 0, we will not have a real-valued solution for q2. The above equation is a cubic polynomial that has roots at n = 1, 3 2 , 3. Since D < 0 for all n > 3, q2 will not be real-valued for all n > 3. Thus there are no rational-valued solutions for the quasihomogeneous weights in this case. ∼= BW T ,{0}. Extending by the Group- This shows that there is no W T in two variables satisfying AW (cid:48),G ∼= AW (cid:48),G. n , 1 By this isomorphism, we know that dim (AW,G) = 2n − 2 and the degree of its highest sector is 2(2n−4) . Therefore, by what we have just shown, there cannot not exist any W T in two variables such that AW,G We do have the following solutions for n ∈ {1, 2, 3}. n = 1 yields the solution q = (1, 1), n = 2 yields (cid:1), we have that AW,G solutions q = (1, 0), (0, 1), and n = 3 gives a solution q =(cid:0) 1 be in the interval (0, 1/2], q =(cid:0) 1 (cid:1) is the only valid weight system. (cid:1). However, since each coordinate must ∼= BW T ,{0}. This proves the theorem. (cid:3) 3 , 1 n n 3 3 , 1 3 6 system J =(cid:0) 1 Our original conjecture (Conjecture 1) about the transpose of a noninvertible polynomial was that W W )T = {0}. We will now state a corollary to and W T have the same number of variables and (Gmax demonstrate that one of these assumptions must be false. Corollary. For any n ∈ N, n > 3, let W be a noninvertible polynomial in two variables with weight W = (cid:104)J(cid:105). Then there does not exist a corresponding W T in two variables n , 1 satisfying AW,Gmax The proof follows from the fact that for W (cid:48) = xn + yn + xn−1y we have (cid:104)J(cid:105) = Gmax W (cid:48) Lemma. The polynomial W (cid:48) has Gmax (cid:1)(cid:105) for all n ∈ N, n ≥ 3. (cid:1) and Gmax ∼= BW T ,{0}. W (cid:48) = (cid:104)J(cid:105) = (cid:104)(cid:0) 1 n W . n , 1 n The proof of the lemma relies on a theorem due to Lisa Bendall. We will state Bendall's theorem here, and refer the reader to the Appendix for a proof. Theorem. Let W = xp + yq. If a monomial satisfying the quasihomogeneous weights of W is added to W (cid:48) = (cid:104)(1/p, 1/q), (1/n, 0)(cid:105), where n = gcd(p, r). make the new polynomial W (cid:48) = xp + yq + xrys, then Gmax Alternatively, Gmax (cid:16) (cid:1)(cid:105), since gcd(n, 1) = 1 and the generator (0, 1) ≡ (0, 0) mod 1 contributes nothing. (cid:3) W (cid:48) = (cid:104)(1/p, 1/q), (0, 1/m)(cid:105), where m = gcd(q, s). Proof of Lemma. By Lisa Bendall's theorem (see Appendix), Gmax W (cid:48) = (cid:104)(cid:0) 1 (cid:17)(cid:105) = (cid:104)(cid:0) 1 (cid:1) , n , 1 gcd(n,1) 0, n 1 Since W (cid:48) has Gmax that there does not exist a corresponding W T in two variables satisfying AW (cid:48),Gmax W (cid:48) by the Group-Weights theorem shows that any noninvertible W with weights J and Gmax have a W T in two variables satisfying the mirror symmetry alignment stated in the Corollary. W (cid:48) = (cid:104)J(cid:105), and since W (cid:48) satisfies the hypotheses of the previous theorem, we conclude ∼= BW T ,{0}. Extending W = (cid:104)J(cid:105) fails to n , 1 n 3.2 Evidence Against Conjecture 2 We will now consider finding a suitable W T in a different number of variables. By relaxing the constraint on the number of variables required in Conjecture 1, it is natural to make the following conjecture. Conjecture 2. For any admissible W , there is a corresponding admissible W T satisfying AW,Gmax BW T ,{0}. The following theorem is a start to disproving this conjecture. Theorem. For any admissible polynomial W with weight system J =(cid:0) 1 (cid:1) and G = (cid:104)J(cid:105), there is no corresponding admissible W T in 1, 2, or 3 variables satisfying AW,G Proof. For W as given in the hypothesis, we have previously shown that the degree of the A-model is 8, and the degree of its highest sector is 12/5. ∼= BW T ,{0}. 5 , 1 ∼= W 5 We will rule out the existence of a W T in these three cases. In one variable, we can only have W T = x9 to give us an unorbifolded B-model of dimension 8. Then q1 = 1 5 . The two variable case is done by the previous theorem. Now let n ∈ N, n ≥ 3. We have the following equations for a candidate weight system: 9 (cid:54)= 6 9 = 7 9 , but 1 − 2 (cid:19) (cid:35) = 8, − 1 = 12 5 . (cid:18) 1 qi (cid:19) n(cid:89) n(cid:88) i=3 i=3 (1 − 2q1) + (1 − 2q2) + (1 − 2qi) (cid:18) 1 (cid:19)(cid:18) 1 − 1 − 1 q2 2 q1 (cid:34) (cid:17) , and B = − n(cid:80) i=3 5n − 6 10 Letting A = 1 − (cid:16) 1 8 qi n(cid:81) i=3 − 1 (1) (2) (3) (4) qi, equations (1) and (2) simplify to Aq1q2 − q1 − q2 + 1 = 0, −q1 + B = q2. 7 For any qi ∈ (0, 1/2], we have that 1 (cid:16) 1 qi − 1 ≥ 1. By equation (1), we require that (cid:17) ≤ 8. Therefore we have that −7 ≤ A ≤ 0. qi − 1 From equation (2) we also have that 5 . Subtracting n − 2 from both sides yields − n(cid:80) (1 − 2qi) ≤ 6 i=3 qi ≤ 16−5n 10 . i=3 tells us that 1 ≤ n(cid:81) n(cid:80) qi ≤ 6 2 i=3 i=3 (cid:16) 1 qi n(cid:81) i=3 (cid:17) ≤ 8. This − 1 n(cid:80) 5 . Rewriting the left-hand side gives us (n − 2) − Substituting this into B gives us B = 5n − 6 10 − n(cid:88) i=3 qi ≤ 5n − 6 10 + 16 − 5n 10 = 1. Though we have developed the previous formulas in general, we will now restrict our attention to the case n = 3. When A (cid:54)= 0, we can use the quadratic formula to plot the real-valued solutions of q1. In three variables, the discriminant D = (AB)2 − 4A(B − 1) ≥ 0 for q3 ≤ 1/9. This yields the following: Figure 1: Positive solutions for q1 in the quadratic system (3) and (4) Figure 2: Negative solutions for q1 in the quadratic system (3) and (4) None of these values of q1 is in the interval (0, 1/2], let alone (0, 1/2] ∩ Q. −1 = 8. Therefore by equation (1) we can only have q1 = q2 = 1/2. Now when A = 0, we must have that 1 qi But equations (1) and (2) show that if this is the case, then we could have found a satisfactory weight system in just 1 variable without considering q1 and q2. Since we have already ruled out the case n = 1, we conclude that there are no valid weight systems for W T in three variables. (cid:3) The previous result casts doubt on the validity of Conjecture 2. Using the formulas developed in the last theorem may be useful in proving the following statement. Conjecture 3. For any admissible polynomial W with weight system J =(cid:0) 1 no corresponding admissible W T satisfying AW,G Proving Conjecture 3 will demonstrate that the mirror symmetry construction AW,Gmax not, in general, extend to noninvertible W . ∼= BW T ,{0}. W (cid:1) and G = (cid:104)J(cid:105), there is ∼= BW T ,{0} does 5 , 1 5 4 Conclusion Given a polynomial W fixed by a weight system J =(cid:0) 1 the number of variables in a candidate W T , it is impossible to construct AW,G n (cid:1) and group G = (cid:104)J(cid:105), and m ∈ N representing ∼= BW T ,{0} in the following n , 1 8 cases: m 2 X 1 4 3 . . . n 5 X X X 6 ... X X These results show that our original intuition about invertible polynomials and their transposes does not extend well to the noninvertible case. Even at the level of graded vector spaces, simply allowing an invertible polynomial to have one extra monomial seems to break this mirror symmetry construction. Though we have not completely ruled out the possibility of noninvertible polynomials having a transpose, we have shown that this problem is difficult and will require further research to fully elucidate it. 5 References [1] Michela Artebani, Samuel Boissi`ere, and Alessandra Sarti, The Berglund-Hubsch-Chiodo-Ruan mir- ror symmetry for K3 surfaces, Journal de Math´ematiques Pures et Appliqu´ees, vol. 102, no. 4 (2014), 758 -- 781. [2] P. Berglund and M. Henningson, Landau-Ginzburg orbifolds, mirror symmetry and the elliptic genus, Nucl. Phys. B, 433(2):311 -- 332, 1995. [3] P. Berglund and T. Hubsch, A generalized construction of mirror manifolds, Nucl. Phys. B 393 (1993) 377 -- 391. [4] Huijun Fan, Tyler J. Jarvis, and Yongbin Ruan, The Witten equation, mirror symmetry and quantum singularity theory, Annals of Mathematics, vol. 178, no. 1 (2013), 1 -- 106. [5] Amanda Francis, Tyler Jarvis, Drew Johnson, and Rachel Suggs, Landau-Ginzburg mirror symmetry for orbifolded Frobenius algebras, Proceedings of Symposia in Pure Mathematics, vol. 85 (2012), 333 -- 353. [6] K. Intriligator and C. Vafa, Landau-Ginzburg orbifolds, Nuclear Phys. B 339 (1990), no. 1, 95 -- 120. [7] R. Kaufmann, Singularities with symmetries, orbifold Frobenius algebras and mirror symmetry, Cont. Math. 403, 67-116. [8] R. Kaufmann, Orbifold Frobenius algebras, cobordisms and monodromies, Orbifolds in mathematics and physics (Madison, WI, 2001), 135 -- 161, Contemp. Math., 310, Amer. Math. Soc., Providence, RI, 2002. [9] R. Kaufmann, Orbifolding Frobenius algebras, Internat. J. Math. 14 (2003), no. 6, 573 -- 617. [10] Marc Krawitz, FJRW rings and Landau-Ginzburg mirror symmetry, PhD thesis, University of Michi- gan, 2010. [11] M. Kreuzer and H. Skarke, On the classification of quasihomogeneous functions, Comm. Math. Phys. 150 (1992), no. 1, 137 -- 147. MR 1188500 (93k:32075) [12] Julian Tay, Poincar´e polynomial of FJRW rings and the group-weights conjecture, Master's thesis, Brigham Young University, May 2013. http://contentdm.lib.byu.edu/cdm/singleitem/collection/ETD/id/3667/rec/1 9 6 Appendix We used the following result when proving the corollary in Section 3.1. The theorem and proof are due to Lisa Bendall. We will reproduce the entire proof here because it is not publicly available elsewhere. Theorem. Let W = xp + yq. If a monomial satisfying the quasihomogeneous weights of W is added to W (cid:48) = (cid:104)(1/p, 1/q), (1/n, 0)(cid:105), where n = gcd(p, r). make the new polynomial W (cid:48) = xp + yq + xrys, then Gmax Alternatively, Gmax Proof. Any element (θ1, θ2) ∈ Gmax W (cid:48) = (cid:104)(1/p, 1/q), (0, 1/m)(cid:105), where m = gcd(q, s). W (cid:48) must satisfy the following matrix equation: p 0 (cid:20)θ1 (cid:21) θ2 0 r q s =  ∈ Z3. k1 k2 k3 This yields the following three equations: θ1 = k1 p , θ2 = k2 1 , rθ1 + sθ2 = k3. p We also note that since the new monomial satisfies the weights vector of W , then r q = 1. From this equation, it follows that s = pq−rq , which we can substitute along with the first two equations into the last equation to get the equation r(k1 − k2) = p(k3 − k2). Dividing out by the gcd of r and p, we get the equation r(cid:48)(k1 − k2) = p(cid:48)(k3 − k2), where r(cid:48) and p(cid:48) are relatively prime. From this, we know that p(cid:48) (k1 − k2), or in other words, k1 − k2 = k4p(cid:48) for some k4 ∈ Z. Now, dividing both sides by p, we get k1 From the second equation, we already know that θ2 = k2 form: n . Next, substitute pθ1 for k1. We find that θ1 = k2 q , so we have the following equation in vector p − k2 (cid:16) 1 p = k4 p + s (cid:0) 1 n (cid:1). (cid:17) + k4 p (θ1, θ2) = k2(1/p, 1/q) + k4(1/n, 0), where k2, k4 ∈ Z. Now, to show that this generates the group, we show that anything of form k2(1/p, 1/q) + k4(1/n, 0) satisfies the three original equations for some three arbitrary integers. For the first equation, note that 1/n = p(cid:48)/p, thus θ1 = (k2 + k4p(cid:48))(1/p), so it is satisfied for some integer. The second equation follows immediately. For the final equation, plugging in we get r(k2/p + k4/n) + s(k2/q) = (r/p + s/q)k2 + (r/n)k4 = k2 + r(cid:48)k4 ∈ Z. Therefore, any element of the form k2(1/p, 1/q) + k4(1/n, 0) is in Gmax . Thus W (cid:48) W (cid:48) = (cid:104)(1/p, 1/q), (1/n, 0)(cid:105). Gmax Note: by substituting r = pq−sp and (0, 1/m). (cid:3) in the last equation, we get the alternate set of generators (1/p, 1/q) q 10
1108.0797
4
1108
2013-01-15T12:57:23
A simply connected numerical Campedelli surface with an involution
[ "math.AG", "math.GT" ]
We construct a simply connected minimal complex surface of general type with $p_g=0$ and $K^2=2$ which has an involution such that the minimal resolution of the quotient by the involution is a simply connected minimal complex surface of general type with $p_g=0$ and $K^2=1$. In order to construct the example, we combine a double covering and $\mathbb{Q}$-Gorenstein deformation. Especially, we develop a method for proving unobstructedness for deformations of a singular surface by generalizing a result of Burns and Wahl which characterizes the space of first order deformations of a singular surface with only rational double points. We describe the stable model in the sense of Koll\'ar and Shepherd-Barron of the singular surfaces used for constructing the example. We count the dimension of the invariant part of the deformation space of the example under the induced $\mathbb{Z}/2\mathbb{Z}$-action.
math.AG
math
A SIMPLY CONNECTED NUMERICAL CAMPEDELLI SURFACE WITH AN INVOLUTION HEESANG PARK, DONGSOO SHIN, AND GIANCARLO URZ ´UA ABSTRACT. We construct a simply connected minimal complex surface of general type with pg = 0 and K2 = 2 which has an involution such that the minimal resolution of the quotient by the involution is a simply connected minimal complex surface of general type with pg = 0 and K2 = 1. In order to construct the example, we combine a double covering and Q-Gorenstein deformation. Especially, we develop a method for proving unobstruct- edness for deformations of a singular surface by generalizing a result of Burns and Wahl which characterizes the space of first order deformations of a singular surface with only rational double points. We describe the stable model in the sense of Koll´ar and Shepherd- Barron of the singular surfaces used for constructing the example. We count the dimension of the invariant part of the deformation space of the example under the induced Z/2Z- action. 1. INTRODUCTION One of the fundamental problems in the classification of complex surfaces is to find a new family of complex surfaces of general type with pg = 0. In this paper we construct new simply connected numerical Campedelli surfaces with an involution, i.e. simply con- nected minimal complex surfaces of general type with pg = 0 and K2 = 2, that have an automorphism of order 2. There has been a growing interest for complex surfaces of general type with pg = 0 hav- ing an involution; cf. J. Keum-Y. Lee [13], Calabri-Ciliberto-Mendes Lopes [5], Calabri- Mendes Lopes-Pardini [6], Y. Lee-Y. Shin [18], Rito [24]. A classification of numerical Godeaux surfaces (i.e. minimal complex surfaces of general type with pg = 0 and K2 = 1) with an involution is given in Calabri-Ciliberto-Mendes Lopes [5]. It is known that the quotient surface of a numerical Godeaux surface by its involution is either rational or bira- tional to an Enriques surface, and the bicanonical map of the numerical Godeaux surface factors through the quotient map. However, the situation is more involved in the case of numerical Campedelli surfaces, because the bicanonical map may not factor through the quotient map; cf. Calabri-Mendes Lopes-Pardini [6]. In particular it can happen that the quotient is of general type. More precisely, let X be a numerical Campedelli surface with an involution σ. If σ has only fixed points and no fixed divisors, then the minimal resolution S of the quotient Y = X/σ is a numerical Godeaux surface and σ has only four fixed points; cf. Barlow [2]. Conversely, if S is of general type, then σ has only four fixed points and no fixed divisors; Calabri-Mendes Lopes-Pardini [6]. There are some examples of numerical Campedelli surfaces X with an involution σ having only four fixed points. Barlow [1] constructed examples with π1(X) = Z/2Z ⊕ Date: December 31, 2011; Revised at October 03, 2012. 2000 Mathematics Subject Classification. 14J29, 14J10, 14J17, 53D05. Key words and phrases. surface of general type, involution, Q-Gorenstein smoothing. 1 2 H. PARK, D. SHIN, AND G. URZ ´UA Z/4Z,Z/8Z. Barlow [2] also constructed examples with π1(X) = Z/5Z whose minimal resolution of the quotient by the involution is the first example of a simply connected nu- merical Godeaux surface. Also all Catanese's surfaces [7] have such an involution and π1 = Z/5Z. Recently Calabri, Mendes Lopes, and Pardini [6] constructed a numerical Campedelli surface with torsion Z/3Z⊕Z/3Z and two involutions. Frapporti [11] showed that there exists an involution having only four fixed points on the numerical Campedelli surface with Z/2Z⊕ Z/4Z constructed first in Bauer-Catanese-Grunewald-Pignatelli [3]. It is known that the orders of the algebraic fundamental groups of numerical Campedelli surfaces are at most 9 and the dihedral groups D3 and D4 cannot be realized. Recently, 1 ≤ 9 was settled by the existence question for numerical Campedelli surfaces with πalg 1 = Z/4Z; Frapporti [11] and H. Park-J. Park-D. the construction of examples with πalg Shin [23]. Hence it would be an interesting problem to construct numerical Campedelli 1 = G for each given group G with G ≤ 9. Espe- surfaces having an involution with πalg cially we are concerned with the simply connected case because the fundamental groups of all the known examples with an involution have large order: G ≥ 5. Furthermore the first example of simply connected numerical Campedelli surfaces is very recent (Y. Lee-J. Park [16]), but we have no information about the existence of an involution in their exam- ple. The main theorem of this paper is: Theorem (Corollary 3.6). There are simply connected minimal complex surfaces X of general type with pg(X) = 0 and K2 X = 2 which have an involution σ such that the minimal resolution S of the quotient Y = X/σ is a simply connected minimal complex surface of general type with pg(S) = 0 and K2 S = 1. We also show that the minimal resolution S of the quotient Y = X/σ has a local defor- mation space of dimension 4 corresponding to deformations S of S such that its general fiber St is the minimal resolution of a quotient Xt /σt of a numerical Campedelli surface Xt by an involution σt; Theorem 5.2. In addition, we show that the resolution S should be al- ways simply connected if the double cover X is already simply connected; Proposition 3.7. Conversely Barlow [2] showed that if the resolution S is a simply connected numerical Godeaux surface then the possible order of the algebraic fundamental group of the double cover X is 1, 3, 5, 7, or 9. As far as we know, the example in Barlow [2] was the only one whose quotient is simply connected. It has π1(X) = Z/5Z as mentioned earlier. Here we find an example with π1(X) = 1. Hence it would be an intriguing problem in this context to construct an example with π1(X) = Z/3Z. In order to construct the examples, we combine a double covering and a Q-Gorenstein smoothing method developed in Y. Lee-J. Park [16]. First we build singular surfaces by blowing up points and then contracting curves over a specific rational elliptic surface. These singular surfaces differ by contracting certain (−2)-curves. If we contract all of the (−2)-curves, we obtain a stable surface Y(cid:48) in the sense of Koll´ar -- Shepherd-Barron [15], and we prove that the space of Q-Gorenstein deformations of Y(cid:48) is smooth and 8 dimen- sional; Proposition 2.2. A (Q-Gorenstein) smoothing of Y(cid:48) in this space produces simply connected numerical Godeaux surfaces. In particular, the smoothing of Y(cid:48) gives the exis- tence of a two dimensional family of simply connected numerical Godeaux surfaces with six (−2)-curves; Corollary 2.3. We also prove that a four dimensional family in this space produces simply connected numerical Godeaux surfaces with a 2-divisible divisor consist- ing of four disjoint (−2)-curves; Theorem 2.4 and Theorem 5.2. These numerical Godeaux surfaces are used to construct the numerical Campedelli surfaces with an involution. The A NUMERICAL CAMPEDELLI SURFACE WITH AN INVOLUTION 3 desired numerical Campedelli surfaces are obtained by taking double coverings of the nu- merical Godeaux surfaces branched along the four disjoint (−2)-curves; Theorem 3.4. On the other hand we can also obtain the Campedelli family explicitly from a singular stable surface X(cid:48). It comes from blowing up points and contracting curves over a certain rational elliptic surface; Proposition 3.1. The Q-Gorenstein space of deformations of X(cid:48) is smooth and 6 dimensional; Proposition 3.3. In both Godeaux and Campedelli cases we compute H2(T ) = 0 to show no local-to-global obstruction to deform them; Theorem 2.1 and Theorem 3.2. This involves a new technique (Theorem 4.4) which generalizes a result of Burns-Wahl [4] describing the space of first order deformations of a singular complex surface with only rational double points. Notations. A cyclic quotient singularity (germ at (0,0) of) C2/G, where G = (cid:104)(x,y) (cid:55)→ (ζ x,ζ qy)(cid:105) with ζ a m-th primitive root of 1, 1 < q < m, and (q,m) = 1, is denoted by m (1,m− 1). A (−1)-curve (or (−2)-curve) in a smooth surface is an m (1,q). Am means 1 1 embedded CP1 with self-intersection −1 (respectively, −2). Throughout this paper we use the same letter to denote a curve and its proper transform under a birational map. A singularity of class T is a quotient singularity which admits a Q-Gorenstein one parameter dn2 (1,dna− 1) with 1 < a < n and smoothing. They are either rational double points or (n,a) = 1; see Koll´ar -- Shepherd-Barron [15, §3]. For a normal variety X its tangent sheaf TX is H omOX (Ω1 X , OX ). The dimension of Hi is hi. 1 Acknowledgements. The authors would like to thank Professor Yongnam Lee for helpful discussion during the work, careful reading of the draft version, and many valuable com- ments. The authors also wish to thank Professor Jenia Tevelev for indicating a mistake in an earlier version of this paper, and the referee especially for the remark on the proof of Proposition 3.7 which makes it simpler. Heesang park was supported by Basic Science Re- search Program through the National Research Foundation of Korea (NRF) grant funded by the Korean Government (2011-0012111). Dongsoo Shin was supported by Basic Sci- ence Research Program through the National Research Foundation of Korea (NRF) grant funded by the Korean Government (2010-0002678). Giancarlo Urz´ua was supported by a FONDECYT Inicio grant funded by the Chilean Government (11110047). 2. NUMERICAL GODEAUX SURFACES WITH A 2-DIVISIBLE DIVISOR 82 (1,8 · 5 − 1), 1 In this section we construct a family of simply connected numerical Godeaux surfaces having a 2-divisible divisor consisting of four disjoint (−2)-curves by smoothing a singu- lar surface (cid:101)Y ; Theorem 2.4. This is the key to construct numerical Campedelli surfaces face(cid:101)Y . In fact, we construct a rational normal projective surface Y(cid:48) with four singularities with an involution. In addition, we describe the explicit stable model of the singular sur- 72 (1,7 · 4 − 1) and KY(cid:48) ample. Hence Y(cid:48) is a stable surface (cf. A3, A3, 1 Koll´ar-Shepherd-Barron [15], Hacking [12]). We will prove that the versal Q-Gorenstein deformation space DefQG(Y(cid:48)) (cf. Hacking [12, §3]) is smooth and 8 dimensional, and that the Q-Gorenstein smoothings of Y(cid:48) are simply connected numerical Godeaux surfaces. In particular, this shows that there are simply connected numerical Godeaux surfaces whose canonical model has precisely two A3 singularities; Corollary 2.3. Furthermore a four di- mensional family in DefQG(Y(cid:48)) produces the above simply connected numerical Godeaux surfaces with a 2-divisible divisor consisting of four disjoint (−2)-curves; Theorem 2.4 and Theorem 5.2. 4 H. PARK, D. SHIN, AND G. URZ ´UA 2.1. A rational elliptic surface E(1). We start with a rational elliptic surface E(1) with an I8-singular fiber, an I2-singular fiber, and two nodal singular fibers. In fact we will use the same rational elliptic surface E(1) in the papers H. Park-J. Park-D. Shin [21, 22]. However, we need to sketch the construction of E(1) to show the relevant curves that will be used to build the singular surfaces(cid:101)Y and Y(cid:48). Let L1, L2, L3, and (cid:96) be lines in CP2 and let c be a smooth conic in CP2 given by the following equations. They intersect as in Figure 1. L1 : 2y− 3z = 0, L2 : y + (cid:96) : x = 0, c : x2 + (y− 2z)2 − z2 = 0. √ 3x = 0, L3 : y− √ 3x = 0 FIGURE 1. A pencil of cubic curves We consider the pencil of cubics {λ (y− √ 3x)(y + √ 3x)(2y− 3z) + µx(x2 + (y− 2z)2 − z2) [λ : µ] ∈ CP1} √ generated by the two cubic curves L1 + L2 + L3 and (cid:96) + c. This pencil has four base points √ p, q, r, s, and four singular members corresponding to [λ : µ] = [1 : 0], [0 : 1], [2 : 3 3], [2 : −3 tively. They have nodes at [−√ 3]. The latter two singular members are nodal curves, denoted by F1 and F2 respec- √ 3 : 0 : 1], respectively. 3 : 0 : 1] and [ In order to obtain a rational elliptic surface E(1) from the pencil, we resolve all base points (including infinitely near base-points) of the pencil by blowing-up 9 times as fol- lows. We first blow up at the points p, q, r, s. Let e1, e2, e3, e4 be the exceptional divisors over p, q, r, s, respectively. We blow up again at the three points e1∩L3, e2∩L2, e3∩ (cid:96). Let e5, e6, e7 be the exceptional divisors over the intersection points, respectively. We finally blow up at each intersection points e5 ∩ c and e6 ∩ c. Let e8 and e9 be the exceptional di- visors over the blown-up points. We then get a rational elliptic surface E(1) = CP2(cid:93)9CP2 over CP1; see Figure 2. The four exceptional curves e4, e7, e8, e9 are sections of the elliptic fibration E(1), which correspond to the four base points s, r, p, q, respectively. The elliptic fibration E(1) has one I8-singular fiber ∑8 i=1 Bi containing all Li (i = 1,2,3): B2 = L2, B3 = L3, and B5 = L1; cf. Figure 2. We will use frequently the sum B = B1 + B2 + B3 + B4, which will be shown to be 2-divisible. The surface E(1) has also one I2-singular fiber consisting of (cid:96) and c, and it has two more nodal singular fibers F1 and F2. There is a special bisection on E(1). Let M be the line in CP2 passing through the point √ r = [0 : 0 : 1] and the two nodes [−√ 3 : 0 : 1] of F1 and F2. Since M meets 3 : 0 : 1] and [ every member in the pencil at three points but it passes through only one base point, the proper transform of M is a bisection of the elliptic fibration E(1) → CP1; cf. Figure 2. Note that M2 = 0 in E(1). bbbbpsqrℓcL1L2L3 A NUMERICAL CAMPEDELLI SURFACE WITH AN INVOLUTION 5 FIGURE 2. A rational elliptic surface E(1) A 2-divisible divisor on E(1). Let h ∈ Pic(E(1)) be the class of the pull-back of a line in CP2. We denote again by ei ∈ Pic(E(1)) the class of the pull-back of the exceptional divisor ei. We have the following linear equivalences of divisors in E(1): B1 ∼ e2 − e6, B4 ∼ e1 − e5, B7 ∼ e3 − e7, F2 ∼ 3h− e1 −···− e9, B2 ∼ h− e2 − e3 − e6, B3 ∼ h− e1 − e3 − e5, B5 ∼ h− e1 − e2 − e4, B6 ∼ e6 − e9, B8 ∼ e5 − e8, (cid:96) ∼ h− e3 − e4 − e7. F1 ∼ 3h− e1 −···− e9, Let L := h− e3 − e5 − e6. Note that the divisor B = B1 + B2 + B3 + B4 is 2-divisible because of the relation B = B1 + B2 + B3 + B4 ∼ 2(h− e3 − e5 − e6) = 2L. (2.1) 2.2. A rational surface Z = E(1)(cid:93)7CP2. In the construction of Z, we use only one section S := e4. We first blow up at the two nodes of the nodal singular fibers F1 and F2 so that we obtain a blown-up rational elliptic surface W = E(1)(cid:93)2CP2; Figure 3. Let E1 and E2 be the exceptional curves over the nodes of F1 and F2, respectively. We further blow up at each three marked points • in Figure 3, and we blow up twice at the marked point(cid:74) (that is, we first blow-up(cid:74) and then again on the intersection point of the section and the exceptional curve; see Figure 4). We then get Z = E(1)(cid:93)7CP2 as in Figure 4. There exist two linear chains of the CP1 in Z whose dual graphs are given by: C7,4 = −2◦ − −6◦ − −2◦ − −3◦ , −2◦ − −3◦ − −5◦ − −3◦ , C8,5 = where C8,5 consists of (cid:96), S, F1, E1, and C7,4 contains F2, E2, M. 2.3. Numerical Godeaux surfaces. We construct rational singular surfaces which pro- duce under Q-Gorenstein smoothings simply connected surfaces of general type with pg = 0 and K2 = 1. have a normal projective surface(cid:101)Y with two singularities p1, p2 of class T : 1 We first contract the two chains C8,5 and C7,4 of CP1's from the surface Z so that we 72 (1,7· 4− 1). Denote the contraction morphism by (cid:101)α : Z →(cid:101)Y . Let Y be the surface 82 (1,8· 5− obtained by contracting the four (−2)-curves B1, . . . ,B4 in(cid:101)Y . We denote the contraction morphism by α : (cid:101)Y → Y . Then Y is also a normal projective surface with singularities 1), 1 B1B2B3B4B5B6B7B8ℓcF1F2e4e9e7e8M 6 H. PARK, D. SHIN, AND G. URZ ´UA FIGURE 3. A blown-up rational elliptic surface W = E(1)(cid:93)2CP2 FIGURE 4. A rational surface Z = E(1)(cid:93)7CP2 1 72 (1,7· 4− 1), and two A3 = 1 p1, p2 from(cid:101)Y , and four A1's (ordinary double points), denoted by q1, . . . ,q4. We finally contract C8,5, C7,4, B4 + B8 + B3 and B1 + B6 + B2 in Z to obtain Y(cid:48). It has the singularities 82 (1,8· 5− 1), 1 4 (1,3)'s. Let α(cid:48) : Z → Y(cid:48) be the contraction. of the singular surfaces(cid:101)Y , Y and Y(cid:48) vanish. That is: In Section 4 we will prove that the obstruction spaces to local-to-global deformations Theorem 2.1. H2((cid:101)Y , T(cid:101)Y ) = 0, H2(Y, TY ) = 0, and H2(Y(cid:48), TY(cid:48)) = 0. The singular surface Y(cid:48) is the stable model of the singular surfaces(cid:101)Y and Y : Y(cid:48) = 1, pg(Y(cid:48)) = 0, and KY(cid:48) is ample. The space Proposition 2.2. The surface Y(cid:48) has K2 DefQG(Y(cid:48)) is smooth and 8 dimensional. A Q-Gorenstein smoothing of Y(cid:48) is a simply connected canonical surface of general type with pg = 0 and K2 = 1. Proof. For a surface Y(cid:48) with only singularities of type T we have µp, Y(cid:48) = K2 K2 sp − ∑ Z + ∑ p∈Sing(Y(cid:48)) p∈Sing(Y(cid:48)) B1B2B3B4B5B6B7B8ℓcF1F2E1E2SMbbbb×2−3−5−3−2−2−6−2−3−1−1−1−1B1B2B3B4B6B8 A NUMERICAL CAMPEDELLI SURFACE WITH AN INVOLUTION 7 where sp is the number of exceptional curves over p and µp is the Milnor number of p. In Y(cid:48) = −7 + 4 + 4 = 1. We have pg(Y(cid:48)) = q(Y(cid:48)) = 0 because of the rationality of our case, K2 the singularities. We now compute α(cid:48)∗(KY(cid:48)) in a Q-numerically effective way. Let F be the general fiber of the elliptic fibration in Z. Let E3, E4, E5 be the exceptional curves over M∩ E1, F1 ∩ E1, 7 = −1. F2 ∩ E2 respectively. Let E6, E7 be the exceptional curves over F2 ∩ S with E2 Then, KZ ∼ −F +∑7 i=1 Ei + E3 + E4 + E5 + E7. We also have F ∼ F1 + 2E1 + 2E3 + 3E4 ∼ F2 + 2E2 + 3E5 + E6 + E7. Writing F ≡ 1 2 F in KZ and adding the discrepancies from p1 and p2, we obtain 2 F + 1 α(cid:48)∗(KY(cid:48)) ≡3 8 5 14 E4 + F1 + 1 2 5 8 E5 + F2 + 1 2 5 7 E6 + E1 + 13 14 E2 + E3 3 2 E7 + + 4 7 M + (cid:96) + S. 3 8 6 8 We now intersect α(cid:48)∗(KY(cid:48)) with all the curves in its support, which are not contracted by α(cid:48), to check that KY(cid:48) is nef. Moreover, if Γ.α(cid:48)∗(KY(cid:48) ) = 0 for a curve Γ not contracted by α(cid:48), then Γ is a component of a fiber in the elliptic fibration which does not intersect any curve in the support. This is because F1, E1, E3, and E4 belong to the support, and they are the components of a fiber. One easily checks that Γ does not exist, proving that KY(cid:48) is ample. Therefore any Q-Gorenstein smoothing of Y(cid:48) over a (small) disk will produce canonical surfaces; cf. Koll´ar-Mori [14, p. 34]. To compute the fundamental group of a Q-Gorenstein smoothing we use the recipe in Y. Lee-J. Park [16]. We follow the argument as in Y. Lee-J. Park [16, p. 493]. Consider the normal circles around M and E1. We can compare them through the transversal sphere E3. Since the orders of the circles are 49 and 64, which are coprime, we obtain that both end up being trivial. The smoothness of DefQG(Y(cid:48)) follows from Theorem 2.1 and Hacking [12, §3]. To compute the dimension, we observe that if Y (cid:48) → ∆ is a Q-Gorenstein smoothing of Y (cid:48) 0 = Y(cid:48) and TY (cid:48)∆ is the dual of Ω1 t (tangent bundle of t ) when t (cid:54)= 0, and TY (cid:48)∆Y (cid:48) with cokernel supported at the singular points of Y (cid:48) Y (cid:48) 0 ; cf. Wahl [26]. Then the flatness of TY (cid:48)∆ and semicontinuity in cohomology plus the fact that H2(Y(cid:48), TY(cid:48)) = 0 gives H2(Y (cid:48) t ) = 0 for any t. But then, since Y (cid:48) is of general type, Hirzebruch-Riemann-Roch Theorem says Y (cid:48)∆, then TY (cid:48)∆ restricts to Y (cid:48) ⊂ TY (cid:48) t as TY (cid:48) t , TY (cid:48) 0 0 t H1(Y (cid:48) t , TY (cid:48) t ) = 10χ(Y (cid:48) t , OY (cid:48) t )− 2K2 Y (cid:48) t This proves the claim. = 10− 2 = 8. (cid:3) Corollary 2.3. There is a two dimensional family of simply connected canonical numerical Godeaux surfaces with two A3 singularities. Proof. We consider the sequence 0 → H1(Y(cid:48), TY(cid:48)) → T 1 QG,Y(cid:48) → H0(Y(cid:48), T 1 QG,Y(cid:48)) → 0 at the end of Hacking [12, §3]. We just proved that T 1 QG,Y(cid:48) is 8 dimensional, and we know that H0(Y(cid:48), T 1 QG,Y(cid:48) ) is 8 dimensional, since each A3 gives 3 dimensions and each pi gives 1 dimension. Therefore H1(Y(cid:48), TY(cid:48)) = 0. To produce the claimed family we need to smooth (cid:3) up at the same time p1 and p2. 8 H. PARK, D. SHIN, AND G. URZ ´UA A simply connected numerical Godeaux surfaces with a 2-divisible divisor consisting of four disjoint (−2)-curves is obtained from a Q-Gorenstein smoothing of the singular surface(cid:101)Y : Theorem 2.4. (a) There is a Q-Gorenstein smoothing (cid:102)Y → ∆ over a disk ∆ with central fiber (cid:102)Y0 =(cid:101)Y and an effective divisor B ⊂ (cid:102)Y such that the restriction to a fiber(cid:101)Yt over Bt := B ∩(cid:101)Yt = B1,t + B2,t + B3,t + B4,t is 2-divisible in(cid:101)Yt consisting of four disjoint (−2)-curves and B0 = B. minimal resolution of Yt is the corresponding fiber(cid:101)Yt of (cid:102)Y → ∆. (b) There is a Q-Gorenstein deformation Y → ∆ of Y with central fiber Y0 = Y such that a fiber Yt over t (cid:54)= 0 has four ordinary double points as its only singularities and the t ∈ ∆ Proof. We apply a similar method in Y. Lee-J. Park [17]. Since any local deformations of the singularities of Y can be globalized by Theorem 2.1, there are Q-Gorenstein deforma- tions of Y over a disk ∆ which keep all four ordinary double points and smooth up p1 and p2. Let Y → ∆ be such deformation, with Y as its central fiber, and Yt (t (cid:54)= 0) a normal projective surface with four A1s as its only singularities. We resolve simultaneously these four singularities in each fiber Yt. We then get a family (cid:102)Y → ∆ that is a Q-Gorenstein smoothing of the central fiber(cid:101)Y , which shows that a Q-Gorenstein smoothing of Y can be lifted to a Q-Gorenstein smoothing of the pair (Y,B), i.e. the 2-divisible divisor B on(cid:101)Y is extended to an effective divisor B ⊂ (cid:102)Y . We finally show that the effective divisor Bt is 2-divisible in(cid:101)Yt for t (cid:54)= 0. According to Manetti [19, Lemma 2], the natural restriction map rt : Pic((cid:102)Y ) → Pic((cid:101)Yt ) is injective for every t ∈ ∆ and bijective for 0 ∈ ∆. Here we are using that pg((cid:101)Y ) = q((cid:101)Y ) = 0. Since that Bt ∼ 2Lt, where L is extended to a line bundle L ⊂ (cid:102)Y and Lt is the corresponding the divisor B is nonsingular, Bt is also nonsingular. Since B ∼ 2L in (2.1), it follows (cid:3) restriction. 3. NUMERICAL CAMPEDELLI SURFACES WITH AN INVOLUTION The main purpose of this section is to construct simply connected numerical Campedelli surfaces with an involution. Along the way, we will introduce a rational normal projective 72 (1,7·4−1)) and KX(cid:48) surface X(cid:48) with 6 singularities (two A1, two 1 ample. A certain four dimensional Q-Gorenstein deformation of X(cid:48) will produce numerical Campedelli surfaces with an involution. 82 (1,8·5−1), and two 1 Recall that the rational surface Z has a 2-divisible divisor B = B1 + B2 + B3 + B4; cf. (2.1). Let V be the double cover of Z branched along the divisor B, where the dou- ble cover is given by the data B ∼ 2L, L = h− e3 − e5 − e6. We denote the double covering by ψ : V → Z. The surface V has two C8,5's and two C7,4's. On the other hand the surface V can be obtained from a certain rational elliptic surface by blowing-ups, as we now explain. The morphism ψ : V → Z blows down to a double cover ψ(cid:48) : V(cid:48) → E(1) branched along B. The ramification divisor ψ(cid:48)−1(B) consists of four disjoint (−1)-curves R1, . . . ,R4. We blow down them from V(cid:48) to obtain a surface E(1)(cid:48); cf. Figure 5. In Figure 5 the pull-back of the Bi are the B(cid:48) i, of the curve (cid:96) is (cid:96)1 + (cid:96)2, of the section S is S1 + S2, and of the double section M is M1 + M2. Each I2 in E(1)(cid:48) is the pull-back of each I1 in E(1). Note that the surface E(1)(cid:48) has an elliptic fibration structure with two I4-singular fibers and two I2-singular fibers. In fact, the surface E(1)(cid:48) can be obtained from the pencil of A NUMERICAL CAMPEDELLI SURFACE WITH AN INVOLUTION 9 FIGURE 5. Relevant curves in the elliptic rational surface E(1)(cid:48) cubics in CP2 {λ x(y− z)(x− 2z) + µy(x− z)(y− 2z) [λ : µ] ∈ CP1} where the I4-singular fibers come from x(y−z)(x−2z) = 0 and y(x−z)(y−2z) = 0, and the I2-singular fibers come from (x +y−2z)(xy−yz−xz) = 0 and (x−y)(xy−xz +2z2−yz) = √−1)z and x = (1−√−1)z. In summary: 0. The two double sections M1 and M2 are defined by the lines x = (1 + Proposition 3.1. The surface V(cid:48) is the blow-up at four nodes of one I4-singular fiber of the rational elliptic fibration E(1)(cid:48) → CP1. Hence the surface V can be obtained from V(cid:48) by blowing-up in the obvious way. Let (cid:101)X be the double cover of the singular surface(cid:101)Y branched along the divisor B. Note that the surface (cid:101)X is a normal projective surface with four singularities of class T whose resolution graphs consist of two C8,5's and two C7,4's. The ramification divisor in (cid:101)X con- sists of the four disjoint (−1)-curves R1, . . . ,R4. Let(cid:101)φ :(cid:101)X →(cid:101)Y be the double covering. On the other hand the surface(cid:101)X can be obtained from the rational surface V by contracting the two C8,5's and two C7,4's. Let(cid:101)β : V →(cid:101)X be the contraction morphism. (cid:101)X. We denote the blowing-down morphism by β :(cid:101)X → X. Then there is a double covering Let X be the surface obtained by blowing down the four (−1)-curves R1, . . . ,R4 from φ : X → Y branched along the four ordinary double points q1, . . . ,q4. Finally, let X(cid:48) be the contraction of the (−2)-curves B(cid:48) 6 in X. Let β(cid:48) : V → X(cid:48) be the contraction. We then get a double covering X(cid:48) → Y(cid:48). To sum up, we have the following commutative diagram: 8 and B(cid:48) We will show in Section 4 the obstruction spaces to local-to-global deformations of the E(1) Z singular surfaces (cid:101)X, X and X(cid:48) vanish: Theorem 3.2. H2((cid:101)X, T(cid:101)X ) = 0, H2(X, TX ) = 0, and H2(X(cid:48), TX(cid:48)) = 0. The singular surface X(cid:48) is the stable model of (cid:101)X and X: E(1)(cid:48) V(cid:48) ψ(cid:48) V ψ (cid:101)β (cid:101)α / (cid:101)X (cid:101)φ /(cid:101)Y β α / X(cid:48) / X φ / Y / Y(cid:48) B′5B′6B′7B′8S1S2M1M2ℓ2ℓ1o o   o o   /   /   /   o o / / / 10 H. PARK, D. SHIN, AND G. URZ ´UA Proposition 3.3. The surface X(cid:48) has K2 X(cid:48) = 2, pg(X(cid:48)) = 0, and KX(cid:48) ample. The space DefQG(X(cid:48)) is smooth and 6 dimensional. A Q-Gorenstein smoothing of X(cid:48) is a simply connected canonical surface of general type with pg = 0 and K2 = 2. Proof. The proof goes as the one for Y(cid:48) in Proposition 2.2, using the explicit model we have for V by blowing-up E(1)(cid:48) in Proposition 3.1. One can check that an intersection (cid:3) computation as in Proposition 2.2 verifies ampleness for KX(cid:48). The proof of the next main result follows easily from Theorem 2.4. Theorem 3.4. There exist Q-Gorenstein smoothings (cid:102)X → ∆ of (cid:101)X and X → ∆ of X that are compatible with the Q-Gorenstein deformations of (cid:102)Y → ∆ of(cid:101)Y and Y → ∆ of Y in Theorem 2.4, respectively; that is, the double coverings (cid:101)φ :(cid:101)X →(cid:101)Y and φ : X → Y extend to the double coverings(cid:101)φt :(cid:101)Xt →(cid:101)Yt and φt : Xt → Yt between the fibers of the Q-Gorenstein deformations. Remark 3.5. By Theorem 3.2, the obstruction H2(X, TX ) to local-to-global deformations of the singular surface X vanishes. The point of the above theorem is that there is a Q- Gorenstein smoothing of the cover X that is compatible with the Q-Gorenstein deformation of the base Y . Corollary 3.6. A general fiber Xt of the Q-Gorenstein smoothing X → ∆ of X is a sim- ply connected numerical Campedelli surface with an involution σt such that the minimal resolution of the quotient Yt = Xt /σt is a simply connected numerical Godeaux surface. 3.1. The fundamental group of the quotient by an involution. Let X be a minimal complex surface of general type with pg = 0 and K2 = 2. Suppose that the group Z/2Z acts on X with just 4 fixed points. Let Y = X/(Z/2Z) be the quotient and let S → Y be the minimal resolution of Y . Barlow [2, Proposition 1.3] proved that if πalg 1 (S) = 1 then πalg 1 (X) = 1,3,5,7,9. Conversely: Proposition 3.7. If X is simply connected, then so is S. Proof. Let f : X → Y be the quotient map. Then f is a double covering which is branched along the four ordinary double points of Y . Let Y0 ⊂ Y be the complement of the four branch points (i.e., the four A1-singularities) of Y and let X0 = f −1(Y0), that is, X0 ⊂ X is the complement of the four fixed points of the involution σ. Then we get an ´etale double covering fX0 : X0 → Y0. Since π1(X0) = π1(X) = 1, we have π1(Y0) = Z/2Z. Note that the boundary ∂U of an arbitrary small neighborhood U of one of the four nodes of Y is a Lens space L(2,1). Let [γ] be a generator of π1(∂U) ∼= Z/2Z represented by a loop γ contained in ∂U. Since the lifting of γ by the covering fX0 : X0 → Y0 is not a closed path and π1(Y0) = Z/2Z, π1(Y0) is generated by [γ]. Then it follows by van Kampen theorem that π(Y ) is trivial. Hence π1(S) is trivial because S is obtained from Y by resolving only four A1-singularities. (cid:3) 4. THE OBSTRUCTION SPACES TO LOCAL-TO-GLOBAL DEFORMATIONS In this section we prove Theorem 2.1 which says that the obstruction spaces to local- to-global deformations of the singular surfaces(cid:101)Y , Y , and Y(cid:48) vanish. That is, we will prove that H2((cid:101)Y , T(cid:101)Y ) = H2(Y, TY ) = H2(Y(cid:48), TY(cid:48)) = 0. At the end, we also prove the analogues, Theorem 3.2, for (cid:101)X, X, and X(cid:48). A NUMERICAL CAMPEDELLI SURFACE WITH AN INVOLUTION 11 At first the vanishing of the obstruction spaces of a singular surface can be proved by the vanishing of the second cohomologies of a certain logarithmic tangent sheaf on the minimal resolution of the singular surface: Proposition 4.1 (Y. Lee-J. Park [16, Theorem 2]). If π : T → S be the minimal resolution of a normal projective surface S with only quotient singularities, and D is the reduced exceptional divisor of the resolution π, then h2(S, TS) = h2(T, TT (−logD)). Proposition 4.2 (Flenner-Zaidenberg [10, Lemma 1.5]). Let T be a nonsingular surface and let D be a simple normal crossing divisor in T . Let f : T(cid:48) → T be the blow-up of T at a point p of D. Let D(cid:48) = f ∗(D)red. Then h2(T(cid:48), TT(cid:48)(−logD(cid:48))) = h2(T, TT (−logD)). We can add or remove disjoint (−1)-curves. Proposition 4.3. Let T be a nonsingular surface and let D be a simple normal crossing divisor in T . Let E be a (−1)-curve in T such that D + E is again simple normal crossing. Then h2(T, TT (−log (D + E))) = h2(T, TT (−logD)). We can also add or remove disjoint exceptional divisors of rational double points. The following theorem may give a new general way to prove unobstructedness for deformations of surfaces. q1, . . . ,qn as singularities. Let π : (cid:101)S → S be the minimal resolution of S with exceptional Theorem 4.4. Let S be a normal projective surface with only rational double points C∩ M = ∅. Then h2((cid:101)S, T(cid:101)S(−log(C + M))) = h2((cid:101)S, T(cid:101)S(−logC)). i=1 π−1(qi). Let C be a simple normal crossing divisor such that reduced divisor M = ∑n Proof. Let M = ∑i Mi and C = ∑iCi be the prime decompositions of M and C. We have three short exact sequences: 0 → T(cid:101)S(−logM) → T(cid:101)S → ⊕N 0 → T(cid:101)S(−log(C + M)) → T(cid:101)S(−logM) → ⊕N 0 → T(cid:101)S(−log(C + M)) → T(cid:101)S(−logC) → ⊕N Mi/(cid:101)S → 0, Ci/(cid:101)S → 0, Mi/(cid:101)S → 0. We then have the following commutative diagram of cohomologies: 0 0 0 0 / H1(T(cid:101)S(−log(C + M))) H1(T(cid:101)S(−logC)) / H1(T(cid:101)S(−logM)) ⊕H1(N Ci/(cid:101)S) ξ H1(T(cid:101)S) ζ ⊕H1(N Ci/(cid:101)S) φ ψ ⊕H1(N Mi/(cid:101)S) ⊕H1(N Mi/(cid:101)S) / 0 Here all horizontal and vertical sequences are exact. Especially the second row is a short exact sequence, which we explain now briefly: It is shown in Burns-Wahl [4, pp. 70 -- 72] (see also Wahl [25, §6]) that the composition M(T(cid:101)S) → H1(T(cid:101)S) → ⊕H1(N Mi/(cid:101)S) H1     / / /   / /   / / /   / /   / 12 H. PARK, D. SHIN, AND G. URZ ´UA is an isomorphism because the qi's are rational double points; hence, one has a direct sum decomposition H1(T(cid:101)S) = H1(T(cid:101)S(−logM))⊕ H1 M(T(cid:101)S) and an isomorphism H2(T(cid:101)S) ∼= H2(T(cid:101)S(−logM)). Therefore the second row is exact. In order to prove the assertion, it is enough to show that (4.1) φ : H1(T(cid:101)S(−logC)) → ⊕H1(N Mi/(cid:101)S) β = γ + α(cid:48) composition H1 Mi/(cid:101)S). Since ψ is surjective, we have ψ(β ) = α for some M(T(cid:101)S) such that α(cid:48) is mapped to α under the Mi/(cid:101)S). Since α(cid:48) is supported on M and C∩M = Ci/(cid:101)S) vanishes. Therefore ζ (β − γ) = is surjective. Let α ∈ ⊕H1(N β ∈ H1(T(cid:101)S). By (4.1) we have for some γ ∈ H1(T(cid:101)S(−logM)) and α(cid:48) ∈ H1 M(T(cid:101)S) → H1(T(cid:101)S) → ⊕H1(N ∅, its image ζ (α(cid:48)) under ζ : H1(T(cid:101)S) → ⊕H1(N ζ (α(cid:48)) = 0, and so β − γ = ξ (δ ) for some δ ∈ H1(T(cid:101)S(−logC)); hence, φ (δ ) = ψ(ξ (δ )) = α, which shows that φ is sur- (cid:3) jective. Proposition 4.5 (cf. Esnault-Viehweg [9, 2.3]). Let E = ∑n ing divisor on a smooth surface T . Then one has the following exact sequences: (a) 0 → Ω1 (b) 0 → Ω1 Proof of Theorem 2.1. We first claim that T (logE) → n(cid:77) T → Ω1 T (logE) → Ω1 T (log(E −C1))(C1) → Ω1 i=1Ci be a simple normal cross- C1(EC1) → 0. OCi → 0. i=1 H2(W, TW (−log(F1 + F2))) = 0. By duality, we have to show that h2(W, TW (−log(F1 + F2))) = h0(W,Ω1 W (log(F1 + F2))(KW )) = 0. Since KW ∼ −F2 + E1 − E2, it follows by Proposition 4.5 that H0(W,Ω1 W (log(F1 + F2))(KW )) = H0(W,Ω1 = H0(W,Ω1 ⊂ H0(W,Ω1 = H0(W,Ω1 W (logF1)(KW + F2)) W (logF1)(E1 − E2)) W (logF1)(E1)) W (log(F1 + E1))). W (log(F1 +E1))) = 0. On the other hand, we obtain Hence it suffices to show that H0(W,Ω1 a long exact sequence from Proposition 4.5: δ−→ H1(W,ΩW ) H0(W,ΩW ) → H0(W,Ω1 Since H0(W,ΩW ) = 0, it is enough to show that the connecting homomorphism δ is injec- tive. Note the map δ is the first Chern class map. But F1 and E1 are linearly independent in the Picard group of W ; hence, the map δ is injective. Therefore the claim follows. W (log(F1 + E1))) → H0(F1, OF1)⊕ H0(E1, OE1) Let B(cid:48) = B4 + B8 + B3 + B1 + B6 + B2. By Theorem 4.4 we have h2(W, TW (−log (F1 + F2))) = h2(W, TW (−log (F1 + F2 + B(cid:48) + M + (cid:96)))). A NUMERICAL CAMPEDELLI SURFACE WITH AN INVOLUTION 13 We use Propositions 4.3 and 4.2 to obtain h2(W, TW (−log (F1 + F2 + B(cid:48) + M + (cid:96)))) = h2(Z, TZ(−log (F1 + F2 + B(cid:48) + M + (cid:96) + S + E(cid:48)))), where E(cid:48) = E1 + E2 + E6. In this way, it follows by the above claim that h2(Z, TZ(−log (F1 + F2 + B(cid:48) + M + l + S + E(cid:48)))) = h2(W, TW (−log (F1 + F2))) = 0. Then, by Proposition 4.1, we have H2(Y, TY(cid:48)) = 0. Notice we can modify B(cid:48) to obtain (cid:3) vanishing for H2((cid:101)Y , T(cid:101)Y ) and H2(Y, TY ) as well. We now prove that H2((cid:101)X, T(cid:101)X ) = H2(X, TX ) = H2(X(cid:48), TX(cid:48)) = 0. Proof of Theorem 3.2. We will use our explicit model of X(cid:48) in Proposition 3.1. The proof goes along the same lines as the proof of the above Theorem 2.1. We may only need to mention that we start with the elliptic fibration E(1)(cid:48), and 2 I2 fibers (instead of 2 I1's). (cid:3) 5. THE INVARIANT PART OF THE DEFORMATION SPACE βt αt Xt φt / Yt the four ramification points. We then have the following commutative diagram where the curves B1,t , . . . ,B4,t and the corresponding ramification divisor Rt consists of four disjoint The involution of a general fiber Xt induced by the double covering φt : Xt → Yt extends to a Z/2Z-action on the deformation space of Xt. We will count the dimension of the subspace of H1(Xt , TXt ) which is fixed by the Z/2Z-action; Theorem 5.2. Let αt :(cid:101)Yt → Yt be the minimal resolution and let βt : (cid:101)Xt → Xt be the blowing-up at vertical morphisms are double covers: (cid:101)Xt (cid:101)φt (cid:101)Yt Recall that the branch divisor Bt of the double covering (cid:101)φt consists of four disjoint (−2)- (−1)-curves R1,t , . . . ,R4,t. As before the involution of (cid:101)Xt induced by the double covering (cid:101)φt :(cid:101)Xt →(cid:101)Yt extends to a Z/2Z-action on the deformation space of (cid:101)Xt. Lemma 5.1. dimH1((cid:101)Xt , T(cid:101)Xt Proof. By Pardini [20, Lemma 4.2], the invariant part of ((cid:101)φt )∗T(cid:101)Xt under the Z/2Z-action is T(cid:101)Yt Z/2Z ∼= H1((cid:101)Yt , T(cid:101)Yt (−logBt )) ∼= H1((cid:101)Yt ,Ω1(cid:101)Yt H1((cid:101)Xt , T(cid:101)Xt (logBt )(K(cid:101)Yt We know that (cid:101)φt∗(Ω1(cid:101)Xt (K(cid:101)Yt (logBt )(K(cid:101)Yt (K(cid:101)Xt )) = Ω1(cid:101)Yt )⊕ Ω1(cid:101)Yt )) = H0((cid:101)φt∗(Ω1(cid:101)Xt the double cover (cid:101)φt. Therefore H2(T(cid:101)Xt (K(cid:101)Xt (K(cid:101)Xt ) = H0(Ω1(cid:101)Xt (logBt )(K(cid:101)Yt Theorem 3.3. Then we have H0(Ω1(cid:101)Yt )) → H1((cid:101)Yt ,Ω1(cid:101)Yt (logBt )(K(cid:101)Yt (K(cid:101)Yt + Lt ), where 2Lt ∼ Bt defines ))) = 0 by )) = 0. Hence, by Proposition 4.5, there is )) → H1((cid:101)Yt ,Ω1(cid:101)Yt (−logBt ). Therefore we have 0 → H0(Bt , OBt (K(cid:101)Yt a short exact sequence Z/2Z ) = 4. ) )). )) → 0. / /     / 14 H. PARK, D. SHIN, AND G. URZ ´UA Since Bt consists of four disjoint (−2)-curves, we have h0(Bt , OBt (K(cid:101)Yt tion 2.2, we know that h1((cid:101)Yt ,Ω1(cid:101)Yt )) = 8. Therefore h1((cid:101)Yt ,Ω1(cid:101)Yt (K(cid:101)Yt (logBt )(K(cid:101)Yt (cid:3) Theorem 5.2. The subspace of the deformation space of Xt invariant under the Z/2Z- action is four dimensional. Proof. We apply a similar strategy in Werner [27, §4]. We have the exact sequence )) = 4. By Proposi- )) = 4. (−logR) → T(cid:101)Xt Since each Ri,t is a (−1)-curve, we have H1((cid:101)Xt , T(cid:101)Xt hand, it follows by Catanese [8, Lemma 9.22] that → ⊕N Ri,t ,(cid:101)X → 0. ) = H1((cid:101)Xt , T(cid:101)Xt 0 → T(cid:101)Xt H1((cid:101)Xt , T(cid:101)Xt (5.1) where I is the ideal sheaf of the four points in Xt obtained by contacting the exceptional divisors R1,t , . . . ,R4,t. (−logR)) = H1(Xt , TXt ⊗ I ) Let P be the set of these four points. From the ideal sequence, we have 0 → H0(P, TXt ⊗ OP) → H1(Xt , TXt ⊗ I ) → H1(Xt , TXt ) → 0. Therefore the invariant parts of each space satisfies: Z/2Z → H1(Xt , TXt ⊗ I ) 0 → H0(P, TXt ⊗ OP) Z/2Z → H1(Xt , TXt ) Z/2Z → 0. (−logR)). On the other According to Werner [27, p. 1523], we have H0(P, TXt ⊗ OP) by (5.1) and Lemma 5.1 that dimH1(Xt , TXt ) = dimH1(Xt , TXt ⊗ I ) = dimH1((cid:101)Xt , T(cid:101)Xt Z/2Z Z/2Z Z/2Z = 0. Therefore it follows (−logR)) Z/2Z = 4. (cid:3) 6. ANOTHER EXAMPLE We briefly describe another rational surface Z which makes it possible to construct simply connected numerical Campedelli surfaces with an involution as before. The as- sociated Godeaux surfaces come from a rational surface Y(cid:48) with KY(cid:48) ample having three A1-singularities, one A3-singularity, and only one singularity of class T . 6.1. A rational surface Z = E(1)(cid:93)8CP2. The elliptic fibration E(1) is the one in Sec- tion 2. In the construction of Z, we will use the sections e4, e7, e8 among the four sections of E(1). We denote the sections e4, e7, e8 by S1, S2, S3, respectively. We first blow up at the two nodes of the nodal singular fibers F1 and F2 so that we obtain a blown-up rational elliptic surface W = E(1)(cid:93)2CP2; Figure 6. Let E1 and E2 be the exceptional curves over the nodes of F1 and F2, respectively. We further blow up at each two marked points • and blow up four times at the marked point(cid:74) in Figure 6. We then get a rational surface Z = E(1)(cid:93)8CP2; Figure 7. There exists one linear chain of CP1s in Z whose dual graph is −6◦ F1 − −1◦ S3 − −8◦ F2 − −2◦ S1 − −2◦ (cid:96) − −3◦ S2 − −2◦ E5 − −2◦ E6 − −2◦ . E7 Notice that the (−1)-curve S3 is contracted in the way down, which fixes the configuration so that we obtain one singular point of class T whose resolution graph is given by C24,5 := −5◦ − −7◦ − −2◦ − −2◦ − −3◦ − −2◦ − −2◦ − −2◦ . A NUMERICAL CAMPEDELLI SURFACE WITH AN INVOLUTION 15 The divisor B1 + B2 + B3 + B4 is the 2-divisible one as before. The C24,5 and the (−2)- curves B3, B4, B1 + B6 + B2, and M are contracted to obtain a singular surface Y(cid:48). One can use again Theorem 4.4 to show that the space DefQG(Y(cid:48)) is smooth (of dimension 8). FIGURE 6. A blown-up rational elliptic surface W = E(1)(cid:93)2CP2 FIGURE 7. A rational surface Z = E(1)(cid:93)8CP2 REFERENCES [1] R. Barlow, Some new surfaces with pg = 0, Duke Math. J. 51 (1984), 889 -- 904. [2] [3] I. Bauer, F. Catanese, F. Grunewald, R. Pignatelli, Quotients of products of curves, new surfaces with pg = 0 , A simply connected surface of general type with pg = 0, Invent. Math. 79 (1985), 293 -- 301. and their fundamental groups. Amer. J. Math. 134 (2012), no. 4, 993 -- 1049. [4] D. Burns, J. Wahl, Local contributions to global deformations of surfaces, Invent. Math. 26 (1974), 67 -- 88. [5] A. Calabri, C. Ciliberto, M. Mendes Lopes, Numerical Godeaux surfaces with an involution, Trans. Amer. Math. Soc. 359 (2007), no. 4, 1605 -- 1632. [6] A. Calabri, M. Mendes Lopes, R. Pardini, Involutions on numerical Campedelli surfaces, Tohoku Math. J. (2) 60 (2008), no. 1, 1 -- 22. [7] F. Catanese, Babbage's conjecture, contact of surfaces, symmetric determinantal varieties and applications, Invent. Math. 63 (1981), no. 3, 433 -- 465. [8] [9] H. Esnault, E. Viehweg, Lectures on vanishing theorems, DMV Seminar 20, Birkhauser Verlag. , Moduli of algebraic surfaces, Lecture Notes in Math. 1337, Springer, 1988. B1B2B3B4B6ℓcF1F2E1E2S1S3S2Mbbb×4B1B2B3B4B6ℓF1F2S1S3S2ME1E2E3E4E5E6E7E8−6−1−8−2−2−3−2−2−2−2−2−2−2−2−2−1−1−1−1−1 16 H. PARK, D. SHIN, AND G. URZ ´UA [10] H. Flenner, M. Zaidenberg, Q-acyclic surfaces and their deformations, Contemp. Math. 162 (1994), 143 -- 208. [11] D. Frapporti, Mixed surfaces, new surfaces of general type with pg = 0 and their fundamental group, arXiv:1105.1259. [12] P. Hacking, Compact moduli spaces of surfaces of general type, Contemp. Math. 564 (2012), 1 -- 18. [13] J. Keum, Y. Lee, Fixed locus of an involution acting on a Godeaux surface, Math. Proc. Cambridge Philos. Soc. 129 (2000), no. 2, 205 -- 216. [14] J. Koll´ar and S. Mori, Birational geometry of algebraic varieties, Cambridge Tracts in Mathematics, 134, Cambridge University Press, Cambridge, 1998. [15] J. Koll´ar, N. I. Shepherd-Barron, Threefolds and deformations of surface singularities, Invent. Math. 91 [16] Y. Lee, J. Park, A simply connected surface of general type with pg = 0 and K2 = 2, Invent. Math. 170 [17] , A construction of Horikawa surface via Q-Gorenstein smoothings, Math. Z. 267 (2011), no. 1-2, (1988), 299 -- 338. (2007), 483 -- 505. 15 -- 25. 89 -- 118. appear in Osaka J. Math. [18] Y. Lee, Y. Shin, Involutions on a surface of general type with pg = q = 0, K2 = 7, arXiv:1003.3595, to [19] M. Manetti, Normal degenerations of the complex projective plane, J. Reine Angew. Math. 419 (1991), [20] R. Pardini, Abelian covers of algebraic varieties, J. Reine Angew. Math. 417 (1991), 191 -- 213. [21] H. Park, J. Park, D. Shin, A simply connected surface of general type with pg = 0 and K2 = 3, Geom. Topol. 13 (2009), no. 2, 743 -- 767. [22] [23] 1483 -- 1494. appear in Trans. Amer. Math. Soc. , A simply connected surface of general type with pg = 0 and K2 = 4, Geom. Topol. 13 (2009), no. 3, , A complex surface of general type with pg = 0, K2 = 2 and H1 = Z/4Z, arXiv:1012.5871, to [24] C. Rito, Involutions on surfaces with pg = q = 0 and K2 = 3, Geom. Dedicata 157 (2012), 319 -- 330. [25] J. Wahl, Vanishing theorems for resolutions of surface singularities, Invent. Math. 31 (1975), 17 -- 41. [26] [27] C. Werner, A four-dimensional deformation of a numerical Godeaux surface, Trans. Amer. Math. Soc. 349 , Smoothings of normal surface singularities, Topology 20 (1981), no. 3, 219 -- 246. (1997), no. 4, 1515 -- 1525. SCHOOL OF MATHEMATICS, KOREA INSTITUTE FOR ADVANCED STUDY, SEOUL 130-722, KOREA E-mail address: [email protected] DEPARTMENT OF MATHEMATICS, CHUNGNAM NATIONAL UNIVERSITY, DAEJEON 305-764, KOREA & SCHOOL OF MATHEMATICS, KOREA INSTITUTE FOR ADVANCED STUDY, SEOUL 130-722, KOREA E-mail address: [email protected] FACULTAD DE MATEM ´ATICAS, PONTIFICIA UNIVERSIDAD CAT ´OLICA DE CHILE, SANTIAGO, CHILE. E-mail address: [email protected]
1604.02567
3
1604
2017-12-24T17:33:58
Quartic surfaces with icosahedral symmetry
[ "math.AG", "math.GR" ]
We study quartic surfaces that admit a group of projective automorphisms isomorphic to icosahedron group.
math.AG
math
QUARTIC SURFACES WITH ICOSAHEDRAL SYMMETRY IGOR V. DOLGACHEV ABSTRACT. We study smooth quartic surfaces in P3 which admit a group of projective automor- phisms isomorphic to the icosahedron group. 1. INTRODUCTION Let A5 be the icosahedron group isomorphic to the alternating group in 5 letters. Starting from Platonic solids, it appears as an omnipresent symmetry group in geometry. In this article, complementing papers [4], [5], we discuss families of smooth quartic surfaces in P3 that admit the group A5 as its group of projective symmetries. It follows from loc.cit. that any smooth quartic surface S with a faithful action of A5 belongs to one of the two pencils of invariant quartic surfaces. One of them arises from a linear irreducible 4-dimensional representation of A5 and was studied with great details by K. Hashimoto in [10]. It contains a double quadric and four surfaces with 5,10,10, or 15 ordinary double points. The other pencil arises from a faithful linear representation of the binary icosahedron group 2.A5. As was shown in [4] and [5], it contains two singular surfaces singular along its own rational normal cubic and two surfaces with 10 ordinary double points. In this paper we show first that one of the surfaces with 10 ordinary double points from Hashimoto's pencil can be realized as a Cayley quartic symmetroid defined by a A5-invariant web W of quadrics.1 We also show that the Steinerian surface of this web parametrizing singular points of singular quadrics in the web coincides with one of two smooth member of the second pencil that admits a larger group of projective symmetries isomorphic to S5. To see this we give an explicit equation of the second pencil. We also show that the apolar linear system of quadrics to the web W contains two invariant rational normal curves that give rise to two rational plane sextics with symmetry A5 discovered by R. Winger [20], [21]. It is my pleasure to thank I. Cheltsov, B. van Geemen, K. Hulek and S. Mukai for their help in collecting the known information about the subject of this paper. I would like also to thank the referee for careful reading of the manuscript and useful remarks. 2. TWO IRREDUCIBLE 4-DIMENSIONAL REPRESENTATIONS Let E be a 4-dimensional linear space over an algebraically closed field k of characteristic 6= 2, 3, 5 and E ∼= P3 be the projective space of lines in E. Assume that E is a non-trivial projective linear representation space for A5 ∼= PSL(2, F5) in E. Then it originates from either linear representation of A5 in E, or from a linear representation of its central double extension 2.A5 ∼= SL(2, F5), the binary icosahedral group. Then E is either an irreducible representation, or contains one-dimensional trivial representation, or decomposes into the sum of two irreducible two-dimensional representations. 1This result was independently obtained by S. Mukai. 1 2 IGOR V. DOLGACHEV Proposition 1. Suppose there exits a smooth quartic surface S in E which is invariant with respect to a non-trivial projective representation of A5. Then E is an irreducible representation of G = A5 or G = 2.A5. Proof. Suppose E has one-dimensional trivial summand. It is known that an element of order 3 acting on a smooth quartic surface has 6 isolated fixed points (see [18]). A glance at the character table of the groups shows that an element g of order 3 in G in its action on the 3-dimension summand has three different eigenvalues, one of them is equal to 1. This shows that g has a pointwise fixed line in E and two isolated fixed points. The line intersects the quartic at 4 points, so all the fixed points are accounted for. In particular, one of them is fixed with respect to the whole group A5. However, the assumption on the characteristic implies that G acts faithfully on the tangent space of S at this point, and the group A5 has no non-trivial 2-dimensional linear representations. Now, let us consider case when E decomposes into the sum of two irreducible 2-dimensional representations. This could happen only when G = 2.A5. In this case G has two invariant lines in E, hence the union of two invariant sets of ≤ 4-points on S. The known possible orders of subgroups of A5 shows that this is impossible. (cid:3) So, we are dealing with projective representations of the group A5 coming from an irreducible four-dimensional representation W4 of A5 or from an irreducible faithful 4-dimensional linear representation U4 of 2A5. It is known that the group 2.A5 has two irreducible 2-dimensional representation V and V′, both self-dual but one is obtained from another by the composition with an outer automorphism of A5. The center acts as the minus identity. Since it acts as the identity on the symmetric squares S2V or S2V′, the groups A5, 2.A5 admit two 3-dimensional representations, both of them are irreducible and self-dual, they differ from each other by an outer automorphism. The group A5 acts in S2V∨ via the Veronese map of V → S2V of its natural action on V ∼= P1. The restriction of the representation S2V of A5 to its subgroup H of order 10 isomorphic to the dihedral group D10 has trivial one-dimensional summand. It defines a fixed point of H in its action in the plane S2V. Its orbit of A5 consists of six points, called by F. Klein, the fundamental points. The linear system of plane cubics is invariant with respect to A5 and defines an irreducible subrepresentation of S3(S2V)∨ of dimension 4. The image of the plane under the map defined by the linear system of cubics is isomorphic to the Clebsch diagonal cubic surface representing the unique isomorphism class of a nonsingular cubic surface with the automorphism group isomorphic to S5 [9], 9.5.4. The representation W4 is isomorphic to the restriction of the standard irreducible representation of S5 realized in the hyperplane x1 + ··· + x5 = 0 in k5. The representation U4 is realized in the third symmetric power S3V of V. It is self-dual and isomorphic to S3V′. The projective representation of A4 in U4 is obtained via the Veronese map V → S3V from the natural action of A5 on V. Using the character table for G one obtains the following. Proposition 2. Let S4W∨ and S4U4 be the fourth symmetric power of W4 and of U4, and let ()G denote the subspace of G-invariant elements. Then dim(S4W∨)G = dim(S4U∨4 )G = 2. Thus we have two pencils of invariant quartic polynomials (S4W4)A5 and (S4U4)2.A5, so our quartic surface S is a member of one of them. QUARTIC SURFACES 3 3. THE PENCIL (S4W∨4 )A5 Since W4 is the restriction of the standard representation of S5, the invariant theory says that the space (S4W4)A5 is isomorphic to the linear space of symmetric polynomials of degree 4 in variables x1, . . . , x5 and the discriminant anti-symmetric polynomial. Since the degree of the latter is larger than 4, we obtain that (S4W4)A5 = (S4W4)S5. Thus any A5-invariant pencil of quartic surfaces in W4 consists of the quartics given by equations (3.1) s4 + λs2 2 = s1 = 0, where si are elementary symmetric functions. It will be more convenient to rewrite this equation in terms of power symmetric functions: x4 1 + ··· + x4 6 − t(x2 1 + ··· + x2 6)2 = x1 + ··· + x6 = 0. where t = (2λ + 1)/2. (3.2) Computing the partial derivatives, we find that the S5-orbits of singular points must be rep- resented by points with coordinates (1, 1,−1,−1, 0), or (1,−1, 0, 0, 0), or (2, 2, 2,−3,−3), or (1, 1, 1, 1,−4) corresponding to the parameters t = 1/4, 1/2, 7/30, 13/20, respectively. This gives the following. 30 , 13 Proposition 3. The surface St given by equation (3.2) is nonsingular if and only if t 6= 1 20 . If t = 1/4, it has 15 singular points, if t = 1/2, 7/30, it has 10 singular points, if t = 13/20, it has 5 singular points. Each singular point is an ordinary node. 4 , 7 2 , 1 Example 1. The surface S1/4 is the intersection of the Castelnuovo-Richmond-Igusa quartic threefold (see [9], 9.4.4) given by equations in P5 1 + ··· + x4 x4 6 − 1 4 1 + ··· + x2 (x2 6)2 = x1 + ··· + x6 = 0. (3.3) and the hyperplane x6 = 0. Its singular points are the intersection of the hyperplane with the fifteen double lines of the threefold. Example 2. If t = 1/2, then λ = 0 and the surface can be rewritten by the equation 5 Xi=1 1 xi = x1 + ··· + x5 = 0. in P4. We recognize the equation of the Hessian surface of the Clebsch diagonal cubic surface C3. Its 10 nodes are the vertices of the Sylvester pentahedron. Its edges lie on the surface and together with nodes form the Desargues symmetric configuration (103) (see [11], III, §19). By definition of the Hessian surface, S1/2 is a quartic symmetroid (see [9], 4.2.6), the locus of points in E such that the polar quadric of the cubic surface C3 is singular. This symmetric determinantal representation of S1/2 is defined by a linear map of S5-representations W4 → S2W∨4 . It is the polar map associated with the cubic surface (see [9]). It allows one to rewrite the equation of the Hessian of the Clebsch cubic surface as the symmetric determinant: det  L − x1 L L L where L = x1 + x2 + x3 + x4. L L − x2 L L L L L − x3 L L L L L − x4   = 0, 4 IGOR V. DOLGACHEV It is known that the surface S1/2 is isomorphic to the K3-cover of an Enriques surface X with Aut(X) ∼= S5 (of type VI in Kondo's classification, [16]). The covering involution is defined by the standard Cremona transformation (x1, . . . , x5) 7→ (x−1 Example 3. Assume t = 7/30. Projecting from the node q0 = (2, 2, 2,−3,−3), we get the equation of S7/30 as the double plane 1 , . . . , x−1 5 ). w2 = (xy + xz + yz)2(x + y + z)2 − 3(x − y)2(x − z)2(y − z)2 = 111x2y2z2 + 80(x3y2z + x3yz2 + x2y3z + x2yz3 + xy3z2 + xy2z3) +13(x4y2 + x4z2 + y4z2 + x2z4 + y2z4 + x2y4)+ 10(x4yz + x3y3 + x3z3 + xy4z + xyz4 + y3z3). The branch curve B is the union of two cubic curves intersecting at 9 points [1, 0, 0], [0, 1, 0], [0, 0, 1], [2,−1, 2], [−1, 2, 2], [2, 2,−1], [−2, 1, 1], [1,−2, 1], [1, 1,−2]. It is well-known and goes back to A. Cayley (see a modern exposition in [8]) that this implies that the surface S7/30 is a quartic symmetroid. We will return to this example in the next section. Example 4. Let S3 be the Segre cubic primal (see [9], 9.4.4) given by equations x3 1 + ··· + x3 6 = x1 + ··· + x6 = 0 (3.4) (see [9]). It is isomorphic to the image of a rational map f : P3 99K P4 given by the linear system of quadrics through 5 points p1, . . . , p5 in general linear position. The images of the lines hpi, pji are the ten nodes on S3. Consider the surface S in S3 given by the additional equation 1 + ··· + x2 x2 6 = 0. Obviously, it has S6-symmetry. The pre-image of S under the map f is a quartic surface in P3 with 5 nodes at p1, . . . , p5. One can make the map f to be S5-invariant by viewing P3 as the hyperplane y0 + . . . + y5 = 0 in P4 and choosing the points p1, . . . , p5 to be the points in the hyperplane with coordinates [1, 0, 0, 0,−1], [0, 1, 0, 0, −1], etc. The group S5, acts naturally in P3 by permuting the five points. The restriction of this representation to A5 is the projectivization of the linear representation space W4. Via this action, the map f becomes a S5-invariant birational map from P3 to the Segre cubic C3. Thus f−1(S) is a 5-nodal quartic in W4 with S5-symmetry. Note that the action of S5 on the Segre cubic primal is closely related to its known action on the del Pezzo surface of degree 5 via the following commutative diagram explained in [19], Proposition 4.7: X + φ+ 0 }④④④④④④④④ φ χ S3 X φ0 ❆❆❆❆❆❆❆❆ σ >⑥ ⑥ f ⑥ P3 ⑥ /❴❴❴❴❴❴❴❴ D Here χ is a flop, φ and φ+ surface of degree 5 and φ is a P1-bundle. 0 are small contractions to the Segre cubic primal, D is a del Pezzo The pencil (3.2) was intensively studied by K. Hashimoto in [10]. So, it is appropriate to refer to it as the Hashimoto pencil. In particular, Hashimoto computed the lattice of transcendental cycles T (Xt) of a minimal nonsingular model Xt of a singular member St of the pencil.   / /   } / > Theorem 1. Assume k = C. For any singular member of the Hashimoto pencil, the lattice T (Xt) is of rank 2 and is given by the matrix QUARTIC SURFACES 5 0 0 10(cid:19) (t = (cid:18)4 1 2 ), (cid:18)4 1 1 4(cid:19) (t = 1 4 ), (cid:18)4 2 16(cid:19) (t = 2 7 30 ), (cid:18)6 0 20(cid:19) (t = 0 13 20 ). The transcendental lattice of a generic member of the pencil is of rank 3 and is given by the matrix  4 1 0 1 4 0  0 0 −20   . Remark 1. It is known that the quartic surface defined by equation F = x4 + y4 + z4 + w4 + 12xyzw = 0 (3.5) admits a group of projective automorphisms isomorphic to 24.S5 (see [17]). According to S. Mukai [18], the subgroup 24.A5 is isomorphic to the Mathieu group M20 and is realized as one of maximal finite groups of symplectic automorphisms of a complex K3 surface. The equa- tion defining the surface is an invariant of the Heisenberg group H2 acting in its Schrodinger 4-dimensional irreducible representation (see [3]). The linear space of invariant quartic polynomi- als is 5-dimensional and it has a basis that consists of quartic polynomials p0 = x4 + x4 + z4 + w4, p1 = x2w2 + y2z2, p2 = x2z2 + y2w2, p3 = x2y2 + z2w2, p4 = xyzw. Let M1 = −2p0 + 24p4, M2 = p0 − 6(p1 + p2 + p3), M3 = p0 + 6(−p1 + p2 + p3), M4 = p0 + 6(p1 − p2 + p3), M5 = −2p0 − 24p4, M6 = p0 + 6(p1 + p2 − p3). be a spanning set of this linear space. The fifth of them defines the surface from (3.5). We propose to call the polynomials Mi the Maschke quartic polynomials (not to be confused with the Maschke octic polynomial from [2]). It is shown in [17], p. 505, that they satisfy 6 ( Xi=1 M2 i )2 − 4 6 Xi=1 M4 i = 6 Xi=1 Mi = 0 We recognize the equations of the Castelnuovo-Richmond-Igusa quartic threefold from Example 1. Thus, for any Maschke polynomial Mi, the surface V (Mi) is a Galois 24-cover of the pre-image of a coordinate hyperplane section of the quartic threefold isomorphic to the surface S1/4 from the Hashimoto pencil. This shows the appearance of 24.S5 in the group of projective automorphisms of V (Mi). It was communicated to me by Bert van Geemen that the projective transformations defined by the matrices M =    1 −1 −1 1 1 −1 −1 1  −1 1 −1 1 −1 −1 −1 −1 , N =    1 −1 −i −i 1 −i −i −1  1 −1 i i i −1 1 i , (3.6) 6 IGOR V. DOLGACHEV define automorphisms of orders 5 and 2 that generate a subgroup of automorphism of V (F ) that splits the extension 24.S5. This shows that the Maschke quartic surface V (Mi) admits S5 and hence A5 as its group of projective automorphisms. By computing the traces of M and N we find that S5 originates from its linear standard irreducible 4-dimensional representation, and the surface must be isomorphic to a member of the Hashimoto pencil. According to computations made by Bert van Geemen, the surface V (M1) corresponds to the parameter t = 3 20 (3 − i). Note that according to S. Mukai the transcendental lattice of the Maschke surface is given by the diagonal 2× 2-matrix with the diagonal entries 4 and 40. So it is different from the transcendental lattice of a general member of the pencil. 4. THE PENCIL (S4U∨4 )A5 Recall that the linear representation space U4 of G = 2.A5 is isomorphic to the space S3V. Since there is only one isomorphism class of irreducible faithful 4-dimensional representations of G, we have an isomorphism U4 ∼= S3V ∼= S3V′. Let e1, e2 be a basis in V and (u, v) be the coordinates in V with respect to this basis, i.e. the dual basis of (e1, e2) in V∨. The space SdV (resp. SdV∨) has a natural monomial basis (ed 2) (resp. (ud, ud−1v, . . . , vd)). The polarization isomorphism e2, . . . , ed 1, ed−1 1 SdV∨ → (SdV)∨ assigns to ud−ivi the linear function on SdV that takes the value 1 zero on all other monomials. This shows that the basis ((cid:0)d 1, ed−1 2). Thus any binary form f ∈ SdV∨ can be written as e2, . . . , ed (ed 1 i(cid:1)ud−ivi)i=0,...,d is the dual basis of d! (d − i)!i! on ed−i 1 di 2 and d f = Xi=0 (cid:0)d i(cid:1)aiud−ivi, (4.1) widely used in the invariant theory, we will switch to the basis ((cid:0)d so that (a0, . . . , ad) are the natural coordinates in the space SdU∨. Although, this notation is 2) of SdV to get the dual basis formed by the monomials ud−ivi. So, we will drop the binomial coefficients in (4.1). This will help us to agree our formulas with ones given in Klein's book. Let us clarify the coordinate-free definition of the Veronese map i(cid:1)ed−i 1 ei It is defined by assigning to a vector αe1 + βe2 the linear function f 7→ f (α, β) on SdV∨ = (SdV)∨. It follows that νd : V → SdV. d νd(αe1 + βe2) = Xi=0 (cid:0)d i(cid:1)αd−iβied−i 1 ei 2 = (αe1 + βe2)d. In coordinates, this is the map Passing to the projective space, we get the map (u, v) 7→ (ud, ud−1v, . . . , uvd−1, vd). (4.2) νd : V → SdV that is given by the complete linear system SdV∨ = OV(d). The image Rd of this map is a Veronese curve of degree d, or a rational normal curve of degree d. If we re-denote the coordinates ud−ivi by (x0, . . . , xd), a hyperplane V (Pd i=0 aixi) intersects Rd along the closed subscheme isomorphic, under the Veronese map νd, to the closed subscheme V (Pd i=0 aiud−ivi) of V. Dually, we have the Veronese map QUARTIC SURFACES 7 ν∗d : V∨ → SdV∨ which assigns to a linear function l = au + bv ∈ V∨ the linear function νd(l) ∈ SdV∨ that takes value on ed−i 1 2 equal to ad−ibi. It follows that ei Xi=0 (cid:0)d ν∗d(αu + βv) = d i(cid:1)αd−iβiud−ivi = (αu + βv)d. So, we get a familiar picture: points of SdV∨ are non-zero binary forms of degree d up to proportionality, and points of the Veronese curve are powers of linear forms, up to proportionality. In coordinates, the dual Veronese map is given by The image R∗d of the corresponding map (e1, e2) 7→ (ed 1, . . . ,(cid:0)d i(cid:1)ed−i 1 ei 2, . . . , ed 2). ν∗d : V∨ = P(V) → SdV∨ = P(SdV) is the dual Veronese curve of degree d. The duality can be clarified more explicitly. For any point x ∈ Rd, one can consider the osculating hyperplane at x, the unique hyperplane in SdV that intersects Rd at one point x with multiplicity d. The dual Veronese curve R∗d in the dual space SdV∨ is the locus of osculating hyperplanes. One can use the isomorphism V∨ → V defined by assigning to a linear function l = au + bv ∈ V∨ its zero V (l) = [−b, a] ∈ V. In other terms, it is defined by the exterior product pairing V × V → V2 V ∼= k. Thus we have two Veronese maps with the images Rd and R∗d. The map of V → SdV∨ is given, in coordinates, by ν∗d : V → SdV∨ νd : V → SdV, (u, v) → ((−1)ded 2, . . . , (−1)d−i(cid:0)d 1ed−i i(cid:1)ei 2 , . . . , ed 1). Let ρ : G → GL(V) be a linear representation of a group G. By functoriality, it defines a linear representation Sd(ρ) : G → GL(SdV). The dual linear representation ρ∗ : G → GL(V∨) defines a representation Sd(ρ∗) : G → GL(SdV∨). It follows from the polarization isomorphism that the representations Sd(ρ) and Sd(ρ∗) are dual to each other. After we fix these generalities, it is easy to describe irreducible linear representations of G = 2.A5. We start with the 2-dimensional representations V and V∨. We choose a basis e1, e2 in V and its dual basis (u, v) in V∨ as above to assume that the group preserves the volume forms e1∧e2 and u∧v. Thus in these bases we represent the matrices of the representation by unimodular matrices. According to [15], p. 213, the group 2.A5 is generated by the transformations S, T, U of orders 5, 4, 4, respectively, and its representation in V is given in terms of coordinates as follows: where ǫ = e2πi/5 and We have S : (u, v) T : (u, v) U : (u, v) 7→ (ǫ3u, ǫ2v), 7→ 7→ (−v, u), 1 √5 (−cu + dv, du + cv), c = ǫ − ǫ−1, d = ǫ2 − ǫ−2. λ = c/d = ǫ + ǫ−1 + 1 = 1 + √5 2 8 IGOR V. DOLGACHEV 2 is the golden ratio. It satisfies λ2 = λ + 1. Note that the trace of S is equal to ǫ3 + ǫ2 = λ2 − 2 = λ − 1 = −1+√5 . This is denoted by b5 in [7], so we can identify this representation with the one given by the character χ6. The representation V′ is given by the same formulas as above, where ǫ is replaced with ǫ2. The trace of S becomes b5∗ := −λ. the basis (e2 Next we consider the 3-dimensional irreducible representations realized in S2V and S3V′. In 2) of V, the first representation is given by the matrices 1, 2e1e2, e2 S :  0 ǫ 0 0 1 0  0 0 ǫ−1   , T : 1 √5   ǫ + ǫ4 2 ǫ2 + ǫ3 1 ǫ2 + ǫ3 2 ǫ + ǫ4 1 1  , U :    0  0 1 0 −1 0  . 0 1 0 are slightly different. Note that Klein uses slightly different coordinates (A0, A1, A2) = (−uv, v2,−u2) so his matrices The second irreducible 3-dimensional representation (S2V)′ is obtained by replacing ǫ with ǫ2. The invariant theory for the icosahedron group A5 in this representation is discussed in Klein's book (see also [9], 9.5.4). The linear representation U4 of 2.A5 realized in S3V is given by the formulas d3 0 ǫ3 0 0 0 0 ǫ2 0 S : (cid:18) ǫ4 0 0 0 0 0 0 ǫ(cid:19), 5√5(cid:18) −λ3 U = (cid:18) 0 0 0 −1 1 0 0 0 (cid:19). 0 0 1 0 0 −1 0 0 T : 3λ2 −3λ 1 λ2 −2λ+λ3 1−2λ2 λ −λ 1−2λ2 2λ−λ3 λ2 1 3λ2 λ3(cid:19) = 3λ d3 5√5(cid:18) −(1+2λ) 3(λ+1) −3λ −2λ−1 −1 −2λ−1 3(λ+1) 2λ+1(cid:19), λ+1 −λ 1 λ+1 1 λ 3λ 1 The dual representation is given by the same formulas, where ǫ is replaced with ǫ2. We will use the coordinates (x0 = u3, x1 = u2v, x2 = uv2, x3 = v3). Let N1 ⊂ S2U∨4 be the linear space of quadratic forms such that N1 is the linear system of quadrics with the base locus equal to the Veronese curve R3. Obviously, it is an irreducible summand of S2U∨4 . It is generated by the quadratic forms q2 = x0x2 − x2 1, q1 = x0x3 − x1x2, q3 = x1x3 − x2 2. Under the transformation S they are multiplied by 1, ǫ, ǫ4, respectively. The trace of S is equal to 1 + ǫ + ǫ4 = λ. Thus, we can identify the space N1 with the linear representation S2V. Let N ∗2 be the linear space of quadratic forms in the dual space U∨4 such that the linear system of quadrics N ∗2 has the base locus equal to the dual Veronese curve R∗3. It is spanned by the quadratic forms q′1 = 9ξ0ξ3 − ξ1ξ2, 1e2, ξ2 = e1e2 q′2 = 3ξ0ξ2 − ξ2 1, q′3 = 3ξ1ξ3 − ξ2 2, 2, ξ3 = e3 1, ξ1 = e2 2 are the dual coordinates. The representation N ∗2 where ξ0 = e3 is isomorphic to S2V′. Consider the dual space (N ∗2 )⊥ ⊂ S2U∨4 of apolar quadratic forms. It is spanned by quadratic forms p1 = x0x3+9x1x2, p2 = 2x0x2+3x2 The linear span is a 7-dimensional summand of S2U4. Computing the character of S2U∨, we find the decomposition 1, p3 = 2x1x3+3x2 2, p4 = x2 0, p5 = x2 3, p6 = x0x1, p7 = x2x3. S2U∨4 ∼= S2U4 ∼= W4 ⊕ S2V ⊕ S2V′. (4.3) QUARTIC SURFACES 9 This shows that (N ∗2 )⊥ = W4 ⊕ S2V′. Observe that the quadratic forms pi are eigenvectors of S with eigenvalues 1, ǫ, ǫ4, ǫ3, ǫ2, ǫ2, ǫ3, respectively. Since S has trace −1 on W4, we find that the summand V1 of (N ∗2 )⊥ isomorphic to W4 is spanned by eigenvectors with eigenvalues ǫ, ǫ2, ǫ3, ǫ4. Under the transformation T , the forms (p1, . . . , p7) are transformed to (−p1, p3, p2, p5, p4,−p7,−p6). Now we use the following known fact due to T. Reye (see [14], Lemma 4.3). Theorem 2 (T. Reye). A general 6-dimensional linear space L of quadrics in P3 = E contains precisely two nets N1,N2 with the base loci equal to Veronese curves C1, C2 of degree 3. The dual space L⊥ of apolar quadrics in E∨ contains 10 quadrics Qi with one-dimensional singular locus ℓi. Each line ℓi is a common secant of C1, C2. It follows from the lemma that the summand of S2U∨4 isomorphic to S2V′ is isomorphic to the linear space N2 such that the base locus of the net of quadrics N2 is a rational normal cubic curve. It is easy to see that this is possible only if N2 is generated by 0 + ax2x3, r3 = x2 r1 = x0x3 + 9x1x2, r2 = x2 3 − ax0x1, where a2 = 9. Since the image of the point [0, 1, 0, 0] on the curve under the transformation T must be on the curve, we check that a = 3. Thus, the base locus is a rational normal cubic curve with parametric equations U → U4, (u, v) 7→ (9u2v, 27v3,−u3, 27uv2). (4.4) Note that one can avoid using Reye's Theorem by using decomposition (4.3) and the fact there exists no A5-invariant lines, A5-invariant conics, and A5-orbits are of length ≤ 8. Now we are ready to write the pencil S4U∨4 of invariant quartics. We take as generators of the pencil the tangential ruled surfaces of the invariant rational normal curves defined by the nets N1 and N2. They are obviously invariant with respect to the action of G in W4. The net N1 defines a map from W4 to N ∨1 = S2V∨ whose image is an invariant conic, the fundamental conic. Its equation in Klein's coordinates A0, A1, A2 is A2 0 + A1A2 = 0. The equation of the dual conic in the dual coordinates A′0, A′1, A′2 in the dual plane S2U∨ is A′0 2 + 4A′1 A′2 = 0. It follows that the equation of the tangential ruled surface must be of the form Q2 0 + 4Q1Q2, where Qi = 0 are equations of quadrics from the net N1. We know that A0, A1, A2 are eigen- vectors for the transformation S acting in S2U with eigenvalues 1, ǫ, ǫ−1. Thus A′0, A′1, A′2 are eigenvectors for the action of S in S2U∨ with the eigenvalues 1, ǫ−1, ǫ. Therefore our quadrics Q0, Q1, Q2 are also eigenvectors for the action of S in S2W∨4 with eigenvalues 1, ǫ−1, ǫ. We find that (Q0, Q1, Q2) = (q1, λq3, λ′q2) for some constants λ, λ′. Since the equation is invariant with respect to the transformation U , and A′1 7→ −A′2, we must have λ = −λ′. We find the condition 0 − λ2Q1Q2 = 0 is the equation of the tangential ruled on λ that guarantees that the equation Q2 surface of the Veronese curve C1 = R3. 0] the tangent line to the Veronese curve R3 is spanned by the vector 0) of the tangent 1 + c2q2q3 = 0, we obtain that c2 − 4 = 0. Thus the equation of the quartic 0]. Plugging in the parametric equation s(1, t0, t2 [0, 1, 2t0, 3t2 line in the equation q2 tangential ruled surface is At each point [1, t0, t2 0) + r(0, 1, 2t0, 3t2 0, t3 0, t3 S1 : = x2 (x0x3 − x1x2)2 − 4(x0x2 − x2 0x2 3 − 6x0x1x2x3 + 4x0x3 1)(x1x3 − x2 2) 3 − 3x2 1x2 2 + 4x1x3 2 = 0. (4.5) 10 IGOR V. DOLGACHEV The second invariant quartic is the tangential ruled surface of the rational normal curve C2 defined parametrically in (4.4). The net of quadrics containing this curve is G-equivariantly isomorphic to S2V′. It defines the map from U4 to (S2V′)∨. The invariant conic in this space is the same conic as in the previous case. Similarly to this case we get the equation (x0x3 + 9x1x2)2 + d2(x2 0 + 3x2x3)(x2 3 − 3x0x1) = 0 (4.6) The parametric equation of the tangent line to the base curve C2 is s(9t2 0, 27,−t3 0, 27t0) + r(6t0, 0,−t2 0, 9) = 0. Plugging this in the equation (4.6), we obtain that d2 + 4 = 0 and the equation of the second tangential surface is S2 (x0x3 + 9x1x2)2 − 4(x2 0 + 3x2x3)(x2 3 + 18x0x1x2x3 + 27x2 So, our pencil (S4U∨4 )2.A5 can be explicitly written in the form : = 3(4x3 0x1 − x2 0x2 3 − 3x0x1) 1x2 2 − 4x2x3 3) = 0. where and λF1 + µF2 = 0, F1 = x2 0x2 3 − 6x0x1x2x3 + 4x0x3 2 + 4x1x3 3 − 3x2 1x2 2 (4.7) F2 = 4x3 0x1 − x2 0x2 3 + 18x0x1x2x3 + 27x2 1x2 2 − 4x2x3 3. Consider the linear transformation K : (x0, x1, x2, x3) 7→ (√3x2, 1 √3 x0,− 1 √3 x3,√3x1). (4.8) We have K 2 = U ∈ G, and the group generated by K and G is isomorphic to 2.S5. We immediately check that it transforms 0 + 3x2x3 7→ 3(x2 3 − 3x0x1 7→ 3(x2 x0x3 + 9x1x2 7→ 3(−x0x3 + x1x2), x2 This shows that K(S2) = S1, more precisely, we get K(F2) = 3F1. Thus the surfaces 0x2 S3 = V (3F1 + F2) = V (x2 S4 = V (F2 − 3F1) = V (−x2 are S5-invariant. One checks, using MAPLE, that they are nonsingular. 2 − 2x2x3 3 + 2x3 3 + 9x0x1x2x3 − 3x0x3 0x2 3 + 6x0x3 2 − 3x3 1 − x0x2), x2 2 + 6x3 1x2 1x3), 2 + x3 1x3 + 9x2 0x1 + 9x2 1x2 (4.9) 0x1 − x2x3 3). 2 − x1x3). The following result is proven in [5]. Proposition 4. The pencil generated by the quartic surfaces S1 and S2 contains two more singular members, each of them has 10 ordinary double points. The base curve B of the pencil is an irreducible curve of degree 16 with 24 singular points, each of them is an ordinary cusp. It also contains a unique A5-orbit of 20 points. It is easy to see these two orbits of singular points of B. Each of them is the intersection of one of the rational normal cubic curves C1 and C2 with a nonsingular surface from the pencil. If we take the latter to be the surface S3 or S4, we obtain that the union of the two orbits is invariant with respect to S5. In the next section we will be able to see explicitly the unique orbit of 20 points. QUARTIC SURFACES 5. AN A5-INVARIANT WEB OF QUADRICS IN U4 Consider the linear 2.A5-equivariant map φ : W4 → S2U∨4 11 (5.1) defined by the decomposition (4.3). The corresponding map of the projective spaces defines a web of quadrics in U4. Recall that some of its attributes are the determinantal surface D(φ) parametrizing singular quadrics in the web, and the Steinerian surface St(φ) parametrizing singular points of singular quadrics from the web. Let us find the equation of the discriminant surface. We choose a basis of φ(W4) formed by the quadrics Q1 = 2x1x3 + 3x2 2, Q2 = x2 0 − 2x2x3, Q3 = x2 3 + 2x0x1, Q4 = 3x2 1 + 2x0x2. They are transformed under the representation of 2.A5 (with the center acting trivially) as follows S : (Q1, Q2, Q3, Q4) 7→ (ǫ4Q1, ǫ3Q2, ǫ2Q3, ǫQ4), T :   Q1 Q2   Q3 Q4 −λ 1 λ λ −λ λ2 1 λ2 −λ 1 λ λ2 1 −λ λ 7→ c  Q1 Q2 Q3 Q4       U : (Q1, Q2, Q3, Q4) 7→ (Q4, Q3, Q2, Q1), λ2 , where c is some constant which will not be of concern for us. Choose a basis of φ(W4) ⊂ S2U∨4 spanned by the following quadratic forms: Q′1 = ǫ4Q1 − ǫ3Q2 − ǫ2Q3 + ǫQ4, Q′2 = −ǫ3Q1 − ǫQ2 − ǫ4Q3 + ǫ2Q4, Q′3 = ǫ2Q1 − ǫ4Q2 − ǫQ3 + ǫ3Q4, Q′4 = ǫQ1 − ǫ2Q2 − ǫ3Q3 + ǫ4Q4. (5.2) The symmetric matrix defining the quadratic form y1Q′1 + y2Q′2 + y3Q′3 + y4Q′4 is equal to the matrix where , 0 a11 a12 a13 a24 a21 a22 0 0 a31 a33 a34 0 a42 a43 a44 A(y1, y2, y3, y4) =     a11 = −(ǫ3y1 + ǫy2 + ǫ4y3 + ǫ2y4, a12 = a21 = −(ǫ2y1 + ǫ4y2 + ǫy3 + ǫ3y4), a13 = a31 = ǫy1 + ǫ2y2 + ǫ3y3 + ǫ4y4, a22 = 3(ǫy1 + ǫ2y2 + ǫ3y3 + ǫ4y4), a24 = a42 = ǫ4y1 + ǫ3y2 + ǫ2y3 + ǫy4, a33 = 3(ǫ4y1 + ǫ3y2 + ǫ2y3 + ǫy4), a34 = a43 = ǫ3y1 + ǫy2 + ǫ4y3 + ǫ2y4, a44 = −(ǫ2y1 + ǫ4y2 + ǫy3 + ǫ3y4). 12 IGOR V. DOLGACHEV Plugging in these expressions and computing the determinant, we obtain (using MAPLE) the equation of the discriminant surface, where y5 = −y1 − y2 − y3 − y4, yiyjy2 det(A(y1, y2, y3, y4)) = 25( yiyjykyl), yiy3 5 y4 i − X1≤i<j≤5 j + X1≤i<j≤k≤5 k−3 X1≤i<j<k<l≤5 5 Xi=1 where y5 = −(y1 + y2 + y3 + y4). We can rewrite this in terms of power functions to obtain the equation of the discriminant surface in the form 30 5 Xi=1 5 y4 i − 7( Xi=1 y2 i )2 = y1 + y2 + y3 + y4 + y5 = 0. (5.3) This is the equation of the 10-nodal surface S7/30 from the Hashimoto pencil.2 Remark 2. The 10 nodes of the symmetric discriminant quartic (also known as a Cayley sym- metroid quartic) correspond to the singular lines of 10 reducible quadrics in the web. According to A. Coble [6], p. 250, they are the ten common secants of the rational normal curves C1 and C2. Let us now compute the equation of the Steinerian surface of the web of quadrics defined by the map (5.1). If we choose the basis of φ(W4) given by (5.2), then the equation of the Steinerian surface is given by the determinant of the matrix with columns Q′1 · x, Q′2 · x, Q′3 · x, Q′4 · x, where we identify Q′i with the associated symmetric matrix and x is the column of the coordinates x0, x1, x2, x3 in U4: A =  ǫ4x3 − ǫ2x0 + 3ǫx1 −ǫ3x0 − ǫ2x1 + ǫx2 −ǫ4x1 + ǫ2x2 − ǫx0 −ǫ4x0 + ǫ3x3 + 3ǫ2x1  3ǫ3x1 + ǫ2x3 − ǫx0 −ǫ4x0 + ǫ3x2 − ǫx1 ǫ4x3 − ǫ3x1 − ǫ2x0 3ǫ4x2 − ǫ3x0 + ǫx3 ǫ4x1 + ǫ3x2 − ǫ2x3 3ǫ4x2 + ǫ3x3 + ǫx0 3ǫ3x2 + ǫ2x0 + ǫx3 −ǫ4x3 + ǫ3x1 + ǫx2 ǫ4x2 + ǫ2x1 − ǫx3 ǫ4x3 + ǫ3x0 + 3ǫ2x2 ǫ4x1 + ǫ2x3 + 3ǫx2 −ǫ3x3 + ǫ2x2 + ǫx1   t Computing the determinant of the matrix A, we find that the equation of the Steinerian surface coincides with the equation of the surface S4 from (4.9). The Steinerian surface S4 contains 10 lines, the singular lines of ten reducible members of the web corresponding to singular points of the discriminant surface S7/30. These are of course the ten common secants of the two rational normal curves C1 and C2. The unique orbit of 20 points from the same Proposition is of course the intersection points of the 10 common secants with the curves C1 and C2. Remark 3. Recall that a minimal nonsingular model of a Cayley quartic symmetroid is isomorphic to the K3-cover of an Enriques surface. In its turn, the Enriques surface is isomorphic to a Reye congruence of bidegree (7, 3) in the Grassmannian G(2, 4) (see [8]). This applies to our surface S4. The embedding j : S4 ֒→ U4 defines an invertible sheaf L1 = j∗OU4(1). The birational morphism from s : S4 → S7/30 ⊂ W4 whose image is the quartic symmetroid S7/30 that inverts the rational map from the discriminant surface to the Steinerian surface defines another invertible sheaf L2 = s∗OW4(1). It is known that the covering involution of S4 preserves the tensor product L1⊗L2. Since the both maps are A5-equivariant, we obtain that the action of A5 on S7/30 descends to an action on the Enriques surface. The following is a list of open questions: • Find the values of the parameters corresponding to the 10-nodal members. Are these sur- faces determinantal? What is the transcendental lattice of its minimal nonsingular member. • Find more facts about the S5-invariant surface S3. Is it determinantal? What is its tran- scendental lattice? 2This identification of the discriminant surface with the surface S7/30 was also confirmed to me by S. Mukai QUARTIC SURFACES 13 • Find the transcendental lattices of the general member of the pencil. 6. THE CATALECTICANT QUARTIC SURFACE There is another view of the Cayley symmetroid quartic surface S7/30. First we have the decomposition of the linear representations of A5 Comparing this with (4.3), we find an isomorphism S6V∨ ∼= S6V ∼= W4 ⊕ S2V. S2(U4)∨ ∼= S6V ⊕ S2V′. The summand S6V is isomorphic to the image of S2(U4)∨ under the restriction to the Veronese curve R6 of degree 3. The kernel S2V′ is isomorphic to the (linear) space of quadrics vanishing on the Veronese curve. Recall that each f ∈ S2dV∨ defines a linear map a : SdV → SdV∨ (the apolarity map). We view a basis of SdV as partial derivatives of the coordinates (u, v) in SdV and apply the differential operator fd( ∂ ∂v ) to f to obtain a binary form of degree d in u, v. In these coordinates, the determinant of the map is a polynomial of degree d + 1 in coefficients of the form f , called the catalecticant (see [9],1.4.1). A general zero of this polynomial is a binary form of degree 2d that can be written as a sum of less than the expected number (equal to d) of powers of linear forms. In the projective space S2dV∨ this corresponds to the variety of d-secant subspaces of dimension d − 1 of the Veronese curve R2d. In our case where d = 3 and the basis in S6V∨ is ∂u , ∂ taken as in (4.1), we get a quartic polynomial Cat = det  a0 a1 a2 a3 a1 a2 a3 a4 a2 a3 a4 a5 a3 a4 a5 a6 .   The zeros of this polynomial in S6V∨ is the variety Tri(R6) of trisecant planes of the Veronese curve R6. Let S = Tri(R6)] ∩ W4, this is a quartic surface in W4. Let us see that it coin- cides with the Cayley symmetroid S7/30 studied in the previous section. In coordinates, a direct computation shows that binary form f = P as(cid:0)6 0t6−s 1 ∈ S6V∨ corresponds to the quadric s(cid:1)ts 2 + a6y2 3 + 2a1y0y1 + 2a2y0y2 + 2a3y0y3 Q = a0y2 0 + a2y2 1 + a4y2 +2a5y2y3 + 2a4y3y1 + 2a3y1y2. The condition that Cat(f ) = 0 becomes the condition that Discr(Q) = 0. This shows that the catalecticant quartic becomes isomorphic to the discriminant quartic in P6 = P(N⊥). The singular locus of the variety Tri(R6) is of degree 10 and it is isomorphic to the secant variety of R6. Intersecting it with W4 we obtain 10 singular points of our symmetroid S7/30. respect to the dual of the fundamental conic. Note that another model of S6V is the space of harmonic cubics used in [12] in S2V with 7. A5-INVARIANT RATIONAL PLANE SEXTIC Let N1 and N2 be the nets of quadrics with base loci rational normal curves C1 and C2 defined by parametric equations (4.2) and (4.4). Restricting N2 to C1 we obtain a map C1 → N ∨2 ∼= S2V′∨. We identify the plane S2V′∨ with the plane S2V ′ via the A5-invariant conic. Using the basis of N2 formed by the quadrics V (x0x3 + 9x1x2), V (x2 3 − 3x0x1), the map P1 → C1 → S2V′∨ is given by 0 + 3x2x3) and V (x2 (u, v) 7→ (z0, z1, z2) = (10u3v3, u6 + 3uv5, v6 − 3u5v). 14 IGOR V. DOLGACHEV Recall that the quadratic polynomials r1, r2, r3 defining the basis are eigenvectors of S with eigen- values (1, ǫ3, ǫ2) and hence proportional to Klein's dual coordinates (A′0, A′1, A′2). Since they are also transformed as (r1, r2, r3) 7→ (−r1, r3, r2) under U , they are equal to (c′A′0, cA′1,−cA′2) for some constants c, c′. Also, we know that the dual conic A′0 A′2 = 0 parameterizes the singular quadrics in the net, and hence the equation of S2 shows that we may assume that c′ = ±1, c = 2. We noticed before that in our coordinates u, v, we have to choose c′ = −1. (z0, z1, z2) = (−A′0, 2A′1,−2A′2). The polarity isomorphism S2U → S2U′∨ defined by the conic A2 (−2A′0, A′2, A′1). Thus, in the Klein coordinates, the image of P1 → C1 → S2U′∨ → S2U′ is equal to the curve Γ1 with parametric equation 2 + 4A′1 0+A1A2 = 0, gives (A0, A1, A2) = Γ1 : (u, v) 7→ (A0, A1, A2) = (−5u3v3,−v6 + 3u5v, u6 + 3uv5). This agrees with the parametric equation of a A5-invariant rational sextic curve found by R. Winger in [20], [21]. He uses slightly different coordinates (x, y, z) = (A1, A2, A0). The pencil of A5-invariant sextics is spanned by the triple conic and the union of the six fun- damental lines: µ(xy + z2)3 + λz(x5 + y5 + z5 + 5x2y2z − 5xyz3 + z5). According to R. Winger [21], the curve Γ1 is a unique rational curve in this pencil with its 10 nodes forming an orbit of A5. It corresponds to the parameters (λ : µ) = (27, 5). The equation of Γ1 becomes 27z(x5 + y5) + 5x3y3 + 150x2y2z2 − 120xyz4 + 32z6 = 0. The action of A5 in the dual plane S2U∨ has an orbit of 6 points. The corresponding lines in the original plane S2U are the six lines with equations A0 = 0, A0 + ǫν A1 + ǫ−ν A2 = 0, ν = 0, . . . , 4. Each of these fundamental lines intersects the curve Γ1 at 2 points with multiplicity 3. The 12 intersection points are the images of the orbit of A5 acting in P1 that can be taken as the vertices of the icosahedron. They are the zeros of the polynomial The ten pairs of branches of the singular points of Γ1 correspond to the A5-orbit of 20 points in P1. They are the zeros of the polynomial Φ12 = uv(v10 + 11u5v5 − u10). Φ20 = u20 + v20 + 288(u15v5 − u5v15 − 494u10v10. The dual curve Γ∗1 of Γ1 is a rational curve of degree 10 with parametric equation (x∗, y∗, z∗) = (−10u7v3 − 5u2v8, 5u8v2 − 10u3v7, u10 − 14u5v5 − v10). Its equation can be found in [12], p. 83. Not that, via the fundamental conic, we can identify the dual planes S2V and S2V∨. Thus we have the second rational sextic curve Γ2 and its second dual curve Γ∗2. It is a rational curve Γ∗2 of degree 10 with parametric equation (x, y, z) = (−10(u7v3 − u2v8, 10(u8v2 − 2u3v7, u10 − 14u5v5 − v10). The pair of these curves corresponds to the pair of rational normal curves C1, C2. These pairing of rational normal sextics is discusses in details in [6], Chapter 4. Since the line z = 0 intersects Γ1 at two points with multiplicity 3, the Plucker formulas show that Γ∗1 and, hence Γ∗2 has six 4-fold multiple points. They are the fundamental points of A5.3 3Note that there is another A5-invariant rational curve of degree 10 with 36 double points, we will not be concerned with it. QUARTIC SURFACES 15 Remark 4. The blow-up of the 10 nodes of Γ1 (or of Γ2) is a rational Coble surface C with − KC = ∅ but − 2KC 6= ∅ and consists of the proper transform of the sextic Γ1. It inherits the A5-symmetry of Γ1. It is known that a Coble surface with an irreducible anti-bicanonical curve is a degenerate member of a pencil of Enriques surfaces. As I was informed by S. Mukai, there is a pencil of Enriques surfaces containing A5 in its automorphism group (acting linearly in its Fano embedding in P5) that contains among its members the Enriques surfaces with K3-covers S1/2 and S7/30 as well as the Coble surface C. Remark 5. The parametric equation of the A5-invariant rational sextic Γ1 appears in [13], p. 122. It is shown there that the curve Γ1 is equal to the intersection of the tangent lines at the origin of the modular family of elliptic quintic curves in P4 with the eigenplane of the negation involution. It is also discussed in [1], pp. 751-752, where it is shown that there is a (3 : 1)-map from the modular Bring's curve X0(2, 5) (isomorphic to the intersection of the Clebsch diagonal cubic surface with the quadric x2 5 = 0 to the normalization of the curve Γ1 that coincides with the forgetting map X0(2, 5) → X(5). The twelve cusps of Γ1 are its intersection points with the fundamental conic. 1 +···+x2 REFERENCES [1] W. Barth, K. Hulek, R. Moore, Degenerations of Horrocks-Mumford surfaces. Math. Ann. 277 (1987), 735 -- 755. [2] G. Bini, B. van Geemen, Geometry and arithmetic of Maschke's Calabi-Yau three-fold. Commun. Number Theory Phys. 5 (2011), 779 -- 820. [3] B. L. Cerchiai, B. van Geemen, From qubits to E7. J. Math. Phys. 51 (2010), no. 12, 122203, 25 pp. [4] I. Cheltsov, V. Przyjalkowski, C. Shramov, Quartic double solids with icosahedral symmetry, European Journal of Math. 2 (2016), 96 -- 119. [5] I. Cheltsov, C. Shramov, Two rational nodal quartic threefolds. Quart. J. Math. 67 (2016), 573 -- 601. [6] A. Coble, Algebraic geometry and theta functions, Amer. Math. Soc. Coll. Publ. vol. 10, Providence, R.I., 1929 (4d ed., 1982). [7] J. Conway, R. Curtis, S. Norton, R. Parker, R. Wilson, Atlas of finite groups. Maximal subgroups and ordinary characters for simple groups. With computational assistance from J. G. Thackaray. Clarendon Press, Oxford Uni- versity Press, 1985. [8] F. Cossec, Reye congruences, Trans. Amer. Math. Soc. 280 (1983),737 -- 751. [9] I. Dolgachev, Classical algebraic geometry. A modern view, Cambridge Univ. Press, 2012. [10] K. Hashimoto, Period map of a certain K3 family with an S5-action. With an appendix by Tomohide Terasoma. J. Reine Angew. Math. 652 (2011), 1 -- 65. [11] D. Hilbert, S. Cohn-Vossen, Anschauliche Geometrie, Berlin, Springer, 1932. [12] N. Hitchin, Vector bundles and the icosahedron. Vector bundles and complex geometry, 71 -- 87, Contemp. Math., 522, Amer. Math. Soc., Providence, RI, 2010. [13] K. Hulek, Projective geometry of elliptic curves. Ast´erisque No. 137 (1986), 143 pp. [14] K. Hulek, K. Ranestad, Abelian surfaces with two plane cubic curve fibrations and Calabi-Yau threefolds. Com- plex analysis and algebraic geometry, 275 -- 316, de Gruyter, Berlin, 2000. [15] F. Klein, Vorlesungen Uber das Ikosaeder und die Auflosung der Gleichungen vom funften Grade. LeipReprint of the 1884 original. Edited, with an introduction and commentary by Peter Slodowy. Birkhauser Verlag, Basel; B. G. Teubner, Stuttgart, 1993 [16] S. Kond¯o, Enriques surfaces with finite automorphism group, Japan J. Math. 12 (1986), 192 -- 282. [17] H. Maschke, Uber die quaternare endliche lineare Substitutions Gruppe der Borhardt'schen Moduln, Math. Ann. 30 (1887), 486 -- 515. [18] S. Mukai, Finite groups of automorphisms of K3 surfaces and the Mathieu group. Invent. Math. 94 (1988), 183 -- 221. [19] Y. Prokhorov, Fields of invariants of finite linear groups. Cohomological and geometric approaches to rationality problems, 245-273, Progr. Math., 282, Birkhauser Boston, Inc., Boston, MA, 2010. [20] R. Winger, Self-Projective rational sextics. Amer. J. Math. 38 (1916), no. 1, 45 -- 56. [21] R. Winger, On the invariants of the ternary icosahedral group. Math. Ann. 93 (1925), no. 1, 210 -- 216. 16 IGOR V. DOLGACHEV DEPARTMENT OF MATHEMATICS, UNIVERSITY OF MICHIGAN, 525 E. UNIVERSITY AV., ANN ARBOR, MI, 49109, USA E-mail address: [email protected]
1206.6517
1
1206
2012-06-27T20:32:13
On algebraic equivalences among the 27 Abel-Prym curves on a generic abelian 5-fold
[ "math.AG" ]
This article shows that on a generic principally polarized abelian variety of dimension five the $\mathbb{Q}$-vector space of algebraic equivalences among the 27 Abel-Prym curves has dimension 20.
math.AG
math
ON ALGEBRAIC EQUIVALENCES AMONG THE 27 ABEL-PRYM CURVES ON A GENERIC ABELIAN 5-FOLD MAXIM ARAP AND ROBERT VARLEY Abstract. This article shows that on a generic principally polarized abelian variety of di- mension five the Q-vector space of algebraic equivalences among the 27 Abel-Prym curves has dimension 20. 2 1 0 2 n u J 7 2 ] . G A h t a m [ 1 v 7 1 5 6 . 6 0 2 1 : v i X r a 1. Introduction Let X be a principally polarized abelian variety (ppav) over C. Let A∗(X) denote the Chow ring of X modulo algebraic equivalence with Q-coefficients (this notation differs from the one in [Fu98, 10.3, p.185]). Besides the intersection product, the ring A∗(X) is endowed with Pontryagin product defined by x1 ∗ x2 = m∗(p∗ 1x1 · p∗ 2x2), where m : X × X → X is the addition morphism, and pj : X × X → X is the projection onto the jth factor, cf. [BL, p.530]. Moreover, A∗(X) carries a bi-grading, A(X) = M Al(X)(s). l,s The l-grading is by codimension. The Beauville grading (s) is defined by the condition x ∈ Al(X)(s) if and only if k∗x = k2l−sx for all k ∈ Z, where k also denotes the endomorphism of X given by x 7→ kx (the second grading was originally defined in [B86] for rational equivalence but it is well-defined for algebraic equivalence as well). The (s)-component of a cycle Z is denoted by Z(s). Also, let us recall the Fourier transform FX : A∗(X) → A∗(X) that is given 1z · eℓ), where ℓ is the class of the Poincar´e bundle on X × X (here we identify X by z 7→ p2,∗(p∗ and its dual abelian variety using the principal polarization). September 21, 2018 2010 Mathematics Subject Classification: 14C25, 14H40. 2 on algebraic equivalence classes of abel-prym curves Let P := Prym( C/C) be the Prym variety of a connected ´etale double cover C → C of In the sequel we assume that C is not hyperelliptic and we let smooth curves, see [Mu74]. p := dim P . After fixing a base-point o ∈ C we have the Abel-Prym map ψ : C → P whose image will be denoted by C as well and is called an Abel-Prym curve (ψ is a closed embedding, cf. [BL04, Cor.12.5.6, p.380]). Given an Abel-Prym curve C ⊂ P , by [A12, Def.1, p.708] there is an associated tautological (sub)ring T (P, C) ⊂ A∗(P ), defined to be the smallest subring of A∗(P ) for the intersection product that contains [ C] and is stable under ∗, FP and k∗ for all k ∈ Z. By [A12, Thm.4, p.710], T (P, C) is generated by the cycles ζn := FP ([ C](n−1)) for 1 ≤ n ≤ p − 1 odd (the components of [ C] of even Beauville degree vanish because C has a symmetric translate). Also, by [A12, Rem.3, p.711] the ring T (P, C) does not depend on the choice of an Abel-Prym curve if P is a generic ppav of dimension p 6= 5. As a corollary of the main result of this article (Theorem 2.10), we obtain that for a generic ppav P of dimension 5 the tautological ring T (P, C) does depend on the choice of an Abel-Prym curve C ⊂ P . This gives 27 tautological rings, one for each choice of an Abel-Prym curve (see the next paragraph for the explanation of the number 27). However, by [A11, p.35] each of these 27 rings is isomorphic to the quotient Q[x1, x3]/(x6 1x3) of the polynomial ring in two variables via the homomorphism ζi 7→ xi 1, x2 3, x2 (the cycles ζn vanish for n > 3 by [A11, Cor.3.3.8, p.34]). Let P : R6 → A5 be the Prym map from the moduli space R6 of connected ´etale double covers C → C of smooth genus 6 curves to the moduli space A5 of 5-dimensional ppav's. By [DS81], the morphism P is generically finite of degree 27. By the tetragonal construction described in [Do92, 2.5, p.76], given ( C → C) ∈ R6 and a degree 4 linear pencil g 1 4 on C there exist ´etale double covers C0/C0 and C1/C1 of tetragonal curves whose associated Prym is isomorphic (as a ppav) to P = Prym( C/C). Furthermore, by [Do92, Thm.4.2, p.90], the correspondence on the generic fiber of P induced by the tetragonal construction is isomorphic to the incidence correspondence of the 27 lines on a smooth cubic surface. Also, the monodromy group of the cover R6 → A5 is the Weyl group W of the E6 lattice. It follows from [Fa96] and [A12] that the 27 Abel-Prym curves on a generic ppav of dimension 5 are not pairwise algebraically equivalent, see Lemma 2.4. In this article we determine all the 3 maxim arap and robert varley algebraic equivalences (with Q-coefficients) among these 27 effective 1-cycles. The main result is the following. Theorem (=Theorem 2.10). The Q-vector space of algebraic equivalences of the 27 Abel-Prym curves on a generic principally polarized abelian 5-fold has dimension 20. The key ingredient for proving the above theorem is Lemma 2.9. This lemma implies that the action of the monodromy group of the cover R6 → A5 on the 27 Abel-Prym curves on a generic ppav of dimension 5 preserves algebraic equivalences among the Abel-Prym curves (see [Har79] for a discussion of monodromy in algebraic geometry and its applications). Throughout the paper we work over the field of complex numbers. 2. Algebraic equivalences Proposition 2.1 (Connectedness principle). Let Y be a smooth connected but not necessarily complete curve and let f : X → Y be a proper morphism. If X is connected then the locus CF(f ) := {y ∈ Y Xy is connected} parametrizing connected fibers is Zariski closed in Y . Proof. Let f = h ◦ g be the Stein factorization, where g : X → Z has connected fibers and h : Z → Y is a finite morphism. Since Z = g(X) and X is connected then Z is connected. If Z is a point, the conclusion holds trivially. Assume that Z is not a point, let Z1, . . . , Zk be the irreducible components of Z and let hi be the restriction of h to Zi. If k ≥ 2 then there are at most finitely many points with connected fibers since each Zi is mapping onto Y . Thus it remains to consider the case k = 1, i.e., the case when Z is an irreducible curve and h : Z ։ Y is a finite surjective morphism. We may and shall assume that Z is reduced. Consider the normalization Z → Z, which is a finite birational morphism onto Z, and the composed morphism h : Z → Z → Y . It suffices to show that the locus CF(h) is Zariski closed in Y . If h has degree 1 then h is an isomorphism and CF(h) = Y . If h has degree ≥ 2, then CF(h), which must be contained in the branch locus, is at most a finite set of points and hence is Zariski closed in Y . (cid:3) 4 on algebraic equivalence classes of abel-prym curves Remark 2.2. The conclusion of the above proposition becomes false in general if Y is a reducible curve, dim Y > 1 or if X is not connected. Lemma 2.3. If C1/C1, C2/C2, C3/C3 is a tetragonally related triple on a Prym variety P then [ C1](2) + [ C2](2) + [ C3](2) = 0 in A∗(P ). Proof. Taking C1/C1 as a reference double cover, the curves C2, C3 ⊂ P are the special subva- 1 rieties associated to some g 4 on C1, see [B82, 3(c), p.366]. By [A12, Ex.1, p.724] we have the relation [ C2](2) + [ C3](2) = −[ C1](2). (cid:3) Lemma 2.4. On a generic ppav of dimension 5 there is a pair of Abel-Prym curves that are not algebraically equivalent. Proof. By [Fa96], for a generic ppav P ∈ A5 with Prym realization P = Prym( C/C) we have [ C](2) 6= 0. It is well known that C has five g 1 1 4's. Choosing a g 4 on the base curve C, the special 1 subvarieties V0, V1 ⊂ P associated to the g 4 are also Abel-Prym curves, [B82, 3(c), p.366]. By Lemma 2.3 we have the relation [V0](2) + [V1](2) + [ C](2) = 0. If the Abel-Prym curves V0, V1 and C are pairwise algebraically equivalent then we have 3[ C](2) = 0. Since we are working with Q-coefficients this implies [ C](2) = 0, which is a contradiction. (cid:3) Lemma 2.5. If C1, . . . , C27 are the 27 Abel-Prym curves on a generic ppav P ∈ A5 then the vector space of algebraic equivalences among C1, . . . , C27 induced by the tetragonal construction has dimension 20. Proof. Let V be the Q-vector space with basis C1, . . . , C27 and consider the linear map cl : V → A4(P ) that takes an element of V to its class modulo algebraic equivalence. Let Vtet ⊂ V be the subspace spanned by the vectors Ci + Cj + Ck, where Ci, Cj, Ck are tetragonally related. The vector space Rtet := ker(cl : Vtet → A4(P )) is the space of algebraic equivalences induced by the tetragonal construction. It follows immediately from [Do92, Thm.4.2, p.90] that tetragonally related triples are in bijection with the 45 triples of lines forming triangles on a smooth cubic surface in P3. Let us fix an ordering L1, . . . , L27 of the 27 lines on the cubic surface. Each of the triangles determines 5 maxim arap and robert varley a vector in a 27 dimensional Q-vector space with basis L1, . . . , L27 with 1's in the coordinates corresponding to the lines forming the triangle, see [Dol10] and [Hun96, Ch.4] for details on the geometry of cubic surfaces. An explicit calculation shows that the dimension of the span of these 45 vectors is 21. Therefore, dim Vtet = 21. For each i ∈ {1, . . . , 27} we have the formula [ Ci] = [ Ci](0) + [ Ci](2) in A4(P ). The vanishing of the odd degree Beauville components of [ Ci] is a consequence of the fact that Ci has a symmetric translate. The vanishing of components of degree ≥ 4 was proven in [A11, Cor.3.3.8]. Furthermore, using the Fourier transform on A∗(P ) it is easy to see that [ C1](0) = . . . = [ C27](0) in A4(P ). The image of [ Ci](0) under the cycle class map A4(P ) → H 8(P, Q) is twice the minimal cohomology class, and therefore, [ Ci](0) 6= 0 for all i ∈ {1, . . . , 27}. Using Lemma 2.3, this implies that for every tetragonally related triple Ci, Cj, Ck the image vector cl( Ci + Cj + Ck) lies in the 1-dimensional subspace of A4(P ) spanned by [C1](0). Therefore, dim Rtet = 20. (cid:3) Let L1, . . . , L27 be the lines on a smooth cubic surface. It is well known that the group of symmetries of these lines is the Weyl group W of the E6 lattice. Lemma 2.6. The representation of W on L27 representations of dimensions 1,6 and 20. i=1 QLi decomposes into a direct sum of irreducible Proof. The Weyl group W has order 51840 and 25 conjugacy classes. Computing the character table for W and using the orthogonality relations between irreducible characters, it can be seen that L27 calculations were carried out on Magma. i=1 QLi decomposes into irreducible representations of dimensions 1, 6 and 20. Our (cid:3) Remark 2.7. The vectors determined by the 45 triangles in the cubic surface span a 21- dimensional subrepresentation, whose quotient is the representation of W on the Q-span of the E6 lattice, which is well-known to be absolutely irreducible, [Hum73, Lem.B, p.53]. Under the map of the free abelian group on L1, . . . , L27 to the Picard group of the cubic surface, each triangle is mapped to −K (the anti-canonical divisor class) and the perp of K is precisely the E6 lattice. 6 on algebraic equivalence classes of abel-prym curves Let A ◦ 5 be the complement in A5 of the union of the branch divisor of R6 → A5 and the 5 over which singular locus of A5. Let V be a non-empty connected Zariski open subset of A ◦ the Prym map R6 → A5 is finite. We have the following key lemma. Lemma 2.8. A general (smooth, connected but not complete) curve section S of V has the property that the map π1(S) → π1(V ) induced by the inclusion is surjective. Proof. If V were projective, the result of the lemma would follow immediately by the Lefschetz hyperplane theorem for homotopy groups. Since V is not projective, we have to compactify V and resolve the resulting space before applying Lefschetz theorem. Although the argument is standard, we include it here for completeness. Let ¯V be a smooth projective variety containing V and such that the complement ¯V − V is a simple normal crossings (snc) divisor. To obtain ¯V , we may take the closure of V in some projective space and apply Hironaka's theorem to resolve the singularities and make the complement of V an snc divisor (we identify V with its proper transform in ¯V ). Let D1, . . . , Dn be the irreducible components of ¯V − V . For a general curve section ¯S of ¯V we have ¯S ∩ Di 6= ∅ for 1 ≤ i ≤ n and S := ¯S ∩ V is a non-empty Zariski open set in ¯S. Also, by the theorem of Lefschetz (cf. π1( ¯S) ։ π1( ¯V ). [Mi63, Thm.7.4, p.41]) and induction we have a surjection Given a smooth connected complex manifold M and a smooth connected divisor Γ ⊂ M, using van Kampen's theorem we may check that there is an exact sequence ZℓΓ → π1(M − Γ) → π1(M) → 1, where ℓΓ is the class of a loop in M that goes once around Γ. Since ¯S ∩ D1 6= ∅ then ℓD1 is in the image of π1(S) → π1( ¯V − D1). Therefore, using the above exact sequence we have a natural diagram with all maps being surjective π1(S) π1( ¯S) π1( ¯V − D1) / π1( ¯V ). Now, removing Di's one at a time and applying the above argument, we conclude by induction that there is a surjection π1(S) ։ π1(V ). (cid:3) / / / /         / / / 7 maxim arap and robert varley Recall that a connected cover f : X → Y of topological spaces is said to be Galois (or regular, or normal, see [Hat02, p.70]) if for every y ∈ Y and every pair of lifts x1, x2 ∈ X of y there is a deck transformation of X taking x1 to x2. Given a connected cover f : X → Y of topological spaces we shall denote its Galois closure by X → Y . More precisely, X is the connected cover of Y corresponding to the normal subgroup of π1(Y ) obtained as the intersection of all the conjugates of f∗(π1(X)) in π1(Y ). Also, the group of deck transformations of X over Y will be referred to as the monodromy group of X over Y and will be denoted by M(X/Y ) in the sequel. Let U := P −1(V ) be the inverse image of V under the Prym map P : R6 → A5. By our construction, the cover U → V is finite ´etale of degree 27. Unfortunately V does not carry a universal family of ppav's, which we shall need in the sequel. In what follows, we shall show that there is a finite cover V ′ → V such that V ′ carries a universal family of ppav's and the monodromy group M(U ′/V ′) of the pull-back U ′ → V ′ of U → V surjects onto W . Let us argue that taking ppav's with level 3 structure suffices for this purpose, i.e., we may take the connected ´etale cover λ : V ′ → V with Galois group G := Sp(10, Z/3)/(±I). First, it is well known that the moduli space of 5-dimensional ppav's with level 3 structure carries a universal family (see [BL04, Prop.8.8.2, p.233]), and in particular, so does V ′. Therefore, it suffices to prove that M(U ′/V ′) surjects onto W . By [Do92], there is a surjection π1(V ) ։ W whose kernel N := ker(π1(V ) ։ W ) is the image of π1( V ) under the natural homomorphism. The group M := λ∗(π1(V ′)) = ker(π1(V ) → G) determines the Galois cover V ′ → V . The situation is summarized in the following commutative diagram π1( U ′) π1( U ) π1(V ′) λ∗ / π1(V ) W, where the cokernel of the vertical map on the left is the monodromy group M(U ′/V ′). To prove surjectivity of M(U ′/V ′) → W induced by the homomorphisms in the above diagram, it suffices to show that M surjects onto W . By considering the indices of M and N in π1(V ) we / /     /     8 on algebraic equivalence classes of abel-prym curves see immediately that N is not contained in M. Indeed, the index of M is the order of G and is given by a standard formula. However, it suffices to note that H = GL(5, Z/3)/(±I) injects as a subgroup of G and H = 1 2(35 − 1)(35 − 3)(35 − 32)(35 − 33)(35 − 34), which is already much larger than 51840. Therefore, the image ¯N of N in the quotient π1(V )/M ≃ G is non-trivial. Since N is a normal subgroup of π1(V ), ¯N is normal in the quotient group G. Since G is a simple group (cf. [Di48, Thm.1, p.12]) and ¯N is non-trivial, we must have ¯N = G. Thus, we have MN = π1(V ), which shows that the quotient map M → π1(V )/N ≃ W is surjective. By Lemma 2.8, there exists a smooth connected curve section S of V whose topological fundamental group surjects onto that of V . Let S ′ := λ−1(S) and let R′ be the pull-back of U ′ to S ′. The covering µ : S ′ → S is determined by the inverse image Q of λ∗(π1(V ′)) in π1(S) under the homomorphism π1(S) → π1(V ) induced by the inclusion (in particular, Q = µ∗π1(S ′)). Therefore, since λ∗ is injective, from the following commutative diagram λ∗ µ∗ π1(V ′)  π1(S ′) / π1(V ) π1(S) we see that the vertical map on the left is surjective. Using the following commutative diagram π1( U ′) / π1(V ′) π1( R′) π1(S ′) and surjectivity of π1(S ′) ։ π1(V ′) we conclude that the monodromy group of R′ → S ′ still surjects onto W . To summarize some of the constructions so far, we now have over the smooth curve S ′, a degree 27 finite ´etale map R′ → S ′ and a family P ′ → S ′ of ppav's (with level 3 structure), such that for each point s′ ∈ S ′, each one of the 27 points in the fiber of R′ → S ′ over s′ (equivalently, in the fiber in R6 over the image point of V ⊂ A5) corresponds to exactly one of 27 Abel-Prym curves in the abelian variety P ′ up to translations (since the automorphism group of the ppav P ′ s′. Each Abel-Prym curve is uniquely determined s′ is {±1}, and an Abel-Prym curve always has some translates that are invariant under −1). Since translation in the abelian  / / / O O O O O O / / / O O O O 9 maxim arap and robert varley variety defines an obvious algebraic equivalence between a 1-cycle and any of its translates, maintaining a careful distinction between an Abel-Prym curve Ci and the collection of all its translates, is not the essential point in our arguments; we are going to suppress the distinction, for simplicity. What is critical is that the mapping R′ → S ′ still has monodromy group W . Additionally, for simplicity of notation, we shall now write R and S for R′ and S ′, respectively, in the sequel. Thus, with this updated notation, let P be the universal family of 5-dimensional ppav's (with level 3 structure) over S. The fiber over s ∈ S will be denoted by Ps. Let V be the local system on S whose fiber over a point s ∈ S is the 27-dimensional vector space spanned by the Abel-Prym curves in Ps. Given a contractible open set U ⊂ S in the analytic topology we fix a trivialization V (U) ≃ 27 M i=1 QCi, and let Ci,s denote the fiber of Ci over s ∈ U. Thus, the fiber of V over s ∈ U is Vs = L27 To simplify the notation, for q = (q1, . . . , q27) ∈ Q27 define i=1 QCi,s. Zs(q) := 27 X i=1 qiCi,s. Since S has a base for the analytic topology consisting of contractible sets, the assignment U 7→ T (U) := (cid:8)Zs(q) Zs(q) ∼alg 0 for all s ∈ U(cid:9) defines a presheaf on S in the analytic topology. In the sequel, we let T denote the associated sheaf on S. Lemma 2.9. The subsheaf T of V is a local subsystem of V . Proof. Given a contractible open set U ⊂ S in the analytic topology, a trivialization V (U) ≃ L27 i=1 QCi and q = (q1, . . . , q27) ∈ Q27 as above, we claim that the locus {s ∈ U Zs(q) ∼alg 0} is either all of U or a countable set of points in U. Assuming this claim, let us prove the lemma. Fix s ∈ S for the remainder of the proof and consider the monodromy representation ρ : π1(S, s) → Aut(Vs). 10 on algebraic equivalence classes of abel-prym curves It is well-known that there is a natural bijection between isomorphism classes of local systems and monodromy representations up to conjugation, see [V07, Cor.3.10, p.71]. Under this bijec- tion subrepresentations correspond to local subsystems. Therefore, to show that T is a local subsystem of V , it suffices to prove that for each [γ] ∈ π1(S, s) the automorphism ρ([γ]) of Vs maps Ts to itself. The image of γ : [0, 1] → S can be covered by finitely many contractible open sets U1, . . . , Un for the analytic topology. Assume s ∈ U1 and Ui ∩ Ui+1 6= ∅ for i = 1, . . . , n − 1. Given an element Zs(q) ∈ Ts, by the above claim we know that Zt(q) ∈ Tt for all t ∈ U1. Applying the same argument to U2, . . . , Un, we conclude that the parallel transport of Zs(q) stays in the fibers of T . Therefore, ρ([γ]) maps Ts to itself. In the remainder of the proof we show the above claim. Let R → S be the Galois closure of the cover R → S. Let V be the pull-back of V to R under the composition R → R → S. Since the claim is about algebraic equivalences in the fibers of V , it suffices to show the claim for V globally over R. The local system V trivializes over R and we may write V ≃ 27 M i=1 QCi. Let P := P ×S R → R be the pull-back of the universal family. By clearing the denominators and rearranging the terms we may rewrite Zs(q) ∼alg 0 as X aiCi,s ∼alg X bjCj,s, i∈I j∈J (2.1) where ai, bj are non-negative integers, I ∪ J = {1, . . . , 27} and I ∩ J = ∅. Furthermore, (2.1) holds if and only if there exists an effective cycle Es on Ps such that the cycles Es +Pi∈I aiCi,s and Es + Pj∈J bjCj,s lie in the same connected component of the Chow variety of Ps, see [K96, 4.1.3, p.122]. In what follows we may and shall assume that the effective cycle Es is effectively algebraically equivalent to a sufficiently high multiple of the cycle Ξ4 s, where Ξs is the theta divisor on Ps (see [K96, 4.1.2, p.121] for the definition of effective algebraic equivalence). More precisely, we may arrange, by taking another finite cover if necessary, that there is a family over R of theta divisors Ξs ⊂ Ps, and we take Es to be the intersection of four general elements of the linear system nΞs for n ≫ 0. To emphasize the role of n we write En,s for such Es. This assumption is needed for the following two reasons. First, to ensure that En,s extends to a cycle over all of R, i.e., that there is an effective cycle En flat over R whose fiber over s ∈ R is En,s. 11 maxim arap and robert varley Second, to ensure that is suffices to consider at most countably many effective relative cycles En (indexed by n) in order to exhibit algebraic equivalence (provided it holds) of Pi∈I aiCi,t and Pj∈J bjCj,t on Pt by adding on the fiber of En over t. Consider the relative Chow variety of P over R that parametrizes effective cycles on P which lie over 0-dimensional subschemes of R, see [K96, Def.3.1.1, p.41 and Def.3.20, p.51]. Let En,t denote the fiber of En over t ∈ R. It is easily seen that the cycles En,t + Pi∈I aiCi,t and En,t + Pj∈J bjCj,t are algebraically equivalent on P (but not necessarily on Pt), and therefore, lie in the same connected component X of the relative Chow variety of P over R . Applying Proposition 2.1 to the natural morphism X → R we conclude that the set of t ∈ R such that En,t + Pi∈I aiCi,t and En,t + Pj∈J bjCj,t lie in the same connected component of the Chow variety of Pt is Zariski closed in R. If Pi∈I aiCi,t and Pj∈J bjCj,t are algebraically equivalent on Pt, there exists m ≥ n and a relative cycle Em (constructed as above) such that Em,t + Pi∈I aiCi,t and Em,t + Pj∈J bjCj,t lie in the same component of the Chow variety of Pt. Furthermore, since the choice of the four effective divisors in mΞt that define Em,t is unrestricted (as long as they intersect properly), it suffices to consider at most countably many such cycles Em in order to exhibit algebraic equivalence of Pi∈I aiCi,t and Pj∈J bjCj,t on Pt (for any t). Therefore, we conclude that the locus of t ∈ R such Pi∈I aiCi,t and Pj∈J bjCj,t are algebraically equivalent on Pt is either all of R or a countable union of points in R. (cid:3) Theorem 2.10. The Q-vector space of algebraic equivalences of the 27 Abel-Prym curves on a generic principally polarized abelian 5-fold has dimension 20. Proof. By Lemma 2.5 the fibers of T have dimension ≥ 20. By definition of T , for each s ∈ S the 1-dimensional sub-representation Ls of Vs spanned by the Beauville degree 0 graded piece of an Abel-Prym curve intersects Ts trivially. By Lemma 2.4, the composition Ts ֒→ Vs → Vs/Ls is not surjective for generic s ∈ S. Therefore, for generic s ∈ S we have dim Ts ≤ 25. By Lemma 2.9, T is invariant under the monodromy of the Galois closure of R → S, which by our construction surjects onto the Weyl group W . By Lemma 2.6 we conclude that the fibers of T have dimension 20. (cid:3) 12 on algebraic equivalence classes of abel-prym curves Corollary 2.11. On a generic ppav P of dimension 5 with Abel-Prym curves C1, . . . , C27 we have: (1) T (P, Ci) 6= T (P, Cj) if i 6= j; (2) [ Ci](2) 6= 0 in A∗(P ) for 1 ≤ i ≤ 27. Proof. (1) For each i ∈ {1, . . . , 27}, the bi-graded piece of T (P, Ci) of codimension 4 and Beauville degree 2 is spanned by [ Ci](2). If T (P, Ci) = T (P, Cj) then [ Ci](2) and [ Cj](2) are proportional, which is easily seen to contradict the structure of the vector space of relations among Abel-Prym curves from the proof of Theorem 2.10. (2) We may check that the class [ Ci](2) does not belong to the 20-dimensional Q-vector space of algebraic equivalences among C1, . . . , C27 from Theorem 2.10. (cid:3) Acknowledgements The first named author expresses his gratitude to David Swinarski for help with Magma. Both authors are thankful to Mitchell Rothstein for a helpful suggestion. The first named author is grateful to the second named author and to Roy Smith for sharing a question by Emanuele Raviolo that is closely related to the subject of the present article, which was then in progress. References [A11] Arap, M.: Tautological rings of Prym varieties. PhD thesis, University of Georgia (2011). [A12] Arap, M.: Algebraic cycles on Prym varieties. Math. Ann. 353 (2012), 707 -- 726. [B82] Beauville, A.: Sous-vari´et´es sp´eciales des vari´et´es de Prym. Comp. Math. 45 (1982), no. 3, 357 -- 383. [B86] Beauville, A.: Sur l'anneau de Chow d'une vari´et´e ab´elienne. Math. Ann. 273 (1986), no. 4, 647 -- 651. [BL04] Birkenhake, C., Lange, H.: Complex Abelian Varieties. 2nd ed. Grund. der Math. Wis., 302. Springer- Verlag, Berlin (2004). [Di48] Dieudonn´e, J.: Sur les Groupes Classiques. Hermann & Cie, Editeurs, Paris (1948). [Dol10] Dolgachev, I.: Classical Algebraic Geometry: a Modern View. Book to be published by Cambridge University Press. [DS81] Donagi, R., Smith, R. C.: The structure of the Prym map. Acta Math. 146 (1981), no. 1-2, 25 -- 102. [Do92] Donagi, R.: The fibers of the Prym map. Curves, Jacobians, and Abelian Varieties (Amherst, MA, 1990), 55 -- 125, Contemp. Math., 136, Amer. Math. Soc., Providence, RI (1992). [Fa96] Fakhruddin, N.: Algebraic cycles on generic Abelian varieties. Comp. Math. 100 (1996), no. 1, 101 -- 119. [Fu98] Fulton, W.: Intersection Theory. Ergeb. der Math. und ihrer Grenz. Vol. 2, 2nd edition. Springer-Verlag, Berlin (1998). [Har79] Harris, J.: Galois groups of enumerative problems. Duke Math. J. 46 (1979), no. 4, 685 -- 724. 13 maxim arap and robert varley [Hat02] Hatcher, A.: Algebraic topology. CUP, Cambridge (2002). [Hum73] Humphreys, J.E.: Introduction to Lie Algebras and Representation Theory. GTM 9, Springer (1973). [Hun96] Hunt, B.: The Geometry of Some Special Arithmetic Quotients. LNM 1637. Springer (1996). [K96] Koll´ar, J.: Rational Curves on Algebraic Varieties. Ergeb. der Math. und ihrer Grenz. 3, 32. Springer- Verlag, Berlin (1996). [Mi63] Milnor, J.: Morse Theory. Princeton University Press, (1963). [Mu74] Mumford, D.: Prym varieties. I. Contributions to analysis, pp. 325 -- 350. Acad. Press, New York, 1974. [V07] Voisin, C.: Hodge theory and Complex Algebraic Geometry II. Translated from the French by Leila Schneps. Cambridge Studies in Advanced Mathematics, 77. CUP, Cambridge, (2007). Johns Hopkins University, Department of Mathematics, 404 Krieger Hall, 3400 N. Charles Street, Baltimore, MD 21218, USA E-mail address: [email protected] University of Georgia, Department of Mathematics, Boyd Graduate Studies Research Cen- ter, Athens, GA 30602, USA E-mail address: [email protected]
1612.09406
1
1612
2016-12-30T06:55:56
On derived categories of nonminimal Enriques surfaces
[ "math.AG" ]
By Orlov's formula, the derived category of blow up must contain the original variety as a semiorthogonal component. This arises an interesting question: does there exist a variety $X$ such that $\operatorname{D}^{\sf b}(X)$ does not admit an exceptional collection of maximal length, but $\operatorname{D}^{\sf b}(\operatorname{Bl}_{x} X)$ admits such a collection? We give such an example where $X$ is a minimal Enriques surface.
math.AG
math
ON DERIVED CATEGORIES OF NONMINIMAL ENRIQUES SURFACES YONGHWA CHO Abstract. By Orlov's formula, the derived category of blow up must contain the original variety as a semiorthogonal component. This arises an interesting question: does there exist a variety X such that Db(X) does not admit an exceptional collection of maximal length, but Db(Blx X) admits such a collection? We give such an example where X is a minimal Enriques surface. 1. Introduction This short note is to give an answer to the question posed in [7, Remark 3.12] on the existence of exceptional collections of maximal length in the derived categories of nonminimal Enriques surfaces. In his recent article [7], Vial proves that an algebraic surface S with pg = q = 0 admits numerically exceptional collections of maximal length if and only if one of the following is true: (1) S is minimal and of Kodaira dimension −∞ or 2; (2) S is one of the Dolgachev surfaces of type X9(2, 3), X9(2, 4), X9(3, 3), X9(2, 2, 2); (3) S is nonminimal. By this criterion, an (minimal) Enriques surface never has an exceptional collection of maximal length, but there still remains a possibility that its blow up has an exceptional collection of maximal length. Indeed, it turns out that there exist such examples: Theorem 1.1 (see Theorem 3.2). There exist an Enriques surface S(cid:48) such that the blowing up at a general point gives a surface S whose derived category admits a semiorthogonal decomposition of 13 line bundles together with a triangulated category A satisfying K0(A) = Z/2Z. By the formula due to Orlov [6], we have two very different-looking semiorthogonal decompositions Db(S) = (cid:104)OE(1), Db(S(cid:48))(cid:105) = (cid:104)A, E1, . . . , E13(cid:105), where E is the exceptional divisor of S → S(cid:48). It seems a very intriguing question to ask how these semiorthogonal components can be compared. In Section 2 we briefly explain notions related to the exceptional collections on algebraic surfaces. Section 3 deals with the construction method of the nonminimal Enriques surfaces which appear in Theorem 1.1 and the technical parts, including proofs, are discussed in Section 4. The theoretical backgrounds on Sections 3 -- 4 are developed in [1], but these have been applied to simpler setup in this article. For this reason, we expect that this example is more comprehensible than the one in [1]. Notations 1.2. (1) Everything is defined over the field of complex numbers, except YQ in the proof of Lemma 4.9. (2) Let µr = (cid:104)ζr(cid:105) be the multiplicative group which is generated by the primitive rth root of unity. The group action µr × Cn → Cn, ζr · (x1, . . . , xn) = (ζ a1 r x1, . . . , ζ an r xn) 1 2 YONGHWA CHO defines the quotient space Cn/µn, which we will denote by Cn/ 1 r (a1, . . . , an). (3) For a scheme T of finite type over C and a point P ∈ T , (P ∈ T ) denotes the analytic germ. (4) Except stated otherwise, the equality between divisors indicates the linear equivalence relation. Also, we say D is effective if D is linearly equivalent to an effective divisor, or equivalently, h0(D) > 0. 2. Preliminaries Let X be a nonsingular projective variety over C and let Db(X) be the bounded derived category of coherent sheaves on X. An exceptional collection is an ordered collection of objects E1, . . . , Ek ∈ Db(X) satisfying the following conditions:  C i = j, p = 0 i = j, p (cid:54)= 0 0 i > j 0 HomDb(X)(Ei, Ej[p]) This notion is motivated from the decomposition problem in derived categories. In general, a triangulated category T admits a semiorthogonal decomposition (cid:104)T1, . . . ,Tk(cid:105) if (1) T1, . . . ,Tk are full triangulated subcategories of T ; (2) the smallest full triangulated subcategory containing T1, . . . ,Tk is T ; (3) HomT (Ti, Tj) = 0 for each i > j and Ti ∈ Ti, Tj ∈ Tj. If T = Db(X) and E1, . . . , Ek is an exceptional collection, then there exists a semiorthogonal decompo- sition T = (cid:104)A, E1, . . . , Ek(cid:105) where A = (cid:104)E1, . . . , Ek(cid:105)⊥ is the full triangulated subcategory generated by the objects {A ∈ Db(X) : HomDb(X)(Ei, A[p]) = 0, i = 1, . . . , k, p ∈ Z} In practical situations, the Hom-groups in the derived category has a geometric interpretation. Indeed, for any coherent sheaf F on X, we can regard F as an objects in Db(X) by considering the complex (. . . → 0 → F → 0 → . . .) concentrated at degree zero, then for coherent sheaves F1 and F2 one has HomDb(X)(F1,F2[p]) (cid:39) Extp (2.1) An exceptional object in Db(X) contributes a Z-direct summand in the group K0(X). Thus, if E1, . . . , Ek is an exceptional collection, then K0(X) = Z⊕k ⊕ K0(A) where A = (cid:104)E1, . . . , Ek(cid:105)⊥. For this reason, the length of an exceptional collection is bounded by the rank of K0(X). If X is an algebraic surface with CH2(X) (cid:39) Z, then it is known that K0(X) (cid:39) Z⊕2 ⊕ Pic X (see [3, Lemma 2.7]). Hence, for S as in Theorem 1.1, X (F1,F2). K0(S) (cid:39) Z⊕2 ⊕ Pic S (cid:39) Z⊕13 ⊕ Z/2Z, thus the length of any exceptional collection does not exceed 13. Also, once we establish such a collection, say E1, . . . , E13, then the orthogonal category A := (cid:104)E1, . . . , E13(cid:105)⊥ must satisfy K0(A) = Z/2Z, and in particular, A (cid:54)(cid:39) 0. ON DERIVED CATEGORIES OF NONMINIMAL ENRIQUES SURFACES 3 3. Construction method We begin with explaining the method to construct Enriques surfaces by Q-Gorenstein smoothing. The method is originally developed in [5]. Also, the paper contains the construction of Enriques surfaces as an example (see [5, Example 2]). Here, we give a minimal description to establish the notations for the future use. Let h1, h2 ∈ S := C[x, y, z] be homogeneous cubics which define nodal cubics on the plane. Assume h1 ∩ h2 are nine distinct points. Then the pencil p := λ1h1 + λ2h2 defines ϕp : P2 (cid:57)(cid:57)(cid:75) P1 whose resolution of indeterminacy is the rational elliptic fibration f(cid:48) : Y (cid:48) → P1. There are two special fibers which corresponds to (hi = 0) for i = 1, 2. Let Y → Y (cid:48) be the blow up at the nodal points of (hi = 0). It ends up with the rational elliptic surface f : Y → P1 with the following dual graph of divisors. (−4) A1 (−1) B1 E1 E2 E3 E4 E5 E6 E7 E8 E9 B2 (−1) A2 (−4) Figure 3.1. Divisors on Y and their intersections Here, Ai is the proper transform of (hi = 0) along Y → P2, and Bi is the exceptional divisor obtained by the blowing up Y → Y (cid:48). Also, E1, . . . , E9 are the divisors which appear in the blowing up Y (cid:48) → P2. An edge between two nodes implies that corresponding divisors meets with intersection number 1. For example, it can be read (Ai.Bi) = 2 from the graph. From Y , we can produce Enriques surfaces as follows. 4 (1, 1) singularities. Step 1. Contract A1 and A2 to gain a singular surface X with two 1 Step 2. Consider a Q-Gorenstein smoothing X /(0 ∈ ∆) of X, i.e. a proper flat morphism X → (0 ∈ ∆) such that X0 (cid:39) X and Xt is smooth for general t ∈ ∆ \ {0}. Step 3. For general t ∈ ∆ \ {0}, S := Xt is an Enriques surface. One possible way to perform a blow up S is to blow up the surface Y in advance, and proceeds to Steps 1 -- 3 described above. Hence, we add: Step 0. Blow up a point in Y \ (A1 ∪ A2 ∪ B1 ∪ B2 ∪ E1 ∪ . . . ∪ E9). By a slight abuse of notations, we keep call the resulting surface Y and the respective divisors A1, A2, B1, B2, E1, . . . , E9. Also, let E0 be the exceptional divisor produced in Step 0. Notations 3.1. Let π : Y → X be the contraction of A1, A2, let p : Y → P2 be the blow down morphism, and let H ⊂ P2 be a line. Then, Pic Y is the free abelian group generated by (cid:8)p∗H, E0, E1, E2, . . . , E9, B1, B2 (cid:9). 4 YONGHWA CHO Also define divisors Q := p∗(2H), (cid:96)i := p∗H − Ei for i = 1, . . . , 9, and  Di = 0 −(cid:96)i + E0 + B1 −B1 + E0 −Q + 3E0 + 2B1 2D11 i = 0 1 ≤ i ≤ 9 i = 10 i = 11 i = 12 . The push forward along π defines the Q-Cartier Weil divisors on X. Our main claim in this article is the following. Theorem 3.2. Assume h1, h2 are general. There exist divisors Dg correspond to Di, such that OS(Dg 0), OS(Dg 1), . . . , OS(Dg 12) i ∈ Pic S (i = 0, . . . , 12), which is an exceptional collection in Db(S). Our aim is to find divisors on S which are comparable to the divisors on X. One of the natural attempts is to find a line bundle L ∈ PicX so that the line bundles L(cid:12)(cid:12)S and L(cid:12)(cid:12)X share some information through L. Unfortunately, this cannot be done for some line bundles on S. 4. The proof Proposition 4.1. There exists a short exact sequence 0 → Pic S → Cl X → H1(M1, Z) ⊕ H1(M2, Z), where Mi is the Milnor fiber of the smoothing (Pi ∈ X )/(0 ∈ ∆). In particular, the image of Pic S → Cl X is exactly the set of Weil divisors DX on X such that (D.Ai) ≡ 0 (mod 2) where D ∈ Pic Y is a proper transform of DX along π. i = 1, 2 Proof. For the proof, we refer to the arguments in [1, §3.1]. (cid:3) Indeed, we have divisors on Y which satisfies (D.Ai) ≡ 0 (mod 2) but π∗D is not Cartier, thus it is impossible to find L ∈ PicX such that L(cid:12)(cid:12)X = O(π∗D). We need a workaround, which modifies the total space X so that it is able to assign a line bundle on modified family X to each line bundle on S. This birational modification trick is developed in [4, §3]. There are two singularities, say P1, P2 in X, and these are isomorphic to (0 ∈ C2 4 (1, 1)). By the change of variables x = u2, y = v2, z = uv, we get u,v/ 1 (Pi ∈ X) (cid:39) (0 ∈ (xy = z2)) ⊂ (0 ∈ C3 x,y,z/ 1 2 (1, 1, 1)). The versal Q-Gorenstein deformation is given by X ver := (xy = z2 + t) ⊂ C3 x,y,z/ (1, 1, 1) × ∆ver t , 1 2 t t ⊂ C is a small complex disk centered at the origin. Locally, the ambient space C3 where ∆ver ∆ver Z4 inside the lattice N = Z4 +Z· 1 to Σ. The resulting toric variety C admits the birational morphism Φ(cid:48) : C → C3 2 (1, 1, 1)× can be identified to the toric variety associated with the fan Σ generated by the standard basis of 2 (1, 1, 1, 2) 2 (1, 1, 1) × ∆ver . 2 (1, 1, 1, 2). Let Σ be the fan obtained by adding the ray Z· 1 x,y,z/ 1 x,y,z/ 1 t ON DERIVED CATEGORIES OF NONMINIMAL ENRIQUES SURFACES 5 Let X ver be the proper transform of X ver and let Φver : X ver → X ver be the birational morphism induced by Φ(cid:48). Proposition 4.2 (cf. [4, §3]). Let Φ : X → X be the pullback of Φver along X /(0 ∈ ∆) → X ver/(0 ∈ ∆ver). Then, X satisfies the following properties; (1) Over Pi, Wi := Φ−1(Pi) is a projective plane and Φ is an isomorphism outside {P1, P2}. (2) The central fiber over 0 ∈ ∆ is the union Y ∪ W1 ∪ W2, where Y is the rational elliptic surface introduced above, and the scheme-theoretic intersection Wi ∩ Y is realized as Ai in Y , while it is a conic in Wi. then since Ai is a conic curve in Wi, OY (D)(cid:12)(cid:12)Ai Now, suppose that a divisor D ∈ Pic Y satisfies the conditions in Proposition 4.1. Let 2di := (D.Ai), . Let D0 be the glueing of OY (D), OW1(d1), (cid:39) OWi(di)(cid:12)(cid:12)Ai OW2(d2), i.e. the kernel of OY (D) ⊕ OW1(d1) ⊕ OW2 (d2) → OA1 (2d1) ⊕ OA2(2d2), (s, s1, s2) (cid:55)→ (s − s1, s − s2). Then, it can be easily proved that D0 is an exceptional line bundle on the reducible surface X0 = Y ∪ W1 ∪ W2. It is well-known that an exceptional vector bundle extends to a small neighborhood of = D0. Now, deformation, thus shrinking ∆, we can say that there exists a line bundle D such that D(cid:12)(cid:12) X0 D(cid:12)(cid:12)S is a line bundle on S. Notations 4.3. For a divisor D ∈ Pic Y as above, the divisor associated with the line bundle D(cid:12)(cid:12)S is denoted by Dg. Since D is a flat family of line bundles, we have χ(Dg) = χ( D0). The latter can be computed via the short exact sequence 0 → D0 → OY (D) ⊕ OW1 (d1) ⊕ OW2(d2) → OA1(2d1) ⊕ OA2 (2d2) → 0. By Riemann-Roch formula, χ( D0) = χ(D) + 1 2 d1(d1 − 1) + d2(d2 − 1). 1 2 Lemma 4.4. For any divisor D ∈ Pic Y such that there exists an associated divisor Dg ∈ Pic S, (4.2) (4.3) (Dg.KS) = (D.E0). Proof. Since KS is numerically equivalent to Eg Roch formula, one can compute 0, it suffices to prove that (Dg.Eg 0) = (D.E0). By Riemann (Dg.Eg 0) = χ(Dg + Eg 0) − χ(Dg). Let D0 be the line bundle obtained by glueing OY (D), OW1(d1), OW2(d2), and E0 be the line bundle obtained by glueing OY (E0), OW1, OW2. Then, using the formula (4.3) we get (Dg.Eg 0) − χ(Dg) = χ( D0 ⊗ E0) − χ( D0) 0) = χ(Dg + Eg =(cid:0)χ(D + E0) + d1(d1 − 1) + = χ(D + E0) − χ(D) = (D.E0). 1 2 1 2 d2(d2 − 1)(cid:1) −(cid:0)χ(D) + + d1(d1 − 1) + 1 2 1 2 d2(d2 − 1)(cid:1) (cid:3) 6 YONGHWA CHO Combining all together, we get the intersection formula: (Dg)2 = (D.E0) + 2χ(D) + d1(d1 − 1) + d2(d2 − 1) − 2. Note that all the information from the right hand side can be read in Y . Now, it is just a matter of computations to derived the following intersection table: let 1 ≤ i (cid:54)= j ≤ 9 Qg 22 10 10 3 0 (cid:96)g i 10 1 Eg j Bg (cid:96)g 0 3 10 0 2 3 1 0 3 2 1 0 1 1 0 0 0 0 0 −1 Qg (cid:96)g i (cid:96)g j Bg 1 Eg 0 (4.4) 1 − Bg Note that Bg Proposition 4.5. The intersection matrix of divisors {Dg 2 is numerically trivial, so the above table also includes the intersections involving Bg 2. (cid:0)(Dg j )(cid:1) i .Dg 1≤i,j≤11 =  −1 ... 0 0 0 0 i }i=1,...,11 is given by ··· ... ... ··· −1 ··· 0  ... 1 0 . Also, it holds that KS =num Dg relation. 1 + . . . + Dg 10 − 3Dg 11, where =num indicates the numerical equivalence i=1 (cid:96)g i=1 Dg i − 3Dg and(cid:80)9 Proof. The intersection table is obtained immediately from (4.4). Furthermore, 0) − 3(−Qg + 2Bg 1 + 10Eg 1 + Eg i + 2Bg 0, 0 where Ag = 3Qg + Ag (cid:80)10 i=1 (cid:96)i =(cid:0)p∗(6H)(cid:1)g 1, we have(cid:80)10 11 = (−(cid:80)9 = 3Qg −(cid:80)9 +(cid:0)p∗(3H) −(cid:80)9 i=1 Ei)(cid:1)g (cid:80)10 i=1 Dg i − 3Dg S induced by the elliptic fibration of Y → P1. This leads to 11 = −Ag 0 =num KS. i − 3Dg 11 = Eg i + 8Bg i=1 (cid:96)g 0 + 2Bg 1 + Eg 0. 0 = 2Bg Since Ag Corollary 4.6. For 0 ≤ j < i ≤ 12, χ(−Dg i=1 Dg i + Dg j ) = 0. 1 + 3Eg 0) 0 is the general elliptic fiber of (cid:3) To prove Theorem 3.2, we have to prove that hp(−Dg Proof. This is an immediate consequence of Proposition 4.5 and Riemann-Roch theorem. (cid:3) j ) = 0 for each p and 0 ≤ j < i ≤ 12. By the above corollary, it suffices to prove only for p = 0, 2. By Serre duality, h2(−Dg i −Dg j ), thus understanding h0 of divisors is enough to prove Theorem 3.2. This can be done using the short exact sequence (4.2); it is easy to see that h0(OWi (di)) → h0(OAi(2di)) is always surjective, thus j ) = h0(KS +Dg i +Dg i +Dg h0(Dg) ≤ h0( D0) = h0(D) + h0(OW1(d1)) + h0(OW2(d2)) − h0(OA1(2d1)) − h0(OA2(2d2)). (4.5) Before proceed to the proof, we give one remark explaining why we need to use a computer-based approach. ON DERIVED CATEGORIES OF NONMINIMAL ENRIQUES SURFACES 7 Example 4.7. The divisor −Dg p∗H − E9 − E0 − B1. Since (D.A1) = 0 and (D.A2) = 2, 9 + Dg 0 = (p∗H − E9 − E0 − B1)g can be obtained by deforming D := h0(Dg) ≤ h0(D) + h0(OW1) + h0(OW2(1)) − h0(OA1) − h0(OA2(2)) = h0(D). Because of its divisor form, h0(D) depends on the configuration of (hi = 0). Indeed, h0(D) = 1 if the points p(E9), p(E0) and p(B1) are colinear and is zero otherwise. If these three points are colinear, we just have h0(Dg) ≤ 1, thus we cannot see the desired vanishing. In this simple example, we can present the possible values of h0(D) together with exact criterion, but it is getting complicated for other pairs. For example, to conclude h0(−Dg 10) = 0, we will see that it suffices to prove that h0(D) = 0 where 12 + Dg p∗(16H) − 4(cid:80) i≤9 Ei − 6E0 − 6A2 − 6B2. This means that we have to show there is no plane curve of degree 16 which passes through p(E1), . . . , p(E9) 4 times for each, p(E0) 6 times, p(A2) 6 times, and p(B2) 6 times. Lemma 4.8. Let D be a divisor of the form p∗(dH) − (positive sum of E0, E1, . . . , E9, B1, B2). (4.6) For given particular h1, h2, assume h0(D) = N. Then, h0(D) ≤ N for general h1, h2. of plane curves passing through the prescribed positions. Consider the homogeneous equation(cid:80) Proof. As explained in Example 4.7, counting h0(D) reduces down to count the dimension of the space α cαxα of degree d, where α = (αx, αy, αz) is a 3-tuple with αx + αy + αz = d and xα = xαx yαy zαz. Then the positional conditions given by D will imposes linear conditions on (cα), thus we get a system of linear equations, say M c = 0. Now, h0(D) is the dimension of the space of solutions of this system. If we perturb h1, h2 then it perturbs M, but the rank of matrices is a lower-semicontinuous function, so (cid:3) dim ker M = h0(D) is upper-semicontinuous with respect to h1, h2. Now it suffices to prove Theorem 3.2 for the particular choice of h1, h2. However, since we use computer-based approach, we still need another obstruction to overcome. Imagine the situation that we are given a divisor D of the form (4.6), and h1, h2 are explicitly given. We have to tell the computer the conditions imposed by the plane curve p∗D as an ideal sheaf of P2. The problem is that it is extremely hard (perhaps impossible) to find an example of h1, h2 such that the points corresponding to E1, . . . , E9, B1, B2 are defined over a subfield of C which is solvable by radicals over Q. Unless this is possible, we cannot define the explicit ideals in computers. This problem can be resolved by observing some symmetric nature between E1, . . . , E9. In the end, we will see that it suffices to find cubics such that only p(E9), p(B1), p(B2) are defined over Q. Lemma 4.9. Assume h1, h2 define general nodal plane cubics. Let D be a divisor in the rational elliptic surface Y , and assume that in the expression of D in terms of the Z-basis {p∗H, E1, . . . , E9, A2, B2, B3}, the coefficients of E1, . . . , E9 are same. Then, hp(D + Ei) = hp(D + Ej) for any p and any 1 ≤ i, j ≤ 9. Proof. The statement is a slight variation of [1, Lemma 5.8], so we do not give a precise proof here. The main idea is the following: first, pick h1, h2 so that (1) the point p(E0), the cubics h1, h2 ∈ Q[x, y, z], and the nodes of them are defined over Q; 8 YONGHWA CHO (2) the ideal (h1, h2) is prime in Q[x, y, z]; (3) the points in the intersection is written as the form [xi, yi, 1] ∈ P2C, where x1, . . . , x9 are Galois con- jugate to each other, y1, . . . , y9 are Galois conjugate to each other, and the irreducible polynomials of xi and yj are not the same. Now, we take τ ∈ Aut(C/Q) such that τ (xi) = xj. By condition (1), Y is indeed defined over Q, i.e. there exists a variety YQ defined over Q such that Y = YQ ×Q C. Consider the automorphism τY := IdYQ × τ of Y . Then, τY permutes E1, . . . , E9 (sending Ei to Ej), and fixes E0, B1, B2. Thus τY fixes D. In particular, hp(D + Ej) = hp(τ∗ Y (D + Ej)) = hp(D + Ei). (cid:3) By Lemmas 4.8 and 4.9, the main part of the proof reduces to the following statement: Proposition 4.10. There exists h1, h2 ∈ C[x, y, z] such that OS(Dg 0), OS(Dg 9), OS(Dg 10), OS(Dg 11), OS(Dg 12) is an exceptional collection in Db(S). Proof. We choose h1 = (y − z)2z − x3 − x2z and h2 = x3 − 2xy2 + 2xyz + y2z, and pick p(E0) = [4, 9, 6]. We denote the number hp(−Dg 1 is nef, hence any divisor Dg ∈ Pic S with (Dg.Bg i − Dg 1) > 0 implies h2 i +Dg 1) < 0 must have vanishing h0. Also, by Serre duality, (Dg ij = 0. Using this criterion, we get the vanishing of the following numbers: i,j. First of all, we observe that the divisor Bg j ) by hp j . Bg In what follows, we prove hp D ∈ Pic Y such that D deforms to either −Dg h2 9,0, h0 11,0, h2 11,9, h2 10,9, h2 ij = 0 for i, j ∈ {0, 9, 10, 11, 12} with j < i by taking suitable divisor 11,10, h2 12,10, h2 12,9, h2 12,0, h2 12,11. i + Dg j of KS + Dg i − Dg j , and by using (4.5). choice of D p∗(H) − E9 − E0 − B1 p∗(H) − E9 + E0 − B1 − B2 result h0 9,0 = 0 h2 10,9 = 0 h0 11,0 = 0 h0 11,9 = 0 h0 h0 12,0 = 0 h0 12,9 = 0 h0 12,11 = 0 same D as in h0 h0 p∗(5H) −(cid:80) p∗(4H) −(cid:80) 11,10 = 0 p∗(5H) −(cid:80) p∗(16H) − 4(cid:80) p∗(12H) − 3(cid:80) 12,10 = 0 p∗(10H) − 3(cid:80) 11,0 i≤9 Ei − 3E0 − 2B1 − 2B2 i≤8 Ei − 2E0 − B1 − 2B2 i≤9 Ei − 2E0 − 3B1 − 2B2 i≤9 Ei − 6E0 − 6B1 − 6B2 i≤8 Ei − 2E9 − 5E0 − 6B1 − 4B2 i≤9 Ei − 5E0 − 5B1 − 6B2 For the first two in the table, it is easy to verify h0(D) = 0 if the triples ( p(E9), p(E0), p(B1) ) and ( p(E9), p(B1), p(B2) ) are not colinear. For the rest part of the table, we use computer to find h0(D) = 0. (cid:3) The Macaulay2 scripts can be found in [2]. i,j = 0 for 1 ≤ j < i < 9. This can be easily Proof of Theorem 3.2. The only thing remains to prove is hp shown since (4.5) reads i,j ≤ h0(−Ei + Ej) = 0, h0 i,j ≤ h0(KY + E1 − E2) = 0. h2 ON DERIVED CATEGORIES OF NONMINIMAL ENRIQUES SURFACES Lemmas 4.8, 4.9 and Proposition 4.10 imply that 1), . . . , OS(Dg 12) is an exceptional collection after a slight perturbation of h1 and h2. 0), OS(Dg OS(Dg 9 (cid:3) Acknowledgement. The main part of this work has been done when the author was visiting Universität Bayreuth. He would like to thank the members of Lehrstuhl Mathematik VIII -- Algebraische Geometrie for their hospitality and for providing conducive working atmosphere during his visit. References [1] Y. Cho and Y. Lee, Exceptional collections on Dolgachev surfaces associated with degenerations. preprint. arXiv:1506.05213v3. [2] , Macaulay2 scripts for the paper "on derived categories of nonminimal Enriques surfaces". https://sites. google.com/site/yhchoag/home/repository/. [3] S. Galkin and E. Shinder, Exceptional collections of line bundles on the Beauville surface, Adv. Math. 244 (2013), 1033 -- 1050. [4] P. Hacking, Exceptional bundles associated to degenerations of surfaces, Duke Math. J. 162 (2013), no. 6, 1171 -- 1202. [5] Y. Lee and J. Park, A simply connected surface of general type with pg = 0 and K2 = 2, Invent. Math. 170 (2007), no. 3, 483 -- 505. [6] D. Orlov, Projective bundles, monoidal transformations, and derived categories of coherent sheaves, Russian Acad. Sci. Izv. Math. 41 (1993), 133 -- 141. [7] C. Vial, Exceptional collections, and the Néron-Severi lattice for surfaces, Adv. Math. 305 (2017), 895 -- 934. Department of Mathematical Sciences, KAIST, 291 Daehak-ro, Yuseong-gu, Daejeon 305-701, Korea E-mail address: [email protected]
1012.1437
5
1012
2011-05-10T07:26:01
Tate properties, polynomial-count varieties, and monodromy of hyperplane arrangements
[ "math.AG", "math.AT" ]
The order of the Milnor fiber monodromy operator of a central hyperplane arrangement is shown to be combinatorially determined. In particular, a necessary and sufficient condition for the triviality of this monodromy operator is given. It is known that the complement of a complex hyperplane arrangement is cohomologically Tate and, if the arrangement is defined over $\Q$, has polynomial count. We show that these properties hold for the corresponding Milnor fibers if the monodromy is trivial. We construct a hyperplane arrangement defined over $\Q$, whose Milnor fiber has a nontrivial monodromy operator, is cohomologically Tate, and has not polynomial count. Such examples are shown not to exist in low dimensions.
math.AG
math
TATE PROPERTIES, POLYNOMIAL-COUNT VARIETIES, AND MONODROMY OF HYPERPLANE ARRANGEMENTS ALEXANDRU DIMCA1 Abstract. The order of the Milnor fiber monodromy operator of a central hy- perplane arrangement is shown to be combinatorially determined. In particular, a necessary and sufficient condition for the triviality of this monodromy operator is given. It is known that the complement of a complex hyperplane arrangement is coho- mologically Tate and, if the arrangement is defined over Q, has polynomial count. We show that these properties hold for the corresponding Milnor fibers if the mon- odromy is trivial. We construct a hyperplane arrangement defined over Q, whose Milnor fiber has a nontrivial monodromy operator, is cohomologically Tate, and has not polynomial count. Such examples are shown not to exist in low dimensions. 1 1 0 2 y a M 0 1 ] . G A h t a m [ 5 v 7 3 4 1 . 2 1 0 1 : v i X r a Contents Introduction 1. 2. Proof of Theorem 1.4 and of Corollary 1.7 3. Proof of Theorem 1.1 and of Corollary 1.8 4. A purity result and the key example 5. Finite field computations References 1 6 10 11 13 17 1. Introduction Let A be a central arrangement of d hyperplanes in Cn+1, with d ≥ 2 and n ≥ 1, given by a reduced equation Q(x) = 0. Consider the corresponding global Milnor fiber F defined by Q(x) − 1 = 0 in Cn+1 with monodromy action h : F → F , h(x) = exp(2πi/d) · x. A general investigation line is to check whether certain properties of the associated projective hyperplane arrangement complements M(A) in Pn extend to the Milnor fiber F . For instance, note the following. 2000 Mathematics Subject Classification. Primary 32S22, 32S35; Secondary 32S25, 32S55. Key words and phrases. hyperplane arrangement, Milnor fiber, monodromy, polynomial-count, cohomologically Tate. 1 Partially supported by the ANR-08-BLAN-0317-02 (SEDIGA). 1 2 ALEXANDRU DIMCA (i) The cohomology ring H ∗(M(A), Z) is determined by the combinatorics, see [20], but the same question for the Betti numbers of the Milnor fiber F is widely open, see for instance [16]. (ii) H ∗(M(A), Z) is torsion free, see [20], and there is the open question about the torsion freeness of H ∗(F, Z), see [3]. (iii) The complement M(A) is formal in the sense of Sullivan [28], [4], and it was recently shown that the Milnor fiber F may be not even 1-formal, see [30]. We say that a complex variety Y is cohomologically Tate if for any cohomol- ogy group H m(Y, C), one has the following vanishing of mixed Hodge numbers: hp,q(H m(F, C)) = 0 for p 6= q. The fact that the hyperplane arrangement com- plements M(A) are cohomologically Tate is known for a long time: any cohomology group H m(M(A), Q) is a pure Hodge structure of type (m, m), see [14], [12], [25]. When the monodromy action h∗ is trivial on all the cohomology groups H ∗(F, C), it follows that we have an equality H m(F, Q) = H m(M(A), Q) for any 0 ≤ m ≤ n, and hence in this case F is cohomologically Tate. One may ask whether this is the only possibility for a hyperplane arrangement Milnor fiber F to be cohomologically Tate. The claim that F cohomologically Tate implies h∗ trivial is shown to be true in the case n = 1 (obvious, using the MHS on the Milnor fiber of an isolated homogeneous hypersurface singularity, given by Steenbrink in [27] and recalled in [5], pp.243-244) and n = 2, i.e. for plane arrangements, and negative in general. We do not know whether there is a similar result to Theorem 1.1 for 2 < n < 7, and this explains why we go to dimension 7 to construct our example. To state our results in this direction, we need some preliminaries. In studying the cohomology H ∗(F, Q) of the Milnor fiber, the monodromy action h∗ : H ∗(F, Q) → H ∗(F, Q) or the number of points in F (Fp), we can, without any loss of generality, suppose that the arrangement A is essential, i.e. ∩H∈AH = 0. This is the same as supposing that the polynomial Q involve in an essential way all the variables x0, ..., xn, i.e. one can not choose the coordinates x on Cn+1 such that Q(x0, ..., xn) = R(x0, ..., xu) for some 0 ≤ u < n and a homogeneous polynomial R ∈ C[x0, ..., xu] . The properties of the monodromy h∗ : H ∗(F, Q) → H ∗(F, Q) are rather misterious, and many things that we know in general are related to the spectrum (1.1) mαtα, with mα = Pj(−1)j−n dim Grp Sp(A) = Xα∈Q H j(F, C)β where p = [n+1−α] and β = exp(−2πiα), which is combinatorially determined, see [1]. Surprinsingly, note that for most ar- rangements the situation is rather simple, namely hm : H m(F, Q) → H m(F, Q) is trivial (i.e. the identity) for 0 ≤ m < n and dim H n(F, C)β = χ(M(A)), for any β ∈ µd = {z ∈ C zd = 1} with β 6= 1, see for instance [2], [15], as well as Prop. 2.5.4, Prop.6.4.6, Example 6.4.14 and Theorem 6.4.18 in [6]. F Theorem 1.1. Let A be an essential central arrangement of d planes in C3. The following conditions are equivalent. TATE PROPERTIES, POLYNOMIAL-COUNT VARIETIES, AND ARRANGEMENTS 3 (i) The mixed Hodge numbers hp,q(H 2(F, C)) vanish for p 6= q. (ii) The arrangement A is reducible. (iii) The monodromy action h∗ is trivial on all the cohomology groups H ∗(F, C). (iv) The spectral numbers mα in Sp(A) vanish for all α ∈ (0, 1). Geometrically, the property (ii) means the following: in the projective line arrange- ment A′ associated to A, (d − 1) lines meet in one point, say A, and the remaining line Ld does not contain A. In terms of coordinates, this means that one may choose the coordinates (x : y : z) on P2 such that A = (0 : 0 : 1) and Ld : z = 0. With this choice one has Q(x, y, z) = Q1(x, y)z, where Q1 is a degree (d − 1) reduced homogeneous polynomial in x, y. This property is exactly the definition of a reducible arrangement when n = 2. To discuss the higher dimensional case we need the following precise characteriza- tion of arrangements with a trivial monodromy. Theorem 1.2. For an essential central arrangement A, the following conditions are equivalent. (i) The monodromy action h∗ is trivial on all the cohomology groups H ∗(F, C). (ii) The arrangement A is reducible and satisfies the following: if A = A1 × ... × Aq is the decomposition of A as a product of irreducible arrangements and if dj = Aj denotes the number of hyperplanes in Aj, then G.C.D.(d1, ..., dq) = 1. Moreover, if A is defined over Q (i.e. each hyperplane in A is defined over Q), then all the irreducible arrangements Aj are also defined over Q. We show below, see Lemma 2.1, that in the (unique) decomposition A = A1 × ... × Aq of A as a product of irreducible arrangements Aj, the integers q, d1,...,dq are determined by the combinatorics, i.e. by the intersection lattice L(A), see [21]. We recall that a central arrangement A as above is reducible if one can choose the coordinates x on Cn+1 such that Q(x0, ..., xn) = R1(x0, ..., xu)R2(xu+1, ..., xn) for some 0 ≤ u < n and homogeneous non-constant polynomials R1 and R2. We write then: A = A1 × A2, with Aj : Rj = 0. It is known that an essential arrangement A is reducible if and only if χ(M(A)) = 0, see [24]. On the other hand, a trivial monodromy action h∗ implies χ(M(A)) = 0, as a simple consequence of the following general formula (1.2) (−1)j dim H j(F, C)β = χ(M(A)) Xj for any β ∈ µd, see Prop. 2.5.4 and Prop.6.4.6 in [6]. See also (1.4.2) in [1]. Remark 1.3. All compact Kahler manifolds are formal spaces, see [4]. When the monodromy h∗ is trivial, the corresponding Milnor fiber is clearly a formal space in Sullivan's sense, see [28], [9]. Hence Theorem 1.2 yields a wealth of new examples of formal spaces in the class of smooth affine varieties. Note also that there are examples of non-formal smooth affine surfaces, either related to Milnor fibers of central plane arrangements, see [30], or to isolated weighted homogeneous singularities [10]. 4 ALEXANDRU DIMCA Theorem 1.2 is an obvious consequence of the following Thom-Sebastiani type result, where the sum of polynomials in disjoint sets of variables is replaced by their product. This result plays also a key role in the construction of our examples below. Theorem 1.4. Let A = A1 × ... × Aq be the decomposition of the central essential arrangement A as a product of irreducible arrangements, let dj = Aj denotes the number of hyperplanes in Aj and let d0 = G.C.D.(d1, ..., dq). Then the following hold. (i) There is a natural identification of graded MHS defined over R H ∗(F, C) = H ∗(T, C) ⊗ (⊕β∈µd0 (H ∗(F1, C)β ⊗ · · · ⊗ H ∗(Fq, C)β)). More precisely, for any β ∈ µd0, there is an identification H ∗(F, C)β = H ∗(T, C) ⊗ H ∗(F1, C)β ⊗ · · · ⊗ H ∗(Fq, C)β. (ii) The monodromy operator h∗ : H ∗(F, C) → H ∗(F, C) has order d0, which is determined by the intersection lattice L(A). Since each Aj is irreducible, it follows from (1.2) that each H ∗(Fj)β which occur in Theorem 1.4 is nonzero. Remark 1.5. There are a number of papers dealing with Thom-Sebastiani type results for the product of two polynomials f and g (or, more generally, for h(f, g), with h a function of two variables), see for instance Oka [18], Sakamoto [22], N´emethi [17] and Tapp [29]. The decomposition in our Theorem 1.4, (i), appears in Tapp [29]. However, in all of these papers there is no reference to the mixed Hodge structures involved, and the methods used, being purely topological, do not allow in a direct way to derive such conclusions. Since these MHS play a central role in the sequel of our paper, we give below a proof of this decomposition rather different from that in [29], carefully handling the corresponding MHS. We have noticed already that F is a cohomologically Tate variety when the mon- odromy action h∗ is trivial. We give below an example showing that the converse claim is false in general, see Example 4.3. This Example is also interesting since it shows that the part H <top(F, C)6=1 of the Milnor fiber cohomology can be rather big, unlike all the previously known examples. Corollary 1.6. Consider the central rational hyperplane arrangement Au,v in Cn+1 defined in Example 4.3, for any u, v ∈ Z>0 . Then n = 3u + 5v − 1, the only eigenvalues of the monodromy operator h∗ are ±1 and dim H ∗(Fu,v, C)−1 = 2u+v−1. More precisely, the nontrivial (−1)-eigenspaces are exactly H 2u+4v+j(Fu,v, C)−1 for 0 ≤ j ≤ u + v − 1 and dim H 2u+4v+j(Fu,v, C)−1 = (cid:18)u + v − 1 j (cid:19). The smallest n for which our construction yields a counterexemple is n = 7; the Milnor fiber of the corresponding hyperplane arrangement A1,1 is cohomologically Tate, but h∗ is not trivial. TATE PROPERTIES, POLYNOMIAL-COUNT VARIETIES, AND ARRANGEMENTS 5 Recall that the (compactly supported) Hodge-Deligne polynomial (or E-polynomial) associated to a complex variety Y is given by (1.3) HDX(x, y) = Xu,v (Xj (−1)jhu,v(H j c (Y, C)))xuyv. Assume that Y is in fact defined over Q. We say that Y has polynomial count with count polynomial PY , if there is a polynomial PY ∈ Z[t], such that for all but finitely many primes p, for any finite field Fq with q = ps and s ∈ N∗, the number of points of Y over Fq is precisely PY (q). As an example, if the hyperplane arrangement A is defined over Q , then M(A) has polynomial count, see for instance [7], section (5.3) or [26], Theorem 5.15. The above rationality assumption is essential: indeed, the line arrangement A : x2 + y2 = 0 is defined over Q(i), and the corresponding complement M(A) satisfies M(A)(Fp) = p − 1 if p ≡ 1 mod 4 and M(A)(Fp) = p + 1 if p ≡ 3 mod 4. Note that in this case Q(x) has integer coefficients, A is reducible, but the corresponding splitting A = A1 × A2 is not defined over Q. A theorem of N. Katz in [11] says that if Y has polynomial count with count polynomial PY , then HDY (x, y) = PY (xy). See Theorem 2.1.8 in [11], and the remark at the bottom of page 563 and Example 2.1.10 explaining that there can be allowed finitely many 'bad characteristics' p. In particular, such a variety is not too far from being cohomologically Tate, i.e. the not Tate part of the cohomology should cancel out in HDY (x, y). One may ask what happens to the Milnor fibers of hyperplane arrangements. Again we need a rationality assumption: the Milnor fiber F : x2 + y2 = 1 satisfies F (Fp) = p − 1 if p ≡ 1 mod 4 and F (Fp) = p + 1 if p ≡ 3 mod 4. Note that in this example the monodromy operator h∗ is trivial and F is cohomologically Tate. When A is defined over Q, we will always choose the defining equation Q(x) = 0 with integer coefficients. It turns out that the arithmetic properties of the corre- sponding Milnor fiber F : Q(x) − 1 = 0 do not depend on the choice of Q, see for instance Corollary 1.7. So a first naive idea when trying to construct cohomologically Tate Milnor fibers of rational arrangements which have not polynomial count is to look for a hyperplane arrangement A with a trivial monodromy h∗ and such that the corresponding Milnor fiber F has not polynomial count. Such an attempt cannot succeed in view of the following result, implied by Theorem 1.2. Corollary 1.7. If the monodromy action h∗ is trivial on all the cohomology groups H ∗(F, C) of the Milnor fiber F of an essential central arrangement A defined over Q, then F has polynomial count, with the same count polynomial as the corresponding projective complement M(A). In particular, this property of the Milnor fiber of having polynomial count, and the corresponding count polynomial, do not depend on the choice of defining equation Q in Z[x]. For n = 2 we have the following. 6 ALEXANDRU DIMCA Corollary 1.8. Consider the following conditions. (i) Y has polynomial count. (ii) Y is cohomologically Tate. If Y is a smooth affine surface, then (i) =⇒ (ii). If in addition Y is the Milnor fiber of a central rational plane arrangement in C3, then one also has (ii) =⇒ (i). The hyperplane arrangement A1,1 introduced in Example 4.3 is defined over Q, and the corresponding Milnor fiber is cohomologically Tate, but has not polynomial count, see Theorem 5.2. We would like to thank Nero Budur, Denis Ibadula, Mark Kisin, Gus Lehrer and Morihiko Saito for useful discussions in relation to this work. Special thanks are due to Gabriel Sticlaru who provided expert help in some numerical experiments with rather large numbers, see Remark 5.4. 2. Proof of Theorem 1.4 and of Corollary 1.7 Proof. We can always, up to a linear coordinate change, write Q(x) = Q1(y1) · · · Qq(yq), where x = (y1, ..., yq) ∈ Cn+1, yj ∈ Cnj such that: n1+...+nq = n+1, dj = deg Qj > 0 and Aj : Qj = 0 an irreducible and essential arrangement in Cnj for j = 1, ..., q. Note that the existence of such a decomposition is equivalent to the following property: there is a partition of the hyperplanes in A in q subsets A1,...,Aq, such that if we define Vj = ∩H /∈Aj H, then there is a direct sum decomposition Cn+1 = V1 + V2 + ... + Vq, and this partition is the finest with this property. More precisely we have the following result. Lemma 2.1. Let A = {Hi}i∈I be a central, essential hyperplane arrangement in Cn+1 = V . Consider the set P of partitions I = I1 ∪ ... ∪ Im of I satisfying the following condition: Pj codim(∩i∈Ij Hi) = dim V . Then the following hold. (i) P is non-empty, since the trivial partition I = I is in P . (ii) If I = I1 ∪ ... ∪ Im and I = I ′ m′ are two partitions in P , then their intersection I = ∪i,jIi,j where Ii,j = Ii ∩ Ij for i = 1, m, j = 1, m′ (the empty intersections Ii,j are discarded), is again a partition in P . 1 ∪ ... ∪ I ′ (iii) In the (unique) decomposition A = A1×...×Aq of A as a product of irreducible arrangements Aj, the integers q, d1,...,dq are determined by the combinatorics, i.e. by the intersection lattice L(A). Proof. The claim (i) is obvious. Let E be the dual of V and, for a given partition I = I1 ∪ ... ∪ Im in P , let Ei be the vector subspace in E spanned by the equations of the hyperplanes in Ii. Since A is essential, it follows that Pj Ej = E. Since dim Ej = codim(∩i∈Ij Hi), it follows that the above sum is in fact a direct sum. When we have two partitions as above, we define Ei,j to be the vector subspace in j. Then it follows immediately E spanned by the equations of the hyperplanes in Ii ∩I ′ that the sum Pj Ei,j = Ei is again direct sum for all i's. Hence the sum Pi,j Ei,j = V is a direct sum, which proves the claim (ii). TATE PROPERTIES, POLYNOMIAL-COUNT VARIETIES, AND ARRANGEMENTS 7 Since the set of partitions P is defined only in terms of the intersection lattice L(A), it follows that the unique minimal element of P (with respect to the partial order given by refinement), is combinatorially determined. This minimal element is denoted above by I = A1 ∪ ... ∪ Aq. The direct sum decomposition Cn+1 = V1 + V2 + ... + Vq mentionned above is just the dual of the direct sum decomposition Pj Ej = E. Since dj = Aj, they are also determined by the combinatorics. (cid:3) Remark 2.2. When A is defined over Q, it follows that all the vector subspaces Vj are defined over Q. Then the arrangement Aj, which is essentially given by the traces of H ∈ Aj on Vj, is clearly defined over Q. Moreover, the coordinate change from the coordinates x to the coordinates y is defined over Q. This shows that when doing computations for almost all primes p, we may replace the equation Q(x) = 0 (resp. Q(x) = 1) by the corresponding equations Q1(y1) · · · Qq(yq) = 0 (resp. Q1(y1) · · · Qq(yq) = 1). Proof of Theorem 1.4. Let Fj : Qj = 1 and hj : Fj → Fj be the corresponding Milnor fibers and mon- odromy homeomorphisms. Let us consider the corresponding least common multiple m = L.C.M.(d1, ..., dq) and set wj = m/dj for j = 1, ..., q. Our first aim is to obtain a description of the (total) Milnor fiber F in terms of the collection of Milnor fibers F1,...,Fq. For this we consider the affine torus (2.1) T = {t = (t1, ..., tq) ∈ (C∗)q t1t2 · · · tq = 1}. Consider the mapping (2.2) given by (2.3) f : T × F1 × · · · × Fq → F (t, y1, ..., yq) 7→ (tw1 1 y1, ..., twq q yq). It is easy to check that this mapping f is surjective and one has f (t, y1, ..., yq) = f (t′, y′ 1, ..., y′ q) if and only if the points (t, y1, ..., yq) and (t′, y′ the group 1, ..., y′ q) are in the same G-orbit, where (2.4) G = {g = (g1, ..., gq) ∈ µq m g1g2 · · · gq = 1} acts on X = T × F1 × · · · × Fq via (2.5) g · ((t1, ..., tq), y1, ...yq) = ((g−1 1 t1, ..., g−1 q tq), gw1 1 y1, ..., gwq q yq). It follows that F = X/G and in particular H ∗(F, Q) = H ∗(X, Q)G, the G-fixed part of the cohomology of X under the induced G-action. This is an isomorphism of MHS (mixed Hodge structures), since the G-action is algebraic. Note that (2.6) H ∗(X, C) = H ∗(T, C) ⊗ H ∗(F1, C) ⊗ · · · ⊗ H ∗(Fq, C). 8 ALEXANDRU DIMCA Moreover, the group G acts trivially on the factor H ∗(T, C), since T is a connected algebraic group and G ⊂ T. If we set then it follows that H ∗ = H ∗(F1, R) ⊗ · · · ⊗ H ∗(Fq, R) H ∗(F, R) = H ∗(T, R) ⊗ (H ∗)G. The G-action on H ∗ η1 ⊗ · · · ⊗ ηq, then C, the complexification of H ∗, is given by the following: if η = (2.7) gη = (h1)k1(η1) ⊗ · · · ⊗ (hq)kq(ηq). Here g = (λk1, ..., λkq) with λ = exp(2πi/m) and k1 + ... + kq is divisible by m, the kj being otherwise arbitrary integers. Let ηj ∈ H ∗(Fj, C) be now chosen such that for any j = 1, ..., q there is a βj ∈ µdj ⊂ µm with h∗ j ηj = βjηj and look at η = η1 ⊗ · · · ⊗ ηq. Such elements form a C)G is the same as finding all η's of this C-basis of H ∗ form which are fixed under the G-action. By choosing kq = m − k1 − ... − kq−1 we get from (2.7) the following C and hence to determine (H ∗ gη = ( β1 βq )k1 · · · ( βq−1 βq )kq−1η where now there is no condition on the integers k1,...,kq−1. By taking one of them C)G implies β1 = ... = βq. Call this equal to 1 and the rest zero, we see that η ∈ (H ∗ common value λ0 and note that λ0 ∈ µd0 = ∩j=1,qµdj . Conversely, for any λ0 ∈ µd0 and any η ∈ H ∗(F1, C)λ0 ⊗ · · · ⊗ H ∗(Fq, C)λ0, we see by using (2.7) that η ∈ (H ∗ C)G. Moreover, set Eλ0 = H ∗(F1, C)λ0 ⊗ · · · ⊗ H ∗(Fq, C)λ0 and note that Eλ0 = Eλ0. It follows that if we set Mλ0 = Eλ0 when λ0 ∈ R and Mλ0 = Eλ0 + Eλ0 when λ0 /∈ R, then Mλ0 is endowed with a natural R-MHS, see also the formula (4.1). To prove the second part of the claim (i) in Theorem 1.4, we construct a nice G-equivariant lifting h : X → X of the monodromy morphism h : F → F . We set (2.8) h(t1, ..., tq, y1, ..., yq) = (γ1t1, ..., γqtq, β1y1, y2, ..., yq) where β1 = exp(2πi/d1) and γj = exp(2πiaj) with w1a1 = 1/d−1/d1 and wjaj = 1/d for j > 1. Then Pj aj = 0, i.e. (γ1t1, ..., γqtq) ∈ T and f ◦ h = h ◦ f . Then h∗ acts as identity on all the factors in the tensor product (2.6), except on H ∗(F1, C), where it acts via h∗ 1. We get the second part of the claim (i) by using the description of the cohomology H ∗(F, C) given in the first part of the claim (i). To prove the claim (ii), note that the affine torus T acts on the Milnor fiber F by (2.9) t(y1, ..., yq) = (tw1 1 y1, ..., twq q yq). Hence to show that (h∗)d0 is trivial, it is enough to show the existence of an el- ement t = (t1, ..., tq) ∈ T such that twj j = exp(2πid0/d) for j = 1, ..., q. Since TATE PROPERTIES, POLYNOMIAL-COUNT VARIETIES, AND ARRANGEMENTS 9 G.C.D.(d1, ..., dq) = d0, there are integers kj such that (2.10) k1d1 + ...kqdq = (m − 1)d0. For j = 1, ..., q we set (2.11) tj = exp[2πi( d0 dwj + kj wj )]. The relations twj j = exp(2πid0/d) are clearly satisfied. Moreover t1t2 · · · tq = exp[2πi(Xj ( d0dj dm + kjdj m ))] = exp[2πid0( 1 m + m − 1 m )] = 1. Hence t ∈ T, it follows that (h∗)d0 is trivial. We conclude using following fact: since each Aj is irreducible, it follows from (1.2) that each H ∗(Fj)β which occur in Theorem 1.4 is nonzero. Hence each H ∗(F )β is nonzero, i.e. the order of h∗ is indeed d0. (cid:3) Remark 2.3. Consider the central essential hyperplane arrangement A in Cn+1 and its decomposition A = A1 × ....Aq as a product of irreducible arrangements Aj for j = 1, ..., q. Then it is easy to show that M(A) = T × M(A1) × ... × M(Aq). This implies the following for the cohomology of the corresponding projective com- plements H ∗(M(A)) = H ∗(C∗)⊗(q−1) ⊗ H ∗(M(A1)) ⊗ ... ⊗ H ∗(M(Aq)) i.e. the case β = 1 in Theorem 1.4, (iii). See also Tapp [29]. Now we pass to the proof of Corollary 1.7. Since the monodromy h∗ is trivial, we know by Theorem 1.2 that G.C.D.(d1, ...dq) = 1. Hence there exist integers mj such that m1d1 + ...mqdq = 1. Let F be a finite field and consider the mapping Q : Fn+1 → F induced by the polynomial Q. For a ∈ F, denote by F (a) the fiber Q−1(a). Denote also M(A, F) the correspond- ing (projective) hyperplane arrangement complement over F. It is clear that (2.12) Fn+1 \ F (0) = (F − 1) · M(A, F). Consider the following F∗-action on Fn+1: (2.13) t · x = t · (y1, y2, ..., yq) = (tm1y1, tm2y2, ..., tmqyq). The relation Q(t · x) = tQ(x) shows that all the fibers F (a) for a ∈ F∗ have the same cardinal. Since their disjoint union is exactly Fn+1 \ F (0), the equation (2.12) yields (2.14) F (1) = M(A, F). This equality completes the proof of Corollary 1.7. 10 ALEXANDRU DIMCA 3. Proof of Theorem 1.1 and of Corollary 1.8 Proof. It follows from the discussion just after Theorem 1.1 that, assuming (ii), the Milnor fiber F is isomorphic to the complement of the central line arrangement given by Q1 = 0 in C2 (indeed, the only partitions of 3 are 1 + 2 = 2 + 1 = 1 + 1 + 1 = 3). Hence the implication (ii) ⇒ (i) is obviously true. Note that for n = 2 and α ∈ (0, 1), the corresponding spectral number is by definition (3.1) with β = exp(−2πiα). Indeed, h2,2(H 2(F, C)β = 0, as follows from Theorem 1.3 in [8]. mα = h2,0(H 2(F, C)β) + h2,1(H 2(F, C)β) Since H 2(F, C)1 is known to be of type (2, 2), the equivalence of the claims (i) and (iv) in Theorem 1.1 follows. Moreover, the equivalence of the claims (ii) and (iii) follows from Theorem 1.2, using again the partions of 3. So from now on we assume that (iv) holds and we prove (ii). The fact that the arrangement is essential implies that d ≥ 3 and that the lines in A′ do not pass all through the same point, i.e. there is no point s of multiplicity ms = d. To prove (ii) we have to show the existence of a point of multiplicity d − 1. For d = 3 there are only two type of arrangements, described better in terms of their associated projective line arrangements A′: (a) three lines forming a triangle, in which case one may take Q = xyz and F = C∗ × C∗, and (b) three lines meeting at one point. The claim (i) ⇒ (ii) is clear by the previous remark. We assume from now on that d ≥ 4. Next we recall the following key formula from [1], Theorem 3 (rewritten slightly for our needs). If 0 < α = j d < 1, then (3.2) mα = (cid:18)j − 1 2 (cid:19) − Xs;ms≥3 (cid:18)⌈jms/d⌉ − 1 2 (cid:19), where the sum is over all multiple points s in A′ with multiplicity ms ≥ 3. By convention (cid:0)a b(cid:1) = 0 if a < b. If we use the above formula for j = 3, we get that the corresponding vanishing mα = 0 is equivalent to the existence a unique point s of multiplicity ms > 2d/3 in A′. For d = 4 (resp. d = 5), this means a point of multiplicity ms ≥ 3 (resp. ms ≥ 4). As above (case d = 3, (b)), the case ms = 4 (resp. ms = 5) is discarded since A′ is essential. Hence ms = 3 (resp. ms = 4), which gives exactly an arrangement A′ as claimed in (ii). From now on we assume d > 5. We apply the formula (3.2) for j = d − 1. Since one clearly has m − 1 < (d − 1)m d < m, TATE PROPERTIES, POLYNOMIAL-COUNT VARIETIES, AND ARRANGEMENTS 11 it follows that ⌈(d − 1)m/d⌉ = m and hence the vanishing mα = 0 in this case is equivalent to the equality (3.3) (cid:18)d − 2 2 (cid:19) = Xs;ms≥3 2 (cid:19). (cid:18)ms − 1 Similarly, for j = d − 2 we get from mα = 0 the following equality (3.4) (cid:18)d − 3 2 (cid:19) = Xs;3≤ms<d/2 (cid:18)ms − 1 2 (cid:19) + Xs;ms≥d/2 2 (cid:19). (cid:18)ms − 2 By taking the difference of (3.3) and (3.4) we get the equality (3.5) d − 3 = Xs;ms≥d/2 (ms − 2). It follows that, if the set S = {s; ms ≥ d/2} contains exactly one element, then the corresponding multiple point s satisfies ms = d − 1 and we are done. Suppose now that the set S contains at least two elements. Since one of them has to be the point s with multiplicity ms > 2d/3 obtained above for j = 3, we get in this case (3.6) d − 3 > (2d/3 − 2) + (d/2 − 2). This is equivalent to d < 6, a contradiction with our hypothesis d > 5. (cid:3) The proof of Corollary 1.8 is very easy now. If (i) holds, it follows from Katz' Theorem that the Hodge-Deligne polynomial of Y contains only the monomials 1, xy, (xy)2. The possible non-zero mixed Hodge numbers in this situations are: h2,2(H 4 h1,1(H 2 c (Y, C)), h1,2(H 3 c (Y, C)), h0,2(H 2 c (Y, C)), c (Y, C)). c (Y, C)), h2,1(H 3 c (Y, C)), h0,1(H 2 c (Y, C)), h1,1(H 3 c (Y, C)), h1,0(H 2 c (Y, C)), h2,0(H 2 c (Y, C)), h0,0(H 2 It follows that the only cancellations in the Hodge polynomial can occur in the coefficient of xy. Hence all the mixed Hodge numbers hu,v(H m c (Y, C)) vanish for u 6= v, i.e. Y is cohomologically Tate. Now for Milnor fibers of a central plane arrangement we have seen in Theorem 1.1 that (ii) implies that Y is isomorphic to a line arrangement complement in C2, and hence it has polynomial count. 4. A purity result and the key example Let A be a central arrangement of d hyperplanes in Cn+1, with n ≥ 1, given by a reduced equation Q(x) = 0. Then clearly H n(F, Q)1 and H n(F, C)−1 are mixed Hodge substructures in H n(F, Q). Moreover, for β ∈ µd, β 6= ±1, the same is true for the subspace (4.1) H n(F, C)β,β = H n(F, C)β ⊕ H n(F, C)β = ker[(hn)2 − 2Re(β)hn + Id] which is in fact defined over R (as the last equality shows). For β = −1, we set H n(F, C)β,β = H n(F, C)−1 for uniformity of notation. 12 ALEXANDRU DIMCA Let D = Q−1(0) = ∪H∈AH. For a point x ∈ D, x 6= 0, let Lx = ∩H∈A,x∈H H and denote by Ax the central hyperplane arrangement induced by A on a linear subspace Tx, passing through x and transversal to Lx. We may choose dim Tx = codim Lx and identify x with the origin in the linear space Tx. Let h∗ x : H ∗(Fx, C) → H ∗(Fx, C) be the corresponding monodromy operator at x. With this notation we have the following result. Proposition 4.1. Let β ∈ µd, β 6= 1 be a root of unity which is not an eigenvalue for any monodromy operator h∗ x for x ∈ D, x 6= 0. Then the corresponding eigenspace H n(F, C)β,β is a pure Hodge structure of weight n. In particular, if β = exp(−2πiα) for some α ∈ Q, then the coefficients in the corresponding spectrum Sp(A) have the following symmetry property: (4.2) mα = mn+1−α. Proof. This result is a direct consequence of Lemma 3.6 in [23]. Indeed, our hy- pothesis on β implies that the nearby cycle sheaf ψQ,βC is supported at the origin and hence it is identified to H n(F, C)β. This identification in turn implies that the logarithm of the unipotent part of the monodromy N is trivial on ψQ,βC (as this holds for H n(F, C)β, the monodromy h∗ being semisimple). The last claim about the symmetry property in (4.2) is proved in the usual way. In view of the purity result, we have hp,q(H n(F, C)β) = 0 for p + q 6= n. For example, assume that α = j d with 0 < j < d. It follows from the definition of the spectrum (1.1) that mα = dim hn,0(H n(F, C)β), mn+1−α = dim h0,n(H n(F, C)β). The claimed equality follows by considering the action of the complex conjugation on the pure Hodge structure H n(F, C)β,β. (cid:3) Corollary 4.2. Assume that A is a generic central hyperplane arrangement, i.e. the associated projective divisor in Pn is a divisor with normal crossing. Then H n(F, C)6=1 = ⊕β6=1H n(F, C)β is a pure Hodge structure of weight n and for any rational number α ∈ Q \ Z one has mα = mn+1−α. This symmetry of the coefficients of the spectrum Sp(A) of a generic central ar- rangement is alluded to in [23], see the remarks just after Corollary 1 in the Introduc- tion. The result above follows from Proposition 4.1 using the simple fact that in this case all the monodromy operators h∗ x are the identity. For more on the monodromy of generic arrangements, see pp. 209-210 in [21], (3.2) in [1] and section 3 in [2]. Now we pass to our example. Example 4.3. For n ≥ 2 let Gn be the central arrangement in Cn+1 given by the equation Qn(x) = Qn(x0, ..., xn) = x0 · x1 · ... · xn · (x0 + x1 + ... + xn). Hence the degree dn of Qn is n+2 and clearly Gn is a generic, irreducible arrangement. Assume that n = 2k is even and use Theorem 3.2 in [2] (or refer to the original paper TATE PROPERTIES, POLYNOMIAL-COUNT VARIETIES, AND ARRANGEMENTS 13 [19]) to get that (4.3) for 0 ≤ m < n and dim H n(F, C)−1 = 1. H m(F, C) = H m(F, C)1 It follows that all the spectral coefficients in Sp(Gn) corresponding to the mon- odromy eigenvalue −1 vanish except for mk+ 1 = 1 (which is the only auto-dual element in the sum with respect to the symmetry given by Corollary 4.2). It follows that the eigenspace H n(F, C)−1 is spanned by a cohomology class ωn of Hodge type (k, k). 2 Let us return now to the setting of the proof of Theorem 1.2 in Section 2. Let Au,v be the central hyperplane arrangement obtained by taking the product of u > 0 copies of the arrangement G2 and v > 0 copies of G4. In follows that n = 3u + 5v − 1, d = 4u + 6v, q = u + v, d0 = G.C.D.(d1, ..., dq) = 2. In this case, the cohomology of the corresponding total Milnor fiber Fu,v can be described via Theorem 1.4 as the following direct sum H ∗(Fu,v, C) = H ∗(Fu,v, C)1 ⊕ H ∗(Fu,v, C)−1, where (4.4) and (4.5) H ∗(Fu,v, C)1 = H ∗(T, C) ⊗ H ∗(M(G2), C)⊗u ⊗ H ∗(M(G4), C)⊗v H ∗(Fu,v, C)−1 = H ∗(T, C) ⊗ (Cω2)⊗u ⊗ (Cω4)⊗v. It follows that Fu,v is a cohomologically Tate variety of dimension n = 3u + 5v − 1, but the corresponding monodromy action h∗ is not trivial. By choosing various values for u, v we can get n = 7 as a minimal value as well as any integer n ≥ 15. Remark 4.4. If one is interested only in irreducible arrangements, then such exam- ples with large cohomology H <top(F, C) can be obtained by taking a generic hyper- plane section of the arrangement Au,v. 5. Finite field computations In this section we use the following notation. Let F = F1,1 be the variety defined over Z by Q(x) = 1, where Q(x) = x1 · x2 · x3 · (x1 + x2 + x3) · x4 · x5 · x6 · x7 · x8 · (x4 + x5 + x6 + x7 + x8). For any prime p we denote by A(p) the number of points of F over Fp = Z/pZ, i.e. the number of solutions of the equation Q(x) = 1 in F8 p, consider the varieties p. For a ∈ F∗ F1(a) = {(x1, x2, x3) ∈ F3 p x1 · x2 · x3 · (x1 + x2 + x3) = a} and F2(a) = {(x4, x5, x6, x7, x8) ∈ F5 p a · x4 · x5 · x6 · x7 · x8 · (x4 + x5 + x6 + x7 + x8) = 1}. Let n1(a) = F1(a), n2(a) = F2(a) and note that obviously one has (5.1) A(p) = Xa∈F∗ p n1(a)n2(a). 14 ALEXANDRU DIMCA From now on we assume that p is a prime number in the arithmetic progression 12b + 11, where b ∈ N. Lemma 5.1. Consider the group morphism p(k) : F∗ the following hold. p → F∗ p given by t 7→ tk. Then (i) There is an index 2 subgroup H ⊂ F∗ (ii) If a and a′ have the same class in F∗ p such that p(4)(F∗ p) = p(6)(F∗ p) = H. p/H, then n1(a) = n1(a′) and n2(a) = n2(a′). Proof. Since the multiplicative group F∗ p is cyclic, of order p − 1 = 12b + 10, it follows that to prove (i) it is enough to show that p(4)(F∗ p) have both cardinal (p − 1)/2. This is the same as proving that the corresponding kernels have order 2. Note that the equation t4 = 1 (resp. t6 = 1) are both equivalent to t2 = 1 (look at the order of t, which must be a divisor of 12b+10). Hence ker p(4) = ker p(6) = {±1} and hence have order 2. p) and p(6)(F∗ To prove (ii), consider the action of F∗ p on F3 p (resp. F5 tiplication of a vector by a scalar. This multiplication by t ∈ F∗ F1(a) = F1(t4a) (resp. F2(a) = F2(t6a)). This completes the proof. p) given by the usual mul- p induces a bijection We denote n′ 1 = n1(1), n′ p \ H). We also write p = 12b + 11 = 4k + 3, i.e. we set k = 3b + 2. With this notation, (5.1) may be rewritten as 2 = n2(a), for a ∈ (F∗ 1 = n1(a) and n′′ 2 = n2(1), n′′ (cid:3) (5.2) A(p) = H(n′ 1n′ 2 + n′′ 1n′′ 2) = (2k + 1)(n′ 1n′ 2 + n′′ 1n′′ 2). The (affine or projective) hyperplane arrangement complements have polynomial count, see for instance [7], section (5.3) or [26], Theorem 5.15. So we apply this fact 2 in P4) to the projective hyperplane arrangement complements A′ corresponding to x1x2x3(x1 + x2 + x3) = 0 (resp. x4x5x6x7x8(x4 + x5 + x6 + x7 + x8) = 0). Using the formulas for the Betti numbers in [2], the fact that each cohomology group H m is pure of type (m, m) and the duality between H m and H 2d−m (where d is the corresponding (complex) dimension), it follows that 1 in P2 (resp. A′ c (5.3) and (5.4) HDM (A′ 1)(x, y) = (xy)2 − 3(xy) + 3 HDM (A′ 2)(x, y) = (xy)4 − 5(xy)3 + 10(xy)2 − 10(xy) + 5. It follows that the corresponding counting polynomials are (5.5) and (5.6) PM (A′ 1)(t) = t2 − 3t + 3 PM (A′ 2)(t) = t4 − 5t3 + 10t2 − 10t + 5. Using these two polynomials, it follows that, for almost all p = 4k + 3, one has (5.7) N ′ 1(p) = M(A′ 1)(Fp) = p2 − 3p + 3 ≡ 4k + 3 mod 8 TATE PROPERTIES, POLYNOMIAL-COUNT VARIETIES, AND ARRANGEMENTS 15 and N ′ 2(p) = M(A′ 2)(Fp) = p4 − 5p3 + 10p2 − 10p + 5 ≡ 4k + 3 mod 8. (5.8) Let A1 (resp. A2) denote the corresponding central hyperplane arrangements in C3 (resp. C5) and M(A1) (resp. M(A2)) the associated affine complements. Then, using (5.7) and (5.8), we get (5.9) N1(p) = M(A1)(Fp) = (p − 1)(p2 − 3p + 3) ≡ 4k + 6 mod 8 and (5.10) N2(p) = M(A2)(Fp) = (p − 1)(p4 − 5p3 + 10p2 − 10p + 5) ≡ 4k + 6 mod 8. Now, note that M(A1) (resp. M(A2)) is the disjoint union of all the hypersurfaces F1(a) (resp. F2(a)) for a ∈ F∗ p. Lemma 5.1, (5.9) and (5.10) yield (5.11) and n′ 1 + n′′ 1 = 2(p2 − 3p + 3) ≡ 6 mod 8 2 = 2(p4 − 5p3 + 10p2 − 10p + 5) ≡ 6 mod 8. n′ 2 + n′′ (5.12) It follows that n′′ 1 ≡ 6 − n′ (5.13) A(p) = (2k + 1)(n′ 1 mod 8 and n′′ 1n′ 2 + n′′ 1n′′ 2 ≡ 6 − n′ 2 mod 8. Using (5.2), we get 2) ≡ 2(2k + 1)(2 + n′ 1n′ 2 − 3n′ 1 − 3n′ 2) mod 8. Now we can state and prove our main result of this section. Theorem 5.2. The variety F satisfies the following. (i) F is a cohomologically Tate variety. (ii) F has not polynomial count. Proof. The proof of the first claim was given already in Example 4.3. Moreover, it follows from the description of the cohomology given there, that the corresponding Hodge-Deligne polynomial is (5.14) HDF (x, y) = (xy)7 − 9(xy)6 + 36(xy)5 − 82(xy)4 + 119(xy)3 − 110(xy)2 + 60(xy) − 15. Assume that (ii) fails, i.e. that F has polynomial count with polynomial PF . Then according to Katz Theorem [11], we must have (5.15) PF (t) = t7 − 9t6 + 36t5 − 82t4 + 119t3 − 110t2 + 60t − 15. We will reach a contradiction by showing that PF (p) 6= A(p), for infinitely many primes, namely for all the primes p in the progression 12b + 11 (Dirichlet prime number theorem). Let p be such a prime and write p = 4k + 3 as above. Then a simple computation (e.g. using Maple) shows that (5.16) PF (4k+3) = 8(2k+1)(1024k6+2560k5+2752k4+1632k3+584k2+129k+15). Hence, to complete the proof, it is enough to show that A(p) 6≡ 0 mod 8, for any prime p as above. Using the formula (5.13), this is equivalent to showing (5.17) 2 − 3n′ This in turn follows from the following. 2 + n′ 1n′ 1 − 3n′ 2 6≡ 0 mod 4. 16 ALEXANDRU DIMCA Lemma 5.3. Let p be a prime number of the form 4k + 3. Then both n′ divisible by 4. 1 and n′ 2 are Proof. Consider the group G = {±1}×{1, τ } acting on the (finite) Milnor fiber F1(1) by and −1 · (x1, x2, x3) = (−x1, −x2, −x3) τ · (x1, x2, x3) = (x2, x1, x3) −τ · (x1, x2, x3) = (−x2, −x1, −x3). There are two types of G-orbits. First we have the orbits Gx, corresponding to points x such that x1 6= x2. Then the isotropy group Gx is trivial, and the orbit Gx consists of 4 points. Since we are interested in a computation modulo 4, we can forget about such orbits. The second type of orbit corresponds to points x such that x1 = x2 = t. Then x ∈ F1(1) is equivalent to (x3 + t)2 = ∆(t) (5.18) where ∆(t) = t2 + t−2. There are two possibilities: (a) ∆(t) /∈ H, i.e. ∆(t) is not a square in F∗ p. Then the equation (5.18) is impossible and we do not get a point in F1(1). (b) ∆(t) is a square in F∗ p. Then the equation (5.18) has two solutions, namely x3 = −t + y and x3 = −t − y, with y satisfying y2 = ∆(t). In this way we get two points in F1(1). Note that ∆(t) = ∆(−t), hence for the point (−t) we are exactly in the same situation as for the point t. It follows that for each pair {t, −t} (i.e. orbit of the obvious {±1}-action on F∗ p), we get either 0 points in the (finite) Milnor fiber F1(1), or we get 4 points. This proves the claim for n′ 1. Consider next the group G = {±1} × {1, τ } × {1, σ} acting on the (finite) Milnor fiber F2(1) by and −1 · (x4, x5, x6, x7, x8) = (−x4, −x5, −x6, −x7, −x8) τ · (x4, x5, x6, x7, x8) = (x5, x4, x6, x7, x8) σ · (x4, x5, x6, x7, x8) = (x4, x5, x7, x6, x8) (and their obvious consequences). Since we are interested in a computation modulo 4, we can forget about all orbits corresponding to points having an isotropy group of order at most 2. Let now x be a point such that Gx = {1, τ } × {1, σ}. Then x = (s, s, t, t, x8) and x ∈ F2(1) is equivalent to (x8 + s + t)2 = ∆(s, t) (5.19) where ∆(t) = (s + t)2 + (st)−2. Note that ∆(s, t) = ∆(−s, −t), hence for the point (−s, −t) we are exactly in the same situation as for the point (s, t). It follows that p)2, we get either 0 points in the for each orbit of the obvious {±1}-action on (F∗ (finite) Milnor fiber F2(1), or we get 4 points. This proves the claim for n′ 2. TATE PROPERTIES, POLYNOMIAL-COUNT VARIETIES, AND ARRANGEMENTS 17 (cid:3) Remark 5.4. Surprisingly, it seems that PF (p) = A(p) for many primes of the form p = 4k + 1. Indeed, by a computer aided computation, we have obtained the following. PF (5) = A(5) = 11160, PF (13) = A(13) = 30575400 , PF (17) = A(17) = 237920544, PF (29) = A(29) = 12579682248, PF (37) = A(37) = 74188920024, PF (41) = A(41) = 155950465680, PF (53) = A(53) = 989657318520, PF (61) = A(61) = 2708356179720, PF (73) = A(73) = 9757738115280. However one has PF (89) = 39954467578608 6= A(89) = 39843984220188, and PF (97) = 73603528860864 6= A(97) = 72706366451444. Remark 5.5. M. Kisin and G. Lehrer have considered a notion of mixed Tate variety (imposing conditions on the eigenvalues of Frobenius action on p-adic ´etale cohomol- ogy), see Definition (2.6) in [13] and shown that if a variety Y is mixed Tate then Y is cohomologically Tate (see Theorem (2.2) (2) in [13]). Then in Remark (2.4) and in a footnote on p. 213, they discuss the conjectural equivalence between cohomologically Tate and mixed Tate conditions. Any hyperplane arrangement complement is mixed Tate, see [13], Proposition (3.1.1). We say that Y has weak polynomial count with count polynomial PY , if there is a polynomial PY ∈ Z[t], such that for all but finitely many primes p, there is an integer kp > 0 satisfying the following: if we set q = pkp, then for any finite field Fqs with s ∈ N∗, the number of points of Y over Fqs is precisely PY (qs). M. Kisin and G. Lehrer have shown that if a variety Y is mixed Tate then Y has weak polynomial count (see Proposition (3.4) in [13]). References [1] N. Budur and M. Saito, Jumping coefficients and spectrum of a hyperplane arrangement, Math. Ann. 347 (2010), 545 -- 579. 1, 1, 3, 4 [2] D. C. Cohen, A. I. Suciu, On Milnor fibrations of arrangements, J. London Math. Soc. 51 (1995), no. 2, 105 -- 119. 1, 4, 4.3, 5 [3] D. C. Cohen, G. Denham, A. I. Suciu, Torsion in Milnor fiber homology, Alg. Geom. Topology 3 (2003), 511535. 1 [4] P. Deligne, P. Griffiths, J. Morgan, D. Sullivan, Real homotopy theory of Kahler manifolds, Invent. Math. 29(1975), no. 3, 245 -- 274. 1, 1.3 [5] A. Dimca, Singularities and Topology of Hypersurfaces, Universitext, Springer-Verlag, 1992. 1 [6] A. Dimca, Sheaves in Topology, Universitext, Springer-Verlag, 2004. 1, 1 [7] A. Dimca and G.I. Lehrer, Purity and equivariant weight polynomials, dans le volume: Alge- braic Groups and Lie Groups, editor G.I.Lehrer, Cambridge University Press, 1997. 1, 5 [8] A. Dimca, G. Lehrer, Hodge-Deligne equivariant polynomials and monodromy of hyperplane arrangements, arXiv: 1006.3462. 3 [9] A. Dimca and S. Papadima, Finite Galois covers, cohomology jump loci, formality properties, and multinets, arXiv:0906.1040. 1.3 18 ALEXANDRU DIMCA [10] A. Dimca, S. Papadima, A. Suciu, Quasi-Kahler groups, 3-manifold groups, and formality, Math. Z., online DOI 10.1007/s00209-010-0664-y. 1.3 [11] T. Hausel and F. Rodriguez-Villegas, Mixed Hodge polynomials of character varieties, with an appendix by Nicholas M. Katz, Invent. Math. 174 (2008), 555 -- 624. 1, 5 [12] M. Kim, Weights in cohomology groups arising from hyperplane arrangements, Proc. Amer. Math. Soc. 120(1994), 697 -- 703. 1 [13] M. Kisin and G.I. Lehrer, Eigenvalues of Frobenius and Hodge numbers, Pure Appl. Math. Q. 2(2006), 497 -- 518. 5.5 [14] G.I. Lehrer, The ℓ-adic cohomology of hyperplane complements, Bull. London Math. Soc. 24(1992), 76 -- 82. 1 [15] A. Libgober, Eigenvalues for the monodromy of the Milnor fibers of arrangements, in: Trends in singularities, Trends Math., Birkhauser, Basel, 2002, pp. 141 -- 150. 1 [16] A. Libgober, On combinatorial invariance of the cohomology of Milnor fiber of arrangements and Catalan equation over function field, arXiv:1011.0191. 1 [17] A. N´emethi, Generalized local and global Sebastiani-Thom type theorems, Compositio Math- ematica 80 (1991), 1-14. 1.5 [18] M. Oka, On the homotopy type of hypersurfaces defined by weighted homogeneous polynomials, Topol- ogy 12 (1973), 19-32. 1.5 [19] P. Orlik and R. Randell, The Milnor fiber of a generic arrangement, Ark. Math. 31(1993), 71 -- 81. 4.3 [20] P. Orlik, L. Solomon, Combinatorics and topology of complements of hyperplanes, Invent. Math. 56(1980), 167 -- 189. 1 [21] P. Orlik and H. Terao, Arrangements of Hyperplanes, Springer-Verlag, Berlin Heidelberg New York, 1992. 1, 4 [22] K. Sakamoto, Milnor fiberings and their characteristic maps, in: Manifolds - Tokyo 1973, Univ. Tokyo Press, Tokyo 1975, pp. 1456150. 1.5 [23] M. Saito, Multiplier ideals, b-functions, and spectrum of a hypersurface singularity, Compositio Math. 143 (2007), 1050 -- 1068. 4, 4 [24] V. Schechtman, H. Terao, A. Varchenko: Local systems over complements of hyperplanes and the Kac-Kazhdan condition for singular vectors, J. Pure Appl. Alg. 100(1995), no. 1-3, 93 -- 102. 1 [25] B.Z. Shapiro, The mixed Hodge structure of the complement to an arbitrary arrangement of affine complex hyperplanes is pure, Proc. Amer. Math. Soc. 117(1993), 931 -- 933. 1 [26] R. P. Stanley, An Introduction to Hyperplane Arrangements, Park City Mathematics Series, volume 14: Geometric Combinatorics (2004). 1, 5 [27] J. Steenbrink, Intersection form for quasi-homogeneous singularities, Compositio Math. 34(1977), 211-223. 1 [28] D. Sullivan, Infinitesimal computations in topology, Inst. Hautes ´Etudes Sci. Publ. Math. 47 (1977), 269 -- 331. 1, 1.3 [29] D. Tapp, Picard-Lefschetz monodromy of products, Journal of Pure and Applied Algebra 212 (2008), 2314-2319. 1.5, 2.3 [30] H. Zuber, Non-formality of Milnor fibres of line arrangements, Bull. London Math. Soc. 42 (2010), no. 5, 905911. 1, 1.3 Laboratoire J.A. Dieudonn´e, UMR du CNRS 6621, Universit´e de Nice Sophia An- tipolis, Parc Valrose, 06108 Nice Cedex 02, France E-mail address: [email protected]
1801.08286
1
1801
2018-01-25T06:31:34
Perverse schobers and birational geometry
[ "math.AG" ]
Perverse schobers are conjectural categorical analogs of perverse sheaves. We show that such structures appear naturally in Homological Minimal Model Program which studies the effect of birational transformations such as flops, on the coherent derived categories. More precisely, the flop data are analogous to hyperbolic stalks of a perverse sheaf. In the first part of the paper we study schober-type diagrams of categories corresponding to flops of relative dimension 1, in particular we determine the categorical analogs of the (compactly supported) cohomology with coefficients in such schobers. In the second part we consider the example of a "web of flops" provided by the Grothendieck resolution associated to a reductive Lie algebra g and study the corresponding schober-type diagram. For g=sl(3) we relate this diagram to the classical space of complete triangles studied by Schubert, Semple and others.
math.AG
math
Perverse schobers and birational geometry Alexey Bondal, Mikhail Kapranov, Vadim Schechtman January 26, 2018 To Sasha Beilinson on his 60th birthday `A travers le brouillard, il contemplait des clochers, des ´edifices, dont il ne savait pas les noms. Flaubert, L' ´Education sentimentale. Abstract Perverse schobers are conjectural categorical analogs of perverse sheaves. We show that such structures appear naturally in Homo- logical Minimal Model Program which studies the effect of birational transformations such as flops, on the coherent derived categories. More precisely, the flop data are analogous to hyperbolic stalks of a perverse sheaf. In the first part of the paper we study schober-type diagrams of categories corresponding to flops of relative dimension 1, in particular we determine the categorical analogs of the (compactly supported) cohomology with coefficients in such schobers. In the second part we consider the example of a "web of flops" provided by the Grothendieck resolution associated to a reductive Lie algebra g and study the corresponding schober-type diagram. For g = sl3 we relate this diagram to the classical space of complete triangles studied by Schubert, Semple and others. 1 Contents 0 Introduction 1 Flops and schobers on the disk 2 Cohomology of 1-dimensional flobers 3 The web of flops. Role of arrangements. 4 The Grothendieck resolution 5 The g-web of flops and partial blowdowns. 6 Fiber products and varieties of simplices 7 The flop diagram for sl3 and Schubert's variety of complete triangles 8 Properties of the sl3-flober 0 Introduction 2 8 17 28 33 40 46 54 71 A. Goals of the paper. Perverse schobers are conjectural categorical analogs of perverse sheaves. The possibility of a meaningful categorified theory of perverse sheaves was suggested in [36]. It gradually becomes clear that such a theory must indeed exist and have applications to various areas of mathematics. In some simple cases a precise definition of perverse schobers can be given using quiver description of perverse sheaves as a starting point and then replacing quivers by analogous diagrams of triangulated categories and exact functors. One important role of perverse schobers is that they can serve as natural coefficient data for forming Fukaya categories [27], just like sheaves can serve as coefficient data for forming cohomology. The goal of this paper is to investigate a different, perhaps "mirror dual" to the above, appearance of pervese schobers: in birational geometry. We are talking especially about the Homological Minimal Model Program which studies derived categories of coherent sheaves on algebraic varieties related 2 by flops. Our starting point was the observation that the basic framework of the derived equivalence corresponding to a flop [14] matches very precisely the "hyperfunction" description (given in [35] and recalled in Proposition 1.1 below) of Perv(C, 0), the category of perverse sheaves on C smooth outside 0. We do not know any a priori reason for this remarkable match. In the first part of the paper, we study B. Brief summary of the paper. various features of the schober-type diagrams of categories corresponding to flops of relative dimension 1. In this case the corresponding perverse schobers F (we call it them flobers) can be seen as categorifications of objects from Perv(C, 0). We describe the categories having the meaning of Hi(C, F) and Hi c(C, F). In particular, we see an appearance of the "categorical Poincar´e duality" between H0 and H2 it corresponds to Toen's Morita duality [50] c: between the coherent derived category Db(Z) and the category of perfect complexes Perf(Z) of a singular projective variety Z. At the same time, the description of Perv(C, 0) mentioned above, is a particular case of a much more general classification result [35] for perverse sheaves on Cn smooth with respect to a stratification given by an arrangement of hyperplanes H with real equations. That result (recalled as Theorem 3.1 below) is formulated in terms of combinatorics of the chamber structure of the real part Rn ⊂ Cn (including walls between chambers and, more generally, faces of all dimensions) given by H. Again, we note that similar kinds of chamber structures appear in birational geometry [37] in the study of the cone of movable divisors and the corresponding flops. In the second part of the paper, we consider the prototypical example where the movable cone with its chamber structure can be investigated in great detail: that of the simultaneous Grothendieck resolution associated to a reductive Lie algebra g. In this case the chamber structure is given by the Weyl chambers in the lattice of integer weights h∗ Z. C. Relation to earlier work. Whatever possible approach one chooses, a perverse schober on a stratified variety (M, {Mα})) amounts to the following levels of structure: (0) A local system L of (enhanced) triangulated categories on the open stratum M0. (1, 2,· · · ) "Singularity data" extending L to strata of higher codimension. 3 In several situations, the local system L is known; our observation is that (some of) the singularity data are also naturally present. We note the recent work of Donovan and Wemyss [23] who associated, to a system of 3-dimensional flops, a local system L of triangulated categories on the complement of a hyperplane arrangement closely related to the root arrangements of types A-D-E. In the context of Grothendieck resolutions, the work of Bezrukavnikov [7] and Bezrukavnikov-Riche [9] has, quite some time ago, constructed the braid group action on the corresponding coherent derived category. This gives a (W -equivariant) local system L of triangulated categories on the open stratum h∗ reg in the space of complex weights. These results can be seen as "quasi-classical" analogs of the following fundamental fact: the braid group associated to a semisimple Lie algebra g, acts by Schubert correspondences, in Db constr(G/B), the constructible derived category of the flag variety G/B. Our point of view is that this local system L should admit a natural extension to a perverse schober on entire h∗ smooth with respect to the stratification given by the arrangement of the coroot hyperplanes. Indeed, the transition functors of our flober diagram along real codimension 1 faces give precisely this local system. In fact, [7] [9] have constructed the action of the affine Weyl group, not just the Artin braid group. From our point of view, this should be extended to a schober on the dual torus H ∨ (instead of the Cartan subalgebra) smooth with respect to the stratification given by the arrangement of the "coroot subtori". This indicates that the types of perverse schobers appearing in birational geometry go beyond the realm of hyperplane arrangements. Further, our approach seems closely related to the general philosophy of Bezrukavnikov [8] concerning derived categories of symplectic resolutions. That context is more general than that of the Grothendieck resolution (which is the "universal deformation" of T ∗(G/B), a symplectic resolution of the nilpotent cone) and includes, for instance, all quiver varieties of Nakajima. In that case, the chamber structure also plays a crucial role. Moreover, the point of view based on quantization in characteristic p, provides additional insight into the existence and nature of derived equivalences between resolutions corresponding to adjoining chambers ("symplectic flops"). If we go one categorical level down, we get actions of affine Hecke alge- bras on various Grothendieck groups related to T ∗(G/B) and Grothendieck 4 resolutions which have been, for some time, a classical subject in representa- tion theory [21]. (They are, of course, modern manifestations of the peren- nial principal series intertwiners, familiar in representation theory since the 1940's.) However, such questions end up being quite subtle, due to the fact that various composite correspondences become very singular as algebraic varieties. In our context, the analogs of such correspondences are given by various fiber products XC of the flopped Grothendieck resolutions. They are labelled by cells C of the root arrangements in the space of real weights. We observe that for g = sln the highest fiber product X0 s related to the classical variety of simplices in Pn−1 [34] [3] [4]. In particular, for n = 3, the variety of triangles T has been studied in the XIX century by Schubert [46] who has constructed its desingularization, the variety of complete triangles bT , see [47][45]. Using bT allows us to lift the diagram of fiber products for sl3 to a diagram containing only smooth varieties. The relation of Schubert's complete triangles with correspondences (also bearing Schubert's name) on the Grothendieck resolution has, apparently, not been noticed before. In this paper we consider only "B-model" aspects of perverse schobers, i.e., schobers formed by coherent derived categories only. Relations of such schobers to homological mirror symmetry, cf. the work of Harder-Katzarkov [29], seem very interesting. D. Organization and results of the paper. between flopping diagrams §1 explains an analogy (0.1) X− ←− X0 −→ X+ and perverse sheaves on the disk (or on the complex plane C) smooth out- side of 0. We denote by F and call flober the diagram of coherent derived categories associated to a flopping diagram as above. In §2 we make sense and identify the categories Hi(C, F) and Hi c(C, F) in terms of the singular variety Z (the common flopping contraction of X±). This is done in Theorems 2.5 and 2.12). §3 explains the role of hyperplane arrangements in HMMP and presents our approach to perverse schobers smooth with respect to an arrangement with real equations. This approach is based in the description of perverse sheaves given in [35]. In §4 we give a detailed treatment of the Grothendieck resolution from 5 the point of view of flops. This material is known but we could not find an existing presentation suitable for our purposes. §5 is devoted to constucting the "anti-flober", the diagram of partial blow- downs ZC of various flopped Grothendieck resolutions These blowdowns are labelled by faces C of the coroot arrangement. We pay special attension to the singular nature of ZC for C of small codimension and their relation to similar varieties for Levi subalgebras in g. In §6 we construct fiber products XC of flopped Grothendieck resolutions (the analogs of the variety X0 in (0.1)). These varieties, although extremely beautiful, have not been traditionally considered in representation theory. The closest concept discussed in the literature, is the variety of simplices [3] defined as the space of maps of the Coxeter complex of g to the spherical Bruhat-Tits building. We denote the principal component of this variety by Tg. As for the ZC above, we establish relations of XC for C of small codi- mension, with similar varieties for Levi subalgebras in g but corresponding to the smallest face 0. . In fact, the variety X0 appears as the total space of a natural coherent sheaf on Tg whose fibers, generically are Cartan subalgebras in g but can have higher dimension over certain degenerate points. This gives rise to a natural partial desingularization which we call Cartanization. In §7 we study the first new case g = sl3. Remarkably, in this case, Cartanization of Tg gives a smooth variety which is identified with Schubert's variety of complete triangles (Proposition 7.3). Therefore, the entire flop diagram for sl3 can be interpreted in terms of Schubert's space. We perform a detailed geometric study of various intermediate fiber products (2-ray and 3-ray varieties) which appear in composing the elementary correspondences [44]. These partial fiber products are related to varieties of partial triangles which form classical stepping stones to bT , see [45] and references therein. Finally, in §8 we establish the main properties of the flober diagram for sl3 (Theorem 8.3). We prove all the conditions of the Definition 3.6 defining H-schobers except one which concerns a diagram which we call the Schubert transform. It connects two 1-ray fiber products of flopped Grothendieck resolutions via the Cartanized variety Y0 (the central term in the sl3-flop diagram, which is the total space of a vector bundle over bT ). Studying the effect of the Schubert transform on derived categories might help to understand sln-schobers for higher n. 6 E. Acknowledgements. We are grateful to R. Bezrukavnikov, A. Bodzenta, T. Bridgeland, W. Donovan, A. Efimov, A. Kuznetsov, Yu. Prokhorov, B. Toen and M. Reid for useful discussions and correspondence. We are also grateful to the referees for numerous remarks which have improved the paper. The work of A.B. and M.K. was supported by the World Premier Inter- national Research Center Initiative (WPI Initiative), MEXT, Japan. A.B. is partially supported by Laboratory of Mirror Symmetry NRU HSE, RF Government grant, ag. № 14.641.31.0001. V. S. thanks Kavli IPMU for the support of the visits during which a part of this work was performed. 7 1 Flops and schobers on the disk A. Reminder on perverse sheaves. Let M be a connected complex manifold and S = (Mα) be a complex analytic stratification of M. We assume that each Mα is smooth but the closure M α can be a singular complex analytic space. In particular, there is an open dense stratum M0. We denote by Perv(M, S) the abelian category of S-constructible perverse sheaves of Q-vector spaces on M, with respect to the middle perversity [6]. That is, we view Perv(M, S) as an abelian subcategory in the triangulated category of S-constructible complexes of sheaves on M, defined as the heart of the perverse t-structure. We further normalize the shift of the t-structure by the following requirement: for F ∈ Perv(M, S) the restriction F M0 is a local system in degree 0. Thus a perverse sheaf can be viewed as a particular way to extend a local system on X0 to strata of higher codimension. Perv(M, S) is a Noetherian and Artinian abelian category. In many cases it can be identified with the category of representations of an explicit quiver with relations. The simplest case is M = C, stratified by 0 and C − {0}. Denote the corresponding category of perverse sheaves by Perv(C, 0). Its descrption by linear algebra data was first found in [28]. The following alternative description [35] was the starting point of the present paper. Proposition 1.1. Perv(C, 0) is equivalent to the category of diagrams of Q-vector spaces E− γ− δ− / E0 γ+ δ+ / E+ satisfying the two following conditions: (1) γ−δ− = IdE−, γ+δ+ = IdE+. (2) The maps γ−δ+ : E+ → E−, γ+δ− : E− → E+ are invertible. Explicitly, for F ∈ Perv(C, 0) the corresponding diagram is found as R(F ), the sheaf of the 1st hypercohomology with support follows. Let R = H1 on R ⊂ C. This is a sheaf on the real line R, constructible with respect to the strat- ification into R<0, 0 and R>0. The spaces E−, E0, E+ are the stalks of R at 8 / o o / o o these strata. The maps γ± are the generalization maps for R. The maps δ± can be obtained by considering the dual perverse sheaf F ∗, see [35]. The isomorphism T = (γ+δ−)(γ−δ+) : E+ −→ E+ is the monodromy of the local system F C∗ calculated at any point of R>0. Further, the global (hyper)cohomology and the (hyper)cohomology with compact support of F are given by the following two complexes: (1.2) RΓ(C, F ) = (cid:8)E− ⊕ E+ RΓc(C, F ) = (cid:8)E0 −→ E0(cid:9), −→ E− ⊕ E+(cid:9), (δ−,δ+) (γ−,γ+) concentrated in degrees [0, 1] and [1, 2], respectively. The canonical map RΓc(C, F ) −→ RΓ(C, F ), is the unique morphism of the above complexes, identical on E0. B. The Atiyah flop and its flober. We want to compare Proposition 1.1 with the simplest example of a flop in birational geometry. We will work over an algebraically closed field k. For an quasi-projective scheme X over k we denote by Db(X) the bounded derived category of coherent sheaves on X, and by Perf X the triangulated category of perfect complexes, so Perf X ≃ Db(X) if X is smooth and quasi-projective. When needed, we will consider Db(X) and Perf X as dg-categories with their standard enhancements by the RHom- complexes. The classical (Atiyah) flop is the diagram (1.3) (X−, C−) f−−→ (Z, z) f+←− (X+, C+) formed by two birational desingularizations of a 3-fold Z with one quadratic singular point z. The desingularizations X± are small, i.e., each preimage C± = f −1 ± (z) is identified with P1, in contrast with the blowup q : X0 = BlzZ → Z which has q−1(z) = P1 × P1. In fact, X0 = X− ×Z X+ is the fiber product of X− and X+, so we have a diagram (1.4) X− p−←− X0 p+−→ X+. 9 Example 1.5. In the simplest case, Z is the 3-dimensional quadratic cone, which we can write in the determinantal form Z = (cid:8)z ∈ Mat2(k) det(z) = 0(cid:9). The varieties X± are given explicitly by: X+ =(cid:8)(z, l) ∈ Mat2(k)×P1 z(l) = 0(cid:9), X− =(cid:8)(z, l) ∈ Mat2(k)×P1 zt(l) = 0(cid:9). Here P1 is viewed as consisting of 1-dimensional subspaces l in k2, on which Mat2(k) acts. So X±, as abstract varieties, are each identified with the total space of the vector bundle O(−1)⊕2 on P1. The variety X0 is the total space of the line bundle O(−1, −1) on P1 × P1. It was proved in [14] that Db(X+) and Db(X−) are equivalent. More precisely, each of the functors (now known as the flop functors) (1.6) T+,− = Rp−∗ ◦ p∗ T−,+ = Rp+∗ ◦ p∗ + : Db(X+) −→ Db(X−), − : Db(X−) −→ Db(X+), is an equivalence. Therefore the diagram of categories (1.7) Db(X−) Rp−∗ Lp∗ − / Db(X0) Rp+∗ Lp∗ + / Db(X+) satisfies the categorical analogs of the properties (1) and (2) of Proposition 1.1. We say that (1.7) represents a perverse schober on C. We denote this schober by F. Since F comes from a flop, we refer to it as the flober associated to (1.3). C. General flops of relative dimension 1. The Atiyah flop is a partic- ular case of a more general situation [51] [10] which also leads to schobers on the disk. We recall this context in a partial generality, see [10] for general case. We consider a Cartesian diagram of quasi-projective schemes (also re- ferred to as a flop) (1.8) X− X0 = X− ×Z X+ p− w♦♦♦♦♦♦♦♦♦♦♦♦ '❖❖❖❖❖❖❖❖❖❖❖❖❖❖ f− Z 10 p+ '❖❖❖❖❖❖❖❖❖❖❖❖ w♦♦♦♦♦♦♦♦♦♦♦♦♦♦ X+ f+ / o o / o o w ' ' w with the following properties: (F1) X± and Z are irreducible algebraic varieties (i.e., reduced schemes), with X± smooth and Z normal. (F2) Z is allowed to have canonical hypersurface singularities. Further, the multiplicity of each singular point z ∈ Z is at most 2. That is, in bz, the formal neighborhood of z, the equation of Z can be brought to the form t2 0 = f (t1, · · · , tn). (F3) The morphisms f± are birational, and each fiber of both f± has di- mension ≤ 1. Further, for a singular point z ∈ Z as in (F2), the two preimages f −1 ± (bz) can be identified as formal schemes (but not as formal schemes overbz) via the involution t0 7→ (−t0). (F4) The exceptional loci of f± have codimension ≥ 2. Note the scheme X0 can be singular and even non-reduced. In this setup we have the flop-flop functors T+,− and T−,+ similar to (1.6). More precisely, It was shown in [10] that T+,− and T−,+ are equivalences. In particular, for this situation we have a schober F as in (1.7). This schober further was studied in [10]. Another important feature of a flop (1.8) is the (triangulated) null-category of the morphism f− which is the full subcategory in Db(X−) defined as (1.9) and similarly for C+. C− := (cid:8)A− ∈ Db(X−) R(f−)∗(A−) = 0(cid:9), Remarks 1.10. (a) For an extension to the case when X± have Gorenstein singularities, see [10]. (b) If one relaxes the assumption that the dimensions of the fibers of f± are ≤ 1, then the flop functors (1.6) may no longer be equivalences, see [41]. D. Flober decategorified. For a triangulated category D we denote by K(D) = K0(D) ⊗ Q the Grothendieck group of D made into a Q-vector space. We denote by [F ] ∈ K(D) the class of an object F ∈ D. For an algebraic variety X over k, we write K(X) = K(Db(X)); this is the rational Grothendieck group of coherent sheaves on X. 11 Applying the functor K to the diagram (1.7) of the kind considered in §C, we get a diagram of Q-vector spaces (1.11) E− = K(X−) γ−=p−∗ δ−=p∗ − / E0 = K(X0) γ+=p+∗ δ+=p∗ + / E+ = K(X+). This diagram satisfies the properties of Proposition 1.1 and so represents a perverse sheaf on (C, 0) which we denote K(F). Example 1.12. Consider the situation of Example 1.5. diagram (1.11) has the form In this case the Q2 γ− δ− / Q4 γ+ δ+ / Q2. Indeed, we have the diagram (1.13) X− π− p− p+ X0 π0 / X+ π+ P1 P1 × P1 pr− pr+ / P1 where π± represents X± as the total space of OP1(−1)⊕2 and π0 represents X0 as the total space of OP1×P1(−1, −1). Therefore K(X±) ≃ K(P1) ≃ Q2, K(X0) ≃ K(P1 × P1) ≃ Q4. This means that the perverse sheaf K(F) has generic rank 2. Proposition 1.14. Let j : C∗ ֒→ C be the embedding. Then, in the situation of Example 1.12, K(F) ≃ Q0[1] ⊕ Q C ⊕ j!(Q C∗). Remarks 1.15. (a) In particular, the proposition says that the monodromy of the local system j∗K(F) on C∗ is trivial. This triviality is a feature of our local case: the varieties X± are non-proper, being just neighborhoods of the corresponding (−1)-curves C±. The self-equivalence (flop-flop functor) of Db(X±) underlying the monodromy is, by general results, see [10], the spherical reflection with respect to the spherical object OC±(−1). In our local case the class [OC±(−1)] ∈ K(X±) vanishes, see (1.18) below. 12 / o o / o o / o o / o o   o o   /   o o / (b) The situation will be different in the case of a flop relating proper smooth varieties. For instance, consider the natural compactification of Ex- ample 1.5 that is, the flop between the two resolutions of the projective 3- dimensional quadratic cone. In this case the class [OC±(−1)] and therefore the monodromy of local system is non-trivial. Indeed, as X± is proper, K(X±) carries the Euler form h[F ], [G]i = Xi (−1)i dim Exti(F , G). Let D± ⊂ X± be an effective divisor such that the intersection number (C±.D±) is equal to 1. Then h[OC±(−1)], [OD±]i = 1 and so [OC±(−1)] 6= 0. (c) A way to get perverse sheaves with nontrivial monodromy in our local situation is to pass to equivariant derived categories with respect to the group G = GL2 × GL2 acting on the variety Z ⊂ Mat2(k) (and thus on the entire flop diagram) by left and right multiplications. The monodromy on corresponding equivariant K-groups is the simplest instance of the Hecke algebra action on the equivariant K-theory of the Grothendieck resolutions, see §4 below and [21]. E. Proof of Proposition 1.14: We denote L±(i) = π∗ ± OP1(i), i ∈ Z; L0(i, j) = π∗ 0 OP1×P1(i, j), i, j ∈ Z the line bundles on X± and X0. A basis of K(X±) is given by the classes [L±(i)] with i = 0, 1 while a basis of K(X0) is given by the classes [L0(i, j)] with i, j = 0, 1. We start with identifying all the maps in (1.11). Proposition 1.16. (a) The action of δ± in our bases is given by δ−[L−(i)] = [L0(i, 0)], δ+[L+(i)] = [L0(0, i)], i = 0, 1. (b) The action of γ± in our bases is given by the matrices γ− [L0(0, 0] [L0(1, 0)] [L0(0, 1)] [L0(1, 1)] γ+ [L0(0, 0] [L0(1, 0)] [L0(0, 1)] [L0(1, 1)] [L−(0)] [L−(1)] 1 0 0 1 2 −1 1 0 [L+(0)] [L+(1)] 1 0 2 −1 0 1 1 0 13 Proof of Proposition 1.16. Part (a) reflects the isomorphisms p∗ − L−(i) ≃ L0(i, 0), p∗ + L+(i) ≃ L0(0, i), which follow from the commutativity of the squares in (1.13). Let us focus on (b). We note, first of all, that the standard exact Koszul complex on P1 0 → OP1 → OP1(1)⊕2 → OP1(2) → 0 implies, after tensoring with O(i) and pulling back, the relation (1.17) [L±(i)] − 2[L±(i + 1)] + [L±(i + 2)] = 0 in K(X±). As before, we denote by C± ≃ P1 ⊂ X± the image of the zero section of the bundle O(−1)⊕2. We have the tautological section of the pullback bundle π∗ ±O(−1)⊕2 = L±(−1)⊕2 on X± which vanishes on C±. The corresponding Koszul resolution of OC± together with (1.17) implies that [OC±(j)] = 0 for any j. Denoting by I± ⊂ OX± the sheaf of ideals of C±, we conclude that (1.18) [I± ⊗ L±(j)] = [L±(j)], j ∈ Z. Now, X0 is identified with the blowup of X± along C± that is, X0 ≃ Proj(cid:18)Mn≥0 I n +(cid:19) ≃ Proj(cid:18)Mn≥0 I n −(cid:19). Therefore it carries line bundles O(1)+ representations as Proj so that rel and O(1)− rel coming from these two (1.19) Rp±∗O(1)± rel ≃ I±. Lemma 1.20. Both line bundles O(1)± rel are isomorphic to L0(1, 1). Proof of the lemma: For definiteness, let us consider the "−" case. It will be more convenient for us to identify the dual bundles, i.e., to identify the "tautological" bundle O(−1)− rel with L0(−1, −1). Any line bundle on X0 is isomorphic to some L0(i, j) and (i, j) can be found by restricting to the zero section: L0(i, j)P1×P1 = OP1×P1(i, j). We need to find (i, j) for O(−1)− rel, i.e., to find the restriction of O(−1)− rel to P1 × P1 = C− × P1 which is the 14 exceptional divisor of the blowup. The "vertical" slices {c} × P1 are the projectivizations of the fibers of the normal bundle to C− in X−, so the degree of the restriction of O(−1)− rel to them is, by definition, equal to (−1). Let us look at a "horizontal" slice C−×{p}, where p ∈ P1. It is represented by a rank 1 subbundle Mp ⊂ OC−(−1)⊕2. This subbundle is isomorphic to O(−1). Now, by definition, the fiber of the "tautological" bundle O(−1)− rel at a point (c, p) is identified with the fiber of Mp at c. This shows that the restriction of O(−1)− rel to C− × {p} is isomorphic to OC−(−1). Lemma 1.20 is proved. Applying the projection formula and (1.19), we get: Corollary 1.21. We have, for any i ∈ Z: Rp−∗ L0(i, 0) ≃ L−(i), Rp−∗ L0(i, 1) ≃ I− ⊗ L−(i − 1), Rp+∗ L0(0, i) ≃ L+(i), Rp+∗ L0(1, i) ≃ I+ ⊗ L+(i − 1). Applying this corollary and invoking the identity (1.18) and the recursion (1.17), we obtain the proof of Proposition 1.16. Corollary 1.22. The monodromy of K(F) around 0 is trivial. Proof: Indeed, each of the half-monodromy maps ϕ−+ = γ+δ− : E− −→ E+, ϕ+− = γ−δ+ : E+ −→ E− is given, in the above bases, by the same matrix(cid:18)1 squares to the identity, the monodromy is trivial. 2 0 −1(cid:19). Since this matrix We now recall [5] [28] the more standard description of Perv(C, 0) by linear algebra data. Proposition 1.23. (a) Perv(C, 0) is equivalent to the category of diagrams of finite-dimensional Q-vector spaces Φ uo v / Ψ such that TΦ = IdΦ − uv and TΨ = IdΨ − vu are invertible. (b) If F corresponds to such a diagram, then RΓ(C, F ) and RΓc(C, F ) are identified with the two term complexes situated in degrees [0, 1] and [1, 2] respectively. (cid:8)Ψ u−→ Φ(cid:9), (cid:8)Φ v−→ Ψ(cid:9), 15 / o In particular, we note the following (Φ, Ψ)-diagrams and the correspond- ing perverse sheaves: (cid:8) Q 0 Id / Q / 0 (cid:9) ∼ Q0[−1], (cid:8) 0 (cid:9) ∼ Rj∗(Q C∗), (cid:8) Q (cid:8) Q / Q / Q Id 0 (cid:9) ∼ QC, (cid:9) ∼ j!(Q C∗). (1.24) Lemma 1.25. We have the following identifications of the dimensions: H0(C, K(F)) ≃ H1(C, K(F)) ≃ Q, H1 c(C, K(F)) ≃ H2 c(C, K(F)) ≃ Q2. Proof of the lemma: Indeed, the identifications of the first line follow imme- diately from part (a) of Proposition 1.16 together with (1.2). The identifica- tions of the second line follow from part (b) of Proposition 1.16, because the 4 × 4 matrix obtained by stacking the matrices of γ+ and γ−, is easily found to have rank 2. As explained in [35], §9A, to pass from the description of Proposition 1.1 (i.e., from the diagram as in (1.11)) to that of Proposition 1.23, we need to form Φ = Ker(γ−) ⊂ E0, Ψ = E+. This shows that in our case both Φ and Ψ are 2-dimensional. Comparing now the second line of the identifications in Lemma 1.25 with part (b) of Proposition 1.23, we find that the complex {Ψ v→ Φ} formed by two 2-dimensional vector spaces, has 2-dimensional cohomology spaces, so v = 0. Further, comparing the first line of identifications in Lemma 1.25 with the same part (b) of Proposition 1.23 we find that the complex {Φ u→ Ψ} has 1-dimensional cohomology spaces, so u has rank 1. This means that our (Φ, Ψ)-diagram is isomorphic to the direct sum of the first, second and fourth diagrams in (1.24). This finishes the proof of Proposition 1.14. 16 / o o / o o / o o / o o 2 Cohomology of 1-dimensional flobers A. Setup of the problem. of the 2-term complexes (1.2) for the flober In this section we study categorical analogs (2.1) F = (cid:8) Db(X−) Rp−∗ Lp∗ − / Db(X0) Rp+∗ Lp∗ + / Db(X+) (cid:9) discussed in §1C from which we keep the assumptions and notations. A literal interpretation would give 2-term complexes of triangulated cat- egories (i.e., just exact functors) (2.2) (Lp∗ RΓ(C, F) = (cid:8)Db(X−) ⊕ Db(X+) RΓc(C, F) = (cid:8)Db(X0) −→ Db(X0)(cid:9), −→ Db(X+) ⊕ Db(X−)(cid:9), (Rp−∗,Rp+∗) −,Lp∗ +) with the sources and targets formally put in degrees 0, 1, resp. 1, 2. We are interested in the "cohomology" of these complexes, a concept which needs to be defined. For a 2-term complex V • = {V 0 d→ V 1} of vector spaces, the cohomology of V • are Ker(d) and Coker(d). In our case, the correct replacement of "ker- nel" for the "differential" (Lp∗ +) in RΓ(C, F) would be the (homotopy) fiber product. Similarly, the "cokernel" of (Rp−∗, Rp+∗) in RΓc(C, F) is nat- urally understood as the (homotopy) cofiber product, or pushout. We recall these concepts. −, Lp∗ B. Model structures in dgCat. The proper framework for homotopy lim- its and colimits is that of model categories see, e.g., [25] [30] for background. We denote by dgCat the category of small k-linear dg-categories. For A ∈ dgCat we denote by H 0(A) the k-linear category with the same objects as A and the Hom-spaces obtained by taking H 0 of the Hom-complexes in A. We also denote by Perf A the category of perfect contravariant dg-functors A → dgVect. We will use two model structures on dgCat introduced by Tabuada [48] (see also a short summary in [26] §1.1). The first is the quasi-equivalence model structure, where weak equivalences are quasi-equivalences. We denote the category dgCat equipped with the quasi-equivalence model structure by Qeq. 17 / o o / o o The second is the Morita model structure, where weak equivalences are Morita equivalences. Recall that a dg-functor f : A → B is called a Morita equivalence, if f∗ : Perf A → Perf B is a quasi-equivalence. We denote dgCat equipped with the Morita model structure by Mrt. We recall that fibrant objects in Mrt are perfect dg-categories A, i.e., pre- triangulated [13] dg-categories A such that the associated triangulated cat- egory H 0(A) is closed under direct summands. In particular, for a quasi- projective scheme X, the dg-categories Db(X) and Perf X are perfect. As shown in [48], the model category Mrt is the left Bousfield localization of Qeq with respect to the class of Morita equivalences. This implies: Proposition 2.3. (a) Cofibrations in Mrt are the same as in Qeq. (b) A dg-functor f : A → B between perfect dg-categories is a fibration in Mrt iff it is a fibration in Qeq. (c) A dg-functor f : A → B between perfect dg-categories is a Morita equivalence iff it is a quasi-equivalence. For perfect dg-categories being a quasi-equivalence is equivalent to H 0(f ) : H 0(A) → H 0(B) being an equiva- lence of triangulated categories. Proof: (a) is the definition of left Bousfield localization, see [30], Def. 3.3.1. Part (b) is [30], Prop. 3.3.16(1). Part (c) holds because for a perfect A the Yoneda functor y : A → Perf A is a quasi-equivalence. Each of the two model structures on dgCat gives the corresponding con- Mrt etc. cept of homotopy limits and colimits which we will denote by holim←−−− Homotopy (co)fiber products are particular cases of these. C. Global sections. We define the category of "global sections" of the flober F as the homotopy fiber product in Mrt: H0(C, F) := Db(X−) ×h Db(X0) Db(X+) = (2.4) = holim←−−− Mrt(cid:8)Db(X−) Lp∗ −−→ Db(X0) Lp∗ +←− Db(X+)(cid:9). Theorem 2.5. H0(C, F) is identified with Perf Z, where Z is as in (1.8). Proof: At the level of triangulated categories, each of the Lp∗ ± embeds Db(X±) into Db(X0) as a full triangulated subcategory. This is because the composition R(p±)∗ ◦ Lp∗ ± is identified with the identity of Db(X±). Indeed, according to Proposition 2.7 in [10], Rp±∗OX0 = OX±. 18 We note further that Db(X±) ≃ Perf X± because X± are assumed smooth. Let us now consider Db(X0) as a dg-category, and Perf X± as dg-subcategories there. Let us further consider each Perf X± as a homotopy strictly full dg- subcategory in Db(X0). That is, we assume that it includes, with each object, all objects quasi-somorphic to it. Doing this does not change the quasi-equivalence classes of our dg-categories. However, it assures that the embeddings α± : Perf X± → Db(X0) are fibrations in Qeq, as follows from the explicit description of fibrations in [48], Prop. 1.13 which we recall. Proposition 2.6. A dg-functor g : A → B is a fibration in Qeq, if and only if it is surjective on Hom-complexes and the following lifting condition holds: (F) Let A ∈ Ob(A) and u ∈ Hom0 B(g(A), B)) be a closed degree 0 morphism in B which becomes an isomorphism in the category H 0(B). Then there is A′ ∈ Ob(A) such that g(A′) is equal (not just isomorphic) to B and there is a closed u′ ∈ Hom0 A(A, A′) with g(u′) = u which becomes an isomorphism in H 0(A). Because all the three categories in question are perfect, the embeddings ± by α± are also fibrations in Mrt. So we have replaced the original Lp∗ fibrations and therefore H0(C, F) = Perf X+ ∩ Perf X− (intersection of homotopy strictly full subcategories in Db(X0)). So our state- ment reduces to: Proposition 2.7. Perf X+ ∩ Perf X− is equal to Perf Z, embedded via L(f+p+ = f−p−)∗ and extended to a homotopy strictly full subcategory. Proof of the proposition: It is clear that the Perf Z is contained in the in- tersection, so only the opposite inclusion needs to be proved. Since our statement is about inclusion of classes of objects, we will work in the context of triangulated, not dg, categories. Suppose A0 ∈ Db(X0) lies in the intersection, so there are two objects +(A+). A− ∈ Perf X− and A+ ∈ Perf X+ such that A0 ≃ Lp∗ Without loss of generality we can assume that A0 = Lp∗ −(A−) ≃ Lp∗ −(A−). Define AZ = Rf−∗ Rp−∗(A0) = Rf−∗(A−) ∈ Db(Z). 19 −(A−) ≃ Lp∗ The isomorphism φ : Lp∗ +(A+) implies that AZ ≃ R(f+)∗A+. Now, because of the possibly singular nature of Z and X± the functors Lf ∗ ± take values in the left unbounded derived categories D−(X±), and we can form the exact triangles in these categories induced by the adjunction mor- phisms: Lf ∗ Lf ∗ −(AZ) −→ A− −→ C−, +(AZ) −→ A+ −→ C+. Note that C± ∈ C± are objects of the D− versions of the null-categories. Further, the morphisms φ and Id (the latter expressing the commutativity of the diagram (1.8)), fit into a commutative square which, by axioms of triangulated categories, extends to a morphism of exact triangles: Lp∗ −Lf ∗ −(AZ) / Lp∗ −(A−) Lp∗ −(C−) Id φ Lp∗ +Lf ∗ +(AZ) / Lp∗ +(A+) / Lp∗ +(C+). This morphism is an isomorphism since φ and Id are. That is, Lp∗ Lp∗ +(C+). It follows that −(C−) ≃ T−,+(C−) = R(p+)∗Lp∗ −(C−) ≃ C+, and similarly T+,−(C+) ≃ C−. This means that the flop-flop functor F = T+,− ◦ T−,+ takes C− into itself. On the other hand, it is known that F acts on the null-category C− by the shift by 2 (see [10] Th. 5.7). This implies that C− = 0 = C+. Therefore, A− ≃ Lf ∗ −(AZ), i.e. A0 is isomorphic to an object in the image of Db(Z) in Perf X0. −(AZ), A+ ≃ Lf ∗ +(AZ), A0 ≃ Lp∗ −Lf ∗ It remains to show that AZ lies in Perf Z. Since Lf ∗ −(AZ) = A−, we have by adjunction: RHomZ(AZ, Oz) = R(f−)∗ RHomX−(A−, Ox), for any closed point x ∈ X− and its image z ∈ Z. (Here we used the obvious fact that R(f−)∗Ox = Oz.) Further, since the object in the right hand side is clearly in Db(Z) and the morphism f− is surjective, it follows that RHomZ(AZ, Oz) ∈ Db(Z) for any closed point z ∈ Z. This implies that AZ is perfect. Proposition 2.7 and Theorem 2.5 are proved. 20   / / /     / / Remark 2.8. The only essential geometric feature of the situation used in the proof of Theorem 2.5 is that T+,− ◦ T−,+ acts on the null-category C− by a shift. So the argument can possibly generalize to other situations. D. H1 c and H1. We define H1 c(C, F) = K := Ker(cid:8)Db(X0) (Rp−∗,Rp+∗) −→ Db(X+) ⊕ Db(X−)(cid:9), i.e., as the full subcategory in Db(X0) consisting of objects A such that Rp−∗(A) and Rp+∗(A) are quasi-isomorphic to 0. This category has been used in [10] to construct a spherical pair. We further define H1(C, F) = Db(X0)(cid:14)hLp∗ +(Db(X+)), Lp∗ −(Db(X−))i, the quotient by the minimal thick subcategory generated by the pullbacks of the Db(X±). Remark 2.9. Although H0(C, F ) is defined as the homotopy limit of the diagram of categories Lp∗ −−→ Db(X0) (cid:8)Db(X−) Lp∗ +←− Db(X+)(cid:9), the category H1(C, F ) is not defined as the homotopy colimit of the same diagram: that homotopy colimit is Db(X0). Our definition, which is a direct categorification of the cokernel of the map of vector spaces of the form E− ⊕ E+ → E0, can be perhaps seen as some "derived functor of the homotopy limit". Example 2.10. For the Atiyah flop (Example 1.5), the category K is equiv- alent to Db(Vectk) and is generated by the sheaf OP1×P1(−1, −1), where P1 × P1 is the exceptional fiber in X0. The category H1(C, F) is in this case also identified with Db(Vectk). In general, K = H1 c(C, F), if considered as a dg-category, is not smooth. The relation between it and H1(C, F) probably fits into the "categorical Poincar´e duality", see Remark 2.13. 21 E. H2 with compact support We define H2 in Mrt: c(C, F) as the homotopy pushout hGDb(X0) R(p−)∗←− Db(X0) (2.11) H2 c(C, F) := Db(X−) Db(X+) = = holim−−−→ Mrt(cid:8)Db(X−) R(p+)∗−→ Db(X+)(cid:9). Theorem 2.12. H2 c(C, F) is identified with Db(Z), where Z is as in (1.8). Remark 2.13. Suppose that Z and X± are projective. The relations H0(C, F) and H2 c(C, F), i.e., between Perf(Z) and Db(Z), is then that of duality. More precisely, let us consider them as dg-categories. Toen [50] has introduced a duality operation on dg-categories called the Morita duality: A 7→ A∨ = RHom(A, Perf k), where RHom is an approproately defined internal Hom functor in dgCat. Here Perf k is the dg-category of cochain complexes over k with bounded, finite-dimensional cohomology (a particular case of the notation Perf A from §2B). It follows from [16], Thm. A.1, that for any projective variety Z we have Z ≃ Db(Z) and so Perf ∨ H2 c(C, F) ≃ H0(C, F)∨. Note that a perverse schober F can be seen as a categorical analog of a perverse sheaf F with an identification F ∼→ F ∗, where F ∗ is the Verdier dual perverse sheaf. This corresponds to the Hom-pairing on any category. So the above identification can be seen as an instance of a categorical Poincar´e duality. Proof of Theorem 2.12: We will use the concept of the Drinfeld quotient C/A of a dg-category C by a full dg-subcategory A, see [24]. This construction is invariant under quasi-equivalences in the following sense. Suppose A α B / C β / D 22   /   / be a commutative diagram of dg-functors with horizontal arrows being fully faithful embeddings and α and β being quasi-equivalences. Then we have a quasi-equivalence of Drinfeld quotients C/A → D/B compatible with α and β. To relate this with our geometric situation, we prove the following theo- rem of independent interest. Theorem 2.14. Let f : X → Y be a projective morphism of quasi-projective schemes over k such that each geometric fiber of f has dimension ≤ 1 and Rf∗(OX ) ≃ OY . Let Cf ⊂ Db(X) be the kernel of Rf∗. Let us endow Db(X), Db(Y ) and Cf with natural structures of dg-categories. Then Rf∗ lifts to a quasi-equaivalence between Db(Y ) and the Drinfeld quotient Db(X)/Cf . The proof will be given in §F below. Remark 2.15. Theorem 2.14 has been stated in [15] (in the equivalent terms of Verdier rather than Drinfeld quotient) with no restriction on the dimension of fibers of f . The authors of [15] later realized that they do not have the proof of this statement. To the best of our knowledge, the problem remains unsolved for this general case. We apply Theorem 2.14 to the two rows in the diagram of dg-categories L+ h C− / Db(X0) Rp+∗ / Db(X+) Rp−∗ Rf+∗ / Db(X−) Rf−∗ / Db(Z), Here C− is the null-category (1.9). We conclude that Db(Z) ≃ Db(X−)/C− (Drinfeld quotient). Similarly, L+ is the kernel of Rp+∗, and we conclude that Db(X+) = Db(X0)/L+. The dg-functor h is induced by Rp−∗. We now note the following. Lemma 2.16. Let g : A → B be any dg-functor. Then the homotopy pushout A 0 (both in Qeq and Mrt) is quasi-equivalent to the Drinfeld quotient B/ Im(g), where Im(g) is the full dg-subcategory on objects from the image of g. BFh 23   /   /   / / Proof: In any model category, the homotopy pushout of a diagram of cofi- brant objects can be obtained by replacing one of the functors by a cofibration and then taking the usual push out. So we first replace A and B by cofibrant (in Qeq or Mrt) dg-categories, and, after changing the notation, assume that A and B are already cofibrant. Second, we replace the dg-functor A → 0 by A → A/A, which is a cofibration (because the natural functor to any Drinfeld quotient is a cofibration). Now, the usual pushout BFA(A/A) in dgCat is quasi-equivalent to B/ Im(g). Lemma 2.17. Let u : S1 → T1 be a dg-functor of perfect dg-categories such that H 0(u) : H 0(S1) → H 0(T1) is fully faithful and let T1 → T2 be any dg- functor. Let S2 ⊂ T2 be the strictly full dg-subcategory on objects from the (T1/S1) is identified with T1 T2/S2. image of S1. Then the homotopy pushout T2Fh T1(cid:19)GS1 0(cid:19) ≃ (cid:18)T2 hGT1(cid:18)T1 hGS1 T2 (T1/S1) ≃ T2 hGT1 hGT1 Proof: Using associativity of pushouts and Lemma 2.16, we rewrite: 0 ≃ T2 hGS1 2.16 ≃ T2/S2. 0 To apply Lemma 2.17 , we take T1 = Db(X0), S1 = L+ and T2 = Db(X−). This gives Db(X−)/S2 ≃ Db(X−) (Db(X0)/L+), hGDb(X0) where S2 is the strictly full subcategory on objects from the image of L+, i.e., the strictly full subcategory in C− on the objects from the image of the functor h. By Theorem 2.14 we have Db(Z) = Db(X−)/C− and Db(X+) = Db(X0)/L+. Thus to finish the proof of Theorem 2.12, it suffices to establish the following fact. Lemma 2.18. S2 = C−, i.e., the functor h is essentially surjective on objects. Proof: Let A− = C− ∩ Coh(X−), an abelian category. It was shown in [10] that: (1) There is an essentially surjective (spherical) functor Ψ− : Db(A−) → C− which extends the embedding of A−. 24 (2) There is an exact functor ep− : Db(A−) → Db(X0)/K such that Ψ− = R(p−)∗ ◦ep−. Here K is the kernel in §D above. (3) We also have R(p+)∗ ◦ep− = 0. Our statement follows from these properties. This finishes the proof of Theorem 2.12. F. Proof of Theorem 2.14. Note that the homotopy category of the Drinfeld quotient by a full dg-subcategory whose homotopy category is a thick subcategory of the ambient homotopy category is the Verdier quotient of the corresponding homotopy categories. For this reason, it is enough to prove the equivalence of Db(Y ) regarded as triangulated category, with the Verdier quotient Db(X)/Cf . We have the obvious functor Φ : Db(X)/Cf → Db(Y ). We will prove that Φ is an equivalence. This statement consists of three steps (I)-(III) below. (I) Φ is injective on Hom-spaces. Take a pair of objects A, B ∈ Db(X) and a morphism α : A → B. Suppose that Φ(α) = Rf∗(α) = 0. As in the proof of Proposition 2.7, we consider the functor Lf ∗ taking values in D−(X) and form an exact triangle with the first arrow the counit c of the adjunction: (2.19) Lf ∗Rf∗A c−→ A −→ C, C ∈ D−(X). By applying Rf∗ to this triangle, we see that C belongs to C− f , the D− version of the null-category for f . The composition of c with α gives the morphism γ : Lf ∗Rf∗A → B that corresponds by adjunction to Rf∗α, hence γ is zero. Then the long exact sequence on Hom-spaces from the triangle (2.19) to B shows that α factors via a morphism C → B. If C were in Cf (the bounded version of the null category), then this would imply that α is the zero morphism in the quotient category. As C is, in general, an object of C− f only, we proceed to find an appropriate bounded replacement for C. To this end, we consider a t-structure on D−(X) which restricts to a bounded t-structure on Db(X) and such that functor Rf∗ is t-exact for this t-structure. For instance, we can choose any of the Bridgeland's t-structures on the unbounded category D(X) [17], which are numbered by p ∈ Z, by fixing p, and restrict it to D−(X). 25 Consider the exact triangle of truncations for C in this t-structure: τ<kC → C → τ≥kC, k ∈ Z. Since B is in the bounded derived category, we can find k ≪ 0 such that Hom(τ<kC, B) = 0. Then, applying functor Hom(−, B) to the above triangle of truncations gives a lift of our morphism C → B to a morphism τ≥kC → B. Together with the composite A → C → τ≥kC, this gives a factorization of α via τ≥kC. Since Rf∗ is t-exact and C is in C− f , it follows that τ≥kC ∈ Cf . Hence α = 0 in the quotient category. This proves the claim (I). (II) Φ is surjective on Hom-spaces. Let β ∈ Hom(Rf∗A, Rf∗B). It induces a morphism β : Lf ∗Rf∗A → B, such that Rf∗eβ = β. As before, the source of eβ lies, a priori, in D−(X), and we consider a triangle of truncations: τ<k(Lf ∗Rf∗A) → Lf ∗Rf∗A → τ≥k(Lf ∗Rf∗A) with respect to the t-structure as in the proof of (I). Again, since A and B are bounded, we can choose k negative enough so that Hom(τ<k(Lf ∗Rf∗A), A) = 0 = Hom(τ<k(Lf ∗Rf∗A), B), which implies that the canonical morphism Lf ∗Rf∗A → A has a lift to a morphism δ : τ≥k(Lf ∗Rf∗A) → A and β has a lifting to a morphism β′ : τ≥k(Lf ∗Rf∗A) → B. By considering the octahedron related to the commutative triangle Lf ∗Rf∗A τ≥kLf ∗Rf∗A 'PPPPPPPPPPPPPP δ A, we conclude that the cone of the morphism δ lies in Cf . Hence we got a pair of morphisms (a roof): A δ←− τ≥k(Lf ∗Rf∗A) β′ −→ B, which can be interpreted as a morphism A 99K B in the quotient category. Moreover, one can easily see that it is mapped by Φ into β. This proves the claim (II). Together, (I) and (II) mean that Φ is is a fully faithful functor. 26 / / '   (III) Φ is essentially surjective (on objects). For any coherent sheaf F on Y , consider the zeroth cohomology sheaf with respect to our chosen t-structure as above: A = H0(Lf ∗(F )) Since Rf∗Lf ∗(F ) = F and Rf∗ is t-exact, it follows that Rf∗(A) = F , i.e. F lies in the image of the functor. Since the functor is fully faithful, the d´evissage of any object B ∈ Db(Y ) into its t-cohomology sheaves proves that B lies in the essential image of Φ. This ends the proof of Theorem 2.14. 27 3 The web of flops. Role of arrangements. A. Perverse sheaves on arrangements. Proposition 1.1 is the simplest instance of a more general result [35] which we now recall. Let H be an arrangement of hyperplanes in Rn. The complexification HC defines a natural stratification of Cn, and we denote by Perv(Cn, H) the corresponding category of perverse sheaves. In [35] it was described in terms of the chamber structure on Rn given by the arrangement H. More precisely, H subdivides Rn into locally closed real cells (of various dimensions). Open cells are called chambers. We have a poset (C = CH, ≤) formed by cells and inclusions of their closures. Theorem 3.1. Perv(Cn, H) is equivalent to the category of diagrams formed by vector spaces EC, C ∈ C, and linear maps EC γCC′ δC′ C / EC′, C ≤ C ′, satisfying the following relations: (0) Transitivity: if C ′′ ≤ C ≤ C ′, then γCC′′ = γC′C′′γCC′ and δC′′C = δC′CδC′′C′. (1) Idempotency: γCC′δC′C = Id for any C ≤ C ′. This, together with (0), allows us to unambiguously define the map ϕCC′ : EC → EC′ for any C, C ′ as the composition ϕCC′ = γDC′δCD where D is any cell such that D ≤ C, C ′. (2) Collinear transitivity: if three cells C1, C2, C3 are collinear, i.e., there are points ci ∈ Ci such that c2 lies in the closed straight line interval [c1, c3], then ϕC1C3 = ϕC2C3ϕC1C2. (3) Invertibility: if C, C ′ are two cells of the same dimension r lying in a linear subspace of dimension r and separated by a cell D ≤ C, C ′ of dimension (r − 1) ("wall"), then ϕCC′ is an isomorphism. Proposition 1.1 is obtained when n = 1 and H consist of one "hyperplane" {0} ⊂ R. B. Web of flops: the movable cone picture. General construction of 3-dimensional flops, of which the Atiyah flop (1.3) is the simplest instance, 28 / o o can be considered as an algebro-geometric analog of the procedure of surgery on a knot in 3-dimensional topology. That is, we remove a curve C+ from a variety X+ and fill X+ − C+ in a new way, by a curve C−, to get X−. The curves C+ (in smooth X) for which this is possible (flopping curves) have been classified. If C+ is irreducible, then C+ ≃ P1 and the normal bundle NC+/X+ can be one of the three types: O(−1)⊕2 (the Atiyah flop), O⊕O(−2) (the so-called pagoda flops), and O(1) ⊕ O(−3), see [42]. More generally, we can start with a reducible curve C+ =S Ci in a 3-fold X = X+ such that we can make a flop along some component Ci, getting a new 3-fold X1, then it may be possible to flop X1 along the strict preimage of some other component Cj, getting X2 and so on. We get in this way a "web of flops", a system of 3-folds Xν and flops connecting them. This system of flops can have loops: we may be able to obtain the same Xν by two or more different sequences of flops. According to Y. Kawamata [37], the structure of iterated flops is governed by the chamber structure of the movable cone MX . By definition, MX ⊂ Pic(X) ⊗ R is the cone generated by line bundles L on X such that the locus of base points of the linear system L has codimension ≥ 2. The open part of this cone is subdivided into chambers (certain open cones). For L inside each chamber, the variety XL is the same; in particular, the sign of the degree (i.e., the property of being ample or not) of the restriction of L to any fixed component of C+ is the same. Considering the closures of the chambers, their intersections etc., one gets a decomposition of the movable cone into cells of all dimensions. In particu- lar, when we cross a wall (a cell of real codimension 1) between neighboring chambers, the XL undergoes a flop. Further, cells of codimension ≥ 2 can be considered as relations, syzygies etc. among the flops. C. The elephant picture. The role of arrangements. The schober HMMP. It is a fundamental fact [14][17] that any 3-dimensional flop results in a derived equivalence: Db(X+) ∼→ Db(X−). The Homological Minimal Model Program (HMMP) studies, in particular, such derived equivalences and relations among them. If we have a web of flops (Xν), then all Xν have equivalent derived categories, and loops give self-equivalences. We get a local system of categories on the oriented graph whose vertices are the Xν and edges are the flops. In fact, it was proved by Donovan-Wemyss [23] that this extends to a local system of categories on Cn − HC, the complement of 29 a certain hyperplane arrangement on Cn with equations of the hyperplanes having real coefficients. Individual Xν correspond thereby to chambers of the real arrangement. More precisely, contracting a (possibly reducible) flopping curve C+ ⊂ X+ produces a 3-fold Z with an isolated singular point z. According to M. Reid's "elephant" picture [43], hyperlplane sections of Z through z are partial desingularizations of a Kleinian (ADE) singularity A2/Γ. So Z, near z, can be seen in terms of the total space of a 1-parameter deformation of A2/Γ, which can be obtained by base change from the universal deformation of A2/Γ as constructed by Brieskorn [18]. Brieskorn's deformation is defined over the Cartan subalgebra h of the corresponding ADE Lie algebra, with singular fibers forming the root arrangement of hyperplanes in h, and the Donovan- Wemyss arrangement is a certain subarrangement of this, corresponding to a partial, not full, desingularization of A2/Γ. Therefore, Theorem 3.1 appears very suggestive from the point of view of 3-dimensional birational geometry of flops. It leads to the following. Proposal 3.2. Local systems of triangulated categories appearing in HMMP typically defined on open strata of some stratified spaces, admit natural extensions to perverse schobers on the entire spaces. D. The general concept of H-schobers. Note that the conditions of Theorem 3.1 do not involve addition in the relations on the γ's and δ's and so can be formulated for a diagram of triangulated categories, not vector spaces. More precisely, let H be an arrangement of hyperplanes in Rn. We con- sider the poset (C = CH, ≤) of cells of H as a category. We denote by Cmin ∈ C the intersection of all the hyperplanes of H; one can assume that Cmin = {0} without changing the combinatorics, so we make this assumption in the sequel. Definition 3.3. By a triangulated 2-functor on C we will mean a 2-functor F from C to the 2-category of triangulated categories and exact functors. That is, F consists of the data: • For each C ∈ C, a triangulated category EC. • For each pair C ≤ C ′ in C, an exact functor γCC′ : EC → EC′. 30 • For each length 3 chain C ≤ C ′ ≤ C ′′ in C, an isomorphism of func- tors κC,C′,C′′ : γC′C′′γCC′ → γCC′′, and these isomorphisms satisfy the compatibilty condition for each length 4 chain of cells. Let now F = (EC, γCC′) be a triangulated 2-functor on C. Assume, in addition, that each γCC′ has a right adjoint δC′C. Then (EC, δC′C) form a triangulated 2-functor on Cop, the opposite poset of C. We now impose the following additional condition: (1) (Idempotency) Each natural transformation IdEC → γCC′δC′C (the unit of the adjunction), is an isomorphism of functors. This implies, first of all, that each δCC′ is an embedding of a left admissible [12] full triangulated subcategory, so we can think of all the EC as such subcategories in E0, the category correspond to the minimal cell {0} = Cmin. Let us define the flopping functors (3.4) ϕCC′ = γ0C′δC0 : EC −→ EC′ for any two cells C, C ′ ∈ C. The idempotency condition provides the canoni- cal identification ϕCC′ ≃ γDC′δCD where D is any cell such that D ≤ C, C ′. In particular, ϕCC′ = γCC′, C ≤ C ′; ϕCC′ = δCC′, C ≥ C ′. Further, given three arbitrary faces C, C ′, C ′′, the counit δC′0γ0C′ → IdEC′ induces a natural transformation (not necessarily an isomorphism) κCC′C′′ : ϕC′C′′ϕCC′ −→ ϕCC′′. For any four cells C, C ′, C ′′, C ′′′ the evident square of κ's commutes: (3.5) κCC′C′′′ ◦ (κCC′′C′′′ · ϕCC′) = κCC′′C′′′ ◦ (ϕC′′C′′′ · κCC′′C′′′) as transformations ϕC′′C′′′ϕC′C′′ϕCC′ −→ ϕCC′′′. Let [C] be the category whose objects are all elements of C and there is a unique morphism between any two objects. We can express (3.5) by saying that (EC, ϕCC′, κCC′C′′) form a lax 2-functor [C] → Cat. Definition 3.6. Let H be an arrangement of hyperplanes in Rn. An H- schober is a triangulated 2-functor F = (E, γCC′, κCC′C′′) on CH with each γCC′ admitting a right adjoint δC′C, satisfying the idempotency condition (1) above, as well as the following two conditions: 31 (3) (Collinear transitivity) If three cells C, C ′, C ′′ are collinear, then the natural transformation κCC′C′′ is an isomorphism of functors. (4) (Invertibility) if C, C ′ are two cells of the same dimension r lying in a linear subspace of dimension r and separated by a cell D ≤ C, C ′ of dimension (r − 1) ("wall"), then ϕCC′ is an equivalence of categories. We will use H-schobers as a possible avatar of a perverse schober on Cn smooth with respect to HC. Given an H-schober F, forming the ratio- nal Grothendieck groups K(E) gives a diagram which gives a perverse sheaf K(F) ∈ Perv(Cn, H). One way to interpret Proposal 3.2 is to construct H- schobers formed by derived categories of varieties appearing in a web of flops. We will refer to them as flobers. 32 4 The Grothendieck resolution A. Definition of the resolution and the g-web of flops. Let k be an algebraically closed field of characteristic 0. Let g be a reductive Lie algebra over k, corresponding to a reductive algebraic group G, and F ≃ G/B be its flag variety. It can be seen as parametrizing all the Borel subalgebras b ⊂ g, so there is a tautological bundle π : b → F . The Grothendieck resolution, see, e.g. [7] [9] [21] is the total spaceeg of this bundle, i.e., It comes with a natural projection g :eg → g, a proper map. eg = (cid:8)(x, b) ∈ g × F(cid:12)(cid:12)x ∈ b(cid:9). Let grss ⊂ g be the open subset of regular semisimple elements ("matrices with distinct eigenvalues"). Over grss, the projection g is an unramified Galois covering with Galois group W (the Weyl group of g). In fact, g factors into the composition (Stein factorization) of a finite morphism and a map f with connected fibers (which, in our case, happens to be birational, since these fibers generically consist of one point): (4.1) f−→ Z := g ×h/W h −→ g. eg Here h is the Cartan subalgebra in g, and we use the identification g/ Ad(G) = h/W to construct the "characteristic polynomial map" χ : g −→ g/ Ad(G) = h/W. The variety Z is singular whileeg is smooth. In particular, we can view points of Z as pairs (x, K), where x ∈ g and K is a connected component of g−1(x), the variety of all Borel subalgebras containing x. Example 4.2. Let g = sl2, then F = P1, and b ≃ O(−1)⊕2. The variety Z is the 3-dimensional quadratic cone (the 2-sheeted covering of g ≃ A3 ramified along the quadratic cone in A3). So f is the flopping contraction for the Atiyah flop (local model, Example 1.5). Remarks 4.3. (a) The morphism eg → h implicit in (4.1) comes from the following well known but important phenomenon [21]. The vector bundle bab = b/[b, b] of abelianized Borel subalgebras on F , is trivial, and its space of global sections h = H 0(F, bab) can be seen as the universal Cartan subalgebra 33 of g. More precisely, for any particular Cartan h ⊂ g, a choice of a Borel b ⊃ h identifies h with h; a different choice of b through h changes this a more intrinsic definition of Z is Z = g ×h/W h. identification by an element of W . So we have a canonical mapeg → h, and In the sequel we fix a distinguished Cartan h ⊂ g together with a Borel b+ ⊃ h so writing Z as in (4.1) becomes unambiguous. This choice fixes ∆ ⊃ ∆+ ⊃ ∆sim, the systems of roots, positive roots and simple roots of g, which we will use whenever needed. (b) Since Z is affine and f is proper, Z = Spec H 0(eg, O) is the affinization ofeg. The Grothendieck resolution gives rise to a very explicit "g-web of flops" constructed as follows. Note that the variety Z is acted upon by W , so we define the w-flopped contraction as the base change fw : Xw := w∗eg −→ Z, w ∈ W. Although all the Xw are isomorphic as algebraic varieties, they are connected by nontrivial birational isomorphisms coming from the birational identifica- tions fw with Z. This is similar to defining a 3d flop by base change along an involution (which is z 7→ zt in Example 1.5). Remark 4.4. According to Brieskorn [19], his simultaneous resolution of an ADE singularity is obtained from the Grothendieck resolution for the corre- sponding g, by a base change: restriction to the Brieskorn slice, a transversal slice to the subregular nilpotent orbit. Combining this with the elephant picture of 3-dimensional flops, see §3C, we can say that the Grothendieck resolution is "the mother of all flops" (at least of 3-dimensional ones). B. Singular nature of Z. By Example 4.2, the variety Z ⊂ g × h can be considered as the g-analog of the 3-dimensional quadratic cone. This analogy makes it natural to introduce the following varieties which we will use later. First, let pr : h → h/W be the projection. Inside h we have the root hyperplanes hα = α⊥, α ∈ ∆+. We denote by ∇ = pr(cid:0) [α∈∆+ hα(cid:1) ⊂ h/W the discriminantal hypersurface in h/W , i.e., the locus of points represented by non-free orbits of W . We further denote by D = χ−1(∇) ⊂ g the preimage 34 of ∇ and call it the discriminantal hypersurface in g. Thus g−D = grss is the open set of regular semisimple elements ("matrices with distinct eigenvalues"). By construction, the hypersurface D is the branch locus of the finite morphism : Z → g. Ramification points of Z, i.e., points lying over D, are generically smooth. For g = sl2, the variety D becomes a 2-dimensional quadratic cone in sl2 = A3 (the cone of nilpotent elements). Singularities of Z are governed by the closed subvariety gnr ⊂ g formed by non-regular elements. We recall several equivalent characterizations of this concept. (R) An element x ∈ g is called regular, if the centralizer gx of x is abelian, and non-regular otherwise, see [21]. The sets of regular and non-regular elements of g are denoted greg and gnr. (R') Alternatively, gnr consists of the critical points of χ, i.e., of x ∈ g such that the differential dxχ is not of full rank, see [38], Th. 0.1. (R") Any element x ∈ g has the (unique) Jordan (or Cartan) decomposition x = y + z where y is semisimple, z is nilpotent and [y, z] = 0. In this case gy, the centralizer of y, is reductive, and it is classical ([38], Prop. 0.4) that x is regular iff z is a principal nilpotent element in gy, i.e., lies in the open orbit in the nilpotent cone of gy. (R" ') x is regular if and only if F x = π(g−1(x)) ⊂ F , the variety of Borel subalgebras through x, is 0-dimensional, i.e., consists of finitely many points. Proposition 4.5. (a1) Dsing, the singular locus of D, is the union of two irreducible components: Dsing = gnr ∪ χ−1(∇sing). (a2) The variety gnr has codimension 3 in g and 2 in D. The variety χ−1(∇sing) is empty for g = sl2, and in all other cases has codimension 2 in g and codimension 1 in D. (a3) The intersection of D with the transversal slice at a generic point of gnr has a 2-dimensional quadratic cone singularity. 35 (b) The variety Zsing coincides with the preimage −1(gnr). It has codi- mension 3 in Z. The intersection of Z with the transversal slice at a generic point of gnr has a 3-dimensional quadratic cone singularity. Proof: (a1) We first prove that Dsing is contained in gnr ∪ χ−1(∇sing), that is, a point x lying in D ∩ greg and mapped by χ into a smooth point of ∇, is a smooth point of D. Indeed, by (R') the map χ is a smooth map near x, so near x the hypersurface D = χ−1(∇) is smooth. To see the reverse inclusion, gnr ∪ χ−1(∇sing) ⊂ Dsing, we first note that, by the same argument using (R'), the the intersection greg ∩ χ−1(∇sing) is contained in Dsing. So it remains to prove that gnr ⊂ Dsing. This will follow from the part (a3) below. (a2) The statement about codimension of gnr is well known. The variety ∇sing is the image, under pr : h → h/W , of the locus of points lying on more than one root hyperplane. So it is empty for g = sl2 and has codimension 2 in h in all other cases. Our statement follows from this by applying χ−1. (a3) Using the Jordan decomposition in (R"), we see that the generic non- regular case is when gy ≃ sl2 ⊕(abelian) ("simple coincidence of eigenvalues") and z = 0. Denote by g◦ nr the open part of gnr formed by such semisimple x = y. For y ∈ g◦ nr, a transversal slice to gnr at y can be taken to consist of y + t where t lies in the sl2-part of gy. Being identified with sl2, it is 3-dimensional, and its intersection with D is the cone of nilpotent elements in sl2. (b) First, let us prove that −1(greg) consists of smooth points of Z. Indeed, let x ∈ greg. Then by (R') there is a neighborhood U of x in g such that χ : U → h/W is a smooth morphism. Therefore −1(U) is the pullback, under this smooth morphism, of h mapping to h/W . Since h is smooth, so is the pullback. Next, let us prove that any point of Z lying over any x ∈ gnr is singular. For this it is enough to assume that x ∈ g◦ nr ("generic" case) and to prove, in this case, the last statement of (b). Assuming this, we see that over such x the ramification of is simple quadratic, as any preimage z ∈ Z of x has the stabilizer in W identified with Z/2. Together with (a3), this means that near such z, the variety Z looks like the product of a smooth manifold and a double covering of A3 ramified along a quadratic cone. Denote S = Zsing. We now analyze S in terms of the projection κ : Z → h. 36 Proposition 4.6. Irreducible components of S are in bijection with positive roots α ∈ ∆+. The component Sα corresponding to α, projects surjectively to the root hyperplane hα ⊂ h. Proof: Inside S = −1(gnr) we have the open dense subset S◦ = −1(g◦ where g◦ elements y with gy ≃ sl2 ⊕(abelian). An element y ∈ h lies in g◦ to h◦ necessarily disjoint) union of "conjugation sweeps" Ad(G) · h◦ consider the action of Ad(G) on g × h through the first factor, and let nr), nr is, as in the proof of Proposition 4.5(a), the variety of semi-simple nr iff it belongs nr is the (not α. Accordingly, α = hα −Sβ6=α hβ for some α ∈ ∆+. This means that g◦ diag(h◦ α) ⊂ h × h ⊂ g × h be the image of h◦ is the (disjoint!) union of the S◦ closure of S◦ α, we get the statement. α under the diagonal embedding. We conclude that S◦ α). Denoting by Sα the α = Ad(G) · diag(h◦ S12 (cid:4) Z (cid:4) S13 S23 Figure 1: Structure of Z up to codimension 4. Red lines represent compo- nents Sα of Zsing, with labels given for g = sl3. Remark 4.7. Several important properties of the variety Z have been estab- singularities. lished in [31] using the morphism f :eg → Z, which is a small resolution of C. The g-web of flops up to codimension 4. Let g(cid:4) nr be the closed subset in gnr formed by non-regular elements more complicated nr = gnr − g◦ 37 (i.e, other) than semisimple elements y with gy ≃ sl2 ⊕ (abelian). Using the Jordan decomposition, it is easy to see that g(cid:4) nr has codimension 1 in gnr and codimension 4 in g. Denote be the complement of this codimension 4 subvariety. Let also g(cid:3) = g − g(cid:4) nr Z (cid:4) = −1(g(cid:4)), Z (cid:3) = Z − Z (cid:4) = −1(g(cid:3)), eg(cid:3) = f −1(Z (cid:3)). The open subvariety Z (cid:3) is invariant under the W -action on Z, so we have the restriction of the g-web of flops to it: f (cid:3) w : X (cid:3) w = w∗eg(cid:3) −→ Z (cid:3). This web of flops can be described directly in terms of the standard Atiyah flops. Indeed, the singular locus of Z (cid:3) is the disjoint union of smooth varieties S◦ α, α ∈ ∆+, and transversely to each Sα, the variety Z (cid:3) looks like a 3- dimensional quadratic cone. Therefore, a neighborhood of Sα ⊂ Z (cid:3) has two small desingularizations corresponding to the two desingularizations of the 3d quadratic cone. One of them is given by the restriction of f (cid:3) : eg(cid:3) → Z (cid:3). Let us call it the positive desingularization of Z (cid:3) along S◦ α, and the other one will be called the negative one. Thus we have 2∆+ possible small desingularizations of Z (cid:3). Among these, we have W desingularizations X (cid:3) w . They are described as follows. Proposition 4.8. Let w ∈ W and α ∈ ∆+. Then f (cid:3) S◦ α: (a positive desingularization, if w(α) ∈ ∆+; a negative desingularization, if w(α) ∈ ∆−. w : X (cid:3) w → Z (cid:3) is, along Proof: We analyze the sets naturally labeling the combinatorial objects we want to compare. Let Irr(S) be the set of irreducible components of S. It has a W -action coming from the W -action on Z. From the proof of Proposition 4.6 we see that, as a W -set, Irr(S) is identified with the set of root hyperplanes hα, the set ∆/± of pairs [α] := {α, −α} of opposite roots. In the remainder of this proof we will use the more precise notation S[α] for components of S. a choice of one of two of small desingularizations of Z near that component. This set is also equipped with a W -action and we have a W -equivariant Let further fIrr(S) be the set of pairs consisting of a component of S and surjective map fIrr(S) → Irr(S) whose fibers have cardinality 2. 38 Lemma 4.9. We have a commutative diagram with vertical arrows being isomorphisms of W -sets: fIrr(S) ∆ Irr(S) / ∆/ ± . The lemma implies Proposition 4.8. Indeed, ∆+ ⊂ ∆ is a fundamental domain for the action of {±1} on ∆. So if we call a positive desingularization of S[α] the one corresponding to the positive root α inside [α], then the action of w ∈ W will send it to a positive or negative desingulariation according to whether w(α) is a positive or negative root. Proof of Lemma 4.9: We first consider the two small desingularizations eQ± → Q of a 3-dimensional quadratic cone Q ⊂ A4. They are naturally labelled by the two families of planes (2-dimensional linear subspaces) lying on Q, i.e., by the two families of generators of the 2-dimensional quadric P(Q) ≃ P1 × P1. Call them the (+)- and (−)-families. Planes from the (±)-family lift bijectively into eQ± while planes from the (∓)-family lift to eQ± as blowups. We now look at the two families of 2-planes in the transverse slices to the S[α] and show that they are naturally labelled by the roots (positive or negative) α ∈ ∆ themselves inside the pairs [α] of opposites. For this it is sufficient to take the slice at a point (x, K) ∈ S[α] such that x is a generic point of the root hyperplane hα ⊂ h ⊂ g in the chosen Cartan and K is the component of the variety of Borels through x such that the standard Borel b+ belongs to K. This slice is provided by the variety Z(sl2[α]) corresponding to the sl2-subalgebra sl2[α] ⊂ g with the set of weights [α]. The two families of 2-planes in Z(sl2[α]) are naturally labelled by the two Borels of sl2[α] through the chosen Cartan, i.e., by the two roots in [α]. Indeed, such a Borel is a 2-dimensional subspace and its lift to Z(sl2[α]) is a 2-plane representing the corresponding family. 39 / /     / 5 The g-web of flops and partial blowdowns. and cone picture A. The movable We now explain the analog, for the Grothendieck resolution, of the relation between flops and cham- bers of the movable cone. This relation is particu- larly transparent in our "local" situation, as all di- visors are movable. We keep the notation of the previous section. The space h∗ partial blowdowns. Cw h∗ R C+ R of real weights carries the coroot arrangement H = H∆ of hyperplanes Hα, α ∈ ∆+. Explicitly, Hα is the orthogonal complement of the coroot α∨ corresponding to α. Therefore we have the decomposition of h∗ R into cells of H. The cham- bers (open cells) Cw are labelled by w ∈ W via Cw = w−1(C+), where C+ is the dominant chamber, on which all α∨, α ∈ ∆+ are positive. Figure 2: Cells of the coroot arrangement. C This decomposition is the g-analog of the chamber structure of the mov- able cone for 3d flops. That is, denote X =eg. We have Pic(X) = Pic(G/B) = h∗ Z, the lattice of integer weights, the isomorphism given by pullback along π. For λ ∈ h∗ Z we denote by L(λ) = π∗O(λ) the corresponding line bundle. Then all L(λ) are movable. We denote Zλ = ProjZ(cid:18) ∞Mn=0 f∗ L(nλ)(cid:19). This is the local analog of "the image of the rational map to a projective space defined by the linear system L(λ)". Thus, if λ ∈ C+, then Zλ = X, since L(λ) is ample. If λ = 0, then Zλ = Z is the affinization of X. Proposition 5.1. (a) Zλ depends only on the cell C of H containing λ. So we will use the notation ZC for it. (b) For any two cells C ≤ C ′ we have a regular birational morphism qC′,C : ZC′ → ZC. These morphisms are transitive whenever C ≤ C ′ ≤ C ′′. (c) For any w ∈ W we have Zw−1(C) = w∗ZC as a variety over Z. 40 In particular, ZCw = Xw is the flopped Grothendieck resolution. The diagram formed by the ZC and qC′C is the g-analog of the diagram (1.3) defining the Atiyah flop. B. W -equivariance of the Zλ. To prepare for the proof of Proposition 5.1, we start with the following statement. Proposition 5.2. For any w ∈ W and λ ∈ h∗ variety over Z. Z we have Zw−1(λ) = w∗Zλ as a Proof: The smooth locus of Z is, by Proposition 4.5, equal to Zreg = −1(greg). It is a W -invariant open subset whose complement Zsing has, the generic part of S = Zsing given by the disjoint union of the S◦ α, it is a P1-bundle (coming from the small desingularizations of the transverse 3d by Proposition 4.5, codimension 3 in Z. Let egreg = f −1(Zreg) ⊂ eg. The Grothendieck resolution map f :egreg → Zreg is an isomorphism. The com- plement eg −egreg has, by the analysis of §4C, codimension 2 in eg. Over quadratic cones). This means that each vector bundle M on egreg extends canonically to a torsion free sheaf j∗M oneg, where j :egreg ֒→eg is the em- N fromeg to f −1(Zsm), the j∗M = N . phisms of the open subvarietyegreg ⊂eg. Proposition 5.2 now reduces to the bedding. In particular, if M was already the restriction of a vector bundle Because of the identification with Zreg, the group W acts by automor- following fact. Lemma 5.3. Let λ ∈ h∗ Z and w ∈ W . Then w∗(cid:0)L(λ)egreg(cid:1) ≃ L(w−1λ)egreg. Indeed, the lemma implies that for any n ≥ 0, the coherent sheaf f∗L(w−1(nλ)) on Z is identified with w∗(f∗L(nλ)), and so Zw−1λ ≃ w∗Zλ as a scheme over Z. Proof of Lemma 5.3: This is well known. We are grateful to R. Bezrukavnikov for the explanation regarding this fact and for suggesting the proof below. Let L and R be the line bundles in the left and right hand side of the isomorphism claimed in the proposition. They are naturally G-equivariant bundles, so it suffices to establish the equality [L] = [R] of their classes in the G-equivariant Picard group PicG(greg). Becauseeg −egreg has codimension 41 Z. 2, the restriction map PicG(eg) → PicG(egreg) is an isomorphism. On the other hand, the projectioneg → F identifies PicG(eg) with PicG(F ) = h∗ Insideegreg, consider the locally closed subvariety eO = (f )−1(O), where O is the adjoint orbit of a regular (semisimple) element x ∈ h. Then eO = Fy∈W Oy is the disjoint union of isomorphic lifts of O labelled by y ∈ W . Z and so the restriction PicG(egreg) → To see this, we note that the action of w onegreg permutes the Oy, and For each y we have that PicG(Oy) = h∗ PicG(Oy) is an isomorphism. It is therefore enough to show that for some (or, what is equivalent, for any) y the classes, in PicG(Oy), of the restrictions of L and R, are equal. the map from PicG(O1) ≃ h∗ the action of w on h∗ Z. Z induced by w, coincides with Z to PicG(Ow) ≃ h∗ C. Blowdowns via parabolic Grothendieck resolutions. For the proof of Proposition 5.1, and for future use, we give an independent construction of the varieties ZC satisfying parts (b) and (c), and then identify ZC with Zλ for any λ ∈ C. We start with the following standard fact. Proposition 5.4. (a) The poset C of cells in H is anti-isomorphic to the the poset of parabolic subalgebras in g containing h. Explicitly, the parabolic subalgebra pC corresponding to C is spanned by h and the root vectors eα for all α ∈ ∆ such that αC ≥ 0. (b) Under the above identification, cells C lying in the closure of the domi- nant chamber C+ correspond to parabolic subalgebras containing the standard Borel b+. For C ∈ C let PC ⊂ G be the parabolic subgroup corresponding to pC and FC = G/PC be the variety of conjugates of pC. We then have the parabolic Grothendieck resolution π−1 C ([p]) = p. egC = (cid:8)(x, p) ∈ g × FC(cid:12)(cid:12)x ∈ p(cid:9) πC−→ FC, Here [p] ∈ FC is the point corresponding to a parabolic p conjugate to pC. For any such p let np be its nilpotent radical, and lp = p/np the Levi quotient. Thus lp is a reductive Lie algebra, and we can form its own Grothendieck resolutionelp → lp and its affinization Z(lp) = Spec H 0(elp, O). Denote Z(p) the fiber product, which induces a finite covering of p, Z(p) := p ×lp Z(lp) −→ p. 42 Such covering can be formed for any parabolic subalgebra p conjugate to pC, and we denote by ZC the universal family of these coverings: (5.5) ZC := Z(p) = G ×PC Z(pC) −→ FC. If C ≤ C ′, then pC′ ⊂ pC, so PC′ ⊂ PC, and we have the projection Thus Z0 = Z while ZC+ =eg. ρCC′ : FC′ → FC. ConsideringegC andegC′ as universal bundles of parabolic αC′C :egC′ → ρ∗ CC′egC, where ρ∗ subgroups over FC and FC′ respectively, we have the G-equivariant map CC′ means the induced (pullback) bundle. The map αC′C is normalized uniquely by the requirement that over the base point [pC′] of FC′, it is equal to the embedding pC′ → pC. The map αC′C gives rise to a morphism qC′C : ZC′ → ZC satisfying part (b) of Proposition 5.1. Part (c) (W -equivariance of the ZC) also follows directly from the construction. Proposition 5.1 now reduces to the following fact. Proposition 5.6. If λ ∈ h∗ ZC as a variety over Z. Z lies in a cell C ∈ C, then Zλ is identified with D. Proof of Propositions 5.6. Because of W -equivariance of the ZC and Zλ, it is enough to consider the case when C ≤ C+ lies in the closure of the dominant chamber. As is well known, such faces are labelled by subsets I ⊂ ∆sim. The face CI ≤ C+ is given by the conditions (α∨, c) = 0, α ∈ I, (α∨, c) > 0, α ∈ ∆sim − I. We denote the corresponding standard parabolic subalgebra pI = pCI ⊃ b+, ZCI . Let ρI : F → FI be the canonical projection. Intrinsically, any Borel b is contained in a unique p conjugate to pI. This gives rise to the morphism the parabolic Grothendieck resoluton associated to CI byegI and set ZI = gI :eg →egI. Proposition 5.7. The variety ZI fits into a Stein factorization fI ❂❂❂❂❂❂❂❂ gI eg ZI I with I being finite and fI having connected fibers (and, in our case, being birational). egI 43 / /    ((x, p), K ′) where (x, p) ∈egI and K ′ is a connected component of g−1 Proof: The proposition means that k-points of ZI are identified with triples I (x, p), the variety of Borels b such that x ∈ b ⊂ p. Now, such Borels are in bijection with Borel subalgebras in lp containing x, the image of x in lp. So K ′ corresponds (in a bijective fashion) to a connected component K ′′ of the varieties of Borels in lp through x. Now, the set of such components K ′′ is, by definition, the fiber of Z(p) → p over x ∈ p. We now prove Proposition 5.6. Consider the commutative diagram Z qI ZI f fI ②②②②②②②②② I gI eg = b /egI = p π / / F = {b} φI / FI = {p}. πI If λ ∈ CI, then the line bundle O(λ) on F is the pullback O(λ) = φ∗ of a very ample line bundle OFI (λ) on FI. Let LI(λ) = π∗ IOFI (λ) I OFI (λ), a line I (z), z ∈ ZI. Indeed, any q−1 I (z) is identified with a closed subvariety of FI, the variety of all parabolics of type I, and the restriction of ∗ I LI(λ) is identified with the restriction of OFI (λ) from FI. Now, we identify the summand f∗ L(nλ) in the definition of Zλ: bundle onegI. Note that ∗ I LI(λ) is very ample on each fiber q−1 f∗ L(nλ) ≃ qI∗ fI∗ g∗ I LI(nλ) ≃ qI∗∗ I LI(nλ), where the second isomorphism comes from the fact that fI is proper with connected fibers. We also use that fI∗f ∗ I LI(λ) is very ample on fibers of qI, we have that I = Id. Now, since ∗ Zλ = ProjZ(cid:18) ∞Mn=0 f∗ L(nλ)(cid:19) ≃ ProjZ(cid:18) ∞Mn=0 qI∗ (∗ I LI(λ))⊗n(cid:19) ≃ ZI. Proposition 5.6 and therefore Proposition 5.1 are proved. E. Blowdowns up to codimenion 4. We now continue the analysis of §4C from which we retain the notation. For a cell C of H let Z (cid:3) C = q−1 C (Z (cid:3)). Recall that the singular locus of Z (cid:3) is the disjoint union of smooth components S◦ α in Z (cid:3) is a 3d quadratic cone with two possible small desingularizations, positive and negative. The morphism qC : Z (cid:3) C → Z (cid:3) is therefore a partial desingularization of Z (cid:3). More precisely: α, α ∈ ∆+, and the transverse slice to S◦ 44 o o     O O / / Proposition 5.8. Let α ∈ ∆+ and C be a cell of H. Then, in the variety ZC: • The components Sα with α∨C = 0 remain singular (nothing happens). • The components Sα with α∨C > 0 are resolved in the positive way. • The components Sα with α∨C < 0 are resolved in the negative way. (x, K) ∈ Sα with x ∈ hα generic and K being the component of Borels Proof: As with Proposition 4.8, this follows by looking at a point ex = through x such that b+ ∈ K. Since the the transversal slice to Sα at ex is identified with Z(sl2[α]), our statement reduces to the case of sl2. 45 6 Fiber products and varieties of simplices A. Fiber products and horizontal components. The system of partial blowdowns, constructed in §5, is the g-analog of the diagram (1.3), describing the Atiyah flop. However, the derived equivalence associated to the flop, is constructed using the diagram (1.4) featuring the fiber product of the two small desingularizarions. It is now our goal to construct such a diagram in the g-situation. As before, we denote by H the coroot arrangement in h∗ R. For any cell C of H we define the C-incidence variety This variety can be reducible. At the same time, all ZC′ project to Z = Z0 in a compatible, birational way, in fact biregularly over IXC = lim←− C′≥C ZC′ ⊂ YC′≥C ZC′. Zrss = −1grss ⊂ Z. Therefore, IXC has the horizontal component XC defined as the closure of the image of Zrss under the system of the rational maps inverse to the qC′ : ZC′ → Z. Recall that the chambers (open cells) Cw of H are labelled by w ∈ W , and qCw = fw : ZCw = Xw → Z is the flopped Grothendieck resolution. Proposition 6.1. XC is identified with the closure of the image of (f −1 w ) : Zrss −→ Yw: Cw≥C Xw. Proof: We show that the projectionQC′≥C ZC′ →QCw≥C ZCw is an isomor- phism of XC onto its image, which we temporarily denote by J. Indeed, the inverse map is given by sending a point (xw)Cw≥C ∈ J to the system (xC′)C′≥C ∈ XC where xC′ is defined as qCw,C′(xw) for any Cw ≥ C ′. Here we note that for any C ′ ≥ C, the value qCw,C′(xw) is, for any (xw) ∈ J, independent of the choice of Cw ≥ C ′, because it is so for (xw) lying in the image of the map (fw)−1 on Zrss. The following is clear by construction. Proposition 6.2. The XC form an representation of the poset CH of faces of H, by algebraic varieties. More precisely, for any C ≤ C ′ we have a regular, proper, birational map pC,C′ : XC → XC′, and these maps are transitive: for any C ≤ C ′ ≤ C ′′ we have pC,C′′ = pC′,C′′ ◦ pC,C′. 46 B. Role of the varieties XC. The natural next step would be to pass from the diagram (XC) to an H-schober (Definition 3.6), a diagram F of tri- angulated categories (EC, γCC′, δC′C). Intuitively, EC should be some coherent derived category of XC, while γCC′ and δC′C should be the pushforward and pullback along pCC′. The diagram F should represent a perverse schober on h∗ C smooth with respect to HC. However, the varieties XC can be singular, so a straightforward implemen- tation of this idea is difficult. For instance, in order to have both inverse and direct images well defined in the singular case, we would be forced to work with (left) unbounded derived categories which will destroy the Grothendieck groups and meaningful decategorifications. One possible approach is to construct natural desingularizations eXC of the singular XC and to define EC = Db(eXC). We make first steps in this a choice of eXC. However, Theorem 2.5 suggests the following conjecture describing the C, F). This conjecture can be formulated without reference to Conjecture 6.3. The intersection of the pullbacks of the categories Db(Xw) inside Db(X0) is equivalent to Perf(Z). direction in §7. analog of H0(h∗ C. Reduction of XC to X0 for a Levi. In order to indicate the de- pendence of our varieties XC, ZC as well as the arrangement H, the Car- tan subalgebra h etc. on g we will use, when necessary, the notations XC(g), ZC(g), H(g), h(g), etc. Let C be a cell of H(g), and pC be the corresponding parabolic subalgebra containing h, see Proposition 5.4. Let lC be the Levi quotient of pC and mC be the semi-simplification (quotient by the center) of lC. We realize mC as the subalgebra in g generated by the root generators eα with αC = 0. At the same time, let R · C ⊂ h∗ R(g) be the real subspace spanned by C. Then we have identifications and R(lC) ≃ h∗, h∗ R(mC) ≃ h∗ h∗ R(g)(cid:14)R · C. H(lC) ≃ H(g)≥C, H(mC) ≃ H(g)/C. Here H(g)≥C is the sub-arrangement of H(g) consisting of the hyperplanes which contain C, while H(g)/C is the quotient arrangement in the quotient 47 R(g)(cid:14)R · C whose hyperplanes are the images of the hyperplanes from space h∗ H(g)≥C, see, e.g., [35]. Accordingly, the cells of H(g)/C are the images, under the quotient map, of the cells C ′ of H satisfying C ′ ≥ C. We will denote such images by C ′/C. In particular, for any C ′ ≥ C we have the varieties ZC′(lC) and ZC′/C(mC) associated to the reductive groups lC and mC. We have the composite pro- jection pC → lC → mC. Proposition 6.4. For any C ′ ≥ C we have an identification ZC′(g) ≃ G ×PC(cid:0)pC ×mC ZC′/C(mC)(cid:1). These identifications take, for any C ′′ ≥ C ′ ≥ C, the projection ZC′′/C(mC) → ZC′/C(mC), to the projection ZC′′ → ZC′. The proposition means that ZC′(g) is obtained from ZC′/C(mC) by first performing a pullback onto pC and then forming the universal family over FC. Proof: This is a direct consequence of the definition of ZC′ as the universal family over FC′ given by (5.5). Corollary 6.5. (a) For any cell C of H = H(g) we have an identification XC(g) ≃ G ×PC(cid:0)pC ×mC X0(mC)(cid:1). Here X0(mC) is the principal component of the fiber product of all the flopped Grothendieck resolutions of mC over Z(mC). (b) In particular, if C is a wall (codimension 1 cell), then XC(g) is It coincides with the fiber product Xw′(g) ×ZC (g) Xw′′(g), where smooth. w′, w′′ ∈ W are such that C separates two chambers Cw′ and Cw′′. The corollary implies that each XC(g) locally looks like the product of X0(mC) and a smooth manifold, because it is obtained from X0(mC) by first forming a pullback under a smooth map and then forming a universal family over a smooth base. Proof: (a) follows directly from Proposition 6.4. Part (b) follows from (a) since mC ≃ sl2 in this case, and the case of sl2 corresponds to the Atiyah flop where the statement is well known, see §1B. 48 D. The variety of g-simplices. By an (ordered) g-simplex we mean a pair h ⊂ b ⊂ g formed by a Cartan subalgebra h and a Borel b containing it. Denote by h, b the dimensions of the h's and b's. Then we have the quasi- projective variety T ◦ g ⊂ Fl(h, b, g) formed by ordered g-simplices h ⊂ b. It is nothing but the quotient G/H where H ⊂ G is the maximal torus. Example 6.6. Let g = sln. A Cartan subalgebra h ⊂ g is the same as a choice of n points {xi}i∈I, I = n in Pn−1 which are in (linearly) general position, i.e., form an unordered simplex. Explicitly, h consists of linear operators which are diagonal in a basis of kn formed by vectors ei that represent xi. A choice of a Borel b ⊃ h amounts to ordering the xi, i..e, writing them as (x1, · · · , xn) (an ordered simplex). The variety T ◦ g is in this case identified with the open set (Pn−1)n gen ⊂ (Pn−1)n formed by n-tuples of points in general position. We recall that G-orbits on F × F are parametrized by elements of W (the Bruhat decomposition). In other words, for any two Borel subalgebras b, b′ we have their relative position d(b, b′) ∈ W . Thus d(b, b′) = 1 iff b = b′ while d(b, b′) = w0 (the maximal element in W ), iff b and b′ are in general position (i.e., (b, b′) lies in the open G-orbit in F × F ). In particular, given h ⊂ b, for any w ∈ W we have a unique Borel bw = bw(h) ⊃ h such that d(bw, b) = w. Let F W be the product of W copies of F ; its points will be written as systems (bw)w∈W of Borels. We then have a regular embedding u : T ◦ g −→ F W , (h ⊂ b) 7→ (bw(h))w∈W . If we view T ◦ then the w-component of u is the second component of the w-action map g as G/H and W as the quotient of the normalizer N(H) by H, Ad(w) : G/H −→ G/H, (h, b) 7→ (h, bw(h)). Definition 6.7. The variety of g-simplices Tg ⊂ F W is defined as the closure of the image of u. In the case g = sln there is a more direct description of Tg, cf. [3]. Instead of F W = F n!, consider the product of Grassmannians Gr(I, kn) labelled by all subsets I ⊂ {1, · · · , n}. We then have the regular embedding (6.8) v : (Pn−1)n Gr(I, kn), gen −→ YI⊂{1,··· ,n} (x1, · · · , xn) 7→ (cid:0)Span(xi)i∈I(cid:1)I⊂{1,··· ,n}, 49 which associates to (x1, · · · , xn) the system of the subspaces spanned by all possible subsets of {x1, · · · , xn}. Proposition 6.9. Tsln is identified with the closure of the image of v. Proof: Denote GrI = Gr(I, kn) the factor in the target of v corresponding to I. The flag variety F for sln is, classically, embedded in the product of Grassmannians Gr(p, kn), p = 1, · · · , n − 1 (as the incidence variety). Let us think of the factor Gr(p, kn) in this embedding as Gr{1,··· ,p}. Further, for each w ∈ W we consider the w-th factor Fw = F in F W and embed in into the product of Grassmannians Grw({1,··· ,p}), p = 1, · · · , n − 1. By taking the direct product of these embeddings over all w ∈ W , we embed F W into the following product of Grassmannians: η : F W −→ n−1Yp=1 YI⊂{1,··· ,n}, I=p W I Gr I {1,··· ,p} where W I w({1, · · · , p}) = I. We now notice: {1,··· ,p} ⊂ W is the set of permutations w ∈ W = Sn such that Lemma 6.10. The composition η ◦ u takes values in the product, over all W I p, I, of the small diagonals GrI ⊂ Gr I sln = (Pn−1)n T ◦ Proof of the lemma: Let x = (x1, · · · , xn) ∈ (Pn−1)n sln. Then, for w ∈ Sn, the Borel subalgebra bw(h) corresponds to the flag gen toQI GrI is identified with v. . The resulting morphism from gen correspond to h ⊂ b ⊂ {1,··· ,p} Span(xw(1)) ⊂ Span(xw(1), xw(2)) ⊂ · · · Span(xw(1), · · · , xw(n−1)) ⊂ kn. Our statement follows from this immediately. Proposition 6.9 is proved. E. X0 and the variety of simplices. We now realize the biggest hori- zontal component X0 ⊂Qw∈W Xw as a "fibration in cones" over the variety of simplices Tg. We denote by (6.11) h gen ֒→ Tg × g = (cid:8)(cid:0)(h ⊂ b), x(cid:1) ∈ T ◦ g × g(cid:12)(cid:12) x ∈ h(cid:9) α the universal bundle of Cartan subalgebras over T ◦ g . 50 in Tg × g. In Proposition 6.12. X0 is identified with the closure of h particular, it is a closed subvariety in Tg × g, conic (dilation invariant) in the g-direction. gen gen Proof: We note that we have a projection Zrss → T ◦ is hreg, the open subset of regular elements in h. So we write Zrss = h which exhibits it as an open dense subset in h g whose fiber over (h ⊂ b) reg . and, as such, is embedded into F W × gW . By definition, X0 is the closure (in Next, each Xw = w∗eg is identified witheg ⊂ F × g, but with projection to Z twisted by w. So X0 can be conisdered as a subvariety inegW =Qw∈Weg, egW ⊂ F W × hW ) of the image of Zrss ⊂ Z under the system of maps inverse to the projections Xw → Z (biregular over Zrss). We now notice that the composite map Zrss → F W × gW lands in Tg × g (with g ⊂ gW being the small diagonal). This means that X0 is identified with the closure of h in Tg × g which is the same as the closure of the vector bundle h , in which it is open and dense. gen reg F. The variety of reductions and the Cartanization of X0. Propo- sition 6.12 gives an affine morphism ρ : X0 → Tg whose fiber over a generic point b = (h ⊂ b) ∈ T ◦ g is the Cartan subalgebra h. The fiber ρ−1(b) over an arbitrary point b ∈ Tg can be seen as the union of the limit positions of such Cartans for all 1-parameter curves (k[[t]]-points) b(t) in Tg such that b(0) = b and b(t) ∈ T ◦ g for t 6= 0. In particular, if we write b ∈ Tg ⊂ F W as a system of Borels (bw)w∈W , then ρ−1(b) is contained in the intersection Tw∈W bw. We now produce partial desingularization of Tg and X0. Definition 6.13. [32, 33] Let R◦ g ⊂ Gr(h, g) be the variety of all Cartan subalgebras. The variety of reductions for g is defined to be the closure Rg = R◦ g ⊂ Gr(h, g). By construction, Rg carries the universal bundle a ⊂ Rg × g of rank h whose fibers are "limit positions of Cartan subalgebras". They are abelian subalgebras in g. We then have the embedding (6.14) g −→ F W × Rg, bu : T ◦ (h ⊂ b) 7→(cid:0)(bw(h))w∈W , h(cid:1) and we define the Cartanized variety of g-simplices bTg to be the closure of the image ofbu. Thus we have a regular birational map τ : bTg → Tg which is 51 the composition of the closed embedding bTg ⊂ Tg × Rg and the projection of the product to Tg. Similarly, we consider the embedding h gen bα ֒→ Tg × g × Rg, ((h ⊂ b), x) 7→(cid:0)α((h ⊂ b), x), h(cid:1), where α is as in (6.11). Define the variety bX0 as the closure of the image of bα. Proposition 6.15. bX0 lies (after permuting the second and third factors in the target of bα) in bTg × g ⊂ Tg × Rg × g. Further, we have a commutative diagram a σ where: Rg / X0 ρ / Tg τ bρ bX0 bTg (a) σ and τ are regular, birational and proper. (b) The left square is Cartesian, in patricular, bρ exhibits bX0 as the total space of a rank h vector bundle on bTg. (c) For b ∈ Tg the preimage ρ−1(b) ⊂ g is the union, over all bb ∈ τ −1(b), of the vector subspaces bρ−1(bb). Proof: The first statement (about bX0) follows from the construction of bTg as the closure of the lift of T ◦ g into Tg ×Rg. The construction and commutativity of the diagram is just a restatement of the steps above. Let us prove the claimed properties of the diagram. (a) σ is regular and proper since it is obtained from a proper morphism (projection) Tg × g × Rg → Tg × g by taking the closure of a lift of h to the source and mapping in to the closure of the image of this lift in the target. . The argument for τ is It is birational since it is an isomorphism on h similar. gen gen 52   o o   /   o o / total space of the pullback of the vector bundle a to Tg × Rg. (b)This follows because we can also define bX0 as the closure of h in the gen (c) By definition of X0 as the closure, ρ−1(b) is the union of all the limit positions of the Cartans associated to b′ ∈ T ◦ g approaching b in 1-parameter families. These limit positions will represent points of Rg and thus points positions will become vectors in the fibers of the tautological bundle a over of bTg, more precisely of bb ∈ τ −1(b). The individual elements of these limit Rg, i.e., vectors in the vector subspaces bρ−1(bb). 53 7 The flop diagram for sl3 and Schubert's vari- ety of complete triangles A. Complete triangles vs. Cartanization. From now on we restrict the discussion to g = sl3. The variety T = Tsl3 of sl3-simplices is the classical variety of triangles [46, 47, 34]. It has dimension 6 and is embedded into (P2)3 × (P2)3, see (6.8). Here P2 = Gr(2, k3) is the projective plane of lines in P2. Below are some of the well known properties of T . Proposition 7.1. (a) T coincides with the incidence variety (cid:8)(x1, x2, x3, l12, l13, l23) ∈ (P2)3 × (P2)3(cid:12)(cid:12) xi, xj ∈ lij(cid:9). (b) The singular locus T sing consists of (xi, lij) such that x1 = x2 = x3 and l12 = l13 = l23. It is thus identified with F , the flag variety for sl3. (c) Near T sing = F , the variety T locally looks like the product of a 3- dimensional affine space and a 3-dimensional quadratic cone. This has been proved in [47], Th. I, where explicit charts were con- structed. For a more modern treatment of (a) and (b), see [34], Th. 1(b) and Lemma 7. Because of the 3d quadratic cone nature of the singularities of T , it has two small desingularizations, connected by a flop: (7.2) T Sch τSch−→ T τFM←− T FM. Thus for any b ∈ T sing = F , the preimages τ −1 FM(b) are isomorphic to P1. The variety T Sch is known as Schubert's variety of complete triangles and T FM is the Fulton-MacPherson blowup of (P2)3. They are both acted upon by SL3. Intrinsically, they are distinguished by the fact that the action of the maximal torus (Gm)2 ⊂ SL3 on T Sch has isolated fixed points, while the fixed points in T FM can be non-isolated, see [34, 45]. Sch(b) and τ −1 We recall [34, 45, 47] the definition of T Sch going back to Schubert [46]. Consider the 6-dimensional space S2(k3∗) of quadratic forms on k3. Every triple (x1, x2, x3) ∈ (P2)3 gen defines a 3-dimensional subspace Nx1,x2,x3 = (cid:8)q ∈ S2(k3∗)(cid:12)(cid:12) q(x1) = q(x2) = q(x3) = 0(cid:9) 54 (the net of quadrics through x1, x2, x3). We then consider the embedding vSch : (P2)3 gen −→ (P2)3 × (P2)3 × Gr(3, S2(k3∗)), By definition, T Sch is the closure of the image of vSch. The fact that T Sch thus defined, is smooth, was proved directly in [47], Th. II. (x1, x2, x3) 7→(cid:0)(x1, x2, x3), (Span(x1, x2), Span(x1, x3), Span(x2, x3)), Nx1,x2,x3(cid:1). Proposition 7.3. The Cartanized variety of triangles bT is isomorphic to corresponds to τSch). In particular, bT is smooth, while T Sch carries a rank 2 Schubert's variety T Sch (as a variety over T , so that τ in Proposition 6.15 bundle of abelian Lie subalgebras in sl3. Proof: This is a consequence of the interpretation, given in [32], of the variety R = Rsl3 of reductions for sl3. A Cartan subalgebra h ⊂ sl3 can be seen as corresponding to some x = (x1, x2, x3) ∈ (P2)3 gen defined uniquely up to permutation; we write h = hx. Thus, tautologically, R is the closure of the image of the embedding (P2)3 gen/S3 −→ Gr(2, sl3), x 7→ hx. It was shown in [32] Prop. 4.1 that R is equal to the subvariety in Gr(2, sl3) formed by all 2-dimensional abelian Lie subalgebras, i.e., to the intersection in the Plucker embedding of Gr(2, sl3). Gr(2, sl3) ∩ P(cid:0)Ker(cid:8)Λ2sl3 [−,−] −→ sl3(cid:9)(cid:1) At the same time, any x as above gives a 3-dimensional subspace Nx ⊂ S2(k3∗), so we have the embedding It was proved in [32], Th. 4.2, that the closure of the image of ν is identified with R. This is based on the isomorphism of sl3-modules x 7→ Nx. ν : (P2)3 gen/S3 −→ Gr(3, S2(k3∗)) ⊂ P(cid:0)Λ3(S2(k3∗))(cid:1), −→ sl3(cid:9) ≃ Λ3S2(k3∗). Ker(cid:8)Λ2sl3 [−,−] describing the ambient spaces of the two Plucker embeddings, see [32], Lemma 4.3. We can therefore compare the embeddings bu from (6.14) defining bT and the embedding vSch defining T Sch. We see that they both factor through the same map into (P2)3 × (P2)3 × R with R embedded in Gr(2, sl3) in the first case and in Gr(3, S2(k3∗)) in the second case. 55 Corollary 7.4. The Cartanized variety bX0 for g = sl3 is smooth. Proof: Indeed, bX0 is the total space of a rank 2 vector bundle (of abelian Lie subalgebras in sl3) over the smooth variety bT = T Sch. Using Proposition 6.15, we can now get a more detailed picture of the original variety X0 ⊂ T × sl3 in terms of the projection ρ : X0 → T . Corollary 7.5. (a) If b ∈ T is a smooth point, then ρ−1(b) is a 2-dimensional abelian Lie subalgebra in sl3. (b) If b ∈ T sing = F corresponds to a Borel subalgebra b ⊂ sl3, then ρ−1(b) = [b, b] is 3-dimensional. (c) In the situation (b), the preimage τ −1(b) ⊂ bT is identified with the variety of all 2-dimensional abelian Lie subalgebras in n = [b, b]. Such sub- algebras are precisely the 2-dimensional vector subspaces in b containing the center [n, n]. In particular, there are P1 of them. (d) The variety X0 is singular along T sing × {0} ⊂ T × sl3. Proof: (a) follows from the identification of τ with τSch and the known fact that τSch is bijective outside of T sing. Part (b) is a consequence of the following elementary lemma which ex- plains the appearance of n. Lemma 7.6. Let pi(t), i = 1, · · · , 3, be three k[[t]]-points of P2 with the following properties: (1) For t 6= 0, i.e., as k((t))-points, the pi(t) are in general position. (2) At t = 0, all three pi(t) evaluate to the same point p ∈ P2(k), and the three lines (pi(t), pj(t)) evaluate to the same line l ⊂ P2 (containing p). Let b ⊂ g = sl3 be the Borel subalgebra fixing the flag (p, l) and n = [b, b]. Let further x(t) ∈ sl3(k[[t]]) be a family of matrices such that for t 6= 0 the matrix x(t) has three points pi(t) as eigendirections. Then x(0) ∈ n. Moreover, each element of n can be obtained as x(0) for an appropriate pi(t), i = 1, · · · , 3 and x(t) as above. Proof of the lemma: Clearly x(0) fixes p, l so x(0) ∈ b. Let ξ be the gobal vector field on P2 given by x(0). The two independent eigenvalues 56 of x(0) ∈ sl3 can be read off the two eigenvalues of the transformation in- duced by ξ on the tangent space TpP2. Now, the first eigenvalue (on the line Tpl) vanishes because the points p1(t) and p2(t) specialize for t → 0, to p and the line (p1(t), p2(t)) specializes to l. The second eigenvalue (on the quotient TpP2/Tpl) vanishes because the third point p3(t) is such that the lines (pi(t), p3(t)), i = 1, 2 specialize, as t → 0, to the same line l. This shows that x(0) ∈ n. The fact that any element of n can be obtained like this is obvious and left to the reader. Part (c) follows from (b), from Proposition 6.15(c) and the fact ([32] Prop. 4.1) that R coincides with the variety of all 2-dimensional abelian Lie subalgebras in sl3. To see (d), consider the Zariski tangent space to X0 at (b, 0) ∈ T sing×{0}. It contains, as transverse direct summands, first, the tangent space to T at b (of dimension 7), and, second, the fiber of the projection ρ over b, i.e., b (of dimension 3). So it is 10-dimensional, while X0 is 8-dimensional. This finishes the proof of Corollary 7.5. B. The flop diagram for sl3. The coroot arrangement H for sl3 consists of three lines in the real plane h∗ It has 13 cells: six chambers Cw, six walls (open half-lines) and the point 0. Accordingly, the diagram (XC, pCC′)C≤C′ consists of 13 varieties: R ≃ R2. (2) Six flopped Grothendieck resolutions Xw = XCw , w ∈ S3. They are smooth. (1) Six binary fiber products XC corresponding to the half-lines C. They are smooth, see Corollary 6.5. (0) The "central" variety X0, singular. Definition 7.7. (a) The flop diagram for sl3 is the diagram (YC, lCC′)C≤C′ of proper birational regular maps between smooth varieties, obtained from σ→ X0 and leaving the other varieties unchanged. That is, we put (XC, pCC′) by replacing X0 with its Cartanization bX0 lCC′ =(pCC′, if C 6= 0, if C = 0; if C 6= 0, if C = 0 < C ′. p0C′ ◦ σ, YC =(XC, bX0, 57 (b) The Schubert transform is the diagram of birational morphsms Y−C l0,−C←− Y0 l0,C−→ YC associated to any 1-dimensional face (ray) C of H. The remainder of this section prepares the ground for the study of the diagram formed by the categories Db(YC) in the next section 8. C. Central fibers. The partial triangle picture. We first summarize the main features of the varieties YC and XC for arbitrary C. Proposition 7.8. (a) We have a diagram FC ρC YC τC / g FC / {0} in which the square is Cartesian. (b) The map ρC represents YC as the total space of a vector bundle on FC which we denote LC. Every fiber of LC is a Lie subalgebra in g. (c) The map (ρC, τC) : YC → FC × g embeds YC as a closed subvariety in FC × g. In other words, it realizes LC as a vector subbundle of the trivial bundle FC × g. The variety FC = τ −1 C (0) will be called the central fiber of YC. We now summarize the structure of the vector bundles LC. Denote by F = F (1, 2, k3) ⊂ P2 × P2 the flag variety of SL3. Proposition 7.9. (a) If dim(C) = 2, then FC ≃ F has dimension 3, the bundle LC has rank 5 and every fiber of LC is a Borel subalgebra in g. (b) If dim(C) = 1, then FC is isomorphic, as a SL3-manifold, to F ×P2 F or F ×P2 F and has dimension 4. The bundle LC has rank 4. (b1) Let q = (p, l, l′) ∈ F ×P2 F be a point with l 6= l′. Then, in the above identification, the fiber of LC over q is identified with the intersection of two Borels corresponding to the flags (p, l) and (p, l′), i.e. with the space 58 o o / O O / O O of global vector fields on P2 which preserve p, l and l′. At q = (p, l, l) the fiber Lq is the (well defined) limit position of the above intersections of Borels corresponding to neighboring points q′. Explicitly, Lq consists of global vector fields on P2 which preserve p and l and whose restricttion to l has vanishing linear part at p. (b2) Let q = (p, p′, l) ∈ F ×P2 F is a point with p 6= p′. Then, in the above identification, the fiber of LC over q is identified with the intersection of the Borels corresponding to the flags (p, l) and (p′, l), i.e., with the space of global vector fields on P2 preserving p, p′ and l. At q = (p, p, l) he fiber Lq is the (well defined) limit position of the above intersections of Borels corresponding to neighboring points q′. Explicitly, Lq consists of global vector fields on P2 whose linear part of p (an endomorphism of TpP2 preserving the subspace Tpl) induces zero endomorphism of the quotient TpP2/Tpl. (c) If C = {0}, then FC = bT is the Schubert space of complete triangles and has dimension 6. The bundle LC has rank 2 and is formed by Cartan subalgebras in g and their limit positions. It is convenient to depict the central fibers FC and the varieties YC symbolically by "partial trian- gles", i.e., by certain parts of the picture (Fig. 3) consisting of three points p1, p2, p3 and three lines l12, l13, l23 joining them. This is depicted in Fig. 4. The six chambers labelled by permutations of 1, 2, 3, correspond to six flags in the triangle, the six rays correspond to parts formed by either two vertices and an edge through them, or by two edges and a vertex common to them. The face {0} correspond to the full triangle of Fig. 3. l23 l13 • p3 p1 • p2 • l12 Figure 3: The sym- bolic triangle. 59 l23 p3 • p2 • l23 p2 • l12 l23 l12 p2 • p1 • p2 • l12 •p2 213 p1 • l12 l13 231 321 123 132 312 •p3 l13 •p3 l23 •p3 l13 l23 •p1 l12 l13 • p1 l13 • p3 • p1 Figure 4: The partial triangle notation for faces C and varieties YC, FC. 60 If C 6= {0}, then the central fiber FC is sim- ply the incidence variety associated to the corre- sponding picture. For instance, if C is the right horizontal ray in Fig. 4, then FC is formed by (p1, l12, l23) ∈ P2 × P2 × P2 such that p1 ∈ l12 and p1 ∈ l13. If C = {0}, then the incidence variety associated to the picture is T , the space of triangles, and the central fiber F0 is its desingularization bT . D. Fiber products: 1-ray and 2-ray varieties. Let us call any variety YC = XC associated to C which is a ray (1-dimensional face) of H, a 1-ray variety, and the corresponding central fiber FC = F ×P2 F or F ×P2 F a 1-ray central fiber. Note that FC is just an elementary Schubert correspondence (i.e., the correspondence associated to a simple reflection in the Weyl group) in F ×F , resolution. These are precisely the correspondences used by Bezrukavnikov- the action of the generators of Brg. while YC ⊂eg ×eg is the lift of such a correspondence to the Grothendieck Riche [9] to construct an action of the braid group Brg on Db(eg): they give In order to prove the collinear transitivity conditions for the sl3-flober in §8, we will need certain fiber products of the correspondences YC. These products have been studied in [44], §2. At the same time, related varieties (the central fibers of the fiber products) appear in the classical study of trian- gle varieties [45]. We present a treatment which emphasizes this connection. Let A0, · · · , A3 be four consecutive chambers of H and A01, A12, A23 be the rays between them, see Fig. 5. We denote by Y(2) = Y12,23 := YA12 ×YA2 YA23 and call the 2-ray variety the fiber product of two 1-ray varieties over YA2 ≃eg. (The chamber A0 is not used in this definition but will be used in defining the 3-ray varieties later.) 61 A23 A12 A2 A3 A1 A0 A01 l23 p1 • p2 • l12 Figure 5: Four consecutive chambers and the partial triangle code for a 2-ray variety. We now extend the partial triangle notation to 2-ray varieties. If we assume that the positions of A0, · · · , A3 in Fig. 5 correspond to Fig. 4 (i.e., A1 corresponds to 123, A2 corresponds to 213 etc.) then Y(2) is encoded by the partial triangle in Fig. 5 on the right. We define the 2-ray central fiber as the fiber product of the central fibers of the factors of Y(2), i.e., as the incidence correspondence associated to the partial triangle code: F(2) = F12,23 := FA12 ×FA2 FA23 = = (cid:8)(p1, p2, l12, l23) ∈ P2 × P2 × P2 × P2(cid:12)(cid:12) p1, p2 ∈ l12, p2 ∈ l23(cid:9). This is nothing but the Demazure resolution of the Schubert correspondence in F × F associated to the reduced decomposition (23)(12) in W = S3. By construction we have a diagram i Y(2) / F(2) × g $❍❍❍❍❍❍❍❍❍ ρ π F(2) where i is a closed embedding and π is the projection. Each fiber of ρ is a Lie subalgebra in g. In the following it will be convenient to identify elements ξ ∈ g with global regular vector fields on P2. Proposition 7.10. (a) F(2) is a smooth variety of dimension 5, which is a P1-bundle over FA12, as well as a P1-bundle over FA23. 62 $ /   (b) The map ρ realizes Y(2) is the total space of a rank 3 vector bundle L(2) = L12,23 over F(2) = F12,23, whose fibers are Lie subalgebras in g. In particular, Y(2) is smooth. (c) The projection σ : Y(2) → YA12 exhibits Y(2) as the blowup of YA12 (the total space of a rank 4 bundle LA12 on FA12) along the total space of a rank 2 subbundle E. Explicitly, the fiber of E at (p1, p2, l) consists of global vector fields ξ on P2 which belong to (LA12)(p1,p2,l) and whose linear part at p2 is the scalar operator in Tp2P2. In the case p1 = p2 this linear part must be 0, since the correspinding elements of g = sl3 must be nilpotent. Similarly for the projection to YA23. (d) The square Y(2) / YA12 YA23 / YA2 is Tor-independent, i.e., Y(2) is identified with the derived fiber product YA12×h YA23. YA2 Proof: (a) is obvious. Let us prove (b). By definition of the fiber product, the fiber of ρ over q = (p1, p2, l12, l23) is the intersection (L(2))q := (LA12)(p1,p2,l12) ∩ (LA23)(p2,l12,l23) ⊂ g of two Lie subalgebras in g. If q is a generic point (i.e., p1 6= p2 and l12 6= l23), this intersection is simply the subalgebra in g which preserves the points p1, p2 and the lines l12 and l23, and this subalgebra has dimension 3. We now claim that for any other q ∈ F(2) the above intersection has the same dimension 3. By P GL3-equivariance, it suffices to check one point in each P GL3-orbit on F(2). We consider only the "most degenerate" case p1 = p2 = p and l12 = l23 = l. In this case, by Proposition 7.9(b), the subalgebra (L(2))q consists of global vector fields ξ on P2 which preserve p and l and whose linear part at p has trivial restrictions to the tangent space Tpl and to the quotient Tp(P2)/Tpl. Global ξ preserving p and l form a Borel subalgebra b ⊂ g, and the linear parts in question correspond to the diagonal elements of a triangular matrix. So the condition on the linear parts simply means that ξ lies in the commutant n = [b, b]. This commutant has dimension 3. (c) We look at the fibers of σ. Suppose we have a point (q′, ξ) ∈ YA12, so q′ = (p1, p2, l12) ∈ FA12 and ξ ∈ (LA12)q. Then σ−1(q′, ξ) consists of 63   /   / (q′, ξ′) such that (q = p1, p2, l12, l23) projects to q′ and ξ′ = ξ ∈ (L(2))q. In other words, the freedom consists in choosing the new line l23 which must be invariant under ξ. Now, let Λ ∈ End(Tp2P2) be the linear part of ξ at p2. If Λ is not a scalar operator, there is precisely one possible l23: the line tangent to the other eigen-direction of Λ (or to the unique eigen-direction, if Λ is not semisimple). If, however, Λ is a scalar operator, then any l23 through p2 is good and we have a P1 worth of choices. So σ has fibers P1 over the total space of the subbundle E and is an isomorphism on the complement. Since both the source and target of σ are smooth varieties of the same dimension 8, the identification with the blowup follows. Finally, part (d) follows from the next general proposition. It has been formulated in [40] with the reference to [39], but the latter text does not explicitly contains the proof of the fact. We are thankful to A. Kuznetsov for explanations and reproduce the proof here for the sake of completeness. Proposition 7.11. Let XT = X ×S T ϕ′ ψ′ T ϕ / X ψ / S be a Cartesian square in which X, T, S are smooth irreducible varieties. Sup- pose further that each component of XT has the same dimension which is equal to dim(XT ) = dim(X) + dim(T ) − dim(S). Then the square is Tor-independent. Proof of Proposition 7.11: The statement is obvious, if ϕ is a smooth mo- prhism. Let us point another particular case. Lemma 7.12. [39] Suppose, in addition, that ϕ is a closed embedding (of a smooth subvariety into a smooth variety). Then the square is Tor-independent. Proof of the lemma: The statement is local. Locally T ⊂ S is given by n independent equations f1 = · · · = fn = 0. So OT has a Koszul resolution OT ∼(cid:0)OS[ξ1, · · · , ξn], d(cid:1), 64 deg(ξi) = −1, d(ξi) = fi.   /   / Accordingly, OT ⊗L OS OX is represented by the Koszul complex K• := OX ∼(cid:0)OS[ξ1, · · · , ξn], d(cid:1), d(ξi) = efi := fi ◦ ψ ∈ O(X). But the condition on the dimension of XT implies that ef1, · · · ,efn form a regular sequence and so K• ∼ OXT . End of proof of Proposition 7.11: We factor ϕ as the composition γ −→ S × T πS−→ S, T γ(t) = (ϕ(t), t), πS(s, t) = s of a regular embedding and a smooth projection. So we decompose our square as the concatenation of two other Cartesian squares XT / X × T πX / / X T γ / S × T / S πS of which the right one is Tor-independent because πS is smooth an the left one is Tor-independent by the lemma. E. 3-ray varieties. Let A0, · · · , A3 be four consecutive chambers as in Fig. 5. We then have two 2-ray varieties Y01,12 = YA01 ×YA1 YA12, Y12,23 = YA12 ×YA2 YA23, . both isomorphic to Y(2). The 3-ray variety is defined as their fiber product over YA12: Y(3) = Y01,12,23 := Y01,12 ×YA12 Y12,23 = YA01 ×YA1 YA12 ×YA2 YA23, so that we have the Cartesian square (7.13) Y(3) ζ1 ζ2 Y01,12 σ1 Y12,23 / YA12. σ2 We encode Y(3) by the partial triangle in Fig. 6. Note the absence of the point p3. By the 3-ray central fiber we will mean the fiber product of the central 65   /     / / / /     / fibers of the factors of Y(3), which is the same as the incidence correspondence associated to the partial triangle code: F(3) = F01,12,23 = F01,12 ×FA12 F12,23 = FA01 ×FA1 FA12 ×FA2 = (cid:8)(p1, p2, l12, l13, l23) ∈ (P2)2 × (P2)3(cid:12)(cid:12) p1 ∈ l12, l12, p2 ∈ l12, l23(cid:9). FA23 = This is nothing but the Demazure resolution of the Schubert correspondence in F × F associated to the reduced decomposition (23)(12)(23) in W = S3. It has been used classically, see [45], as a tool to analyze Schubert's variety bT of complete triangles. By construction, we have a diagram i Y(3) / F(3) × g $❍❍❍❍❍❍❍❍❍ ρ π F(3) l23 l13 p1 • p2 • l12 Figure 6: The par- tial triangle code for the 3-ray vari- ety. where i is a closed embedding and π is the projec- tion. Each fiber of ρ is a Lie subalgebra in g. Inside F(3) we have the locus of degenerate partial triangles F = (cid:8)(p1, p2, l12, l13, l23) ∈ F(3)(cid:12)(cid:12) p1 = p2, l12 = l12 = l23(cid:9) identified with the flag variety SL3/B. Proposition 7.14. (a) The variety F(3) is smooth, of dimension 6. It can be represented as a P1 × P1-bundle over FA12. F(3)) is birational. (b) The projection T → F(3) (and therefore the composite projection bT → (c) Let q = (p1, p2, l12, l13, l23) ∈ F(3). Then the Lie subalgebra ρ−1(q) ⊂ g In the latter case, it is is 2-dimensional, unless p1 = p2 and l13 = l23. 3-dimensional. (d) The 3-ray variety Y(3) has all components of dimension 8 (in fact, we will see that it is irreducible). Proof: (a) and (b) are obvious. To see (c), we denote by L01,12 and L12,23 the rank 3 bundles of Lie algebras over F01,12 and F12,23 respectively, whose 66 $ /   total spaces are the 2-ray varieties Y01,12 and Y12,23, see Proposition 7.10 (b). They are subbundles in the trivial bundles with fiber g, i.e., their fibers are Lie subalgebras in g. Then, by definition of the fiber product, ρ−1(q) = (L01,12)(p1,p2,l13,l23) ∩ (L12,23)(p1,p2,l12,l23) ⊂ g. If p1 = p2 and l13 = l23, then both terms in the intersection are equal. In any other case the two terms in the intersection are two distinct 3- dimensional subspaces contained in the 4-dimensional space (LA12)(p1,p2,l12), so the intersection is of dimension 2. (d) Note that the 2-ray variety Y01,12, being the blow-up of YA12 in a smooth subvariety of codimension 2, is, at least locally over YA12, a Cartier divisor in a P1 bundle over YA12. Let W be the pull-back of this P1 bundle from YA12 to Y12,23. Then the fiber product Y(3) is a Cartier divisor in it. Hence all components of Y(3) are of the same dimension 8. Corollary 7.15. The Cartesian square (7.13) is Tor-independent. Proof: Follows from Proposition 7.11, since Y01,12, Y12,23 and Y12 are smooth of dimension 8, and Y(3) has all components of dimension 8 by Proposition 7.14(d). F. 3-ray variety as a fiber product of blowups. We now recall that the maps σ1 and σ2 in (7.13) realize the 2-ray varieties as blowups of the 1-ray variety YA12. More precisely, denote for short F(1) = FA12 = F ×P2 F = (cid:8)(p1, p2, l12) ∈ P2 × P2 × P2(cid:12)(cid:12) p1 ∈ l12, p2 ∈ l12(cid:9) the central fiber of YA12. We further denote by L = LA12 the rank 4 bundle of Lie algebras on F(1) whose total space is YA12, see Proposition 7.9(b). We write simply YA12 = L. Then, by Proposition 7.10 (c), Y01,12 = BlE1(L), Y12,23 = BlE2(L), where E1, E2 ⊂ L are two rank 2 subbundles of Lie algebras in L defined as follows. As before we identify elements ξ of g = sl3 with global vector fields on P2. Then the fibers of Ei, i = 1, 2, are defined by: (Ei)(p1,p2,l12) = (cid:8)ξ ∈ L(p1,p2,l12) ⊂ g(cid:12)(cid:12) the linear part of ξ at pi is scalar(cid:9). 67 Proposition 7.16. Let q = (p1, p2, l12) ∈ F(1). (a) If p1 6= p2, then the fibers (E1)q, (E2)q ⊂ Lq are transversal, i.e., intersect only at 0. (b) If p1 = p2, then the fibers (E1)q and (E2)q are equal. Proof: (a) Suppose p1 6= p2 and ξ ∈ (E1)q ∩ (E2)q. We need to prove that ξ = 0. Consider the dual plane P2, with the lines Pi corresponding to the points pi and let η be the vector field on P2 corresponding to ξ. It suffices to prove that η = 0. The fact that ξ preserves pi and has scalar linear part at pi, means that η preserves each point of Pi, i.e., vanishes on Pi identically. Now, for any global vector field ζ on any Pn, any connected component of the zero locus of ζ is a projective subspace (since it corresponds to the eigenspace of the corresponding matrix). So the fact that η vanishes on P1 ∪ P2 means that η = 0. Part (b) is obvious. It follows from this proposition that the intersection of the two blow-up centers E1 ∩ E2 has two irreducible components: one of codimension 4, the zero section of L, and the other one of codimension 3, which is the locus of all points in the fibers in the part (b) of the above proposition. Proposition 7.17. The variety Y(3) is irrducible and has rational singular- ities. That is, for any proper birational morphism k : M → Y(3) with M smooth, we have Rk∗(OM ) ≃ OY(3). In fact, the singular locus Q of Y(3) is a smooth 5-dimensional variety and the singularity in the transversal direction to it is a 3-dimensional quadratic cone. The variety Q is identified via the map Y(3) → YA12 with the blow up in the zero section of the total space of the rank 2 vector bundle E1F = E2F over the flag variety F ⊂ F(1). Proof: The map ζ1 : Y(3) → Y12,23 is an isomorphism outside the preimage of the intersection of the two blow-up centers LA12 ∩ LA23. The later has one component of codimension 3 and another one of codimension 4, as it was mentioned above. Then the preimage of the first component has an irreducible component of minimal possible codimension 2 in Y12,23. Since the dimension of fibers for ζ1 : Y(3) → Y12,23 has maximal possible dimension 1, the irreducibility of Y(3) follows if there is no component of Y(3) over this component of codimension 2. This follows from local calculations below. The variety F(1), the bunlde L on it and the subbundles E1 and E2 are all acted upon by the group SL3. The action on F(1) has two orbits: the 68 diagonal F ⊂ F ×P2 F , a codimension 1 submanifold, and the complement to it. So to unerstand the singularities of the fiber product of the two blowups it is enough to take a 1 dimensional transversal slice T to F , consider the restriction L′ = LT , its blowups along the E′ i = EiT and look at their fiber product. The following local model describes this situation: • T = A1 t := Spec k[t], the affine line with coordinate t (we will use similar notation below). • L′ = A5 x1,y1,x2,y2,t is the trivial rank 4 bundle over T with coordinates in the fiber being x1, y1, x2, y2. • E′ 1, E′ 2 are two rank 2 subbundles which are transversal outside t = 0 and coincide for t = 0, given explicitly by E′ E′ 1 = (cid:8)(x1, y1, x2, y2, t)(cid:12)(cid:12) x2 = 0, y2 = 0(cid:9), 2 = (cid:8)(x1, y1, x2, y2, t)(cid:12)(cid:12) x2 = tx1, y2 = ty1(cid:9). (L′) explicitly in local charts. We recall We now write the blowups BlE′ general formulas. i Let S be a smooth algebraic variety and x, y ∈ k[S] be two regular func- tions defining smooth divisors meeting transversally. Then Blx=y=0(S) is the union of two charts, one given in S × A1 z by the equation y = xz, the other z′ by the equation x = yz′. given in S × A1 Accodingly, BlE′ (L′) has a local chart in A6 x1,y1,x2,y2,t,z1 given by y2 = x2z1 2(L′) (and another chart with the roles of x2, y2 interchanged). Similarly, BlE′ has a local chart in A6 x1,y1,x2,y2,t,z2 given by (y2 − ty1) = (x2 − tx1)z2 (and another chart with the roles of x's and y's interchanged). So the fiber product 2(L′) has 4 charts, of which we write one, it is given by two BlE′ equations 1(L′) ×L′ BlE′ 1 y2 = x2 · z1, (y2 − ty1) = (x2 − tx1) · z2 in the affine space A7 being similar. x1,x2,y1,y2,t,z1,z2. We analyze this chart, the other charts Eliminating y2 from the first equation, we represent this chart as a hy- persurface Σ ⊂ A6 x1,x2,y1,t,z1,z2 given by the equation t(y1 − x1z2) = x2(z1 − z2). 69 In other words, Σ = h−1(Q), where Q ⊂ A4 cone {ad = bc} and h : A6 → A4 is the regular map given by a,b,c,d is the 3-dimensional quadratic a = t, b = x2, c = z1 − z2, d = y1 − x1z2. We verify at once that h is smooth in any point lying over Q. Therefore Σsing, the singular locus of Σ, is a smooth subvariety given (in A6) by the equations a = b = c = d = 0 and near any point of Σsing the variety Σ locally looks like the product of Σsing and Q. So it has rational singularities. By the the similar analysis of the other charts of the fiber product, we get the structure of singularities of the blow-up. The locus of singularities is a smooth surface which is mapped birationally on the plane (t = 0, x2 = 0, y2 = 0) and is in fact the blow-up of this plane at the origin. The singularities of the fiber product in the transversal direction of every point of this surface are three dimensional quadratic cones. Since the plane (t = 0, x2 = 0, y2 = 0) corresponds exactly to the fiber of E1 over a point in the flag variety F ⊂ F(1), the description of the variety Q follows. 70 8 Properties of the sl3-flober A. Definition and main result. We recall the flop diagram (YC, lCC′)C≤C′ for sl3, see Definition 7.7. Definition 8.1. The sl3-flober is the diagram consisting of the coherent de- rived categories EC := Db coh(YC) of the varieties from the sl3-flop diagram (Definition 7.7) and the functors γCC′ = R(lCC′)∗ : EC −→ EC′, δC′C = L(lCC′)∗ : EC′ −→ EC, C ≤ C ′. We denote this diagram by F = Fsl3. The rough size of the categories EC can be understood from the following. Proposition 8.2. Let EC = K(EC) be the rational Grothendieck group of EC (or, what is the same, of YC). Then: dimQ(EC) = 6 12 72 if dim(C) = 2; if dim(C) = 1; if dim(C) = 0. Proof: As each YC is the total space of a vector bundle over its central fiber FC, we have K(YC) ≃ K(FC). Now, for dim(C) = 2 we have FC = F is the flag variety for sl3, so its K-group has rank W = 6. For dim(C) = 1 we have FC = F ×P2 F is the total space of a P1-bundle over F , so its K-group complete triangles. From its explicit representation as an iterated blowup has rank 12. Finally, for C = 0 we have FC = bT is the Schubert space of of (P2)3 recalled in [45], Ex. 7.7.3, it follows that K0(bT ) is free and has the same rank as the total Chow group A•(bT ). The structure of the latter group is also known, see [45] Th. 3.2 and references therein, as well as [20]. It is free of rank 72. The rest of the paper is devoted to the proof of the following. Theorem 8.3. The diagram of categories (EC, γCC′, δC′C) satisfies all the conditions of Definition 3.6 of H-schobers except, possibly, Condition (4) (invertibility) for the case C ′ = −C, dim(C) = 1, D = 0. Studying Condition (4) in te remaining case, i.e., investigating the effect l0,C−→ YC, (see Definition 7.7(b)) on of the Schubert transform Y−C derived categories, is an interesting question which we plan to address in a future work. l0,−C←− Y0 71 B. Idempotency and invertibility. We now start the proof of Theorem 8.3. The fact that the γCC′ form a 2-functor from C (the poset of faces) to triangulated categories, is a consequence of the fact that the lCC′ form a functor from C to algebraic varieties. Let us prove Condition (1) (idem- potency). Each lCC′ : YC → YC′ is a regular birational, proper morphism between smooth varieties. Therefore we have R(lCC′)∗(OYC ) = OYC′ . So by the projection formula, for each F ∈ EC′ = Db coh(YC′) we have a natural identification γCC′δC′C(F ) ≃ R(lCC′)∗(lCC′)∗F ≃ R(lCC′)∗(OYC ) ⊗ F = F . Next, let us look at Condition (4) (invertibility) of Definition 3.6). Apart from Conjecture ??, the only case we need to consider is dim(C) = dim(C ′) = 2, dim(D) = 1. In this case the diagram YC ← YD → YC′ is, by Corollary 6.5 locally a product of a smooth manifold and of the diagram (1.4) for the Atiyah flop (case g = sl2). So the fact that the corresponding flop functor is an equivalence, follows from the case g = sl2. C. Collinear transitivity. We now turn to verifying Condition (4) (collinear transitivity) of Definition 3.6. That is, we define the flopping functor ϕCC′ : EC → EC′ as in (3.4) and verify that ϕBC ◦ ϕAB = ϕAC whenever A, B, C are collinear faces of H. For this we distinguish several cases as to the pos- sible dimensions of A, B, C. We will call Case (i, j, k) the situation when dim(A) = i, dim(B) = j, dim(C) = k. We only need to consider nonzero i, j, k. Each such case may have several subcases as to the relative positions of A, B, C with given dimensions. D. Case (2,2,2). Our arguments in this case are parallel to those of [44] which establish the braid group relations between the flop functors acting on the EC for C running over chambers. In fact, one has a description of the fundamental groupoid of the open stratum h∗ C − HC in terms of morphisms ϕCC′ for dim(C) = dim(C ′) = 2 satisfying collinear transitivity, see [35], Appendix, so Case (2,2,2) also proves these braid relations. We consider several possibilities. First, we consider the neighboring case. That is, A, B, C are positioned as the faces A1, A2, A3 in Fig. 5 (in one of the two possible orientations). By definition, ϕAB and ϕBC are defined by pullback and pushforward in the diagrams YA ←− Yα −→ YB, YB ←− Yβ −→ YC, 72 where α and β are the rays between A and B and between B and C, re- spectively. By Proposition 7.10(d), the composition ϕBCϕAB is defined by pullback and pushforward in the diagram pA←− Yα ×YB Yβ pC−→ YC YA (the middle term Yα ×YB Yβ is the 2-ray variety Y(2) studied in §7D and the square defining the fiber product is Tor-independent). On the other hand, the functor ϕAC is defined by pullback and pushfor- ward in the diagram YA qA←− Y0 qC−→ YC. The identification of ϕBCϕAB and ϕAC is implied by Proposition 7.10 (which says that Y(2) = Yα ×YB Yβ is smooth) which we can combine with the com- mutative diagram qA yttttttttttt pA YA Yα ×YB Yβ / YC Y0 σ qC %❏❏❏❏❏❏❏❏❏❏❏ pC Indeed, smoothness implies that Rσ∗(O) = O and so Rσ∗◦σ∗ = Id. Therefore ϕBCϕAB = RpC∗p∗ C = RpC∗Rσ∗σ∗p∗ A = RqC∗q∗ C = ϕAC. Next, we consider the non-neighboring cases. That is, A, B, C are posi- tioned as A0, A2, A3 or as A0, A1, A3 in Fig. 5 (in one of the two possible orientations). For definiteness, let us consider the first of these possibili- ties. Then by the neighboring case above, ϕA0,A2 is given by pullback and pushforward through the 2-ray variety, so ϕA2,A3ϕA0,A2 is, by Corollary 7.15 (Tor-independence of the Cartesian square) given by pullback and pushfor- ward through the 3-ray variety Y(3). Our statement then follows, similarly to the neighboring case above, from Proposition 7.17 applied to the morphism Y0 → Y(3). E. Case (2, 1, 2) (neighboring). Assume that A, B, C are positioned as A1, A12, A2 in Fig. 5 (in one of the two possible orientations). In this case the transitivity follows formally from idempotency. Indeed, by definition, ϕAC = ϕ0CϕA0, while substituting Id = ϕ0BϕB0 and re-arranging the brackets, we get (8.4) ϕBCϕAB = ϕBC(cid:0)ϕ0BϕB0(cid:1)ϕAB = (cid:0)ϕBCϕ0B(cid:1)(cid:0)ϕB0ϕAB(cid:1) = ϕ0CϕA0. 73 y   % o o / F. Case (1, 2, 1) (neighboring). Assume that A, B, C are positioned as A01, A1, A12 in Fig. 5 (in one of the two possible orientations). Thus YB is isomorphic to the Grothendieck resolution. We can assume that YB =eg is the standard Grothendieck resolution, i.e., B is the dominant Weyl chamber. Now, the functor ϕAC = ϕ0CϕA0 is defined by the pullback and push forward through Y0. This is the same as pullback and pushforward through the 2-ray variety YA ×YB YC, as follows from a diagram similar to (8.4): ysssssssssss Y0 σ %❑❑❑❑❑❑❑❑❑❑❑ YA ×YB YC s / YC. p YA On the other hand, the functor ϕBCϕAB is defined by pushforward and pull- back through YB =eg, i.e., as r∗Rq∗ in the square YA ×YB YC / YC p s YA r / YB q So if we show that the square is Tor-independent, we will identify r∗Rq∗ with Rp∗s∗ = ϕAC. By Corollary 6.5, the varieties YA and YC are the blowups of YB along two codimension 2 subvarieties ΛAB, ΛCB which are transversal. More precisely, YB =eg is the total space of the tautological bundle b → F of Borels over F , and ΛAB, ΛCB are total spaces of two transversal SL3-equivariant subbundles whose fibers over the standard base point of F = G/B are two subalgebras in the standard Borel b of the form b ∩ si(b), where si, i = 1, 2, are two simple reflections in the Weyl group. Now, it is standard (and easily follows from Proposition 7.11) that the fiber square of two transversal blowups is Tor-independent. G. Case (1, 1, 1) (neighboring). Assume that A, B, C are positioned as A01, A12, A23 in Fig. 5 (in one of the two possible orientations). The identity ϕAC = ϕBCϕAB in this case follows from the Tor-independence of the square defining the 3-ray variety Y(3) (Corollary 7.15) and from the fact that Y(3) has rational singularities (Proposition 7.17). 74 y   % o o /   /   / H. Case A ≥ B or B ≤ C. If A ≤ B, then for any C we have ϕBCϕAB = ϕ0CϕB0ϕAB A≥B≥0 = ϕ0CϕA0 = ϕAC. We have a similar statement and argument when B ≥ C. I. Reduction to neighboring cases. We now give an inductive argument which reduces all the remaining cases to the neighboring cases that have been already considered. For any two faces A, B of H of dimensions 6= 0 we denote by d(A, B) the incidence distance between A and B, i.e., the minimal m such that there exists a chain of faces of dimensions 6= 0 D0 = A, D1, · · · , Dm = B such that for any i we have Di < Di+1 or Di > Di+1. For example, if A and B are two adjacent chambers, then d(A, B) = 2. Suppose we know transitivity ϕAC = ϕBCϕAB for all collinear A, B, C with d(A, C) < L. Let us prove it in the case d(A, C) = L. Assume first that dim(A) = 2. In this case we take the 1-dimensional face D < A in the direction of B, C. We can then write by inductive assumption ϕBCϕAB ind.= (ϕBCϕDB)ϕAD = ϕBC(ϕDBϕAD) ind.= ind.= ϕDCϕAD (H) = ϕAC, where the last equality is an instance of Case H above. A similar argument works when dim(C) = 2. So our statement reduces to the case dim(A) = dim(C) = 1 which we now assume. Now, if dim(B) = 2, then we take a 1-dimensional face D ≤ B in the direction of A or C (there must be a gap in one of the directions, otherwise we are in a neighboring case). Suppose D is in the direction of A. Then, by geometry of our arrangement (3 lines only), the faces D, B, C must be neighboring, i.e., fit into Case F and the faces A, D, C fit into Case G above, so we write ϕBCϕAB ind.= ϕBC(ϕDBϕAD) = (ϕBCϕDB)ϕAD (F ) = (F ) = ϕDCϕAD (G) = ϕAC. This finishes the proof of Theorem 8.3. 75 References [1] R. Anno Spherical functors, arXiv:0711.4409. [2] R. Anno, T. Logvinenko. Spherical DG-functors, arXiv:1309.5035. [3] E. Babson, P. E. Gunnells, R. Scott. A smooth space of tetrahedra. Adv. Math. 165 (2002) 285–312. [4] E. Babson, P. E. Gunnells, R. Scott. Geometry of the tetrahedron space. Adv. Math. 204 (2006) 176–203. [5] A. Beilinson. How to glue perverse sheaves. In: K-theory, arithmetic and geometry (Moscow, 1984), Lecture Notes in Math. 1289, Springer- Verlag, 1987, 42 - 51. [6] A. Beilinson, I. Bernstein, P. Deligne. Faisceaux Pervers, Ast´erisque 100, 1982. [7] R. Bezrukavnikov. Noncommutative counterparts of the Springer reso- lution. arXiv math/0604445. [8] R. Bezrukavnikov. Commutative and noncommutative symplectic res- olutions and perverse sheaves. Lecture May 18 2015. [9] R. Bezrukavnikov, S. Riche. Affine braid group actions on derived categories of Springer resolutions. Ann. Sci. ´Ec. Norm. Sup´er. 45 (2012), 535-599. [10] A. Bodzenta, A. Bondal. Flops and spherical functors. arXiv:1511.00665. [11] A. Bodzenta, A. Bondal. Canonical tilting relative generators, Advances in Mathematics, 323 (2018) 226-278. [12] A. I. Bondal, M. M. Kapranov. Representable functors, Serre functors and mutations. Math. USSR Izv. 35 (1990) 519–541. [13] A. I. Bondal, M. M. Kapranov. Enhanced triangulated categories. Math. USSR Sbornik, 70 (1991) 93-107. [14] A. Bondal, D. Orlov. Semiorthogonal decomposition for algebraic va- rieties. arXiv alg-geom/9506012. 76 [15] A. I. Bondal, D. O. Orlov. Derived categories of coherent sheaves. Pro- ceedings of the International Congress of Mathematicians, Vol. II (Bei- jing, 2002), (2002) 47-56, Beijing, Higher Ed. Press. [16] A. I. Bondal, M. van den Bergh. Generators and representability of functors in commutative and noncommutative geometry. Mosc. Math. J. 3 (2003) 1-36. [17] T. Bridgeland. Flops and derived categories. Invent. Math. 146 (2002) 613-632. [18] E. Brieskorn. Die Auflosung der rationalen Singulariaten holomorpher Abbildungen, Math. Ann. 178 (1968), 255–270. [19] E. Brieskorn. Singular elements of semi-simple algebraic groups. Actes Congr´es Intern. Math. Nice, 1970, Tome 2, 279-284. [20] A. Collino, W. Fulton. Intersection rings of spaces of triangles. M´em. Soc. Math. France, 38 (1989) 75-117. [21] N. Chriss, V. Ginzburg. Representation Theory and Complex Geome- try. Birhkauser, Boston, 2010. [22] W. Donovan. Perverse schobers and wall crossing. arXiv:1703.00592. [23] W. Donovan, M. Wemyss. Twists and braids for general 3-fold flops. arXiv:1504.05320. [24] V. Drinfeld. DG quotients of DG categories. J. Algebra 272 (2004) 643-691. [25] W. G. Dwyer, P. S. Hirschhorn, D. M. Kan, J. H. Smith. Homotopy Limit Functors on Model Categories and Homotopical Categories. AMS Publ. 2004. [26] T. Dyckerhoff, M. Kapranov. Triangulated surfaces in triangulated cat- egories. arXiv:1306.2545. [27] T. Dyckerhoff, M. Kapranov, V. Schechtman, Y. Soibelman. Fukaya categoris with coefficients, in preparation. 77 [28] A. Galligo, M. Granger, P. Maisonobe. D-modules et faisceaux pervers dont le support singulier est un croisement normal. Ann. Inst. Fourier Grenoble, 35 (1985), 1-48. [29] A. Harder, L. Katzarkov. Perverse sheaves of categories and some ap- plications. arXiv 1708.01181. [30] P. S. Hirschhorn. Model Categories and Their Localizations. AMS Publ. 2003. [31] R. Hotta, M. Kashiwara. The invariant holonomic system on a semisim- ple Lie algebra. Invent. Math. 75 (1984) 327-358. [32] A. Iliev, L. Manivel. Severi varieties and their varieties of reductions. J. Reine Angew. Math. 585 (2005), 93–139. [33] A. Iliev, L. Manivel. Varieties of reductions for gln. arXiv:math/0501329. [34] W. van der Kallen, P. Magyar. The space of triangles, vanishing theo- rems, and combinatorics. J. Algebra, 222 (1999) 17-50. [35] M. Kapranov, V. Schechtman, Perverse sheaves over real hyperplane arrangements, Ann. Math. 183 (2016) 617-679. [36] M. Kapranov, V. Schechtman, Perverse Schobers, arXiv:1411.2772. [37] Y. Kawamata. On the cone of divisors of Calabi-Yau fiber spaces. Int. J. Math. 8 (1997), 665-687. [38] B. Kostant. Lie group representations on polynomial rings. Bull. Amer. Math. Soc. 69 (1963) 518-526. [39] A. Kuznetsov. Hyperplane sections and derived categories Russ. Math. Izv. 70 (2006), 447–547. [40] A. Kuznetsov. Homological projective duality. Publ. Math. Inst. Hautes ´Etudes Sci. 105 (2007), 157–220. [41] Y. Namikawa. Mukai flops and derived categories. II. In: "Alge- braic structures and moduli spaces", CRM Proc. Lecture Notes 38, p.149–175, AMS Publ. 2004. 78 [42] H. Pinkham. Factorization of birational maps in dimension 3, in: Sin- gularities (P. Orlik, ed.), Proc. Symp. Pure Math. 40 pt. 2, p. 343-371, American Math. Soc., Providence, 1983. [43] M. Reid. Young person's guide to canonical singularities, in: "Algebraic geometry" (Bowdoin, 1985), Proc. Sympos. Pure Math. 46 Part 1, AMS 1987, p. 345–414. [44] S. Riche. Geometric braid group actions on derived categories of coher- ent sheaves (with a joint appendix with R. Bezrukavnikov). Represen- tation Theory 12 (2008) 131-169. [45] J. Roberts. Old and new results about the triangle varieties. Lecture Notes in Math. 1311, p. 197-219, Springer, Berlin, 1988. [46] H. Schubert. Anzahlgeometrische Behandlung des Dreiecks. Math. Ann. 17 (1880) 1213-1255. [47] J. G. Semple. The triangle as a geometric variable. Mathematika 1 (1954) 80-88. [48] G. Tabuada. Th´eorie homotopique des DG-categories. arXiv:0710.4303. [49] G. Tabuada. Homotopy theory of dg categories via localizing pairs and Drinfeld's dg quotient. Homology, Homotopy and Appl. 12 (2010) 187- 219. [50] B. Toen. The homotopy theory of dg-categories and derived Morita theory. Invent. Math. 167 (2007) 615–667. [51] M. van den Bergh. Three-dimensional flops and noncommutative rings. Duke Math. J. 122 (2004) 423–455. A.B.: Kavli Institute for Physics and Mathematics of the Universe (WPI), 5-1-5 Kashi- wanoha, Kashiwa-shi, Chiba, 277-8583, Japan, Steklov Institute of Mathematics, Moscow, Russia, National Research University Higher School of Economics, Russian Federation Email: [email protected] M.K.: Kavli Institute for Physics and Mathematics of the Universe (WPI), 5-1-5 Kashi- wanoha, Kashiwa-shi, Chiba, 277-8583, Japan. Email: [email protected] V.S.: Universit´e Paul Sabatier, Institut de Math´ematiques de Toulouse, 118 Route de Narbonne, 31062 Toulouse, France. Email: [email protected] 79
1106.3633
1
1106
2011-06-18T10:20:58
Pentagramma Mirificum and elliptic functions
[ "math.AG" ]
We give an exposition of fragments from Gauss [G] where he discovered, with the help of some work of Jacobi, a remarkable connection between Napier pentagons on the sphere and Poncelet pentagons on the plane. As a corollary we find a parametrization in elliptic functions of the classical dilogarithm five-term relation.
math.AG
math
PENTAGRAMMA MIRIFICUM AND ELLIPTIC FUNCTIONS (Napier, Gauss, Poncelet, Jacobi, ...) Vadim Schechtman FAUST. Das Pentagramma macht dir Pein? Ei sage mir, du Sohn der Hole, Wenn das dich bannt, wie kamst du denn herein? Introduction In this note we give an exposition of fragments from Gauss [G] where he dis- covered, with the help of some work of Jacobi, a remarkable connection between Napier pentagons on the sphere and Poncelet pentagons on the plane. This gives rise to a parametrization of the variety of Napier pentagons using the division by 5 of elliptic functions. As a corollary we will find the classical five-term relation for the dilogarithm in a somewhat exotic disguise, cf. 6.4. The author gave talks on these subjects in Toulouse on January, 2011 and in Moscow on May, 2011. §1. Napier’s rules John Napier of Merchiston (1550 - 1617) was a Scottish mathematician and astrologer. His main work is Mirifici Logarithmorum Canonis Descriptio (1614). The reader will find in [C] some historical remarks on what follows. 1.1. For a spherical triangle with sides a, b, c and angles α, β, γ, cos c = cos a cos b + sin a sin b cos γ 1 2 Consider a right-angled spherical triangle with sides a, b, c and angles α, β, γ = π/2. Let us call its parts five quantities τ = (a, b, π/2 − α, π/2 − c, π/2 − β) We consider them as positioned on a circle, i.e. ordered in the cyclic order. If we turn them we get the parts of another right-angled triangle: τ ′ = n(τ ) = (a′, b′, . . .) = (b, π/2 − α, π/2 − c, π/2 − β, a) (Napierian rotation). Obviously n5 = 1. Repeat this once more: g(τ ) := n2(τ ) = (a′′, b′′, . . .) = (π/2 − α, π/2 − c, π/2 − β, a, b) In other words, (α′′, b′′, c′′) = (β, π/2 − c, π/2 − a), this triangle is obtained by a reflection of the first one in the vertex β; call it Gaussian reflection. 1.2. Napier rules: I. sine(middle part) = product of tangents of adjacent parts: sin b = tan a cot α II. sine(middle part) = product of cosines of opposite parts: sin a = sin α sin c §2. Pentagramma Gaussianum 2.1. Let us draw a spherical right-angled triangle which we denote P3Q1P4 with the angles ∠Q1P3P4 = p3, ∠P3Q1P4 = π/2, ∠Q1P4P3 = p4 and sides P3Q1 = π/2 − p5, Q1P3 = π/2 − p2, P3P4 = p1. Let us use an abbreviated notation p′ = π/2 − p. So, the Napierian parts of this triangle are τ1 = (p′ 2, p′ 5, p′ 3, p′ 1, p′ 4) Applying the Gaussian reflexion we get the triangles i+2, p′ τi := gi−1(τ1) = (p′ i+4, p′ i+1, p′ i+5, p′ i+3) where we treat the indices modulo 5. We denote the i-th triangle Pi+2QiPi+3, where ∠Pi+2QiPi+3 = π/2, pi = Pi+2Pi+3. So we get a spherical pentagon P1P2P3P4P5; its characteristic property is PiPi+2 = π/2, cf. the picture on p. 481 of [G]. 3 Set Gauss’ notation: Napier rules give: αi = tan2 pi (α, β, γ, δ, ǫ) = (α1, . . . , α5) cos pi = sin pi+1 cos pi−1 1 = cos pi tan pi+2 tan pi+3 It follows: Relations: (γδ, δǫ, ǫα, αβ, βγ) = (sec2 p1, sec2 p2, sec2 p3, sec2 p4, sec2 p5) Out of two quantities one can build up the remaining three, for example: 1 + α = γδ, 1 + β = δǫ, etc. (2.2.1) β = γ = 1 + α + γ αγ , δ = 1 + α γ , ǫ = 1 + γ α 1 + α αβ − 1 , δ = αβ − 1, ǫ = 1 + β αβ − 1 , , etc. and also etc. One has 3 + α + β + γ + δ + ǫ = αβγδǫ =p(1 + α)(1 + β)(1 + γ)(1 + δ)(1 + ǫ) 2.2. Gauss’ favorite example: (α, β, γ, δ, ǫ) = (9, 2/3, 2, 5, 1/3), αβγδǫ = 20 2.3. Regular pentagram. If α = β = . . . then whence α5 = Also α2 = sec2 i, so α = 11 + 5√5 2 α + 1 = α2, 1 + √5 2 = 11.0901699... α 1 + α = 1 + √5 2 cos pi = α−1 = Note that We have α−1 = α 1 + α = −1 + √5 2 = 0.618033988... sin 2π/5 = 2s · c 4 where we have denoted s = sin π/5, c = cos π/5. On the other hand, sin 2π/5 = sin 3π/5 = s(3 − 4s2) = s(4c2 − 1), cos π/5 = 1 + √5 4 = α 2 2 cos pi · cos π/5 = 1 c′ := cos 2π/5 = −1 + √5 4 cc′ = 1/4, c − c′ = 1/2 whence 2c = 4c2 − 1, so It follows that We set Then A cone 2.5. Let M be the centrum of the sphere. The vertices Pi all lie on a quadratic cone whith vertex M. Namely, we have So MP3 is orthogonal to MP1 and to MP5. ∠Pk−1MPk+1 = π/2 We define the coordinates in such a way that M lies ar the origin, Then and P3 = (1, 0, 0), P1 = (0, 1, 0) P5 = (0, cos p3, sin p3) P4 = (cos p1, 0, sin p1) Finally, the ray MP2 ⊥ P4MP5, so The coordinates of Pk satisfy the equation P2 = (cos p5, cos p4,− cos p3 sin p5) or Note that cos p2 · (cos p1 · z − sin p1 · x)(cos p3 · z − sin p3 · y) = xy, z2 − √αxz − √γyz − 1 + α + γ √αγ xy := = z2 + pxz + qyz + rxy = 0 (C) r = −β√αγ Reduction to principal axes. 5 2.6. In general, given a quadratic form defined by a symmetric matrix A, v = (x, y, z)∗ 7→ vtAv, if we want to find an orthogonal matrix B such that for v = Bv ′, v ′ = (x′, y ′, z′)∗, v ∗Av = v ′∗B ∗ABv ′ = v ′∗A′v ′ such that A′ = diag(G, G′, G′′) then A′ = B ∗AB = B−1AB and the numbers G, G′, G′′ are the eigenvalues of A. The columns of B are eigenvectors of A. 2.7. In our case, for the quadrics (C), the matrix is A =  r/2 p/2 0 0 q/2 r/2 p/2 q/2 1   The characteristic polynomial is det(t · I − A) = t3 − t2 − p2 + q2 + r2 4 r(pq − r) 4 , − so the characteristic equation takes the form where This is our main equation. t(2t − 1)2 = ω(t − 1) ω := αβγδǫ (E) 2.8. Let us investigate the real roots of (E). We suppose that ω > 0. The straight line allways intersects the cubic parabola ℓ : u = ω(t − 1) at one negative point t = G < 0. P : u = t(2t − 1)2 (2.8.1) (2.8.2) On the other hand, ℓ intersects P at two points t = G′, G′′ > 1 iff ω is greater than some critical value ω0; if ω = ω0 then ℓ touches P at t = G′ = G′′, if ω < ω0 then there are no points of intersection of ℓ with P other than t = G. The critical value is ω0 = α5 0 where α0 = 1 + √5 2 , cf. 2.3. In that case G = −α0, G′ = G′′ = α2 c0 = cos π/5 0/2 = −c0G, 6 Thus, if ω ≥ α5 G′′ which coincide if ω = α5 0. 0 then (E) has one negative root, G, and two positive roots, G′, One has: GG′G′′ = −ω/4 (G − 1)(G′ − 1)(G′′ − 1) = −1/4 (2G − 1)(2G′ − 1)(2G′′ − 1) = −ω 2.9. Example. For αβγδǫ = 20, G = −2.197, G′ = 1.069, G′′ = 2.128. §3. Gauss’ coordinates 3.1. Consider an ellipse E : x2 a2 + x2 b2 = 1, a > b. Inscribe E into the circle C with the centrum O and radius a. For a point P = (x, y) ∈ E, let P ′ = (x, y ′) ∈ C. The eccentric anomaly (anomalie excentrique) of P is the angle φ = ∠XOP ′ where X = (a, 0). Then x = a cos φ, y = b sin φ 3.2. Return to the Napier pentagon P1 . . . P5. Let us draw a plane tangent to the sphere at the point of intersection with the axe of the cone and take the central projection of our pentagon on this plane. Let Ri be the projection of Pi to this plane. The points Ri, i = 1, . . . , 5, lie on an ellipse with semi-axes p−G/G′ and p−G/G′′. We put the coordinate axes x, y of the plane along the axes of the ellipse. Let Ri have the coordinates (xi, yi). Let φi be the eccentric anomaly of Ri. We have (3.2.1) xi = g ′ cos φi, yi = g ′′ sin φi where Let M be the centrum of the sphere, with coordinates (x, y, z) = (0, 0, 1). g ′ =p−G/G′, g ′′ =p−G/G′′ Attention: these coordinates differ from 2.5. Set ψi = ∠PiMPi+1 = ∠RiMRi+1; ri = (xi, yi, 1). Then and cos ai = (ri, ri+1) riri+1 = p(x2 αi := tan2 ψi = sec2 ψi − 1 = xixi+1 + yiyi+1 + 1 i+1 + y2 i + 1)(x2 i + y2 i+1 + 1) = (xi − xi+1)2 + (yi − yi+1)2 + (xiyi+1 − yixi+1)2 (xixi+1 + yiyi+1 + 1)2 = ri × ri+12 (ri, ri+1)2 For the future use, note the quantities 3.3. We have whence βi := sin2 ψi = αi αi + 1 = ri × ri+12 ri2ri+12 ∠Ri−1MRi+1 = π/2 xi−1xi+1 + yi−1yi+1 + 1 = 0 Solving (3.3.1)i−1 and (3.3.1)i+1 for xi, yi, we get xi = yi+2 − yi−2 xi+2yi−2 − yi+2xi−2 , yi = xi−2 − xi+2 xi+2yi−2 − yi+2xi−2 3.4. Next, xixi+1 2G′ − 1 + yiyi+1 2G′′ − 1 + 1 2G − 1 = 0 This will be proven in 3.6. It follows: 2G′ − 1 2G − 1 · 2G′′ − 1 2G − 1 · xi = − yi = , yi+1 − yi−1 xi−1yi+1 − xi+1yi−1 xi+1 − xi−1 xi−1yi+1 − xi+1yi−1 3.5. Theorem (Gauss, April 20, 1843). (a) sin((φi−2 + φi+2)/2) cos((φi−2 − φi+2)/2) = G G′′ sin φi; cos((φi−2 + φi+2)/2) cos((φi−2 − φi+2)/2) = G G′ cos φi (b) 7 (3.2.2) (3.2.3) (3.3.1)i (3.3.2) (3.4.1) (3.4.2) (3.5.1) sin((φi−1 + φi+1)/2) cos((φi−1 − φi+1)/2) cos((φi−1 + φi+1)/2) cos((φi−1 − φi+1)/2) =s G(G − 1) G′′(G′′ − 1) =s G(G − 1) G′(G′ − 1) sin φi = G(2G − 1) G′′(2G′′ − 1) sin φi cos φi = G(2G − 1) G′(2G′ − 1) cos φi (3.5.2) 3.6. Proof, cf. [F]. (a) follows from (3.3.2) and (3.2.1), taking into account the formulas sin a − sin b = 2 cos((a + b)/2) sin((a − b)/2 cos a − cos b = −2 sin((a + b)/2) sin((a − b)/2 (3.6.1) Proof of (3.4.1). 8 It follows from (3.5.1) (replacing i + 2 by i) that G′2 cos2((φi + φi+1)/2) + G′′2 sin2((φi + φi+1)/2) = G2 cos2((φi − φi+1)/2). Using cos2 a = 1 + cos 2a 2 , sin2 a = 1 − cos 2a 2 we get G2 cos(φi − φi+1) + (G′′2 − G′2) cos(φi + φi+1) = G′2 + G′′2 − G2 From (3.2.1) we get (G2 − G′2 + G′′2)G′xixi+1 + (G2 + G′2 − G′′2)G′′yiyi+1+ The numbers G, G′, G′′ are the roots of +(−G2 + G′2 + G′′2)G = 0 (3.6.2) (3.6.3) It follows that Whence t(2t − 1)2 = ω(t − 1) 1 + ω G2 + G′2 + G′′2 = 2 (−G2 + G′2 + G′′2)G =(cid:18)1 + ω 2 − 2G2(cid:19)G = ω 2 · 1 2G − 1 Similarly we compute the coefficients at xixi+1 and at yiyi+1 of (3.6.2) and arrive at (3.4.1). Proof of (b). The second equalities follow from (3.6.3). The first ones follow from (3.4.2) in the same manner as one has deduced (a) from (3.3.2). (cid:3) §4. Poncelet’s problem and division of elliptic functions (Jacobi) In this Section we describe the work of Jacobi [J2]. For a modern treatment of it see [GH] and references therein. 4.1. One considers a circle with center C of radius R, inside it a smaller circle with center c of radius r, a will be the distance between their centra. The line cC intersects the bigger circle at a point P , so CP = R, cP = R + a. One takes a point A1 on the bigger circle, draws the tangent to the smaller one till the intersection with the bigger one at the point A2, and continues similarly. Denote 2φi = ∠AiCP Let us suppose for simplicity that C is inside the smaller circle. Let B be the point where A1A2 touches the smaller circle, B ′ ∈ A1A2 the base of the perpendicular from C, and D ∈ cB the base ofthe perpendicular from C. Then DB = CB ′ = R cos(φ2 − φ1) 9 (4.1.1)i (4.1.2) and so It follows that or cD = a cos(φ2 + φ1), r = cD + DB = R cos(φ2 − φ1) + a cos(φ2 + φ1) R cos(φi+1 − φi) + a cos(φi+1 + φi) = r (R + a) cos φi+1 cos φi + (R − a) sin φi+1 sin φi = r Substracting (4.1)i − (4.1)i−1 and using cos x − cos y sin y − sin x we get 2 (cid:19), = tan(cid:18)x + y R − a R + a tan((φi+1 + φi−1)/2)) = tan φi 4.2. Jacobi elliptic functions. Cf. [J1]. Fix a number 0 ≤ k < 1 and define the period K =Z π/2 0 dx p1 − k2 sin2 x =Z 1 0 dy p(1 − y2)(1 − k2y2) (4.2.1) The function ”amplitude” is defined by φ = am(u), u =Z φ 0 Thus am : dx [0, K] −→ [0, π/2] =Z sin φ 0 p1 − k2 sin2 x am(K) = π/2 dy p(1 − y2)(1 − k2y2) We extend am to a function R −→ R by am(−u) = − am(u), am(u + 2K) = am(u) + π. We set Modern notations: ∆ am u := d am u du =p1 − k2 sin2 am u sn u = sin am u, cn u = cos am u, dn u = ∆ am u 10 Thus sn is an odd function and cn and dn are even. 4.3. Addition formulas. sn(u + v) = cn(u + v) = sn u cn v dn v + cn u sn v dn u cn u cn v − sn u sn v dn u dn v 1 − k2 sn2 u sn2 v 1 − k2 sn2 u sn2 v dn u dn v − k2 sn u sn v cn u cn v 1 − k2 sn2 u sn2 v dn(u + v) = cf. [J1], §18. Main formula: , (4.3.1) Next, whence cn(u − v) = cn u cn v + sn u sn v dn(u − v) (4.3.2) sn(u + v) − sn(u − v) = cn(u + v) − cn(u − v) = − 2 cn u sn v dn u 1 − k2 sn2 u sn2 v 2 sn u sn v dn u dn v 1 − k2 sn2 u sn2 v = tan((am(x) + am(y))/2) = cn x − cn y sn y − sn x dn((x − y)/2) tan am((x + y)/2) , (4.3.3) 4.4. If φn = am(φ + nt) then tan((φ0 + φ2)/2) = ∆ am t tan φ1 So if then (4.1.2) is satisfied. We also have ∆ am t = R − a R + a cos φi cos φi+1 + sin φi sin φi+1 ·p1 − k2 sin2 α = cos α where α = am t. So we can find k and α from the equations and p1 − k2 sin2 α = R − a R + a cos α = r R + a wherefrom It follows: k2 = 4Ra (R + a)2 − r2 = 1 − (R − a)2 − r2 (R + a)2 − r2 11 (4.4.1) 4.5. Theorem. If the process closes up after n steps and m full turns then, defining k through (4.4.1), we have Z α 0 dφ p1 − k2 sin2 φ = m n Z π 0 dφ p1 − k2 sin2 φ §5. Back to pentagramma 5.1. Let us return to (3.3.1). Denoting φ = φi, φ′ = φi+2 and using (3.2.1) we get or g ′2 cos φ cos φ′ + g ′′2 sin φ sin φ′ = 1 cos φ cos φ′ + G′ G′′ sin φ sin φ′ = − G′ G (5.1.1) Now compare this with the Main formula (4.3.2). We see that (5.1.1) will be satisfied if φ = am(u), φ′ = am(u + w), α = am(w), and It follows that and We can go backwards. dn w =p1 − k2 sin2 α = G′ G′′ cos α = − cn w = − G′ G . G′ G k =r G′−2 − G′′−2 G′−2 − G−2 (5.1.2) (5.1.3) (5.1.4) 5.2. Theorem. For 0 ≤ k < 1, let K the corresponding complete elliptic integral (4.2.1). Define vectors in R3: rj(k, u) =(cid:18) cn(u + 4jK/5) pcn(2K/5) ,pdn(2K/5) sn(u + 4jK/5) pcn(2K/5) , 1(cid:19), 12 u ∈ R, j ∈ Z. Then rj(k, u) = rj+5(k, u). Set αj(k, u) = rj(k, u) × rj+1(k, u)2 (rj(k, u), rj+1(k, u))2 Then we have 1 + αj(k, u) = αj−2(k, u)αj+2(k, u) (5.2.1) In the degenerate case k = 0 we have rj(0, u)2 = √5, αj(0, u) = 1 + √5 2 for all j, u. The ends of vectors rj(0, u) form a regular plane pentagon, cf. 2.3. 5.3. The idea of uniformising relations in the spherical (resp. hyperbolic) geometry by elliptic functions (cf. [S] for a nice review) plays an important role in Statistical Physics. Onsager uses it in his 1944 paper [O] on the solution of the Ising model. Baxter remarks that the same idea underlies the solution of his star-triangle relations, cf. [B], 7.13. Baxter cites a book [Gr] in this connection; apparently this is the very same ”Greenhill’s very odd and individual Elliptic functions” which, according to Lit- tlewood cited by Hardy, was the Ramanujan’s textbook, cf. [H], Ch. XII. §6. Dilogarithm five-term relation 6.1. Euler dilogarithm: Li2(x) = 0 ≤ x ≤ 1. Rogers dilogarithm: ∞ Xn=1 xn n2 = −Z x 0 log(1 − t) t dt, L(x) = Li2(x) + 1 2 log x log(1 − x) = − 0 < x < 1. 6.2. Theorem (W.Spence, 1809). 1 2Z x 0 (cid:18)log(1 − t) t − log t 1 − t(cid:19)dt, L(x) + L(1 − x) = π2 6 (6.2.1) L(x) + L(y)− L(xy)− L(x(1− y)/(1− xy))− L(y(1− x))/(1− xy)) = 0, (6.2.2) 0 < x, y < 1. 6.3. One can reformulate (6.2.2) as follows (cf. [GT]): L(x) + L(1 − xy) + L(y) + L((1 − y)/(1 − xy)) + L((1 − x)/(1 − xy)) = Define 13 π2 2 (b1, . . . , b5) = (x, 1 − xy, y, (1 − y)/(1 − xy), (1 − x)/(1 − xy)) and for an arbitrary n ∈ Z define bn by periodicity bn = bn+5. Then Define a new sequence an by bn−1bn+1 = 1 − bn, n ∈ Z. Then an = bn 1 − bn , bn = an 1 + an an−2an+2 = 1 + an Combining this with 5.2 and (3.2.3) we get 6.4. Corollary. Let rj(k, u) be as in 5.2. Set βj(k, u) = rj(k, u) × rj+1(k, u)2 rj(k, u)2rj+1(k, u)2 L(βj(k, u)) = π2 2 5 Xj=1 Then In the regular case k = 0 (6.4.1) takes the form (Landen), cf. [K]. Here as usually α = (1 + √5)/2. L(α−1) = π2 10 (6.4.1) (6.4.2) References [B] R.J.Baxter, Exactly solved models in statistical mechanics, Academic Press, 1982. [C] H.S.M.Coxeter, Frieze patterns, Acta Arithm. XVIII (1971), 297 - 304. [G] C.F.Gauss, Pentagramma Mirificum, Werke, Bd. III, 481 - 490; Bd VIII, 106 - 111. [F] R.Fricke, Bemerkungen zu [G], ibid. Bd. VIII, 112 - 117. 14 [GT] F.Gliozzi, R.Tateo, ADE functional dilogarithm identities and integrable models, hep-th/9411203. [Gr] A.G.Greenhill, Applications of elliptic functions, 1892 (Dover, 1959). [GH] P.Griffiths, J.Harris, On Cayleys explicit solution to Poncelet’s porism. [H] G.H.Hardy, Ramanujan, Cambridge, 1940 (AMS Chelsea, 1991). [J1] C.G.J.Jacobi, Fundamenta nova theoriae functionum ellipticarum. [J2] C.G.J.Jacobi, Uber die Anwendung der elliptischen Transcendenten auf ein bekanntes Problem der Elementargeometrie, Crelles J. 3 (1828) [K] A.Kirillov, Dilogarithm identities, hept-th/9408113. [N] John Napier, Mirifici Logarithmorum canonis descriptio, Lugdini, 1619. [O] L.Onsager, Crystal statistics. I. A two-dimensioal model with an order- disorder transition, Phys. Rev. 65 (1944), 117 - 149. [S] J.Snape, Applications of elliptic functions in classical and algebraic geom- etry, Dissertation, Durham. Institut de Math´ematiques, Univ´ersit´e Paul Sabatier, 118 route de Narbonne, 31062 Toulouse, France
0903.3359
7
0903
2010-07-31T16:55:22
On ramified covers of the projective plane I: Segre's theory and classification in small degrees, with Appendix by Eugenii Shustin
[ "math.AG" ]
We study ramified covers of the projective plane. Given a smooth projective surface S and a generic enough projection of S to the projective plane, we get a cover of the plane ramified over a plane curve. The branch curve is usually singular, but is classically known to have only cusps and nodes as singularities for a generic projection. Several questions arise. First, what is the_geography_ of branch curves among all nodal-cuspidal curves? Second: what is the_geometry_ of branch curves? In other words, how can one distinguish a branch curve from a non-branch curve with the same numerical invariants? For example, a plane sextic curve with six cusps is known to be a branch curve of a generic projection iff its six cusps lie on a conic curve, i.e., form a special 0-cycle on the plane. We start with reviewing what is known about the answers to these two questions, mentioning both simple and some non-trivial results. We continue with study of classical work of Beniamino Segre which gives a complete answer to the second question in the case when S is a smooth surface in a three-dimensional projective space. We give an interpretation of Segre's work in terms of a study of Picard group of 0-cycles on a singular plane curve B. We also review examples of small degree. The Appendix written by Eugenii Shustin shows the existence of many new Zariski pairs of plane curves. We hope to continue this paper with a generalization to the case of smooth surfaces in a projective space of any dimension.
math.AG
math
ON RAMIFIED COVERS OF THE PROJECTIVE PLANE I: SEGRE'S THEORY AND CLASSIFICATION IN SMALL DEGREES (WITH AN APPENDIX BY EUGENII SHUSTIN) MICHAEL FRIEDMAN AND MAXIM LEYENSON1 Abstract. We study ramified covers of the projective plane P2. Given a smooth surface S in Pn and a generic enough projection Pn → P2, we get a cover π : S → P2, which is ramified over a plane curve B. The curve B is usually singular, but is classically known to have only cusps and nodes as singularities for a generic projection. Several questions arise: First, What is the geography of branch curves among all cuspidal-nodal curves? And second, what is the geometry of branch curves; i.e., how can one distinguish a branch curve from a non-branch curve with the same numerical invariants? For example, a plane sextic with six cusps is known to be a branch curve of a generic projection iff its six cusps lie on a conic curve, i.e., form a special 0-cycle on the plane. We start with reviewing what is known about the answers to these questions, both simple and some non-trivial results. Secondly, the classical work of Beniamino Segre gives a complete answer to the second question in the case when S is a smooth surface in P3. We give an interpretation of the work of Segre in terms of relation between Picard and Chow groups of 0-cycles on a singular plane curve B. We also review examples of small degree. In addition, the Appendix written by E. Shustin shows the existence of new Zariski pairs. Contents Introduction ´Etale covers 1. 2. Ramified covers 2.1. 2.2. Ramified covers of complex analytic spaces 2.3. Ramified covers of P2 3. Moduli of branch curves and their geography 3.1. Severi-Enriques varieties of nodal-cuspidal curves 3.2. Geography of branch curves 3.3. Construction of the variety of branch curves B(d, c, n) 4. Surfaces in P3 4.1. Smooth surfaces in P3 4.2. Adjoint curves to the branch curve 4.3. Adjoint curves and Linear systems 4.4. Segre's theorem 4.5. Special 0-cycles 4.6. Dimension of B(d, c, n) 4.7. Projecting surfaces with ordinary singularities. 5. Classification of singular branch curves in small degrees 6. Appendix A : New Zariski pairs (By Eugenii Shustin) 6.1. Introduction 2 3 3 3 4 4 5 7 11 13 13 16 19 24 28 29 30 34 38 38 1This work is partially supported by the Emmy Noether Research Institute for Mathematics (center of the Minerva Foundation of Germany), the Excellency Center "Group Theoretic Methods in the Study of Algebraic Varieties" of the Israel Science Foundation. 1 2 M. FRIEDMAN, M. LEYENSON 6.2. Construction 7. Appendix B : Picard and Chow groups for nodal-cuspidal curves 8. Appendix C : Bisecants to a complete intersection curve in P3 References 38 43 45 46 1. Introduction Let S be a non-singular algebraic surface of degree ν in the complex projective space Pr. One can obtain information on S by projecting it from a generically chosen linear subspace of codimension 3 to the projective plane P2. The ramified covers of the projective plane one gets in this way were studied extensively by the Italian school (in particular, by Enriques, who called a surface with a given morphism to P2 a "multiple plane", and later by Segre, Zariski and others.) The main questions of their study were which curves can be obtained as branch curves of the projection, and to which extent the branch curve determines the pair (S, π : S → P2). In the course of this study, Zariski studied the fundamental groups of complements of plane curves and in particular introduced what later became known as Enriques-Zariski-Van Kampen theorem (see Zariski's foundational paper [6]). It was also discovered by the Italian school that a branch curve of generic projections of surfaces in characteristics 0 has only nodes and cusps as singularities, though we were not able to trace a reference to a proof from that era (but see [63] for a modern proof). Segre-Zariski criterion (see [6] and also Zariski's book [22]) for a degree 6 plane curve with 6 cusps to be a branch curve is well known and largely used, but Segre's generalization ([8]), where he gave a necessary and sufficient condition for a plane curve to be a branch curve of a ramified cover of a smooth surface in P3 in terms of adjoint linear systems to the branch curve, was largely forgotten (see Theorem 4.32). Two recent surveys, by D'Almeida ([41]) and Val. S. Kulikov ([60]), were written on Segre's generalization. However, our approach is different, for we give an interpretation of Segre's work in terms of studying various equivalence relations of 0-cycles on nodal-cuspidal curves. We also em- phasize the logic of passing from adjoint curves for a plane singular curve to rational objects which become regular on the normalization of this curve, (what would be called "weakly holomorphic functions" in analytic geometry [18, Chapter VI] ), and control geometry of its space models. The geometry of ramified covers in dimension two is very different from the geometry in dimension In dimension one, for any set of points B in the projective line P1 we always have many one. possible (non isomorphic) ramified covers of P1 for which B is the branch locus. In terms of monodromy data, the fundamental group G = π1(P1 −B) is free, and thus admits many epimorphic representations G → Symν for multiple values of ν; and, moreover, every such a representation is actually coming from a ramified cover due to the Riemann-Grauert-Remmert theorem (see Theorem 2.2). However, in dimension two Chisini made a surprising conjecture (circa 1944, cf. [14]) that the pair (S, π : S → P2), where π is generic, can be uniquely determined by the branch curve B, if deg π ≥ 5 (and in the case of generic linear projections, if this curve is of sufficiently high degree). This conjecture was proved only recently by Kulikov ([52],[65]). In terms of monodromy data, by Grauert-Remmert theorem, even though it is true that every representation ρ : π1(P2 − B) → Symν comes from a ramified cover S → P2 of degree ν with a normal surface S, one has to ensure certain "local" conditions on the representation ρ which ensure that S is non-singular, which sharply reduce the number of admissible representations into the symmetric group. In fact, the Chisini's conjecture implies that once the degree of the ramified cover sufficiently high, there is only one such representation. The structure of the paper is as follows. Sections 2 and 3 contain preliminary material. In section 2 we recall some facts about ramified coverings and Grauert-Remmert theorems, and in the following section 3 we look at V (d, c, n) (resp. B(d, c, n)) the variety of degree d plane curves (resp. ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 3 branch curves) with c cusps and n nodes (in addition, we prove the following interesting fact: in coordinates (d, c, χ), with geometric Euler characteristic χ, the duality map (d, c, χ) → (d∗, c∗, χ∗) becomes a linear reflection). We also recall a number of necessary numerical conditions for a curve to be a branch curve. In the main section, Section 4 we re-establish the results of Segre for smooth surfaces in P3 and discuss the geography of surfaces with ordinary singularities and their branch curves. Looking at the variety of nodal cuspidal plane curves, we also compute the dimension of the component which consists of branch curves of smooth surfaces in P3 (see Subsection 4.6). In Section 5 we classify admissible (i.e. nodal-cuspidal irreducible) branch curves of small degree. Appendix A is written by Eugenii Shustin, where new Zariski pairs are found. Each Zariski pair consists of a branch curve of a smooth surface in P3 and a nodal -- cuspidal curve which is not a branch curve. In the other Appendices we recall some facts on the Picard and Chow groups of nodal cuspidal curves we use, and on the bisecants to complete intersection curves in P3. In the subsequent papers (see [67]) we will deal with an analogue of the Segre theory for sur- faces with ordinary singularities in P3, and also give a combinatorial reformulation of the Chisini conjecture. Acknowledgments: Both authors are deeply thankful to Prof. Mina Teicher and the Emmy Noether Research Institute at the Bar Ilan University (ENRI) for support during their work, and also thankful to Prof. Teicher for attention to the work, various important suggestions and stimu- lating discussions. The authors wish also to thank deeply E. Shustin for writing the Appendix. This is a major and important contribution to this work. The authors are also very grateful to Valentine S. Kulikov and to Viktor S. Kulikov, for paying our attention to the paper of Valentine S. Kulikov on branch curves in Russian [60]. We are grateful to Fabrizio Catanese for scanning and sending us a rare paper by Segre ([8]). We also thank Tatiana Bandman, Ciro Ciliberto, Dmitry Kerner, Ragni Piene, Francesco Polizzi and Rebecca Lehman for fruitful discussions and advices. The second author also wants to thank the department of Mathematics at the Bar Ilan University for an excellent and warm scientific and working atmosphere. In this section we start with a general discussion on branched coverings, continuing afterwards to investigation of surfaces and generic projections. 2. Ramified covers 2.1. ´Etale covers. Let S be a scheme (of finite type) over C, and San be the corresponding analytic space. Let Etf be the category of finite ´etale complex-anaytic spaces over San. One can verify (cf. [16] and [17, XI.4.3]) that the "analytization" functor S be the category of finite ´etale schemes over S, and Etf San aS : Etf S → Etf San is faithfully flat. The following Grauert-Remmert theorem generalizes the so-called Riemann exis- tence theorem in case dim S = 1: Theorem 2.1 (Grauert-Remmert). If S is normal, then aS is equivalence of categories. 2.2. Ramified covers of complex analytic spaces. Let X be a complex analytic space, Y ⊂ X be a closed analytic subspace in X, and U = X − Y be the complement. Assume that U is dense in X. Theorem 2.2 (Grauert-Remmert). If X is normal, then the restriction functor resU : (normal analytic covers of X ´etale over U ) −→ (´etale analytic covers of U) is an equivalence of categories. 4 M. FRIEDMAN, M. LEYENSON For other formulations of Theorem 2.2 and the proof, see [21, Proposition 12.5.3, Theorem 12.5.4.]. We say that f : X ′ → X is a ramified cover branched over Y if f U is ´etale and the ramification locus of f (i.e. supp(Ω1 X ′/X )) is contained in Y . Note that even if X is smooth, we still have to allow ramified covers X ′ → X with normal singularities in order to get an essentially surjective restriction functor, as seen in the following example. Let X = A2, Y = (xy = 0), U = X − Y , and f : U ′ → U be a degree 2 unramified cover given by the monodromy representation πan 1 (U ) ≃ Z ⊕ Z → Z/2 which sends both generators to the generator of Z/2. It is easily seen in this example that f can not be extended to a ramified cover X ′ → X with smooth X ′, but if we allow normal singularities one gets canonical extension given in coordinated by z2 = xy, a cone with A1 singularity. 2.3. Ramified covers of P2. From now on, we restrict ourselves to char = 0. Let S be a smooth surface in Pr = P(V ). Let W ⊂ V be a codimension 3 linear subspace such that P(W ) ∩ S = ∅ and let p the resulting projection map p : P(V ) → P(V /W ) and π : S → P2 its restriction to S. It is clear that π is a finite morphism of degree equals to deg S. for a generic choice of W , π is called a generic projection map and the following is classical (see, for example, [15], [26] and [63]): (i) π is ramified along an irreducible curve B ⊂ P2 which has only nodes and cusps as singu- larities; (ii) The ramification divisor B ∗ ⊂ S is irreducible and smooth, and the restriction π : B ∗ → B is a resolution of singularities; (iii) π−1(B) = 2B ∗ + Res for some residual curve Res which is reduced. Remark 2.3. Note that not every ramified cover S of P2 with a branch curve B ⊂ P2 can be given as a restriction of generic linear projection Pr → P2 (to a smooth surface S). See, for example, Remark 5.7. Remark 2.4. Note that generically cusps to not occur in a generic projection of a smooth space curve, but do occur for the projection of a ramification curve of surfaces, already in the basic example of smooth surfaces in P3 and its projection to P2. Consider, for example, the case of a smooth surface S in P3 and its generic projection to P2. Since the branch curve B is the projection of the ramification curve B ∗ which is a space curve, it generically has double points corresponding to bisecants of B ∗ containing the projection center O. The cuspidal points are somewhat more unusual for projections of smooth space curves, since they do not occur in the projections of generic smooth space curves. However, the projections of generic ramification curves have cusps. To give a typical example, consider a family of plane (affine) cubic curves z3 − 3az + x = 0 in the (x, z) - plane, where a is a parameter. The real picture is the following: for a > 0 the corresponding cubic parabola has 2 extremum point, for a = 0 one inflection point and for a < 0 no real extremums; the universal family in the (x, z, a) space is the so-called real Whitney singularity , and projection to the "horizontal" (x, a) plane gives a semi-cubic parabola a2 − x3 with a cusp. In other words, substituting y = −3a, we see that the affine cubic surface S can be considered as the "universal cubic polynomial" in z, p(z) = z3 + y · z + x = 0, and its discriminant ∆ = 27y2 + 4x3 has an A2 singularity, which is a cusp. (Recall that in general a discriminant of a polynomial of degree n with an−1 = 0 has singularity of type An−1). 3. Moduli of branch curves and their geography The geography of surfaces was introduced and studied by Bogomolov-Miyaoka-Yau, Persson, Bombieri, Catanese and more. Parallel to the terminology of geography of surfaces, we will use the term geography of branch curves for the distribution of branch curves in the variety of nodal- cuspidal curves. Subsection 3.1 recalls few facts on nodal-cuspidal degree d curves with c cusps ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 5 and n nodes and introduces a more natural coordinate to work with: χ -- the Euler characteristic. The main subsection is Subsection 3.2, which compares the geography of branch curves in the (d, c, χ) coordinates to the geography of surfaces in (c2 1, c2) coordinates. Subsection 3.3 constructs the variety of branch curves. 3.1. Severi-Enriques varieties of nodal-cuspidal curves. Notation 3.1. For a triple (d, c, n) ∈ N3 let V (d, c, n) be the variety of plane curves of degree d with c cusps and n nodes as their only singularities. It is easy to prove that V (d, c, n) is a disjoint union of locally closed subschemes of PN , where N = 1 2 d(d + 3). A curve C ∈ V (d, c, n), has arithmetic genus pa and geometric genus g = pg when (1) (2) pa = 1 2 (d − 1)(d − 2), g = pa − c − n = 1 2 (d − 1)(d − 2) − c − n, and we let χ to be the topological Euler characteristics of the normalization of C (3) . = 2 − 2g. χ We shall use the coordinates (d, c, χ) instead of (d, c, n) since many formulas, such as Plucker formulas, become linear in these coordinates. Note that one can present n in terms of (d, c, χ) as follows: n = 1 2 1 2 1 2 (d − 1)(d − 2) − c + χ − 1 = d(d − 3) − c + χ. 1 2 Let C ∈ V (d, c, n) be a Plucker curve, i.e., a curve that its dual C ∨ is also a curve in some V (d∗, c∗, n∗) (Note that this is an open condition in V (d, c, n) and that (C ∨)∨ = C.) Then the following Plucker formulas hold: (4) (5) d∗ = d(d − 1) − 3c − 2n, g = g∗ where g∗ is the geometric genus of C ∨. The formula from c∗ can be induced from Equations (2), (4) for C ∨, i.e. c∗ = 3d2 − 6d − 8c − 6n. 3.1.1. Linearity of the Plucker formulas. The Plucker formulas become linear in the (d, c, χ) coor- dinates (and also the formulas for the Chern classes of a surface whose branch curve B ∈ V (d, c, χ). See Lemma 3.9 and 3.10.), which is the primarily reason we want to consider them. Namely, (6) (7) (8) d∗ = 2d − c − χ, c∗ = 3d − 2c − 3χ, χ∗ = χ, in other words, in these coordinates projective duality is given by a linear transformation D =   2 −1 −1 3 −2 −3 0 1 0   6 M. FRIEDMAN, M. LEYENSON which is diagonalizable with eigenvalues (−1, 1, 1) where the eigenvector d − c − χ = d∗ − d corre- sponds to the eigenvalue (-1), i.e., gives a reflection in the lattice Z ⊕ Z ⊕ 2Z. We hope to explain this phenomenon elsewhere. The fact that the invariants d, c, n and g of the curve are not negative implies, in the (d, c, χ) coordinates, the following inequalities: (9) (10) (11) (12) (n ≥ 0) ⇒ 2c − χ ≤ d(d − 3), (g ≥ 0) ⇒ χ ≤ 2, (d∗ ≥ 0) ⇒ c + χ ≤ 2d, (c∗ ≥ 0) ⇒ 2c + 3χ ≤ 3d. Zariski also proved ([9, Section 3]) the following inequality (13) c < 1 2 (d − β)(d − β − 3) + 2, where β = [(d − 1)/6]. His proof uses the computation of the virtual dimension of complete linear system of curves of order d − β − 3 passing through the c cusps of C. (see also [22, Chapter VIII]). But his inequality is stronger then the ones given by Plucker formulas only for small d's; we use it once for d = 8 when classifying branch curves of small degree (see Section 5). Remark 3.2. For a nodal -- cuspidal curve C ∈ V (d, c, n) we have the following inequality or, in (d, c, χ) coordinates: 2c + n ≤ (d − 1)2 2c + χ ≤ d2 − d + 2, which is induced from intersecting two generic polars of C and B´ezout theorem. n d(d+3) 1/2.d(d-1) 1/2.(d-1)(d-2) = pa g = 0 d* = 0 1/4.d(d+3) 1/3.d(d-1) 1/2.(d-1)(d-2) = pa c Figure 1 : Geography of admissible plane curves in the (c, n) -- plane for large d. The dashed line is where the expected dimension of {family of degree d curves with n nodes and c cusps} = 1 2 d(d + 3) − n − 2c = 0. ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 7 c 6d 2d d 2 -d(d-3) 3/2d 2d 3d 1/2.d(d-3) c n = 0 c* = 0 d* = 0 Figure 2 : Geography of admissible plane curves in the (c, χ) -- plane for large d. The dashed line is where the expected dimension of {family of degree d curves with n nodes and c cusps} = 3d − 1 2 χ − c = 0. For more obstructions on the existence of singular plane curves and a recent survey on equisin- gular families and, in particular, nodal-cuspidal curves see [59]. 3.2. Geography of branch curves. Notation. Let B(d, c, n) be the subvariety in V (d, c, n) consisting of branch curves of generic linear projections to P2. We discuss it in Subsection 3.3. Let B ∈ B(d, c, n) be the branch curve of a generic linear projection π : S → P2 for a smooth irreducible projective surface S ⊂ Pr. Let ν = deg π, and g = pg(B) be the geometric genus of B. An important invariant of B is the fundamental group of its complement π1(P2 − B). Remark 3.3. Let C ∈ V (d, c, n). If c = 0, i.e., C is a nodal curve, then, by Zariski-Deligne-Fulton's theorem, the fundamental group π1(P2 − C) of the complement of C is abelian ([22],[64],[31]). This theorem was proved by Zariski under the assumption that the Severi variety of nodal curves is irreducible (this was assumed to be established by Severi, but later was found to be mistaken). The correct proof of the irreducibility of the Severi variety V (d, 0, n) was given by Harris [37], which completed Zariski's proof. Independent proofs were given later by Deligne and Fulton ([64],[31]) and others. We begin with a consequence from Nori's result on fundamental groups of complements plane curves. Though the proof is known, we bring it as it is enlightening and brings together various aspects of the subject. Lemma 3.4 (Nori [34]). Let B ∈ B(d, c, n). Then 6c + 2n ≥ d2. Proof. Let ψ : π1(P2 − B) → Symν be the monodromy representation, sending each generator to a permutation, which describes the exchange of the sheets. Since π : S → P2 is a generic projection, the image H = Im(ψ) is generated by transpositions. As S is irreducible, H is acts transitively on a set of n points. This implies that H = Symν and thus ψ is an epimorphism. Thus for ν > 2, the fundamental group π1(P2 − B) is not abelian and therefore c > 0 (by Remark 3.3). 8 M. FRIEDMAN, M. LEYENSON Nori proved ([34]) that for a cuspidal plane curve C with d2 > 6c + 2n and c > 0 the fundamental group of the complement π1(P2−C) is abelian. Thus, by the above discussion, ν = 2. However, there is no smooth double cover of P2 ramified over a singular C (indeed, locally S would be isomorphic to the singular cone z2 = xy in a formal neighborhood of a node of C and to the singular surface z2 = x2 − y3 in a formal neighborhood of a cusp). This implies that Nori's condition cannot hold for a branch curve. Therefore (14) or in (d, c, χ) coordinates: 6c + 2n − d2 ≥ 0, 4c + χ − 3d ≥ 0. In the spirit of the above Lemma, we have the following result of Shimada: Lemma 3.5 (Shimada [51]). Let B ∈ B(d, c, n). Then 2n < d2 − 5d + 8. Proof. Let C ∈ V (d, c, n). By [51], if 2n ≥ d2 − 5d + 8 then π1(P2 − C) is abelian. However, for a branch curve B ∈ B(d, c, n), the corresponding fundamental group is not abelian, and we have (cid:3) (15) or, in (d, c, χ) coordinates: 0 < 1 2 (d2 − 5d + 8) − n, 2c − χ − 2d + 8 > 0. The following conditions on c and n are less obvious then the previous Lemmas: Lemma 3.6. (16) or in (c, χ) coordinates: Proof. see [40]. c = 0 mod 3, n = 0 mod 4 c = 0 mod 3, χ = 2c − d(d − 3) mod 8 (cid:3) (cid:3) 3.2.1. Geography of branch curves in (d, c, χ) versus geography of surfaces in (c2 1, c2). Let S be a smooth algebraic surface and π : S → P2 be a generic ramified cover. Let B be the branch curve of π, B ∈ B(v) for some vector v ∈ L, and let ν = deg π. It is well known that d ≥ 2ν − 2 but for the convenience of the reader we bring the proof of this fact. Lemma 3.7. (17) d ≥ 2ν − 2. Proof. Denote by π : S → P2 the projection map and let C = f −1(l) for a generic line l. The curve C is irreducible and smooth. Applying the Riemann- Hurwitz's formula to the map πC : C → l, we get 2g(C) − 2 = −2ν + d, since all the ramification points of the map πC are of ramification index 2, and (C 2)S = ν, which implies d = 2ν − 2 + 2g(C) ≥ 2ν − 2 A different proof will be given in Subsection 4.7 when we discuss the geometry of surfaces with (cid:3) ordinary singularities in P3. Remark 3.8. Note that the proof of the above Lemma implies that the degree of a branch curve is even. ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 9 We want to express the Chern invariants c2 1(S) and c2(S) in terms of (d, c, χ) and, equivalently, in terms of (d, c, n), so we give 2 formulas for each invariant. Lemma 3.9. (see [52]) (18) (19) c2 1(S) = 9ν − 9 2 d − 1 2 χ c2 1(S) = 9ν − 9 2 d +(cid:18)(d − 1)(d − 2) 2 − n − c(cid:19) − 1 Proof. Let π : S → P2 be the ramified cover, R = B ∗, the ramification curve and C = f −1(l) for l a generic line. First, we want to compute [R]2 and [C]2. By Riemann-Hurwitz, KS = −3f ∗([l]) + [R] = −3[C] + [R]. As π : R → B is a normalization of the branch curve B, we apply adjunction formula to R we get 2g − 2 = (KS + [R]) · R = (−3[C] + 2[R]) · R = −3[C] · R + 2[R] · R = −3f ∗[l] · R + 2[R]2 = = −3[l] · f∗[R] + 2[R]2 = −3 deg B + 2[R]2 = −3d + 2[R]2 and thus We also have (20) [R]2 = 3 2 d + g − 1. [C]2 = f ∗[l] · [C] = [l] · f∗[C] = [l] · (deg f [l]) = ν[l]2 = ν. We can now compute c2 1(S): c2 1(S) = K 2 S = (−3[C] + [R])2 = 9[C]2 − 6[C] · [R] + [R]2 = 9ν − 6d + 3 2 d + g − 1 = = 9ν − 9 2 d + g − 1 = 1 = 9ν − 9 2 d − 1 2 χ The expression in (d, c, n)-coordinates follows easily. (cid:3) Lemma 3.10. (see [52]) (21) (22) c2(S) = 3ν − χ − c, or c2(S) = 3ν + d2 − 3d − 3c − 2n. Proof. To compute c2(S) we use the usual trick of considering a pencil of lines in P2 passing through a generic point p ∈ P2 and its corresponding preimage with respect to π : S → P2 -- the Lefshetz pencil Ct of curves on S. We then apply the following formula on Ct c2(S) = χ(S) = 2χ(generic fiber) + #(singular fibers) − (self-intersection of Ct) (see, for example, [28, section 4.2]). The generic fiber of Ct is a ramified cover of a line l with d simple ramification points (i.e. ramification index 2 at every point), and thus χ(generic fiber) = 2ν − d by the Riemann -- Hurwitz formula. The number of singular fibers in the pencil Ct is clearly equal to the degree d∗ of the curve B ∨ (the dual to the branch curve B), which by the Plucker formulas for B satisfies d∗ = d(d − 1) − 3c − 2n. The self-intersection [Ct]2 of the fiber equals to ν (by (20)). Thus 10 M. FRIEDMAN, M. LEYENSON c2(S) = χ(S) = 2(2ν − d) + d∗ − ν = 2(2ν − d) + (d(d − 1) − 3c − 2n) − ν = = 3ν + d2 − 3d − 3c − 2n = 3ν − χ − c. (cid:3) Remark 3.11. Equation (21) can be written as an analog to Riemann-Hurwitz formula for the map S → P2 c2(S) − νc2(P2) = −χ − c, as Iversen described in [20]. Remark 3.12. Inverting the formulas above, we get n and c in terms of c2 1, c2, ν and d: n = −3c2 1(S) + c2(S) + 24ν + d2 2 − 15d, c = 2c2 1(S) − c2(S) − 15ν + 9d We use these formulas below in Subsection 4.7.1. The next two results are rather surprising, as one gets an inequality for the branch curve which is independent of the degree of the projection: Lemma 3.13. (see, .e.g., the introduction of [35]) Let B ∈ B(d, c, n) a branch curve of a linear projection to P2 of a surface of general type, where d, c, n, χ and ν as above. Then (23) 5χ + 6c − 9d ≤ 0 or, equivalently, in (d, c, n) coordinates: 10n + 16c − 5d2 + 6d ≤ 0 Proof. Substituting the expressions for c2 (d, c, χ) and (d, c, n) into the Bogomolov inequality c2 1(S) and c2(S) (from Lemmas 3.9,3.10) in terms of ν and 1(S) ≤ 3c2(S), we get the desired inequality. (cid:3) There is, however, an inequality which is true for every branch curve, restricting the sum of the nodes and the cusps (though it is weaker than inequality (23)). Lemma 3.14. Let B ∈ B(d, c, n). Then or 15d − 5χ − 6c > 0 10n + 16c < 5d2. Proof. Note that Nemirovski's inequality (see [54]) for branch curves is or, equivalently 3d − χ 3d − χ − c < 6 15d − 5χ − 6c > 0. Remark 3.15. The variety B(d, c, n) is not necessarily connected. See, for example, [66], where it is proven that B(48, 168, 840) has at least two disjoint irreducible components. (cid:3) ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 11 3.2.2. Chisini's conjecture. The following theorem was known as Chisini's Conjecture, by now proved by Victor Kulikov (see [52], [65]): Theorem 3.16. Let B be the branch curve of generic projection π : S → P2 of degree at least 5. Then (S, f ) is uniquely determined by the pair (P2, B). Kulikov proved this conjecture for generic covers of degree greater than 11 and for generic linear projections of degree greater than 4. Kulikov considered two surfaces S1, S2 ramified over the same branch curve, and studied the fibred product S1 ×P2 S2, proving that the normalization of this fibred product contradicts Hodge's Index Theorem if (S1, f1) is not isomorphic to (S2, f2). Remark 3.17. We want to mention that a version of a Generalized Chisini's conjecture also exists, for surfaces with normal isolated singular points: Conjecture 3.18. Let fi : Si → P2, i = 1, 2 be two generic coverings with the same branch curve B where Si can have singular points, denoted as Sing Si. Assume f1(Sing S1) = f2(Sing S2). Then either there exists a morphism φ : S1 → S2 s.t. f1 = f2 ◦ φ or (f1, f2) is an exceptional pair. See [57] for the definition of an exceptional pair. This theorem was partially proven by V. S. Kulikov and Vik. S. Kulikov for f1, f2 generic m -- canonical coverings, for m ≥ 5 (see [55]) or when max(deg f1, deg f2) ≥ 12 or max(deg f1, deg f2) ≤ 4 (see [57]). Remark 3.19. One of the theorems induced from the proof of the Chisini's conjecture was the fact that a class of certain factorization associated to the branch curve B (i.e. the Braid Monodromy Factorization) determines the diffeomorphism type of S as a smooth 4-manifold. We refer the reader to [33], [39] for an introduction of this factorization, and to Kulikov and Teicher's proof [53] of the above theorem. 3.2.3. Representation-theoretic reformulation. Let Gi (resp. Γi) be the local fundamental group of P2 − B at the neighborhood of a cusp (resp. a node) of B. Note that each Gi is isomorphic to the group with presentation {a, b : aba = bab} and every Γi is isomorphic to the group with presentation {a, b : ab = ba} = Z2. Let l be a line in P2 in generic position with B, pi (i = 1, . . . , d) be the intersection points of B and l, p∗ be a generically chosen point in l and γi be a small loop around pi starting and ending at p∗. The map F reed → π1(P2 − B) sending generators of F reed to [γi] is epimorphic by Zariski -- Van Kampen theorem, and the classes [γi] are called geometric generators of π1(P2 − B). It is well known (see [44] or [52, Proposition 1]) that given a ramified cover S → P2, the monodromy map ϕ : π1(P2 − B) → Symν satisfies the following three conditions: (i) for each geometric generator γ, the image ϕ(γ) is a transposition in Symν; (ii) for each cusp qi, the image of the two geometric generators of Gi is two non-commuting transpositions in Symν; (iii) for each node pi, the images of two geometric generators of Γi are two different commuting transpositions in Symν. The inverse assertion is a group theoretic reformulation on the Chisini's theorem ([44]): Proposition 3.20. The map associating the monodromy representation with each ramified cover S → P2 gives an isomorphism of the set of the isomorphism classes of generic ramified covers of P2 of degree ν with the branch curve B and the set of isomorphism classes of epimorphisms ϕ : π1(P2 − B) → Symν satisfying the conditions (i),(ii) and (iii) above, with respect to the action of Symν on the set of such representations by inner automorphisms. 3.3. Construction of the variety of branch curves B(d, c, n). Let V = V (d, c, n) be the Severi- Enriques subvariety in dh of degree d plane curves with n nodes and c cusps. Let B = B(d, c, n) ⊆ V the subset consists of branch curves. In this subsection we show that B(d, c, n) is a subvariety 12 M. FRIEDMAN, M. LEYENSON of V (d, c, n). Although it is standard, we have not found it in the literature, though references to its existence can be found in [24] or in [61]. The following lemma proves that the variety of branch curves of ramified covers is a union of connected components of V . Using the same techniques in the following proof, and the fact that the Chisini's conjecture is proven (for generic linear projections), one can prove that also B is a union of connected components of V . Lemma 3.21. Over the field k = C, every connected component Vi of V either does not contain branch curves of generic covers at all, or every curve C ∈ Vi is a branch curve of a generic cover. Explicitly, every component of B is a connected component of V . Proof. Let us fix a connected component V1 of V = V (d, c, n), let p ∈ V1, and let C be the corresponding plane curve. Take q ∈ V1, q 6= p and choose a path I = [0, 1] → V1 connecting p and q. Let us denote GC = π1(P2 − C), GCt = π1(P2 − Ct) with Ct ∈ V1, t ∈ I where C1 corresponds to q. As these curves are equisingular, we get an identification of fundamental groups GCt ∼ → GC. For every t ∈ I. Consider the group Hom(GC , SymN ) and its subgroup Homgeom(GC , SymN ) of geometric homomorphisms -- i.e., homomorphisms which satisfy the conditions (i),(ii),(iii) above -- which can be empty. From the above identification, we get a canonical set bijection from Hom(GCt , SymN ) → Hom(GC , SymN ) preserving the set of geometric homomorphisms. In par- ticular, Homgeom(GC , SymN ) is empty if and only if Homgeom(GC1, SymN ) is empty, and thus C is a branch curve if and only if C1 is. Therefore B(d, c, n) is a union of connected components of V and thus it is a subvariety. (cid:3) Remark 3.22. We want to describe here on the action of the fundamental group π1(V ) on G = π1(P2 − C). Let p ∈ V , C be the corresponding degree d plane curve and U = P2 − C. A loop γ : I → V (starting and ending at p), induces an automorphism of the group G = π1(U ), and thus an automorphism of the set of representations Hom(π1(U ), SymN ) which preserves the set of geometric representations Homgeom(π1(U ), SymN ). To describe it more explicitly, note that we can choose a line l ⊂ P2 in generic position to every Ct, t ∈ I (since the set of lines in special position to a fixed curve in P2 forms a dual curve in the dual plane, and thus the space of lines which are special to some Ct is of real codimension 1 in the dual plane). Note that l − l ∩ C = l ∩ U ≃ P1 − {d points}. Let us now choose a base point a∗ on l not belonging to any of the curves Ct, and a "geometric basis" Γ of π1(U, a∗) = π1(P2 − C, a∗), which gives an epimorphism e(Γ) : π1(l ∩ U, a∗) → π1(U, a∗). Recall that the group of classes of diffeomorphism of (P1 −d points) modulo diffeomorphisms homo- topic to identity can be identified with the commutator of the braid group B ′ (see e.g. [33]). A loop γ ∈ π1(V ) gives a diffeomorphism of l ∩ U , which in turn induces an auto- morphism of π1(l ∩ U ), i.e. an element in Aut(π1(l ∩ U )) or equivalently, an element bγ ∈ B ′ d. It follows that there is a natural diagram d = Braidd/Center(Braidd) π1(V ) α Aut π1(U ) / Aut(Homgeom(π1(U ), SymN )) β (PPPPPPPPPPPP Aut π1(l ∩ U ) ≃ B ′ d and a commutative triangle: π1(V ) α Im(α) ⊆ Aut π1(U ) β (PPPPPPPPPPPP Im(β) ⊆ B ′ d / / ( / / / ( O O ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 13 An element bγ ∈ B ′ generators of B ′ explicitly the action of each xi on Aut(F reed): d, i.e. bγ = x±1 1 d which is the image of γ admits a decomposition of bγ into a product of canonical k . Since π1(l ∩ U ) = F reed = hy1, ..., ydi, we can describe · ... · x±1 xi(yj) = yj if j 6= i, i + 1 xi(yi) = yi+1 xi(yi+1) = y−1 i+1 · yi · yi+1. Thus, the action of an element γ ∈ π1(V ) on the group G = π1(U, a∗) can be expressed as a map on the generators {yi} of G : (yi 7→ bγ(yi) = (x±1 k )·yi) where xj ·yi is given by the above action. Note that this action is non-trivial in general, and thus π1(V, p) acts generically non-trivially on the set of good covers S → P2 ramified over a given curve C. However, in a situation when such a cover is unique up to a deck transformation, like in the case of a high degree ramified cover, (due to Chisini's conjecture), this action reduces to the action of the deck transformation group Aut(S/P2) which is the trivial group, for geometric reasoning. 1 ·...·x±1 4. Surfaces in P3 Let X be a smooth surface in Pr and p : Pr → P2 be generic projection; we decompose p as a composition of projections Pr p1→ P3 p2→ P2 such that S = p1(X) is smooth or has ordinary singularities in P3. We begin in section 4.1 with the examination of branch curves of smooth surfaces in P3 and proceed to singular surface in section 4.7. 4.1. Smooth surfaces in P3. Our goal here is to reformulate and give a more modern proof to a result of Segre [8] published in 1930. Segre proved that the set of singular points of the branch curve of a smooth surface in P3 is a special 0-cycle with respect to some linear systems on P2, i.e., it lies on some curves of unexpectedly low degree. (We remind that a curve passing through the singularities of a given one is called adjoint curve. See Definition 4.7). For example, if deg S = 3, we get the following result Zariski published in 1929 (cf. [6]): the variety of plane 6-cuspidal sextics has two disjoint irreducible components. Every curve in the first component is a branch curve of a smooth cubic surface and all its six cusps are lying on a conic, while the second component does not contain any branch curves. (Miraculously, this condition does not define a subvariety of positive codimension in the variety of all plane curves of degree 6 with 6 cusps, but rather selects one of its two irreducible components, which was probably the most surprising discovery of Zariski concerning this variety.) In the following paragraphs we recall Segre's method for constructing some adjoint curves to branch curves of ramified covers. The main result is the following: a nodal -- cuspidal curve B is a branch curve iff there are two adjoint curves of (some particular) low degree passing through all the singularities of B (see Theorem 4.32). Though this result was presented in [60] (by Val. S. Kulikov) and in [41] (by J. D'Almeida), our point of view is different, as we emphasize the relations between the Picard and Chow groups of 0 -- cycles of the singularities of the branch curve. We also investigate the connections between adjoint curves and the sheaf of weakly holomorphic rational functions on a nodal -- cuspidal curve C. We hope that the study of the Picard group of branch curves and the study of adjunction with values in sheaf of weakly holomorphic rational functions (see [18]) gives a new understanding of the work of Segre. Let S be a smooth surface of degree ν in P3, and let π : P3 → P2 be a projection from a point O which is not on S. Let B ⊂ P2 be a branch curve of π. It is easy to see that the degree of B is d = ν(ν − 1): indeed, B is naturally a discriminant of a homogeneous polynomial of degree ν in one variable. The curve B is in general singular, however, for a generic projection it has only nodes and cusps as singularities (see e.g. [63]). 14 M. FRIEDMAN, M. LEYENSON Assume now that S is given by a homogeneous form f (x0, . . . , x3) of degree ν, and O = (O0, .., O3) is a point in P3 which is not on S. The polar surface PolO(S) is given by the degree ν − 1 form P Oifi, where fi = ∂f ∂xi . The following lemma is well known: Lemma 4.1. Let π : S → P2 be the projection with center O. The ramification curve B ∗ of π is the intersection of S and the first polar surface PolO(S). Indeed, the intersection of S and PolO(S) consists of such points p on S that the tangent plane to S at p, TpS, contains the point O. This implies that the line joining O and p intersects S with multiplicity at least 2 at p. Note that this gives yet another proof that deg B ∗ = deg S · deg(PolO(S)) = ν(ν − 1). Notation: (1) H ∈ A2P3 is a class of hyperplane in P3; (2) h ∈ A1P2 is a class of a line in P2; (4) ℓ∗ = HB∗ , ℓ∗ ∈ A0B ∗; (3) ℓ = hB, ℓ ∈ A0B; We also denote (5) S ′ (6) S ′′ O = PolO(S) ⊂ P3, and O = Pol2 neous form f ′′ = (P Oi ∂ ∂xi O(S) is the second polar surface to S w.r.t. the point O; it is given by a homoge- )2f = P OiOjfij of degree ν − 2. (7) We call a 0-subscheme with length 1 at every point a 0-cycle. (8) Let P ⊂ B be the 0-cycle of nodes on B, and P ∗ be its preimage on B ∗. Note that deg P ∗ = 2 deg P , as can be seen from Lemma 4.2. (9) Let Q ⊂ B be the 0-cycle of cusps on B, and Q∗ be its preimage on B ∗. Note that deg Q∗ = deg Q (see Lemma 4.2). (10) ξ be the 0-cycle of singularities of B. From now on we assume that O is chosen generically for a given surface S. It follows that B ∗ is smooth, and B has only nodes and cusps as singularities. Already in the 19th century the number of nodes and cusps of a branch curve was computed for a smooth surface of a given degree. Lemma 4.2 (Salmon [1]). (a) There is one-to-one correspondence between bisecant lines for B ∗ passing through O and nodes of B. Moreover, the number of bisecant lines through O does not depend on S, and is equal to (24) n = n(ν) = 1 2 ν(ν − 1)(ν − 2)(ν − 3) (b) If Q∗ is the set of points q on B ∗ such that the tangent line TqB ∗ contains the point O, then the set Q = π(Q∗) iis the set of cusps of B. (c) Moreover, Q∗ is the scheme-theoretic intersection of B ∗ and the second polar surface S ′′ O. In other words, they intersect transversally at each point of Q∗, and B ∗ ∩ S ′′ = Q∗. In particular, the class [Q∗] in A0B ∗ is equal to (ν − 2)l∗. (d) It follows that deg Q does not depend on a choice of the surface S, and is equal to (25) c = c(ν) = ν(ν − 1)(ν − 2) Proof. (a) The first statement is geometrically clear; for the second see [1, art. 275, 279]. Yet another proof is given below, in Proposition 4.8. See also [15, Chapter IX, sections 1.1,1.2] for a way to induce the formula for the number of bisecant of a complete intersection curve in P3 (i.e. the number n + c). For (b), see [1, art. 276]. (c) is a straightforward computation, and (d) follows from (c). (cid:3) ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 15 Lemma 4.3. Let ℓ ∈ A0(B) be the class of a plane section on B. Then (2) The equality above can be lifted to Pic B: there is a Cartier divisor Q0 such that can(Q0) = Q with respect to the canonical map [Q] = (ν − 2)ℓ in A0(B), associating Weil divisor with a Cartier divisor, and can : Cartier(B) → Weil(B) in Pic(B). [Q0] = (ν − 2)ℓ Proof. We have Q = π∗(Q∗), and Q∗ = B ∗ ∩ S ′′ [Q∗] = (ν − 2)ℓ∗ in A0B ∗, and thus O. Since [S ′′ O] = (ν − 2)H in A2P3, we have [Q] = [π∗(Q∗)] = π∗([Q∗]) = (ν − 2)π∗ℓ∗ = (ν − 2)ℓ in A0B. To see that π∗ℓ∗ = ℓ it is enough to consider a hyperplane in P3 containing the point O. (2) Consider the rational function r = f ′′ O is by definition the equation of the second polar Pol2(O, S), and H is an equation of a generic hyperplane containing the projection center O. Since the curves B ∗ and B are birational, r can be considered as a rational function on B, where it gives the desired linear equivalence. O/H (ν−2), where f ′′ Remark 4.4. Note that both cusps and nodes on a curve are associated with Cartier divisors on the curve, even though these Cartier divisors are not positive. For example, on the affine cuspidal curve C given by the equation y2 − x3 = 0 the divisor (y/x) = 3[0] − 2[0] = [0] is a principle Cartier, but since y/x is not in the local ring of the point [0], it is not locally given a section of the sheaf (cid:3) OC . For the nodal curve C given by the equation xy = 0, the divisor (cid:16) y−x2 also a principle Cartier, though not positive. y−2x(cid:17) = 3[0] − 2[0] = [0] is 4.1.1. Example: smooth cubic surface in P3. Let S be a smooth cubic surface in P3. Then Lemma 4.2 imply that B is a plane curve with 6 cusps and no other singularities, and Lemma 4.3 implies that [Q] = ℓ in A0B. Q is, of course, not a line section of the curve B; the linear equivalence above implies that the map P H 0(P2, O(1)) ≃ P H 0(B, O(1)) → ℓ is not epimorphic, where ℓ is the set of all Weil divisors linearly equivalent to a generic line section of B. Even though Q is associated with a Cartier divisor b/a, this Cartier divisor is not positive. It is well known that 6 points in general position on P2 do not lie on a conic. As for the 6 cusps Q on the branch curve we have the following result of Zariski and Segre (see [6],[8]). Corollary 4.5. All 6 cusps of a degree 6 plane curve B which is a branch curve of a smooth cubic surface lie on a conic. Remark 4.6. Explicit construction of a branch curve of a cubic. By change of coordinates a cubic surface S is given by the equation f (z) = z3 − 3az + b, 16 M. FRIEDMAN, M. LEYENSON where a and b are homogeneous forms in (x, y, w) of degrees 2 and 3, and the projection π is given by (x, y, w, z) 7→ (x, y, w). In these coordinates the ramification curve is given by the ideal (f, f ′) = (f, z2 − a) = (z3 − 1 2 b, z2 − a) and the branch curve B is given by the discriminant ∆(f ) = b2 − 4a3 In particular, one can easily see that it has 6 cusps at the intersection of the plane conic defined by a and the plane cubic defined by b, as illustrated on the Figure 3. It is also clear that the conic defined by a coincides with one constructed in Corollary 4.5. Figure 3 : The branch curve of a smooth cubic surface The ideal of Q∗ is equal to (f, f ′, f ′′) = (f, f ′, z) = (a, b, z). Note that z equal to b 2a as a rational section of OB∗ (1). We want to explicate the linear equivalence of Q∗ and the intersection of B ∗ with the "vertical" plane (one containing the point O). For this, let l(x, y, w) be a linear form in x, y, w, and consider the rational function on B ∗ z l b 2al φ = = Then φ gives the linear equivalence 0 = (φ) = (b) − (a) − (l) = 3Q − 2Q − (l) = Q − (l), which gives an explicit proof that [Q] = ℓ in A0B. (We used the fact that cubic b is tangent to B at the cusps, while conic a is not.) This example has a "natural" continuation in example 4.30. 4.2. Adjoint curves to the branch curve. We begin with the definition of an adjoint curve. This type of curves will play an essential role when studying branch curve. Definition 4.7. Given a plane curve C, a second curve A is said to be adjoint to C if it contains each singular point of C of multiplicity r with multiplicity at least r − 1. In particular, A is adjoint to a nodal-cuspidal curve C if it contains all nodes and all cusps of C. For more on adjoint curves see [2, § 7], [15, Chapter II, § 2], or [29] for a more recent survey. Below, following Segre, we construct more adjoint curves to B (i.e. W , L, L1) and relate them to the geometry of B ∗ in P3. We continue this subsection with Proposition 4.8 from [8] and we bring its proof for the conve- nience of the reader. Proposition 4.8. (a) One has in A0(B) 2[P ] + 3[Q] = ν(ν − 2)ℓ. (b) The equality above can be lifted to Pic B: there are Cartier divisors P1 and Q1 such that in Pic B: can(P1) = 2P, can(Q1) = 3Q, and [P1] + [Q1] = ν(ν − 2)ℓ ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 17 In fact, Q1 is the canonically defined "tangent" Cartier class Qτ . Proof. (following Segre [8]). Let us choose a plane Π in P3 not containing the point O, and consider the projection with center O as a map to Π. Let us also choose a generic point O′ = (O′ 2) ∈ Π, and let B ′ = PolO′(B) be the polar curve of B, defined as follows: if B is given by the homogeneous form g(x0, x1, x2) of = 0}. Note that the first polar B ′ = PolO B is 1, O′ 0, O′ degree d = ν(ν − 1), then PolO′(B) = {P2 adjoint to B. i=0 O′ i ∂g ∂xi It is clear that (26) [B ∩ B ′] = 2P + 3Q + R, where R (for "residual") is the set of non-singular points p on B such that the tangent line to B at p contains O′, and thus (27) [2P + 3Q + R] = (d − 1)ℓ in A0B (Here we used the fact that O′ is generic, in particular, it does not belong to B and to the union of tangent cones to B at nodes and cusps.) Let R∗ be the preimage of R on B ∗. We claim that R∗ = B ∗ ∩ S ′ O′ = PolO′(S). Indeed, if p ∈ R, then TpB contains the point O′, and if p∗ is the preimage of p on B ∗, then the tangent space to S at p∗ can be decomposed into a direct sum of the line l joining p and p∗ (and containing O) and the tangent line Tp∗B ∗ which projects to the tangent line TpB, (as illustrated on Figure 4 below). O′, where S ′ o p* B* o` p B P Figure 4 : R∗ = B ∗ ∩ S ′ O′ It follows that in A0B, and thus [R] = π∗([R∗]) = π∗([B ∗ ∩ S ′ O′]) = π∗((ν − 1)ℓ∗) = (ν − 1)ℓ [2P + 3Q] = [2P + 3Q + R] − [R] = (d − 1)ℓ − (ν − 1)ℓ = (d − ν)ℓ = ν(ν − 2)ℓ The proof of the second part is parallel, as the Weil divisors 2P and 3Q can be lifted to Pic B. (cid:3) Note that this gives yet another proof for the formula for the number of nodes n = n(ν). Proposition 4.9. There exist a (unique) curve W in the plane Π of degree ν(ν − 2) such that in A0(B) [W ∩ B] = 2[P ] + 3[Q]. 18 M. FRIEDMAN, M. LEYENSON Proof. By the previous proposition, the cycle 2[P ] + 3[Q] is in the linear system ν(ν − 2)ℓ on B. Note that 2P + 3Q is actually a Cartier divisor (see Remark 4.4). Now, since deg B = ν(ν − 1) is greater than ν(ν − 2), there is a restriction isomorphism 0 → H 0(Π, O(ν(ν − 2))) → H 0(B, O(ν(ν − 2))) → 0 which completes the proof. (cid:3) Note that W is an adjoint curve to B which is tangent to B at each cusp of B. Proposition 4.10. Let a = (ν − 1)(ν − 2). (1) We have [2P + 2Q] = aℓ In A0(B). (b) The equality above can be lifted to Pic(B): there are Cartier divisors P2 and Q2 such that in Pic(B): can(P2) = 2P, can(Q2) = 2Q, and [P2] + [Q2] = aℓ (2) There is a (unique) curve L of degree a such that [L ∩ B] = 2P + 2Q. Proof. (1) We have [2P + 2Q] = [2P + 3Q] − [Q] = ν(ν − 2)ℓ − (ν − 2)ℓ = (ν − 1)(ν − 2)ℓ = aℓ. The computation in Pic(B) is parallel: we let P2 = P1 and Q2 = Q1 − Q0. (2) Note that (ν − 1)(ν − 2) < deg B, which completes the proof. (cid:3) Note that L is an adjoint curve to B which is not tangent to B at the cusps of B. Notation 4.11. Let ζL be the Cartier divisor on B given by restricting the equation of L to B. Recall that ζL is supported on the 0 -- cycle of singularities ξ. Definition 4.12. Let V (d, c, n) be the variety of plane curves of degree d with c cusps and n nodes, abd let B(d, c, n) be the subvariety in V (d, c, n) consisting of branch curves of all ramified covers of P2. Example 4.13. By substituting ν = 3 and ν = 4 we get the classical example of a sextic with six cusps on a conic we discussed above, and the example of a degree 12 curve with 24 cusps and 12 nodes, all of them are on a sextic: (1) The branch curve C of a smooth cubic surface in P3 is a sextic with six cusps, C ∈ B(6, 6, 0). We have deg L = 2; two different constructions of this conic was given above in Corollary 4.5 and Remark 4.6. See also Figure 3 above. (2) The branch curve C of a smooth quartic surface in P3 is of degree 12, and has 24 cusps and 12 nodes, i.e., C ∈ B(12, 24, 12). We have deg L = 6. Moreover, the 24 cusps lie on the intersection of a quartic and a sextic curves (see e.g. [8]). ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 19 4.3. Adjoint curves and Linear systems. We start with the following easy Lemma: Lemma 4.14. (Adjunction for a flag (ξ, ζ, K, P )) Assume we are given a flag of 4 (arbitrary) schemes (ξ, ζ, K, P ). Then there is a diagram 0 0 Jξ,ζ / Jξ,ζ 0 0 / JK / JK / Jξ / Jζ 0 where JX = JX,P , and X is either ξ, ζ, or K. 0 0 Jξ,K Jζ,K 0 Corollary 4.15. Coming back to our standard notations, let K be a plane curve, ξ ⊂ ζ ⊂ K be a flag of 0-subschemes on K such that ζ is given by a positive Cartier divisor, and ξ = supp ζ. In this case Jζ,K = OK (−ζ). Then, given an integer n < deg K, the diagram above gives isomorphisms 0 0 0 / H 0(P2, Jξ,ζ(n)) / H 0(K, Jξ,ζ (n)) / H 0(P2, Jξ(n)) / H 0(K, Jξ,K (n)) / H 0(P2, Jζ (n)) / H 0(K, OK (−ζ)(n)) / 0 / 0 / 0 0 0 Corollary 4.16. Assume that there is an integer a and a positive Cartier divisor ζ = ζ0 on K such that there is a linear equivalence ζ0 ∼ al, where l is the class of a line section on K. (Such is the case of a branch curve and the class ζL constructed above.) Then, setting n = a + i, we get isomorphisms jK,ζ0(i) : H 0(P2, Jζ0 (a + i)) → H 0(K, OK (i)) for every i ≥ 0. In other words, adjoint curves on P2 with given tangent conditions at the singularities of the curve K correspond to homogeneous forms on K with given tangent conditions at the singularities of the curve K correspond to homogeneous forms on K. We only need this isomorphism for i = 0; it implies that there is a curve L0 ∈ H 0(P2, Jζ0 (a)) of degree a corresponding to the element 1 ∈ H 0(K, OK ), and ζ0 is locally given by the equation of L0. (This is exactly the case of a branch curve K = B, where ζ0 = ζL and L0 = L). The isomorphism jK,ζ0(i) is given by h 7→ h fL0 , where fL0 is an equation of the curve L0. O O / o / O O / / / / O O / / O O / / / / O O / / O O O O O O / / / / O O / O O / / O O / O O / O O O O 20 M. FRIEDMAN, M. LEYENSON Our next goal is to study curves of various degrees n > a containing the 0-cycle ξ but restricting to different Cartier divisors with support on ξ, not necessarily coinciding with ζ0. Assume that we are given a positive Cartier divisor ζ1 on K; We will study adjoint curves restricting to K as ζ1 . Note that Jζ1,K = OK (−ζ1), and consider the restriction map resK : H 0(P2, Jζ1(a + i)) → H 0(K, OK (−ζ1)(a + i)) To introduce notations we need to recall some basic facts about linear equivalence of Cartier divisors. Assume that we are given two positive Cartier divisors D1 and D2 on a scheme X and a linear equivalence D1 − D2 = (r) for a meromorphic function r. We realize both OX (D1) and OX (D2) as subsheaves of the sheaf MX of meromorphic functions on X, and describe the isomorphism OX (D1) → OX (D2) given by the function r explicitly. Locally, on a small enough affine open set U ⊂ X, U ≃ Spec A, D1 and D2 are given by equations f1 and f2, fi ∈ A, f1/f2 = r in the full ring of fractions MA of A, and O(Di) is given by the A-submodule 1 A in MA, i = 1, 2. fi The isomorphism jr : 1 A, a/f1 7→ r · (a/f1) = a/f2 gives rise to an automorphism of the f1 sheaf MX given by the multiplication by r. Thus, globally, the sheaf automorphism jr : MX → MX given by the multiplication by r takes O(D1) to O(D2). A → 1 f2 Now, using the linear equivalence ζ0 ∼ al on K, we get an isomorphism jr : OK (−ζ1)(a + i) → OK (−ζ1)(ζ0)(i) ≃ OK (ζ0 − ζ1)(i) given by multiplication with the rational function r = f a l /fL0 where fl is an equation of a line l, and fL0 is the equation of L0 ∈ H 0(P2, Jζ0 (a)). Thus the image for an adjunction belongs to the sheaf OK (ζ0 − ζ1) ⊗ OK(i), which is the sheaf of meromorphic functions on K with zeroes at ζ1 and poles at ζ0, shifted by i. 1 + ζ res 1 , where ζ ξ Since we want to study adjoint curves to K, we are interested in positive Cartier divisors of the form ζ1 = ζ ξ (res for "residual") is supported on the set of smooth points of K. Note that the sections of the sheaf OK (ζ0 − ζ1) can locally be given by r = h1/h0 · g, where hi is the local equation for the Cartier divisor ζi, and g is regular, i.e., g ∈ OK,p. 1 is supported on ξ, i.e., supp(ζ ξ 1 ) = supp(ζ0) = ξ, and ζ res 1 Thus we introduce the following module and sheaf: Definition 4.17. For a commutative ring A, we define an A-submodule RA in the full ring of fractions MA, A ⊂ RA ⊂ MA, as the set of all fractions r = g1/g0 such that ordp(g1) ≥ ordp(g0) for each height one ideal p of A. Given a scheme X, one can define the sheaf RX ; this sheaf is the subsheaf of the sheaf of meromorphic functions MX given locally by fractions r = g1/g0 such that ordZ (r) = ordZ(g1) − ordZ(g0) ≥ 0 for each codimension one subvariety Z of X. The sheaf RX coincides with the structure sheaf OX at the set of smooth points of X, and there is a filtration Moreover, we have the following easy Lemma: OX ⊂ RX ⊂ MX . Lemma 4.18. The normalization NA of A in the full ring of fractions MA is a submodule of RA. I.e., there is a filtration A ⊂ NA ⊂ RA ⊂ MA Note that sheaf NX coincides with the sheaf π∗(OX ∗ ), the pushforward of the structure sheaf along the normalization X ∗ → X. ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 21 Combining this all together, we get an adjunction sequence aK,i,ζ1 : Jζ1,P(a + i) resK→ OK (−ζ1)(a + i) f a l fL ∼ → OK (ζ0 − ζ1)(i) = = OK (ζ0 − ζ ξ 1 )(i)(−ζ res 1 ) ⊂ OK (ζ0 − ζ ξ 1)(i) ⊂ RK(i) and, taking union over all positive Cartier divisors ζ1, we finally get our main adjunction (28) where aK,i : Jξ,P(a + i) r→ RK(i), r = f a l /fL. Now we study the image of the map aK,i. Definition 4.19. Let C be a plane curve. We say that a line l containing a cuspidal or nodal point p of C is strictly tangent to C at p if l intersects C with multiplicity 3 at p. We also say that a curve C1 containing p is strictly tangent to C at the nodal or cuspidal point p of C if C1 intersects C with multiplicity at least 3 at p. Assume from now on that the adjoint curve L0 to K is not (strictly) tangent to K at its singular points, and does not intersect K elsewhere. We want to introduce a sheaf of rational functions with denominator vanishing exactly along L0. This sheaf is clearly the image of the adjunction map aK defined above. Definition 4.20. Let RL0 r = f /fL0, where f is a homogeneous polynomial on P2, and fL0 is an equation of the curve L0. Proposition 4.21. If K is a nodal-cuspidal curve and L0 is an adjoint curve not tangent to K at the singularities of K and not intersecting it elsewhere, then the natural inclusion K be a subsheaf of RK consisting of sections r which can be given by RL0 K ⊂ RK is an equality. Moreover, they both coincide with the sheaf π∗OK ∗. Proof. The proof follows easily from the fact that nodal and cuspidal singularities of curves are resolved by a single blow-up, and, moreover, we can take t = f1/f0 or t = f1/fL0 as a local coordinate on the resolution, where f1 and f0 vanish at the singular points of K and have separated tangents to K at the singularities of K. In this way, both of the sheaves are equal to π∗OK ∗, and thus they coinside. (cid:3) Remark 4.22. This proposition is an example for the analytic theory of weakly holomorphic functions and universal denominator theorem (see, for example, [18]) in case our base field is the field of complex numbers. In this case the equation of the adjoint curve L0 works as the universal denominator for the sheaf of weakly holomorphic functions at each point of K. Combining the proposition above and the construction of the adjunction map aK,i (which is essentially a division by the equation of L0), we get the following theorem: Theorem 4.23. For a nodal-cuspidal curve K and an adjoint curve L0 as above, the map aK,i is epimorphic onto RK (i), and there is an exact sequence 0 / JK,P(a + i) / Jξ,P(a + i) aK,i / RL0 K (i) 0 RK (i) / / / / /  O   22 M. FRIEDMAN, M. LEYENSON In other words, adjoint curves of degree a + i to the curve K on the plane induce rational functions on the curve K for which ordp(r) ≥ 0 for each point p ∈ K. The map aK,i is an isomorphism modulo ideal spanned by the equation of K. Passing to the global sections for a + i < deg K, we get the following theorem: Theorem 4.24. For a + i < deg K, there are isomorphisms M H 0(P2, Jξ(a + i)) ∼ → M H 0(K, RK (i)) ∼ → M H 0(K ∗, OK ∗(i)) For higher degrees i ≥ deg K − a one can modify these isomorphisms readily to get a correct version including adjoint curves containing K as a component. Proof. This theorem follows immediately from Theorem 4.23 and Proposition 4.21 if we take into account the projection formula for π : K ∗ → K, π∗(OK ∗(i)) ≃ π∗(OK ∗ ⊗ π∗OK(i)) ≃ π∗(OK ∗) ⊗ OK(i) ≃ RK ⊗ OK (i). The meaning of the theorem is that plane curves through ξ exactly correspond to homogeneous (cid:3) functions on K ∗. Remark 4.25. (Graded algebras interpretation) Assume we are given a smooth space curve K ∗ not contained in a plane in P3 and a projection p : K ∗ → K to a plane curve K. Since K ∗ is birational to K, in order to reconstruct K ∗ from K, we have to say what is the "vertical coordinate z" on K ∗ in terms of K. Since K ∗ and K are birational, the regular (holomorphic) objects on K ∗ are rational (meromorphic) objects on K, and thus we should have an equality of the form z = fn+1/fn for some integer n and plane curves fn and fn+1 of degrees n and n + 1. More precisely, let S = ⊕Si, Si = H 0(K, O(i)) be the graded algebra of homogeneous functions on K, and T be the graded algebra of homogeneous functions on K ∗. The inclusion S → T gives an isomorphism of fraction fields Q(S) → Q(T ), since K and K ∗ are birational. Now T1 = S1 ⊕ kz for some element ("vertical coordinate") z ∈ T1; since T1 ⊂ Q(T ) ≃ Q(S), we would have z = fn+1 fn for some integer n and plane curves fn and fn+1 of degrees n and n + 1, both passing through the singularities of K. Corollary 4.26. As in the previous remark, assume that we are given a smooth space curve K ∗, a projection p : P3 → P2 with center O not on K ∗ such that K = p(K ∗) is a nodal-cuspidal curve, and an adjoint curve L0 of degree a to K which is smooth at the singularities of K and is not (strictly) tangent to K there. Then the "vertical coordinate" z on K ∗, z ∈ H 0(K ∗, OK ∗(1)), is the image of a uniquely defined plane curve L1 of degree a + 1 under the adjunction map aK,1 defined by the formula (28). In other words, we can choose n = a in the remark above, and z = fL1 fL0 , where fC is an equation of a plane curve C, C = L0 or L1, deg L1 = a + 1, and the curve L1 is not a union of L0 and a line, i.e., is a "new" adjoint curve. The curves L0 and L1 are smooth at the points of ξ and have different tangents at every point p ∈ ξ. Proof. There are two ways to prove it. First, this statement is a corollary of the theorem 4.24. The fact L1 is "new", i.e., not a union of L0 and a line, follows from the fact that z is "new", i.e., does not come from a linear form on P2 (explicitly, z ∈ H 0(P3, O(1)) ≃ H 0(K ∗, OK ∗(1))). The fact that ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 23 L1 is smooth at the singularities of K follows from the fact that the fraction z = fL1/fL0 resolves the singularities of K. A more straightforward proof is the following: let S be the graded homogeneous algebra of K and T be the graded homogeneous algebra of K ∗; and consider the element t = z · fL0 of Ta+1. It is enough to prove that t actually belongs to Sa+1, since then we can let fa+1 = t and z = fa+1/fL0. Now this is an easy local computation for each singular point of K, since the exact sequence 0 → Sa+1 → Ta+1 → Ta+1/Sa+1 → 0 is obtained from the sheaf exact sequence 0 → OK (a + 1) → p∗OK ∗(a + 1) → F (a + 1) → 0, where F is by definition the factor sheaf p∗OK ∗/OK , by passing to global sections: 0 → H 0(K, OK (a + 1)) p∗ → H 0(K ∗, OK ∗(a + 1)) → coker p∗ → 0 Since the factorsheaf F is a product of sheaves supported at singular points of K, this makes computing the image of t in H 0(K, F (a + 1))) an easy local computation at nodes and cusps. The intuitive meaning of this computation is that fL0 vanishes at the singularities of K, which (cid:3) implies that t = zfL0 is a regular (holomorphic) object on K, and thus belongs to Sa+1. In particular, this is the case when K = B is a branch curve of a smooth surface S in P3, where ξ is the 0 -- cycle of singularities of K. In this case we can take L = L0, a = (ν − 1)(ν − 2), where ν = deg S. Segre refers to the existence of the second adjoint curve L1 as something known from the Cayley's "monoıde construction" (see [3, pg. 278]). Remark 4.27. Summarizing what is written above, the branch curve B has an adjoint curve L of degree equal to a. In this case, we have z = fL1 fL The curves L and L1 are smooth at the points of ξ = P + Q and have separated tangents at every point p ∈ ξ. Remark 4.28. Note that if the plane nodal-cuspidal curve K has two adjoint curves of degrees n and n + 1 with separated tangents at Sing K for any integer n, then K is the image of a smooth space curve K ∗ under the projection from P3, but it is only n = a = (ν − 1)(ν − 2) that K may actually be a branch curve of a surface projection. Remark 4.29. We the following isomorphisms: H 0(P2, Jξ(a + 1)) ≃ H 0(K ∗, OK ∗(1)) ≃ H 0(P3, O(1)). I.e., linear forms on K ∗ correspond to adjoint curves of degree equal to a + 1 on K. Example 4.30. For a cubic surface f = z3 − 3az + b the branch curve B = b2 − 4a3. The six cusps of B are given by the intersection of a conic and a cubic (a = b = 0), and in this case L = a is a conic in general position to B at the cusps, the cubic W = b is strictly tangent to B at the cusps (see definition 4.19), and both of them do not intersect B elsewhere. We claim that L1 = W in this case. Indeed, we have on B ∗ f = z3 − 3az + b = 0, f ′ = 3(z2 − a) = 0 and thus on B ∗. It follows that L1 is given by b. z = 1 2 b a 24 M. FRIEDMAN, M. LEYENSON Remark 4.31. In the previous example we can choose the curve L1 as any of the curves W + l0L, where l0 is a linear form on P (perhaps 0). An easy computation shows that L1 is strictly tangent to K at q ∈ Q iff l0 contains the point q (or if l0 = 0), but even in this case l0 the curves L and L1 have different tangents at q. 4.4. Segre's theorem. Consider again a smooth surface S in P3 and a projection π : S → P2 with a center O ∈ P3 − S. Let B be the branch curve of p, and ξ be the 0-cycle of singularities of B. Consider now the graded vector space ⊕H 0(P2, Jξ(n)). It follows from the Segre's computation that a = (ν − 1)(ν − 2) is the smallest integer such that there are adjoint curves of degree a to B. The vector space H 0(P2, Jξ(a)) is one-dimensional and generated by the the curve L. Let ζL = LB be the corresponding divisor class in Pic(B). Note that for n = a the class ζL gives a canonical lifting of 2ξ = 2P + 2Q to Pic B, and thus H 0(P2, Jξ(a)) ≃ H 0(P2, Jζ (a)). We have (29) (30) (31) (32) ζL ∈ al, [ζL] = 2ξ in A0(B), k = kL ∼ → H 0(P2, JζL(a)) ∼ → H 0(P2, Jξ(a)), H 0(P2, JζL(a)) ≃ H 0(B, OB(−ζL)(a)) ≃ H 0(B, OB) Now L is smooth at the points of ξ and is not strictly tangent to B at these points by Remark 4.27, and thus ζL is given by a tangent vector to p at each point p ∈ ξ, which follows from the descriptio of Cartier divisors supported at nodes and cusps. The picture for the branch curve of a smooth cubic surface is drawn below. Segre proves that this data is sufficient to reconstruct the surface S: Figure 5 : Cartier divisor ζL Theorem 4.32 (Segre). A plane curve B of degree d = ν(ν − 1) is a branch curve of a smooth surface of degree ν in P3 if and only if (1) B has n = 1 (2) B has c = ν(ν − 1)(ν − 2) cusps; (3) There are two curves, L of degree a = (ν − 1)(ν − 2) and L1 of degree a + 1, which both 2 ν(ν − 1)(ν − 2)(ν − 3) nodes; contain the 0-cycle ξ of singularities of B and have separated tangents at the points of ξ. Proof. The necessity of these conditions was proved in the preceding sections. We now prove that they are sufficient. Let B be such a curve in the plane P2. First, since L is adjoint to B, the 0-cycle associated with the scheme-theoretic intersection L ∩ B contains 2ξ = 2P + 2Q, but by conditions of the theorem deg B · deg L = 2 deg ξ = ν(ν − 1)2(ν − 2) ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 25 It follows that the 0-cycle associated with L ∩ B is [L ∩ B] = 2P + 2Q. Let us denote ξ = P + Q. It follows immediately that 2ξ is in the linear system aℓ on B, where ℓ is the linear system associated with the given plane embedding of B. In particular, we conclude that ξ ∈ (cid:12)(cid:12)(cid:12)(cid:12) 1 2 a · ℓ(cid:12)(cid:12)(cid:12)(cid:12) Note also that [L1 ∩ B] = 2P + 2Q + R, where deg R = d = ν(ν − 1). Now the space H 0(P2, Jξ(a + 1))) contains a 4 -- dimensional subspace of the form kf1 + kxf + kyf + kwf , where f1 is the equation of L1 and f is the equation of L. (Recall that k is our base field.) Now consider the linear system on B given by restriction of (f1, xf, yf, wf ) = kL1⊕H 0(P2, O(1)) ⊗ kL. It has ξ as a set of base points. It follows that it defines a rational map φ : B − ξ → P3. Let π : B ∗ → B be the normalization of B. We claim that the rational map φ can be lifted to give a regular map φ∗ : B ∗ → P3. Indeed, we have the following lemma: Lemma (A). Let B be a plane nodal-cuspidal curve with the set of singularities ξ, and let f ∈ H 0(B, Jξ(j)) and f1 ∈ H 0(B, Jξ(j + 1)) be non-zero elements determining adjoint curves C = Z(f ) and C1 = Z(f1) on the plane, such that TpC 6= TpC1 at any point p ∈ ξ. Let Then the rational map φΩ : B 99K P3 can be resolved as Ω = kf1 ⊕ H 0(P2, O(1)) ⊗ kf = (f1, xf, yf, wf ). B ∗ π B P3 pr / P2 where π : B ∗ → B is the normalization of B. Note that TpC 6= TpC1 implies that f1 /∈ H 0(P2, O(1)) ⊗ kf , and also that Ω → TpC is epimorphic at every point p ∈ ξ. Proof. It is clear that we only have to verify the statement at nodes and cusps of B as well as smooth points p on B such that f1(p) = f (p) = 0. For a node p we can choose coordinates in the local ring of P2 at p such that B is given by the equation xy = 0. Assume that f1 is given by the equation a1,0x + a0,1y + (order 2 terms), and f is given by the equation b1,0x + b0,1y + (order 2 terms). Note that φΩ = (f1, f x, f y, f w) = (f1/f, x, y, w). One can easily see that φΩ maps the point p on the branch (y = 0) of B to a1,0/b1,0, and the same point on the branch (x = 0) to a0,1/b0,1. Thus, if a1,0b0,1 − a0,1b1,0 6= 0, then φW can be lifted to a regular map B ∗ → P3 with a smooth image in the neighborhood of p. In the same way, in a neighborhood of a cups B can be given by the local equation y2 − x3 = 0, and thus f1/f = a1,0x + a0,1y + (order 2) b1,0x + b0,1y + (order 2) = a1,0 + a0,1t + (order 2) b1,0 + b0,1t + (order 2) , where t = y/x is the coordinate on the exceptional divisor in the resolution of the cusp. Now it is clear that if a1,0/b1,0 6= a0,1/b0,1, then φW lifts to an embedding of the exceptional divisor and thus the normalization of the curve as well. / /        / 26 M. FRIEDMAN, M. LEYENSON If now p is a smooth point of B such that f1(p) = f (p) = 0, then it is a standard fact that the map (B − p) → P3 can be uniquely extended to the map B → P3 in a neighborhood of the point p, since P3 is proper. (Note also that we do not have any such points in the application of this Lemma below, due to the intersection multiplicity computation for C1 and C.) (cid:3) This gives a non-singular model C ⊂ P3, and a projection π : C → B with some center O. Note that if we start from a given ramification curve B ∗, the curve we reconstruct from B coinsides with B ∗. Lemma (B). If B is a branch curve of the generic projection π : S → P2, where S is a smooth surface in P(V ) ≃ P3, and B ∗ is the ramification curve of π, then there is an isomorphism P(V ) → P(kL1 ⊕ H 0(P2, O(1)) ⊗ kL) which takes B ∗ ⊂ P(V ) to C. In other words, the linear system (f1, xf, yf, wf ) reconstructs the curve B ∗. The idea of the proof is, as in the previous lemma, to set z = f1/f on B ∗. Recall that preimages of the nodes of B belong to the bisecant lines to B ∗ containing ithe point O, and preimages of cusps belong to the tangent lines to B ∗ containing the point O. Considering tangent lines to B ∗ as a limiting case of bisecants to B ∗, we see that B ∗ has n + c = 1 2 ν(ν − 1)2(ν − 2) of bisecants (and tangents) containing the point O, which belong to a cone of order (ν − 1)(ν − 2) above L with vertex O. Lemma (C). B ∗ does not belong to a surface of degree m < ν − 1. Proof. Assume that S1 is such a surface of degree m; we can assume that it is irreducible. Consider S ′ 1 = PolO(S1). First, if S1 is smooth, note that S ′ 1 contains the preimage of the 0-cycle of cusps Q∗, since at each point q ∈ Q∗, the tangent line l to B ∗ is contained in TqS1, and also l contains O, since q projects to a cusp of B. It follows that q ∈ S1 ∩ S ′ 1. Secondly, if S1 is not smooth, then S ′ 1 still contains q. However, then it follows that the number of cusps c ≤ ν(ν − 1) · (m − 1), which contradicts to (cid:3) assumption that c = ν(ν − 1)(ν − 2). We now have to prove that the model B ∗ we constructed is a complete intersection of a surface S of degree ν and its polar PolO(S) of degree ν − 1 with respect to the (fixed) point O which is the center of the projection π : B ∗ → B. For these, following Segre, we apply the following theorem belonging to Halphen (See [3, pg. 359]): Theorem (Halphen). Let C be a space curve of order a·b in P3 s.t. a < b which has 1 2 a(a−1)b(b−1) bisecants all lying on a cone of degree (a − 1) · (b − 1). Assume also that C is not on a surface of degree smaller than a. Then C is a complete intersection of two surfaces of degree a and b. The inverse statement to the Halphen's theorem is easy; see [1, art. 343] or [15, Chapter IX, sections 1.1, 1.2]. Alternatively, instead of invoking Halphen's theorem, one can invoke a theory of Gruson and Peskine, as it is done by D'Almeida in [41]; we cite his reasoning for the convenience of the reader: Lemma (D). [41, pg. 231] The curve B ∗ constructed above is a complete intersection of two surfaces of degrees ν and ν − 1. Proof. To prove the lemma, we introduce first the following definition: Definition 4.33. Given a space curve C, we define its index of speciality as s(C) = max{n : h1(C, OC (n)) 6= 0}. ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 27 Now we state the following Speciality Theorem of Gruson and Peskine [27]: Let C be an integral curve in P3 of degree d, not contained in a surface of degree less than t. Let s = s(C). Then s ≤ t + d t − 4, with equality holding if and only if C is a complete intersection of type (t, d t ) (and thus OC(s) is special, i.e., h1(OC(s)) 6= 0). Let now p : B ∗ → B be the projection from the point O. The conductor of the structure sheaf OB∗ in OB is Ann(p∗OB∗ /OB), which by duality is isomorphic to Ann(ωB/p∗(ωB∗ )) (see e.g. [19, Chap- ter 8]). By the definition of the conductor, we get that Ann(ωB/p∗(ωB∗)) = Hom(ωB, p∗(ωB∗)) = p∗(ωB∗ ) ⊗ ω B. It is well known that for a nodal-cuspidal curve, H is a global section of the con- ductor sheaf iff H passes through the nodes and the cusps of the curve (see e.g. [32, Proposition 3.1]). ∨ By Serre duality, for all i, H 1(OB∗ (i)) = H 0(ωB∗(−i)). Thus, the minimal degree of the curve containing the singular points of B is ν(ν − 1) − 3 − s(B ∗). Indeed, for a curve to pass through the singular points of B, the conductor has to have sections, i.e. p∗(ωB∗) ⊗(ωB) has sections. Since we know that the minimal degree of the curve containing the singular points of B is (ν − 1)(ν − 2), we get s(B ∗) = 2ν − 5. ∨ As B ∗ does not lie on any surface of degree ν − 2 (by Lemma (C)), then the Speciality The- orem shows that B ∗ is a complete intersection of two surfaces of degrees ν and ν − 1 (taking t = ν − 2, d = ν(ν − 1)). (cid:3) Either way, by results of Halphen or Gruson-Peskine, the curve B ∗ is a complete intersection of two surfaces, say, Sν and F ν−1 of degrees ν and ν − 1. We still have to prove that B ∗ can be written as an intersection of a surface of degree ν and its polar with respect to the given point O. Let W = H 0(P3, JB∗ (ν)) be the linear system of surfaces of degree ν containing B ∗, W = kS ⊕(cid:0)H 0(P3, O(1)) ⊗ kF(cid:1) , as for any complete intersection of type (ν, ν − 1). For a point t ∈ PW , let St be the corresponding surface of degree ν containing B ∗. (here we also denoted by S and F some particular equations for the surfaces S and F , even though they are defined only up to Gm action). Consider now the linear map ∂O : W = H 0(P3, JB∗ (ν)) → H 0(P3, O(ν − 1)), which maps f to P olOf = P Oi∂if , its polar with respect to the fixed point O. We claim that ∂0 is injective. Indeed, if ∂0(f ) = 0, then f vanishes on a cone of degree ν, containing the curve B ∗. Note that F ν−1 vanishes on B ∗ but also gives a degree ν − 1 form on every line generator of the cone (f = 0), which implies that the projection map B ∗ → B has degree ν − 1, which is not the case. Now, for every t ∈ P(W ) and the corresponding surface St of degree ν, consider the triple intersection ηt = St ∩ F ν−1 ∩ PolO St First, we have St ∩ F = B ∗. Let Rt = St ∩ PolO St. Rt is a ramification curve for the surface St with respect to the projection with the given center O. We have ηt = B ∗ ∩ Rt. Note that Q∗ ⊂ Rt for every t, since B ∗ belongs to St and has all tangent lines at the points of Q∗ contain the projection center O. 28 M. FRIEDMAN, M. LEYENSON Thus for every t either the polar surface PolO St contains the curve B ∗, or we have a decompo- sition of 0-cycles on B ∗ of the form ηt = Q∗ + rt Also note that Q∗ ⊆ B ∗ ∩ PolO F by the same geometric argument, i.e., since at the points of Q∗ the tangent lines to B ∗ contain the projection center O, these points are on the intersection of B ∗ with O-polar of every surface containing B ∗. But since these two 0-cycles have the same degree, they coinside. It follows that Q∗ ∈ (ν − 2)h on the curve B ∗, where h is a class of hyperplane section. Now, since ηt ∈ ∂0St = (ν − 1)h on B ∗, we have rt ∈ h on B ∗ whenever PolO St intersects non-trivially with the curve B ∗, i.e., does not contain it. Since B ∗ is complete intersection, it is linearly normal (which follows easily from the cosideration of Koszul complex). It follows that rt gives a map W → H 0(B ∗, O(1)) from the 5-dimensional space W to the 4-dimensional vector space H 0(B ∗, O(1)) ≃ H 0(P3, O(1)). Such a map must have a kernel, and let S0 be the corresponding surface in the linear system W . It follows that Pol(O, S0) contains the curve B ∗, and thus B ∗ = S0 ∩ Pol(O, S0), i.e., B ∗ is a ramification curve for the projection of the surface S0 to P2 with the given center O. This finishes the proof. (cid:3) Remark 4.34. We generalize Segre's theory for smooth surfaces in PN , N > 3, in the subsequent paper [67]. Let us notice that the 0 -- cycle of singularities of the branch curve B is special. We would like to emphasize this in the next subsection. 4.5. Special 0-cycles. Let ξ be a 0-cycle in P2. Define the superabundance of ξ (relative to degree n curves) as: δ(ξ, n) = h1Jξ(n) We have the following Lemma 4.35. If deg ξ ≤ dim nh, then in other words, δ(ξ, n) is the speciality index of the 0-cycle ξ with respect to the linear system nh. dim nh − ξ = (dim nh − deg ξ) + δ(ξ, n), Also note that δ(ξ, n + 1) ≤ δ(ξ, n). Let now ξ = P + Q - the zero cycle of singularities of B, and, as before, a = (ν − 1)(ν − 2). Proposition 4.36. (Speciality index of ξ) There are following identities for the speciality index of ξ: δ(ξ, a) = δ(ξ, a + 1) = 1 2 1 2 (ν − 1)(ν − 2)(2ν − 5) (ν − 3)(2ν2 − 7ν + 4) In particular, the 0-cycle ξ is special with respect to ah for all surfaces of degree at least 3, and special with respect to (a + 1)h for all surfaces of degree at least 4. ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 29 Proof. For the expected dimension vdim Jξ(a) we have vdim Jξ(a) = dim ah − deg ξ = 1 2 a(a + 3) − 1 2 ν(ν − 1)2(ν − 2) Since a = (ν − 1)(ν − 2), we get vdim Jξ(a) = 1 2 (ν − 1)(ν − 2)(5 − 2ν) Since, by definition of speciality index, dim Jξ(d) = vdim Jξ(d) + δ(ξ, d) and since Jξ(a) = {L}, we get the first equality. The proof of the second formula is parallel; we use isomorphism Jξ(a + 1) ≃ PH 0(P3, O(1)) (see (cid:3) Remark 4.29). Example 4.37 (6-cuspidal sextic). Let ξ6 be a 0-cycle of degree 6 on a plane which is an intersec- tion of conic and cubic curves in P2, given by a degree 2 (resp. 3) polynomial f2 (resp. f3). Note that generic 0-cycle of degree 6 is not like this, because generic 6 points do not belong to a conic. Note that for ξ6 given by (f2, f3) there is a Koszul resolution 0 → OP2(n − 5) " f3 f2 # → OP2(n − 2) ⊕ OP2(n − 3) [−f2 f3] → Jξ6(n) → 0 An easy computation shows that δ(ξ6, 2) = 1, δ(ξ6, 3) = 0, δ(ξ6, 4) = 0 (see Subsection 4.5 for the definition of δ(·, ·)), and that H 0Jξ6(2) = kf2, H 0Jξ6(3) = kf3 + kxf2 + kyf2 + kwf2. Note that we start the computation from n = 2, since δ(ξ, 1) = 3 is not a defect w.r.t. the linear system. Also note that For a generic 0-cycle ξ of degree 6, δ(ξ, 2) = 0, otherwise it would lie on conic. 4.6. Dimension of B(d, c, n). In this subsection, let d(ν) = ν(ν − 1), c(ν) = ν(ν − 1)(ν − 2), n(ν) = 1 2 ν(ν − 1)(ν − 2)(ν − 3). Motivated by Segre's theory and the Chisini conjecture, We want to compute the dimension of the component B3(ν) of B(d(ν), c(ν), n(ν)) which consists of branch curves of smooth surfaces in P3 of degree ν with respect to generic projection. Let S(ν) be the variety parameterizes smooth surfaces in P3 of degree ν. It is well known that dim S(ν) = 1 6 (ν + 1)(ν + 2)(ν + 3) − 1. Let B ∈ B3(ν), a branch curve in the plane Π of a smooth surface S in P3 of degree ν, when projected from the point O = (0 : 0 : 0 : 1) (we work with the coordinates (x : y : w : z)). Now, B is also the branch curve of a smooth surface S ′ iff there is a linear transformation in P GL4(C) that fixes that point O, fixes the plane Π (with coordinates (x : y : w)) and takes S to S ′. It is easy to see that the dimension of this subgroup of transformations G is 5 (in GL4(C)), but as we are in a projective space, dim P (G) = 4. By the Chisini's conjecture (proven completely for a generic projection, see [65]), the branch curve B determines the surface uniquely up to an action of P (G). Thus, (33) dim S(ν) − 4 = dim B3(ν). Denote by V (ν) = 1 curves with n(ν) nodes and c(ν) cusps. 2 d(ν)(d(ν) + 3) − n(ν) − 2c(ν) the virtual dimension of a family of degree d(ν) Example 4.38. (1) For ν = 3, 4 , dim B3(ν) = S(ν) − 4 = V (ν), as expected (as for these branch curves, c(ν) < 3d(ν). See [22, p. 219]). (2) For ν ≥ 5, dim B3(ν) = S(ν) − 4 > V (ν). This gives examples of nodal cuspidal curves, whose characteristic linear series is incomplete (for other examples see e.g. Wahl [24]). 30 M. FRIEDMAN, M. LEYENSON 4.7. Projecting surfaces with ordinary singularities. We bring here a short subsection on surfaces in P3 with ordinary singularities, as we use it in the next section, where we classify branch curves of small degree. The generalization of Segre's theory for these surfaces will be presented in [67]. It is classical that (see e.g. [28]) any projective surface in characteristics 0 can be embedded in P3 in such a way that its image has at most so-called ordinary singularities, i.e., a double curve with some triple and pinch points on it. Any projection S ⊂ Pn → P2 can be factorized then as a composition of projections S ⊂ Pn → P3 → P2 such that the image S1 of S in P3 has ordinary singularities in P3. However, if we project S to P3 first, and then from P3 to P2, we get an extra component of the branch curve, which would be the image of the double curve. Assume now that we are given a degree ν surface S ⊂ P3 = P(V ) with ordinary singularities and a point O not on S. Let E∗ be the double curve of S. Consider the projection map π : S → P(V /lO) ≃ P2. We define the ramification curve B ∗ of the projection as an intersection of S and the polar surface S ′ O is the support of the sheaf Ω1 O. (To justify this definition, one can check that S ∩ S ′ S/P2.) One can now see that B ∗ can be decomposed as B ∗ = B ∗ res + F ∗, where [F ∗] = 2[E∗], when [F ∗] is the Weil divisor associated with the 1-dimensional Cartier divisor 2[E∗]. Note that B ∗ res in its intersection with the smooth locus of S is set-theoretically the set of smooth points p on S such that the tangent plane Tp(S) contains O. (To be more careful, B ∗ res is the scheme-theoretical support of the kernel sheaf of the canonical map Ω1 S/P2 → 0, where i is the embedding of F ∗ to S. For a different scheme-theoretic description of E∗ and B ∗ res, see [26, Section 2]). S/P2 → i∗i∗Ω1 It follows that the branch curve B can also be decomposed as B = Bres + 2E, where E is the image of E∗. Let e = deg E∗ and d = deg B ∗ res = ν(ν − 1) − 2e. Now a generic hyperplane section of S, S ∩ H, is a plane curve of degree ν with nodes at the finite set E∗ ∩ H, and thus there is a restriction 0 ≤ e ≤ (ν − 1)(ν − 2) 2 , since the number of nodes of a plane curve can not exceed its arithmetic genus. It follows that the pair (ν, d) satisfies 2(ν − 1) ≤ d ≤ ν(ν − 1), as illustrated on Figure 6 below. ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 31 Figure 6 : Geography of surfaces in P3 with a double curve. We examine in Subsection 4.7.1 the cases where ν = 3, 4. What is important here is that for a given d there is only a finite number of possible ν's such that a plane curve C of degree d can be a pure branch curve of degree ν surface in P3 with ordinary singularities. As before, we define Q∗ to be an intersection of B ∗ and the second polar surface S ′′ O, i.e., as an O. However, for a singular surface S not all points of Q∗ form cusps on intersection of S, S ′ the branch curve. This is shown, for example, at [15, Chapter IX, section 3.1]. O and S ′′ Notation 4.39. Denote by v∗ ∈ E∗ a point, such that the tangent plane to S at v∗ contains the center of projection O. These points are called vertical points (or points of immersion) and we denote the set of such points as V ∗. Denote by T ∗ the set of triple points of E∗, and by t the number of these points. Let also P i∗ be the set of pinch points of E∗ and let p be the number of these points. Remark 4.40. Note that the number of pinch points p is always positive (see [30]). We will use this fact to prove the inexistence of branch curves in V (8, 12, 0) in Section 5. The following Lemma is proved at [15, Chapter IX, sections 3.1, 3.2]. This Lemma is the base for generalizing Segre's theory for singular surfaces, a generalization which will be presented in [67]. Lemma 4.41. (1) Q∗ = S ′′ O ∩ B ∗ can be decomposed as Q∗ = (S ′′ O ∩ 2E∗) + Q∗ res Note that the images of (S ′′ O ∩ E∗) under the projection are smooth points on Bres. (2) points in B ∗ res ∩ E∗ do not form cusps of the branch curve, i.e., their images are smooth points on Bres. Explicitly, (3) S ′′ O ∩ E∗ can be decomposed as B ∗ res ∩ E∗ = P i∗ + V ∗. S ′′ O ∩ E∗ = V ∗ + 3T ∗ and S ′′ O ∩ B ∗ res can be decomposed as S ′′ O ∩ B ∗ res = V ∗ + Q∗ res 32 M. FRIEDMAN, M. LEYENSON Remark 4.42. Denote by e∗ the degree of E∨ the dual curve of E in P2. Given a surface S in P3, we can express the number of nodes and cusps of its branch curve Bres by terms of ν, e, e∗ and t. The following result is proved at [15, Chapter IX, section 3]: c = ν(ν − 1)(ν − 2) − 3e(ν − 2) + 3t, n = 1 2 ν(ν − 1)(ν − 2)(ν − 3) − 2e(ν − 2)(ν − 3) − 2e∗ − 12t + 2e(e − 1). Remark 4.43. Let u be the number of components of E∗, and g = Pu i=1 gi the geometric genus 1, c2 and the number of pinch points p by terms of of E∗. By [28, pp. 624, 628] we can express c2 ν, e, t and (g − u): c2 1 = ν(ν − 4)2 − 5νe + 24e + 4(g − u) + 9t, c2 = ν2(ν − 4) + 6ν + 24e − 7νe + 8(g − u) + 15t, p = 2e(ν − 4) − 4(g − u) − 6t. 4.7.1. Examples. We survey the well known examples of surfaces of degree 3 and 4 in P3 with ordi- nary singularities and use the results from Remarks 3.12 and 4.42 in order to calculate the number of nodes and cusps of the branch curve Bres of the surface S. These numbers can be expressed in terms of c2 1(S), c2(S), deg(S) and deg(Bres) or in terms of ν, e, e∗ and t. Degree 3 surfaces We know from the inequality above that 0 ≤ deg E∗ = e ≤ 1, in other words, the only cubic surfaces with ordinary singularities are those with double line. (1) e = 0. This is a smooth cubic surface, with the branch curve B being a 6-cuspidal sextic. (2) e = 1. Such a surface has a double line, and thus d = deg Bres = 4. Since we consider only generic projections, we can choose coordinates (x, y, w, z) in P3 in such a way that the projection center O = (0, 0, 0, 1) and the double line E∗ = l∗ is given by equations (z = w = 0). In these coordinates the projection is given by the (rational) map (x, y, w, z) 7→ (x, y, w), and E = l is the "line at infinity" (w = 0) in the "horizontal" plane (z = 0). It is easy to see that such a cubic surface can be given by a degree 3 form f = z3 + a1z2 + b1wz + c1w2, where (a1, b1, c1) are homogeneous forms in (x, y) of degree 1. One can see from the definition of the normal cone ([48]) that the normal cone to l∗ in S is given by the degree 2 part of f in (z, w), i.e., by the form We can consider [f ]2 as a section of O(1, 2) on the ruled surface [f ]2 = a1z2 + b1wz + c1w2 P Nl∗/P3 ≃ l∗ × P1. Note that [f ]2, being a quadratic form of the variables (z, w) with coefficients in k[x, y], degenerates in the zeroes of its discriminant ∆([f ]2) = b2 1 − 4a1c1. It follows that there are 2 points p1 and p2 on l∗ where this quadratic form degenerates into a double line, which proves that a cubic surface with a double line has 2 pinch points. Note also that B ∗ res ∩ l∗ consists of such points p on l∗ such that one of the normal lines to l∗ in S at p is the "vertical" line (one that contains the point O): it is the only point of immersion. In the normal plane to the line l∗ with coordinates (z, w) this vertical line is given by the equation (w = 0). It follows that such points p are exactly those where a1 ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 33 vanishes. This gives just one point p0, different from the two pinch points p1 and p2 defined above, and a decomposition S ′′ O ∩ B ∗ res = p0 + Q∗ res O ∩ B ∗ res) = deg B ∗ We have deg(S ′′ res = 3. It follows that the pure branch curve Bres has 3 cusps. Note also that Bres has no nodes, since a plane quartic with 3 cusps is rational and can not have any other singularities; i.e., we obtain a point [Bres] in B(4, 3, 0). res = 6 − 2 = 4, and thus deg Q∗ Degree 4 surfaces We should have 0 ≤ e = deg E∗ ≤ 3. (1) e = 0. This is the case of a smooth quartic surface with degree 12 branch curve, which belongs to B(12, 24, 12). (2) e = 1. Let S be a quartic surface with a double line l∗. We have B ∗ = 2l∗ + B ∗ res, where d = deg B ∗ res = 4 · 3 − 2 = 10. Arguing as above we can see that the normal cone to l∗ in S can be given by the equation a2z2 + b2wz + c2w2 = 0 for some homogeneous forms (a2, b2, c2) of degree 2 of variables (x, y). It follows that S has 4 pinch points on the line l∗, and the intersection of l∗, B ∗ res and S ′′ O consists of two (different) points p1, p2 which are the points of immersion. It follows that res = p1 + p2 + Q∗ res, S ′′ O ∩ B ∗ O ∩ B ∗ res) − 2 = 18. where deg Q∗ res = deg(S ′′ It is known that a quartic surface with a double line is the image of P2 blown up at 9 points (see [28, pg. 632]). Computing its Chern invariants c2 1 and c2 and using the formulas from Remark 3.12, one can check that the number of cusps is indeed 18, and the number of nodes is 8. Thus [Bres] ∈ B(10, 18, 8). Alternatively, since e∗ = t = 0, by remark 4.42 we find out that indeed [Bres] ∈ B(10, 18, 8). Remark 4.44. It is easy to see from the above the classical fact that for a singular surface of degree ν in P3 with a double line, the number of pinch points is p = 2(ν − 2) and the number of the vertical points is ν − 2. (3) e = 2. In this case we have deg E∗ = 2. A curve of degree 2 in P3 is either a smooth conic contained in a plane, or a union of two skew lines, or a union of two intersecting lines, or a double line. By definition, the last two curves can not be double curves of a surface with ordinary singularities. Both of the two remaining cases are actually realized, as explained, for example, in [28]. If the double curve is a smooth conic, then it is classical that the surface S is a projection to P3 of the intersection of two quadrics in P4, (cf. [28]), and one can check (using remark 3.12 or 4.42) that the branch curve is in B(8, 12, 4). If the double curve is a union of 2 skew lines, then it is known that S is a ruled surface over elliptic curve (cf, say, [28], who deduces it from the classification of surfaces with q = 1). 1(S) = 0, and, using the From this classification one can conclude now that c2(S) = c2 same formulas as before, that the branch curve Bres gives a point in B(8, 12, 8). (4) e = 3. A double curve E∗ of a surface S with ordinary singularities is either smooth, or has some triple points. Thus E∗ can be either (a) a rational space cubic, or (b) a non-singular plane cubic, or (c) a union of a conic and a non-intersecting line, or (d) a union of 3 skew 34 M. FRIEDMAN, M. LEYENSON lines, or (e) union of 3 lines intersecting in a point. It is explained, for example, in [28], that only cases (a) and (e) are realized. In the case (a) the surface S is the projection of P1 × P1 embedded with the linear system ℓ1 + 2ℓ2 to P5, and the branch curve is in B(6, 6, 4). We discuss this case in details in Subsection 5. In the case (e) the surface S is the projection of the 2-Veronese-embedded P2 in P5, and the branch curve is in B(6, 9, 0); see the discussion in Subsection 5. 5. Classification of singular branch curves in small degrees For B a smooth curve of even degree d defined by the equation {fB = 0}, let π : S → P2 a degree ν cover ramified over B which is generic in sense of Subsection 2.3. We have that d is even (see Remark 3.8). By Zariski-Van Kampen theorem [11], π1(P2 − B) ≃ Z/dZ which is abelian. Since the monodromy representation of a generic cover into the symmetric group should be epimorphic (see the proof of Lemma 3.4), we conclude that π is of degree 2, i.e., isomorphic to the double cover given by z2 = fB in the total space of the line bundle OP2(d/2). Remark 5.1. As was stated in subsection 2.3, we study generic linear projections p : PN = P(V ) → P(V /W ) (where W ⊂ V be a codimension 3 linear subspace) where P(W ) ∩ S = ∅. Explicitly, if S is a surface in P3, then the projection is from a point O 6∈ S. However, see remark 5.7. All possible non-smooth branch curves of degrees 4 and 6 are known and we list them in the next paragraphs. For each case we give examples (and sometimes complete classification) of coverings with a given branch curve. We then give all the numerical possible singular degree 8 branch curves (see Theorem 5.5). We denote ha, bi = (aba)(bab)−1, ab = bab−1, and for the rest of this section we will use the coordinates (d, c, n) in the variety of the nodal -- cuspidal curves. Degree 4 singular branch curves There is only one branch curve of degree 4, as the following has to be satisfied: 4n, 3c, and the geometric genus g(B) = (d − 1)(d − 2)/2 − n − c ≥ 0. It follows that the only possibility is (c = 3, n = 0). This unique curve is the famous complexification of the classical deltoid curve, which is a cycloid with 3 cusps, i.e., the trace of a point on a circle of radius 1/3 rotating within a circle of radius 1. It is not hard to show that all other curves in V (4, 3, 0) are obtained from the deltoid by linear transformation, since the dual curve belongs to V (3, 0, 1), which is an irreducible space. Zariski computed the braid monodromy for a deltoid using elliptic curves [6] and proved that π1(P2 − B) is isomorphic to the group with presentation (cid:26)a, b : ha, bi = 1, a2b2 = 1(cid:27), where the notation ha, bi was introduced above. This is the dicyclic group of order 12. The monodromy representation is the obvious one: a 7→ (1, 2), b 7→ (2, 3). Zariski [6] noted that the discriminant of a cubic surface S in P3 with a double line is a plane curve of degree 6 which is a union of double line (the image of the double line of S) and a quartic curve (which is straightforward), and moreover proved that the residual quartic has 3 cusps. Thus the variety B(4, 3, 0) is not empty and thus B(4, 3, 0) = V (4, 3, 0) ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 35 Degree 6 singular branch curves (i) The cases (c = 0, n > 0) and c = 3 are not realized. For c = 0, π1(P2 −B) is abelian (by Remark 3.3), and there are no generic covers ramified over C, as we argued in Lemma 3.4. In the second case, c = 3 , Nori's result we cited (see Equation (14)) implies that the group π1(P2 − B) is also abelian. It follows that there are no branch curves with these (d, c, n) triples, even though the corresponding varieties V (6, 0, n) and V (6, 3, n) are not empty. (ii) (c = 6, n = 0): This case was studied by Zariski, as we discussed in the introduction to Section 4.1. If S is a smooth cubic surface in P3, then the branch curve B of a generic projection of S to P2 is in V (6, 6, 0), and Segre's result we discussed (or a direct computation) shows that these 6 cusps lie on a conic. See subsection 4.1.1 and Corollary 4.5. Zariski proved the inverse statement: C ∈ V (6, 6, 0) is a branch curve if and only if 6 cusps of C lie on a conic. It follows from Zariski's work that branch curves form one of the connected components of V (6, 6, 0); and, moreover, Zariski proved the existence of other connected components. Degtyarev proved in [64] that V (6, 6, 0) has exactly two irreducible components. Zariski also proved [6] that for B ∈ B(6, 6, 0) the group π1(P2 − B) is isomorphic to Z/2 ∗ Z/3, whereas for C ∈ V (6, 6, 0) \ B(6, 6, 0) the group π1(P2 − C) is isomorphic to Z/2 ⊕ Z/3. Remark 5.2. As a generalization of the above result, Moishezon [33] proved that the funda- mental group of the complement of B in P2 is isomorphic to the quotient Braidν/Center(Braidν) of the braid group Braidν by its center. (iii) (c = 6, n = 4). Consider the surface S = P1 × P1 embedded to P5 by linear system ℓ1 +2ℓ2. Then S is of degree 4 in P5, and the image of its generic projection to P3 is a quartic with a rational normal curve (the twisted cubic) as its double curve (see [28, pg. 631]). The branch curve B of S is in B(6, 6, 4) as can be seen from Remark 3.12 or from Remark 4.42 and it is known [7] that the fundamental group π1(P2 − B) is braid group of the sphere with 3 generators (see [58] for an explicit calculation). Note that V (6, 6, 4) is irreducible since it is dual to V (4, 0, 3). (iv) (c = 9, n = 0). First, we describe the variety V = V (6, 9, 0). For a curve B ∈ V , its dual is a smooth plane cubic; this gives an isomorphism of V and an open subset in the linear system of plane cubics 3h consisting of smooth curves. It follows immediately that V is irreducible. In this case B(6, 9, 0) = V (6, 9, 0): there is a direct classical construction of a cover with a given branch curve C ∈ V (6, 9, 0) from the dual smooth cubic, discussed in Remark 5.4. Moreover, every curve in V (6, 9, 0) is a branch curve of exactly four different ramified cover- ings, the construction of which was given by Chisini ([14]). This is the only counterexample to the Chisini's conjecture (see subsection 3.2.2). More precisely, we have the following proposition: Proposition 5.3. (a) Given a sextic B with 9 cusps and no nodes, there are four covers having B as a branch curve. Three of them are degree 4 maps P2 → P2, obtained as three various projections of Veronese-embedded P2 in P5, and the fourth one is of degree 3. The construction of the fourth is given in Remark 5.4. (b) The fundamental group π(P2 − B) has exactly 4 non-equivalent representations into symmetric groups Symν for all ν which rise to smooth generic covers ramified over B. Proof. (a) See [36]. (b): Note that G = π1(P2 − B) was already calculated by Zariski in [12], showing that: G ≃ ker(B3(T ) → H1(T )), 36 M. FRIEDMAN, M. LEYENSON where B3(T ) is the braid group of the torus. We compute it here in a different method, using the degeneration techniques explained in [42]. Let S be the image of the Veronese embedding of P2 into P5; the branch curve B of a generic projection S → P2 belongs to V (6, 9, 0) (see e.g. [38]). Since V (6, 9, 0) is irreducible, it is enough to look at B. Now S can be degenerated into a union of four planes is P5, with combinatorics shown on the Figure 7 below, as explained in [42]. Figure 7 : degeneration of V2 Using the techniques of [43],[45], one can prove that G has a presentation with generators {γ1, γ1′, γ2, γ2′, γ3, γ3′} and relations {hγ2, γ1i, hγ2, γ1′ i, hγ1′ , γγ2 2′ i, γ−1 1′ hγ3, γ1i, hγ3, γ1′ i, hγ3, γ 1 i, hγ2′, γ3i, hγ2′ , γ3′i, hγ2′ , γγ3 3′ i, γ−1 2 · γγ2γ1′ γ1 γ1γ1′γ2γ2′γ3γ3′} 2′ , γ−1 3 · γγ3γ1′ γ1 3′ , γ−1 2 γ−1 3 γ−1 3′ 2′ · γ , Using GAP [62] one can prove that G is actually generated by the set {γ1′, γ2, γ3, γ3′}. and having the following relations 3 γ2γ3γ2γ−1 3 γ−1 2 , γ−1 1′ γ3′γ1′γ3′γ−1 1′ γ−1 3′ , γ−1 2 γ−1 3′ γ−1 2 γ3′γ2γ3′, γ2γ1′γ2γ−1 1′ γ−1 2 γ−1 1′ , (cid:26)γ−1 γ3γ1′γ3γ−1 1′ γ−1 3 γ−1 1′ , γ3′γ−1 3′ γ−1 1′ γ−1 3′ γ−1 3 γ3′γ−1 2 γ−1 3 γ−1 1′ γ−1 1′ γ−1 2 γ−1 3′ γ−1 3′ γ2γ−1 1′ γ1 3 γ−1 γ3γ3′γ2γ−1 3 γ−1 2 , γ−1 3′γ3γ−1 2 γ−1 3′ γ1′γ3γ−1 3′ γ2γ−1 1′ γ3′γ1′γ2γ3γ3′γ2, 3′ (cid:27). 3 γ−1 It follows that it suffices to look for the homomorphisms G → Symν, when ν = 3, 4, 5, since 4 transpositions can generate at most symmetric group on 5 letters. Note also that the homomorphisms, in order to correspond to generic covers, have to satisfy Proposition 3.20. Using GAP again, one shows that the only epimorphisms are the following: π1(P2 − B) → Sym3 : (1) {γ1′, γ2, γ3, γ3′} → {(1, 3), (2, 3), (1, 2), (1, 2)} π1(P2 − B) → Sym4 : (2) {γ1′, γ2, γ3, γ3′} → {(2, 4), (2, 3), (3, 4), (1, 2)} (3) {γ1′, γ2, γ3, γ3′} → {(1, 3), (2, 3), (3, 4), (1, 2)} (4) {γ1′, γ2, γ3, γ3′} → {(2, 4), (2, 3), (1, 2), (1, 2)} and there are no epimorphisms to Sym5. Remark 5.4. We recall a construction of Chisini (see [14] or [36, Section 3]). Let B ⊂ P2 a curve with nodes and cusps only, such that its dual A = B ∨ ⊂ (P2)∨ is a smooth curve of degree d. Let (cid:3) Σ = {(λ, y) ∈ (P2)∨ × P2 : λ(y) = 0, λ ∈ A} and ψ : Σ → P2 be the projection to the second factor. For a given point y ∈ P2 − B, the line lλ in P2 is not tangent to A (i.e. intersects A in d distinct points) iff ψ−1(y) has d points. Hence Σ is a degree d covering of P2 with B as the branch curve. For d = 3 we get the fourth example in Proposition 5.3, a degree 3 ramified cover of P2 branched along a 9-cuspidal sextic. Note that this does not contradict to the fact that the only plane curves which are branch curves of projections of smooth cubic surfaces in P3 are sextics with six ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 37 cusps, since the surface constructed above is naturally embedded into F l line in P2, x ∈ P2, x ∈ l} ⊂ (P2)∨ × P2, and not in P3. . = {(l, x) : l is a Degree 8 singular branch curves Theorem 5.5. The only degree 8 singular branch curves of generic linear projections have either 9 cusps and 12 nodes or 12 cusps and 4 nodes. Proof. By simple calculations (see subsection 3.1,3.2 for the obstructions), one can conclude that there are only finite number of possibilities for c and n for a degree 8 singular branch curve. It was proven by Zariski-Deligne-Fulton theorem 3.3 on nodal curves that the case (c = 0, n > 0) cannot be realized as a branch curve. The cases (c = 3, n > 0), (c = 6, n > 0), (c = 9, n = 0) and (c = 9, n = 4) are ruled out as branch curves by Nori's theorem 3.4 (though the corresponding nodal -- cuspidal varieties are not empty). Moreover, the case (c = 18, n = 0) cannot be realized even as nodal -- cuspidal curve: By the Zariski's inequality (13), the number of cusps of a degree 8 curve should be less then 16, and thus V (8, 18, 0) is empty. By considering the dual curve, it's easy to see that also V (8, 15, 4) is empty. We are left to show that there are no degree 8 branch curves with (c, n) = (9, 8), (12, 0), (15, 0), (12, 8) (although the corresponding nodal -- cuspidal varieties are not empty since the genus of these curves is less than 5). (i) (c = 9, n = 8). Assume that there exists a surface S ⊂ P3 such that its branch curve is B ∈ B(8, 9, 8), such that S is its smooth model in P5 (i.e. S → S by generic projection). Since d ≥ 2ν − 2, ν = 3, 4 or 5. By the examples in subsection 4.7.1 we see that ν = 5 and thus e = 6. In this 1( S) = c2( S) = 12. The degree 6 double case, by Lemma 3.9 and 3.10, we can see that c2 curve (of the quintic) cannot lie on a hyperplane for degree reasons. Therefore the canonical system K S is empty (since it is the pull-back of the linear system cut out by hyperplanes passing through the double curve. See [28, pp. 627]). Therefore pg( S) = 0. But since c2 1 = c2 = 12, we get that χ(O S) = 2. Thus 2 = 1 − q + pg or q = −1 -- contradiction. Thus B(8, 9, 8) is empty. (ii) (c = 12, n = 0) Assume that there exist B ∈ B(8, 12, 0). s.t. it is the branch curve of a surface S in P3. By the same argument as in case (i) we see that ν = 5 and thus e = 6. In this case, by Lemma 3.9 and 3.10, we can see that c2 1(S) = 17, c2(S) = 19. So by Remark 4.43, we can find that the number of pinch points p = 0 -- but this cannot happen, by Remark 4.40. Therefore B(8, 12, 0) is empty. Remark 5.6. Note that Zariski proved ([5]) that the twelve cusps of C ∈ V (8, 12, 0) cannot be the intersection of a cubic and a quartic curves. We conjecture that this restriction is directly linked to the fact that B(8, 12, 0) is empty. Remark 5.7. Considering a quartic surface S with a double line in P3, we can project it from a generic smooth point O ∈ S. The resulting branch curve will be a curve in V (8, 12, 0) (see [10]). However, we do not consider this projection as generic. Note that this phenomena happens also in other cases. For example, a branch curve in V (10, 18, 0) of a generic projection does not exist, but if we project a smooth quartic surface in P3 from a point on the quartic, the branch curve of this projection would be in V (10, 18, 0). (iii) (c = 15, n = 0) As in cases (i) and (ii), we can see that a surface S with such a branch curve could only be a quintic in P3 with a degree 6 double curve E∗ with 3 triple points (by Remark 4.42). Considering Π -- the plane passing through these three points -- and looking at E∗ · Π, we see that E∗ ⊂ Π. However, deg Π ∩ S = 5, so such a surface does not exist. 38 M. FRIEDMAN, M. LEYENSON Note that in this case Zariski demonstrated in ([5]) that the 15 cusps cannot lie on a quartic curve. Remark 5.8. The nonexistence in cases (ii) and (iii) can also be proven by the method indicated in (i). (iv) (c = 12, n = 8). The variety V (8, 12, 8) is irreducible, since it is dual to the V (4, 0, 2). If S is a quartic surface in P3 which double curve is a union of two skew lines, then we prove in Subsection 4.7 that a branch curve of S is in B(8, 12, 8). By [12], π1(P2 − B) ≃ ker(B4(T ) → H1(T )). However the double curve of an image a smooth surface in PN in P3 is an irreducible curve (unless the surface is the Veronese surface, where in this case the double curve is a union of three lines. See e.g. [25, Theorem 3]). Thus B is not a branch curve of generic linear projection. We now shall construct degree 8 branch curves with (c, n) = (9, 12), (12, 4). With this we covered all the possible numerics for the possible number of nodes and cusps of a degree 8 branch curve. (I) (c = 9, n = 12). First, note that V (8, 9, 12) is irreducible, since it is dual to the variety V (5, 0, 6). Now note that if we consider the Hirzebruch surface F1 embedded into P6 by 2f + s (where f is the class of a fiber and s is the class of a movable section, so that f 2 = 1, f · s = 1, s2 = 1.) A projection of this model of F1 to P2 factorizes as a composition of a projection to P3, where the image of F1 is a quintic surface with a double curve of degree 6, and a projection from P3 → P2. One can check that the branch curve B of the resulting map has 9 cusps and 12 nodes (see [46] or Remark 3.12) and that π1(P2 − B) is isomorphic to the braid group of the sphere with 4 generators (see [7]). (II) (c = 12, n = 4). We do not know whether V (8, 12, 4) is irreducible. If S is a smooth intersection of two quadrics in P4, then the branch curve of a projection of S to P2 is in B(8, 12, 4), see Subsection 4.7 for the details. By [49], π1(C2 − B) ≃ Braid4/h[x2, xx1x3 ]i. (cid:3) 2 6. Appendix A : New Zariski pairs (By Eugenii Shustin) 6.1. Introduction. Along Lemma 3.21, The family B(d, c, n) of the plane branch curves of degree d with c cusps and n nodes as their only singularities consists of entire components of V (d, c, n), the space parameterizing all irreducible plane curves of degree d with c cusps and n nodes as their only singularities. In the particular case of the branch curves of generic projections of smooth surfaces of degree ν ≥ 3 in P3 onto the plane, one has (cf. [1] and Lemma 4.2) (34) d(ν) = ν(ν − 1), c(ν) = ν(ν − 1)(ν − 2), n(ν) = 1 2 ν(ν − 1)(ν − 2)(ν − 3) . The celebrated Zariski result [6] says that, in the case ν = 3, d = 6, c = 6, n = 0, the variety V (6, 6, 0) contains a component which is disjoint with B(6, 6, 0). This suggests Conjecture 6.1. For each ν ≥ 3, the variety V (d(ν), c(ν), n(ν)) contains a component disjoint with B(d(ν), c(ν), n(ν)). Here we confirm this conjecture for few small values of ν. Theorem 6.2. Conjecture 6.1 holds true for 3 ≤ ν ≤ 10. We prove Theorem 6.2 explicitly constructing curves C ∈ V (d(ν), c(ν), n(ν))\B(d(ν), c(ν), n(ν)). Our construction is based on the patchworking method as developed in [50]. It seems that this method does not allow one to cover sufficiently large values of ν. 6.2. Construction. ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 39 6.2.1. The main idea. The variety V (d, c, n) is said to be T -smooth at C ∈ V (d, c, n) if the germ of V (d, c, n) at C is the transverse intersection in OP2(d) of the germs at C of the smooth1 [47]). In particular, this equisingular strata corresponding to individual singular points of C (cf. implies that, for any subset S ⊂ Sing(C), there exists a deformation C S 0 = C, C S t ∈ V (d, c′, n′), where c′, n′ are the numbers of cusps and nodes in Sing(C)\S, respectively, and, furthermore, the deformation C S t smoothes out all the singular points of C in S. Clearly, the T -smooth part V T (d, c, n) of V (d, c, n) is an open subvariety (if not empty) of V (d, c, n). t , t ∈ (C, 0), such that C S We derive Theorem 6.2 from Proposition 6.3. (1) Let ν ≥ 3. If V T (d(ν), c(ν), n(ν) + 1) 6= ∅, then V (d(ν), c(ν), n(ν))\B(d(ν), c(ν), n(ν)) 6= ∅. (2) If 3 ≤ ν ≤ 10, then V T (d(ν), c(ν), n(ν) + 1) 6= ∅. 6.2.2. Proof of Proposition 6.3(1). If ν ≥ 5 then as noticed in Subsection 4.6, B3(d(ν), c(ν), n(ν)) is not T -smooth, since a T -smooth family must have the expected dimension. On the other hand, smoothing out one node of a curve C ∈ V T (d(ν), c(ν), n(ν) + 1), we obtain an element of V T (d(ν), c(ν), n(ν)), a component whose dimension differs from that of B3(d(ν), c(ν), n(ν)). This reasoning does not cover the case of ν = 3 and 4. We then provide another argument which, in fact, works for all ν ≥ 3. Let ν ≥ 3, C ∈ V T (d(ν), c(ν), n(ν) + 1). We intend to show that there is a nodal point p ∈ C such that Sing(C)\{p} is not contained in a plane curve of degree d′(ν) = (ν − 1)(ν − 2). This is enough, since by [8] (see also Proposition 4.2), all the singular points of a branch curve D ∈ B(d(ν), c(ν), n(ν)) lie on a plane curve of degree d′(ν), and, on the other hand, a deformation C p t ∈ V (d(ν), c(ν), n(nu)), t ∈ (C, 0), of C which smoothes out the node p, contains curves whose singular points are not contained in a curve of degree d′(ν). We prove the existence of the required node p ∈ C arguing for contradiction. Let p1, p2 be some distinct nodes of C and let C1 ⊃ Sing(C)\{p1}, C2 ⊃ Sing(C)\{p2} be some curve of degree d′(ν). Since d(ν) · d′(ν) = 2(c(ν) + n(ν)), the curves C1 and C2 are non-singular along their intersection with C, in particular, they are reduced. Furthermore, p1 6∈ C1 and p2 6∈ C2. Let D be the (possibly empty) union of the common components of C1 and C2 with deg D = k, 0 ≤ k < d(ν). So, C1 = DD1, C2 = DD2, where the curves D1, D2 of degree d(ν) − k have no component in common. By Bezout's theorem D ∩ C consists of kd(ν)/2 points of Sing(C), and Di ∩ C = Sing(C)\(D ∩ C ∪ {pi}), i = 1, 2. Take two distinct generic straight lines L1, L2 through p1. By Noether's AF + BG theorem (see, for instance, [23])2, • if k ≤ d′(ν) + 1 − d(ν)/2, then there are polynomials A1, A2 of degree d′(ν) + 2 − k and i = AiD1 + BiC, i = 1, 2, • if k ≥ d′(ν) + 2 − d(ν)/2, then there are polynomials A1, A2 of degree d′(ν) + 2 − k such polynomials B1, B2 of degree 2d′(ν) + 2 − d(ν) − 2k such that D2 2L2 that D2 2L2 i = AiD1, i = 1, 2. The latter case is impossible since D2 2(L2 case, we obtain that D1 divides D2 2L2 1B2 − L2 i and D1 have no component in common. In the former 2B1), and hence divides L2 1B2 − L2 2B1. In view of deg D1 = (ν − 1)(ν − 2) − k > 2 + (ν − 1)(ν − 4) − 2k = deg(L2 1B2 − L2 2B1) , we conclude that L2 D2 1D1 + B ′ 2 = A′ 1B2 = L2 1 divides B1, but then L2 1C contrary to the fact that p2 6∈ D2 and p2 ∈ D1 ∩ C. 2B1, in particular, L2 1 divides A1 too, and hence 1In the case of nodes and cusps, the smoothness of these equisingular strata always holds. 2For the sake of notation we denote a plane curve and its defining homogeneous polynomial (given up to a constant factor) by the same symbol, no confusion will arise. 40 M. FRIEDMAN, M. LEYENSON 6.2.3. Proof of Proposition 6.3(2). We suppose that ν ≥ 4. Using the patchworking construc- tion of [50, Theorem 3.1], we obtain curves in V T (12, 24, 16), V T (20, 63, 67), V T (30, 126, 191), V T (42, 216, 435), V T (56, 336, 902), V T (72, 504, 1550), and V T (90, 720, 2526), what suffices for our purposes in view of Proposition 6.3(1), since d(4) = 12, c(4) = 24, n(4) = 12 < 16 , d(5) = 20, c(5) = 60 < 63, n(5) = 60 < 67 , d(6) = 30, c(6) = 120 < 126, n(6) = 180 < 191 , d(7) = 42, c(7) = 210 < 216, n(7) = 420 < 435 , d(8) = 56, c(8) = 336, n(8) = 840 < 902 , d(9) = 72, c(9) = 504, d(9) = 1512 < 1550 , d(10) = 90, c(10) = 720, n(10) = 2520 < 2526 . Referring to [50] for details, we only recall that the patchworking construction uses a convex3 lattice subdivision of the triangle Td = conv{(0, 0), (0, d), (d, 0)}. Pieces ∆1, ..., ∆N of the sub- division will serve as Newton polygons of polynomials in two variables (called block polynomials) which define curves with nodes and cusps leaving in the respective toric surfaces: Ck ⊂ Tor(∆k), k = 1, ..., N . Along [50, Theorem 4.1], the patchworking construction can be performed under the following sufficient conditions: (C1) Any two block polynomials have the same coefficients along the common part of their Newton polygons, and the truncations of block polynomials to any edge is a nondegenerate (quasihomogeneous) polynomial. (C2) The adjacency graph of the pieces of the subdivision can be oriented without oriented cycles so that if, for each polygon ∆k, ≤ k ≤ N , we mark its sides which correspond to the arcs of the adjacency graph coming inside the polygon and denote by Dk ⊂ Tor(∆k) the union of the unmarked toric divisors, then the number of cusps of any component C of Ck is less than CDk. By [50, Theorems 3.1 and 4.1], the resulting curve with the Newton triangle Td belongs to V T (d, c, n), and the numbers c and n are obtained by summing up the numbers of cusps and nodes over all the block curves. Condition (C2) formulated above is, in fact, sufficient for the following transversality property defined in [50, Definition 2.2] and used in the patchworking construction of [50, Theorem 3.1]: the variety VC consisting of the curves in the linear system C on Tor(∆k) which are equisingular to C and intersect D′ k is the union of the marked toric divisors of Tor(∆k), is smooth at C of codimension 2c(C) + n(C) + CD′ k, where c(C), n(C) are the numbers of cusps and nodes of C. We shall call this property the T -smoothness relative to D′ k. Let L be a coordinate line in P2 corresponding to a side σ of Td. Then the patchworking construction produces a curve, where the variety V (d, c, n) is T -smooth relative to L, if one replaces condition (C2) by the following one: k at the same points as C, where D′ (C2') The adjacency graph of the pieces of the subdivision can be oriented without oriented cycles so that if, for each polygon ∆k, ≤ k ≤ N , we mark its sides which correspond to the arcs of the adjacency graph coming inside the polygon or are contained in σ, and denote by Dk ⊂ Tor(∆k) the union of the unmarked toric divisors, then the number of cusps of any component C of Ck is less than CDk. 3Convexity means that the subdivision lifts up to a graph of a convex function linear on each subdivision polygon and having a break along each common edge. ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 41 We consider the subdivisions of T12, T20, T30, T42, T56, T72, and T90 shown in Figures 1, 2(b), 3, where all the slopes are 0, −1, or ∞. We leave to the reader an easy exercise to check that these subdivisions are convex. Next we describe the block polynomials. (1) If d = 12, we take the block polynomial F1(x, y) with Newton triangle T6 defining a curve C1 ∈ V (6, 6, 4) (such a curve is dual to an irreducible quartic with three nodes). The other block polynomials are obtained via affine automorphisms of Z2 which interchange two adjacent triangles of the subdivision keeping their common side fixed. Thus, the patchworking construction gives a curve C12 ∈ V T (12, 24, 16). (2) If d = 20, we take the block polynomial F2(x, y) with Newton triangle T6 defining a curve C2 ∈ V (6, 9, 0) (such a curve is dual to a non-singular cubic). The other block polynomials with the Newton triangles with side length 6 are obtained from F2 by suitable reflections. For any other polygon in the given subdivision, we take a block polynomial splitting into the product of linear polynomials, defining (reducible) nodal curves, and satisfying (C1). It is easy to check that condition (C2) holds, and thus, one obtains a curve C20 ∈ V T (20, 63, 67). (3) If d = 30 or 42, for each triangle in the subdivision having side length 6 and intersecting with a coordinate axis, we take the block polynomial obtained from F2 as described above, and, for any other polygon of the subdivision, we take a suitable polynomial splitting into the product of linear polynomials. This gives us the curves C30 ∈ V T (30, 126, 191) and C42 ∈ V T (42, 216, 435). (4) If d = 56 we use the block polynomial F3 defining a curve C9 ∈ V (9, 16, 10) which is obtained via a slight modification of the construction of a curve C ′ 9 ∈ V (9, 20, 0) in [50, Section 4.3]. We consider the subdivision of T9 shown in Figure 2(a) (cf. [50, Figure 2]) and take the following block curves: those with two symmetric Newton quadrangles have 8 cusps each (as in [50, Proof of Theorem 4.3]), the block curve with Newton triangle has one node, and the block curve with Newton square splitting onto 6 lines, has 9 nodes. Now we subdivide the triangle T56 as shown in Figure 2(b) and take the following block curves: • The block curves with the triangles intersecting with the coordinate axes and the triangle marked with asterisk are defined by the polynomial F3 and its appropriate transforms, • each other block curve is defined by a polynomial splitting into linear factors. Observe that the conditions (C1) and (C2) can be satisfied in this situation, which, finally, gives us a curve in V T (56, 336, 902). (5) Observe that the construction of step (3) gives a curve C30 at which the variety V (30, 126, 191) is T -smooth relative to the y-axis. Clearly, this T -smoothness property at C30 hold relatively to almost all lines in P2, and hence by an appropriate coordinate change we can make V (30, 126, 191) to be T -smooth at C30 relative to each of the coordinate lines. Let F4(x, y) be a defining polynomial of C30. Consider the subdivision of T72 presented in Figure 3(a). For the triangle incident to the origin, we take the above polynomial F4(x, y), and, for the other triangles with side length 30, we take the transforms of F4 as described in step (1). For the other polygons of the subdivision we take appropriate polynomials splitting into linear factors. The relative T -smoothness of V (30, 126, 191) at C30 ensures condition (C2); hence the patchworking procedure is performable, and it gives a curve C72 ∈ V T (72, 504, 902). (6) In the case d = 90 we need a curve C ′ 30 ∈ V (30, 120, 199) at which the variety V (30, 120, 199) is T -smooth relatively to each of the coordinate lines. Assuming that such a curve exists, we take its defining polynomial F5(x, y) and spread it through all the triangles in the subdivision shown in Figure 3(b) except for the right-most one (marked by asterisk). For the latter triangle and for the parallelogram we take suitable polynomials splitting into linear factors. Again, due to the relative 42 M. FRIEDMAN, M. LEYENSON ✻ ✻ ❅❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ 6 ❅ ❅ 12 ✲ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ 6 ❅ ❅ ❅ ❅ ❅ ❅ 12 ❅ ❅ 18 20 ❅ ✲ (a) (b) ❅ ❅ ❅ ❅ ✻ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ 6 ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ✻ ❅❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ 6 ❅ 12 ❅ 18 ❅ ✲ ❅ 30 ❅ 24 ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ 12 ❅ 18 ❅ 24 ❅ 30 ❅ ❅ ❅ 36 ❅ ✲ ❅ 42 (c) (d) Figure 1. Patchworking construction: The case ν = 4, 5, 6, 7 T -smoothness of V (30, 120, 199) at C ′ procedure gives a curve in V (90, 720, 2526). 30, the condition (C2) holds true, and hence the patchworking The required curve C ′ 30 can be constructed using the modified construction of step (3) and a generic coordinate change afterwards. The modification is as follows: for the upper and the right- most triangles with side length 6 in the subdivision shown in Figure 1(c), we take polynomials defining curves in V (6, 6, 4) (instead of V (6, 9, 0) as in the original construction of step (3)). Curves of degree 6 with 6 cusps and 4 nodes do exist: they are dual to rational quartics with 3 nodes. To ensure condition (C1), we need the sextics as above which cross one of the coordinate lines along a prescribed configuration of 6 points. Notice that, given a straight line L ⊂ P2, the varieties V (6, 6, 4) and V (6, 9, 0 are T -smooth elative to L at each curve crossing L transversally (it immediately follows from condition (C2')). In particular, this yields that the rational maps V (6, 6, 4) → Sym6(L) and ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 43 ❅ ❅ ❅ ✁ ❅ ✁ ✟✟✟ ❅ ✁ 3 ❅ ❅ ❅ ❅ ❅ 9 3 4 5 ✲ (a) (b) ✻ ❅ ❅ ❅ ✻ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ 9 ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ∗ ❅ ❅ ❅ ❅ ❅ ❅ 18 ❅❅ 27 ❅ ❅ ❅ ❅ ❅ 36 ❅ ❅ ❅ ❅ ❅ 45 ❅ ❅ ❅ ❅❅ ❅❅ 54 56 ✲ Figure 2. Patchworking construction: The case ν = 8 V (6, 9, 0) → Sym6(L) defined by C 7→ C ∩ L are dominant. Hence we can choose polynomials defining curves a curve in V (6, 6, 4) and a curve in V (6, 9, 0) so that the considered patchworking data will meet condition (C1). 7. Appendix B : Picard and Chow groups for nodal-cuspidal curves In this Appendix we remind the reader the connections between Cartier and Weil divisors and the connection of the Picard and Chow groups on a nodal -- cuspidal curve with c cusps and n nodes. This connection is implicit in Segre [8], and here we recall the explicit formulation. 44 M. FRIEDMAN, M. LEYENSON ✻ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ 30 ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅❅ 60 ❅ ✲ ❅❅ 72 ✻ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ 30 ❅ ❅ ❅ ❅❅ 60 ❅ ∗ ❅ ❅ ✲ ❅❅ 90 (a) (b) Figure 3. Patchworking construction: The case ν = 9 and 10 Let B a nodal -- cuspidal plane curve, with B ∗ its normalization in P3. First, by definitions of Pic and A0, we have for B 0 0 0 Cart. P rinc. B / W eil. P rinc. B 0 0 / GS / GS 0 / Cartier B / W eil B / Pic B 0 / A0B 0 0 / 0 / 0 where the canonical map Pic B → A0B is the map induced by associating the class of a Weil divisor with each Cartier divisor on B, S is the set of singular points of B, and GS is the subgroup of the group of Cartier divisors on B such that their associated Weil divisors are trivial. Note that the map Cartier B → W eil B is surjective since B is a nodal-cuspidal curve. Secondly, there is an exact sequence 0 → H 0QS → Pic B π∗ → Pic B ∗ → 0 where QS = π∗(O∗ B∗ )/O∗ B = Qp∈Sing B ( Qp∗ π →p O∗ p∗)/O∗ p. We also have the following excision diagram:     / /     / / /   /   /   /   / /   /   /   / ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 45 0 ZP 0 / ZP / ∽ / 0 A0ξ∗ A0B ∗ A0ξ A0B A0U ∗ ≃ A0U 0 0 0 0 0 where ξ = P + Q, ξ∗ = P ∗ + Q∗, U = B − ξ, U ∗ = B ∗ − ξ∗, and the map A0ξ∗ → A0ξ can be described as ZP ∗ by a subgroup generated by 2) for a preimage of each node p of B. (For a different proof that the map A0B ∗ → A0B is (p∗ epimorphic, see [48, Example 1.9.5]). Denote T = ZP / ∽. → ZP ⊕ ZQ which is the factorization of ZP ∗ ⊕ ZQ∗ 1 − p∗ Combining the two diagrams together, we get 0 0 T 0 0 / H 0QS Pic B π∗ Pic B ∗ ∼= / GS Pic B A0B 0 0 T ′ 0 0 Where the last column is induced from the fact that Pic B ∗ ≃ A0B ∗ and from the exact sequence (35) Note that the map H 0QS → GS is injective, and cannot be surjective, otherwise the map Pic B ∗ → A0B would be an isomorphism. Thus T ′ ∼= T . 0 → T → A0B ∗ → A0B → 0. 8. Appendix C : Bisecants to a complete intersection curve in P3 We note here that the inverse statement to Theorem 4.4 is easy. Explicitly, we have the following Theorem: Let C be a curve in P = P3 which is a complete intersection of type (µ, ν), and let a be a point in P which is not on C such that C does not admit any 3-secants through a. Then C has 1 2 µν(µ−1)(ν −1) bisecants passing through a. Remark 8.1. The above theorem gives a direct proof that the number of nodes of the branch curve B is indeed n = 1 2 ν(ν − 1)(ν − 2)(ν − 3) (recall that B ∗ is a complete intersection of S and P olOS, i.e., of type (ν, ν − 1) and that the line Oq∗ for each q∗ ∈ Q∗ is also considered as a bisecant of B ∗).       /   /   / /   / /   / /   / /   / /   / /         / / /   / /   / /   / / /   / /   / /   46 M. FRIEDMAN, M. LEYENSON Proof. Our proof is essentially a reformulation of a proof by Salmon, [1, art. 343]. See also [15, Chapter IX, sections 1.1,1.2] for another way to induce this formula. Consider the moduli space M of data {line l in P which is bisecant for C, a point p′ ∈ l ∩ C, a point p ∈ l, p /∈ C}. (see the following Figure) p p’ c l the parameters (l, p′, p) It is clear that the line l can be reconstructed uniquely from p′ and p as lp,p′, and thus M can be embedded into P × P, (l, p′, p) 7→ (p′, p). For a point (l, p′, p) in M , let q be a point in l ∩ C different from p′. Then there is a number t ∈ k such that q = p′ + tp. If C is given by 2 equations u, v, then we have u(q) = 0, v(q) = 0. Let us write u(q) = u(p′ + tp) = u0(p′) + tu1(p′, p) + · · · + tµuµ(p′, p), where ui is of degree µ − i in p′ and i in p. In the same way we can write v(q) = v(p′ + tp) = v(p′) + tv1(p′, p) + · · · + tνvν(p′, p). Consider now two polynomials, a(t) = u1(p′, p) + · · · + tµ − uµ(p′, p), b(t) = v1(p′, p) + · · · + tν − uν(p′, p). Let R(p′, p) be the resultant of a(p′, p, t) and b(p′, p, t) in t. It has (see the the Sylvester definition of resultant) bidegree ((µ − 1)(ν − 1), µν − 1) in (p′, p). Lemma: Let U ⊂ P × P = {(p′, p) : p′ 6= p}. Then U ∩ (R = 0) ∩ (C × P) = M Indeed, let (p′, p) be such that R(p′, p) = 0, p′ a(p′, p, t) = 0, b(p′, p, t) = 0. Let q = p′ + tp. We have 6= p. Then there is a number t ∈ k such that u(q) = u(p′) + ta(p′, p) = u(p′) v(q) = v(p′) + tb(p′, p) = v(p′). Thus q ∈ C iff p′ ∈ C. It follows that (p′, p) is in M iff p′ ∈ C. This proves the lemma. It follows now that R ∩ (C × a) = M ∩ (P ×a) = ( bisecants through a to C with a marked point p′ in l ∩ C ) The order of this set is equal to degp′ C · degC = (µ − 1)(ν − 1)µν. Since C does not have any 3-secants through a, it follows that the number of bisecants through a is one half of the number above. (cid:3) References [1] G. Salmon, A treatise on analytic geometry of three dimensions edition by Hodges, Smith, & Co., Dublin (1862), fourth edition. [2] A. Brill, M. Nother, Uber die algebraischen Functionen und ihre Anwendung in der Geometrie, Math. Ann. 7 (1874) 269-310. [3] G. H. Halphen, M´emoire sur la classification des courbes gauches alg´ebriques J. Ec. Polyt. 52 (1882), 1200. ON RAMIFIED COVERS OF THE PROJECTIVE PLANE 47 [4] S. Lefschetz, On the existence of loci with given singularities, Transactions of the American Mathematical Society, Vol. 14, No. 1 (1913), 23-41. [5] O. Zariski, On the linear connection index of the algebraic surfaces zn = f (x, y), Proceedings of the National Academy of Sciences, vol. 15 (1929), 494-501. [6] O. Zariski, On the Problem of Existence of Algebraic Functions of Two Variables Possessing a Given Branch Curve, American Journal of Mathematics, Vol. 51, No. 2 (1929), 305-328. [7] O. Zariski, On the Poincare group of rational plane curves, American Journal of Mathematics, Vol. 58, (1930), 607-619. [8] B. Segre, Sulla Caratterizzazione delle curve di diramazione dei piani multipli generali Mem. R. Acc. d'Italia, I 4 (1930), 531. [9] O. Zariski, On the irregularity of cyclic multiple planes, Annals of Mathematics (2), vol. 32 (1931), 485 - 511. [10] P. Du Val, On Triple Planes Having Branch Curves of Order Not Greater Than Twelve, J. London Math. Soc., s1-8 (1933), 199-206. [11] E.R. Van Kampen, On the fundamental group of an algebraic curve, Am. J. Math. 55 (1933), 255-260. [12] O. Zariski, On the topological discriminant group of a Riemann surface of genus p, Amer. J. Math., 59, (1937), 335-358. [13] T.R. Hollcroft, Anomalous plane curve systems associated with singular surfaces, Bull. Amer. Math. Soc. Volume 46, Number 4 (1940), 252-257. [14] O. Chisini, Sulla identita birazionale delle funzioni algebriche di due variabili dotate di una medesima curva di diramazione, Rend. Ist. Lombardo 77 (1944), 339-356. [15] J. Semple, L. Roth, Introduction to Algebraic Geometry, Oxford University Press (1949). [16] E.R. Grauert, R. Remmert, Komplexe Raume, Math. Ann. 136 (1958), 245 - 318. [17] A. Grothendieck, Fondements de la G´eometrie alg´ebrique, S´eminaire Bourbaki, Exp. No. 190, (1959/60). [18] R. Narasimhan, Introduction to the Theory of Analytic Spaces, Lecture Notes in Mathematics, Volume 25, Springer-Verlag (1966). [19] A. Altman, S. Kleiman, Introduction to Grothendieck duality theory, Springer (1970). [20] B. Iversen, Numerical invariants and multiple planes, Amer. J. Math. 92 (1970), 968-996. [21] A. Grothendieck, SGA1, Revetements ´etales et groupe fondamental, 1960-1961, Lecture Notes in Mathematics 224 (1971). [22] O. Zariski, Algebraic surfaces Springer, Heidelberg (1971), 2nd. ed. [23] Van der Waerden, B. L.: Einfuhrung in die algebraische Geometrie, 2nd edition. Springer (1973). [24] J. M. Wahl, Deformations of plane curves with nodes and cusps, Amer. J. Math. 96 (1974), 529-577. [25] B. Moishezon, Complex surfaces and connected sums of complex projective planes, Lecture Notes in Mathematics, 603. Springer-Verlag, New York, 1977. [26] R. Piene, Some formulas for a surface in P3, Algebraic geometry (Proc. Sympos., Univ. Tromsø, Tromsø, (1977)), pp. 196 -- 235, Lecture Notes in Math., 687, Springer, Berlin, (1978). [27] L. Gruson, C. Peskine, Genre des courbes de l'espace projectif, Algebraic Geometry (Tromsø 1977), Springer- Verlag, Lect. Notes in Math. 687(1978), 31-59. [28] P. Grithis, J. Harris, Principles of Algebraic Geometry Wiley, New York (1978). [29] S. Greco, P. Valabrega, On the Theory of Adjoints, in "Algebraic Geometry" (K. Lønsted, ed.), Springer-Verlag LNM, vol. 732, (1979), pp. 98-123. [30] W. Fulton, J. Hansen,A Connectedness Theorem for Projective Varieties, with Applications to Intersections and Singularities of Mappings, Annals of Mathematics, Second Series, Vol. 110, No. 1 (1979), 159-166. [31] W. Fulton, On the fundamental group of the complement of a node curve, Ann. of Math. 111 (1980), 407-409. [32] L. Chiantini, On some algebraic and geometric extension of the theory of adjoints, Rendiconti del Seminario Matematico della Universita di Padova, tome 64 (1981). [33] B. Moishezon, Stable branch curves and braid monodromy, Lectur Notes in Math., 862, Springer-Verlag, (1981) pp. 107-192. [34] M. Nori, Zariskis conjecture and related problems, Ann. Sci. ´Ecole Norm. Sup. (4) 16 (1983), 305-344. [35] A. Libgober, Fundamental groups of the complements to plane singular curves, Algebraic geometry, Bowdoin, 1985 (Brunswick, Maine, 1985), Proc. Sympos. Pure Math., vol. 46, Amer. Math. Soc., Providence, RI, 1987, pp. 2945. [36] F. Catanese, On a problem of Chisini, Duke Math. J. 53 (1986), 3342. [37] J. Harris, On a problem of Severi, Invent. Math. 84 (1986), 445-461. [38] B. Moishezon, M. Teicher, Galois covers in theory of algebraic surfaces , Amer. Math. Soc. Pub. PSPM 46 (1987), 47-65. [39] B. Moishezon, M. Teicher, Braid group technique in complex geometry, I, Line arrangements in P2, Contemp. Math. 78, (1988), 425-555. 48 M. FRIEDMAN, M. LEYENSON [40] B. Moishezon, M. Teicher, Braid group technique in complex geometry, II, From arrangements of lines and conics to cuspidal curves, Algebraic Geometry, Lecture Notes in Math., vol. 1479 (1990), 131-180. [41] J. D'Almeida, Courbe de ramification de la projection sur P2 dune surface de P3, Duke Math. J. Volume 65, Number 2 (1992), 229-233. [42] B. Moishezon, M. Teicher, Braid group techniques in complex geometry III: Projective degeneration of V3, Con- temp. Math. 162, (1993), 313-332. [43] B. Moishezon, M. Teicher, Braid group techniques in complex geometry IV: Braid monodromy of the branch curve S3 of V3 → P2 and application to π1 : (P2 − S3), Contemp. Math. 162 (1993), 332-358. [44] B. Moishezon, The arithmetic of braids and a statement of Chisini, Contemporary Math 164 (1994), 151-175. [45] B. Moishezon, M. Teicher, Braid group techniques in complex geometry, V: The fundamental group of comple- ments of a branch curve of Veronese generic projection, Communications in Analysis and Geometry 4, (1996), no. 1, 1-120. [46] B. Moishezon, A. Robb and M. Teicher, On Galois covers of Hirzebruch surfaces, Math. Ann. 305, (1996), 493-539. [47] Shustin, E. Geometry of equisingular families of plane algebraic curves, J. Alg. Geom. 5 (1996), no. 2, 209 -- 234. [48] W. Fulton, Intersection theory, Springer-Verlag (1997). [49] A. Robb, On branch curves of algebric surfaces, AMS/IP Studies in Advanced Mathematics, vol. 5 (1997). [50] Shustin, E. Gluing of singular and critical points. Topology 37 (1998), no. 1, 195 -- 217. [51] I. Shimada, On the commutativity of fundamental groups of complements to plane curves, Math. Proc. Cambridge Philos. Soc. 123 (1998), no. 1, 49 -- 52. [52] Vik. Kulikov, On Chisini's Conjecture Izv. RAN. Ser. Mat. (1999), 63:6, 83116 [53] Vik. Kulikov, M. Teicher,Braid monodromy factorizations and diffeomorphism types, Izv. Ross. Akad. Nauk Ser. Mat. 64(2), 89-120 (2000) [Russian]; English transl., Izvestiya Math. 64(2), 311-341 (2000). [54] S. Nemirovski, A remark on the Chisini conjecture, e-print AG/0001113. [55] V. S. Kulikov, Vik. S. Kulikov Generic coverings of the plane with A-D-E-singularities, Izv. RAN. Ser. Mat., (2000), 64:6. [56] C. Ciliberto, R. Miranda, M. Teicher, Pillow Degenerations of K3 Surfaces, Applications of Algebraic Geometry to Coding Theory, Physics, and Computation, NATO Science Series II, Vol. 36 (2001), 53-64. [57] Vik. S. Kulikov, Generalized Chisini's Conjecture, Tr. MIAN (2003). [58] M. Amram, O. Shoetsu, Degenerations and fundamental groups related to some special Toric varieties, Michigan Math. J. Volume 54, Issue 3 (2006), 587-610. [59] G.-M. Greuel, C. Lossen, E. Shustin, Global aspects of complex geometry, F. Catanese et al., eds., Springer (2006), pp. 171-209. [60] Val. S. Kulikov , On characterization of branch curves of generic coverings, Interuniversity proceedings "Mathe- matics: fundamental problems, applications, teaching", vol.2, 66-102, Moscow State University Publisher (2002) (In Russian). [61] R. Vakil, Murphys law in algebraic geometry: Badly-behaved deformation spaces, Invent. Math. 164 (2006), no. 3, 569-590. [62] The GAP Group, GAP -- Groups, Algorithms, and Programming, Version 4.4.10; 2007. (http://www.gap-system.org) [63] C. Ciliberto, F. Flamini On the branch curve of a general projection of a surface to a plane, arXiv:0811.0467v1 (2008). [64] A. Degtyarev, On deformations of singular plane sextics, J. Algeb. Geom. 17 (2008), 101-135. [65] Vik. Kulikov, On Chisini's Conjecture II Izv. Math. 72 (2008). [66] M. Friedman, M. Teicher, On non fundamental group equivalent surfaces, Algebraic & Geometric Topology 8 (2008) 397-433. [67] M. Friedman, M. Leyenson, R. Lehman, On ramified covers of the projective plane II: Segre's theory for singular surfaces in P3 , in preperation. Michael Friedman, Department of Mathematics, Bar-Ilan University, 52900 Ramat Gan, Israel E-mail address: [email protected] Maxim Leyenson E-mail address: [email protected] Eugenii Shustin, School of Mathematical Sciences, Raymond and Beverly Sackler Faculty of Exact Sciences, Tel Aviv University, Ramat Aviv, 69978 Tel Aviv, Israel E-mail address: [email protected]
1111.0874
1
1111
2011-11-03T15:18:51
Donaldson-Thomas invariants for complexes on abelian threefolds
[ "math.AG" ]
We modify the standard perfect symmetric obstruction theory for moduli spaces of simple perfect complexes, to the situation of complexes on abelian threefolds with fixed determinant and Fourier-Mukai determinant. As outcome we attach nontrivial Donaldson-Thomas invariants to moduli spaces for complexes modulo twist and translation.
math.AG
math
DONALDSON -- THOMAS INVARIANTS FOR COMPLEXES ON ABELIAN THREEFOLDS MARTIN G. GULBRANDSEN Abstract. Donaldson -- Thomas invariants for moduli spaces M of perfect com- plexes on an abelian threefold X are usually zero. A better object is the quotient K = [M/X ×bX] of complexes modulo twist and translation. Roughly speaking, this amounts to fixing not only the determinant of the complexes in M, but also that of their Fourier -- Mukai transform. We modify the standard perfect symmet- ric obstruction theory for perfect complexes to obtain a virtual fundamental class, giving rise to a DT-type invariant of the quotient K . It is insensitive to deforma- tions of X, and respects derived equivalence. As illustrations we examine the case of Picard bundles and of Hilbert schemes of points. 1. Introduction The aim of this paper is to attach nontrivial Donaldson -- Thomas type invariants to abelian threefolds. Donaldson -- Thomas invariants, as defined by Thomas [21], are integers associ- ated to (proper) moduli spaces M of stable coherent sheaves on a projective three- fold X with trivial canonical bundle. By definition, the Donaldson -- Thomas invari- ant is the degree of the virtual fundamental class (of dimension zero) associated to a natural perfect obstruction theory on M. If Pic0(X) is nontrivial, so that line bundles deform, the perfect obstruction the- ory contains a trivial summand, causing the virtual fundamental class and hence its degree to be zero. The standard remedy is to shrink M so that it parametrizes only sheaves E with fixed determinant line bundle det(E ). The resulting invariant makes sense when X is an abelian threefold, but is still almost always zero, as there is another trivial summand present in the obstruction theory. Roughly speaking, this summand controls deformations of the determinant line bundle det(bE ) of the Fourier -- Mukai transform bE . In this text we call det(bE ) the codeterminant of E . In Section 2 we modify the standard perfect obstruction theory to obtain a virtual fundamental class on moduli spaces for sheaves -- and more generally perfect complexes -- on abelian threfolds, with fixed determinant and codeter- minant. This gives rise to a DT-type invariant which is nontrivial in general. In Section 3 we package the data in a more natural way by forming the stack quotient K = [M/X ×bX] and assigning a DT-invariant to K. This invariant is insensitive to deformations of X, and it agrees with Behrend's weighted Euler characteristic of K. In particular it is intrinsic to K. Moreover, it respects derived equivalence in the following sense: if Y is a second abelian threefold such that the derived cat- egories D(X) (cid:27) D(Y ) are equivalent, then M may equally well be considered as a moduli space for complexes on Y , but the quotient K and its DT-invariant stay the same. As indication that the quotient K is a natural object, and that its DT-invariant is nontrivial in general, we examine two examples in Section 4: Picard bundles, and Hilbert schemes of points. 2010 Mathematics Subject Classification. Primary 14N35; Secondary 14K05 14D20. 1 2 MARTIN G. GULBRANDSEN 1.1. Notation. Let π : X → S be a projective abelian scheme over a separated, noe- therian and connected base scheme S, itself defined over an algebraically closed field k of characteristic zero. In some sections we let k = C to apply a result of Mukai (Section 2.1), relating the Chern character of a complex with that of its Fourier -- Mukai transform. Fur- thermore, the characteristic zero assumption is used to ensure that line bundles deform freely. The group law is written m : X ×S X → X, and translation along an S ′-valued point x ∈ X(S ′) is written Tx : XS ′ → XS ′ . We let P denote the normalized Poincaré x ∈ X(S ′). L −1 ⊗ p∗ 2 m∗L ⊗ p∗ 1 L −1. We denote by Pξ the restriction (1 × ξ)∗P for a valued bundle on X ×S bX, wherebπ : bX → S is the dual abelian scheme. When L is a line bundle on X, we let φL : X → bX be the S-morphism defined by the line bundle point ξ ∈ bX(S ′). Abusing notation slightly, we also write Px for (x × 1)∗P when We write the Fourier -- Mukai transform of a complex E of OX -modules as bE ; it is a complex on bX. This is a deviation from the literature: the notation bE is usu- ally reserved for WIT-sheaves. The relative version of Mukai's Fourier inversion theorem [15, Theorem 1.1] gives a functorial isomorphism where g is the relative dimension of X over S. Since the Fourier -- Mukai transform may map a sheaf to an honest complex, i.e. having nontrivial cohomology in more than one degree, it is natural to work with complexes from the start. Complexes will be considered as objects in the derived category, and functors f∗, f ∗, ⊗, Hom etc. will mean their derived versions throughout. In particular, restriction of a complex to a subscheme, such as a fibre of a morphism, means derived restriction. We write H i and Exti for hyperco- homology and hyperext. The cohomology sheaves of a complex will be denoted hp(−). Thus, the classical p'th derived functor of, say f∗, will be written hpf∗(−). We use line bundle and vector bundle as synonyms for invertible sheaf and locally free sheaf. 1.2. Moduli spaces. A complex E of sheaves on a scheme X/S is perfect if it is Zariski locally isomorphic to a bounded complex of locally free sheaves of finite rank. It is simple if (1) bbE (cid:27) (−1)∗E ⊗ ωX/S[−g] Extp Xs (Es, Es) = p = 0, k 0 p < 0 for all k-valued points s ∈ S. (To save ink we thus let simplicity also encompass the property of having no negative Exts.) By a moduli space M/S for simple perfect complexes on X, we mean a subfunc- tor, representable by scheme or algebraic space of finite type/S, of the functor sending a scheme T to (cid:26)perfect simple com- plexes on X ×S T (cid:27),twist by Pic(T ) and quasi-isomorphism and which is locally complete: whenever a T -valued point E ∈ M(T ) extends to a complex E on X ×S T for an infinitesimal thickening T ⊂ T , then E is a T -valued point of M. Weaker notions of moduli spaces suffice, and we will allow Simpson moduli spaces M for (locally free resolutions of) stable sheaves, with fixed Hilbert poly- nomial. This is not quite covered by the definition above: a universal family E on M × X may only exist (étale) locally. Still the sheaf Hom(E , E ) does exist globally, DONALDSON -- THOMAS ON ABELIAN THREEFOLDS 3 and we will ignore that the family E itself may not exist as a sheaf. We are not concerned with stability conditions for complexes in general, but take the moduli space M as given. Eventually, we will assume that the moduli space M is proper, as we want to take degrees of zero cycles on it. 1.3. Obstruction theory. For the following construction, we refer to Huybrechts -- Thomas' treatment [9] of obstruction theory for perfect complexes, which we fol- low closely. Let LM/S denote the truncation to degrees ≥ −1 of the cotangent complex of M/S. If M embeds as a closed subscheme of a smooth scheme W over S, with ideal I ⊂ OW , this is when identifying (3) via Poincaré duality. bch = χ − γ + · · · + (−1)g−1c1 + (−1)g r H 2p(X, Z) (cid:27) H 2(g−p)(X, Z)∨ (cid:27) H 2(g−p)(bX, Z) To the divisor class c1 there is an associated homomorphism φc1 : X → bX. Via (3), the curve class γ corresponds to a divisor class on bX, (the first Chern class of bE , up to sign); denote the associated homomorphism by ψγ : bX → X. LM/S : (cid:16)I /I 2 → Ω W /S(cid:12)(cid:12)(cid:12)M(cid:17) with nonzero objects in degree −1 and 0. Associated to the universal complex E on X ×S M there is an element, the Atiyah 2LM/S is a direct summand of L X×S M/S , class of E , in Ext1(E , E ⊗L X×S M/S ). Since p∗ the Atiyah class projects to an element in Ext1(E , E ⊗ p∗ 2LM/S ) (cid:27) H 1(Hom(E , E ) ⊗ p∗ 2LM/S ) (cid:27) H 1(p2∗Hom(E , E ) ⊗ LM/S ) (cid:27) Ext1((p2∗Hom(E , E ))∨, LM/S ) and thus defines a morphism (2) (p2∗Hom(E , E ))∨[−1] → LM/S in the derived category. (By relative duality, the source may be identified with p2∗Hom(E , E ⊗ p∗ 1ωX/S )[g − 1], as is commonly done in the literature; we prefer not to.) Replacing Hom with the kernel of the trace map, and M with its subspace M(L ) of complexes with fixed determinant, the morphism (2) becomes a relative ob- struction theory [9, Theorem 4.1] in the sense of Behrend -- Fantechi [3]. 2. Deformations with fixed determinant and codeterminant 2.1. Chern class condition. In this section, S = Spec C and X is an abelian vari- ety of dimension g. Let ch be the Chern character in Lp H 2p(X, Z) of a perfect complex E , with homogeneous decomposition We view the rank r and the Euler characteristic χ as integers. By a result of Mukai ch = r + c1 + · · · + γ + χ. [15, Corollary 1.18], the Chern character of the Fourier -- Mukai transform bE is 4 MARTIN G. GULBRANDSEN Now let M be a moduli space of perfect complexes with Chern character ch, big enough to be stable under the actions of X by translation andbX = Pic0(X) by twist: a (valued) point (x, ξ) ∈ X ×bX acts on complexes E ∈ M by −x(E ) ⊗ Pξ , E 7→ T ∗ where the sign on x is for cosmetic reasons. The following is modelled on the constructions of Beauville [1, Section 7] and Yoshioka [23, Section 4.1] in the case of sheaves on abelian surfaces. The deter- (4) M × X define a morphism minant det(E ) and codeterminant det(bE ) line bundles of the universal family on (γ is viewed as a divisor class on bX). We want to use the action of X ×bX on M to δ : M → Pic−γ (bX) × Picc1(X) ensure that all fibres of δ are isomorphic. Condition 2.1. Assume that is an isogeny. χ −φc1 −ψγ r ! ∈ End(X ×bX) The condition is rather weak, as we will indicate in Proposition 3.5. Recall that an isotrivial fibration is a morphism that becomes the projection from a product after an étale base change. Proposition 2.2. Impose Condition 2.1. Then δ is an isotrivial fibration. Proof. Let E ∈ M and (L ′, L ) = δ(E ). Then det(T ∗ −x(E ) ⊗ Pξ ) = (T ∗ −x L ⊗ L −1) ⊗Prξ ⊗ L det( [T ∗ −x(E ) ⊗ Pξ ) = (T ∗ ξ ⊗Pχx ⊗ L ′ φc1 (−x) L ′ ⊗ (L ′)−1) {z } {z } ψ−γ (ξ) (use that the Fourier -- Mukai transform exchanges twist and translation, up to sign [15]). Thus, if X ×bX acts on the base of δ via the matrix in Condition 2.1 followed by addition in Pic(bX) × Pic(X), then δ is X ×bX-equivariant. −−−−−−→ X ×bX = Pic0(bX) × Pic0(X) y Pic−γ (bX) × Picc1 (X), It follows that, for any fibre M(L ′, L ) = δ−1(L ′, L ) of δ, there is a Cartesian X ×bX isogeny diagram (X ×bX) × M(L ′, L ) projection ❄ isogeny X ×bX action ✲ M δ ❄ ✲ Pic−γ (bX) × Picc1 (X) where the isogeny at the bottom is defined by the action on (L ′, L ). (cid:3) Recall the endomorphism construction studied by Morikawa [12] and Mat- if σ and τ are cycles of complementary dimension on an abelian susaka [11]: variety X, then α = α(σ, τ) is the endomorphism α(x) =Pστx −Pστ, DONALDSON -- THOMAS ON ABELIAN THREEFOLDS 5 where τx = Tx(τ) is the translated cycle, and the sum means addition of zero cycles using the group law on X. This endomorphism depends only on the numerical equivalence classes of σ and τ [11, Theorem 1]. Condition 2.1 may be checked on the "determinant" of the matrix appearing: Lemma 2.3. Condition 2.1 holds if and only if the homomorphism is an isogeny. rχ − α(γ, c1) ∈ End(X) Proof. The homomorphism in Condition 2.1 is an isogeny if and only if is an isogeny, and their composition is φc1 r r ! r φc1 −ψγ ψγ χ ! ∈ End(X ×bX) χ ! = χr − ψγ φc1 0 ψγ χ −φc1 0 χr − φc1 ψγ! . Thus Condition 2.1 holds if and only if this diagonal matrix is an isogeny. The two entries on the diagonal are duals of each other, so to prove the lemma it suffices to see that (5) ψγ ◦ φc1 = α(γ, c1). Both sides of (5) are Z-linear in γ. Thus it suffices to treat the case where γ is the class of an integral curve C ⊂ X. Let ν : eC → X be the normalization of C, and let J be the Jacobian of eC. There are induced Picard and Albanese maps ρ : bX → J It is straight forward to verify (see proof of [5, Prop. 11.6.1], where the Morikawa -- and σ : J → X. Matsusaka endomorphism is defined with opposite sign of ours) that α(γ, c1) = −σ ◦ ρ ◦ φc1 . Now choose a line bundle L on eC (taking OeC is ok). By Grothendieck -- Riemann -- Roch for ν, we have h ch(ν∗L )ig−1 = γ where the subscript g − 1 denotes the homogeneous component in H 2(g−1)(X, Z). Thus the first Chern class of [ν∗L corresponds to −γ via Poincaré duality (3), so with L ′ = det([ν∗L ), the homomorphism φL ′ : bX → X coincides with −ψγ . But now [19, Prop. 17.3] φL ′ = σ ◦ ρ, and we have established (5). (cid:3) Remark 2.4 (Matsusaka [11, Proposition 1]). If γ is proportional to c then α(γ, c1) is multiplication by the integer deg(γc1)/g. So in this case the endo- morphism in Lemma 2.3 is either zero or an isogeny. g−1 1 in H 2(g−1)(X, Q), 2.2. Diagonals and traces. Return to the relative situation of an abelian scheme X/S. Lemma 2.5. Let E and F be perfect complexes on X. (i) There is a canonical isomorphism of complexes in the derived category D(S). π∗Hom(E , F ) (cid:27)bπ∗Hom(bE ,cF ) 6 MARTIN G. GULBRANDSEN (ii) The induced isomorphism sends an element on the left, viewed as a derived category morphism E → F [i], (bE ,cF ) Proof. Consider the fibre diagram Exti X (E , F ) (cid:27) Exti bX to its Fourier -- Mukai transform bE → cF [i]. p2✲ bX ✲ Sbπ X ×S bX ❄ π X p1 ❄ and form the complex Hom(p∗ Push forward to bX to get 2bE , p∗ p2∗(Hom(p∗ 1 1 2bE , p∗ F ) ⊗ P ) (cid:27) p2∗Hom(p∗ F ⊗ P ) F ⊗ P )) F ) ⊗ P on X ×S bX. 2bE , p∗ (cid:27) Hom(bE , p2∗(p∗ (cid:27) Hom(bE ,cF ) 1 1 and push forward to X to get, by relative duality over p1, p1∗Hom(p∗ 1 2bE , p∗ F ) ⊗ P (cid:27) p1∗Hom(p∗ 2 (cid:27) Hom(p1∗(p∗ (cid:27) Hom(E , F ) E ⊗ P ∨, p∗ 1 F ) 2bE ⊗ P ∨ ⊗ ωp1[g]), F ) where g is the relative dimension of X/S, and where we used Fourier -- Mukai in- version in the last step. Thus Hom(p∗ 1 2bE , p∗ F )⊗P pushes forward to Hom(E , F ) on X and to Hom(bE ,cF ) The induced map in (ii) now arises as a composition Exti X (E , F ) (cid:27) Exti on bX. Push further down to S to obtain (i). 2bE , p∗ X×SbX adj −−→ p∗ 1 and the second isomorphism takes g : p∗ The first isomorphism takes f : E → F [i] to 2p2∗(p∗ 1 E ⊗ P ) (p∗ p∗ 1 F ⊗ P ) (cid:27) Exti bX p∗ 1(f )⊗P −−−−−−−−→ (p∗ 1 F ⊗ P [i] to (bE ,cF ). F ⊗ P )[i] E ⊗ P 2bE → p∗ 1 2bE = p∗ bE The composition of these is the Fourier -- Mukai transform. (cid:3) adj −−→ p2∗p∗ p2∗(g) −−−−−→ p2∗(p∗ 1 2bE F ⊗ P )[i] = cF [i]. Remark 2.6. The isomorphism β(E , F ) : π∗Hom(E , F ) →bπ∗Hom(bE ,cF ) from Lemma 2.5(i) is compatible with relative duality: the relative canonical sheaf ωX/S has trivial fibres, and hence can be written π∗η, where η is the restriction of ωX/S to the zero section. On the dual side, the canonical sheaf ωbX/S is bπ∗η DONALDSON -- THOMAS ON ABELIAN THREEFOLDS 7 for the same η. Thus relative duality gives the vertical arrows in the diagram of isomorphisms Hom(bπ∗Hom(bE ,cF ), OS ) bπ∗Hom(cF ,bE ) ⊗ η[g] ❄ β(E ,F )∨ ✲ Hom(π∗Hom(E , F ), OS ) β(F ,E )−1⊗η[g] ❄ ✲ π∗Hom(F , E ) ⊗ η[g] and this diagram is commutative. We leave out the (formal) verification of this statement. Let E be a perfect complex on the abelian scheme X/S, and let ι : π∗OX → π∗Hom(E , E ), tr : π∗Hom(E , E ) → π∗OX denote the homomorphisms induced by the canonical "diagonal" map OX → Hom(E , E ) Lemma 2.7. Form the composition and the trace map Hom(E , E ) → OX . Letbι andbtr denote the corresponding maps associated to bE on bX. bπ∗ObX bι and take first cohomology sheaves. The result coincides with the negative of the homo- morphism tr −→ π∗OX . −→bπ∗Hom(bE ,bE ) (cid:27) π∗Hom(E , E ) dφdet(E ) : h1bπ∗ObX → h1π∗OX Remark 2.8. The notation dφdet(E ) is a reminder of the following [6, Theorem 8.4.1]: if σ : S → X denotes the zero section, then the restriction σ ∗TX/S of the induced by φdet(E ) : X →bX. relative tangent bundle is canonically isomorphic to h1bπ∗ObX . The induced homo- morphism in the lemma can be viewed as the derivative of φdet(E ) along the zero section. Proof. The claim is local on S, so we may assume S = SpecR is affine. Then the map in question is the composition of certain (R-module-) homomorphisms (6) Ext1 bX (ObX , ObX ) bι −→ Ext1 bX (bE ,bE ) (cid:27) Ext1 X (E , E ) tr −→ Ext1 X(OX , OX ). View elements of these Ext1-groups as first order infinitesimal deformations of the argument, i.e. families of perfect complexes over S[ǫ] = SpecR[ǫ], where ǫ2 = 0. The key point in what follows is that the trace map sends a first order deformation to its determinant. Since X is the dual of bX, first order deformations of ObX correspond to mor- phisms f : S[ǫ] → X extending the zero section σ : S → X. In these terms, the homomorphism dφdet(E ) in the Lemma sends f to φdet(E ) ◦ f . By definition of φdet(E ), the corresponding deformation of OX is (7) f (q∗ det(E )) ⊗ q∗ det(E )−1 ∈ Pic(X ×S S[ǫ]) T ∗ where q is projection to X. We want to compare this with the image of f through the string (6) of homomorphisms. The Fourier -- Mukai transform yields an isomorphism Let F = (f , 1) : S[ǫ] → X ×S S[ǫ]. Then the first order deformation of σ∗OS corre- sponding to f is F∗OS[ǫ]. Ext1 X (σ∗OS , σ∗OS ) (cid:27) Ext1 bX (ObX , ObX ). 8 MARTIN G. GULBRANDSEN Underbι, the deformation [F∗OS[ǫ] is sent to the tensor product [F∗OS[ǫ] ⊗dq∗E ∈ Ext1 X (bE ,bE ). The Fourier -- Mukai transform exchanges tensor product and Pontryagin product (cf. Mukai [15, §3.7], where the argument applies also relative to a base), defined as A ⋆ B = m∗(p∗ 2B). Thus we arrive at 1A ⊗ p∗ Using the shorthand X[ǫ] = X ×S S[ǫ] and with an eye at the commutative diagram (F∗OS[ǫ]) ⋆ (q∗E ) ∈ Ext1 X (E , E ). p2 f m X[ǫ] ✛ Tf ❄ X[ǫ] ✛ S[ǫ] ×S[ǫ] X[ǫ] F×1 ❄ X[ǫ] ×S[ǫ] X[ǫ] p1 ✲ S[ǫ] F ❄ p1 ✲ X[ǫ] we write out what this means: (F∗OS[ǫ]) ⋆ (q∗E ) = m∗(cid:16)p∗ 1(F∗OS[ǫ]) ⊗ (p∗ 2q∗E )) = m∗ ((F × 1)∗(p∗ = Tf ∗(q∗E ). 2q∗E )(cid:17) Now Tf ∗ = T ∗ yields its determinant, i.e. −f , and the trace map applied to the first order deformation T ∗ −f (q∗E ) This is, by the theorem of the square, the inverse to the invertible sheaf (7). (cid:3) −f (q∗ det(E )) ⊗ q∗ det(E )−1. T ∗ 2.3. Splitting off traces. To define Donaldson -- Thomas invariants, we are going to construct a perfect symmetric obstruction theory on M(L ′, L ). However, to obtain deformation invariance of the resulting invariant, which is a key point, a relative version of the obstruction theory is needed. So we extend the setup from Section 2.1 to the relative situation of an abelian scheme X/S. say that a complex E on X ×S T has first Chern class L if φdet(E ) = φL ×S T as Fix integers r and χ together with line bundles L on X and L ′ on bX. We morphisms X ×S T →bX ×S T . Let M/S be a moduli space of perfect complexes on X/S with rank r, first Chern class L , and with Fourier -- Mukai transform of rank χ and first Chern class L ′. Let E denote the universal family (possibly defined only locally on M, as in Sec- tion 1.2). Recall that an isogeny of abelian schemes means a finite and surjective morphism of group schemes. Such a morphism is necessarily flat [17, Lemma 6.12]. Condition 2.9. Assume that χ −φL is an isogeny. φL ′ r ! ∈ End(X ×S bX) When S = Spec C, the endomorphism in Condition 2.9 agree with the one in Condition 2.1. Proposition 2.10. If the endomorphism in Condition 2.9 restricts to an isogeny in some closed fibre Xs ×bXs, then it is an isogeny globally. DONALDSON -- THOMAS ON ABELIAN THREEFOLDS 9 Proof. Let f denote the endomorphism in Condition 2.9. The locus of points s ∈ S such that fs is an isogeny is open (as the locus where Ker(f ) → S is finite) and closed (as the locus over which fs is surjective), and nonempty by assumption. Thus fs is an isogeny for all s ∈ S. So f is finite and surjective, i.e. an isogeny of abelian schemes. (cid:3) Lemma 2.5 applies to the projection p2 : X ×S M → M, which itself is an abelian scheme, with dual p2 : bX×S M → M. Thus we identify p2∗Hom(E , E ) = p2∗Hom(bE ,bE ), and then there are canonical (co)diagonal and (co)trace maps ι : p2∗OX×S M → p2∗Hom(E , E ), tr : p2∗Hom(E , E ) → p2∗OX×S M , cohomology sheaves of the complex p2∗OX×S M are hi (π∗OX ) ⊗OS on the dual side, the result can be written OM → Exti bι : p2∗ObX×S M → p2∗Hom(E , E ), Form the composition (btr, tr) ◦ (bι + ι) and take cohomology sheaves. Since the (cid:16)hi (bπ∗ObX ) ⊕ hi (π∗OX )(cid:17) ⊗OS For i = 1, view h1(bπ∗ObX ) ⊕ h1(π∗OX ) in (8) as the relative tangent space to X ×SbX btr : p2∗Hom(E , E ) → p2∗ObX×S M . p2(E , E ) →(cid:16)hi (bπ∗ObX ) ⊕ hi(π∗OX )(cid:17) ⊗OS along the zero section, as in Remark 2.8. OM , and similarly OM (8) Lemma 2.11. For i = 1, the composition (8) is the endomorphism (9) χ of h1(bπ∗ObX ) ⊕ h1(π∗OX ), tensored with OM . −dφL Proof. The bottom right entry is h1p2∗ applied to tr −→ OX×S M , ι −→ Hom(E , E ) OX×S M dφL ′ r ! , but this is multiplication by the rank r of E already before the application of the derived push forward. Similarly, the top left entry is multiplication by the rank of The bottom left entry is −dφdet(E ) by Lemma 2.7, and φdet(E ) = φL ×S M. Ex- bE . changing the roles of (X, E ) and (bX,bE ), we see that the top right entry is dφdet(bE ) = (The sign change comes from the (−1)∗ in Mukai's isomorphism (cid:3) d(φL ′ ×S M). (1).) Lemma 2.12. Assume X/S has dimension g = 3 and Condition 2.9 holds. Then the truncation τ[1,2](cid:16)p2∗Hom(E , E )(cid:17) → τ[1,2](cid:16)p2∗ObX×S M ⊕ p2∗OX×S M(cid:17) of (btr, tr) is a split epimorphism. Proof. It suffices to show that the truncation of (btr, tr) ◦(bι +ι) to [1, 2] is an automor- phism of τ[1,2](cid:16)p2∗ObX×S M ⊕ p2∗OX×S M(cid:17) in the derived category. In fact we claim: for X/S of arbitrary dimension g, the composition (8) is an isomorphism for i = 1 and i = g − 1. For i = 1, the claim follows from Lemma 2.11, since Condition 2.9 implies that (9) is an automorphism of h1(bπ∗ObX ) ⊕ h1(π∗OX ). The case i = g −1 follows from duality: the complexes OX×S M , ObX×S M Hom(E , E ) and Hom(bE ,bE ) are all self dual in the derived sense, and the duals of the diagonal 10 MARTIN G. GULBRANDSEN maps ι andbι, upstairs on X ×S M andbX ×S M, are canonically the trace maps tr and btr. Thus, by relative duality over M and Remark 2.6, the dual of β (cid:27) p2∗Hom(E , E ) tr −→ p2∗OX×S M is, canonically, p2∗ObX×S M bι p2∗ObX×M[g] btr[g] −→ p2∗Hom(bE ,bE ) ←−−−− p2∗Hom(bE ,bE )[g] β[g] (cid:27) p2∗Hom(E , E )[g] ι[g] ←−−− p2∗OX×M [g] tensored with a line bundle η (as in Remark 2.6). So, suppressing β again, the dual of tr ◦bι is (btr ◦ ι) ⊗ η[g]. It follows that the dual of (btr, tr) ◦ (bι + ι) is (tr,btr) ◦ (ι +bι) ⊗ η[g]. Knowing that the first cohomology of (btr, tr)◦(bι+ι) is an isomorphism, we conclude that its g − 1'st cohomology is an isomorphism, too. Remark 2.13. By working with moduli spaces for complexes with fixed determi- nant, and the corresponding trace free obstruction theory, we only get rid of one of the trivial summands in Lemma 2.12, and the associated virtual fundamental class is still zero. So in order to get a nontrivial invariant, we will fix the codeterminant, too. (cid:3) 2.4. The obstruction theory. Let M(L ′, L ) ⊂ M denote the sub moduli space parametrizing complexes with fixed determinant L and codeterminant L ′, i.e. the fibre product M(L ′, L ) ✲ M δ ❄ ❄ S ✲ Pic(bX/S) ×S Pic(X/S) where the lowermost horizontal arrow is defined by (L ′, L ), and δ is defined by (det(bE ), det(E )). Recall that Behrend -- Fantechi [4] say that a perfect obstruction theory F → LM over k is symmetric if it is equipped with an isomorphism θ : F → F ∨[1] satisfying θ∨[1] = θ. We extend the definition to the relative situation by working modulo line bundles from the base. Definition 2.14. A relative perfect obstruction theory F → LM/S is symmetric if there is a line bundle η on S and an isomorphism θ : F → F ∨ ⊗OS η[1] satisfying θ∨[1] ⊗OS η = θ. (10) Let F denote the kernel of the split epimorphism in Lemma 2.12, so that τ[1,2]p2∗Hom(E , E ) (cid:27) F ⊕ τ[1,2](cid:16)p2∗ObX×S M ⊕ p2∗OX×S M(cid:17) . Theorem 2.15. Assume Condition 2.9. The morphism (2) induces a morphism F ∨[−1] → LM/S , whose restriction to M(L ′, L ) is a relative perfect symmetric obstruction theory. Proof. The arguments of Huybrechts -- Thomas [9, Theorem 4.1 and Section 4.4] work with minor additions. Simplicity of E combined with Serre duality shows that the diagonal and trace maps induce isomorphisms OM = h0p2∗OX×S M ∼→ h0p2∗Hom(E , E ), h3p2∗Hom(E , E ) ∼→ h3p2∗OX×S M (cid:27) η ∨ ⊗OS OM DONALDSON -- THOMAS ON ABELIAN THREEFOLDS 11 (the last isomorphism by relative duality for π; as before η is the line bundle for which ωX/S = π∗η) and thus there are distinguished triangles OM → p2∗Hom(E , E ) → τ≥1p2∗Hom(E , E ), τ[1,2]p2∗Hom(E , E ) → τ≥1p2∗Hom(E , E ) → η ∨ ⊗OS OM [−3]. (11) By these triangles (and the fact that LM/S is, by definition, concentrated in de- grees [−1, 0]), the morphism (2) induces (12) (τ[1,2]p2∗Hom(E , E ))∨[−1] → LM/S and hence, by the direct sum decomposition (10), F ∨[−1] → LM/S . (13) For every deformation situation over S, f : T → M, T ⊂ T where T ⊂ T is an affine square zero thickening with ideal I ⊂ O gives rise to a class T , the map (12) ω ∈ Ext2(f ∗(τ[1,2]p2∗Hom(E , E ))∨, I ) (cid:27) H 2(Hom(ET , ET ) ⊗OT I ) which is in fact an obstruction class [9, Main theorem], i.e., it vanishes if and only if f extends to T . Now ω decomposes as ω = (ω0, ω1, ω2) ∈ H 2(f ∗F ⊗ I ) ⊕ H 2(I ) ⊕ H 2(I ). Here, ω0 ∈ Ext2(f ∗(F ∨), I ) (cid:27) H 2(f ∗F ⊗ I ) is the class induced by (13), and ω1 and ω2 are the obstructions to deforming det(ET ) and det(bET ), as the trace map sends the obstruction class of a perfect com- plex to the obstruction class of its determinant. Since line bundles deform freely in characteristic zero, ω1 and ω2 vanish, and so ω0 vanishes if and only ω does. Thus ω0 is an obstruction class. If in addition det(ET ) is constant (in the sense of being a pullback from the base), then the set of extensions T → M of f with constant determinant is a torsor under the natural action of the trace free part of Ext1(f ∗(τ[1,2]p2∗Hom(E , E ))∨, I ) (cid:27) H 1(Hom(ET , ET ) ⊗OT I ). Thus, if f factors through M(L ′, L ), so that both det(ET ) and det(bET ) are con- stant, then extensions T → M(L ′, L ) of f form a torsor under the trace- and cotrace free part of H 1(Hom(ET , ET ) ⊗OT I ), which is H 1(f ∗F ⊗ I ). Thus the restriction of (13) to M(L ′, L ) is a relative obstruction theory. Whenever two terms in a distinguished triangle is perfect, so is the third, and thus F is perfect. For every point i : Spec k → M, the vector space hp(i∗F ) is the trace- and cotrace free part of Extp(i∗E , i∗E ), and in particular vanishes for p < [1, 2]. Thus, by an application of Nakayama's Lemma, the perfect complex F has amplitude ⊆ [1, 2]. By the two distinguished triangles (2.4), it follows that relative duality (p2∗Hom(E , E ))∨ (cid:27) p2∗Hom(E , E ) ⊗OS η[3] induces (τ[1,2]p2∗Hom(E , E ))∨ (cid:27) (τ[1,2]p2∗Hom(E , E )) ⊗OS η[3]. Since the direct summand F is both the kernel of the trace/cotrace map and the cokernel of the diagonal/codiagonal map, and these two maps are dual, there is an induced isomorphism F ∨ (cid:27) F ⊗OS η[3]. Thus the obstruction theory (13) (re- stricted to M(L ′, L )) is perfect and symmetric. (cid:3) 12 MARTIN G. GULBRANDSEN 3. Perfect complexes modulo twist and translate Let X be an abelian threefold over S = Spec k. With X × bX acting on a moduli space M of perfect complexes as before, by translation and twist, define the stack quotient K = [M/X ×bX]. In this section we rephrase the results from the previous section, and as outcome we obtain a numerical invariant of K. This is our Donaldson -- Thomas invariant for abelian threefolds. We pause to explain our point of view: suppose Y is a second abelian variety, derived equivalent to X, i.e. there is an equivalence F : D(X) ∼→ D(Y ) of triangu- lated categories. By Orlov [18], there is an object C ∈ D(X × Y ) on the product, such that F is the Fourier -- Mukai functor induced by C : F(−) = p2∗(p∗ 1(−) ⊗ C ) Consequently the equivalence F extends in a natural way to families, i.e. there are equivalences FT : D(X × T ) → D(Y × T ), functorial in T . In particular, if E is the universal family on X × M, then we may replace E by FM (E ) and view M as a moduli space for complexes on Y . We wish to treat (X, E ) and (Y , F(E )) on an equal footing, in particular our Donaldson -- Thomas invariants should be invariant under derived equivalence. The equivalence F induces an isomorphism such that [18, Corollary 2.13] T ∗ y F(−) ⊗ ζ (cid:27) F(T ∗ x (−) ⊗ ξ) f : X ×bX ∼→ Y ×bY whenever (y, ζ) = f (x, ξ). In particular the actions of X ×bX and Y ×bY on M are compatible, and there are induced isomorphisms [M/X ×bX] (cid:27) [M/Y ×bY ], so the construction of K is invariant under derived equivalence. (In contrast, the space M(L ′, L ) is not.) This setup seems to be natural also for not necessar- ily abelian varieties: Rosay [20] has shown that, for any smooth projective va- riety X, the identity component of the algebraic group Aut(D(X)) of autoequiv- alences is in fact Aut0(X) × Pic0(X), and by an unpublished result of Rouquier [8, Proposition 9.45], any equivalence F : D(X) ∼→ D(Y ) induces an isomorphism Aut0(X) × Pic0(X) ∼→ Aut0(Y ) × Pic0(Y ). Proposition 3.1. Assume Condition 2.9, and let G be the kernel of the isogeny appear- ing there. Then there is an isomorphism In particular K is Deligne-Mumford. [M(L ′, L )/G] ∼→ K. Proof. As we saw in the proof of Proposition 2.2, the determinant/codeterminant morphism (4) is X ×bX-equivariant, when the action on the base is defined via the matrix in Condition 2.9. The statement follows. Although it is easy to verify that the obstruction theory in Theorem 2.15 is G-equivariant as an element in the derived category of M(L ′, L ), this does not suffice to conclude that there is an induced obstruction theory on the quotient K. Bypassing this difficulty, we define the Donaldson -- Thomas invariant directly, as what it is if the obstruction theory does descend: let [M(L ′, L )]vir be the virtual fundamental class associated to the obstruction theory of Theorem 2.15. (cid:3) DONALDSON -- THOMAS ON ABELIAN THREEFOLDS 13 Definition 3.2. Assume Condition 2.9 and that M(L ′, L ) is proper over k. The Donaldson -- Thomas invariant of the Deligne-Mumford stack K is DT(K) = 1 G deg[M(L ′, L )]vir. Corollary 3.3 (of Theorem 2.15). Let X/S be an abelian scheme of relative dimension 3, and let M/S be a proper moduli space for perfect complexes, satisfying Condition 2.9. Then the Donaldson -- Thomas invariant DT(Ks) associated to Ks = [Ms/Xs ×bXs] is independent of s ∈ S. Proof. Since the endomorphism in Condition 2.1 is an isogeny, its kernel G is flat and finite over S, so its order Gs is constant in s. The existence of a relative obstruction theory implies (see [3, Proposition 7.2] and [9, Corollary 4.3]) that the degree of the virtual fundamental class of Ms(L ′ is constant in s. Hence the Donaldson -- Thomas invariant s , Ls)]vir. deg[Ms(L ′ DT(Ks) = s , Ls) 1 Gs is independent of s. (cid:3) Recall Behrend's theorem [2] on weighted Euler characteristics: whenever a proper k-scheme (or Deligne-Mumford stack) M admits a perfect symmetric ob- struction theory, the degree of the associated virtual fundamental class agrees with the weighted Euler characteristic χ(M) =Xn∈Z nχ(ν −1(n)) where χ is the usual topological Euler characteristic, and ν : M → Z is Behrend's invariant of singularities [2, Definition 1.4]. Corollary 3.4 (of Behrend's theorem). Assume Condition 2.1. The Donaldson -- Thomas invariant DT(K) agrees with Behrend's weighted Euler characteristic χ(K), and hence is an intrinsic invariant of the Deligne-Mumford stack K. Proof. The quotient map M(L ′, L ) → K is étale of degree G, so DT(K) = 1 G χ(M(L ′, L )) = χ(K). (cid:3) We return to Condition 2.1, with the viewpoint that we are free to replace (X, E ) with a derived equivalent pair. A coherent sheaf E , on an abelian variety of di- mension g, is semi-homogeneous, if the locus (14) Φ(E ) =n(x, ξ) ∈ X ×bX T ∗ x E (cid:27) E ⊗ Pξo has dimension g (the maximal possible) [13, 16]. The proof of Proposition 2.2 shows that Condition 2.1 implies that Φ(E ) is finite, so semi-homogeneous sheaves fail the condition. Conversely: Proposition 3.5. Let X be an abelian threefold over C with Picard number 1. If E is a coherent sheaf on X, such that F(E ) fails Condition 2.1 for all autoequivalences F of D(X), then E is semi-homogeneous. Proof. Semi-homogeneity is invariant under derived equivalence, so we are free to replace E with any transform F(E ). The twist of any sheaf with a sufficiently positive divisor is IT0, and the Fourier -- Mukai transform of an IT0-sheaf is a vector bundle. Thus we may assume E is a vector bundle, using the derived equivalence F defined as the composition of a 14 MARTIN G. GULBRANDSEN very positive twist, the Fourier -- Mukai transform, another very positive twist on bX, and the Fourier -- Mukai transform back to X. Semi-homogeneous vector bundles are numerically characterized [22] by hav- ing Chern character ch = r exp(c1/r). In dimension 3, this means that rγ = c2 1/2 and rχ = γc1/3. Since the Picard number is 1, we have α(σ, τ) = στ/3 for any two complementary cycles σ and τ, by Remark 2.4. Thus rχ − α(γ, c1) = 0 implies rχ = γc1/3. Let H be an ample divisor, denote its cohomology class by h, and use primed symbols to denote the components of the Chern character ch(E ) exp(h) of E (H). A small computation, using basic properties of α [11, Propo- sition 1], shows that (cid:16)rχ − α(γ, c1)(cid:17) −(cid:16)r ′χ′ − α(γ ′, c′ 1)(cid:17) = (rγ − c2 1/2)h − α(rγ − c2 1/2, h) (this holds without any condition on the Picard number). By assumption, the left hand side is zero, and the vanishing of the right hand side implies that rγ = c2 1/2. (cid:3) In relation to the proposition, one may ask whether there exist honest simple semi-homogeneous complexes. Here is a weaker, but easy observation: Remark 3.6. Let X be an abelian variety of dimension g, with a perfect complex E having no twisted negative Ext-groups: Ext−p(E , E ⊗ Pξ ) = 0, ∀ p > 0 and ξ ∈bX Suppose E is semi-homogeneous in the sense that the locus Φ(E ) in Equation (14) has dimension g. We claim that E is isomorphic to a single semi-homogeneous sheaf, up to shift. In fact, since Φ(E ) ⊆ Φ(hp(E )), each hp(E ) is semi-homogeneous as well. By Mukai's classification of semi-homogeneous vector bundles [13, Section 6] (and a little care to cover non locally free semi-homogeneous sheaves), each hp(E ) admits a Jordan-Hölder filtration with simple semi-homogeneous factors, and all these hp(E ) , 0 for p = a, b, there are a simple semi-homogeneous subsheaf F ⊆ ha(E ), factors agree up to twist by bX. Thus if a is minimal and b is maximal such that an element ξ ∈bX, and a surjection hb(E ) ։ F ⊗ Pξ . The composition E [−b] → hb(E ) ։ F ⊗ Pξ ֒→ ha(E ⊗ Pξ ) → E ⊗ Pξ [−a] is an element in Exta−b(E , E ⊗Pξ ), and it is nonzero since it induces a nonzero map on h0. By hypothesis we cannot have a − b < 0, so a = b. 4.1. Picard bundles. Let i : C ֒→ X be a smooth projective curve embedded in its Jacobian X = J(C) via the Abel-Jacobi map, and let L be a line bundle on C. 4. Examples (possibly shifted) vector bundle: such bundles are called Picard bundles. One of Mukai's original applications for his transform [14] was to equate deformations of If L is sufficiently positive or negative, the Fourier -- Mukai transform di∗L is a Picard bundles di∗L with deformations of i∗L : let M be the connected component containing i∗L , of the Simpson moduli space for stable sheaves with respect to an arbitrary polarization. The action (15) X ×bX → M, (x, ξ) 7→ T ∗ −x(i∗L ) ⊗ Pξ DONALDSON -- THOMAS ON ABELIAN THREEFOLDS 15 is an isomorphism whenever C is nonhyperelliptic or has genus two [14, Theorem 4.8]. Thus K = [M/X ×bX] is a reduced point and χ(K) = ν(point) = 1 (C nonhyperelliptic or g(C) = 2) For hyperelliptic C of genus g > 3, the map (15) is an isomorphism onto Mred, but M is everywhere nonreduced [15, Example 1.15] (this comes from obstructed first order deformations of C in X). Thus K is a point with a nonreduced scheme structure and χ(K) = ν(nonreduced point) > 1 (C hyperelliptic and g(C) > 2). There is, of course, nothing "virtual" about these counts, which make sense for any dimension g. 4.2. Hilbert schemes of points. Let M be the moduli space for coherent sheaves with rank 1, c1 = 0, γ = 0 and χ = −n, on an abelian threefold X. Any such sheaf is Thus M =bX × Hilbn(X). a twisted ideal IZ ⊗ Pξ , where Z ⊂ X is a finite subscheme of length n and ξ ∈bX. Projection to bX agrees with the determinant map on M, and projection to the agrees, up to sign, with the codeterminant map. Thus M(ObX , OX ) is the fibre over 0 for (16), and will henceforth be denoted K n(X) (this would have been Beauville's generalized Kummer variety if X were a surface). Condition 2.1 holds, and the kernel of the isogeny there is the group of n-torsion points Hilbert scheme followed by the summation map Hilbn(X) → X (16) Thus K = [K n(X)/Xn]. Xn = Xn × 0 ⊂ X ×bX. Behrend -- Fantechi [4] show that for Hilbert schemes of points on threefolds, the weighted Euler characteristic and the topological Euler characteristic agree up to sign. Their method is easily adapted to K n(X) and gives χ(K n(X)) = (−1)n+1χ(K n(X)). We also mention that Cheah's formula [7] for the Euler characteristic of any Hilbert scheme of points leads us to conjecture that, for X an abelian variety of dimension g, exp(cid:18) ∞Xn=1 χ(K n(X)) n2g tn(cid:19) = ∞Xn=0 Pg−1(n)tn where Pd(n) is the number of d-dimensional partitions of n. (For g = 1 and g = 2, the formula does hold.) Using MacMahon's generating function for plane parti- tions, the conjectured formula for g = 3 can be written χ(K n(X)) = n5Xdn d2 giving, still conjecturally, DT(K) = (−1)n+1 Xn χ(K n(X)) = (−1)n+1 n Xdn d2. 16 MARTIN G. GULBRANDSEN References 1. A. Beauville, Variétés Kähleriennes dont la première classe de Chern est nulle, J. Differential Geom. 18 (1983), no. 4, 755 -- 782 (1984). 2. K. Behrend, Donaldson-Thomas type invariants via microlocal geometry, Ann. of Math. (2) 170 (2009), no. 3, 1307 -- 1338. 3. K. Behrend and B. Fantechi, The intrinsic normal cone, Invent. Math. 128 (1997), no. 1, 45 -- 88. 4. , Symmetric obstruction theories and Hilbert schemes of points on threefolds, Algebra Number Theory 2 (2008), no. 3, 313 -- 345. 5. C. Birkenhake and H. Lange, Complex abelian varieties, second ed., Grundlehren der Mathema- tischen Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 302, Springer- Verlag, Berlin, 2004. 6. S. Bosch, W. Lütkebohmert, and M. Raynaud, Néron models, Ergebnisse der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)], vol. 21, Springer-Verlag, Berlin, 1990. 7. J. Cheah, On the cohomology of Hilbert schemes of points, J. Algebraic Geom. 5 (1996), no. 3, 479 -- 511. 8. D. Huybrechts, Fourier-Mukai transforms in algebraic geometry, Oxford Mathematical Monographs, The Clarendon Press Oxford University Press, Oxford, 2006. 9. D. Huybrechts and R. P. Thomas, Deformation-obstruction theory for complexes via Atiyah and Kodaira-Spencer classes, Math. Ann. 346 (2010), no. 3, 545 -- 569. 10. J. Li and G. Tian, Virtual moduli cycles and Gromov-Witten invariants of algebraic varieties, J. Amer. Math. Soc. 11 (1998), no. 1, 119 -- 174. 11. T. Matsusaka, On a characterization of a Jacobian variety, Memo. Coll. Sci. Univ. Kyoto. Ser. A. Math. 32 (1959), 1 -- 19. 12. H. Morikawa, Cycles and endomorphisms of abelian varieties, Nagoya Math. J. 7 (1954), 95 -- 102. 13. S. Mukai, Semi-homogeneous vector bundles on an Abelian variety, J. Math. Kyoto Univ. 18 (1978), 14. 15. 16. no. 2, 239 -- 272. , Duality between D(X) and D( X) with its application to Picard sheaves, Nagoya Math. J. 81 (1981), 153 -- 175. , Fourier functor and its application to the moduli of bundles on an abelian variety, Algebraic ge- ometry, Sendai, 1985, Adv. Stud. Pure Math., vol. 10, North-Holland, Amsterdam, 1987, pp. 515 -- 550. , Abelian variety and spin representation, Proceedings of symposium "Hodge theory and al- gebraic geometry" (Sapporo), 1994, [In Japanese, English translation: Univ. of Warwick preprint, 1998], pp. 110 -- 135. 17. D. Mumford, Geometric invariant theory, Ergebnisse der Mathematik und ihrer Grenzgebiete, Neue Folge, Band 34, Springer-Verlag, Berlin, 1965. 18. D. O. Orlov, Derived categories of coherent sheaves on abelian varieties and equivalences between them, Izv. Ross. Akad. Nauk Ser. Mat. 66 (2002), no. 3, 131 -- 158. 19. A. Polishchuk, Abelian varieties, theta functions and the Fourier transform, Cambridge Tracts in Mathematics, vol. 153, Cambridge University Press, Cambridge, 2003. 20. F. Rosay, Some remarks on the group of derived autoequivalences, 2009, arXiv:0907.3880v1 [math.AG]. 21. R. P. Thomas, A holomorphic Casson invariant for Calabi-Yau 3-folds, and bundles on K 3 fibrations, J. Differential Geom. 54 (2000), no. 2, 367 -- 438. 22. J.-H. Yang, Holomorphic vector bundles over complex tori, J. Korean Math. Soc. 26 (1989), no. 1, 117 -- 142. 23. K. Yoshioka, Moduli spaces of stable sheaves on abelian surfaces, Math. Ann. 321 (2001), no. 4, 817 -- 884. Stord/Haugesund University College, Norway E-mail address: [email protected]
1911.01796
2
1911
2019-11-06T13:10:23
Hilbert Schemes, Donaldson-Thomas Theory, Vafa-Witten and Seiberg Witten theories
[ "math.AG", "hep-th", "math.DG" ]
This article provides a summary of arXiv:1701.08899 and arXiv:1701.08902 where the authors studied the enumerative geometry of nested Hilbert schemes of points and curves on algebraic surfaces and their connections to threefold theories, and in particular relevant Donaldson-Thomas, Vafa-Witten and Seiberg-Witten theories.
math.AG
math
HILBERT SCHEMES, DONALDSON-THOMAS THEORY, VAFA-WITTEN AND SEIBERG WITTEN THEORIES ARTAN SHESHMANI Abstract. This article provides the summary of [GSY17a] and [GSY17b] where the authors studied the enumerative geometry of "nested Hilbert schemes" of points and curves on algebraic surfaces and their connections to threefold the- ories, and in particular relevant Donaldson-Thomas, Vafa-Witten and Seiberg- Witten theories. Contents 1. Introduction Acknowledgment 2. Nested Hilbert schemes on surfaces 2.1. Special cases 2.2. Nested Hilbert scheme of points 3. Nested Hilbert schemes and DT theory of local surfaces References 1 3 3 4 5 8 10 1. Introduction In recent years, there has been extensive mathematical progress in enumerative geometry of surfaces deeply related to physical structures, e.g. around Gopakumar- Vafa invariants (GV); Gromov-Witten (GW), Donaldson-Thomas (DT), as well as Pandharipande-Thomas (PT) invariants of surfaces; and their "motivic lifts". There is also a tremendous energy in the study of mirror symmetry of surfaces from the mathematics side, especially Homological Mirror Symmetry. On the other hand, physical dualities in Gauge and String theory, such as Montonen-Olive duality and heterotic/Type II duality have also been a rich source of spectacular predictions about enumerative geometry of moduli spaces on surfaces. For instance an extensive research activity carried out during the past years was to prove the modularity properties of GW or DT invariants as suggested by the heterotic/Type II duality. The first prediction of this type, the Yau-Zaslow conjecture [YZ96], 1 NESTED HILBERT SCHEMES AND LOCAL DT THEORY 2 was proven by Klemm-Maulik-Pandharipande-Scheidegger [KMPS10]. Further re- cent developments in this particularly fruitful direction include: Pandharipande- Thomas proof [PT16] of the Katz-Klemm-Vafa conjecture [KKV99] for K3 sur- faces, Maulik-Pandharipande proof of modularity of GW invariants for K3 fibered threefolds [MP13], Gholampour-Sheshmani-Toda proof of modularity of PT and Gholampour-Sheshmani proof of modularity of DT invariants of stable sheaves on K3 fibrations [GST17, GS18] (moreover, the generalizations of the latter in [BCDDQS16]), and finally Gholampur-Sheshmani-Thomas [GS14] proof of mod- ular property of the counting of curves on surfaces deforming freely (in a fixed linear system) in ambient CY threefolds. SU (2)-Seiberg-Witten (SW) / DT correspondence. The introduction of other versions of gauge theories in dimensions 6 started numerous exciting developments in physical mathematics and mathematical physics involving enumerative geom- etry, mirror symmetry, and related physics of 4 dimensional manifolds, realized as a complex two dimensional subvariety which could exhibit deformations inside ambient higher dimensional target varieties, such as CY threefolds. The SW/DT correspondence is one of such platforms to find interesting structural symmetries between theory of surfaces and theory of threefolds: having fixed a spin structure on a complex surface S the SW invariant of S roughly counts with sign the num- ber of points in the parametrizing space (the moduli space) of solutions to SW equations defined on S [D96, Section 1]. The focus of SW theory in physics is the study of moduli space of vacua in N = 2, D = 4 super Yang-Mills theory, and in particular certain dualities of the theory such as electric-magnetic duality (Montonen-Olive duality). In [GLSY17] Gukov-Liu-Sheshmani-Yau conjectured a relation between SW invariants (associate to SU (2) gauge invariant theory) of a projective surface S and DT invariants of a noncompact threefold X obtained by total space of a line bundle L on S. In a later work [GSY17a, GSY17b] , which will be elaborated below, Gholampour-Sheshmani-Yau proved the conjecture in [GLSY17, Equation 56]. The key object which bridges the geometry of complex sur- face to the ambient noncompact complex threefold is the "nested Hilbert scheme of surface". This space parametrizes a nested chain of configurations of curves and points in the surface (pic- ture on the right). Let us assume that S, as shown in the picture, is a projective simply connected complex surface and let L be a line bundle on S. In [GSY17a], Gholampour-Sheshmani-Yau analyzed the moduli space of stable compactly supported sheaves of modules on the noncompact threefold X. In physics terms these are D4-D2-D0 branes wrap- ping the zero section of X (i.e. sheaves are supported on S or a fat neighborhood of S in X). The authors showed in [GSY17a] that the DT invariants of X satisfy an equation in terms of SW invariants of S and certain correction terms governed by invariants of nested Hilbert schemes: NESTED HILBERT SCHEMES AND LOCAL DT THEORY 3 DT invariants of X= (SW invariants of S)· (Combinatorial coefficients) + Invariants of nested Hilbert scheme of S Their main exciting realization was when L = KS, the canonical bundle of S. In this case the above DT invariants of X recover some well known invariants in physics, the Vafa-Witten (VW) invariants [VW94], which are known to have modularity property, following the work of Vafa-Witten [VW94] (and verified by Tanaka-Thomas [TT17), TT18(2)] who defined and computed these invariants mathematically). On the other hand the authors also showed that in some cases the nested Hilbert scheme invariants also have modular property [GSY17a, Theorem 7]. Therefore, we were able to show that in some cases, the SW invariants can be described in terms of modular forms. Acknowledgment The author was partially supported by NSF DMS-1607871, NSF DMS-1306313 and Laboratory of Mirror Symmetry NRU HSE, RF Government grant, ag. No 14.641.31.0001. The author would like to further sincerely thank the Center for Mathematical Sciences and Applications at Harvard University, the center for Quantum Geometry of Moduli Spaces at Aarhus University, and the Laboratory of Mirror Symmetry in Higher School of Economics, Russian federation, for the great help and support. 2. Nested Hilbert schemes on surfaces Hilbert schemes of points and curves on a nonsingular surface S have been vastly studied. Their rich geometric structures have proved to have many applications in mathematics and physics (see [N99] for a survey). The current article is a survey of the articles [GSY17a] and [GSY17b], where the authors studied the enumerative geometry of "nested Hilbert schemes" of points and curves on algebraic surfaces. Given the sequence n := n1, n2, . . . , nr, r ≥ 1 of nonnegative integers, and β := β1, . . . , βr−1, a sequence of classes in H 2(S, Z) such that βi ≥ 0, we denote the corresponding "nested " Hilbert scheme by S[n] β . A closed point of S[n] β corresponds to (Z1, Z2, . . . , Zr), (C1, . . . , Cr−1) where Zi ⊂ S is a 0-dimensional subscheme of length ni, and Ci ⊂ S is a divisor with [Ci] = βi, and Zi+1 is a subscheme of Zi ∪ Ci for any i < r, or equivalently (1) IZi(−Ci) ⊂ IZi+1. In [GSY17a] and [GSY17b], in order to define invariants for the nested Hilbert schemes (see [GSY17a, Definitions 2.13, 2.14]), the authors constructed a vir- tual fundamental class [S[n] β ]vir and then considered cases where one integrates NESTED HILBERT SCHEMES AND LOCAL DT THEORY 4 appropriate cohomology classes against it. More precisely, they constructed a natural perfect obstruction theory over S[n] β . This is done by studying the defor- mation/obstruction theory of the maps of coherent sheaves given by the natural inclusions (1) following Illusie. It turned out that this construction in particular provides a uniform way of studying all known obstruction theories of the Hilbert schemes of points and curves, as well as the stable pair moduli spaces on S. The first main result of [GSY17a] is as follows (see also [GSY17a, Propositions 2.5, and Corollary 2.6]): Theorem 1. ([GSY17a, Theorem 1]) Let S be a nonsingular projective surface β with r ≥ 2 over C and ωS be its canonical bundle.The nested Hilbert scheme S[n] carries a natural virtual fundamental class r−1(cid:88) i=1 β ]vir ∈ Ad(S[n] [S[n] β ), d = n1 + nr + 1 2 βi · (βi − c1(ωS)). 2.1. Special cases. In the simplest special case, i.e. when r = 1, we have S[n] β = S[n1] is the Hilbert scheme of n1 points on S which is nonsingular of dimension 2n1, and hence it has a well-defined fundamental class [S[n1]] ∈ A2n1(S[n1]). For r > 1 and βi = 0, S[n] := S[n] (0,...,0) is the nested Hilbert scheme of points on S parameterizing the flag of 0-dimensional subschemes Zr ⊂ ··· ⊂ Z2 ⊂ Z1 ⊂ S, which is in general singular of actual dimension 2n1. β = S[n1,n2] for some β ∈ H 2(S, Z). Interestingly, in the following cases the invariants of nested Hilbert schemes coincide with the Poincar´e and the stable pair invariants of S that were previously studied in the context of algebraic Seiberg-Witten invariants and curve counting problems. The following theorem is proven in [GSY17a, Section 3]. The authors were specifically interested in the case r = 2, that is: S[n] β Theorem 2. ([GSY17a, Proposition 3.1]) The virtual fundamental class of The- orem 1 recovers the following known cases: 1. If β = 0 and n1 = n2 = n then S[n,n] β=0 ∼= S[n] and fundamental class of the Hilbert scheme of n points. ∼= S[n] and [S[n,n] 2. If β = 0 and n2 = 0, then S[n,0] β=0 β=0 ]vir = [S[n]] is the β=0 ]vir = (−1)n[S[n]] ∩ cn(ω[n] [S[n,0] S ), where ω[n] S canonical bundle ωS of S.1 is the rank n tautological vector bundle over S[n] associated to the 1We were notified about this identity by Richard Thomas. NESTED HILBERT SCHEMES AND LOCAL DT THEORY 5 3. If β = 0 and n = n2 = n1 − 1, then it is known that S[n+1,n] ∼= P(I [n]) is nonsingular, where I [n] is the universal ideal sheaf over S[n] × S [L99, Section 1.2]. Then, β=0 4. If n1 = n2 = 0 and β (cid:54)= 0, then S[0,0] β=0 [S[n+1,n] ]vir = −[S[n+1,n] ] ∩ c1(OP(1) (cid:2) ωS). is the Hilbert scheme of divisors in class ]vir coincides with virtual cycle used to define Poincar´e invariants β=0 β β, and [S[0,0] in [DKO07]. β 5. If n1 = 0 and β (cid:54)= 0, then S[0,n2] β is the relative Hilbert scheme of points on the universal divisor over S[0,0] , which as shown in [PT10], is a moduli space of sta- ble pairs and [S[0,n2] ]vir is the same as the virtual fundamental class constructed in [KT14] in the context of stable pair theory. If Pg(S) = 0 this class was used in [KT14] to define stable pair invariants. β β In certain cases, the authors constructed a reduced virtual fundamental class for by reducing the perfect obstruction theory leading to Theorem 1 ([GSY17a, S[n1,n2] β Propositions 2.10, 2.12]): Theorem 3. ([GSY17a, Theorem 3]) Let S be a nonsingular projective surface with pg(S) > 0, and the class β be such that the natural map H 1(TS) ∗∪β−−→ H 2(OS) is surjective, β ]vir = 0. then, [S[n1,n2] reduced virtual fundamental class red ∈ Ad(S[n1,n2] ]vir [S[n1,n2] β β ), In this case the nested Hilbert scheme S[n1,n2] carries a β d = n1 + n2 + β · (β − KS) + pg. 1 2 The reduced virtual fundamental classes [S[0,0] ]vir red match with the re- duced virtual cycles constructed in [DKO07, KT14] in cases 3 and 4 of Proposition 2. [S[0,n2] ]vir red was used in [KT14] to define the stable pair invariants of S in this case. red and [S[0,n2] ]vir β β β 2.2. Nested Hilbert scheme of points. One interesting specialization of the nested Hilbert schemes is the case where β = 0 and n = n1 ≥ n2. In [GSY17a] the authors studied the nested Hilbert schemes of points S[n1≥n2] := S[n1,n2] β=0 in much more details. Let ι : S[n1≥n2] (cid:44)→ S[n1] × S[n2] be the natural inclusion. For the case where S is toric with the torus T and the fixed set ST, the authors provided a purely combinatorial formula for computing [S[n1≥n2]]vir by torus localization along the lines of [MNOP06]: NESTED HILBERT SCHEMES AND LOCAL DT THEORY 6 Theorem 4. ([GSY17a, Theorem 4]) For a toric nonsingular surface S the T-fixed set of S[n1,n2] is isolated and given by tuple of nested partitions of n2, n1: (cid:8)(µ2,P ⊆ µ1,P )P P ∈ ST, µi,P (cid:96) ni (cid:9) . Moreover, the T-character of the virtual tangent bundle T of S[n1≥n2] at the fixed point Q = (µ2,P ⊆ µ1,P )P is given by (cid:88) P∈ST trT vir Q (t1, t2) = VP , where t1, t2 are the torus characters and VP is a Laurent polynomial in t1, t2 that is completely determined by the µ2,P , µ1,P and is given by the right hand side of the following formula +(cid:0)Z1 · Z2 − Z1 · Z1 − Z2 · Z2 (cid:1) (1 − t1)(1 − t2) . (2) trT vir I1⊆I2 = Z1 + Z2 t1t2 t1t2 It turns out that when S is toric and Fano, by torus localization, one can express [S[n1≥n2]]vir in terms of the fundamental class of the product of Hilbert schemes S[n1] × S[n2]: Theorem 5. ([GSY17a, Theorem 5]) If S is a nonsingular projective toric Fano surface, then, ι∗[S[n1≥n2]]vir = [S[n1] × S[n2]] ∩ cn1+n2(En1,n2), where En1,n2 is the rank n1 + n2 vector bundle on S[n1] × S[n2] obtained by the first relative extension sheaf of the universal ideal sheaves I [n1] and I [n2]. Theorem 5 holds in particular for S = P2, P1 × P1, which are the generators of the cobordism ring of nonsingular projective surfaces. The authors used this fact together with a degeneration formula developed for [S[n1≥n2]]vir ([GSY17a, Proposition 5.1]) to prove: Theorem 6. If S is a nonsingular projective surface, and α is a cohomology class in H n1+n2(S[n1] × S[n2]) with the following properties: • α is universally defined for any pair of a nonsingular projective surface and • α is well-behaved under good degenerations of S, a line bundle on it, and En1,n2 is the rank n1 + n2 virtual vector bundle on S[n1] × S[n2] obtained by taking the alternating sum (in the K-group) of all the relative extension sheaves of I [n1] and I [n2]. then (cid:90) (cid:90) ι∗α = [S[n1≥n2]]vir S[n1]×S[n2] α ∪ cn1+n2(En1,n2), NESTED HILBERT SCHEMES AND LOCAL DT THEORY 7 The operators (cid:90) − ∪ cn1+n2(En1,n2 M ) S[n1]×S[n2] were studied by Carlsson-Okounkov in [CO12]. Here M ∈ Pic(S), and En1,n2 is the rank n1 + n2 virtual vector bundle over S[n1] × S[n2] obtained by taking the alternating sum of all the relative extensions of I [n1] and I [n2] (cid:2) M defined as follows: M Definition 2.1. ([GSY17a, Definition 4.3]) For any line bundles M on S, let En1,n2 M ∈ K(S[n1] × S[n2]) be the element of virtual rank n1 + n2 defined by ∗p∗M ] −(cid:2)RHomπ(cid:48)(I [n1],I [n2] ⊗ p∗M )(cid:3) , En1,n2 M := [Rπ(cid:48) where p and π(cid:48) are respectively the projections from S × S[n1] × S[n2] to the first and the product of last two factors. Let i be the inclusion of the closed point (I1, I2) ∈ S[n1] × S[n2], then, we define M (I1,I2) := [Li∗Rπ(cid:48) En1,n2 If M = O, we sometimes drop it from the notation. We also define the following generating series ∗p∗M ] −(cid:2)Li∗RHomπ(cid:48)(I [n1],I [n2] ⊗ p∗M )(cid:3) ∈ K(Spec(C)). (cid:88) (cid:90) c(En1,n2) ∪ c(En1,n2 M ). Zprod(S, M ) := qn1 1 qn2 2 n1≥n2≥0 S[n1]×S[n2] Note that by equation (5) in [CO12], we know ci(En1,n2 so the integrand in the definition of Zprod(S, M ) can be replaced by M ) = 0 for i > n1 + n2, and cn1+n2(En1,n2) ∪ cn1+n2(En1,n2 M ). Carlsson and Okounkov were able to express these operators in terms of explicit vertex operators. As an application of Theorem 6 and the result of [CO12], the following explicit formula was proven: Theorem 7. ([GSY17a, Theorem 7]) Let S be a nonsingular projective surface, ωS be its canonical bundle, and KS = c1(ωS). Then, (cid:88) n1≥n2≥0 (−1)n1+n2 (cid:90) (cid:89) [S[n1≥n2]]vir (cid:0)1 − qn−1 2 qn 1 ι∗c(En1,n2 2 = 1 qn2 M )qn1 (cid:1)(cid:104)KS ,KS−M(cid:105) (1 − qn 1 qn 2 ) (cid:104)KS−M,M(cid:105)−e(S) , where (cid:104)−,−(cid:105) is the Poincar´e paring on S and En1,n2 M is as in Theorem 5. n>0 NESTED HILBERT SCHEMES AND LOCAL DT THEORY 8 3. Nested Hilbert schemes and DT theory of local surfaces In this section we give an overview of the results of [GSY17b]. Let (S, h) be a nonsingular simply connected projective surface with h = c1(OS(1)). Let ωS be the canonical bundle of S with the projection map q to S, and X be the total 2. X is a noncompact Calabi-Yau threefold and one can define the DT space of ωS invariants of X by using C∗-localization, where C∗-acts on X by scaling the fibers of ωS. More precisely, let v = (r, γ, m) ∈ ⊕2 i=0H 2i(S, Q) be a Chern character vector, and MωS h (v) be the moduli space of compactly sup- ported 2-dimensional stable sheaves E on X such that ch(q∗ E) = v. Here stability is defined by means of the slope of q∗ E with respect to the polarization h. In [GSY17b] the authors provided MωS h (v) with a perfect obstruction theory by re- ducing the natural perfect obstruction theory given by [T98]. The fixed locus MωS of the moduli space is compact and the reduced obstruction theory gives a virtual fundamental class over it, denoted by [MωS ]vir red. They defined two types of DT invariants: h (v) h (v) C∗ C∗ (cid:90) DTωS h (v; α) = [MωS h (v)C∗ h (v) = χvir(MωS ]vir red h (v) e(Norvir) ) ∈ Z, C∗ DTωS 1 ∈ Q[s, s−1], α ∈ H∗ C∗(MωS h (v) C∗ )s where Norvir is the virtual normal bundle of MωS virtual Euler characteristic [FG10], and s is the equivariant parameter. C∗ ⊂ MωS h (v), χvir(−) is the h (v) If α = 1 then the authors were able to show that DTωS h (v; 1) = s−pg VWh(v), C∗ h (v) The C∗-fixed locus MωS where VWh(−) is the Vafa-Witten invariant (which was mathematically defined and studied in detail by Tanaka and Thomas in [TT17), TT18(2)]) and is expected to have modular properties based on S-duality conjecture (see [VW94]). consists of sheaves supported on S (the zero section of ωS) and its thickenings. One can write MωS as a disjoint union of several types of components, where each type is indexed by a partition of r. Out of these component types, there are two types of particular importance; One of them (we call it type I) is identified with Mh(v), the moduli space of rank r torsion free stable sheaves on S. The other type (we call it type II) can be identified with the nested Hilbert scheme S[n] β for a suitable choice of n, β depending on v. The reason that types I and II are more interesting, is the following result proven in [GSY17b]: h (v) C∗ 2In [GSY17b], we consider a more general case in which X is the total space of an arbitrary line bundle L with H 0(L ⊗ ω−1 S ) (cid:54)= 0. NESTED HILBERT SCHEMES AND LOCAL DT THEORY 9 Theorem 8. ([GSY17b, Theorem 2]) The restriction of [MωS h (v) I component Mh(v) is identified with [Mh(v)]vir perfect obstruction theory over Mh(v). The restriction of [MωS II component S[n] ]vir red to the type induced by the natural trace free C∗ ]vir red to a type β is identified with [S[n] β ]vir. h (v) C∗ 0 When r = 2, then types I and II are the only component types of MωS C∗ . h (v) This leads us to the following result: Theorem 9. ([GSY17b, Theorem 3]) Suppose that v = (2, γ, m). Then, DTωS h (v; α) = DTωS h (v; α)I + h (v) = χvir(Mh(v)) + DTωS DTωS h (v; α) II,S[n1,n2] β , χvir(S[n1,n2] β ), (cid:88) (cid:88) n1,n2,β n1,n2,β where the sum is over all n1, n2, β (depending on v) for which S[n1,n2] component of MωS I and II components to the invariant DTωS is a type II , and the indices I and II indicate the contributions of type h (v) C∗ β h (v; α). The stability of sheaves imposes a strong condition on n1, n2, β appearing in the summation in the theorem above. For example, if S is a generic complete intersection in a projective space, then for any n1, n2, β for which S[n1,n2] is a type II component of MωS , the condition in Theorem 3 (leading to the vanishing [S[n1,n2] ]vir = 0) is not satisfied. h (v) C∗ β β The invariants χvir(S[n1,n2] β ) and DTωS h (v; α) II,S[n1,n2] β (for a suitable choice of class α e.g. α = 1) appearing in the theorem above are special types of the invariants NS(n1, n2, β;−) β which are defined as follows: Definition 3.1. ( [GSY17a, Definition 2.13]) Let M ∈ Pic(S). Define the follow- ) of virtual ranks respectively n1 + n2 and −β · βD/2 + ing elements in K(S[n1,n2] β · c1(M ): β;M := [Rπ∗M (Zβ)] −(cid:104) S[ni] := [Rπ∗M ] −(cid:2)RHomπ(I [ni],I [ni] ⊗ M )(cid:3) =(cid:2)Ext1 Kn1,n2 If β = 0 we will instead use the notation K[n1≥n2] 5.4]). We also define the rank 2ni twisted tangent bundles (cid:105) β ⊗ M ) , Gβ;M :=(cid:2)Rπ∗M (Zβ)Zβ (cid:3) . (cid:0)I [ni],I [ni] ⊗ M(cid:1) 0;M (see [GSY17a, Definition RHomπ(I [n1],I [n2] := Kn1,n2 (cid:3) . TM M π 0 Note that if M = O then TM S[ni] = TS[ni]. NESTED HILBERT SCHEMES AND LOCAL DT THEORY 10 Let P := P(M, β, n1, n2) be a polynomial in the Chern classes of Kn1,n2 β;M , Gβ;M , TS[n1], and TS[n2], then, we can define the invariant Moreover under the condition in Theorem 3, the authors defined the reduced in- variants (cid:90) (cid:90) NS(n1, n2, β;P) := P. ]vir [S[n1,n2] β S (n1, n2, β;P) := Nred P. [S[n1,n2] β ]vir red Mochizuki in [M02] expresses certain integrals against the virtual cycle of Mh(v) in terms of Seiberg-Witten invariants and integrals A(γ1, γ2, v;−) over the product of Hilbert scheme of points on S. Using this result the authors were able to prove the following : Theorem 10. ([GSY17b, Theorem 4]) Suppose that pg(S) > 0, and v = (2, γ, m) is such that γ · h > 2KS · h and χ(v) :=(cid:82) h (v; 1) = − (cid:88) h (v) = − (cid:88) γ1+γ2=γ γ1·h<γ2·h DTωS DTωS γ1+γ2=γ γ1·h<γ2·h (cid:88) S v · tdS ≥ 1. Then, (cid:88) n1,n2,β n1,n2,β SW(γ1) · 22−χ(v) · A(γ1, γ2, v; P1) + SW(γ1) · 22−χ(v) · A(γ1, γ2, v; P2) + NS(n1, n2, β;P1). NS(n1, n2, β;P2). Here SW(−) is the Seiberg-Witten invariant of S, Pi and Pi are certain universally defined explicit integrands, and the second sum in the formulas is over all n1, n2, β (depending on v) for which S[n1,n2] is a type II component of MωS C∗ . h (v) β If S is a K3 surface or S is isomorphic to one of the five types of generic complete intersections (5) ⊂ P3, (3, 3) ⊂ P4, (4, 2) ⊂ P4, (3, 2, 2) ⊂ P5, (2, 2, 2, 2) ⊂ P6, the DT invariants DTωS of integrals over the product of the Hilbert schemes of points on S. h (v; 1) and DTωS h (v) can be completely expressed as the sum References [BF97] Kai Behrend, and Barbara Fantechi. "The intrinsic normal cone." Inventiones Mathe- maticae 128, (1997): 45 -- 88. [BCDDQS16] , Vincent Bouchard, Thomas Creutzig, Duiliu-Emanuel Diaconescu, Charles Do- ran, Callum Quigley, and Artan Sheshmani, "Vertical d4-d2-d0 bound states on k3 fibrations and modularity", Communication in Mathematical Physics, 350(3), (2016): 1069 -- 1121. [BFl03] Ragnar-Olaf Buchweitz, and Hubert Flenner. "A semiregularity map for modules and applications to deformations." Compositio Mathematica, 137, (2003): 135 -- 210. [CO12] Erik Carlsson, and Andrei Okounkov. "Exts and vertex operators." Duke Mathematical Journal, 161, (2012): 1797 -- 1815. NESTED HILBERT SCHEMES AND LOCAL DT THEORY 11 [C98] Jan Cheah."Cellular decompositions for nested Hilbert schemes of points." Pacific Journal of Mathematics , 183, (1998): 39 -- 90. [D96] Simon Donaldson. "The Seiberg-Witten equations and 4-manifold topology", Bulletin of American Mathematical Society (N.S.), 33(1), (1996): 45 -- 70. [DKO07] Markus Durr, , Alexandre Kabanov, and Christian Okonek. "Poincar invariants." Topology, 46, (2007): 225 -- 294. [ES87] Geir Ellingsrud, and Stein Arild Strømme. "On the homology of the Hilbert scheme of points in the plane." Inventiones Mathematicae 87, (1987): 343 -- 352. [EGL99] Geir Ellingsrud, and Lothar Gottsche and Manfred Lehn. "On the cobordism class of the Hilbert scheme of a surface." Journal of Algebraic Geometry, 10, (2001): 81 -- 100. [FG10] Barbara Fantechi, and Lothar Gottsche. "Riemann-Roch theorems and elliptic genus for virtually smooth schemes." Geometry & Topology, 14, (2010): 83 -- 115. [F13] William Fulton. "Intersection theory" Vol. 2. Springer Science & Business Media, (2013). [G90] Lothar Gottsche. "The Betti numbers of the Hilbert scheme of points on a smooth pro- jective surface", Mathematische Annalen, 286, (1990): 193 -- 207. [GP99] Tom Graber, and Rahul Pandharipande. "Localization of virtual classes." Inventiones Mathematicae 135, (1999): 487 -- 518. [GS16] Amin Gholampour, and Artan Sheshmani. "Intersection numbers on the relative Hilbert schemes of points on surfaces." Asian Journal of Mathematics, 21 (3), (2016): 531 -- 542. [GS18] Amin Gholampour, and Artan Sheshmani. "Donaldson-Thomas Invariants of 2- Dimensional sheaves inside threefolds and modular forms." Advances in Mathematics, 326 (21), (2018): 79 -- 107. [GS14] Amin Gholampour, Artan Sheshmani, and Richard P Thomas, "Counting curves on surfaces in calabi-yau 3-folds", Mathematische Annalen, 360, (2014): 67 -- 78. [GST17] Amin Gholampour, Artan Sheshmani, and Yukinobu Toda, "Stable pairs on nodal k3 fibrations", International Mathematics Research Notices, 2017(00), (2017): 1 -- 50. [GSY17a] Amin Gholampour, Artan Sheshmani, and Shing-Tung Yau. "Nested Hilbert schemes on surfaces: virtual fundamental class." arXiv:1701.08899. [GSY17b] Amin Gholampour, Artan Sheshmani and Shing-Tung Yau. "Localized Donaldson- (To Appear), surfaces", American Journal of Mathematics Thomas arXiv:1701.08902 (Preprint). theory of [GLSY17] Sergei Gukov, Artan Sheshmani, Chiu-Chu. Melissa Liu, and Shing-Tung Yau. "Topo- logical approach to local theory of surfaces in Calabi-Yau threefolds", Advances in Theoret- ical and Mathematical Physics 21(7), (2017): 1679 -- 1728. [HL10] Daniel Huybrechts, and Manfred Lehn. "The geometry of moduli spaces of sheaves." Cambridge University Press, (2010). [HT10] Daniel Huybrechts, and Richard P. Thomas. "Deformation-obstruction theory for com- plexes via Atiyah and Kodaira-Spencer classes." Mathematische Annalen 346, (2010): 545 -- 569. [Ill] Illusie, L. "Complex Cotangent et Deformations I." Lecture Notes in Math 239. [JS12] Dominic Joyce, and Yinan Song. "A theory of generalized Donaldson-Thomas invariants." Memoirs of American Mathematical Society, 217, (1020). [KKV99] Sheldon Katz, Albrecht Klemm, and Cumrun Vafa.,"M-theory, topological strings and spinning black holes", Advances in Theoretical and Mathematical Physics, 3, (1999): 1445 -- 1537. [KL13] Kiem, Young-Hoon, and Jun Li. "Localizing virtual cycles by cosections." Journal of the American Mathematical Society, 26, (2013): 1025 -- 1050. [KMPS10] A. Klemm, D. Maulik, R. Pandharipande, and E. Scheidegger,"Noether-Lefschtz the- ory and the Yau-Zaslow conjecture",J. Amer. Math. Soc. , 23, (2010):1013 -- 1040. NESTED HILBERT SCHEMES AND LOCAL DT THEORY 12 [K90] J´anos Koll´ar. "Projectivity of complete moduli." Journal of Differential Geometry, 32, (1990): 235 -- 268. [KM77] Finn Knudsen, and David Mumford. "The projectivity of the moduli space of stable curves I: preliminaries on "det" and "Div"." Mathematica Scandinavica, 39, (1977): 19 -- 55. [KST11] Martijn Kool, Vivek Shende, and Richard P. Thomas. "A short proof of the Gttsche conjecture." Geometry & Topology, 15, (2011): 397 -- 406. [KT14] Martijn Kool, and R. P. Thomas. "Reduced classes and curve counting on surfaces I: theory." Algebraic Geometry, 1, (2014): 334 -- 383. [L99] Manfred Lehn. "Chern classes of tautological sheaves on Hilbert schemes of points on surfaces." Inventiones Mathematicae, 136, (1999): 157 -- 207. [L04] Manfred Lehn. "Lectures on Hilbert schemes." Algebraic structures and moduli spaces, 38, (2004): 1 -- 30. [LP12] Yaun-Pin Lee, and Rahul Pandharipande. "Algebraic cobordism of bundles on varieties." Journal of European Mathematical Society, 14(4), (2012):1081-1101. [LP09] Marc Levine, and Rahul Pandharipande. "Algebraic cobordism revisited." Inventiones Mathematicae, 176, (2009): 63 -- 130. [L01] Jun Li. "Stable morphisms to singular schemes and relative stable morphisms." Journal of Differential Geometry 57, (2001): 509 -- 578. [L02] Jun Li. "A degeneration formula of GW-invariants." Journal of Differential Geometry 60, (2002): 199 -- 293. [LT14] Jun Li, and Yu-jong Tzeng. "Universal polynomials for singular curves on surfaces." Compositio Mathematica, 150, (2014): 1169 -- 1182. [LW15] Jun Li, and Baosen Wu. "Good degeneration of Quot-schemes and coherent systems." Communications in Analysis and Geometry, 23, (2015): 841 -- 921. [M02] Takuro Mochizuki. "A theory of the invariants obtained from the moduli stacks of stable objects on a smooth polarized surface." arXiv preprint math/0210211 (2002). [MP13] Davesh Maulik, and Rahul Pandharipande."Gromov-Witten theory and Noether- Lefschetz theory", Clay Math. Proc., 8, (2013). [MNOP06] Davesh Maulik, Nikita Nekrasov, Andrei Okounkov, and Rahul Pandharipande. "Gromov -- Witten theory and Donaldson -- Thomas theory, I." Compositio Mathematica, 142, (2006): 1263 -- 1285. [MNOPII] Davesh Maulik, Nikita Nekrasov, Andrei Okounkov, and Rahul Pandharipande. "Gromov -- Witten theory and Donaldson -- Thomas theory, II." Compositio Mathematica, 142, (2006): 1286 -- 1304. [MPT10] Davesh Maulik, Rahul Pandharipande, and Richard P. Thomas. "Curves on K3 surfaces and modular forms." Journal of Topology, 3, (2010): 937 -- 996. [N99] Hiraku Nakajima. "Lectures on Hilbert schemes of points on surfaces." American Mathe- matical Soc., 18, (1999). [PT16] Rahul Pandharipande and Richard P. Thomas,"The Katz-Klemm-Vafa conjecture for K3 surfaces", Forum Math. Pi, 4:e4, 111, (2016). [PT09] Rahul Pandharipande and Richard. P. Thomas, "The 3-fold vertex via stable pairs", Geometry and Topology, 13, (2009): 1835 -- 1876. [PT10] Rahul Pandharipande, and Richard P. Thomas. "Stable pairs and BPS invariants." Jour- nal of the American Mathematical Society, 23, (2010): 267-297. [TT17)] Yuuji Tanaka, and Richard P. Thomas. "Vafa-Witten invariants for projective surfaces I." arXiv:1702.08487 (Preprint). [TT18(2)] Yuuji Tanaka, and Richard P. Thomas. "Vafa-Witten invariants for projective surfaces II." Pure and Applied Mathematics Quarterly, 13(13), (2017): 517 -- 562. NESTED HILBERT SCHEMES AND LOCAL DT THEORY 13 [T98] Richard P. Thomas. "A holomorphic Casson invariant for Calabi-Yau 3-folds, and bundles on K3 fibrations." Journal of Differential Geometry, 54, (2000): 367 -- 438. [T12] Yu-Jong Tzeng. "A proof of the Gttsche-Yau-Zaslow formula." Journal of Differential Geometry 90, (2012): 439 -- 472. [YZ96] Shing-Tung Yau and Eric Zaslow, "BPS states, string duality, and nodal curves on K3", Nuclear Phys. B, 471(3), (1996): 503 -- 512. [VW94] Cumrun Vafa, and Edward Witten. "A strong coupling test of S-duality." Nuclear Physics B, 431, (1994): 3 -- 77. Center for Mathematical Sciences and Applications, Harvard University, De- partment of Mathematics, 20 Garden Street, Room 207, Cambridge, MA, 02139 Centre for Quantum Geometry of Moduli Spaces, Aarhus University, Depart- ment of Mathematics, Ny Munkegade 118, building 1530, 319, 8000 Aarhus C, Denmark National Research University Higher School of Economics, Russian Federation, Laboratory of Mirror Symmetry, NRU HSE, 6 Usacheva str., Moscow, Russia, 119048 [email protected],